eCAADe 2016

Complexity & Simplicity

Volume 2

Volume 2 - eCAADe 34 | 1
 Editors
Aulikki Herneoja
Toni Österlund
Piia Markkanen
Oulu School of Architecture
University of Oulu

1st Edition, August 2016

Complexity & Simplicity - Proceedings of the 34th International Conference on Education and
Research in Computer Aided Architectural Design in Europe, Oulu, Finland, 24.-26. August
2016, Volume 2. Edited by Aulikki Herneoja, Toni Österlund and Piia Markkanen. Brussels: Ed-
ucation and Research in Computer Aided Architectural Design in Europe; Oulu: Oulu School
of Architecture, University of Oulu.

ISBN
9789491207112
978-94-91207-11-2

Copyright © 2016

Publisher: eCAADe (Education and Research in Computer Aided Architectural Design in Eu-
rope) and Oulu School of Architecture, University of Oulu.

Cover Design: Piia Markkanen and Toni Österlund

All rights reserved. Nothing from this publication may be produced, stored in computerised
system or published in any form or in any manner, including electronic, mechanical, repro-
graphic or photographic, without prior written permission from the publisher.

2 | eCAADe 34 - Volume 2
 eCAADe 2016

Complexity & Simplicity

Volume 2
Proceedings of the 34th International Conference on Education and Research in
Computer Aided Architectural Design in Europe

24.-26. August 2016

Oulu, Finland
Oulu School of Architecture
University of Oulu

Edited by
Aulikki Herneoja
Toni Österlund
Piia Markkanen

Volume 2 - eCAADe 34 | 3
 Sponsors of the eCAADe 2016 Conference

Alfred Kordelin Foundation

4 | eCAADe 34 - Volume 2
Theme

Complexity & Simplicity

The second volume of the conference proceedings of the 34rd eCAADe conference, 24.-26.
August 2016, at Oulu School of Architecture, University of Oulu, Oulu, Finland contains 76 pa-
pers grouped under 13 sub-themes. Both volumes contain altogether 157 accepted papers.

The theme of the 34th eCAADe Conference is Complexity & Simplicity. We invited the eCAADe
community to address the multifaceted notions of complexity and simplicity, which are en-
countered in architectural design processes. Approaches discussing the theme from the per-
spective of computer aided design education; design processes and methods; design tool de-
velopments; and novel design applications, as well as real world experiments and case studies
were welcomed. What is the role of complexity or simplicity as part of the design process?
Does the use of complex design methods offer simplicity to the design process itself? Is it
possible to design complexity with simple methods? Does the use of computation in design
necessitate complexity or offer means to control it?

Recent development in digital technologies and digital design tools enable us to address com-
plex situations in architectural environments, ranging in scale from structures and buildings
to urban contexts. We often expect technology to better help us manage the complexity of
life, to simplify our daily lives and tasks. However, these developments also raise the question
of whether design technologies encourage complexity at the expense of simplicity in both
the design process and lived environments. Does computation cause complexity? Or does it
enable simplicity?

In addition to the accepted papers, the first volume contains Keynote Papers, including keynote
speakers contribution concerning the themes of their keynote lectures and the Workshop Con-
tributions, including the papers summarizing the contents of workshops given.

Theme - Volume 2 - eCAADe 34 | 5

 All the papers of these proceedings will be accessible via CuminCAD - Cumulative Index of
Computer Aided Architectural Design, http://cumincad.scix.net

eCAADe 2016 Acting Conference Chairs

Aulikki Herneoja, Toni Österlund, Piia Markkanen

6 | eCAADe 34 - Theme - Volume 2

Acknowledgements

Welcome to eCAADe 2016, the 34th eCAADe conference, in Oulu, Finland. It has been exactly
fifteen years since the previous eCAADe conference was held in Finland, hosted by Helsinki
University of Technology (HUT) in 2001. HUT was also in charge of organizing the 3rd eCAADe
conference in 1984. We are now very happy to welcome you back to Finland, this time to the
most northern architect school in the world, the Oulu School of Architecture, University of
Oulu.
The original idea of bringing the eCAADe conference to Oulu dates back to September
2012 after the eCAADe conference in Prague. Officially, the eCAADe Council granted us the
permission to organize the 34th eCAADe conference in Oulu in March 2013. Over the last four
years several people have helped us to make this conference happen. We thank especially
the former Dean of Oulu School of Architecture, now the Vice Rector for Education, University
of Oulu, Professor Helka-Liisa Hentilä and the present Dean of Oulu School of Architecture,
Professor Rainer Mahlamäki for their positive and supportive attitude.
During the lengthy process of organizing the eCAADe 2016 we have had the privilege to
experience the supportive and kind atmosphere of eCAADe Council, whose members have
helped us with all aspects of the organizing. We warmly thank both Presidents Johan Verbeke
(until 2016) and Joachim Kieferle (since 2016) for encouragement and unswerving support
during the organizing of this conference. We most warmly thank Bob Martens, the liaison with
the conference host and the previous conference organizer, for always kindly and patiently
guiding us through the multistage arrangement process of the conference. Also, we wish to
thank all the other previous conference organizers, Henri Achten, Rudi Stouffs and Emine Mine
Thompson, for sharing their experience and knowledge without hesitation.
Quality control is an emerging issue concerning the publishing of a conference proceed-
ings. We are grateful for eCAADe Council for being able to use the OpenConf system through-
out the publishing process. Authors uploaded their extended abstracts (length of 1000 to
1500 words and one optional image) for the double blind peer review process. Each abstract

Acknowledgements - Volume 2 - eCAADe 34 | 7

 was evaluated by three reviewers. With the help of the OpenConf system we could, for exam-
ple, easily ensure that none of the reviewers came from the same institution as the authors.
Altogether, we received 241 abstracts from authors from 44 different countries. After the peer
review process, 165 papers were accepted for full paper submission. In the end, altogether
157 papers were presented in eCAADe 2016 and published in the proceedings. We are very
grateful for all the 98 reviewers from 29 different countries (see the List of Reviewers) for their
constructive and thorough comments for each author. We also continued the practice started
in eCAADe 2015 conference in Vienna of having all the session chairs to give prospective com-
ments of the papers to evoke the discourse at early stage between the author and session
chair for the 27 sessions of the conference. All the session chairs also participated the peer
review process of the extended abstracts. We owe the session chairs great gratitude for their
commitment and long term contribution to the process until the final paper presentations.
Parallel to these prospective comments editorial team gave comments to the authors too. We
thank and congratulate all authors for their hard work and support on using the ProceeDings
tool and finalizing their full papers carefully in time. In this last phase of editing the full pa-
pers we may not thank enough Gabriel Wurzer, the Master of the ProceeDings, who patiently
and relentlessly worked together with us and enabled us to successfully produce high quality
proceedings in time.
We owe great thanks to the keynote speakers; Professor Branko Kolarevic (Chair in Inte-
grated Design and co-directs the Laboratory for Integrative Design (LID), University of Calgary;
Canada), Professor Mette Ramsgaard Thomsen (Chair in Architecture and Digital Technologies,
Head of CITA, Centre for Information Technology and Architecture, Royal Academy of Fine Arts,
School of Architecture, Design and Conservation (KADK), Denmark) and Professor Toni Kotnik
(Chair in Design of Structures at Department of Architecture in Aalto ARTS, Finland), and their con-
tribution of writing the keynote papers concerning their keynote lecture themes.
The Panel Session is a traditional Thursday afternoon event in eCAADe conferences. This
year the panelist were Finnish colleagues, the top level practicing architects. The visionary
discussion was moderated by Director Hanna Harris (Architecture Information Centre Finland).
The panelists were; Professor Rainer Mahlamäki (Lahdelma & Mahlamäki Architects), Architect

8 | eCAADe 34 - Acknowledgements - Volume 2

Antti Nousjoki (ALA Architects Ltd) and Architect Toni Österlund (Geometrics Ltd). We owe great
thanks to the moderator and the panelists for their rewarding presentations and discussions.
Workshops are an integral part of eCAADe conferences and eCAADe 2016 made no excep-
tion. We thank all the workshop organizers for the event and for their contribution of short
papers (non-peer reviewed) about the contents of the workshop. We are also grateful for Wolf-
gang Dokonal and the eCAADe council for organizing the traditional PhD workshop for young
researchers and supporting the participants with a subsidy for traveling to Oulu.
In addition to printed and digital conference proceedings, the team supervised by Wolf-
gang Dokonal provides the eCAADe community video streams of the paper presentations.
Harri Hämeenkorpi and Ari Orajärvi from University of Oulu were locally in charge of recording
and video streaming with the help of architecture students from Oulu School of Architecture.
Great thanks for the whole team.
We would like to express our gratitude for the administrative help in organizing this con-
ference. eCAADe council and especially Nele De Meyere, has provided us valuable input and
lessons learned from past conferences. We have also had Konffa Ltd as highly-appreciate part-
ner for managing the conference services, ranging from the registration process to the actual
on-site registration services. A big thank goes you to Pia Hanski, Hanna Hentilä and Jenni
Ylisirniö. We will also thank Erweko Ltd and Mikko Träskelin for smooth process and high qual-
ity printing of the proceedings.
Organizing an international conference of this scale requires financial help. We are in-
debted for our sponsors – The Federation of Finnish Learned Societies, Alfred Kordelin General
Progress and Education Fund, Autodesk Inc., Bentley Systems International Ltd., and the Oulu
School of Architecture, University of Oulu. Their sponsorship enabled us to fulfill the quality
standards of eCAADe conference. As a special form of sponsorship, the members of the local
conference staff donated their time to help prepare and organize this conference. Thank you
all for helping us out.

eCAADe 2016 Acting Conference Chairs

Aulikki Herneoja, Toni Österlund, Piia Markkanen

Acknowledgements - Volume 2 - eCAADe 34 | 9

Contents
5 Theme
7 Acknowledgements

17 CITY MODELING | Urban Planning Approach

19 Crowd-Sourced Neighborhoods
Carlos Sandoval Olascoaga, Wenfei Xu, Hector Flores
31 Computational Approach for the Assessment of Transit Oriented
Development Principles
Fernando Lima, Nuno Montenegro, José Nuno Beirão, Jose Kos
41 Agent-Based Urban Growth Simulation
Cemal Koray Bingöl, Birgül Çolakoğlu
49 Towards Understanding the Complexity of Urban Culture
Malgorzata Hanzl
59 Parametric Modelling with GIS
Patrick Janssen, Rudi Stouffs, Akshata Mohanty, Elvira Tan, Ruize Li

69 CITY MODELING | Applications

71 Autopoietic Features of the Urban Body's Elements
Eirini Androutsopoulou
79 Understanding the Role of Spatial Connectivity in Integrating Informal
Settlements, through the Case of Medellin's Urban Cable-cars.
Paul Goodship
89 Visualizing Urban Sports Movement
Ozgun Balaban, Bige Tuncer
95 Designing for Urban Microclimates: Towards A Generative
Performance-based Approach to Wind Flow Optimization
Mohamed Khallaf, Julie Jupp

107 DESIGN TOOLS | Evaluations

109 Shifting Design Work from Production to Evaluation
Günter Barczik, Rolf Kruse
117 Integrated Spatial-Structural Optimization in the Conceptual Design Stage of
Project
Subhajit Das, Samaneh Zolfagharian, Mehdi Nourbakhsh, John Haymaker

Contents - eCAADe 34 | 11
 127 Tangible Grasshopper
Boris Plotnikov, Gerhard Schubert, Frank Petzold
137 A Simple System for Complex Mass Housing Design Collaborations
Tian Tian Lo, Marc Aurel Schnabel, Tane Moleta
147 Evaluating Daylighting Analysis of Complex Parametric Facades
Mohamed Gomaa, Wassim Jabi

157 DESIGN TOOLS | Applications

159 How Do Small and Medium Architectural Firms Deal with Architectural
Complexity? A Look Into Digital Practices
Adeline Stals, Sylvie Jancart, Catherine Elsen
169 Estimating Potential Event Occurrence Areas in Small Space based on
Semi-supervised Learning
Atsushi Takizawa
179 In Search of Design Parameters for Well-Being and Creativity in Knowledge
Work Environments
Piia Markkanen, Aulikki Herneoja
189 Manifold Façades
Camilla Guerritore, José Pinto Duarte
199 Assisting Early Architectural Planning Using a Geometry-Based Graph Search
Torsten Thurow, Christoph Langenhan, Frank Petzold
209 ColorTracker
Stefanie Holzheu, Sang Lee
217 Automated Service Core Generator in Autodesk Dynamo
Subhajit Das, Colin Day, Michael Dewberry, Varvara Toulkeridou, Anthony Hauck

227 DESIGN TOOLS | Explorations

229 SpaceBook
Gabriel Wurzer, Wolfgang E. Lorenz
239 Surveying Design Spaces with Performance Maps
Thomas Wortmann
249 Spatial Optimisations
Reinhard Koenig, Tasos Varoudis
255 Structural Optimisation Methods as a New Toolset for Architects
Sebastian Białkowski

12 | eCAADe 34 - Contents
265 Soft Modelling
Manuel Jimenez Garcia
275 Esquis'sons ! Sound Sketch : A Parametric Tool to Design Sustainable
Soundscapes
Théo Marchal, Nicolas Remy, Grégoire Chelkoff, Jean-Luc Bardyn, Noha Gamal,
Hengameh Pirhosseinloo

285 SPATIAL REASONING AND ONTOLOGIES

287 Abstract Object in the World of Data
Miro Roman
297 Architecture of Intermodal Complex
Sang Lee
305 Ontologies and Shape Grammars
Rui de Klerk, José Beirão
315 Human Behaviour Simulation to Enhance Workspace Wellbeing and
Productivity
Armando Trento, Antonio Fioravanti

327 SHAPE, FORM AND GEOMETRY | Grammars and Concepts

329 An Algebraic Approach to Implementing a Shape Grammar Interpreter
Rudi Stouffs
339 Style and Type in a Generic Shape Grammar
Garcia Sara, Romão Luís
349 Wood Mass-Customized Housing
Krystian Kwiecinski, Filipe Santos, Ana de Almeida, Bruno Taborda, Sara Eloy
359 Utilizing Gradient Analysis within Interactive Genetic Algorithms
Matthias Kulcke, Wolfgang E. Lorenz
365 3D Reconstruction Survey of Complex Informal Settlements
Debora Verniz, Luis Mateus, José Pinto Duarte, Victor Ferreira
371 Optimization of the Building in Relation to the Insolation Conditions of
Premises in Adjacent Buildings
Agata Pasternak
379 Daylighting Based Parametric Design Exploration of 3D Facade Patterns
Amartuvshin Narangerel, Ji-Hyun Lee, Rudi Stouffs

Contents - eCAADe 34 | 13
 389 SHAPE, FORM AND GEOMETRY | Applications
391 Geometric Versatility of Abeille Vault
Irina Miodragovic Vella, Toni Kotnik
399 Origami Explorations
Yomna Saad ElGhazi, Ayman Hassaan Ahmed Mahmoud
409 Structural Architectural Elements Made of Curved Folded Sheet Metal
Vlad Andrei Raducanu, Vasile Danut Cojocaru, Doina Raducanu
417 Collaborative Modeling with Symbolic Shape Grammars
Martin Ilčík, Michael Wimmer
427 Revising Stereotomy through Digital Technology
Pedro de Azambuja Varela, José Pedro Sousa

435 SPATIAL ANALYSIS | Urban Context

437 Mapping Planned and Emerging Art Places in Singapore through Social
Media Feeds
Ludovica Tomarchio, Bige Tuncer, Linlin You, Bernhard Klein
447 Social Media as Complementary Tool to Evaluate Cities
Nai Chun Chen, Takehiko Nagakura, Kent Larson
457 Big Data for Urban Design
Gideon Aschwanden
463 Development of The Method for Estimating Traffic Volume of Pedestrians in
An Underground Mall by Use of Watch Cameras
Naoya Takagi, Atsushi Takizawa
473 Synthetic Modelling of Pedestrian Movement
Renee Puusepp, Damiano Cerrone, Martin Melioranski

483 DIGITAL HERITAGE

485 Prototyping a Temporospatial Simulation Framework
Augustus Wendell, Burcak Ozludil Altin, Ulysee Thompson
493 Modelling Medieval Vaults: Comparing Digital Surveying Techniques to
Enhance our Understanding of Gothic Architecture
Nicholas Webb, Alexandrina Buchanan, John Robert Peterson
503 Behavioural Simulation for Built Heritage Use Planning
Davide Simeone, Ugo Maria Coraglia, Stefano Cursi, Antonio Fioravanti

14 | eCAADe 34 - Contents
511 3D Digital Reconstructions of Lost Buildings
Danilo Di Mascio, Michele Chiuini, John Fillwalk, Pieter Pauwels
521 Simplifying Architectural Heritage Visualization
Derya Gulec Ozer, Takehiko Nagakura
529 An Intuitive Heritage Education System for Learning Architectural Structures
and Styles
Jongwook Lee, Aram Min, Jihyun Lee

539 VIRTUAL REALITY

541 Development of a Virtual Reality Solution for End User Involvement in
Interior Design
Kjeld Svidt, Jesper Bendix Sørensen
547 High quality Virtual Reality for Architectural Exhibitions
Anette Kreutzberg
555 Design Tools and Complexity: Mobile Games and Collective Imagination
Alexander Holland, Stanislav Roudavski
565 BIM Collaboration in Virtual Environments
Uwe Woessner, Joachim B. Kieferle
573 VR or Not VR - No Longer a Question?
Wolfgang Dokonal, Michael Knight, Ernst Dengg

581 BIM | Concepts

583 Automated Generation of BIM Models
Patrick Janssen, Kian Wee Chen, Akshata Mohanty
591 Information and Construction: Advanced Applications of Digital Prototyping
in the Housing Industry
Basem Eid Mohamed, Frederic Gemme, Aaron Sprecher
601 Algorithm-Aided BIM
Harri Humppi, Toni Österlund
611 Divide and Conquer, Mix and Match
Helga Tauscher, Raimar J. Scherer
621 Developing a Tangible User Interface for Parametric and BIM Applications
Using Physical Computing Systems.
Emad Al-Qattan, Philip Galanter, Wei Yan

Contents - eCAADe 34 | 15
 631 BIM | Applications
633 Captivity or Flexibility: Complexities in a Dimensional Customization System
Salman Khalili-Araghi, Branko Kolarevic
643 Tracking Changes in Buildings over Time
Martin Tamke, Mateusz Zwierzycki, Henrik Leander Evers, Sebastian Ochmann,
Richard Vock, Raoul Wessel
653 Toward Space Oriented BIM Practices
Aida Siala, Najla Allani , Gilles Halin, Mohamed Bouattour
663 TopoBIM: Web-based Spatial Topology for Early Design Participation
Mark Meagher, Phillip Langley
673 Parametric Architectural Design with Point-clouds
Mateusz Zwierzycki, Henrik Leander Evers, Martin Tamke

16 | eCAADe 34 - Contents
CITY MODELING | Urban Planning
Approach
Crowd-Sourced Neighborhoods
User-Contextualized Neighborhood Ranking

Carlos Sandoval Olascoaga1 , Wenfei Xu2 , Hector Flores3

1,2,3
Massachusetts Institute of Technology
1
csandova@mit.edu 2,3 {xu.wenfei|hectorcflores}@gmail.com

Finding an attractive or best-fit neighborhood for a new resident of any city is not
only important from the perspective of the resident him or herself, but has larger
implications for developers and city planners. The environment or mood of the
right neighborhood is not simply created through traditional characteristics such
as income, crime, or zoning regulations - more ephemeral traits related to
user-perception also have significant weight. Using datasets and tools previously
unassociated with real-estate decision-making and neighborhood planning, such
as social media and machine learning, we create a non-deterministic and
customized way of discovering and understanding neighborhoods. Our project
creates a customizable ranking system for the 195 neighborhoods in New York
City that helps users find the one that best matches their preferences. Our team
has developed a composite weighted score with urban spatial data and social
media data to rank all NYC neighborhoods based on a series of questions asked
to the user. The project's contribution is to provide a scientific and calibrated
understanding of the impact that socially oriented activities and preferences have
towards the uses of space.

Keywords: Textual Semantic analysis, machine learning, participatory planning,

community detection, neighborhood definition

INTRODUCTION putation process, and going beyond traditional ana-

With contemporary computational advances urban lytical models.
models have become richer, and represent more as-
A Contextual Urban Computing
pects of the city, they have not necessarily become
The city is constructed iteratively and in an inter-
more accurate in their predictions (Batty 2016). We
connected manner by its different inhabitants; the
believe that the kind of problem the city is cannot be
most quantitatively optimal representation may not,
addressed through a single comprehensive model,
in fact, be the most socially accurate one. As Herbert
but through contextual individual representations of
Simon suggested in "The Sciences of the Artificial", ur-
the world -our platform presents a first step towards
ban growth is comparable to the process of painting
that by harnessing personal participation in the com-
(Simon 1981), where the construction of the overall

CITY MODELING | Urban Planning Approach - Volume 2 - eCAADe 34 | 19

 result does not occur in a single homogenous layer.
Rather, local social interactions and their accretions
constantly reframe, modify, and construct the larger
whole. Cities are continuously reframed and reinter-
preted by their inhabitants.
In order to construct a holistic representation
that allows the contextualization of the vast amount
of urban datasets and layers there is a need to
incorporate the use of granular, socially produced
datasets and a direct interaction with the user. Cities
and neighborhoods ought to be understood be-
yond data; by conjoining traditional data with user
produced data, we hypothesize that it is possible
to embed interrelated spatial and non-spatial con-
ditions and construct personal representations of
place. Open-ended online-systems can turn urban
models into instruments that facilitate interaction
between designers and stakeholders. Current anal-
ysis tools and urban models are largely based on dig-
ital mapping; however, an attempt of urban legibility
through digital modeling can only present a single,
incomplete static representation of the world.
As an incomplete description of the city, urban
models aiming to predict urban processes such as
neighborhood ranking ought to be complemented
by alternative approaches. By incorporating crowd-
sourced datasets and user participation in the con-
struction of heterogeneous urban models, it is pos-
sible to contextualize urban information: exploring
and reframing urban information with different lev-
els of aggregation highlights outliers and serendip-
ity. In this project we explore bottom-up user driven
computation to give planners and users the capac-
ity to continuously reframe urban datasets and con-
ditions, discovering and creating new relationships
within the urban system, and allowing engagement
with different actors of city-making. To understand
the complex dynamics of cities, models should be
transparent and open to transformation, contextual-
ization and collaboration; rather than simulating an
average, or optimal urban condition, their assump-
tions should be tuned to the contextual and hetero-
geneous nuances of the city.
20 | eCAADe 34 - CITY MODELING | Urban Planning Approach - Volume 2
BACKGROUND
Our project is part of a collaborative effort to under-
stand how open data and social media can be lay-
ered above traditional municipal data sources in or-
der to provide a more fine-grained and individual-
ized understanding of neighborhood characteristics.
In the following sections, we provide an overview
of the qualitative and methodological research and
projects on top of which our research is built. The
relevant qualitative context are those projects that
address the topic of personalized neighborhood rec-
ommendations based on an extensive set of fea-
tures. Our methodological context consists primar-
ily of research that fine-tunes clustering algorithms
from both a spatial and lexical perspective.
Qualitative Context
To help us frame our research question and to bet-
ter understand the landscape of neighborhood char-
acterization, we first consulted popular tools and in-
terfaces that help users personalize their real estate
decisions. Through these resources, we gain a bet-
ter understanding what is currently available to users
and what is understood to be important for those
people such as home buyers and renters, who seek
to relocate.
Livability Calculator. The Livability Calculator devel-
oped by Nate Silver (Silver n.d.) is a web application
that ranks New York City neighborhoods according
to urban dimensions like affordability, school acces-
sibility, housing quality and green space. It also pro-
vides a number of pre-defined profiles, like "Young,
Single, and Cash-Strapped", or "Double Income, No
Kids", that sets the score given to each of the livability
dimensions to specific values and provides a custom
ranking of the neighborhoods.
Walk Score. Walk Score is a static neighbor 'walkabil-
ity' assessment tool. It calculates walking distances to
nearby amenities, which indirectly assesses the prox-
imity to amenities, and decays these distances by
time to create a composite score. Additionally, Walk
Score aggregates scores within an entire neighbor-
hood in order to rank the 'walkability' of that neigh-
borhood. Walk Score is an earlier commercialized
method for assessing livability, but aided in generat-
ing interest in this field.
Trulia and Zillow. Increasingly, real estate database
companies such as Trulia and Zillow are offering ser-
vices which give evaluative powers to the user in the
decision-making process. These tools, such as Zil-
low's Zestimate or Trulia Estimates, are primarily used
for assessing the value of a building, but include such
features for surrounding schools, nearby home val-
ues, historical housing prices in the neighborhood,
and Walk Scores.
Methodological Context
A line of more academic research has focused on re-
fining methods to categorize neighborhoods and ex-
trapolating groups from those data that characterize
a neighborhood. The primary strategy is clustering
through different methods of unsupervised learning
of spatial and semantic data, often with a temporal
element.
Spatial Methods. The baseline spatial clustering
technique used in this field, on top of which most
other methodologies is created, is density-based spa-
tial clustering of activity with noise (Ester et al. 1996).
This algorithm groups points that are close together,
while marking as outliers those points that are in a
low point-density region. More recent neighborhood
description techniques build upon the idea of clus-
tering. We look at two main methods of spatial differ-
entiation: One is premised on the primacy of activity
mobility, and the other is premised on homogeneity
of points-of-interest.
CitySense (Loecher and Jebara 2008) is a discov-
ery tool for temporal and spatial hot spots of activ-
ity in the city that has been implemented as a mo-
bile application. The CitySense algorithm, in its first
iteration, was a clustering algorithm based on GPS
data from different sources such as taxis and peo-
ple. The model was intended to not only create hot
spots, but also to match users who have similar ac-
tivities during similar points in the day in order to
create predictions on future behavior or preferences.
CITY MODELING | Urban Planning Approach - Volume 2 - eCAADe 34 | 21
The Hoodsquare project (Zhang et al. 2013) de-
veloped an algorithm for extracting neighborhood
boundaries in cities, including New York City based
on social media data. The algorithm works by us-
ing data related to Foursquare venue types, spatial
distribution of local and tourists in the city. Their
neighborhood definitions goes beyond the neigh-
borhood as administrative or politically defined units
in order to unearth geographies that are much more
in accord with temporal activities in the city. The
Livehoods project (Cranshaw et al. 2012) combines
both points-of-interest clustering with the cluster-
ing of similar activity 'schedules'. It is based on the
development of an algorithm that uses Foursquare
data in order to produce neighborhoods classifica-
tions based on spatial and social proximity of venues
in New York City. The Livehood project has devel-
oped a compelling method to unearth a classification
of New York City neighborhoods by taking advantage
of massive data sets and unsupervised learning ap-
proaches. A significant result of this approach is the
Livelihood's definition of a neighborhood as an "an
urban area [...] defined not just by the type of places
found there, but also by the people that choose
to make that area part of their daily life"(Cranshaw
et al. 2012). Vaca, Quercia, Bonchi, and Fraternali
(2015) recently looked at a neighborhood description
technique that involves both clustering and labeling
simultaneously. Whereas previous techniques first
cluster nearby points on a map and then assign labels
to the points, Vaca et al. use a technique for hierar-
chical clustering that merges branches only when it
increases an objective function.
Semantic Methods. Unique in our methodology for
neighborhood definition is the usage both tradi-
tional geo-spatial data and user-generated social me-
dia data. Because much of the social media data
is both textual and spatial, we explored methods of
parsing tweets and tags that would allow us to spec-
ify neighborhood qualities. Urban dynamics as ex-
plained by geolocated social media data has been ex-
plored by the likes of Noulas et al. 2011, and Chang
et al. 2011 using geolocated data on categories or
 tags. Other papers such as Chang et al. 2012 and
Frias-Martinez and Frias-Martinez 2014, have looked
more closely at Twitter tweets texts and patterns.
METHODS
Discussed here are the structural components of our
ranking system for the 195 neighborhoods in New
York City. We present the practicalities of construct-
ing the ranking system, producing personalized user
results through machine learning techniques, and
representing the data for user interaction.
The project proposes the use of machine learn-
ing techniques to analyze aggregate datasets, and
predict regularities across regions, such as mood,
theme, or related trends. In particular, we make ex-
tensive use of natural language processing (NLP) (Blei
2012, Blei et al. 2003) to characterize and contextu-
alize geo-located social media datasets. The diverse
set of analysis techniques implemented and devel-
oped for the project result in a non-spatial represen-
tation of the city, where, in contrast to geospatial
analysis, which prioritizes Euclidian distances, rela-
tionships are discovered across spatially diverse loca-
tions through a sameness of social and environmen-
tal characteristics within the city. The development
of a publicly available web interface provides agency
to the user, allowing the analysis and construction
of the city to be made according to their individual
preferences, encouraging a continuous rediscovery
of the city.
EXPLORATORY ANALYSIS
In order to determine the differences of neigh-
borhood classifications when using conventional
datasets or the non-conventional datasets and
methodologies proposed by the project, we per-
formed an initial exploratory data analysis with con-
ventional GIS datasets. The initial exploratory data
analysis also served to uncover basic relationships in
the data that were later used for the development of
the composite neighborhood ranking. Among oth-
ers methods, we used unsupervised algorithms such
as Principal Component Analysis, K-Nearest Neigh-
22 | eCAADe 34 - CITY MODELING | Urban Planning Approach - Volume 2
bors, K-Means and hierarchical clustering algorithms,
to investigate possible classifications of the urban
and social data we had, as well as the neighborhoods
themselves. The initial study allowed us to identify
the dimensions or urban and social media data that
would be the most helpful for users to decide which
neighborhoods to live in.
GIS. Over more than month in November 2014, the
authors gathered social media and live datasets from
publicly available web APIs; datasets like tabular and
GIS files were acquired from state and municipal
open-source repositories (see Table 1).
We used New York City's Socrata Open Data por-
tal to understand the static, underlying character-
istics of the city. The portal contains a wide vari-
ety of municipal data including housing, transporta-
tion, business, and environmental data. Addition-
ally, much of the administrative and neighborhood
boundary was also from Socrata.
The 311 information includes all the complaints
that were reported to the city. The relevant fields
that were used for the final analysis are complaints re-
garding construction noise, dirty conditions, graffiti,
and derelict vehicles, as determined that these fac-
tors combined would result in a generalized indica-
tion of an undesirable place to live. From these data,
we were able to achieve a better understanding of
how these neighborhoods compare in terms of the
more traditional desirable housing qualities.
Hierarchical Clustering. We utilized K-Nearest
Neighbors and K-Means, unsupervised learning al-
gorithms that iteratively learn and classify datasets
based on a number of predefined clusters. Validat-
ing and calibrating the number of predefined clus-
ters proved to be "sticky", as the traditional datasets
could not capture the diversity of the actual neigh-
borhoods. We additionally implemented a hierarchi-
cal clustering algorithm, a bottom-up greedy search
algorithm that pairs similar clusters together, moving
up on the cluster hierarchy (see Figure 2). Beyond the
broad clustering performed through the PCA analy-
sis, the hierarchical clustering allowed us to provide
an initial classification of the neighborhoods. The
Figure 1
Exploratory data
analysis of urban
data of NYC
neighborhoods.
Zoom (left).

Figure 2
Unsupervised
Clustering of
Neighborhoods
(Zoom).

CITY MODELING | Urban Planning Approach - Volume 2 - eCAADe 34 | 23

 hierarchical clustering algorithm suggested that the
optimal number of clusters for the 195 New York City
neighborhoods was between 3 and 7 clusters.
The results of these classification methods were
used to develop the overall weights for the compos-
ite ranking that will be described in the next sections,
suggesting which data dimensions were the most
meaningful to urban dynamics. We identified which
of the datasets (i.e., safety, urban density, social) pro-
vided useful dimensions for users to rank (in terms of
how much relevance they placed in them) and define
distinguishable groups of neighborhoods.
Semantic Analysis
The project's methodology proposes the develop-
ment of a personalized ranking through the use
of non-traditional and dynamic datasets to un-
derstand different perspectives of urban livability.
Our methodology uses machine-learning techniques
combining supervised and unsupervised classifiers
and clustering algorithms to make sense out of such
disaggregated datasets. At the same time, such ma-
chine learning techniques are used to construct spa-
tial measures of wellness that are hard to capture
with standardized social measurements.
Semantic analysis is used to generalize the syn-
tactic structures of text enabling the creation of re-
lationships among large amount of social data ob-
tained through web APIs. Specifically, we imple-
mented a Naïve Bayes algorithm for sentiment anal-
ysis classification and a Latent Dirichlet Allocation
(LDA) clustering algorithm for parent-child classifica-
tion of the social media datasets. The results of the
machine learning algorithms were then combined
with the rest of the datasets utilized in the project
to inquire about the implications of social dynamics
in the interpretation and construction of urban form
and city dynamics.
Furthermore, the classification algorithms allow
users to develop subjective and personal choices
about the way they differentiate neighborhood con-
ditions. By interacting with the web interface they
can refine their queries and neighborhood classifica-
24 | eCAADe 34 - CITY MODELING | Urban Planning Approach - Volume 2
tions based on the real-time feedback obtained from
the system.
Supervised Learning. The platform uses a senti-
ment analysis algorithm, to determine the overall
contextual polarity of a textual dataset. We imple-
mented a Naive Bayes Classifier (Norvig 2003), a prob-
abilistic classifier that was previously trained with
a corpus of over 2,000 document rating classifica-
tions. While Naïve Bayes classifiers have been em-
ployed since the 1960s, their use on social media
data is inherently problematic -the inconsistencies
and highly contextual and specific lexicon difficulties
the use of popular out-of-the-box classifiers. To deal
with such textual inconsistencies, the media posts
were cleaned and parsed -we eliminated non-English
words, since our classifier was trained with an En-
glish word corpus words not in the English dictio-
nary, and we eliminated excessive punctuation. Al-
though other tests involved the analysis of hashtags,
ultimately, only English words in the tweet were an-
alyzed in order to be consistent with the training set
used for the algorithm.
After training the classifier, we evaluate every
data point with the classification algorithm, which re-
turns a normalized polarity value for each social me-
dia data point that is then aggregated spatially by
neighborhood. The aggregated values are then used
as part of the composite neighborhood ranking that
will be introduced in the next section. Additionally,
through the web interface, users are able to relate dif-
ferent neighborhood sentiment thresholds to a num-
ber of social activities and physical characteristics of
the space.
Unsupervised Learning. We used topic modeling
to discover the abstract "topics" that occur in a col-
lection of social media posts -essentially a collection
of textual data points are classified into k-number
of topics that share similar topics in their content.
Through topic modeling a set of text data points are
evaluated; based on the statistical similarities of the
words in each text data point their topics are eval-
uated and classified. We used Latent Dirichlet Allo-
cation (LDA), an unsupervised learning algorithm -
Table 1
GIS Sources.

Figure 3
Tweets
Classification.

CITY MODELING | Urban Planning Approach - Volume 2 - eCAADe 34 | 25

 the most common topic modeling method (Blei et al.
2003). The algorithm iteratively divides a dataset of
text data points into a number of k-subgroups with
common topics.
Once the entire dataset of social media data
points was classified, the most common words of
each subgroup and their overall topics were assigned
to each individual data point within the group. Ev-
ery data point in the dataset was based on hashtags
of tweets; this attempted to reduce the noise in the
messages due to the wide use of emoticons, and non-
English words in longer tweet texts. The advantage
of such classification algorithms over a Naïve Bayes
classifies is that since it does not require a pre-trained
dataset, non-traditional textual elements such as so-
cial media hashtags can be iteratively classified ac-
cording to their statistical similarities to the rest of the
set.
Once the LDA model was trained, the tweets
were divided in k-number of groups. Essentially, each
group is classified based on how related the hash-
tags are to each other. Some of the topics that were
extracted from the subgroups of the dataset were:
trendy, foodie, nightlife, public space, music, social
justice, and fitness. The individual topics of each data
point were aggregated within the NYC neighbor-
hoods, and then used to quantify the occurrences of
different topics within neighborhoods -every neigh-
borhood got an aggregated count of different emer-
gent topics extracted from the conversations hap-
pening in the area through social media. By interact-
ing with the user through the web interface, the users
can discover the relationships among specific topics
within a given neighborhood.
Composite Neighborhood Ranking
In our ranking, we used conventional urban data, like
median home values or access to transit and safety
measures, and also semantic and spatial social me-
dia data to gain insight on the mood and social life
within New York City neighborhoods. In order to do
so, our team developed a composite weighted score
with urban and social media data to rank neighbor-
26 | eCAADe 34 - CITY MODELING | Urban Planning Approach - Volume 2
hoods.
Given the diversity of users and preferences, the
built-in request in our web app for users input is a
way to give them control when defining what mat-
ters most to them, as opposed to providing a unique
ranking system for all of them. The end product for
our user is a ranking of all the 195 neighborhoods
built according to the categories the user cares the
most, and that will help make a better decision on
which New York City neighborhood to live in.
Components of the Ranking. The five dimensions
of the composite score are: social capital, affordabil-
ity, urban density, safety, and the mood or sentiment
within a neighborhood. Each of these dimensions is
a simple average of its components variables, except
for the safety score in which a much higher weight
was assigned to the count of shootings (weighted by
population) in order to reflect the severity of shoot-
ing events in comparison to other crimes that might
be less relevant to deciding where to live (i.e. occur-
rence of graffiti in streets). We realize that our dimen-
sions may have some challenges.
We used average SAT score in each neighbor-
hood as a proxy for education quality, average mean
household income as a proxy of resourcefulness, and
average percentage poverty (in such a way that a
higher average percentage of poverty per neighbor-
hood will decrease the social capital dimension, as
explained in the previous section) These levels are
then averaged to create the social capital compo-
nent.
In the case of the affordability dimension, we in-
corporated data related to the average median home
value, average median gross rental prices, and the
rent price percentage increase, and we include them
as indicators more expensive places to live in. This
does not reflect, however, that higher prices might
reflect simply better value for each dollar spent in the
case of neighborhood with a high level of amenities
and relatively low prices.
For the urban density dimension, we used the
average total number of units, the average floor per
area ratio (FAR), as well as other FARs (for example,
commercial, residential, retail) to reflect neighbor-
hoods with more density. We also included the av-
erage percentage of occupied (as opposed to vacant)
house units and the weighted number of subway sta-
tions. In this dimensions, we have assigned most of
the weight (65%) to the average FAR as we take it to
be the single most important indicator of how dense
a neighborhood is.
For the safety score we used counts of shootings,
graffiti, dirty conditions events, and derelict vehicles
reports, all of which came from the city's 311 data.
Our safety measure was an average of these cate-
gories.
Finally, for the mood dimension, we use the
sentiment analysis data extracted from our tweets
geocoded by neighborhood (the proportion of posi-
tive tweets out of the total tweets in a given location).
Front-end Web Interface
In this section, we present the technical specifica-
tions and properties of the front-end web inter-
face developed to enable the interaction with the
computed neighborhood values, the construction
of neighborhood choices, and the visualization of
the values related to the specific combination of
weighted values. The front-end is a D3 app that lever-
ages D3's data binding to synchronize and compute
the data displayed and edited by the UI with specific
data on the computed matrices. The graphical com-
ponent of the web interface is a choropleth map of
NYC that is constructed from a geoJSON of NYC.
The UI is built with a series of sliders that al-
low the modification and construction of weighted
queries that modify the neighborhood rankings on
the fly according to the user choices; the slid-
ers allow weight the different traditional and non-
conventional neighborhood values that have been
pre-computed through the composite neighbor-
hood ranking. Additionally, a number of text boxes
allow the user to input alternative characteristics of
a neighborhood that can be queried from the social
media datasets on the fly, adding additional weights
to the composite neighborhood ranking. The color
CITY MODELING | Urban Planning Approach - Volume 2 - eCAADe 34 | 27
values of the map are generated dynamically by para-
metrically blending the personalized neighborhood
rankings between two colors.
DISCUSSION
Our research straddles the traditional realm of real
estate research, which is rooted in investigating the
relationship between housing values and income,
race, age, proximities to amenities, and transporta-
tion, and the more dynamic, interactive realm of so-
cial media. An example of this type of research is
Nate Silver's Livability Calculator; it uses these more
traditional characteristics to determine livability, but
allows users to prioritize their own needs and desir-
able living characteristics. This calculator, however,
misses the voice of the neighborhood's residents.
One of the main promises of social media research
is that it gives us a sense of vivid cultural pulse. In
this way, our project aims to differ from the tradi-
tional real estate tools in that it incorporates the more
ephemeral social elements that constitute a neigh-
borhood.
Final Composite Score Results
In order to generate the final score for each of the
195 New York City neighborhoods and retrieve an
ordered list of them for the user, we calculated a
weighted composite score of the five dimensions as
described above (social capital, affordability, urban
density, safety and mood or sentiment). The way the
application builds these scores is by asking the user
to rank all dimensions from 0 to 5. These weights are
calculated by adding the 5 ranking values the user
has given for each dimension and dividing each one
of them by the total. For example, if a user has given a
score of 3, 5, 5, 4 and 4 to the social capital, affordabil-
ity, urban density, safety and sentiment score, their
weights would be 15% (3/20), 25% (5/20), and 20%
(4/20), respectively. Finally, we multiply each dimen-
sion by its score and calculate and overall score that
will rank all 195 neighborhoods in New York City.
We found that, when more traditional measures
such as social capital, safety, affordability, and den-
 Figure 4
Changes in
neighborhood
rankings based on
user-adjusted
preferences.

28 | eCAADe 34 - CITY MODELING | Urban Planning Approach - Volume 2

sity are given preference, our results align fairly
closely to those that Zillow or Trulia would produce.
For instance, preference for high density and afford-
ability above the other measures creates suggestions
for neighborhoods in upper Manhattan, while pref-
erence for social capital produces neighborhoods on
Upper East and West sides on Manhattan. When
we prioritize mood, however, our results become
more specific to particular neighborhood eccentrici-
ties that help shape the quality of those places. For in-
stance, our 'night life' topic brought neighborhoods
such as the Lower East Side higher in rankings, while
the 'public space' highlighted those neighborhoods
with easy access to Prospect Park and Central Park.
Contributions
In this project, we take the notion of crowd-sourcing
qualitative information, following the model of ser-
vices like Yelp or Foursquare, and apply them to
decision-making about environmental preferences
regarding urban spaces. The result is a tool that im-
proves upon current ones like Trulia or Airbnb's rec-
ommendation engine in two ways: the first is that it
gives agency and primacy to the decision making-
abilities of the public which reflects the more sub-
jective aspects of urban planning; the second is that
gives the user a wider set of tools in their decision
making process and incorporates more aspects of
neighborhood description, creating a scientific met-
ric for previously unquantifiable qualities.
Another important concept in our project, which
should be highlighted, is the capacity for discovery:
due to the non-spatial nature of certain types of activ-
ities, sub-communities form around certain lifestyle
and recreational interests, which can exist across dif-
ferent neighborhoods. Our tool follows in the lineage
of Tony Jebara's work on activity-based tribes - we
create a real-world implementation of the concept
that our choice of location can be determined by the
social activities we choose to be involved with. For
instance, a devoted yogi who wants to live close to
an advanced yoga center in addition to being located
near public transit and a holistic foods store, can
CITY MODELING | Urban Planning Approach - Volume 2 - eCAADe 34 | 29
choose between Tribeca, Greenpoint, Sunset Park,
amongst other neighborhoods across the city. In this
way, the yogi using the tool can discover new areas
of the city that satisfy his social or recreational needs.
More importantly, the project is a first step to-
wards the creation of urban models that through
a transparent construction and exploration of ur-
ban meaning can minimize the power structures be-
tween stakeholders and citizens. Urban tools that
promote civic engagement contribute to a collabo-
rative construction of cities.
Future Work
Future work involves testing and refining the tool
through feedback data. Currently, we have a 'test set'
in the sense that our tool is not refined through cal-
ibration with user feedback. Tools in the real estate
realm are beginning to incorporate a more detailed
description of neighborhoods to give a unique feel.
For instance, Airbnb has begun to create neighbor-
hood profiles for some of their host cities which in-
clude labels that summarize the character of the area
as well as qualitative images and descriptions. We en-
vision our tool could directly connect to sites such as
Airbnb, Craigslist, or Zillow and provide a more scien-
tific and qualified version of Airbnb's neighborhood
recommendation tool. This would give users who
start on our site the ability to find places to stay in
the neighborhoods recommended from our tool and
provide us data on the successfulness of engine.
REFERENCES
Blei, D. 2012, 'Probabilistic Topic Models', ACM, 55, pp.
77-84
Blei, D., Ng, A. and Jordan, M. 2003, 'Latent dirichlet allo-
cation', J. Mach. Learn, 3, pp. 993-1022
Chang, J and Sun, E 2011 'Location3: How Users Share
and Respond to Location-Based Data on Social', IC-
SMW
Chrisman, N 2006, Charting the Unknown: How computer
mapping at Harvard became GIS, ESRI Press
Cranshaw, J, Schwartz, R, Hong, J and Sadeh, N 2012 'The
Livelihoods Project: Utilizing Social Media to Under-
stand the Dynamics of the City', Assoc. Adv. Artif. In-
tell
 Frias-Martinez, V and Frias-Martinez, E 2014, 'Spectral
Clustering for Sensing Urban Land Use Using Twit-
ter Activity', Engineering Applications of Artificial In-
telligence, 35, pp. 237-245
Jacobs, J 1961, The Death and Life of Great American Cities,
Vintage Books
Light, J 2003, From Warfare to Welfare: Defense Intellec-
tuals and Urban Problems in Cold War America, John
Hopkins University Press
Loecher, M and Jebara, T 2009 'CitySense: Multiscale
space time clustering of gps points and trajectories',
Proceedings of the Joint Statistical Meeting
Martin, E, Krieger, HP, Sander, J and Xu, X ND, 'A Density-
Based Algorithm for Discovering Clusters in LargeS-
patial Databases with Noise', Kdd, 96, pp. 226-231
McHarg, I 1971, Design with Nature, Natural History Press
Norvig, R 2003, Artificial Intelligence: A Modern Approach,
Prentice Hall
Noulas, A, Scellato, S, Mascolo, C and Pontil, M 2011
'Exploiting Semantic Annotations for Clustering Ge-
ographic Areas and Users in Location-Based Social
Networks', SMW Workshop at ICWSM
Townsend, A 2013, Smart cities: Big data, civic hackers,
and the quest for a new utopia, WW Norton & Com-
pany
Vaca, C, Quercia, D, Bonchi, F and Fraternali, P 2015
'Taxonomy-Based Discovery and Annotation of
Functional Areas in the City', AAAI Conference on
Web and Social Media
Zhang, A, Noulas, A, Scellato, S and Mascolo, C 2013,
'Modeling and Recomending Neighborhoods in
Location-based Social Networks', Hum J, 2, pp. 1-15

30 | eCAADe 34 - CITY MODELING | Urban Planning Approach - Volume 2

Computational Approach for the Assessment of Transit
Oriented Development Principles
A multivariate optimization method for urban planning

Fernando Lima1 , Nuno Montenegro2 , José Nuno Beirão3 , Jose Kos4

1
Universidade de Lisboa; Universidade Federal do Rio de Janeiro; Universidade
Federal de Juiz de Fora 2 Universidade de Lisboa 3 Universidade de Lisboa 4 Universidade
Federal de Santa Catarina; Universidade Federal do Rio de Janeiro
1
fernando.lima@arquitetura.ufjf.br 2 nunomontenegro@fa.ulisboa.pt
3
jnb@fa.ulisboa.pt 4 josekos@gmail.com

This paper presents a computational approach to provide dynamic assessment

and optimization of principles derived from Transit Oriented Development (TOD)
- an urban development model that advocates compact, walkable, and mixed-use
neighborhoods, centered around transport stations. In spite of being increasingly
promoted in several cities of the world, TOD lacks an approach that addresses
multivariate data for optimization of its principles. In this paper, we propose a
methodology backed by an algorithmic-parametric CAD environment, applied to
a neighborhood unit in a case study. The objective is the analysis and
improvement of TOD relevant and measurable features (transit accessibility,
walkability and diversity) in order to optimize neighborhoods' features. The
ultimate goal is to facilitate the management of solutions in TOD planning
processes, supported by a principle-index-tool approach triad.

Keywords: Transit Oriented Development, Multi-objective optimization,

Computational urban planning,

INTRODUCTION less social interactions). Transit Oriented Develop-

According to Global Fuel Economy Initiative [1], fuel
combustion in motor vehicles is responsible for up to
75 percent of urban air pollution, while World Health
Organization [2] associates outdoor air pollution with
3.7 million premature deaths in 2012 . Besides con-
tributing to carbon emissions, the adopted logic of
organizing cities is responsible for great inconve-
niences in contemporary urban centers (e. g. au-
tomobile dependence, fragmented spatial patterns,
ment (TOD) is an urban development model that
became one of the key planning paradigms aimed
at creating compact, walkable, mixed-use communi-
ties, centered around high quality transport systems.
TOD facilitates the creation of vibrant livable commu-
nities without depending on an automobile for mo-
bility, and it is being increasingly promoted in several
cities of the world as a sustainable policy (Calthorpe
1993; Vale 2015). Nevertheless, although there is no

CITY MODELING | Urban Planning Approach - Volume 2 - eCAADe 34 | 31

 universally accepted definition of TOD, it is often de-
scribed in a physical manner: an area of compact
mixed-use buildings served by a multimodal mobility
network within walkable distance (Cervero & Kock-
elman 1997; Calthorpe & Fulton 2001; Suzuki et al
2013; Vale 2015). In this logic, basic urban needs are
easily accessible without demanding automobiles or
spending large amounts of commuting time, with
the intention of supporting more autonomous and
sustainable neighborhoods.
TOD is characterized by a few main features:
proximity to transport stations and a functional re-
lationship with them, as well as compact, mixed-use
neighborhoods that encourage walking, cycling, and
the use of public transit by residents, employees,
shoppers and visitors. In other words, it should en-
courage the use of public transportation by creating
neighborhoods with compact urbanization, diversity
in land use and urban design geared to the pedes-
trian - where one can walk to the stations and other
amenities. The primary principles of TOD consist of:
(i) transit accessibility - locating amenities, employ-
ment, retail shops and housing around transit hubs;
(ii) walkability - the ability that a particular neighbor-
hood has to connect housing and amenities points
through distances that can be traveled on foot; (iii) di-
versity - providing a mix of uses, densities and hous-
ing types in the same district; (iv) density for mass
transit - encouraging infill and redevelopment along
transit corridors within existing neighborhoods, al-
lowing the system to run efficiently (Cervero & Kock-
elman 1997; Calthorpe & Fulton 2001; Dittmar & Oh-
land 2004; Suzuki et al 2013).
According to Dittmar et al (2004), there have
not been sufficiently developed standards or systems
to help the actors involved in the process of bring-
ing successful TOD projects into existence. On the
other hand, TOD is a multi-variable dependent sys-
tem, as long as it relies on measurable parameters
for improved performance. It also considers geomet-
ric principles and objectve features for designing au-
tonomous neighborhoods. In other words, it repre-
sents a kind of complex proposition, derived from
32 | eCAADe 34 - CITY MODELING | Urban Planning Approach - Volume 2
some variables considered as crucial in achieving its
objectives, what makes TOD a potential case for com-
putational implementation (Lima et al 2016).
This study advocates a computational urban
planning approach for evaluation and optimization
of TOD features supported by a triad principle-index-
tool. That is, for each addressed TOD principle, there
is a corresponding index and a calculation algorithm
that makes it possible to objectively quantify and,
therefore, optimize alternatives for urban configu-
rations for a TOD oriented area. In this sense, this
approach proposes the application of the following
tools: (i) Station Proximity Calculator (SPC) - a com-
putational tool that measures the Station Proximity
index (SPI) of one or more locations. SPI is an indica-
tor that calculates the smaller physical path between
a transport station and one (or all) plot(s) in a neigh-
borhood, aiming to assess the Transit accessibility of
a particular plot or whole district; (ii) Walk Index Cal-
culator (WIC) - an algorithm that makes it possible to
measure the Walk Index (WI), an indicator intended
to assess the walkability of a particular location using
three parameters related to physical distances: prox-
imity, diversity and variety; (iii) Mixed-use index cal-
culator (MXIC) - a tool for measuring the Mixed-use
index (MXI) of an area. Proposed by Hoek (2008), this
indicator computes the ratio between residential and
non-residential areas in a location, in order to analyze
the diversity of a neighborhood. These three calcu-
lators (SPC, WIC and MXIC) are intended to support
planning strategies as long as they dynamically as-
sess their respective indexes (SPI, WI and MXI), which,
in turn, are intended to measure their respective TOD
principles (transit accessibility, walkability and diver-
sity).
In this context, this paper presents an approach
that advocates the application of the aforemen-
tioned calculators in optimization tasks that seeks
to improve: (a) the positioning of a transport sta-
tion, aiming at improved transit accessibility (smaller
physical distances to transport nodes); (b) the loca-
tion/distribution of amenities, intending to minimize
commuting distances and provide greater walkabil-
ity, and; (c) the balance of living and working places,
aiming at a greater diversity. Besides, we propose the
assessment of Spacematrix density indicators, pro-
posed by Pont & Haupt (2010), for supporting the
goals definitions and the decision making process.
METHODOLOGICAL APPROACH
The presented approach searches for a set of tools
to assist on TOD oriented planning processes, and
is not intended to act as an independent-automatic
solver. It is also not intended here, to develop a sys-
tem that includes all the variables involved in the var-
ious aspects of TOD, but to develop a set of com-
putational tools that supports decision making pro-
cesses, providing instruments that manage data and
perform complex calculations. The tools themselves
do not provide a better city, but are intended to al-
low the crossing of relevant information, providing
decisions supported by data obtained in a more ef-
ficient way than traditional means. The role of the
actors involved in urban development processes re-
mains central, and it´s still them that will stipulate
goals, feed the system and consider subjective and
"non-programmable" aspects, such as economic and
social dynamics, among others.
Geometric entities were associated with urban
elements, in the construction of the parametric
model, according to the following criteria: (a) points
represented the location of buildings and referred to
different urban functions (e. g. living, educational, re-
tail, food, recreational); (b) curves simulated the exist-
ing network of streets, and; (c) polygons were used to
play the role of buildings. Those entities are intended
to cast as input for the aforementioned tools and re-
sources, explained in more detail in sequence:
Station Proximity Calculator (SPC)
This tool is directly related with Transit Accessibility
issues, as long as it intends to measure the distance
between a transport station and one (or all) plot(s) in
a neighborhood. In this regard, the proposed algo-
rithm calculates the path(s) with smaller physical dis-
tance(s) between a station and one (or all) destina-
CITY MODELING | Urban Planning Approach - Volume 2 - eCAADe 34 | 33
tion(s) in a district, considering slope(s) in the path(s).
The score decreases as the distance approaches 1.6
km (20 min walk) and a 0 index is awarded for dis-
tances greater than 1.6 km (see table 1). The algo-
rithm also applies a penalty factor (according to the
acclivity) for estimating a score to classify the proxim-
ity to the station. For instance, in relation to a partic-
ular locality, if a station is 0.4 km away (5 min walk),
then the maximum index (1) is assigned for this plot.
If the station is a 1 km away from a particular plot,
but with a 10% acclivity, then a 10% penalty is ap-
plied. It establishes a simple relationship to differen-
tiate paths accordingly to their acclivity, but as long
as it works on a parametric environment, it is possible
to adopt other criteria for penalization.
Walk Index Calculator (WIC)
This algorithm works in a similar way of Station Prox-
imity Calculator. However, it is intended to esti-
mate objective features of walkability from a loca-
tion. It adapts some criteria from Walkscore Index
[3], for measuring three features (proximity, diversity
and variety) from seven categories of amenities (ed-
ucational, commercial, food, recreational, entertain-
ment, health and services) in order to assign three
complementary walk indexes to a particular location.
Proximity Index (PI) considers the smallest path (dis-
tance and declivity) for the nearest amenity of a cate-
gory, and uses the same criteria of SPC for assigning
an index to a place. Diversity Index (DI) is intended to
measure the average score from all amenities in a cat-
egory near to this same place, while Variety Index (VI)
expresses the number of amenities within a radius of
1,6 km away from one (or all) plot(s) in a category. In
this sense, while PI gives us a hint of how close one
location is to a particular amenity, DI is important to
evaluate its average distance from all amenities, and
VI gives an overall scenario of how many amenities
are connected to a place by walking. These three fea-
tures are meant to be analyzed together, in order to
get a clearer comprehension of the walkability from
a given plot or a whole area.
In this sense, WIC gives three partial indexes from
 seven categories. Each category has the same weight
for estimating PI, DI and VI global indexes - points are
summed and normalized to produce PI (from 0-1) and
DI (from 0-1) scores. VI depends on the number of
amenities of a district. Therefore, the greater these
indexes are, the better the walkability of the analyzed
area.
Mixed-use Calculator (MXIC)
This algorithm seeks to objectively measure a district
diversity. It is based on the Mixed-use index (MXI)
advocated by Hoek (2008), that calculates the sum
of all residential and non-residential areas in a loca-
tion, performing a comparison of their ratio in a given
neighborhood, as demonstrated in Table 2. In this
context, this tool makes it possible to quantify the
distribution of residential and non-residential build-
ings. The closer the relationship between them is to
50/50, the better diversity a district has.
Spacematrix Indicators Calculator (SIC)
This set of algorithms is intended to calculate the
Spacematrix density indicators proposed by Pont &
Haupt (2010). Spacematrix concept advocates a mul-
tivariate density measuring approach that consists
of three fundamental indicators: intensity (FSI), cov-
erage (GSI) and network density (N). FSI reflects the
building intensity, GSI demonstrates the relationship
between built and non-built space, and N refers to
the concentration of networks in a fabric. Spacema-
34 | eCAADe 34 - CITY MODELING | Urban Planning Approach - Volume 2
trix Indicators Calculators (SIC) evaluate density fea-
tures of an urban area and are important to provide
assessment of density issues (in this particular case
on district aggregation) for a TOD oriented Neighbor-
hood.
Multi-objective optimization
A Multi-Objective Optimization manages a set of ob-
jective functions to be optimized (maximized or min-
imized). It is a supporting tool for multiple criteria
decision making, concerned with mathematical op-
timization of problems that involve more than one
objective to be simultaneously satisfied. Usually, in
multi-objective optimization problems, there is not
a single solution that simultaneously optimizes each
objective. In that case, the objective functions are
said to be conflicting, and there is a number of Pareto
optimal solutions. A solution is called Pareto optimal,
or non-dominated, if none of the objective functions
can be improved in value without decreasing some of
the other objective values. As long as there is no ad-
ditional subjective preference information, all Pareto
optimal solutions are considered equally good.
Multi-objective optimization can be applied in
many situations, where optimal decisions need to be
taken while facing trade-offs between two or more
conflicting objectives. In the context of proposed
methodology, multi-objective optimizations are es-
sential for achieving solutions involving conflicting
objectives related to the aforementioned indexes
and indicators.
Table 1
Reference values for
proximity (and
diversity). This
indicators also
apply for SPC.
Table 2
Meaning of the MXI
(Hoek 2008).
Figure 1 Approach workflow
Approach workflow. This approach relies on a logic of identification, eval-
uation and optimization of relevant TOD features,
aiming at an efficient support approach for TOD ori-
ented planning processes. Within this framework, we
create a Rhinoceros/Grasshopper parametric model
of the selected neighborhood to manage geometric
and measurable features related to TOD principles.
Octopus, a Grasshopper plug-in for applying evolu-
tionary principles to problem solving, was used in or-
der to search for many goals at once, as long as pro-
duces a range of optimized trade-off solutions be-
tween the extremes of each goal. Hence, the afore-
mentioned tools use district data, emulated by geo-
metric entities, as input for the formulation of a se-
quence of algorithmic operations, as follows: Firstly,
the measurement of Station Proximity Index, Walk
Index (proximity, diversity and variety), Mixed-use
index and Spacematrix indicators, for each one of
the plots (when applicable) or for the entire neigh-
borhood, identifying demands for intervention, ac-
cording to the standard indexes; Secondly, the im-
provement of station proximity index, given that op-
timization tasks allow one to identify which posi-
tioning for the station promotes the smaller average
distance from all plots, providing better Transit Ac-
cessibility; Thirdly, the increase of amenities proxim-
ity, diversity and variety indexes, as long as multi-
objective optimization enables one to identify where
new amenities should be inserted, simultaneously
considering these three indexes for each Walk Index
category, providing a larger Walk Index and, conse- THE CASE STUDY
quently, suggesting a greater walkability; Fourthly, The case study comprises the application of the pro-
the evaluation of different scenarios for vacant plots posed approach on the existing district of Cascat-
occupation, automatically calculating the ratio be- inha, in the city of Juiz de Fora, Brazil (see Figure
tween residential and non-residential places, in order 2). The main goal of this study is to evaluate the
to promote a more balanced Mixed-use index and proposed methodology potential to frame an area
a higher diversity for the neighborhood; and finally, within the scope of TOD. Despite of being a predom-
a dynamic assessment of Spacematrix indicators be- inantly residential neighborhood, the chosen dis-
fore and after interventions, with the aim of identi- trict has a great potential to be a more autonomous
fying density limitations or potentialities within TOD and sustainable neighborhood, presenting some fea-
scope. Figure 1 shows the approach workflow. tures that make it an ideal sample for evaluation
of the proposed methodology, such as: a suitable

CITY MODELING | Urban Planning Approach - Volume 2 - eCAADe 34 | 35

 Figure 2
Images from the 3D
model of the
neighborhood.

36 | eCAADe 34 - CITY MODELING | Urban Planning Approach - Volume 2

 extension for TOD implementation (approx. 1 km
diameter); relatively low density; no transport sta-
tion; topographical complexity; available areas for
new buildings; vicinity to important amenities such
as a park, a hospital, a university, and a shopping
center, among others; closeness to the city center,
and; a good placement in the urban network, link-
ing directly the city center to important city regions.
This is a key scenario for assessing our approach to-
wards more efficient TOD oriented planning, because
besides demonstrating typical issues of the sprawl-
ing city paradigm, a situation found in several cities
around the world, it also presents some important
features for evaluating the methodology implemen-
tation.
The aforementioned approach steps were per-
formed, considering the data of Cascatinha district.
In that sense, the following information was obtained
and transferred to the analysis model: the foot-
prints, number of levels, uses (residential and non-
residential) and topographical positioning of each
building inside the district, in order to measure dis-
tances, slopes, possible connecting paths, diversity
and density indicators; the location of each one of
the amenities on the neighborhood, according to
the aforementioned categories considered by Walk
Index Calculator, in order to measure proximity, di-
versity and variety of locations; the identification of
Figure 3 available areas for new buildings (vacant lots and
Station Proximity non-consolidated places), seeking to provide room
graphic. for new constructions and to modify the diversity of
the district; the topographical network of neighbor-
hood´s streets, with the view to consider slopes and
distances for measuring SPC and WIC operations; and
the design of blocks and lots, in order to provide den-
sity (Spacematrix) assessment. Thenceforth, the task
was to apply optimization and simulation operations,
aiming to increase transit acessibility, walkability and
diversity related indexes, by the following actions, re-
spectively: (a) look for the best location for insertion
of a station, as long as the district does not have one;
(b) insert only one amenity from each category, seek-
ing to increase Walk indexes with the lowest number
CITY MODELING | Urban Planning Approach - Volume 2 - eCAADe 34 | 37
of new services possible; (c) evaluate different occu-
pation strategies for vacant lots, considering Mixed-
use index and Spacematrix indicators to help analyz-
ing impacts from new buildings propositions.
RESULTS
The implementation of our approach provided some
changes to the arrangement for the neighborhood,
suggesting a better performance from the scope of
TOD principles. In summary, the new settings pro-
vided: (i) an excellent Transit Accessibility, since the
optimized insertion of a transport station permitted
a high average Station Proximity score (see Table 3
and Figure 3); (ii) a greater walkability, as long as the
optimized addition of new amenities provided an in-
crease of the neighborhoods proximity, diversity and
variety, in all analyzed categories, as shown in Table
3 and Figures 4, 5, 6; (iii) a neighborhood with more
diversity, since the proposition of new buildings and
their functions provided a more balanced Mixed-use
index, seeking a greater equilibrium between resi-
dential and non-residential places - see Table 3; (iv)
a more suitable density for a Transit Oriented Neigh-
borhood, since parametrically controlled urban ge-
ometries made possible to regulate density, in a man-
ner that supports more people (working or living)
closer to transport hubs, as shown in Table 3.
Proximity and diversity indexes (both global and par-
tial) increased after optimization tasks, suggesting
that amenities became more closer and more bal-
anced within the new district´s arrangement, as Fig-
ures 4, 5 and 6 shows.
 limitation of one station for it. Besides, the loca- Table 3
tions with worse Station Proximity scores were eas- General data of case
ily identifiable, and in this sense, this is a good start- study, before and
ing point for further actions for transit accessibility after optimization.
improvement (the adoption of secondary modals for
connecting these locations and the station, for exam-
ple).

Figure 4
Global Proximity
and Diversity
indexes before and
after optimization.

Figure 5
Partial Proximity
indexes before and
after optimization.

Figure 6
Partial Diversity
indexes before and
after optimization.

DISCUSSION AND CONCLUSIONS

The approach proved to be helpful in order to pro-
vide dynamic assessment and optimization of princi-
ples derived from TOD. The positioning of the station
was supported by optimization tasks that allowed to
identify, among hundreds of options, the solution Multi-objective optimization addressed simulta-
that provides the lowest average distance between neously multiple conflicting criteria for walk index
the station and all other lots. In this sense, the ob- answers, crossing data and performing calculations
tained station proximity index was as high as it could that would be more difficult to perform by tradi-
get, considering the district´s configuration and the tional means. While proximity indexes considered

38 | eCAADe 34 - CITY MODELING | Urban Planning Approach - Volume 2

Figure 7
Octopus interface
for multi-objective
optimization. In
yellow, the
pareto-optimal
solution adopted.
only the nearest amenities, diversity indexes consid-
ered all surrounding amenities, seeking a greater bal-
ance (see Figure 7). This trade-off context was a key
for providing greater proximity, diversity and vari-
ety indexes, meaning that services are nearer and in
larger quantities along the district, which suggests
more walkability. These three components of WIC
can play an important role in TOD planning tasks, as
long as they identify which services are more or less
available for a particular plot, street, block or even the
entire district. On the other hand, it also allows one
to see which areas of the district are better or worse
supported in relation to different categories.
Multi-objective optimization provides a set of so-
lutions that are intended to be considered equally
good (pareto-optimal solutions). It is an important
possibility for urban planning processes, as long as
it strengthens the planners' role in considering "non-
programmable" aspects, for stipulating subjective
criteria and priorities for decision making.
Despite of not being directly optimized, density indi-
CITY MODELING | Urban Planning Approach - Volume 2 - eCAADe 34 | 39
cators play an important role in this approach, as long
as they enable the visualization and evaluation of dif-
ferent scenarios for vacant areas occupation, guiding
interventions and giving hints from building poten-
tial and mix of uses distribution. The Mixed-use in-
dex and Spacematrix indicators proved to be useful
in the algorithmic implementation of this approach,
given that they consider objective features for mea-
suring diversity and density, respectively.
Limitations, further developments and final
remarks
Despite the usefulness of a computational tool for
supporting TOD oriented urban planning processes
through evaluation and optimization of specific ur-
ban features, we identify some limitations in the pre-
sented approach. First, we recognize that Walk In-
dex does not fully incorporate the diverse features
that can influence the walkability of an urban area.
Therefore, other variables that measure connectiv-
ity, density of streets network, and pedestrian friend-
 liness features should be tested. Second, it would
be important to identify which streets or stretches
have higher recurrence in the walking paths, in or-
der to identify the main pedestrian axes, which would
be the high-priority places to promote an increase
of pedestrian friendliness and mixed-use diversity.
Third, as shown in our research, density has a funda-
mental role within the scope of TOD. In this context,
variables related to populational density and the im-
plementation of modals in a neighborhood should
also be tested. Finally, considering the influence that
acclivity has in Walk Index calculations, it is impor-
tant to incorporate more specific guidelines in order
to give support to penalty factors calculation.
This article seeks to facilitate the management
of solutions in TOD planning processes, in order
to delineate a starting point for computational ap-
proaches towards more efficient TOD proposals.
Therefore, this study demonstrates the approach's
potential towards a more efficient TOD implementa-
tion methodology.
ISOCARP congress 2008
Lima, FT, Kos, JR and Paraízo, RC 2016, 'Algorith-
mic approach toward Transit-Oriented Develop-
ment neighborhoods: (Para)metric tools for evalu-
ating and proposing rapid transit-based districts', In-
ternational Journal of Architectural Computing, 14(2),
pp. 131-146
Pont, MB and Haupt, P 2010, Spacematrix: Space, Density
and Urban Form, NAi Publishers, Rotterdam
Suzuki, H, Cervero, R and Iuchi, K 2013, Transforming
cities with transit: Transit and land-use integration for
sustainable urban development, Word Bank, Wash-
ington
Vale, DS 2015, 'Transit-oriented development, integra-
tion of land use and transport, and pedestrian acces-
sibility: Combining node-place model with pedes-
trian shed ratio to evaluate and classify station areas
in Lisbon', Journal of Transport Geography, 45, pp. 70-
80
[1] http://www.globalfueleconomy.org/
[2] http://www.who.int/mediacentre/news/releases/20
14/air-pollution/en/
[3] https://www.walkscore.com/

ACKNOWLEDGEMENTS
The authors would like to thank Architecture, Ur-
banism and Design Investigation Center (CIAUD)
from University of Lisbon and the Brazilian National
Council for Scientific and Technological Develop-
ment (CNPq) for supporting this research and the pre-
sentation of this work.

REFERENCES
Calthorpe, P 1993, The Next American Metropolis: Ecology,
Community and the American Dream., Princeton Ar-
chitectural Press, New York
Calthorpe, P and Fulton, W 2001, The regional city: plan-
ning for the end of sprawl, Island Press, Washington
Cervero, R and Kockelman, K 1997, 'Travel demand and
the 3Ds: density, diversity, and design', Transporta-
tion Research Part D: Transport and Environment, 2,
pp. 199-219
Dittmar, H and Ohland, G (eds) 2004, The new transit
town: Best practices in transit-oriented development,
Island Press, Washington
Hoek, JVD 2008 'The MXI (Mixed use Index). An instru-
ment for anti sprawl policy?', Proceedings of the 44th

40 | eCAADe 34 - CITY MODELING | Urban Planning Approach - Volume 2

Agent-Based Urban Growth Simulation
A Case Study on Istanbul

Cemal Koray Bingöl1 , Birgül Çolakoğlu2

1
Yildiz Technical University 2 Istanbul Technical University
1
c.koraybingol@gmail.com 2 bigi@alum.mit.edu

This study aims to create a simulation model for urban growth with agent-based
modeling. The model is based on the theoretical research of Michael Batty on
urban growth simulations. The study explains how the theoretical approach
applied in the model with the parameters. The model in this study is created in an
open-source API called 'Processing' and the simulations executed through the
parameters in the study. The results of the simulation are compared with each
other to find optimal parameters fits in the theoretical approach. Parameters are
tested on an existing urban settlement map, which Is Istanbul. The results of
Istanbul simulation are compared with existing density and urban sprawl maps of
Istanbul and discussed for further studies.

Keywords: agent-based design, urban growth, urban simulation

INTRODUCTION beginning of 70's, cities started to be seen as com-

Cities have become one of the major fields to be ex-
amined and studied in the 20th century. The popu-
lation increase and migration to urban areas from ru-
ral parts have triggered an enormous urban growth.
This rapid growth of urban development showed
that urban design theories are not capable of explain-
ing the dynamics of cities. The developments of cities
were being planned as they were simple and stable
forms of a city. The dynamic relations of cities have
not been perceived as a complex system as it is (Batty
1994).
The complexity of cities brought up to the sur-
face in 60's by Jane Jacobs in her book 'The Death and
Life of Great American Cities' (Jacobs 1961). However,
by that time the complexity of the city was not a com-
putational design subject, Jacobs implemented that
cities have an entwined structures of relations. At the
plex systems and the developments in computer sci-
ences showed that the complex systems in many sci-
ence fields, from biology to physics, can be modeled
through digital computers. Furthermore, cities have
become a research subject for computational mod-
els.
This study aims to create a model that simulates
the complexity of an urban growth regarding the dy-
namics of a city. The model produced in this study
is based upon the theoretical structure of Michael
Batty's research explained in his book 'Cities and
Complexity'. Therefore, design based mathematical
algorithms applied in this research are referred to
Batty's theoretical research.

CITY MODELING | Urban Planning Approach - Volume 2 - eCAADe 34 | 41

 COMPUTATIONAL MODEL
Computational methods for urban growth have been
researched in this study. Social relations we en-
counter in a daily life can be a small example of inter-
action, but scaling these interactions up to a city and
inspecting these interactions at urban scale has enor-
mous numbers of inputs to calculate. The compu-
tational method called Agent-Based Modeling (ABM)
is what enabled researchers to model highly compli-
cated interactions in urban scale by utilizing comput-
ers. ABM changed the way computers can model
complex systems (Gilbert 2008). The development
of 'Object-Based Modeling' method started with Cel-
lular Automata (CA) (Batty 2005). CA brought up a
modeling method that works as a bottom-up system.
The method focused on interactions of the individu-
als with their neighbors and the patterns they form
by these interactions. The model used in this study is
based upon these interactions of 'agents' with them-
selves and the environment they have been mod-
eled. As Batty argues that cities do not have hierar-
chical structures, but network structures. So, ABM is
the method which Batty finds most suitable to model
cities.
Batty defines the urban growth model with five
drivers; Randomness, historical accidents, physical
determinism, natural advantage and comparative
advantage (Batty 2005). The first of the drivers, ran-
domness, as Batty points out, which creates hetero-
geneous urban forms by a variety of entities within
itself to avoid a homogeneous form. As DeLanda
agrees, to simulate an urban model, it needs to com-
pute the decision making, where he defines an artifi-
cial intelligence model inherited to the model. How-
ever, it is almost impossible to create thousands of in-
dividuals with artificial intelligence (DeLanda 2009).
So, a value of randomness in the system may help to
simplify the decision making into heterogeneity.
The second driver is the historical accidents,
which are the critical decisions made in history that
influenced the city's location of initial settlement.
Therefore, in the model, it is the driver which has no
effect due to the input's ambiguity. The third driver,
42 | eCAADe 34 - CITY MODELING | Urban Planning Approach - Volume 2
physical determinism, is the parameter which we de-
clare the use of land. Physical determinism is the
mechanism where we are defining which land cells
are developable for urban growth according to the
capabilities of the land's properties in the model. The
fourth driver, natural advantages, is the parameter
compares the resources at locations and attainabil-
ity to these resources by individuals. Batty defines
this driver as the resources that land provides. The
last one, comparative advantages, is the driver which
is defined as the measurement of the differences be-
tween the locational resources like services, facilities
and accessibility to these features.
Model's Setup
The model in this study aims to create a simulation
based upon the interactions of individual entities like
'agents' and 'land cells'. This bottom-up approach is
the key aspect how Batty defines the urban growth
models in his research. The model is utilized in "Pro-
cessing", an open source application processing in-
terface(API). The abstract mathematical fundamen-
tals of the model researched by Batty have been pro-
grammed in Java (a programming language used by
Processing.)
Therefore, the simulation is executed on a
200x200 2D cellular grid, where every land cell has lo-
cational information and properties. The canvas con-
tains information for geographical properties and lo-
cational information for the agents. In the model,
resolution of the canvas is limited with 200x200 grid.
However, the agents' movements are not limited to
the land resolution. Agents move on the canvas
within a higher resolution.
The elements of the model are shown in Figure
1, where agents and land cells interact with each
other to create a bottom-up feedback mechanism.
The land cells contain the information about whether
they are developable or not, and if any trail has been
left by an agent, trail contains the location of the re-
source cell. The trail only contains the information of
a resource cell left by the agent, which found a re-
source, passed through. Agents start to leave trails as
Figure 1
Elements of the
model.
the information for the location of resource cell they
find. Land cells, besides the information, have states;
it can be a land with a potential value, a resource cell
or an agent can be attended to the cell as its home
location. In Figure 2, it is shown how new entities are
adding up to the simulation at every iteration. At ev-
ery iteration, one new resource and three agents are
being added to the simulation.
Figure 2
Population and
Resource Cell
mechanism.
The growth of the model depends on the interac-
tions of individuals and agents, in time. The new re-
CITY MODELING | Urban Planning Approach - Volume 2 - eCAADe 34 | 43
sources and the new populations of the model will
be spawned at the locations with the highest po-
tentials, where the potential is computed with the
amount of interactions of agents at certain locations,
or cells, and with an additional amount of random
potentials. The random potential value added to the
land cells are the simplification of natural advantages
cells have. These behaviors also will help the simula-
tion grow gradually according to the smallest inter-
actions in the model.
Model's Parameters
The model's mechanism depends on the interactions
of agents (individuals) and where they interact with
each other the most. Therefore, the model also cre-
ates a positive feedback effect on the most interac-
tive parts of the canvas. However, the model still
needs initial parameters to start the simulation. Ini-
tial population of agents (POP), initial number of re-
source (RES) land cells and the amount of resource
each resource cell has (FEED) are the initial parame-
ters of the model.
This model creates a positive feedback mecha-
nism that evaluates the urban growth gradually and
heterogeneously in time. Moreover, the heterogene-
ity is defined in the smallest scale of the model with
the interaction of agents in local neighborhoods,
 while these actions create different clusters of pop- creases as RES parameter increases, which points out
ulations, as cities, in the growth of disparate scales. that the clusters start to emerge not by the interac-
Clusters emerging through time in different scales tions but the resource cells' initial locations.
are deriving comparativeness between the clusters. It is important to conclude this, as the amount
Every cluster forms by the local interactions of indi- of resources in resource cells increase, their life span
viduals on locations. increases as well, so the initial setup defines how
In Figure 3, from (a) to (e), we may observe the the simulation evaluates. If RES parameter increases,
gradual growth on the canvas. In Figure 3(a), It is the emergent patterns formed by the simulation are
the initial setup of the simulation and where there becoming depended on the resources at the initial
are only white points. These white points are the ini- setup, but not the interactions of the agents.
tial resource cells. Following in Figure 3(b), there are
green, and magenta points are adding up to the sim-
ulation, where magenta colored cells are the land cell
appointed to agents as their home and green col-
Figure 3
ored cells are the potential values of the land cells.
Iterations of the
In Figure 3 the iterations of the simulations through
model.
time can be seen, but it is important to see the layers
of the model separately. Here in Figure 3, it shows
the overlay of values on cells which are potential val-
ues, resource cells, trails and homes of agents. In Fig-
ure 4 only the potential values of the land cells are
shown. When we compare this with Figure 5, where
only the trails and homes of agents shown, the trails
Figure 4
and homes of agents start to overlap at near dense
Iterations of the
clusters.
potential values of
In Figure 6, different from Figure 4 and 5, only re-
land cells.
source cells are shown. Here, we may see that only
the centers of clusters become resourceful. The rea-
son of this density is the amount of interactions in the
centers of clusters. Therefore, it emerges with the ef-
fect of individuals, agents, in the model. One param-
eter that examined in the model is durations of the re-
Figure 5
sources, Shown in Figure 7. This parameter questions
Iterations of the
the debate Batty mentions. Batty states that even the
Trails and Homes of
initial resources are located randomly, the city devel-
agents.
ops by the interactions and the locations where in-
teractions are the most (Batty 2005). So, by chang-
ing the value for the amount of resources in each re-
source cell, it is possible to examine if the resource
cells affect the structure of the model. In Figure 7, the
amount of resource in a resource cell (RES) increases
from Figure 7(a) to Figure 7(d). The increase is shown
in Figure 7, also shows that the number of clusters in-

44 | eCAADe 34 - CITY MODELING | Urban Planning Approach - Volume 2

 by two bridges between two continents. Although
Istanbul has many topographical properties, due to
Figure 6 the parameters we discuss in this study, the model
Iterations of applied on Istanbul map only considers the availabil-
resource land cells. ity of land for development.
At Figure 7, there are two map images, where Figure
8 (a) is the map of Istanbul's district centers (Hart-
mann 2010)[1], and Figure 8 (b) re-mapped image,
which adjusted to be recognized by the software,
which reads the colors and assigns land uses at the
initial setup of the simulation. White colored areas
Figure 7 are developable lands, blue colored areas are seas
Comparison of the and lakes, as undevelopable lands, and black dots
amount of resource are the district municipalities of Istanbul, where re-
each resource cell source cells are located. However, in the beginning,
has. the model do not run the function to read the black
dots to locate the resource cells on the map image,
the model assigns resource cells on the landscape at
random locations. These two different approaches to
locate resource cells provide whether if the location
of initial resources affect the emergent patterns that
simulation creates.
At Figure 9, there are four simula-
tions ran through same parameters, which
RES10/FEED10/POP3000, at 3000 iterations. We can
see some similarities over these four simulations,
where all the resources located randomly at the ini-
tial setup. These similarities, besides their emergent
structures, have the potential to produce results. As
Figure 8
shown in Figure 10, these four simulations are su-
District centers map
perposed over each other to take a closer look what
(Hartmann
are the similarities between simulations with same
2010)[1](a),
parameters. It can be seen that few areas have an
Adjusted map
intense development on canvas, where simulation
image (b).
ran. We will not consider every developed cluster in
these four simulations, but multiply the overlapping
developments to get an averaged result.
Even though, the developments over each canvas
CASE STUDY: ISTANBUL
The parameters discussed here were tested to find specifically do not match each other, the fact that
the right set of parameters to simulate the model on these simulations do not acquire any specific data of
the map of Istanbul. Istanbul is highly dense urban where the initial locations for resources are. So, we
settlement, due to its topography and its polycentric have to consider them as emergent systems with ex-
structure. The city is separated by sea and connected actly same parameters.

CITY MODELING | Urban Planning Approach - Volume 2 - eCAADe 34 | 45


Furthermore, in Figure 11, it is shown that the same
simulations ran through the same parameters as Fig-
ure 9 but, this time not only the parameters kept
same, but also the locations for initial resources given
to the simulation are same as well. In Figure 11, it
is possible to see almost the same kind of cluster-
ing as in Figure 9. Therefore in Figure 12, the simula-
tions that ran through the same parameters with dif-
46 | eCAADe 34 - CITY MODELING | Urban Planning Approach - Volume 2
ferent locations of initial resources and exactly at the Figure 9
same locations for initial resource locations have sim- Simulations with
ilar emergent behaviors. This similarity fortifies the random resources
idea of emergent clustering through the potentials at iteration 3000
created by the interactions of the agents and creates with same
similar results even the initial resources given or not. parameters
Furthermore, the main difference in these sim- RES10/FEED10/POP300.
ulations than the simulations in Section 2.3 is the
development restrictions on the canvas. The physi-
cal determinism has a massive impact on the simula-
tions for development clusters. Although, the devel-
opment restrictions only stated as 'developable' and
'undevelopable', the difference is immense.
In order to draw a conclusion, it is needed to
compare and observe the similarities with the real
information gathered and the results of the simula-
tions. Therefore, the population density map of Is-
tanbul, shown in Figure 13, and the superposed re-
Figure 10
sults of simulations, in Figure 12, show the cluster-
Simulations
ing of both maps have similarities. Even though the
superposed over
simulations are primitive to compare results with the
each other with
real information, the simulation created and emer-
random resources
gent structure based on the interaction of individual
at iteration 3000
entities for development. Neither the same initial re-
with same
source locations nor randomized initial resource lo-
parameters
cations created a massive change, but the locational
RES10/FEED10/POP300.
properties changed the way agents interact and how
agents initiate the development with the environ-
mental information. In Figure 14, the historical de-
velopment of urban areas in Istanbul shows gradual
sprawl of the urban areas from darker to lighter, as
old settlements to newer sprawl. As shown in Fig-
ures 12,13&14, the urban sprawl of Istanbul also have
same similarities with the outcomes of simulations.
CONCLUSION
Cities have become a research subject of social sci-
ences in the 20th century by the reasons of migra-
tion to urban trajectories. Therefore, the urban de-
sign theories have become insufficient to provide an
understanding of cities and how they are designed.
The methods used to model the complex systems in
other fields of sciences, particle physics, biology and
Figure 11 tions ran on Processing, which is an open-source API
Simulations with uses Java.
resources located The aim of the simulation is to create emergent
on district centers patterns and developments by the complexity of in-
at iteration 3000 teracting agents, which have basic behaviors. The
with same complexity, which created from simple individuals in-
parameters teracting together and how they behave, is the re-
RES10/FEED10/POP300. search's main subject. Furthermore, the research fo-
cuses on the development created by the interac-
tions of individuals to avoid any influence of initial
conditions at the beginning of simulations.
The case study is separated into 2 phases; the first
phase is testing the algorithm to find suitable param-
eters to simulate, and the second phase is the ap-
plication of the simulation over a city, which is Is-
tanbul in this case. The first phase investigated the
parameters and cross-referenced over each other to
find a parameter combination, which creates emer-
Figure 12 gent clusters of development apart from the initial
(left) Simulations givens. Moreover, the patterns emerged from the
superposed over simulations evaluated by the structures of clusters,
each other with which have polycentric structures, thus not sprawled
random resources over all canvas.
at iteration 3000 The second phase of the case study investigated
with same the possible similarities between the real information
parameters and maps gathered from Istanbul and the results of
RES10/FEED10/POP3000. simulations with the information that restricts the de-
(right) Simulations velopable areas of Istanbul. Although, the second
superposed over phase has two different situations differing the initial
each other with many other, have been implemented in urban de-
locations of resources, one with the random locations
locations of sign to research for a better understanding of cities
for initial resources of simulation and the other one
resources defined and how they behave or sprawl. The study focuses
locates the resources according to the district cen-
by district centers on computational tools and methods to investigate
ters of Istanbul at the current time. In this study, the
of Istanbul at and to understand the complexity of cities and ur-
emergent patterns independent from the initial re-
iteration 3000 with ban sprawl, while making a case study on a compari-
sources, these two situations are compared with each
same parameters son between a current situation and a simulation out-
other. The results of these situations came out to
RES10/FEED10/POP3000. come of a computational model.
have lots of similarities. Moreover, the simulations
The case study was building an urban sprawl
created clusters similar to the information maps of Is-
simulation with emergent properties to create com-
tanbul. Thus, there are many similarities, the simula-
plex outcomes with simple behaviors. The algorithm
tion model is built upon three main parameters and a
is built upon the methods and theory of Batty's re-
two parameters of land use. As a result, observing the
search on Agent-Based Modeling for urban sprawls.
similarities in structures of complexity with a real city
The computational theory and methods are trans-
by just five different parameters shows the potential
lated into an algorithm on Processing and simula-

CITY MODELING | Urban Planning Approach - Volume 2 - eCAADe 34 | 47

 of the model for further studies with more complex
algorithms and results.
48 | eCAADe 34 - CITY MODELING | Urban Planning Approach - Volume 2
REFERENCES
Batty, M 1994, Fractal Cities: A Geometry of Form and
Function, Academic Press, London
Batty, M 2005, Cities and Complexity: Understanding Figure 13
Cities with Cellular Automata, Agent-Based, Models Istanbul density
and Fractals, MIT Press map (Danış et al.
Danış, AD, Dörtbudak, Z, Erten, B, Gerçek, H, Gülümser,
AA, Güvenç, M, Kadıoğlu, M, Lewis, S, Ocak, E, Per- 2009).
ouse, JF, Polat, Z, Sanaç, E, Şahin, K, Üstün, B,
Waldheim, C and Yakut, SÖ 2009, Mapping Istanbul,
Garanti Galeri, Istanbul
DeLanda, M 2009, 'Swarm Urbanism', Architectural De-
sign: Digital Cities, 79, No:4, pp. 56-63
Gilbert, N 2008, 'Agent-Based Models', Quantitative Ap-
plications in the Social Sciences, 153, pp. 1-20
Jacobs, J 1961, The death and life of Great American Cities,
Random House, New York
[1] http://remappingistanbul2010.blogspot.com.tr/20
10/11/normal-0-21-microsoftinternetexplorer4.
html
[2] https://lsecities.net/media/objects/articles/ur
ban-spaces-in-and-around-istanbul/en-gb/ Figure 14
Urban sprawl map
of Istanbul (Guvenc
& E. Unlu-Yucesoy
2009)[2].
Towards Understanding the Complexity of Urban Culture
A case study of Jewish communities in pre-war central Poland

Malgorzata Hanzl1
1
Lodz University of Technology
1
mhanzl@p.lodz.pl

The current paper presents the experience of mapping pre-war Jewish

communities in central Poland during a period of intensive social and cultural
transformations. The initial analytical framework, provided thanks to the GIS
database and concatenation of attributes coming from various sources, makes
some initial observations and conclusions possible. It confirms the thesis that
Jewish communities in pre-war Poland may be considered an example of a
self-organising society, one which could be considered a prototype of
contemporary postmodern cultural complexity. The current study provides the
initial framework to map the morphology and spatial distribution of the
complexity of everyday culture of use of space proper to this extremely diverse
group.

Keywords: Urban design, urban morphology, anthropology, parametric

modelling, outdoor space

INTRODUCTION in urban design. The fragmentation and specialisa-

The current speed of urbanisation, and the develop-
ment of design technology that enables easy and ef-
ficient shaping of urban structures based on mathe-
matically defined parameters has created a need for
a more precise definition of the basic rules underpin-
ning the form of urban settings. The context that has
served so far as the main source of knowledge about
the culture of the usage of space no longer suffices
in the period of rapidly progressing urbanisation and
migrations of large groups of people. The need to
design places that answer actual human needs is be-
coming progressively more urgent. Lefebvre's call for
a single discipline of urban science (2003) requires a
more heuristic approach and a more efficient use of
the knowledge from the various disciplines involved
tion of science as well as historically established divi-
sions result in huge discrepancies in the terminology
and language used by researchers from various back-
grounds. Some of these problems, which exist es-
pecially on the margins of social and technical fields,
have already been addressed in the field of complex-
ity science (Portugali 2006).
The current paper, structured around the case
study of pre-war Jewish communities in central
Poland, presents a method to elaborate the morpho-
logical typology of Jewish settlements. An attempt
at typology shall enable an understanding of the pro-
cesses which influenced the forms of urban settings
inhabited by this group. After this introduction, a the-
ory of complex systems is briefly explained, with em-

CITY MODELING | Urban Planning Approach - Volume 2 - eCAADe 34 | 49

 phasis on context as an element affecting individual
choices, and thus self-organisation processes. The
case study follows, featuring a concise presentation
of social diversity in the studied group. The method
of analysis, key from the point of view of the confer-
ence topic, explained along with a brief summary of
results, has made it possible to elaborate on the fu-
ture development of the method.
THE THEORIES OF COMPLEX SYSTEMS
The term complex systems refers to systems which
are both open, so related to the surrounding envi-
ronment, and complex, which equals the inability to
describe the causal relations between parts of a sys-
tem on the one hand, and the emerging nature of
phenomena associated with their functioning on the
other (Portugali 2006). Self-organisation, understood
as the ability of a system to self-organise its inter-
nal structure (Portugali 2000, 49), is one of the ba-
sic features of complex systems. Self-organisation
is explained by means of such theories as synerget-
ics (Haken and Portugali 1995, Haken 1983), which
deals with mutual relations of parts of a system and
its functioning as a whole, and dissipative struc-
tures theory (Progogine and Stengers 1984), which,
applied to socio-economic and particularly urban
themes, focuses on fluctuations of people, values and
information in cities (Portugali 2006).
While the theories of self-organisation of cities
were elaborated in the second half of the twentieth
century, the phenomena which they describe started
even earlier. Complexity, as Portugali (2000, 315)
asserts, intimately related to the 20th century post-
modern urbanism, defines the phenomena which
emerged along with the dissolution of high context,
traditional cultures. Long established sets of norms,
practices and meanings for a given community trans-
formed and new ones emerged. The relations be-
tween physical settings and the communities living
there dispersed. The order of space, which accord-
ing to Alexander (1977) and Hanzl (2016, in press)
means the cultural framework which structures both
human behaviour and the physical settings, has been
50 | eCAADe 34 - CITY MODELING | Urban Planning Approach - Volume 2
changing in tandem. Although during this process
of transformation from traditional, high context so-
cieties to low context ones it loses some of its for-
mer meaning, still the architectural and urban envi-
ronment enables the screening and filtering of sen-
sory data, which is the way information is conveyed
(Hall 1966, 2). A proper understanding of the concept
of the meaning of urban settings (Hanzl 2013) is es-
sential here because, as Haken (2000, after Portugali
2006) states, the key feature in how complex systems
`self-organize', is that they `interpret', the information
that comes from the environment. It also needs high-
lighting that an important part of the internal struc-
ture of human settlements results from the layout of
pathways, as has been described in the widely recog-
nised theory of Space Syntax by Hillier and Hanson
(1984).
Some light may be cast on the above issues
thanks to the analysis of environments where con-
temporary complex societies started. The structures
left behind when juxtaposed with their cultural back-
ground lets us read the past activities and social rela-
tions. The forms of urban environments, approached
through culture related lenses, may lead to an un-
derstanding of the processes which led to the emer-
gence of contemporary multicultural societies. The
comprehension of the way multicultural communi-
ties developed and lived together through longer
spans of time may help while creating new, more
open environments which cater to the actual needs
of citizens, with regard to their culture related re-
quirements.
CASE STUDY
The current paper presents the experience of map-
ping pre-war Jewish communities in towns and vil-
lages of Mazovian and Lodz voivodeships in central
Poland, which went through a period of intensive
transformations. The focus on a group who lived sur-
rounded by other cultures, gives the opportunity to
follow their spatial threads left in physical settings.
The analysis is preceded by an in-depth study of the
cultural habits in a diachronic approach, with Figure 1
presenting an attempt at visual classification. Jewish
communities in central Poland in the period directly
preceding the Second World War consisted of sev-
eral groups of varied levels of acculturation, system
of values, adjustment to religious beliefs and politi-
cal views. The overall picture was very complex and
was what we understand as a contemporary hetero-
geneous society.
Social structure versus urbanisation pro-
cesses
Jews, who made up a large share of the population,
on the one hand mixed with Polish and German citi-
zens, on the other preserved their own culture. Along
with the acculturation processes, which progressed
with time, Jewish daily lives also changed, together
with the urban settings in which they lived. People,
enticed by the emerging opportunities, tried to im-
prove their living conditions or simply began look-
ing for any kind of income. The period of the noble
economy before the partitions of Poland found many
Jews still inhabiting small villages and manor houses,
many still working as leaseholders, even if no longer
dealing with propination. While in small villages the
picture was quite clear and showed a traditional com-
munity with a few more educated representatives of
intelligentsia, the growth of the population and the
range of available occupations, lifestyles and politi-
cal preferences, religious and social groups, resulted
in a much more complex mosaic.
The larger cities, including Warsaw and Lodz,
were surrounded with satellite cities and towns,
while small - shtetls - villages continued their pace
of life and, when it comes to the Jews living there,
their traditional, religion-based culture. Many towns,
starting from the period of the Kingdom of Poland,
went through intensive development as newly es-
tablished administrative centres. Many others de-
veloped as industrial centres, mainly, but not exclu-
sively, for the textile industry. In many towns, new
districts were developed, both by public authorities
and by private owners and other, mostly private,
towns were started from scratch.
CITY MODELING | Urban Planning Approach - Volume 2 - eCAADe 34 | 51
The situation of development during the indus-
trial era, overlapped with that of merchant urbani-
sation (terminology after Lefebvre 2003 and Portu-
gali 2006, 312). Most of these urban centres, espe-
cially those hosting markets, accommodated large
groups of Jews, their core population dealing with
such crafts as tailoring, dressmaking, shoe-making,
etc. Some towns had a longer or shorter tradition
of Jewish settling, with various conditions for the ac-
commodation of Jews. For instance, many towns for-
merly had the privilege, non tolerandis judeorum,
which forced Jews to settle in the private domains
of rich nobles. In other settlements, the Jewish pres-
ence was, until 1862, limited to specially established
zones. Next to the commerce which developed in all
urban conditions, the Jewish religious centres drew
many, with the important Chasidic courts serving as
major attractors. The Jewish faith and its require-
ments also encouraged specialised services, for in-
stance the making of prayer shawls in Maków Ma-
zowiecki. The acculturation processes of the inter-
war period further added to this mosaic, for instance
the popularity of medicine as a profession was picked
up by Jewish intelligentsia, leading to the develop-
ment of towns which served as health resorts offering
sophisticated and specialised health services. New
forms of urban structures developed, along with the
culture of spending holidays in the countryside.
THESIS
As a result of its historic development, each single
town and each single Jewish community can be char-
acterised with its own specific set of features (Wodz-
iński 2010, 195), which played a defined role in the ur-
banisation system. Each had its idiosyncrasy, which,
in order to be understood, must be mapped spatially
and with regard to time axis. While there is no one
single approach to present all cases, still morpholog-
ical typology needs to be developed to enhance un-
derstanding.
In spatial, social and cultural pluralism, the ques-
tions of human intentionality and socio-spatial emer-
gence remain central to social theory (Portugali 2000,

142). The correlation between individual prefer-
ences, values and intentions, and actual behaviour
and actions, is subject to Portugali's theory of self-
organisation (2000). Compared to Gidden's struc-
turalism, which focuses on society and groups, the
point of departure for Portugali (2000) are individu-
als and their personal choices. The above considera-
tions are in line with the empirical studies of the re-
lations between Jews and Poles, especially in large
cities, where more complex socio-cultural processes
could occur. This is one of the possible paths for the
further development of this research.
The current paper proposes a thesis that Jewish
communities in pre-war Poland represented an ex-
ample of a self-organising society, one which could
be considered a prototype of contemporary post-
modern cultural complexity. The mapping of this
complexity is a challenge, a method for which is ad-
dressed in the current study.
METHOD
The understanding of the relations between the pat-
terns of everyday life and the social habits of the com-
munities needs contemporary analytical tools. The
complexity of patterns has been approached with
52 | eCAADe 34 - CITY MODELING | Urban Planning Approach - Volume 2
Figure 1
Main intellectual
and political
threads, religious
movements and
migration processes
which affected
Jewish lifestyles in
prewar Poland.
the usage of a hierarchy of typologies. Three ba-
sic levels were used as a starting point for mod-
elling - that of region (Figure 2), that of town, and
that of neighbourhood. The analysis led to the pin-
pointing of certain cases which although considered
Jewish because of the high population share, at the
same time represented different classes and varied
significantly, including the way the structure could
be mapped. The basic level was a single neighbour-
hood, with such features analysed as culture related
notions of rhythms and sociometric layouts. The
heuristic method of highlighting those elements of
urban environments which were key for a given case
have been applied. The theory, based on the an-
thropological concept of meaning and the quest for
its reflection in spatial order as a feature attributed
to physical forms of urban structures, supported the
classification and understanding.
Database
While the making of the complete dataset needs
a bottom-up approach with individual case studies,
the general framework may already be defined, with
an assumption however that the classification may
be further extended. In order to achieve comprehen-
sive results which could fully render the reality, the
 database needs to be scalable. pology thresholds need to be introduced, referring to
the quantities of Jewish populations, their percent-
Figure 2
age and the size of towns or villages.
The main patterns
The initial study covering the available statistical
of distribution of
data may not only serve as a framework for further
the settlements
development but also reveals an image which may
with Jewish
be used as a source of valuable insights into the dy-
population in 1827,
namics of Jewish life in central Poland. The ArcGIS
1 – percentage of
10.1 has been used as a platform for mapping, with
Jewish population
shapefiles created describing the situation at a few
in towns and
historical moments based on the data available. The
villages, dots’ sizes
shapefiles were afterwards combined in order to fol-
describe size of a
low the processes of the emergence or collapse of
Jewish population ,
Jewish communities. While the data coming from the
2 – regional centre:
origins of Jewish presence in Poland is fragmentary
Warsaw, 3 –
and refers to the established kahalls only, which is still
remaining towns
the case of 1827 statistics (Figure 2), the most recent
and villages with
available statistics, coming from 1921, contain exact
Jewish population,
numbers of Jewish inhabitants, down to a single per-
4 - contemporary
son. This gives a very detailed picture of urban dy-
voivodeships
The combination of values from several fields would namics at this moment. Later census data of 1931 has
borders, 5 –
give a matrix, determining the type of individual not been published in a form enabling the analysis of
counties’ borders in
record. Concatenation of attributes should be done population dynamics.
1827.
separately for various time periods.
Table 1 The current study, departing from a more heuris-
Features included tic approach, starts by establishing a clear method
in the current set of of classification of lifestyles and values versus choice
attributes. of living places. The requirement to render the ac-
tual complexity of the typology means avoiding any
sort of predefined, closed and hierarchical structure.
Instead, what we need is a scalable system of de-
scription which may accommodate both new cate-
gories - fields - and new record values, as necessary.
Some historical documents have a predefined struc-
ture, which should be followed unaltered to convey
its historical testimony, e.g. classification of settle-
ments based on results of the 1921 census. While
the number of such comprehensive and closed doc-
uments is limited, they may be used as initial charac-
teristics, giving the general picture and introducing
background organisation. Another basic source of
knowledge is statistics, which contains mainly quan-
titative data. For the sake of the final assessment, ty-

CITY MODELING | Urban Planning Approach - Volume 2 - eCAADe 34 | 53

 The statistical data and typology of urban centres
were juxtaposed with the available knowledge on
new industrial and administrative towns and districts
started in the region in the 19th and 20th century, in-
cluding the Jewish population. Here many sources
were used, including databases held by: Polin Mu-
seum in Warsaw: www.shtetl.org.pl, YIVO Institute in
New York, Jewish Gen and others, as well as avail-
able research on the development of industry and ur-
banisation, both during the period of the Kingdom of
Poland and the interwar period. Studies on individual
towns and villages were also consulted, as well as nu-
merous historical elaborations referring to the Jewish
past in Poland. The full bibliography goes beyond the
volume of the current paper and will be presented in
a book containing the final results of the current re-
search.
Chosen features included in the current set of at-
tributes are listed in Table 1, along with a preliminary
set of values. The final list is yet to be elaborated or,
more likely, will be left open for possible extension.
As a result, an initial framework typology of urban
centres of various scales has been obtained, which
shall allow for a more profound understanding and
description of Jewish urban life in Poland in the pe-
riod before World War II.
RESULTS
While the traditional society offered a livelihood in
a highly defined and hierarchical structure, starting
with the abolishment of the noble economy and
along with the increasing industrialisation, the for-
mer stability was lost. Deprived of this economic
base, with the changing conditions of local commu-
nities no longer providing sufficient support for their
poor, Jewish masses were forced to look after them-
selves. Population growth and the shrinking mar-
ket for their services in locations they had lived so
far, together with the opportunities which opened
up thanks to the development of industry and the
growth of new administrative centres, forced cru-
cial changes in Jewish society. This was the situa-
tion when, with the progressing development of sec-
54 | eCAADe 34 - CITY MODELING | Urban Planning Approach - Volume 2
ular education, and, as a consequence, individual-
ism, more and more actors started planning and re-
planning their individual activities, looking for new
routines. For the Jewish population, accustomed to
adjusting to the decisions of governing bodies, in the
period of transition from a high context culture to a
low context one again easily adjusted, however this
time based more on individual and family decisions
than that of community.
The model defined by Portugali (2000, 243) as
a new order-parameter plan applies to the current
case study. With the constant migrations and adap-
tations to the situation of the already established or
newly-formed urban centres being a typical reality
for Jewish citizens in pre-war Poland. Both scenarios
which Portugali distinguishes apply here: Jews be-
came enslaved in the system of norms which ruled in
the place before their arrival, and they affected the
new reality, adjusting it to their needs and system
of rules and values appropriate to their own culture.
We may also establish a time-scale, as some larger
and shorter processes coexisted in the system. Cities,
for instance, which had served as primary or sec-
ondary administrative centres for years, were perma-
nent magnets for Jewish communities, who had lived
there since the Middle Ages. These processes pro-
vided the background for the much speedier devel-
opment of industrial towns, some booming then de-
clining very rapidly, others observing a more stable
development, or, like Lodz, growing into a metropo-
lis. Against the backdrop of economic activities, in-
dividual decisions were made, following individual
preferences and systems of values, as well as the out-
set preconditions, such as economic capacities. We
may presume that there was opposition between the
highly centralised hierarchical and, to a large extent,
oppressive nationalist practices of planning, and "the
highly distributed, diffused and decentralised urban
tradition" (Portugali 2000, 247).
The idea of self-organisation is crucial as it shifts
the understanding of agency. In a historical sense, it
has been up to now recognised that a society, espe-
cially the Jewish society in pre-war Poland, had very
 little influence on the actual appearance of cities, ten counting no more than 5 individuals, which may
or at least it was hard to define. Looking at the point to a single family.
same historical facts from the perspective of self-
Figure 3 organisation, we start to recognise their role in the
The main patterns urbanisation process. As particularly mobile citi-
of distribution of zens, they significantly altered the dynamics of ur-
the settlements ban growth. Performing various professional activ-
with Jewish ities, they undeniably influenced local economies.
population share They also influenced the form of dwellings and tene-
higher than 30% ments, which, originally universal and adaptable, had
and the quantity to now accommodate Jewish families, with their spe-
over 50 people in cific habits. Among the neighbourhoods they picked
the regional scale, 1 to live in, there were linear structures, more concen-
– percentage of trated ones, and even those which adopted much
Jewish population more contemporary models of dispersed urbanisa-
in towns and tion. In certain settlements, differing models devel-
villages, dots’ sizes oped one next to another, representing different cul-
describe size of a tures of the usage of space.
Jewish population
in 1921, 2 – regional The regional level
centres: Łódź and The demographic dynamics are comparatively high
Warsaw, 3 – for the Jewish population, yet quantitative thresh- The classification of a town's profile refers mostly to
remaining towns olds need to be defined in order to assess their repre- those settlements with a population larger than 50,
and villages with sentation in the settlement network of central Poland usually hosting a Jewish kehilla - a religious commu-
Jewish population, and provide an image at a given moment. In 1921, of nity. In the case of smaller settlements, the profile
4 - contemporary the total number of 345 settlements in central Poland was chiefly rural, with Jewish families catering to the
voivodeships with a Jewish population larger than 50, 70 had a basic needs of the surrounding communities in terms
borders, 5 – prewar Jewish population of 50 percent or more. Among the of commerce and tavern-keeping. In larger commu-
counties’ borders. multiplicity of towns and villages in central Poland, nities, there were also other basic crafts offered, tra-
some of them very small, others reaching up to sev- ditionally tailoring and shoemaking. Additionally, in
eral hundred thousand, Jews were usually present in many manors, small groups of Jews resided, perhaps
great numbers in the more urbanised ones - in 1921, being current or past leaseholders (Figure 4).
in the 70 mentioned above, 25 possessed the sta- When analysing the distribution of the Jewish
tus of a town, 27 were urban settlements and 10 still population in the scale of a region (Figure 5), two
villages, most quickly urbanising to become towns main patterns are noticeable, corresponding to the
in future or suburban colonies outside larger settle- two main types of prevailing economies: agrarian
ments. and industrial. The main difference between the two
The quantity of settlements with over 50 Jews patterns refers to the regularity of distribution and its
where they provided for 30 and more percent of the internal dynamics.
total population, reached up to 182 (Figure 3). In the The scale of this phenomena varied between
case of smaller settlements, depending on the over- counties, depending on the importance of the manor
all Jewish population in a given region, Jews were economy in general, and the role of agriculture.
generally equally spread, with individual groups of- Nonetheless, in some counties, such as Łowicz, apart
from major towns, Jews were hardly present.

CITY MODELING | Urban Planning Approach - Volume 2 - eCAADe 34 | 55

 tolerandis judeorum or pogroms purged Jews out of Figure 4
towns. In some towns, Jewish zones were created or Distribution of
collapsed and even in certain cases lost their char- manors with Jewish
ters because of fires. Reasons for growth were mul- dwellers (1921). 1 –
tiple, in the era of industrial production, the most manors, 2 –
common were related to the development of spe- regional centres, 3 –
cific industries, with textile or leather garment pro- contemporary
duction predominating. Despite this, many towns voivodeships, 4 –
and villages continued as before, servicing the neigh- prewar counties.
bouring country with minor crafts and commerce.
There were also numerous religious centres, in most
cases growing around the court of a Hasidic or Or-
thodox rabbi. Among centres of religious cults, the
most spectacular examples of rapid growth were Ger
(Góra Kalvaria), Przysucha and Aleksander (Aleksan-
drów Łódzki), all of them seats of famous Hasidic rab-
bis.
The presence of a specialisation did not however Figure 5
exclude other activities, with commerce and crafts Types of settlement
represented in all Jewish settlements. Smaller settle- patterns in the
ments maintained their unique profile and speciali- regional scale
sation, their development leading to complexity, and (1921): I – industrial,
the overlapping of several characteristics. These pro- II - rural, 2 – regional
cesses, accompanied by progressing acculturation centres: Łódź, 3 –
and assimilation, meant districts inhabited by Jew- contemporary
ish groups had specific features when it comes to so- voivodeships
cial status, education, professional profile, religious borders, 4 – prewar
observation, etc. Jewish prewar society was very di- counties.
verse, the individual preferences strongly affecting
both the actual lifestyle and, based on this, place
of living and form of settings. This is undoubtedly
a feature of contemporary complex societies, with
their specifics of low context culture and high context
ones. This phenomenon though most explicit in the
case of large metropolis, such as Warsaw and Lodz, is
also observable in cities with long traditions of Jew-
The town level
ish presence, such as Radom, Płock or Siedlce.
The variety of Jewish culture in pre-war central
Rapidly developing towns and villages gained
Poland found its reflection in the variety of neigh-
new extensions, such as colonies or suburbs. This
bourhoods inhabited by Jews. Each town had its own
rapid suburbanisation proceeded with Jews, gravitat-
history, its own specific reasons for development and
ing to these new urban centres, taking an active part
growth. The Jewish population might change over
in this, often informal, development, sometimes hap-
time, first growing then declining when the gentile
pening just outside a town's administrative borders,
citizens of many towns obtained the privilege non

56 | eCAADe 34 - CITY MODELING | Urban Planning Approach - Volume 2

on private land, as in the case of Lodz Bałuty. In other
cases, like in Koluszki, a former village which grew
strong, it was due first to the construction of the War-
saw Vienna Railway and later on thanks to its connec-
tions to Lodz and further to Kalisz and Poznań and
the important railway hub there. Several smaller vil-
lages in its direct proximity urbanised, leading to the
growth of the town and attracting Jews. Jews who
settled in the neighbouring villages, in most cases,
profited from the existing community facilities, to-
gether forming a thriving community. Such a cluster
may or may not become an actual town.
The structure of a traditional Jewish neigh-
bourhood
In many towns, the districts inhabited by Jews be-
came more urban than the surroundings, with ur-
banisation pressure and densities higher than else-
where. Jewish citizens often flocked to historic me-
dieval cores, which usually hosted local markets.
There they preferred locations next to the river val-
ley, which, firstly, were available and much cheaper
due to the flood danger and, secondly, catered to
their needs in terms of religious customs. While the
plan of a town after the arrival of Jews remained rela-
tively unaltered, especially when speaking of the net-
work of public streets and squares, the urban blocks
were replaced with new, masonry structures of a
much more urban character, as in Brzeziny near Lodz.
Moreover, the way space was used changed, from a
more fragmented and linear one, into one integrated
around typically Jewish focal points, which reflected
the three main spheres of life: sacrum - represented
by a synagogue, house of study, ritual bath, pro-
fanum - concentrated around work - and represented
by the market place, and domicile - in the twentieth
century represented by a backyard.
The transformation of the former parcellation,
explicit in the plans of towns inhabited by Jews
at the turn of the eighteenth century, followed a
different social organisation. The community ke-
hilla, initially very strong, practically independent
from the town's regulations, functioned separately
CITY MODELING | Urban Planning Approach - Volume 2 - eCAADe 34 | 57
from the surrounding town, even if well integrated
into its structure, while still catering to the needs
of the town's citizens and visitors. This position
had its reflection in the urban structure, with inte-
rior space within traditional Jewish quarters seem-
ingly unordered and uninviting to visitors, while for
Jews serving as extensions of their dwellings and a
space of interior circulation. The last observation
confirmed by the tradition of eruvim - temporary
cords distinguishing private space during Sabbath.
With time this former hierarchical structure released,
giving way to multiplicity of lifestyles, which were
also reflected in the actual form of neighbourhoods.
The features which are listed in the current paper are
explained in more depth in other articles stemming
from the current study (Hanzl 2013, 2014, in press).
CONCLUSIONS AND FUTURE RESEARCH
PATHS
This paper provides a summary of the larger research
on the most typical neighbourhoods inhabited by
Jews in central Poland in the pre-war period. The
initial analytical framework, provided thanks to the
GIS database and concatenation of attributes com-
ing from various sources, makes some initial obser-
vations and conclusions possible. Firstly, it confirms
the thesis that Jewish communities in pre-war Poland
may be considered an example of a self-organising
society, one which could be considered a prototype
of contemporary postmodern cultural complexity.
While the complexity of everyday culture of use of
space is challenging, the current study provides the
initial framework to map its morphology and spatial
distribution. The data collected reveals the follow-
ing features which differed between settlements and
neighbourhoods:
• acculturation level and attitude towards reli-
gion,
• main professional profile of a settlement or of
a neighbourhood,
• size of a community.
 Additionally, characteristics changed over time, with
new layers resulting from industrialisation and the
development of modern administration, overlapping
with the former urban centres coming from the pe-
riod of merchant urbanisation (Lefebvre 2003 and
Portugali 2000, 312). The model of a new order-
parameter plan (Portugali 2000, 243) entirely applies,
Jewish citizens in pre-war Poland used to migrate and
adjust to external conditions, either enslaved in the
system of norms which ruled in the place before their
arrival, or adapting it to their own needs.An attempt
at the classification of larger urban centres is given,
distinguishing such types as industrial, administra-
tive, commercial and religious ones. In the settle-
ment network of smaller Jewish concentrations, two
patterns have been noticed: a dynamic and a static
one, linked to industrial and rural economies. The
detailed analyses of census data in the scale of the
urban region made it possible to follow the urban
growth processes.
Further research aims at more in-depth illustra-
tion of the above characteristics, with the analyses
of individual case studies. The data base framework
may be further completed and extended in a bottom-
up way, thanks to its open structure. Another chal-
lenge may be the examination of the spatial, social
and cultural pluralism through the questions of hu-
man intentionality and socio-spatial emergence with
relation to the meaning conveyed by the form of the
urban settings.
ACKNOWLEDGEMENTS
The current study is a part of the research project
financed by The National Science Centre of Poland
UMO-2011/03/D/HS3/01630, entitled: "Morphologi-
cal analysis of urban structures - the cultural approach.
Case studies of Jewish communities in the chosen set-
tlements of Lodz and Masovian voivodeships". Part
of the methodological research was performed in
SENSEable City Laboratory, Massachusetts Institute
of Technology within the framework of a Fulbright
Senior Research Award.
58 | eCAADe 34 - CITY MODELING | Urban Planning Approach - Volume 2
REFERENCES
Alexander, C, Ishikawa, S and Silverstein, M 1977, Pattern
Language, Oxford University Press, New York
Haken, H 1983, Synergetics. An Introduction, Springer-
Verlag, Berlin
Haken, H and Portugali, J 1995, 'A synergetic approach to
the self-organisation of cities', Environment and Plan-
ning B: Planning and Design, 22, pp. 35-46
Hanzl, M 2014, 'Culture as a Determinant of City Form
- the case of the former Jewish district in Lodz', in
Bovati, M, Caja, M, Floridi, G and Landsberger, M
(eds) 2014, Cities in Transformation Research \& De-
sign: Ideas, Methods, Techniques, Tools, Case Studies,
Politecnico di Milano/eAAe/ARCC, Milano, pp. 625-
634
Hanzl, M 2013, 'The meaning of public spaces', in Bártolo,
H (eds) 2013, Green Design, Materials and Manufac-
turing Processes - Proceedings of the 2nd International
Conference on Sustainable Intelligent Manufacturing,
SIM 2013, CRC Press, Taylor & Francis Group, Boca Ra-
ton, London, New York, Leiden, pp. 39-44
Hanzl, M in press 'Towards Understanding of Spatial Or-
der as a Cultural Issue', Proceedings of ASK. The Con-
ference, Warsaw
Hillier, B and Hanson, J 1984, The Social Logic of Space,
Cambridge University Press, Cambridge
Lefebvre, H 2003, The Urban Revolution, University of
Minnesota Press, Minneapolis, London
Portugali, J 2000, Self-Organization and the City,
Springer-Verlag, Berlin Heidelberg
Portugali, J 2006, 'Complexity theory as a link between
space and place', Environment and Planning A, 38(4),
pp. 647-664
Prigogine, I and Stengers, I 1984, Order out of Chaos, Ban-
tam Books, New York
Wodziński, M 2010, 'Żydzi w okresie zaborów', in
Sienkiewicz, W (eds) 2010, Atlas Historii Żydów Pol-
skich, Demart SA, Warszawa, pp. 155-246
Parametric Modelling with GIS
Patrick Janssen1 , Rudi Stouffs2 , Akshata Mohanty3 , Elvira Tan4 ,
Ruize Li5
1,2,3,4,5
National University of Singapore
1
patrick@janssen.name 2 stouffs@nus.edu.sg
3,4,5
{akshatamohanty|elvira.tan|liruizenus}@gmail.com

Existing urban planning and design systems and workflows do not effectively
support a fast iterative design process capable of generating and evaluating
large-scale urban models. One of the key issues is the lack of flexibility in
workflows to support iterative design generation and performance analyses, and
easily integrate into design and planning processes. We present and demonstrate
a parametric modelling system, Möbius, that can easily be linked to Geographic
Information Systems for creating modular workflows, provides a novel approach
for visual programming that integrates associative and imperative programming
styles, uses a rich topological data structure that allows custom data attributes to
be added to geometric entities at any topological level, and is fully web-based.
The demonstration consists of five main stages that alternate between QGIS and
Möbius, generating and analysing an urban model reflecting on site conditions
and using a library of parametric urban typologies, and uses as a case study an
urban design studio project in which the students sketched a set of rules that
defined site coverage and building heights based on the proximity to various
elements in the design.

Keywords: generative design, urban planning, Geographic Information Systems,

parametric modelling

INTRODUCTION tive, and constrained by time, skills and the availabil-

Existing urban planning and design systems and
workflows do not effectively support a fast iterative
design process capable of generating and evaluating
large-scale urban models.
One of the key issues is the lack of flexibility in
workflows to support iterative design generation and
performance analyses, and easily integrate into de-
sign and planning processes. Especially when such
processes are characterized as dynamic, collabora-
ity of systems. While various systems exist to sup-
port performance analysis within urban planning and
design processes, these tend to be neither simple
to use, nor flexible in their ways of usage, nor can
they be easily integrated with other systems that may
serve other parts of the workflow.
Various systems for parametric urban design al-
ready exist. Beirão et al. (2011) present a parametric
urban design system that provides automatic feed-

CITY MODELING | Urban Planning Approach - Volume 2 - eCAADe 34 | 59

 back on a number of urban indicators and density
measures. König (2015) presents an open source li-
brary for computational analysis and synthesis, de-
noted CPlan, which supports the optimization of spa-
tial configurations. However, both systems only con-
sider buildings as rectangular blocks and do not sup-
port a differentiation in building typologies. Knecht
and König (2012) present a tool that, given a street
network, generates building lots and buildings, the
latter in one of four types: row buildings, courtyard
buildings, ribbon buildings and free-standing blocks.
However, the tool supports only one building type at
a time and treats other parameters, such as building
depth, similarly. As such, it mainly focuses on small
developments.
Esri CityEngine (Müller et al. 2016) considers a
procedural modelling approach for generating 3D
city models. Specifically, it adopts a rule-based ap-
proach inspired by shape grammars but using proce-
dural rules. While CityEngine is very comprehensive
in terms of modelling different building typologies,
including their facades, it primarily targets 3D visu-
alization over analysis and assessment. Additionally,
much of the work in using CityEngine to generate 3D
cityscapes is manual in nature, developing or select-
ing rules that apply to specific plots in order to gener-
ate the relevant buildings. Finally, though CityEngine
supports some analysis, this is fairly limited and the
data generated within CityEngine cannot easily be
exported back into a 2D GIS environment (besides
Esri ArcGIS) for detailed analysis and assessment.
Modular Workflows
Modular workflows that can more easily be adapted
by the user are of special interest. Such work-
flows often rely on various software systems that
must be seamlessly connected in order to allow the
user to switch back and forth between modelling
and analysis. Workflows capable of integrating geo-
graphic mapping and parametric modelling systems
could enable various rapid iterative urban prototyp-
ing methods to be developed. Geographic mapping
uses Geographic Information Systems (GIS), such as
60 | eCAADe 34 - CITY MODELING | Urban Planning Approach - Volume 2
QGIS [1] and Esri ArcGIS [2]. Parametric modelling
is generally supported by Computer Aided Design
(CAD) systems together with various plugins, with
four popular tools being McNeel Grasshopper [3],
Bentley GenerativeComponents [4], Autodesk Dy-
namo [5], and SideFX Houdini [6].
Geographic Information Systems can be used to
integrate geospatial data of various types, to gener-
ate 2D maps as a starting point for the urban design
exploration process, and to perform various types of
2D analysis, including area calculations, buffer analy-
sis, and network analysis. Parametric modelling sys-
tems can be used to generate 2D street layouts and
parcel subdivisions, to generate 3D urban massing
studies and morphologies, and to perform various
types of 3D analysis, including view analysis, solar ra-
diation analysis, daylight analysis, and privacy anal-
ysis. Furthermore, both types of modelling systems
can be linked to other more advanced simulation
programs. For example, Geographic Information Sys-
tems can export data for transport simulation pro-
grams, and parametric modelling systems can export
data for Computational Fluid Dynamic (CFD) simula-
tion programs.
The ability to easily exchange data between Ge-
ographic Information Systems and parametric mod-
elling systems could potentially result in an ecosys-
tem of flexible workflows that could be modified and
adjusted to tackle a wide variety of urban modelling
scenarios. The problem is that, currently, exchanging
data between these systems is difficult at best. The
main reason for this is differences in how data is rep-
resented within these systems.
Modelling with Attribute Data
When comparing the representations used by Geo-
graphic Information Systems and parametric mod-
elling systems, many differences can be identified be-
yond the 2D/3D distinction. For example, paramet-
ric modelling systems typically support a wide vari-
ety of geometric types including spline-based curves
and surfaces. Geographic Information Systems on
the other hand have extensive tools sets for working
with both raster and vector data, but the geometric
types supported by the latter are typically very re-
stricted, often consisting only of three types: points,
polylines, and polygons. While such differences are
important, for this research the key difference relates
to how non-geometric data can be associated with
geometric entities in the model.
In Geographic Information Systems, attribute
data can be attached to geometric entities such as
points, polylines and polygons. This attribute data
can represent any arbitrary information that the user
wishes to associate with the geometric entities. The
ability to work with attribute data is a fundamental
feature of geographic mapping and can be very pow-
erful.
The attribute data can be displayed using at-
tributes tables, where each row in the table repre-
sents a geometric entity and each column represents
an attribute. Users can define attribute values explic-
itly or by using formulas that calculate values based
on the geometry or on other attributes. Attributes
are typed, typically being integer numbers, floating
point numbers, text strings, or dates. Geometric en-
tities and their associated attributes can be exported
in a number of file formats, including Shape files and
GeoJSON files.
In contrast, most CAD systems either do not al-
low attribute data to be attached to geometric enti-
ties, or only support it in limited ways. One example is
Trimble SketchUp [7], which allows data to be added
to groups or components. However, this means that
separate groups or components need to be created
for every geometric entity that needs to have data at-
tached to it. For large models, this can become very
cumbersome.
Parametric modelling systems also do not tend
to support attribute data. Instead, this type of data
needs to be handled as a separate dataflow, which
imposes a significant burden of data management
on the user. For example, in Grasshopper, there are
various plugins for reading and writing GIS data. An
example is the Heron plugin [8], which is able to im-
port GIS data from various sources. The plugin pro-
CITY MODELING | Urban Planning Approach - Volume 2 - eCAADe 34 | 61
vides a node that can read GIS data, but the output
produces disconnected data sets for the geometric
entities and the attribute data. Managing these sep-
arate data sets and keeping them synchronised can
become very complex.
An interesting exception is the Jeneratiff plugin
for Grasshopper (Dritsas 2016). This plugin provides
a library of various advanced computational design
algorithms supported by an infrastructure that al-
lows arbitrary data to be associated with geometric
entities. The data is stored in dictionaries, with the
possibility of nesting dictionaries inside one another,
thereby allowing more complex data structures to be
created. However, due to the limitations of the un-
derlying Rhino platform, the association between ge-
ometry and data only exists within the Grasshopper
environment. Once the model is transferred back to
Rhino, geometric entities lose all their data.
One parametric modeller with more direct sup-
port of attribute data is Houdini. Furthermore, ge-
ometric entities are represented using a topological
data structure and attributes can be added at three
levels: primitives, vertices and points, in addition to
the global level representing all geometries. Prim-
itives are geometric entities such as polylines and
polygons. Each primitive is defined by a set of ver-
tices, and each vertex has a position in space that is
defined by a point. Attribute data can be added at
any of these three levels. This means that a polygon
can have different data added to the points, the ver-
tices, and the polygon itself. The attribute data is dis-
played as tables in a similar way to GIS systems.
The authors have previously developed various
parametric GIS workflows using QGIS for geograph-
ical mapping and Houdini for parametric modelling
[9]. The ability of Houdini to be able to handle at-
tribute data in a similar way to GIS systems allowed
data to be easily exchanged. Moreover, the combi-
nation of the topological data structure with the at-
tribute data in Houdini also proved to be very power-
ful, even though this was not supported in QGIS. For
the geographic mapping, the flat data structure was
found to suffice, but for the parametric modelling the
 topological data structure was very useful.
Despite these advantages, Houdini is still found
to be lacking for more advanced modelling tasks.
First, the parametric modelling approach is predom-
inantly a dataflow approach with rather weak sup-
port for procedural modelling (Janssen and Stouffs
2015). Second, the topological data structure focuses
on individual geometric entities but does not include
more complex objects consisting of multiple geo-
metric entities. Therefore, a new type of parametric
modelling system was developed, called Möbius [10],
which can easily be linked to Geographic Information
Systems for creating modular workflows.
Paper Overview
This paper describes the Möbius system and uses a
case study to demonstrate how Möbius can support
complex parametric GIS workflows. The paper con-
sists of three main parts, with the first part describing
the Möbius system, the second part focusing on the
case study, and the third part discussing future chal-
lenges.
62 | eCAADe 34 - CITY MODELING | Urban Planning Approach - Volume 2
OVERVIEW OF MÖBIUS
Möbius is a web-application that runs in the browser
on the client side. Figure 1 shows the Möbius user
interface. Compared to existing systems, Möbius of-
fers two key advantages. First, it provides a novel ap-
proach for visual programming. Second, it uses a rich
topological data structure that supports objects con-
sisting of multiple geometric entities. Each of these
are described in more detail below.
A Novel Visual Programming Approach
The parametric modelling approach in Möbius in-
tegrates associative and imperative programming
styles. The associative style of programming is
supported through dataflow programming, where
the user constructs networks consisting of nodes
and wires. The imperative style of programming
is supported through blocks-based programming,
where the user constructs procedures by creating se-
quences of code blocks (Resnick et al. 2009). Each
node in the dataflow network has an imperative pro-
cedure that is defined using the blocks-based ap-
Figure 1
The Möbius
interface.
proach. This combined approach allows for the
dataflow network to be conceived at a higher level of
abstraction, allowing for smaller dataflow networks
that can more easily be read and understood.
The node network is defined with the associative
programming style, using panes (b) and (c) in Figure
1. Pane (b) is the node parameters pane, and is used
to set certain input values for the node instance se-
lected in pane (c). Pane (c) is the network pane, and
is used to define the dataflow network. The inputs to
a node instance can either be defined by connecting
a wire to the input port, or by entering a specific pa-
rameter value through the user interface widget. In
the example in Figure 1, the generate_model0 node
instance is selected, and the parameters shown are
therefore for this node.
When the dataflow network is run, the network
is converted into a single JavaScript program that
is then executed in one go. This represents a syn-
chronous type of execution, where the nodes are first
topologically sorted and then mapped to code in a
fixed order. This differs from most other dataflow sys-
tem, where nodes are executed asynchronously in re-
sponse to changes to their inputs.
The generation of an execution program from
the dataflow network has the advantage that it al-
lows for a wide range of more advanced program-
ming mechanisms to be used. Two such mechanisms
are iterative loops and higher-order functions. For a
more detailed description of these mechanisms, see
Janssen et al. (2016).
Iteration refers to mechanisms for repeating a
sub-procedure, and includes for-loops, while-loops
and recursive function calls. Visual Programming
Language (VPL) systems for parametric modelling
typically use a graph-based programming approach
consisting of nodes and wires, and as a result do
not support iteration very well. Nevertheless, with
dataflow (e.g. Grasshopper) and procedural mod-
elling systems (e.g. Houdini), it is possible to cre-
ate models that incorporate iterative procedures
(Janssen and Stouffs 2015). However, creating so-
phisticated types of iterations in these types of sys-
CITY MODELING | Urban Planning Approach - Volume 2 - eCAADe 34 | 63
tems (such as nested loops) remains prohibitively dif-
ficult (Janssen and Stouffs 2015).
In Möbius, iteration is supported by defining
imperative procedures using the blocks-based pro-
gramming. Each node is an instance of a type,
which is defined by specifying a procedure with a set
of inputs and outputs. This procedure can include
looping control-flow constructs such as for-loops and
while-loops. In Figure 1, the procedure in the gen-
erate_model0 node iterates over all the plots in the
GeoJSON file and generates a building on each plot.
Higher-order functions refer to any functions
that take another function as an argument or return
a function as a result. In parametric modelling, such
functions can be very useful as they allow compo-
nents in the design to be represented and manipu-
lated using functions (Leitão 2014). Most VPL sys-
tems do not support higher-order functions. Two
recent exceptions are Autodesk Dynamo and Au-
todesk 3DS MCG [11]. However, the dataflow inter-
face makes it difficult to leverage the full power of
higher-order functions.
In Möbius, higher-order functions are supported
by an additional output port, which returns a refer-
ence to the function that wraps the procedure for
that node. When this output is wired into another
node, it results in a dashed wire, which indicates to
the user that it contains a reference to a function
rather than actual data. These wires can be used
to provide a function as an input to another down-
stream node. That node can then have a procedure
that executes that function. In the case study, alter-
native building typologies are conceived as higher-
order functions. In this way, selecting a particular
building typology only requires the function refer-
ence to be provided as input to the generator node,
rather than embedding the procedure for the func-
tion node into the generation network. For exam-
ple, in Figure 1, the three 'typology' nodes define pro-
cedures that generate buildings with alternative ty-
pologies. These are linked as higher-order functions
into the generate_model0 node, which uses these to
generate the urban model.
 A Topological Data Structure with Attribute
Data
Möbius uses a topological data structure consisting
of six levels: points, vertices, edges, wires, faces, and
objects. Points are similar as in Houdini, and repre-
sent positions in space. Geometric entities are then
defined using the various topological levels. For ex-
ample, a polygon is a face defined by a closed wire, a
wire is defined by a series of edges, an edge is defined
by two vertices, and a vertex is defined by a point.
These topological levels are similar to those found in
the OpenCascade modelling kernel [12].
Custom data attributes can be added to the geo-
metric entities at any of the topological levels. When
creating procedures, users are able to use functions
that directly read and write attribute data at specific
entries in the topological hierarchy.
For example, a polygon representing an urban
plot may have some face attributes that define the
target site coverage and building height. In addition,
one of the edges of that polygon may also have at-
tribute data that specify that the edge is adjacent to
a road. This allows the parametric procedure to ori-
ent the buildings that are generated on that plot, so
as to ensure that they are facing the appropriate di-
rection relative to the road.
CASE STUDY
The case study stems from an international, collab-
orative design studio (Winter School) involving over
170 students and 30 design tutors, organized by
the International Forum on Urbanism (IFoU). Partic-
ipants from all over the world came together to work
for 12 days on proposals for the Jurong Industrial
Estate (JIE), a 5000 ha industrial area in the west
of Singapore. The brief was to develop proposals
for the transformation of JIE from an almost mono-
functional, segregated and fragmented, polluted in-
dustrial area into a major catchment area for future
population growth that integrates clean(ed) indus-
trial plants with green lungs, attractive housing and
vibrant urbanity for one million people.
Some of the results of the IFoU Winter School
64 | eCAADe 34 - CITY MODELING | Urban Planning Approach - Volume 2
were further developed in an urban planning studio
within the Master of Urban Planning program at the
Department of Architecture, National University of
Singapore. The case study focuses on one of these
projects, called Ecotopia. As part of the proposal, the
students developed a set of rules that defined urban
parameters based on the proximity to various ele-
ments in the design.
The site was divided into three zones as shown
in Figure 2: the residential zone, the industrial zone,
and the commercial zone. A set of rules was then
diagrammed, as shown in Figure 3. The rules de-
fine how site coverage and building heights vary in
relationship to roads (classes A, B and C), parks and
other boundaries. The leftmost rules apply to the res-
idential zone, the second set of rules to the indus-
trial zone, and the third to the commercial zone. The
rightmost rules apply to all zones and relate to the
proximity to parks.
Figure 2
Division of the site
into three zones.
The case study
focuses on the area
indicated by the
dotted line.
In the proposed parametric GIS workflow, QGIS is
linked to Möbius in order to support fast iterative
generation and evaluation of large-scale urban mod-
els. The workflow uses parameter fields and relies
heavily on the ability to attach attribute data to geo-
metric entities in the model in both QGIS and Möbius.
The workflow consists of five main stages that alter-
nate between QGIS and Möbius, with data exchange
using the GeoJSON file format.
Stage 1: Geographic Mapping in QGIS
QGIS is used to create a map of the site area, includ-
ing existing buildings and infrastructure. Geographic
data is collected and manually integrated into a sin-
gle geospatial dataset. A series of large parcels are
Figure 3
Ecotopia sketched
rules.

defined in QGIS, indicating the areas of the site where are exported as a GeoJSON file and imported back
future development is proposed. The map is shown into QGIS.
in Figure 4. The parcels are exported from QGIS and
imported into Möbius. Stage 3: Parameter Generation in QGIS
QGIS is used to create parameter fields through a
Figure 4 combination of proximity functions and custom for-
A map of the area mulas that capture the spatial rules shown in Figure
for future 3. A series of additional attributes are defined in the
development in attribute table for the plots for this purpose, in three
QGIS. The shaded steps. First, proximity attributes are created that cal-
area indicates the culate proximity to roads, parks, and the waterfront.
parcels where road Second, parameter attributes are created that calcu-
networks and late the building height and plot coverage for each
building typologies of the different rules shown in Figure 3 based on the
need to be proximity attributes. This results in each plot being
developed. assigned two building heights and plot coverages.
Third, a final pair of attributes is created that calcu-
lates the final value for both parameters using for-
mulas that give priority to certain rules or conditions.
The plots with the attributes are exported from QGIS
and imported into Möbius.
Stage 2: Parcel Subdivision in Möbius
Möbius is used to recursively subdivide each parcel Stage 4: Urban Model Generation in Möbius
into similar size plots, with tertiary roads also inserted Möbius is used to generate urban models using a li-
between the plots. The subdivision is performed by brary of parametric urban typologies. The parame-
a parametric procedure that attempts to create plots ters for each plot are extracted from the attributes
that are as evenly sized as possible. The parameters attached to the plot polygon; namely the building
allow the designer to specify the target size of the height and the plot coverage. These parameters can
plot. The transformation from large parcels to smaller be viewed in Möbius in an attribute table, as shown
plots is shown in Figure 5. The plots and tertiary roads in Figure 6. The urban model is then generated by se-

CITY MODELING | Urban Planning Approach - Volume 2 - eCAADe 34 | 65


lecting and instantiating a parametric urban typol-
ogy on each plot, as shown in Figure 7(a).
Each plot together with the urban model on that
plot is treated as a single object. The polygons in that
object have a 'type' attribute that defines the types of
building elements that they represent, such as
plot, footprint, floor, wall, roof , and window. Floors
are further categorised into three types: residential,
commercial and industrial. Möbius is then used to cal-
culate the floor areas for each of the different func-
tions. These values are then added as attributes to
the footprint polygons. For example, if a plot has two
buildings on it, then it will have two footprint poly-
gons. For each building, the floor areas will be calcu-
66 | eCAADe 34 - CITY MODELING | Urban Planning Approach - Volume 2
Figure 5
The process of
subdividing large
parcels into smaller
plots in Möbius. (a)
The large parcels.
(b) The subdivided
plots.
lated and added as attributes to the respective foot- Figure 6
print polygon. The attribute table
Finally, in order to be able to export the model for the plots in
back to QGIS, the model needs to be flattened to Möbius.
2D. The critical information that needs to be trans-
ferred is the building footprints together with the
floor area attributes. These attributes are important
as they will be used for the QGIS analysis in the next
stage. Möbius exports the building footprints as a 2D
model which can then be imported back into QGIS.
Figure 7(b) shows the resulting 2D model, and Figure
8 shows the attribute table for the plots and building
footprints.
Stage 5: Spatial Analysis in QGIS
QGIS is used to analyse the flattened map of the ur-
ban model. First, a set of additional attributes are
added to the building footprint polygons represent-
ing the number of people in each building for each
of the functions. These values are calculated using a
formula that divides the floor area for each function
by average 'area-per-person' values. The attributes
are then used to calculate the percentage of peo-
ple within a certain walking distance of transport
nodes. This is performed using a simple buffer analy-
sis, which is based on the straight-line distance from
Figure 7
The models in
Möbius. (a) The 3D
urban model. (b)
The 2D flattened
model, for export to
QGIS.
the centre of the building footprint to the closest
transport node. Population densities are also visu-
alised by colouring the polygons. Figure 9 shows the
resulting map in QGIS.
Figure 8
The attribute table
for the building
footprints in
Möbius.
CONCLUSIONS AND FUTURE RESEARCH
A parametric modelling system called Möbius has
been described that supports smooth exchange of
data with Geographic Information Systems. In par-
ticular, Möbius provides a novel approach for visual
programming that integrates associative and imper-
ative programming styles and uses a rich topologi-
cal data structure allowing custom data attributes to
be integrated with geometric entities at any topolog-
ical level. This facilitates the creation of flexible work-
flows that can be modified and adjusted to tackle a
CITY MODELING | Urban Planning Approach - Volume 2 - eCAADe 34 | 67
wide variety of urban modelling scenarios.
The case study demonstrates how Möbius can
support complex parametric GIS workflow. In the
case study, a workflow is developed that switches be-
tween the Geographic Information System and the
parametric modelling system multiple times, using
each tool for what it is best at, mainly analysis and
generation, respectively.
Through this research, some potential benefits
of rule-based modelling have started to emerge. Fu-
ture research will explore the integration of a rule or
grammar-based data synthesis method within the vi-
sual programming approach. This will require the
conception of a relevant description model for the
specification, selection and execution of grammar
rules within Möbius. This will further improve work-
flow flexibility allowing users to choose whether to
express rules, e.g. for determining building height
and plot coverage from the proximity attributes,
within Möbius or the Geographic Information Sys-
tem, as procedural expressions, higher-order func-
tions, or grammar rules. Ideally, workflow flexibility
should allow users to select and develop their own
preferred workflow in support of existing design pro-
cesses and preferential working methods.
 Figure 9
The 2D map with
the buffer analysis
showing the
percentage of
people within a
certain walking
distance from
transport nodes.

REFERENCES Müller, P, Wonka, P, Haegler, S, Ulmer, A and van Gool, L

Beirão, JN, Nourian, P and Mashhoodi, B 2011 'Paramet-
ric urban design: An interactive sketching system
for shaping neighborhoods', Proceedings of eCAADe
2011, Ljubljana, pp. 225-234
Dritsas, S 2016 'An advanced parametric modelling li-
brary for architectural and engineering design', Pro-
ceedings of CAADRIA 2016, Melbourne, pp. 611-620
Janssen, P, Li, R and Mohanty, A 2016 'Möbius: a para-
metric modeller for the web', Proceedings of CAADRIA
2016, Melbourne, p. 157–166
Janssen, P and Stouffs, R 2015 'Types of parametric mod-
elling', Proceedings of CAADRIA 2015, Daegu, pp. 157-
166
Knecht, K and König, R 2012, 'Automatische Grundstück-
sumlegung mithilfe von Unterteilungsalgorithmen
und typenbasierte Generierung von Stadtstruk-
turen', Arbeitspapiere (Working Papers) Informatik in
der Architektur, 15, pp. 3-21
König, R 2015 'CPlan: an open source library for com-
putational analysis and synthesis', Proceedings of
eCAADe 2015, Vienna, pp. 245-250
Leitão, M 2014 'Improving generative design by com-
bining abstract geometry and higher-order pro-
gramming', Proceedings of CAADRIA 2014, Kyoto, p.
575–584
2006, 'Procedural modeling of buildings', ACM Trans-
actions on Graphics, 25(3), pp. 614-623
Resnick, M, et al. 2009, 'Scratch: programming for all',
Communications of the ACM, 52(11), pp. 60-67
[1] http://www.qgis.org/
[2] https://www.arcgis.com
[3] http://www.grasshopper3d.com/
[4] https://www.bentley.com/en/products/product-lin
e/modeling-and-visualization-software/generat
ivecomponents
[5] http://www.autodesk.com/products/dynamo-
studio/overview
[6] https://www.sidefx.com/
[7] http://www.sketchup.com/
[8] http://www.food4rhino.com/project/heron
[9] https://gis4design.wordpress.com/
[10] https://gis4design.wordpress.com
[11] http://mobius.design-automation.net/
[12] http://www.opencascade.com/doc/
occt-7.0.0/overview/html/
technical_overview.html

68 | eCAADe 34 - CITY MODELING | Urban Planning Approach - Volume 2

CITY MODELING | Applications
Autopoietic Features of the Urban Body's Elements
Similarity studies on network elements' attributes

Eirini Androutsopoulou1
1
National Technical University of Athens
1
iandroutsopoulou@gmail.com

The methodology presented in this paper is grounded on the analysis and

relational relocation of attributes of the urban body, deriving from the
reconstruction of the urban body as a network configuration. In contrast to the
hierarchical constructions, network constructions allow for multiple connections
between elements, therefore being closer to the complexity of the associative
forces found in the structure of the urban body.Similarity function is applied in an
attempt to restructure those attributes of the urban body which emerge from the
position of each element (node) in relation to other elements of the network and
not from the Cartesian topology. Being able to represent material elements as
nodes, counter-bodies deriving from autopoietic -network functions emerge,
allowing for an inquiry in what concerns the autopoietic features of the urban
body in general, focusing on the application of autopoietic functions which
generate the urban body parts and components and on the multiplicity of
elements' structure, in terms of association of crowds of elements and sets of
attributes' values, aiming at the redefinition of proximity as similarity and of
remoteness as difference.

Keywords: Similarity, Autopoiesis, urban body, Attributes, network, complex

systems

INTRODUCTION the network structure.

Adopting a methodology towards an understanding
of the urban body whose elements include mate-
rial ingredients (nodes) and their inscribed proper-
ties (attributes) through the construction of a net-
work configuration, the autopoietic dynamics of the
urban body emerge, as properties of things -urban ar-
eas inscribed on the Attic peninsula in Greece (Figure
1), revealed through the application of autopoietic -
network functions, measuring qualities which reflect
We call autopoietic those nodes' features which
emerge from the inner structure of the urban body
and are quantified through the application of com-
munity detection (Blondel et al. 2008, Rosvall et
al. 2009), centrality (betweenness centrality, degree
centrality, closeness centrality, Newman 2010) and
other algorithms which capture the network quali-
tative features of the network nodes and reveal the
real-time self-adaptation of the parts and compo-

CITY MODELING | Applications - Volume 2 - eCAADe 34 | 71

 Figure 1
The network
configuration of the
urban body
describing
supra-local nodes
inscribed on the
peninsula of Attica
in Greece.

72 | eCAADe 34 - CITY MODELING | Applications - Volume 2

Figure 2
The re-adaptation
of the network
configuration as
nodes defining the
body parts and
components are
successively
introduced
according to
autopoietic
attributes which are
assigned at the
elements of the
network. The
nodes' colors
correspond to a
certain type of
identity/autopoietic
attribute .
nents of the urban body, on the basis of the succes- processes, reveals the inner structure of the urban
sive introduction of nodes with different autopoietic body, as a reflection of its inherent dynamics and
features (Figure 1). made possible through the examination of the ur-
We focus on the autopoietic properties of the ban body as a network configuration. Following this
network's nodes as they are quantified through the methodology and in order to reveal further relation-
application of certain autopoietic processes which ships between network and urban body structures,
reveal the nodes' interrelations, translating the re- supra-local nodes are regarded as multidimensional
lationships between things into dimensions (at- objects and the attributes emerging from the net-
tributes' values) of elements. Through the quanti- work configuration appear as a set of values (dimen-
fied expressions of the inner dynamics of the network sions), allowing for a clustering methodology of ob-
configurations the autopoietic properties emerge. jects and their attributes, based on similarity stud-
These define the 'species' of the elements accord- ies and made possible through the application of dif-
ing to their inner relations (network attributes in ferent distance and linkage functions (Hastie, Tibshi-
general), the parts of the urban body as articulated rani, Friedman 2003). The application of the distance
fractions/sub-networks (centrality measures), as well and linkage functions results in a data tree where
as the components of the urban body as areas of nodes, as well as their autopoietic attributes, are vi-
stronger interrelated elements (community detec- sualized in a color mosaic which captures the prox-
tion), where the areas' boundary changes according imity of data values to each attribute's lowest and
to the desired threshold of the interrelated strength. highest extreme values and in an hierarchical tree,
The emergence of those features of the urban body where the position of each element reflects its rela-
elements which are inherent in the network struc- tional similarity following a gradual procedure of se-
ture and the generation of the urban body parts and lection of two sets of elements based on the strength
components through the application of autopoietic and the direction of their relation (Figure 2). This ap-

CITY MODELING | Applications - Volume 2 - eCAADe 34 | 73

 proach suggests the transition from a clear percep-
tion of an object/urban site to the ambiguity of multi-
ple properties of things, a set of data values rendered
in the form of scattergrams, color mosaics and den-
dograms. Among the possibilities originating from
this approach is the understanding of the concurrent
alterations taking place in a given network construc-
tion among all node's characteristics with the shift of
one property, a kind of structural coupling between
body and environment (Maturana 2002).
METHOD
The methodology adopted is considered as a
schematic representation of the structure of network
attributes' association and a resulting dimension de-
crease when dealing with the inquiry of distinctive
autopoietic characteristics which aids in the inter-
pretation of the resulting clusters' function. What is
more, it is a way of mapping the multiplicity of ele-
ments' structure, in terms of crowds of elements and
sets of attributes' values which redefines proximity
as similarity and remoteness as difference. Using
similarity function and being able to further explore
interrelations between things through values which
result from given network structures (Figure 2), it is
possible to acquire an overall understanding of the
autopoietic attributes' association, as well as to re-
trieve a cognitive structure which provides the frame
of interrelations of the properties of things.
This allows for the relocation of the ambiguity
of information, shown in a way which enables us
to arrive at an understanding of information in the
grounds of similarity, to look into the aspect of multi-
plicity of things, to draw upon a potential exploration
of proximity in terms of homogeneity and to cap-
ture an ever-evolving field of difference and therefore
identity, with an altered degree of focus into the ac-
cepted divergence of things.
Supra-local nodes are treated as objects whose
attributes' values reflect the different qualities of the
network structure. The application of the distance
and linkage functions results in a data tree where
nodes, as well as their attributes, are visualized in a
74 | eCAADe 34 - CITY MODELING | Applications - Volume 2
color mosaic which captures the proximity of data
values to each attribute's lowest and highest extreme
values.
Through mapping of the gradual procedure of
nodes' union into clusters and being able to trace the
distance function results at each step, the visualiza-
tion of the proximity of nodes in terms of similarity is
made possible, as well as the schematic plan of these
relations at the resulting hierarchical tree, where the
length of the branches reflects the similarity value be-
tween objects (Hastie, Tibshirani, Friedman 2003).
The interrelation of attributes resulting from the
structure of the network configurations allows for the
estimation of possible divergence or convergence of
other attributes in relation to one known attribute. In
the case where we find, for example, that a node's
value has a proximity to a lowest/highest value of
an attribute, in what concerns its relational position
at the network and defined by the application of al-
gorithms which measure certain qualities of the net-
work structure, other qualities could be estimated
beforehand through association. It is therefore pos-
sible, through the application of similarity/linkage
functions and through the production of scatter-
grams showing convergent/divergent relations be-
tween attributes, to shape the structure of network
attributes' association.
The methodology adopted is considered as a
schematic representation of the structure of network
attributes' association and a resulting dimension de-
crease when dealing with the inquiry of distinctive
characteristics which aids in the interpretation of
clusters' function. Using similarity function and be-
ing able to further explore interrelations between
things through values which result from given net-
work structures, it is possible to acquire an overall
understanding of the network attributes' association,
as well as to retrieve a cognitive structure which pro-
vides the frame of interrelations of the properties of
things.
This approach suggests the transition from a
clear perception of an object/urban site to the am-
biguity of multiple properties of things, a set of data
Figure 3
The cluster analysis
resulting from the
application of
similarity function,
visualized as colour
mosaic and
hierarchical
dendrogram of
supra-local nodes
and their
values rendered in the form of scattergrams, color
autopoietic
mosaics and dendograms, or in the form of a network
attributes .
of one or more levels. Among the possibilities origi-
nating from this approach is the witness through vi-
sualization and understanding of the concurrent al-
terations taking place in a given network construc-
tion among all node's characteristics with the shift of
one property, a kind of structural coupling between
body and environment (Maturana 2002).
Based on the results of the applied similari-
ty/linkage functions in what concerns the nodes'
grouping of the supra-local nodes for 3 (minimum
similarity: 0.758-0.653) and 16 resulting clusters
(minimum similarity: 0.994) (Figure 3) and comparing
them to the community detection results (Blondel et
al. 2008, Rosvall et al. 2009) (Figure 5), the nodes'
integration to clusters is further explored, looking at
the issue this time from the angle of analogy and
identity of objects with certain properties and not fo-
cusing merely on the network's structure.
The resulting clusters, comprised of sets of
nodes, when applied at the network visualization, ap-
pear as sparse entities whose location on the network
map doesn't seem to originate at all from topological
network proximity, presenting a strong difference in
results and, of course, in methodology from Blondel
and Rosvall methods. The implementation of similar-
ity results in the network structure shows that prox-
imities of things this time resides in homogeneity and
heterogeneity and not in topological proximity.
The transition from the physical space to the network
structure means the retention of proximity through
network's connections only, exempting any other
analogy or dependency from the objects' Cartesian's
coordinates at the physical space. This second tran-
sition, from network structure to treating network
characteristics as object's properties and conducting
comparative studies, produces clusters whose con-
figurations do not keep a straight link or analogy to
the network's proximity of objects, in spite of the at-
tributes' emergence from those qualities of proximity
between things.
While focusing on higher values of minimum
similarity and therefore producing a higher number
of clusters, things that belong to each cluster get
more similar to each other and the reasons why one
cluster is comprised from certain nodes and not oth-
ers become clearer. However, even in the case of
the production of three clusters, where the value of
minimum similarity is rather high, the ability to track
and comprehend similarity is less viable but remains
reachable.
In the case where the minimum similarity is de-
fined to contain a range of values between 0.758 and
0.653, the concentration of values referring to each
of the three clusters, for pairs of attributes reflecting
network structure, is rather dinstinct, even suggest-
ing an evident qualitative dissociation, as is shown for
the pairs of betweenness centrality - closeness cen-
trality, eigenvector centrality - betweenness central-
ity, eccentricity - closeness centrality.
Being therefore able to track the concentration
of values for each cluster and detect the interrelation
of attributes reflecting network structure, identity is
being traced and redefined, in this case as a set of dis-
tinctive characteristics for each cluster which show a
CITY MODELING | Applications - Volume 2 - eCAADe 34 | 75
 Figure 4
The scattergrams of
pairs of autopoietic
attributes, for each
one of the three
clusters produced
when the minimum
similarity is set
within a range of
values from 0.758 to
0.653.

76 | eCAADe 34 - CITY MODELING | Applications - Volume 2

Figure 5
The application of
the
similarity/linkage
functions results on
the network's
visualization, for 3
(minimum
similarity:
0.758-0.653) and 16
clusters (minimum
similarity: 0.994)
respectively.

Figure 6
Community
detection (24
communities)
based on
modularity class
(Blondel et al. 2008)
and according to
mapequasion
method (Rosvall et
al. 2009). Colouring
of communities
reflects the
differences
between the two
methods.

CITY MODELING | Applications - Volume 2 - eCAADe 34 | 77

 certain amount of distinction for each group of ele-
ments and as a relocation of elements in a map of as-
sociations deriving from network structure.

CONCLUSIONS
Rather than proposing a discussion on overall equal-
ity, the research presented here is a discourse on the
autopoietic properties of things. The methodology
proposed here is a way of mapping the multiplicity
of elements' structure, in terms of crowds of elements
and sets of attributes' values which redefines proxim-
ity as similarity and remoteness as difference. This
allows for the relocation of the ambiguity of infor-
mation, shown in a way which enables us to arrive
at an understanding of autopoietic features of urban
body's elements in the grounds of similarity, to look
into the aspect of multiplicity of things, to draw upon
a potential exploration of proximity in terms of ho-
mogeneity and to capture an ever-evolving field of
difference and therefore identity, with an altered de-
gree of focus into the accepted divergence of things.

REFERENCES
Blondel, VD, Guillaume, JL, Lambiotte, R and Lefebvre, E
2008, 'Fast unfolding of communities in large net-
works', Journal of Statistical Mechanics-Theory and
Experiment, 10, p. P10008
Hastie, T, Tibshirani, R and Friedman, J 2003, The elements
of Statistical Learning, Springer-Verlag, New York
Hillier, B and Vaughan, L 2007, 'The city as one thing',
Progress in Planning, 67, pp. 205 - 230
Maturana, H 2002, 'Autopoiesis, Structural Coupling and
Cognition', Cybernetics & Human Knowing, 9(3-4), pp.
5-34
Maturana, H and Varela, F 1980, Autopoiesis and Cogni-
tion, The realization of the living, D.Reidel Publishing
Company, Holland
Newman, MEG 2010, Networks: An Introduction, Oxford
University Press, Oxford New York
Rosvall, M, Axelsson, D and Bergstrom, CT 2009, 'The
map equation', Eur. Phys. J. Special Topics, 178, pp.
13-23

78 | eCAADe 34 - CITY MODELING | Applications - Volume 2

Understanding the Role of Spatial Connectivity in
Integrating Informal Settlements, through the Case of
Medellin's Urban Cable-cars.
Paul Goodship1
1
The Bartlett School Of Architecture, University College London
1
ucftpgg@live.ucl.ac.uk

Throughout Latin America urban cable-cars are fast becoming a normal sight
with urban transport systems, taking residents and tourists to and from previously
isolated locations and providing a new form of accessibility that was not possible
before. As its popularity grows, it is important to understand the role that
improved spatial connectivity plays in transforming these communities and
discuss how this can be best measured. This is because the spatial conditions of
informal settlements are often forgotten about when planning large scale
upgrading programs. Therefore, the aim of this article is to explore ways to
interpret the role of spatial connectivity in the transformation process of informal
settlements, through the case of Medellin's Urban Cable-car and the exploration
of 'speed' as a measurable variable.

Keywords: Spatial Connectivity, Space Syntax, Speed, Urban Cable-cars

MEDELLIN AND THE URBAN CABLE-CAR
Medellin, the second largest city in Colombia with
a population of around 2.5 million, has recently be-
come one of the most talked about cities in the world,
due to its rapid transformation from a violent, drug
cartel controlled city to a city that aims to be more
socially inclusive. This transformation has been en-
capsulated by new forms of urbanism that aim to en-
gage the poor and a municipality that inspires to in-
tegrate all of its citizens into one city. When Sergio
Fajardo was elected Mayor in 2004, one of his main
goals was to create an 'equal city for all and were all
citizens can construct relations stimulated by a city
rich in services, culture and public space' (Brand and
Davila 2013). This ambition developed into a new ap-
proach to implementing urban projects combining
simultaneously physical transformations, social pro-
grams and participation. To do this the term 'Social
Urbanism' was used, which was 'a metaphor for an
integrated approach to transport and urban develop-
ment, and for the power of the strategic potential of
this integrated approach to address urban inequal-
ity' (Levy 2013). Yet, whilst it was only a metaphor,
it helped to enable a large-scale urban upgrading
project for the whole city, with a focus on the urban
poor.
The most important urban project to develop
from this was 'Proyecto Urbano Integral' (Integral Ur-
ban Project) or PUI. It was, and still is, an urban up-
grading project created and administrated by the
municipality with the aim of increasing the quality of
life of inhabitants, focusing on areas where poverty

CITY MODELING | Applications - Volume 2 - eCAADe 34 | 79

 and violence is most visible. Its strategy focuses on
an integral and comprehensive approach, providing
each project with three main components - a physi-
cal, social and institutional component. This can be
best observed in PUI Noriental (North-east), which
oversaw the construction of large community facil-
ities such as a Library-Park and Business Develop-
ment Centre, 15 new or upgraded public spaces and
streets, 3 new bridges offering connections between
local neighbourhoods, new housing including the
upgrading and consolidation of existing homes, a
wide range of community meetings, workshops and
events, a series of social and participatory programs
(Calderon 2009). One of the essential elements to
its success was the newly opened urban cable-car
(figure 1), which connected the previously isolated
neighbourhoods.
Yet it became more than just a simple transport con-
nection, it enabled the upgrading projects to be-
come a part of the citywide transformation process
and opened up the neighbourhood (and the reali-
ties within) to all citizens of Medellin. Intriguingly the
cable-car was never conceived as a part of the overall
upgrading process, nor was it implemented by Ser-
gio Fajardo the mastermind of 'Social Urbanism'. It
was instigated by Luis Perez, the previous mayor, and
was initially intended to be a mechanism to trans-
port workers in the north of the city to the factories in
the south, via the heavily underused existing metro
80 | eCAADe 34 - CITY MODELING | Applications - Volume 2
train (Brand 2013a). Soon after it opened in 2004 it
became clear to many that this could became the
focal point for the PUI Noriental, with interventions
seemingly adjoining it. As 'in the four years follow-
ing its introduction, the city invested seven times the
cost of the cable-car system in complementary urban
projects', harnessing its potential to not only spatial
connect the poor, but also socially and economically
improve lives (Brand and Davila 2013). This saw major
changes occurring in some of the most violent and
poor neighbourhoods and with it brought much in-
ternational acclaim.
This predictably led to a second cable-car line in
2007, connecting hillside settlements in the west of
the city to the main metro train line. However, whilst
the first line ended up being supported by compli-
mentary urban projects, this was not the case with
the second line, where a lack of investment in sec-
ondary projects meant there was an over reliance Figure 1
on its transportation value. Peter Brand, a local aca- Urban Cable-car in
demic, points out "it should not be assumed that Medellin, Colombia.
aerial cable-car systems [alone] will lead directly to Linea K open in
the integration of poor neighbourhoods with the rest 2004.
of the city" (Brand 2013b). This is clear when com-
paring Medellin's cable-car lines as 'what has hap-
pened around the K and J lines makes it abundantly
clear that integrated, localised, parallel and coordi-
nated strategies of intervention generate far greater
benefits that the sum of independent, diffuse and se-
quential projects'. Yet, even with this realisation that
cable-cars alone can not automatically resolve prob-
lems of urban poverty, two new lines are currently
being built in the east of the city and another line is
proposed in the northwest.
This symbol of Medellin's transformation and its
initial success led to its replication in other cities
across Latin America, including Caracas, Rio de Jane-
rio and La Paz, and is currently being proposed in
Quito and Lima. Yet, whilst this new form of trans-
portation works with the existing urban grid to cre-
ate a solution that prevents the demolition of homes
and local businesses, there are still many uncertain-
ties, including the role its spatial connections play in
the whole transformation process.
THE SPATIAL NETWORK
Within informal settlements, 'there are certain pa-
rameters and elements which, are increasingly
present and should be part of any serious attempt
to address issues of poverty' (Serge 2009). Such pa-
rameters maybe understood as spatial and can be ex-
plained within Bill Hillier's paper 'A theory of the city
as object', where he explains 'a local process gener-
ates differences in local grid patterns and apparently
reflects differences in spatial culture; and the other
a global process generates a single overriding struc-
ture that seems to reflect a more generic or universal
process'. Highlighting that whilst socio-components
create the underlying pattern of differences, there is
a set of 'autonomous spatial laws that are governing
the affects on spatial configuration' and it is these
'invariants' that allow for greater spatial integration
(Hillier 2002). These principles can be observed in
informal settlements, as often the lack of global in-
tegration leads to isolation, encouraging local socio-
components to dominant, making them turn their
back on the city.
These spatial laws are further explored in Hillier
et al. (2000), extensive study of 17 informal set-
tlements in Santiago, Chile. Were pedestrian and
vehicular patterns of movement and land uses was
studied alongside typical spatial configuration mod-
els, to critical assess the spatial aspects that lead to
consolidation. This established the settlements that
were better equip to develop 'edge orientated com-
mercial activity, can participate in the wider local
economy' and would therefore consolidate quicker,
whereas settlements that have no, or little, edge-
activity consolidate a lot slower. (Hillier et al. 2000.)
Karimi et al. again explore these issues in the pa-
per, 'Evidence-Based Spatial Intervention for Regen-
eration of Informal Settlements'. This examines how
limited physical interventions can facilitate the spa-
tial integration of informal settlements, with the case
of Jeddah. It attempts to demonstrate that with a
deeper understanding of the elements and invariants
that consolidate informality, the upgrading process
can be greater improved. This led to an analysis-
driven, evidence-based proposal that analyses the
different spatial and socioeconomic factors, using a
multi-criterion GIS method to form a composite in-
dexing system, which is then used as a guideline for
design decisions. (Karimi. 2007.)
These types of analysis use Space Syntax to ex-
plain and predict the spatial networks of informal set-
tlements. Space Syntax has a long established con-
nection to urban theory that can 'link physical as-
pects of the urban system with its functional, so-
cial and behavioural aspects, directly and seamlessly'
(Karimi et al. 2013).
However, as Vinicius Netto points out 'few so-
ciospatial approaches have been quite so polemi-
cal and stirred quite so many different responses
from scholars of the city as space syntax'. So whilst
Space Syntax has established itself as a both a reli-
able methodology and a well respected urban the-
ory, there are still many detractors. Netto explains
that many cite its epistemological limitations and 'ac-
cuse the theory of excessive formalization and geo-
metric reduction, questioning its capacity to repre-
sent the city, its morphology and transformations,
and the partial way in which it captures the complex-
ity of the social world'. This suggests that Space Syn-
tax can be seen to create a world that is almost exclu-
sively pure geometry and structure, to allow its anal-
ysis to focus on configuration and hierarchies. Then
another commentator Stephen Read points to two
commonly stated weak features, 'one is its tendency
to treat the urban object as a thing bounded by the
limits of the densely built fabric of the centre; the
other, to treat all movement spaces equally when it is
quite clear that different classes of physical space in
the fabric of the city perform quite differently at the
levels of urban speed and function and at the level
of the human experience of space and time' (Read
2005). This demonstrates that Space Syntax can be
seen to negate the way that urban mobilities and ev-
eryday dynamics affect the space of the city, espe-
cially at the level of the individual (Read 2005).
CITY MODELING | Applications - Volume 2 - eCAADe 34 | 81
 When we consider how we connect between the
spaces, walking is of the most basic form of human
movement. However, people do not simply over-
come distances, 'they move within cities with differ-
ent objectives, such as to do shopping, to socialize,
to go to some particular space or only to wander.
Everyone moves in his own way, with their unique
mental and physics distinctive peculiarities and mo-
tivations. However, human behavior is based on a
number of shared decision-making parameters and
spatial accessibility restrictions imposed by the en-
vironment where it occurs.' Yet, it can be said that
this dynamism and individuality is generalised within
Space Syntax with the analysis instead the focusing
on the larger object (Zampieri et al. 2009). If we do
not consider movement in all its dimensions we are
simplifying the reality by 'reducing the way we un-
derstand the pedestrians behavior, disregarding im-
portant attributes like: ambiance; socialisation ele-
ments; health and rest; and aesthetic' (Zampieri et al.
2009). With this we start to understand the complexi-
ties that surround the simple act of walking from one
space to another and whilst Space Syntax map these
conditions depending on connectivity, for the indi-
vidual there are many local factors that influence the
simple act of walking between two points.
Stephen Law et al. (2012) point out 'in an age
where citizens are connected in an endless city of
intertwining social, transport and communications
networks, there is a need for Space Syntax analy-
sis to look beyond the street network in exploring
how different networks impact upon today's soci-
ety'. Thus arguing that 'the decrease in travel time
through faster modes of transport, allows for a fur-
ther accumulation of towns and the emergence of
the metropolis. Underground stations emerge as
multi-modal hubs with high volumes of pedestrian
activity, land use intensity and specialization. They
play an important role for the spatial economy and
shift town centres from the original location'. (Law
et al. 2012.) To prove this Law et al. (2012) cre-
ated a series of experiments to demonstrate how cer-
tain neighbourhoods in London, located next to an
82 | eCAADe 34 - CITY MODELING | Applications - Volume 2
underground stations, where better integrated. This
is not attempting to reinvent Space Syntax, but in-
stead build upon what is successful and push for a
for stronger integration between the traditional con-
figurational analysis and geographical analysis from
other sources, to create more dynamic analysis. (Law
et al. 2012.)
For many critics Space Syntax is 'not fulfilling
its potential as a manner of thinking the dynam-
ical forms of the contemporary city' (Read 2005).
However, both Read and Netto clearly conclude that
Space Syntax in its ambition is a way to understand
complex effects, on the horizontally distributed so-
cial body of the city and its physical infrastructural
movement. It ability to focus on and fully understand
one key element of city - space - and connecting it to
various socio-political and economic conditions has
made it a very power theory in recent times. There-
fore it should be understood that whilst its neutrality
in measuring space maybe seen as a weakness, it is
also its greatest asset (Netto 2015).
THE OBJECTIVE
To understand the role spatial connectivity plays
in the transformation process of an informal settle-
ment, it is important to understand what is meant
by spatial connectivity. The typical Space Syntax ap-
proach measures connection points, which can then
be associated with certain local conditions. How-
ever, the way each individual moves between these
points varies dramatically. In order to further un-
derstand these connections, this article explores the
worthiness of weighting the dynamic measurement
of 'speed' to the traditional connectivity measure-
ments of Space Syntax. At its purest form speed rep-
resents the simplest method for measuring connec-
tions between of different parts of the city and can
represent multi forms of movement, such as public
transport. This is also an important measurement
when calculating the impact of walking in areas of ex-
treme topologies, of which Medellin is located, allow-
ing for a clearer understanding of how topology may
affect spatial integration. It is for these reasons this
article explores how 'speed' may be combined with
typical Space Syntax analysis to provide a clearer pic-
ture of spatial connectivity.
Therefore this article examines the influence that
extreme topologies and unhindered transport sys-
tems have on the connectivity of the city, through
the case of Medellin and its cable-car systems. This
city provides a topology that is extreme with its steep
valley edges, where many informal settlements are
located and offers an unique form of transport that
floats over homes, providing a non-standard form of
spatial connectivity. The aim is to further our under-
standing of the role space plays in connecting neigh-
bourhood and in doing so how different forms of
connections influence the ways we move. This is not
aiming to create a new methodology or contradict
past ones, but instead explore and discuss how the
dynamism of space can be better understood to al-
low the varying movement of the individual to be in-
cluded in the spatial analysis of a city.
DATA AND METHOD
To achieve this an evidence-based methodology is
used, based upon the methods and principles of
Space Syntax and is formed of three parts.
The first part is a typical Space Syntax analysis
of the city, including the transport connections. It
uses the two most common measurements, 'integra-
tion' and 'choice', which is often referred to as 'to'
and 'through' spaces, respectively. These are typi-
cally used to measure the spatial impact of new in-
terventions.
After the results of the standard measurements
are discussed, the article explores how these can be
weighted against the measurement of speed. To do
this an average 'speed' is calculated for both walk-
ing and public transport. At this point it should be
made clear that this does not include micro-buses, as
whilst these are the most popular form of transport
in Medellin they are very unreliable and difficult to
track, since there are no regular pick-up points, routes
vary between drivers and speed differ hugely. There-
fore they are omitted for the ease of developing ini-
tial principles. Using the same model created for the
Space Syntax analysis, an average speed is calculated
for each segment. First the public transport speed is
simply taken from the average speeds provided on-
line (www.metrodemedellin.gov.co) and verified on-
site. Then secondly, walking speed is calculated us-
ing the 'Tobler's Hiking Function'. This takes into ac-
count steepness, the length of walk and your average
speed on a flat surface (5km/hour) to calculate the av-
erage time needed to walk the route, which can then
be converted into an average speed. This is done for
every segment of the spatial model, by converting a
contour map into an elevational raster image (ie each
pixel has an elevational value), which allows the start
and finish point of each segment to have an eleva-
tional value, then the 'Tobler's Hiking Function' can
calculate the time needed to walk each route (in both
directions). This creates a map of the city showing the
speed values of each segment. Thus, the segments
of the city that are located on the flat have the fastest
speed and the segment located on the steep slopes
the slowest. This map is then weighted against the
Space Syntax results at varies scales, to examine if
there is any improvement in the spatial connectivity
results.
Tobler's Hiking Function = time (hours) to cross 1
meter = 0.000166666 * ( exp ( 3.5 * ( abs ( tan ( radians
( slope_deg )) +0.05 ))))
The final part uses Space Syntax observations
methods to measure pedestrian movement at a local
scale, using the technique of gate counting. This sim-
ply counts the number of people moving in a certain
direction and at different times of the day, through
an imaginary gate. This is used to form the basis for
testing spatial analysis results.
These methods allow the results be examined
against each other, in order to develop a discussion
about the worthiness of the speed of movement be-
ing a part of the analysis of cities like Medellin and
allowing a greater understanding of the impact of dy-
namic forms of spatial connections. The aim is not to
provide a full-scale comprehensive model but to test
the feasibility of the methodology, to enable future
exploration.
CITY MODELING | Applications - Volume 2 - eCAADe 34 | 83
 RESULTS
When the city is modeled using standard Space Syn-
tax, the results are as expected and have been dis-
cussed in previous publications (Goodship 2015).
These show the key global integration zones concen-
trating around the city centre and to the west, reflect-
ing the main central areas where most people con-
gregate. However, the areas to the north and east of
the city centre, are highlighted as having high inte-
gration, yet from local knowledge there are not cen-
tral areas for the city and lie on very steep slopes.
At a more local scale the integration measurement
does a reasonably good job in locating the main ar-
eas of integration, or 'to' space, demonstrating how
the cable-car has helped to improve the integration
of neighbourhoods. This is most potent for the first
cable-car line, whereas with the second the interga-
tional impact is minimal. It is therefore reasonable to
say with this case that integration is accurate when
examining local scales, but less so at a global scale.
When this analysis looks at the measurement of
'Choice', at a global scale all of the main 'through'
spaces are picked up, such as highways and bus
routes. Yet, unlike 'Integration', when the cable-
car is introduced there is very little, if any, affect
on the overall network. This is the result of a well-
established network not being affected by the intro-
duction of two new lines. When these measurements
are examined more closely in the individual neigh-
bourhoods, it becomes apparent that the analysis
does not pick up elements that may hinder move-
ment, such as steepness and road conditions, since
some of the important routes indicated are difficult
to navigate on foot. Therefore what we see here with
'Choice' is that unlike 'Integration' the impact of local
factors affects the results at a neighbourhood scale.
These failings become apparent when looking at
the topology of Medellin, as this unmistakably shows
the city located in a very steep valley, with much of
the urban fabric situated on the steep, hence the ne-
cessity for a cable-car. It is these local conditions that
a 2d analysis is unable to pick (figure 2). This has re-
sulted in the city-scale analysis of 'to' spaces and the
84 | eCAADe 34 - CITY MODELING | Applications - Volume 2
'local' scale 'through' spaces being inaccurate.
Figure 2
City contour map
laid on top of
Normalised
Integration (NAIN)
Map.
Figure 3
Speed segment
map of Medellin,
with central red
lines representing
public transport
due to its superior
speed.
The next part of the analysis examines the speed of
movement, by investigating the results of the speed
map that represents the average speed of each seg-
ment within the city. What is instantly clear from this
is the areas located on steep slopes have the slowest
speed and the area on the flat the fastest and whilst
this is expected it does confirm that the methodology
works (figure 3). Also, in the areas with the slowest
speed is the largest concentration of urban poverty
and almost matches a map of city's informal settle-
ments. Therefore, it is clear that this map represents
a simple and accurate image of movement speeds
across the city. When it is examined more closely
around the local neighbourhoods, the roads that run
perpendicular to the cable-car along flatter and faster
areas, often form high-streets, with a hive of commer-
 cial and pedestrian activities. Whereas the roads that
run down the steep slopes have less commercial ac-
tivities, but a variety of irregular and informal shops.
Whilst the speed map shows some basic urban
information about the city, it would be difficult to say
this demonstrates anything that is not obvious to the
general observer. Yet, it represents an opportunity to
weight these dynamic results against more standard
space syntax measurements, in an effort to create a
more accurate image of the city. This is done using
depthmap (Varoudis 2012), a Space Syntax analysis
software, and using the standard weighting function
within when analysing the system.
When this analysis is run to include the weighted
value of speed for 'Integration' there is a change in
the global scale map, with the hotspot in the city
centre appearing to be smaller with less integration
to the east and north of the city. This suggests
that 'speed' provides a more accurate account of the
whole city (figure 4). At the local scale the integra-
tion values remaining roughly the same. When this
is used with 'Choice', there is very little impact to
Figure 4
Normalised
Integration (NAIN)
weighted 'without'
(left) and 'with'
(right) speed.
the overall network of 'through' spaces at the global
scale, however when the results are looked more
closely at walkable scales, the 'through' space net-
work becomes clearer. Hence, at scales between
1250 and 3000 the main walking routes become
more apparent, demonstrating the steepness of the
topology is having an impact on the results (figure 5).
In these results, the two main forms of analysis
in Space Syntax are affected differently. With 'Inte-
graion' it is at the global scale that speed is most af-
fective, allowing the central hotspots to be more ac-
curate, while at the local scale there is very little dif-
ference. Whereas with 'Choice' the global impact of
speed is minimal, yet at the local scale its introduction
creates a clearer representation of 'through' space.
Therefore, the introduction of speed as a weighted el-
ement starts to demonstrate how a city like Medellin
can be represented more accurately in spatial analy-
sis and opens up the possibility for dynamic measure-
ments.
CITY MODELING | Applications - Volume 2 - eCAADe 34 | 85

With this said, it is important to test these results
against the actual movement patterns obtained by
counting the movement of people around the cable-
car stations. These observations show that the ma-
jority of movement is perpendicular to each station,
with people move to and from the station. At the
stations along the first cable-car line, movement is
mostly perpendicular along the flattish main roads,
which are generally the main 'through' spaces and
are generally highs streets and areas of high pedes-
trian activity and commercial land-uses. The ob-
served movement patterns are less clear around the
second cable-car, since the stations are located on
less well-established urban grids, making the rela-
tionship with movement hard to observe. These ob-
servations of both lines show that where there are ex-
tremely steep paths, there is less movement and with
this less commerce and pedestrian activities.
These results start to demonstrate the role of
speed in understanding the spatial connections of
a non-standard city and how speed as a measure-
ment can help us interpret the role cable-cars have
86 | eCAADe 34 - CITY MODELING | Applications - Volume 2
played in transforming informal settlements within Figure 5
Medellin. Local choice results
surrounding the
first cable-car line.
FINDINGS 'Without' (left) and
In order to test the results, certain correlations be-
'With' (right) speed
tween the varied results are needed. In Space Syntax
calculations.
a standard method to test the intelligibility of a city
is to measure the correlation between 'Connectivity'
and 'Global Integration', the higher the correlation
the more intelligible. This provides a method to test
the speed weighting. When this correlation is tested
with the standard space syntax spatial model (not
weighted with speed) the results is 0.11r² for stan-
dard connectivity and 0.05r² for angular connectivity.
When the same intelligibility test is measured against
the model weighted with speed, the correlation is
slightly higher, at 0.15r² for standard connectivity and
0.07r² for angular connectivity, suggesting this pro-
vides a slightly higher level of intelligibility (figure
6). However, both intelligibility results are very low.
Then potentially, the most relevant correlation test is
the connection between the results of 'Choice' and
onsite movement surveys. This showed a reasonable
correlation between movement and 'Choice' with-
out speed, at a local scale above 5000m and when
the same test is performed with the speed-weighted
model, there is very little difference (figure 7).
Figure 6
Intelligibility test
'without' and 'with'
speed.
From these statistical values it would suggests that
nothing conclusive has resulted from the introduc-
tion of speed. These failings could be a result of many
individual segments creating a conflict between pos-
itive 'speed' and a negative 'integration' values. How-
ever, it is hard to ignore the visual changes to the
overall map of the city when speed is introduced and
by intuitively understanding the city, these maps ap-
 pear to reflect a clearly picture of 'to' and 'though'
spaces. So whilst the standard methods for test-
ing correlations do not add up, the visual results are
enough to suggest that this weighted measurement
could be of value.
Figure 7
Correlation with
movement and
speed weighted
maps.
Whilst these results show the spatial maps represent-
ing a closer image of the city, the statistical results do
not correspond. Therefore, it is worthwhile reflecting
on some of the elements that need refined for im-
proving accuracy.
The only transport system to be included in the
speed analysis was the metro-train and cable-car,
which are fully integrated, yet, the most popular form
of public transport in Medellin is the micro-buses.
Therefore, whilst this form of transport is hard to map,
it is still important to include this in some form in any
further studies. It is also worth noting that the speed
of the public transport represented at average speed
for the whole line (600 metres a minute for the metro
train and 300 metres a minute for the cable-car), but
this does not include the time it take to get on and off
or the waiting times, whereas walking speeds were
calculated exactly. For consistency, these exact val-
ues should be added into the speed map. Another
potential failing is the movement survey. The speed
weighted maps show clearly walking paths at a local
scale and by tested this against the onsite observa-
tions these results can be verified, yet in this article
the movement survey was limited to 16 main junc-
tions within 500m radius of the cable-car station (due
to limited resources). So to improve accuracy, the
movement survey should cover a larger area.
Whilst 'speed' improved the visual representa-
tion of the city's overall integration, there was a high
level of disparity with the statistical values, indicating
that the roughness of the values are hidden at a large
scale. One of the main reasons for this is the large
number of high 'speed' values found on small seg-
ments in poorly integrated zones, sharply contrasting
the low integration values. Also, the use of a stan-
dard intelligibility test should be adjusted to repre-
sent 'speed' more accurately, since the standard con-
nectivity value used is not altered for speed. In gen-
eral the methods and equations used to formulate
the spatial connectivity of the city needs to be fully
evaluated, to determine if they are fully appropriate
for this type of analysis, since the two types of results
often conflict.
It is clear from the visual results that the
weighted measurement of 'speed' has merits and
could provide a better opportunity to understand the
spatial impact new urban connections. This is espe-
cially important when using a methodology that is
designed for 2-dimensions, as by using a standard
variable like speed, the 3-dimensional elements of
landscape and dynamic connections of transport can
be better incorporated into the analysis process.
CONCLUSIONS
As this research strives to understand the impact of
urban cable-cars, it is important to understand and
evaluate the role that space has played in reconnect-
ing isolated areas. One of the most reliable methods
for doing this is Space Syntax. It has an established
and trusted association with measuring the impact of
spatial interventions and in the past has been used to
discuss the transformation process of informal settle-
ments. Yet with the case of Medellin, its steep slop-
ing valleys force people to move differently, making
the use of Space Syntax questionable, since it does
not traditionally pick up these local factors due to its
analysis being grounded in a 2-dimensional method.
It is for these reasons this article has explored
the worthiness of 'speed' in spatial analysis. It pro-
vides a pure value for the connection between dif-
ferent spaces and can represent the way individuals
move along routes, whether steep or flat and accu-
rately represents different forms of transport. This ar-
ticle demonstrates an accurate method for calculat-
ing the walking speeds and discusses how its weight-
ing with a standard Space Syntax model can start
CITY MODELING | Applications - Volume 2 - eCAADe 34 | 87
 to visually represent spatial connectivity more accu-
rately. Yet, whilst it has to be acknowledged that this
does not fully prove that 'speed' is a valuable addi-
tion, it does suggest how it maybe included in the
future, providing a potential dynamic variable that
can be altered for the individual. Therefore, by ac-
cepting that space is something more dynamic that
simple linear connections, this provides an opportu-
nity to explore how it can be perceived in different
locations and through different mechanism that con-
nect us from one space to another. This provides the
opportunity to better understand the transformation
process of complex urban fabrics, like Medellin's.
REFERENCES
Brand, P 2013a, 'The Social Significance of Mobility',
in Davila, J (eds) 2013a, Urban Mobility & Poverty.
Lessons from Medellin and Soacha, Colombia, UCL,
London, pp. 16-22
Brand, P 2013b, 'Political Resonances of the Metrocable',
in Davila, J (eds) 2013b, Urban Mobility & Poverty.
Lessons from Medellin and Soacha, Colombia, UCL,
London, pp. 115-120
Brand, P and Davila, J 2013, 'Metrocables and 'Social Ur-
banism' Two Complementary Strategies', in Davila,
J (eds) 2013, Urban Mobility & Poverty. Lesson from
Medellin and Soacha, Colombia, UCL, London, pp.
46-54
Calderon, C 2009, Learning from Slum Upgrading and Par-
ticipation. A Case Study of Participatory Slum Upgrad-
ing in the Emergence of New Governance in the City
of Medellin, Colombia., VDM Verlag Dr. Muller, Saar-
bruken, Germany
Davila, J (eds) 2013, Urban Mobility and Poverty: Lessons
from Medellin and Soacha, Colombia., UCL, London
Goodship, P 2015 'The Impact of an Urban Cable-car
Transport System on the Spatial Configuration of
and Informal Settlement.', Proceedings of the 10th
International Space Syntax Symposium, London, pp.
131:1-131:17
Hillier, B 2002, 'A theory of the city as object: or, how
spatial laws mediate the social construction of urban
space.', Urban design international., 7(3-4), pp. 153-
179
Hillier, B, Greene, M and Desyllas, J 2000, 'Self-generated
Neighbourhoods: the role of urban form in the con-
solidation of informal settlements.', Urban design in-
ternational., 5(2), pp. 61-96
88 | eCAADe 34 - CITY MODELING | Applications - Volume 2
Karimi, K, Amir, A and Shaiei, K 2007 'Evidence-based
spatial intervention for regeneration of informal set-
tlements: the case of jeddah central unplanned ar-
eas.', Proceedings, 6th International Space Syntax Sym-
posium., Istanbul., pp. 034(01)-034(14)
Karimi, K, Parham, E, Friedrich, E and Ferguson, P 2013
'Origin-Destination Weighted Choice Model as a
New Tool for Assessing the Impact of New Urban
Developments.', Proceedings of Ninth International
Space Syntax Symposium., Seoul, South Korea, pp.
055:1-055:13
Law, S, Chiaradia, A and Schwander, C 2012 'Towards A
Multi-Modal Space Syntax Analysis. A Case Study
Of The London Street And Underground Network.',
Proceedings: Eighth International Space Syntax Sym-
posium., Santiago de Chile., pp. 8021:1-8021:20
Levy, C 2013, 'Transport, Diversity, and the Socially Just
City: The Significance of Gender Relations', in Davila,
J (eds) 2013, Urban Mobility & Poverty. Lessons from
Medellin and Soacha, Colombia., UCL, London, pp.
23-29
Netto, VM 2015, '“What is space syntax not?” Reflections
on space syntax as sociospatial theory.', Urban De-
sign International., Vol. 21, 1., pp. 25-40
Read, S 2005 'Flat city: a space syntax derived urban
movement network model.', Proceedings of the 5th
Space Syntax Symposium., Amsterdam
Serge, R 2010, 'Formal-Informal Connections in the Fave-
las of Rio de Janeiro: The Favela-Bairro Programme.',
in Hernandez, F, Kellett, P and Allen, LK (eds) 2010,
Rethinking the Informal City: Critical Perspectives from
Latin America., Berghahn Books, pp. 163-179
Zampieri, FL, Rigatti, D and Ugalde, C 2009 'Evaluated
model of pedestrian movement based on space
syntax, performance measures and artificial neural
nets.', Proceedings of the 7th international space syn-
tax symposium., Stockholm, pp. 135:1-135:8
Visualizing Urban Sports Movement
Ozgun Balaban1 , Bige Tuncer2
1,2
Singapore University of Technology and Design; ETH Zurich, Future Cities Lab-
oratory, Singapore-ETH Center (SEC), Singapore
1
balaban_ozgun@mymail.sutd.edu.sg 2 bige_tuncer@sutd.edu.sg

In this study, a visualization tool that maps outdoor physical activity such as runs
on a map by specifying time, location, activity, gender, age group, etc. is created.
This tool reveals the usage patterns of streets within a city for outdoor physical
activity. This tool is created within a larger research project that investigates the
influence of streets on the leisure walking activity within cities. For this purpose,
the tool is capable of presenting the collected multi-modal data that includes
personal fitness data, weather data, spatial data, and crime data. Moreover, the
tool creates new analysis capabilities such as displaying usage of streets by urban
joggers. The research project in which this tool will be used is aimed for
designers/planners to improve streets for 'runnability'.

Keywords: Sports Activity, Big Data, Urban Visualization, Fitness Applications

INTRODUCTION be done using several different techniques such as

Urban planning and design requires detailed anal-
ysis that relies also on collections of data concern-
ing the site and various issues such as utilization of
urban spaces, accessibility, sustainability, etc. Often
this data comes from traditional sources such as sur-
veys or interviews (Reades et al 2007). Surveys and
interviews are labor intensive tasks, as it is costly to
collect data personally. Moreover, these are done a
limited number of times, and it is hard to update the
data. Additionally, for running and biking, as these
involve time and path data, people may not remem-
ber the exact time and location of their activities.
There are other methods that are available for an
urban planner especially when planning for leisure.
Behavior observation is observations of people using
urban spaces. It was the important tool for Whyte's
study of utilization of plazas in New York (Whyte and
Underhill 2009). For sports it includes recording of
people using parks or other spaces for sports. It can
taking videos and photos, or watching and recording
of users manually. This provides detailed information
about how residents use urban spaces, such as parks
for running. However, this is resource intensive work,
requires workforce and time for recording.
With the rising popularity of mobile phones and
applications running on 'smart devices', there is an-
other possible method of collecting data about how
residents use cities. Especially, fitness applications
which keep track of location of their users' physical
activities are valuable in understanding sports par-
ticipation within a city. As these applications record
users location and time of activities, it is possible to
know a person's usage patterns of spaces during a
sports activity. Furthermore, by acquiring data from
more than one user, it is possible to record location
specific queries such as who is using a park for run-
ning and where they start their runs, which will in
turn indicate a practical catchment area of a park.

CITY MODELING | Applications - Volume 2 - eCAADe 34 | 89

 The main aim of this research, therefore, is to use
personal physical activity data generated from fitness
applications to visualize how an urban setting is used
for running within a selected neighborhood. The
questions to be answered in this study are:
• How to use fitness application data to indicate
running behavior of residents of a neighbor-
hood?
• How to improve the data collection methods
mentioned above with the help of this new
data?
• How to analyze the influence of streets to ur-
ban running activity with the help of imple-
mented tool?
BACKGROUND
With the recent developments of sensors and their
availability for the masses, it is possible to study be-
haviors of urban residents. Tools that are available
for planners, designers and policy makers create a
new notion of cities, which is generally referred to
as "smart cities". Smart cities are the cities that can
adapt to the behavior of their residents and in a way
their citizens are constantly involved in the design
and planning processes without even realizing it. Tra-
ditional way of understanding public utilization of ur-
ban spaces is having some questionnaires and polls.
However, these methods are costly and often cannot
provide the whole picture.
Use of traces of urban residents, left by the sen-
sors they carry with them while using the urban
space (smart phones) is a recent way of acquiring
knowledge instead of doing surveys and polls. One
example is Mobile Landscapes project, (Ratti et al
2006) which uses mobile phone location data to un-
derstand city residents' movement and their activi-
ties in time. Another example comes from Lee and
Kam's work (2014) on the study of Singapore com-
mute patterns using the data from smart cards used
for train commutes. This provides an overview of
travel patterns during the day which can be used
to plan new lines to relieve congestion during peak
hours.
90 | eCAADe 34 - CITY MODELING | Applications - Volume 2
Visualizations of fitness data collected from different
fitness applications are popular within data visualiza-
tion communities. These generate artistic visualiza-
tions that show the frequency of runs within cities
[1,2]. Strava, one of the fitness application compa-
nies, has even created an interface to show heat maps
of runs in a city [3]. While these visualizations provide
an overall understanding of which paths are popular
within a city, they lack filtering capabilities and are
limited in effectively informing urban designers and
planners.
Collection of fitness data and its usage in urban
planning is discussed in several studies. Clarke and
Steele (2011) suggest combining fitness data with so-
cial media data and discuss how the data should be
collected. Cortes et al take (2014) Endomondo as an
exemplary fitness application, collect the data and
provide some statistical data about overall usage of
the application. Both of these studies focus more on
collection of the data and they suggest how to use
this data for urban planning. However, none of these
are implemented.
Open Data approach, which means sharing of
different kinds of data with the public by organiza-
tions or governments, leads to new possibilities in ur-
ban planning. Availability of data for planners makes
it easier to use this data as sometimes collection of
this data might be tedious. As there are data cov-
ering different aspects, it is possible to bring them
together to form multi-modal data. An example of
multi-modal data is collecting crime data, running
data, and GIS data such as street width in a database
and correlate street choices of the users to get in-
formed about perception of crime. Availability of
many different sensors makes multi-modal approach
possible.
TECHNICAL OUTLINE
In this study, multi-modal data approach is used to
collect different data from different sources. The data
collected is weather data, personal fitness data, spa-
tial data regarding streets of a city such as width of
a street, width of a pedestrian walk, etc. But the pri-
Figure 1
An Endomondo
workout accessed
from its own web
interface.
mary mode of the data comes from fitness applica-
tions, the other data sources complement fitness ap-
plications,
There are many different personal fitness appli-
cations on the market. While their interfaces dif-
fer, what they do is basically same: that is to record
users' location within a timeframe and report it back.
The most important personal fitness applications are
Google Fit, Strava, Endomondo, Nike+, and Runk-
eeper. For our study, initially we choose Endomondo
as data source. The reason for this selection is that,
in Endomondo every workout is presented in a web-
page which is listed sequentially which makes data
collection easier.
Table 1
An example of a
workout data.
Endomondo is one of the most used personal fitness
applications. It has over 20 million users worldwide
[4]. In Endomondo every user has a unique ID and
every workout such as a run is stored with a workout
ID as presented in Table 1. In every workout there
are certain parameters related to that workout that
is stored, these are workout ID, type of the sport,
the date and time of the workout, distance covered,
duration of the workout, calories spent during the
workout, average speed, minimum and maximum el-
evations, and ascent and descent during the work-
out. These are the basic parameters that are available
when users' have basic subscription. If a user has pre-
mium subscription, other features such as heart rate
data are also available.
An Endomondo activity webpage is cre-
ated by the ID of the workout such as: en-
domondo.com/workouts/23007256/. This ID is se-
quential. As of June 2016, there are over 741 mil-
lion workouts however some of these workouts are
deleted or private so it is not possible to access those
workouts. Endomondo presents workouts in its in-
terface as shown in Figure 1.
The same approach is used for users as well, all
Endomondo users have user ID and these IDs are also
sequential. One can list every workout of a unique
user if the user has not selected to be a private user.
Some users also associate their social media accounts
with their identity. Profiles of users are also presented
in a webpage and an example of a URL of a profile
page is endomondo.com/profile/4356/. In the profile
CITY MODELING | Applications - Volume 2 - eCAADe 34 | 91
 page, parameters of users are displayed such as user- Figure 2
name, country, workouts that are done by the user, Mysql table and
birthday, gender, and total distance covered. relationships of
Workouts that include GPS data in Endomondo Endomondo
are divided into parts by time (for running it is around database.
9 secs) (Cortes et al 2014). Accordingly, every 9 sec-
onds the application locates the user by a GPS point
and uploads this data. These are called GPS tuples
and they include longitude, latitude, altitude, dis-
tance, duration, and pace.
The first step to collect the data from En-
domondo is to download all public data. As all
the workouts in Endomondo are sequential one can
browse through these workouts by increasing the
number of ID and saving the data. To automatize this Figure 3
process, a script that can visit every webpage that is Traces of a user's
available to public is created. The only task of this ap- runs in a
plication is to save the html code to a text document. neighborhood.
Runs are structured as JSON objects by Endomondo,
which is helpful in text processing.
After this phase, a huge document containing all
public workouts by the whole Endomondo commu-
nity is created. This document is ever expanding
as the application gets more and more popular and
there are new workouts created by the community
every moment. This document needs to be text pro-
cessed such that every workout data is separated into
new entries that contains information such as the
user ID, username, the time of the workout, the type Figure 4
of the workout, duration of the workout, and dis- Pie chart of a user's
tance of the workout. If the application collecting the preference of street
data is used on a GPS enabled hardware, then there typologies.
will be GPS tuples available. For the purpose of this
study workouts without GPS data are not useful. Af-
ter creating these entries, this data is recorded in a
mySQL database with the respective tables and fields
as shown in Figure 2.
In the last phase, this collected information in the
database is used for visualizing users' movements. A
heatmap showing all the workouts that took placein
a certain city is done by using ggmap and ggplot-
package of R, which is a language and environment
for statistical computing and graphics. Ggmap along

92 | eCAADe 34 - CITY MODELING | Applications - Volume 2

with ggplot is used for plotting spatial data on top
of static maps such as Google Maps [5]. For this pur-
pose, first a query is made in the database for select-
ing runs in a city by filtering out runs that are out of
the city coordinates. Then since every run has a GPS
point that is recorded every 9 seconds, it is easy to
plot these points in the map and draw a path. All
these runs happening in a city form a heatmap which
indicates the intensity of usage of that space for run-
ning. With the database capabilities it is easy to cre-
ate different visualizations for an urban planner, such
as runs that start from a certain location, runs within a
time period, runs by gender or age groups, etc (figure
3). Also in this visualization other modes of data such
as crime and weather are visible to the planner/de-
signer as well. Therefore he/she can see the crime
rate of a street to understand its effect on people's
usage of that particular street.
Crime rate data comes from open data approach
which some cities has already started to participate.
Boston city has a portal for publishing crime incident
reports with the public [6]. Reported crimes are pre-
sented in this portal with GPS coordinate, time, and
the event type such as burglary, assault, etc. Unfor-
tunately, this data is not yet available in Singapore.
Weather data is also included in the database as
it affects people's running behavior. The source of
the data comes from Weather Underground website
which publishes past weather data of most of the
cities worldwide [7].
In addition to heatmaps of activities, there are
additional visualization capabilities. This tool is
planned to be used in a study that inspects the in-
fluence of street configurations in the urban running
behavior. For this purpose, there are specific visual-
izations such as a certain street with the graph that
shows the runs and the time of the day that these
runs occurred.
Another one is used to present the statistics of
street typologies such as boulevard, main road etc.
that users prefer running along. To achieve this,
streets are classified into different street typologies
and every run of each user is divided into segments
of street typologies where occurred. Then the re-
sulting statistics are show both for individual runners
(see figure 4) and the overall population of runners.
This is still in development as it needs the street ty-
pology data which is acquired by manually assigning
the streets into street typology.
DISCUSSION
This study develops a promising approach to use
available fitness data in order to assist urban plan-
ners. It improves traditional techniques such as
surveys and behavior observation, if not replacing
some of them. These traditional techniques however
very valuable, are resource intensive, thus sometimes
omitted. This visualization tool displays the fitness
activities on a map by specifying time, location, ac-
tivity, gender, age group, etc, which in turn indicates
the usage patterns of certain spaces. Moreover, it cre-
ates a tool for studying the influence of streets in ur-
ban running behavior.
This study has some limitations due to using data
methodologies. The first limitation is that the data
is open to errors, Errors from sensors that are used
should be taken into account. However, for the pur-
pose of this study, the accuracy is enough to indi-
cate streets used during the runs. The second limita-
tion may be privacy issues. We handle this by making
the data anonymous and removing all the informa-
tion regarding personal contact such as names and
emails.
Another limitation of the study comes from using
Endomondo. Endomondo is just one personal fitness
application on the market where there are dozens of
other applications as well. It gives a sample set of run-
ners within a city but it might be open to some bias.
To overcome this issue, there will be other fitness ap-
plications integrated in the system as well. Also most
of the data study is done considering this issue, so for
example in the case of checking which street is more
appealing for the runners it is avoided to consider the
number of runs as this might be deceiving. Instead,
runs that use that street are analyzed and divided into
segments and the percentage of usage of that street
CITY MODELING | Applications - Volume 2 - eCAADe 34 | 93
 is presented. That shows the runners' willingness of ning B: Planning and Design, 33(1997), pp. 727-748
choosing that street when he/she has other options. Reades, J, Calabrese, F, Sevtsuk, a and Ratti, C 2007, 'Cel-
As mentioned earlier this visualization tool will lular Census :', Pervasive computing, 6(3), pp. 30-38
Whyte, WH and Underhill, P 2012, City: Rediscovering
be part of study that is inspecting the influence of
the Center, University of Pennsylvania Press, Incorpo-
streets in urban running behavior. Therefore, the ca- rated
pabilities are devised to help understand how streets [1] http://barsukov.net/visualisation/2014/07/25/en
affect runners but they are also instrumental in help- domondo
ing planners and designers understand the usage of [2] https://flowingdata.com/2014/02/05/where-people
streets by runners. -run/
[3] http://labs.strava.com/heatmap
This study is precursor for a detailed analysis
[4] http://blog.endomondo.com/2013/10/16/endomondo
study about people's urban running behavior. In this -fitness-app-runs-past-20-million-users-and-r
stage the analysis part is not finished, as still there is eaches-profitability/
ongoing effort of collecting different modes of data. [5] https://cran.r-project.org/
In the future the results of this study will be published [6] https://data.cityofboston.gov/Public-Safety/Cri
with an analysis of the data that is collected. me-Incident-Reports/7cdf-6fgx
[7] https://www.wunderground.com/history/airport/KB
OS
ACKNOWLEDGEMENTS
The research was conducted at the Future Cities
Laboratory at the Singapore-ETH Centre, which
was established collaboratively between ETH Zurich
and Singapore's National Research Foundation (FI
370074016) under its Campus for Research Excel-
lence and Technological Enterprise (CREATE) pro-
gramme. In addition, it is part of PhD study con-
ducted by the author at Architecture and Sustainable
Design Pillar at Singapore University of Technology
and Design.

REFERENCES
Clarke, A and Steele, R 2011, 'How personal fitness data
can be re-used by smart cities', Proceedings of the
2011 7th International Conference on Intelligent Sen-
sors, Sensor Networks and Information Processing, ISS-
NIP 2011, pp. 395-400
Cortes, R, Bonnaire, X, Marin, O and Sens, P 2014, 'Sport
Trackers and Big Data: Studying user traces to iden-
tify opportunities and challenges', no title given
Ka-Wei Lee, R and Seong Kam, T 2014, 'Time-Series Data
Mining in Transportation: A Case Study on Singa-
pore Public Train Commuter Travel Patterns', Inter-
national Journal of Engineering and Technology, 6(5),
pp. 431-438
Ratti, C, Frenchman, D, Pulselli, RM and Williams, S 2006,
'Mobile landscapes: Using location data from cell
phones for urban analysis', Environment and Plan-

94 | eCAADe 34 - CITY MODELING | Applications - Volume 2

Designing for Urban Microclimates: Towards A Generative
Performance-based Approach to Wind Flow Optimization
Mohamed Khallaf1 , Julie Jupp2
1,2
University of Technology Sydney
1
Mohamed.I.Khallaf@student.uts.edu.au 2 julie.Jupp@uts.edu.au

This paper presents the foundations of a multidisciplinary design optimisation

method that addresses the problem of competing wind flow profiles within urban
microclimates. The simultaneous integration of architectural and urban design
parameters and their aerodynamic constraints are investigated. Differences in
the height of tall buildings, which define the urban canopy layer are accounted
for. The formulation that supports the simulation of aerodynamic forces at the
architectural and urban scales includes multidisciplinary parameter specification
of 2D and 3D building geometry, spatial morphology, spatial topology, wind flow
settings, and wind flow compliance. The MDO framework and its development
are discussed relative to their generative performance-based capacity and
innovative approach to multidisciplinary wind flow optimization

Keywords: Urban microclimate, Multidisciplinary design optimisation,

Generative performance-based design, Systems level perspective

INTRODUCTION of aerodynamic behaviours at both architectural and

Globally the growth of high-density cities is on the
rise (Ng 2011), placing more pressure on the provi-
sion of adequate air ventilation in these spaces. Fur-
ther, over 310 million people live in cities with a high
probability of extreme wind events such as tropical
cyclones and by 2050 these numbers are predicted
to more than double (Lall and Deichmann 2012). The
urban climatic issues of providing adequate urban
ventilation whilst mitigating against the hazardous
impacts of extreme wind events in city environments
is therefore of topical concern to building designers,
urban planners and governments alike. This paper
puts forward a notion that the relative lack of consid-
eration of these interdependent microclimatic issues
is due to the complexity of designing at a systems
level. That is, at a level that accounts for the nature
urban scales.
Consequently, the authors focus on the multi-
disciplinary design optimisation (MDO) problem of
competing wind flow profiles that exist at various
scales within urban microclimates, where a microcli-
mate is defined as the climate that prevails at the
micro-scale level and that differs from the surround-
ing area (Erell et al. 2012). Relative to wind flow,
the microclimates of small or restricted areas of high-
density urban environments can be conceived rela-
tive to both architectural and urban scales. At the ar-
chitectural scale, microclimatic wind flow profiles are
typically modelled relative to 2D and 3D building fea-
tures such as niches, under-crofts, openings, court-
yards, and awnings. At the urban scale, microclimatic
wind flow profiles are modelled relative to 2D and

CITY MODELING | Applications - Volume 2 - eCAADe 34 | 95

 3D spatial features, (i.e the spaces between groups
of buildings), such as street canyons, parks and the
'canopy' of building heights across a city block. From
this perspective, this paper presents a framework for
the development of a MDO method that accounts
for the aerodynamic forces acting upon 2D and 3D
spatial features of individual and groups of tall build-
ings. In addressing this MDO problem, the frame-
work seeks to bridge the architectural and urban de-
sign scales, and therefore takes a systems level ap-
proach to the optimisation of 2D and 3D architectural
and urban design solutions.
The paper proceeds with a review of related liter-
ature focusing on the main types of wind flow stud-
ies and MDO techniques, including generative tech-
niques before then identifying the research gap rel-
ative to the dependencies between the architectural
and urban scales. Section 3 presents the framework
for developing an MDO methodology for a genera-
tive performance-based approach to wind flow opti-
mization. Section 4 closes the paper with a discussion
on future work and how MDO can be used to support
complexity and a systems level approach to the de-
sign of 'favourable wind' conditions in high-density
cities at both the building and city scales.
RELATED LITERATURE
In considering the MDO of satisfying competing wind
flow profiles across architectural and urban scales,
three fields of design science research are of interest:
(i) optimisation of building shape and form, (ii) op-
timisation of spatial morphology and topology, and
(iii) multi-objective optimisation and MDO methods.
Optimisation of Building Geometry in Ar-
chitectural Scale
Numerous research studies on the aerodynamic op-
timization of building morphology have been under-
taken over the past 50 years. Davenport's (1971)
investigation of the shape effects of building forms
documents some of the earliest work that utilises
aerodynamic model tests of tall building structures.
The research work that followed Davenport's pio-
96 | eCAADe 34 - CITY MODELING | Applications - Volume 2
neering research focused on the effects of a five gen-
eral characteristics of building morphology aimed at
reducing aerodynamic forces. They include optimis-
ing for the effects of: (i) building corner modifica-
tions (see e.g Dutton and Isyumov 1990); (ii) taper-
ing and stepping (see e.g Kim and Kanda 2010a, Kim
and Kanda 2010b); (iii) openings and slots (e.g see
Isyumov et al. 1989); (iv) twisting (Xie et al. 2014);
and (v) building configurations and composite mod-
els, (see e.g Tanaka et al. 2013), which explore dif-
ferent building plan shape boundaries (square, circu-
lar, rectangular and elliptic), together with different
corner modifications, tilts, tapers, helical twists, and
openings.
The aerodynamic optimization of building mor-
phology can be classified into two categories, namely
aerodynamic modification and aerodynamic design.
Aerodynamic modification is an approach taken in a
situation when a building's aerodynamic mitigations
are necessary but where only limited shape changes
are permitted in order to keep the building's over-
all design unaffected. Corner modifications, such as
chamfering, slotting and roundness are common ap-
proaches. However, given the confinement in this
category and applicable/feasible aerodynamic mod-
ifications, the level of improvement may not be suf-
ficient to meet all design objectives in some cases.
Structural measures or supplemental damping de-
vices may have to be introduced for further improve-
ment. Aerodynamic design on the other hand is an
approach that integrates architectural design with
aerodynamic considerations in early design stage.
Much more aerodynamic options are therefore avail-
able and the outcomes are more efficient and effec-
tive. However, the challenge with this category is to
quantitatively assess the level of effectiveness of var-
ious aerodynamic options, so that an optimized bal-
ance can be reached between the costs and bene-
fits. Traditionally this requires comprehensive tests
on various configurations.
Although aerodynamic shape plays an important
role in tall building design, its optimization cannot
be reached without compromising other design as-
pects, which limit the number of available options.
As a result, a major challenge in aerodynamic build-
ing design optimization is not to look for the best
aerodynamic shape, but to achieve the best balance
between aerodynamic efficiency and other design
aspects, including aesthetics, cost and urban plan-
ning regulations. The various difficulties in the aero-
dynamic optimization of building morphology sur-
round the compromise between aerodynamic con-
straints with other (potentially competing) architec-
tural design variables. This leads to a compromise
between the benefits and costs of aerodynamic op-
timization (Xie 2014). Assessments of aerodynamic
effectiveness of building shape variables such as ta-
pering, stepping and twisting must be capable of be-
ing measured in the conceptual design stages so as
to be able to assess these compromises effectively,
including their potential to minimise across-wind re-
sponses, maximise possible reductions of wind load,
and reach an equalisation of responses for different
wind directions. Such information provides a valu-
able guideline in building optimization studies when
a compromise between various design aspects is de-
sired. Reasonable assessments of the effectiveness
of various aerodynamic options in the early design
stage can then be made so that the potential pros
and cons can be evaluated in the decision-making
process with regard to other design criteria
Optimisation of Spatial Morphology and
Topology in Urban Scale
The spaces and open areas between buildings, such
as streets, parks and city block courtyards are some
of the most important urban elements where wind
flow, population and traffic density fluctuate signifi-
cantly depending on the surrounding building forms,
and human exposure to good or low quality air con-
ditions (and hazardous substances) can be expected
to reflect such fluctuations (Selberg 1996). As a re-
sult, wind flow regimes and wind related problems
(e.g due to hazardous winds or traffic-related emis-
sions) have aroused much attention. Urban design
guidelines and planning strategies have as a result
been developed for cities subject to low and/or high
wind conditions. Urban design guidelines and plan-
ning strategies that target wind flow are generally
aimed at increasing "comfort levels" by achieving
more "favourable" wind flow profiles. Typically, they
have two main objectives namely, to: (i) maximise of
urban air ventilation in case of stagnant wind flow
conditions (ii) mitigate against hazardous wind flow
profiles in the case of high wind conditions.
However, in spite of growing research and urban
design and planning guidelines, many metropolises
still suffer from poor ventilation and air quality prob-
lems due to improper urban planning. Unstructured
planning of urban canopies is common in areas of
rapid urbanization (Chan and Ellen 2001, Chan and
Au 2003). Therefore, research is aimed at furthering
an understanding of the effects of street geometry on
the local atmospheric environment. The objective of
many research studies in this area is to simulate the
effects of urban morphology and topology relative to
wind flow in the context of pollutant dispersion (e.g
Xia and Leung 2001, Assimakopoulos and Ap Simon
2003) and coastal conditions impacting on wind flow
profiles and the "wall effect" (Ng et al. 2011), which
increases the hazardous conditions for pedestrians
in street canyons with different layouts. The iden-
tification of critical building configurations that en-
hance ventilation and thus provide better conditions
for positive air flows have been the focus of these
studies. The influence of the ratio between leeward
building height and canyon width and the ratio be-
tween leeward building height and windward build-
ing height are shown to be the most significant crite-
ria by these studies.
Accurate prediction of wind flow profiles within
street canyons can help urban planners to take into
account urban geometry with optimal natural ven-
tilation and comfort. As two of the most important
parameters that dominate fluctuations in wind flow
regimes urban environments, is the effects of build-
ing and street layout, which extends building geome-
try and architectural morphology into the domain of
street canyon dimensions. These effects have been
CITY MODELING | Applications - Volume 2 - eCAADe 34 | 97
 extensively studied mainly with wind-tunnel experi-
ments (e.g Kastner-Klein and Fedorovich 2001), and
numerical models (e.g Chan and Dong 2002). Oke
(1998), has also studied the flows and the pollutant
dispersion within a street, and summarized the flow
regimes according to the ratio of the building height
and the street width. However, most of the previ-
ous research works were considered where the two
sides of buildings have an identical height. In the
actual street, the typical case is that the buildings at
both sides of a street are asymmetrical in the height
layout. Xia and Leung (2001), and Assimakopoulos
and Ap Simon (2003) have addressed this gap by con-
ducting investigations on the effects of asymmetrical
street layout on pollutant dispersion. A study con-
ducted by (Moya et al. 2015) on the inner city of
Melbourne a number of architecture design strate-
gies are investigated relative to their potential to re-
duce the negative effects of high winds at the pedes-
trian level. The study utilises CFD as its main analy-
sis technique to measure and predict wind velocity.
The study shows that the average wind velocity at 2m
high is 3.7m/s, but due to the channel effect created
by adjacent buildings, wind velocity through the pas-
sages reaches 4.4m/s. As a result, there is a concomi-
tant increase in the discomfort of inhabitants. Moya's
(2015) study demonstrates the effects of adding ar-
chitecture features (such as windbreaks) to existing
buildings and the level of wind deflection and veloc-
ity reduction that can result.
Multi-Objective Optimisation and MDO
Methods
By nature, design is a multidisciplinary process and
design problem solving is a co-evolution of the prob-
lem and solution spaces (Maher and Tang 2003).
As an evolutionary process, it is akin to a balanc-
ing act between competing objectives all vying for
the greatest influence (Gerber and Lin 2014). With
the advancement of technology and the increase in
information fidelity and availability, the process of
design has become more complex as opposed to
less. Consequently, multidisciplinary design consid-
98 | eCAADe 34 - CITY MODELING | Applications - Volume 2
erations have become more and more unavoidable.
To manage complexity, and the increase in compet-
ing objectives, a systematic problem solving tech-
nique is needed. MDO methods have been explored
by various researchers as an approach to tackle and
manage these problems.
MDO refers to methods to solve design problems
which have several objective functions and incorpo-
rate a number of disciplines, the normative case for
design (Coello et al. 2007). MDO relies on numeri-
cal optimization techniques required to design sys-
tems involving multiple disciplines or components
(Martins and Lambe 2013). As defined by Poloni and
Pediroda (1997), MDO is therefore achieved through
"the art of finding the best compromise". Previ-
ous building design precedents have investigated
the application of multi-objective genetic algorithms
(MOGA) for identifying the optimal in the trade-offs
between quantitative cost related and environmen-
tal performance variables in the optimisation of de-
signs. For example, Flager et al. (2009) adopted a
MDO method to perform a study on a simple class-
room design, focusing on the optimisation of struc-
tural and energy performance. Magnier et al. (2010)
used a MOO algorithm to optimise the energy con-
sumption and thermal comfort of a residential build-
ing. The ''CATBOT'' project is based on MDO meth-
ods that link complex geometry to structural analy-
sis (Keough and Benjamin 2010). In the HDS Beagle
project, an MDO tool was developed by Gerber and
Lin (2014), which associates parametric modelling,
and a GA-based multi-objective optimisation (MOO)
algorithm focusing on energy use intensity, financial
performance net present value and spatial program-
ming compliance. The HDS Beagle tool provides an
integrated platform for enabling rapid iteration and
trade-off analysis across the domains of design, en-
ergy use intensity, and finance (Gerber and Lin 2014).
These previous research efforts illustrate the ef-
fectiveness of MDO in identifying higher perfor-
mance solution sets among multiple competing cri-
teria. However, an important limitation of these ap-
plications to consider surrounds their singular do-
main emphasis, which focuses on either structural
performance, detailed mechanical systems, or sim-
plified geometric application settings. As well, the
application of preliminary energy performance feed-
back to support complex geometry has not been
fully understood and therefore developed. Our re-
view of the literature highlights this significant gap
given the need for integrating architectural and ur-
ban variables and measures into the design simula-
tion of geometry and spatial relationships for optimal
wind flow.
TOWARDS A MDO METHODOLOGY
During the conceptual design process, it is impor-
tant for both architectural design and urban plan-
ning disciplines to identify wind flow conditions at
the earliest possible stage of their respective pro-
cesses. It is generally established that the perfor-
mance of a new building will be impacted by the
design decisions made during the early conceptual
design stages. In the case of architectural and ur-
ban design performance relative to their competing
wind flow profiles, it is no different. However for ur-
ban design professionals, it is not typically possible
to explore the impact of wind flow profiles across
a new city block or precinct at a sufficient level of
detail during the conceptual design phase due to
the lack of information about the physical features of
the buildings that they will contain. Thus, whilst re-
lated disciplines, architectural and urban design have
established differences in terms of their foci, scale,
goals and constraints as well as the guidelines and
standards that support decision-making. From this
perspective, a focus on the goals and constraints of
one discipline (e.g architecture) during the concep-
tual design process may ultimately result in a lack
of attention of another related discipline's (e.g ur-
ban planning, Kroo, 1997). Designing for the satisfy-
ing of competing wind flow profiles across the urban
canopy layer of a high-density environment charac-
terised by tall buildings, such a singular discipline-
based focus can result in adverse effects at the build-
ing and/or city scale.
The formulation of a MDO method relative to
wind flow in high-density urban environments re-
quires the integration of the three disciplines of aero-
dynamics, architectural, and urban design, which all
play an important role in defining and achieving the
multi-objective function. In this section, the authors
describe how MDO can be used to enable the simul-
taneous design of "best compromised" architectural
and urban forms via the analysis of wind flow profiles
at different levels of the urban canopy layer (Voogt
2004), see Figure 1. The criteria of a multi-objective
function are therefore investigated relative to satis-
fying wind flow profiles generated at different lev-
els of the urban canopy. The framework includes
design parameters for simulation that include vari-
ables describing building geometry, and urban spa-
tial morphology and topology. A number of wind
flow compliance parameters are considered includ-
ing wind velocity, pressure, turbulence, flow regime,
and amount of energy
MDO for Aerodynamic Architecture and Ur-
ban Design
The complex dependencies between architectural
and urban design forms and their impact on wind
flow is ill-understood in terms of how to optimise
wind flow profiles across the urban canopy layer. An
understanding of the behaviour and relationship be-
tween wind flow around buildings versus cities is
lacking. However, the relationship between the ge-
ometric and spatial features that exist across these
two different scales is a complex one. There are
mixed dependencies between architectural and ur-
ban elements with different structural qualities and
behaviours. At the building scale, design guidelines
and standards that reference to wind conditions di-
rect design decisions towards optimising for struc-
tural wind resistance and passive cooling. At the ur-
ban scale, planning guidelines are typically directed
towards optimising for 'good' or 'best' wind venti-
lation throughout a city, precinct, neighbourhood,
block, or street. The lack of understanding between
the dependencies between the architectural on ur-
CITY MODELING | Applications - Volume 2 - eCAADe 34 | 99

ban scales may ultimately result in poor design per-
formance in terms of how the physical features of
each scale impact on the wind flow profiles intended
to be realised by the building design and or urban
plan.
Considering the different spatial scales within
the urban microclimate, wind conditions can be
modelled and measured relative to four levels of
physical features relating to a building or a city's: (i)
geographical location, (ii) land topography, (iii) (ur-
ban) spatial morphology and topology and (iv) build-
ing geometry. Two main types of wind conditions
can adversely impact on both the architectural and
urban scales: (a) stagnant-to-low wind flow and (b)
high-to-extreme wind flow. In the case of stagnant-
to-low wind flow, wind velocity and permeability in-
creases the risk of airborne diseases and pollution.
Studies of these conditions have aimed at improving
wind flow and developing urban planning guidelines
to promote ventilation. These studies generally focus
on the interactions between building forms relative
to a defined 'grid' of buildings. The unit of analysis
and definition of the urban microclimate focuses on
the relationship between building- and urban mor-
phology. In the case of high-to-extreme wind flow,
wind velocity and permeability increase the risk of
building damage. Research studies in this regard
have focused on the interface between urban mor-
phology, urban topography and urban topography.
Broadly, these research investigations are aimed at
understanding how high wind conditions can be mit-
igated and controlled. The unit of analysis and defi-
100 | eCAADe 34 - CITY MODELING | Applications - Volume 2
Figure 1
Main components
of urban
atmosphere
(source: Voogt,
2004). Two-layer
classification of
architectural and
urban MDO.
nition of the urban microclimate focuses on the rela-
tionship between a city block, or a small cluster of city
blocks (neighbourhoods) and the wider city topology
and or the natural topography
The different scales of the physical features of
these research studies reflect not only differences in
the units of analysis but also how different architec-
tural and urban features impact on wind flow. Figure
1 illustrates this scale, which defines an architectural-
urban spectrum that accounts for 2D and 3D features
that define the building façade, building envelop,
city block, a cluster of city blocks, neighbourhoods,
precincts, and the city as a whole.
The gap in understanding wind flow profiles rela-
tive to the dependencies between the scales of archi-
tectural and urban physical features reflects the dis-
connect between the architectural and urban plan-
ning disciplines. This 'disconnect' is to the detriment
of meaningful design for urban microclimates and for
achieving the positive effects of wind flow within and
around buildings and cities. The approach of this re-
search therefore acknowledges the need to investi-
gate the mixed dependencies between the architec-
tural and urban scales so as to identify the relation-
ships between beneficial wind flow profiles, the phys-
ical features that can support them across scales and
the resulting design qualities that define the 'urban
microclimate'
Figure 2
Spectrum of design
and planning across
scales of urban
microclimate
relative to
architectural and
urban disciplines.
Framework of a Generative MDO Simula-
tion Sequence for Wind Flow
In order to deal with a complex design problem
and several design objectives, this PhD research
project proposes an integrated approach to cou-
pling parametric modelling techniques with MDO
techniques. The framework combines architectural-
urban-topography parameters in one platform for
performance-driven optimization for wind flow con-
dition. The structure of the proposed framework is
based on five stages, with a series of steps across
them; these stages are carried out in sequential a
manner as shown in Figure 3.
Synthesis module: Using a generative paramet-
ric approach, generate all possible design solutions
using architectural and urban design variables within
a single geometric model so as to manipulate the val-
ues of geometric, morphologic and topologic design
parameters, and the relationship between the differ-
ent parameters.
Analysis module: Direct translation of build-
ing geometry, spatial morphology and topology to-
gether with related wind flow parameter settings into
the wind flow simulation engine that utilises CFD to
test solutions. As a result, analysable wind flow pro-
files can be obtained directly from the model without
additional modification of geometry before analysis
results are then transferred to the Evaluation module
Evaluation module: Refers to overall results of
the design analysis. It compares the outcome re-
sults from analysis module with design constraints to
filters out design solutions. Evaluation module dis-
cards all solutions that do not meet building and city
compliance constraints. It ranks the remains of de-
sign solutions according to their performance based
on wind flow criteria defined at different levels within
the urban canopy layer, e.g., at the pedestrian level,
at ≤100, ≤200, etc.
Sub-Routine - External Constraints module:
Consists of other architectural and urban design con-
straints such as building regulations and codes, and
zoning ordinance. These constraints granted from
outside sources such as city councils and other re-
lated authorities.
Optimization module: This module works as a
space search mechanism, searching for the optimum
design alternatives within the domain of feasible and
performance solutions. The aim of the optimization
module is to evaluate and choose the fittest of the
available and feasible alternative designs based on its
performance. However, if the optimized design solu-
tion does not fit the performance criteria, a designer
can implement changes in the initial design parame-
ters using the synthesis module based on the simula-
tion results.
CITY MODELING | Applications - Volume 2 - eCAADe 34 | 101

MDO Problem Formulation and Parametriza- The approach requires the specification of a
tion parametric rig upfront so as to be able to generate
During the early stage of the design process, the
overall building or urban design both play a vital
role on the final design performance. Design deci-
sions are not typically aiming to satisfy a single ob-
jective rather, it aims at searching for best design
solution that compromises between competing ob-
jectives. It requires finding alternative design solu-
tions and analysing their performance impacts up-
front. However, designers in the early stage of the
design process deal with different domains that they
may not have experience in. Limitations surrounding
experience levels and the fidelity of information may
therefore affect the design decisions. In response, a
variety of design disciplines have adopted a paramet-
ric design approach so as to work within a process
that includes a performance analysis feedback loop
that supports early design decision-making. Utilising
parametric design in the early stage of design process
supports the exploration of a larger solution space
due to the number of alternative solutions generated
via the manipulation of the values of design parame-
ters. It enables the exploration of both architectural
and urban design performance, providing an analysis
feedback that contributes to the designer's decisions
about a complex problem. In addition, it strengthens
the flexibility of the design process.
102 | eCAADe 34 - CITY MODELING | Applications - Volume 2
Figure 3
The seven step
process for
integrating
architecture and
urban design with
wind flow
simulation.
a large pool of architectural and urban design solu-
tions, i.e. two layers within the solution space of al-
ternatives. To automatically generate such a solution
space, it is necessary to first formally define the de-
sign problem into a series of design objective func-
tions, variables, and constraints. These definitions
are then used to generate an associative parametric
design model, which implicitly describes a bounded
and a topologically fixed solution space. The defini-
tion of multiple objective functions provide the ba-
sis for specifying design parameters and constraints
at different levels of the urban canopy layer using
corresponding wind flow measures. The specifica-
tion of these internal layers depends on the city and
building profiles relative to existing heights. Conse-
quently, there will be a trade-off between the differ-
ent objective functions and wind flow profile opti-
mization relative to maximization of air ventilation
versus minimisation of hazardous wind conditions.
Design variables defined as the parameters that the
designer controls influence the design constraints
and objective function and are evaluated in the anal-
ysis phase; where design constraints are the func-
tions that must be satisfied during the optimization
process. In order to create a flexible yet defined de-
sign workflow, there are a total of five categories of
design variables, including: (i) building geometry pa-
rameters, (ii) spatial morphology parameters, (iii) spa-
tial topology parameters, (iv) wind flow setting pa-
rameters, and (v) building and city compliance pa-
rameters.
Building geometry parameters - From the per-
spective of architectural design, the units of analy-
sis correspond to the 2D and 3D geometric elements
that drive the generation of form. Five parameters
controlling building geometry can be identified rel-
ative to their influence on wind flow, namely: (i) 2D
building footprint or shape boundary, (ii) 3D build-
ing profile or form, (iii) maximum building height,
(iv) building perforations, and (v) building orienta-
tion. Within these parameter categories, controls for
manipulating the following four operations are spec-
ified: (a) building corner modifications, (b) tapering
and stepping, (c) openings and slots, and (d) twisting.
The building footprint is used to define the basic 2D
building plan so as to be able to explore polygonal
(triangular, square, pentagon, hexagon, octagonal,
combined between two or more shapes), elliptical or
combined shapes. 3D building profiles define the 3D
form of building, which enables the different opera-
tions (extrusion, twisting, tapering, setback, rounded
corners, etc.) to be manipulated. Building height
falls into three categories including low-rise build-
ings with a height from 0 to 10m, mid-rise build-
ings with a height varying between 10 to 15m, and
high-rise buildings that are more than 15m in height.
Building perforations enable the exploration of build-
ings to be able to mitigate against strong aerody-
namic forces (especially for tall buildings) and this will
also depend on the location, dimension and quan-
tity of openings. Finally, the building orientation can
vary 0 to 360 degrees.
Spatial morphology and topology parame-
ters - In the case of urban design, there are a range
of parameters that can be used to define urban mor-
phology which reflect the same type of geometric at-
tributes that apply at the architectural scale. They in-
clude five main parameters and different configura-
tions of them, including the: (1) city block's footprint,
(2) city block's height, (3) city block form, (4) city block
perforations, and (5) city block orientation. A further
six parameters can be identified relative to urban to-
pographic and geographic conditions that can influ-
ence wind flow. These six parameters include the:
(1) density of city blocks and buildings, (2) configura-
tion of city blocks, (3) extent of open spaces between
city blocks, (4) orientation of city blocks, streets and
grids, (5) topography or terrain of the urban environ-
ment, and (6) its geospatial location. These parame-
ters provide the basis for controls of the main spatial
topographic conditions that influence wind flow. The
density of city blocks and buildings can be defined
relative to the distribution of population density in
an urban city block to classify high density (≥ 60
dwellings per acre), medium-density (30-60 dwelling
per hectare) or low density (≤30 dwelling per hectare.
The configuration of city blocks and extent of open
spaces between city blocks are dependent on the ra-
tio of building mass versus open space and can be de-
scribed in relation to their 'mean wind incidence'. The
orientation of city blocks, streets and city grids and
the urban terrain define parameters that describe the
geospatial topology in terms of whether it is complex
or simple. The location of the city provides a defini-
tion relative to coastal or non-coastal conditions.
Wind flow setting parameters- Pedestrian wind
comfort and safety are important requirements by
many cities government in urban areas. Thus, several
city governments require studies of pedestrian wind
safety for before adding new buildings. These studies
combining statistical meteorological data, aerody-
namic information and criteria for wind comfort and
wind safety. In the architectural design and urban
planning domains it is critical to measure wind ve-
locity, wind pressure, wind turbulence, identify wind
flow profiles and measure the amount of wind flow
across city block in early stage of design process.
A parametric approach to wind flow across the ar-
chitectural and urban design scales offers an inno-
vative model to MDO of wind flow by merging the
definition of both problem and solution in the same
method through manipulation of the variables of dif-
CITY MODELING | Applications - Volume 2 - eCAADe 34 | 103
 ferent conditions and measures. However, in order
to utilise parametric models to generate alternatives,
it is necessary to define the competing wind flow
problems in a series of building and urban design ob-
jectives, variables, and constraints and identify the
maximum and minimum ranges of wind flow across
all levels of the urban canopy layer so as to iden-
tify compliance parameters. Thus, it is essential to
identify the prevailing wind flow conditions in early
stages of the design process relative to whether the
building or the area of the urban environment is sub-
ject to high-to-extreme wind flow or stagnant-to-low
wind flow. High-to-extreme wind flow conditions de-
fine as wind velocity equal or exceed of 5m/sec (Pen-
warden 1973), where stagnant-to-low wind flow con-
ditions define as wind velocity equal or lower than
1.6m/sec (Penwarden, 1973).
Wind flow compliance parameters - Refers to
the result of overall building and/or urban wind flow
performance, which includes multiple analysis calcu-
lations applied relative to the specified wind flow re-
quirements that may be calculated at the individual
building level and/or clusters of buildings. Analysis
is also applied across the different levels of the ur-
ban canopy layer. The disturbance that a building
creates from winds at the pedestrian level is due to
two separate types of pressures and will be different
to the disturbance at 100m, 200m, 300m, etc. (Erell
et al. 2012). Different types of wind flow profiles
resulting from the disturbance that a building cre-
ates will therefore result from the simulation. The
first is wind flow caused by pressure distribution on
the windward face of buildings, which increases with
height and is related to the amount of local dynamic
wind pressure. The second type caused by the pres-
sure differences between the low-pressure wake re-
gions on the building's leeward and side faces, and
the pressure regions at the base on windward face.
In the case of two or more buildings located in close
proximity to each other, wind flows may be sig-
nificantly deformed and cause a much more com-
plex effect than is usual, resulting in higher dynamic
pressures and motions, especially on neighbouring
104 | eCAADe 34 - CITY MODELING | Applications - Volume 2
downstream buildings and it also generate vortices
(Tominaga et al. 2008). Wind flow compliance param-
eters are therefore based on numerical techniques
that measure modifications in wind conditions result-
ing from design solutions. Wind flow compliance
parameters can be defined relative to five variables
including wind: (1) velocity, (2) pressure, (3) turbu-
lence, (4) flow regime, and (5) energy. Buildings per-
mit wind flow around and above their surface.
In the measurement of wind velocity, pres-
sure and turbulence, two approaches are com-
monly utilised, namely Zonal method and Numer-
ical method. These methods typically use prim-
itive governing equations call Reynolds-averaged
Navier-Stokes (RANS) equations that includes 'Zonal'
and 'Numerical' methods (Reynolds 1895). The
Zonal method calculates inter-zonal airflow using the
Bernoulli equation (Chen and Patel 1988). However
this method is impractical to couple with computa-
tional design tools. Numerical methods are therefore
more common and are typically based on computa-
tional fluid dynamics (or CFD) measures. CFD is used
to predict and measure wind velocity, wind pressure
and wind turbulence. The advantages of numerical
simulation surround the efficiencies of simulation in-
cluding the high speeds, low costs and maximisa-
tion of testing flexibility to accommodate changes in
building configurations (Stathopoulos and Baskaran
1996). CFD is an efficient measurement method
cross-different spatial scales relative to architectural-
urban-geographic scale. In the case of identifying
wind flow profiles within a city, a common method
is known as the buildings plan area fraction. Building
plan area fraction indicates the potential flow regime
in in 2D (X, Y) within a city based on Equation (1) be-
low:
Ap
λp = (1)
At
Where building plan area fraction (λp) is defined as
the ratio of the plan area of buildings (Ap) to the to-
tal surface area of the study region (At). This tech-
nique related to the city surface roughness (zo), as
the density of buildings (plan area fraction) increases
so does the city roughness. Three flow regimes de-
velop in idealized urban street canyons: (1) isolated
flow, (2) wake interference flow, and (3) skimming
flow. The isolated flow regime occurs when elements
are spaced relatively far apart (0 < λp < 0.1), the wake
interference flow occurs when elements are spaced
at a medium density level (0.1 < λp < 0.6), and the
skimming flow regime occurs for high-density build-
ing arrangements (λp > 0.6).
In the case of measure the amount of the wind
flow cross a city block, a typical technique utilised is
the 'building frontal index', which identifies the build-
ing skin facing and blocking the wind flow. This mea-
sure is related to the city surface roughness (zo) in 2D
(XZ and/or YZ) based on Equation (2) below.
Aproj
λf (θ) = (2)
At
where building frontal index (λf ) defined as the to-
tal area of buildings projected into the plane normal
to the approaching wind direction (Aproj) divided by
the plan area of the study site.
Consequently, different methods for measuring
wind conditions across the different spatial scales of
architectural and urban design can be utilised rel-
ative to the different parameters specified within
building geometry and spatial morphology and to-
pography. These measurement methods are essen-
tial for optimizing wind flow in different spatial scales
within a city. For example, at pedestrian level be-
ing equal to 2.5m/s during periods of low wind flow,
whilst during periods of extreme or hazardous levels
of wind flow, the objective function to be achieved
should be equal to no more than 5m/s.
FUTURE WORK
This paper presents an investigation of the com-
plexity of architectural and urban design relative to
competing wind flow profiles and the advantages of
adopting a multidisciplinary design optimization ap-
proach in the early design stages based on a para-
metric design approach to performance simulation.
The lack of dependencies between the three do-
mains of architecture, urban design, and aerodynam-
ics was discussed relative to the related literature. We
then defined an integrated framework for MDO fo-
cusing on the relationship between geometric and
spatial features that exist across the different scales of
an urban microclimate. The dependencies between
architectural and urban elements and their impact on
wind flow was explored in the specification of sys-
tem parameters relative to building geometry, spa-
tial morphology, spatial topology, wind flow settings,
and wind flow compliance parameters across the ur-
ban canopy layer.
In future work, this PhD research project will de-
velop the framework into detail a methodology for
MDO and carry out a series of studies at varying levels
of complexity. The implementation of the framework
and its validation will test the design parameters and
constraints so as to identify those that are significant
in that they have the largest impact on wind flow con-
ditions based on design performance.
REFERENCES
Assimakopoulos, V.D and Ap Simon, H.M 2003, 'A nu-
merical study of atmospheric pollutant dispersion in
different two-dimensional street canyon configura-
tions', Atmospheric Environment, 37, p. 4037–49
Capeluto, I. G, Yezioro, A and Shaviv, E 2003, 'Climatic as-
pects in urban design—a case study', Building and
Environment, 38(6), pp. 827-835
Chan, AT and Au, WTW 2003, 'Strategic guideline
for street canyon geometry to achieve sustainable
street air quality', Atmospheric Environment, 37, p.
2761–72
Chan, TL and Dong, G 2002, 'Validation of a two-
dimensional pollutant dispersion model in an iso-
lated street canyon', Atmospheric environment, 36, p.
861–72
Chan, AT and Ellen, SPS 2001, 'Strategic guideline
for street canyon geometry to achieve sustainable
street air quality', Atmospheric Environment, 35, p.
4089–98
Chen, H.C and Patel, V.C 1998, 'Near-wall turbulence
models for complex flows including separation',
AIAA Journal, 26(6), pp. 641-648.
Coello, C. A. C, Van Veldhuizen, D.A and Lamont,
G.B 2002, Evolutionary algorithms for solving multi-
objective problems, Kluwer Academic, New York
Davenport, A.G 1971, 'The response of six building
CITY MODELING | Applications - Volume 2 - eCAADe 34 | 105
 shapes to turbulent wind. Phil. Trans', R. Soc. A, 269,
p. 385 – 394
Dutton, R and Isyumov, N 1990, 'Reduction of tall build-
ing motion by aerodynamic treatments', J. Wind Eng.
Ind. Aerodyn, 36, p. 739–747
Erell, E, Pearlmutter, D and Williamson, T (eds) 2011,
Urban Microclimate: Designing the Spaces Between
Buildings, Earthscan
Flager, F, Welle, B, Bansal, P, Soremekun, G and Hay-
maker, J 2009, 'Multidisciplinary process integra-
tion and design optimization of a classroom build-
ing', Journal of Information Technology in Construc-
tion, 14, pp. 595-612
Gerber, D.J and Lin, S. H. E 2014, 'Designing in complex-
ity: Simulation, integration, and multidisciplinary
design optimization for architecture', Simulation,
90(8), pp. 936-959
Isyumov, N, Dutton, R and Davenport, A.G 1989 'Dav-
enport, Aerodynamic methods for mitigating wind-
induced building motion', In: Proceedings of the ASCE
Structures Congress
Kastner-Klein, P and Fedorovich, E. A 2001, 'Wind tun-
nel study of organized and turbulent air motions in
street canyons', Journal of Wind Engineering and In-
dustrial Aerodynamic, 89, p. 849–61
Keough, I and Benjamin, D 2010 'Multi-objective opti-
mization in architectural design.', In Proceedings of
the 2010 Spring Simulation Multiconference, p. 191
Kim, Y,C and Kanda, J 2010(a), 'Characteristics of aerody-
namic forces and pressures on square plan buildings
with height variations', J. Wind Eng. Ind. Aerodyn, 98,
p. 98,449–465
Kim, Y.C and Kanda, j 2010(b), 'Effects of taper and set-
back on wind force and wind-induced response of
tall buildings', J. Wind Struct, (98)6, p. 499–517 (13)
Kroo, I.M 1997, 'MDO for large-scale design', Multidis-
ciplinary Design Optimization, State-of-the-Art, pp.
22-44
Lall, S.V and Deichmann, U 2012, 'Density and disasters:
economics of urban hazard risk', The World Bank Re-
search Observer, 27(1), pp. 74-105
Magnier, L and Haghighat, F 2010, 'Multiobjective opti-
mization of building design using TRNSYS simula-
tions, genetic algorithm, and Artificial Neural Net-
work', Building and Environment, 45(3), pp. 739-746
Maher, M and Tang, H.H 2003, 'Co-evolution as a compu-
tational and cognitive model of design', Research in
Engineering Design, 14(1), pp. 47-64
Martins, J.R and Lambe, A.B 2013, 'Multidisciplinary de-
sign optimization: a survey of architectures', AIAA
Journal, 51(9), pp. 2049-2075
106 | eCAADe 34 - CITY MODELING | Applications - Volume 2
Moya, R 2015 'Empirical evaluation of three wind analy-
sis tools for concept design of an urban wind shel-
ter', Proceedings of Computer-Aided Architectural De-
sign Research in Asia CAADRIA, pp. 313-322.
Ng, E, Yuan, C, Chen, L, Ren, C and Fung, J.C 2011, 'Im-
proving the wind environment in high-density cities
by understanding urban morphology and surface
roughness: a study in Hong Kong', Landscape and Ur-
ban Planning, 101(1), pp. 59-74.
Penwarden, A. D 1973, 'Acceptable wind speeds in
towns', Building Science, 8(3), pp. 259-267
Poloni, C and Pediroda, V 1997, 'GA coupled with compu-
tationally expensive simulations: tools to improve
efficiency', Genetic Algorithms and Evolution Strate-
gies in Engineering and Computer Science, pp. 267-
288
Reynolds, O 1895, 'On the Dynamical Theory of Incom-
pressible Viscous Fluids and the Determination of
the Criterion', Philosophical Transactions of the Royal
Society of London, 186, pp. 5-47
Selberg, K 1996, 'Road and traffic environment', Land-
scape and urban planning, 1996;35, p. 153–72
Stathopoulos, T and Baskaran, B 1996, 'Computer sim-
ulation of wind environmental conditions around
buildings', Engineering Structures, 18(11), pp. 876-
885
Tamura, Y, Y.C, Kim, Tanaka, H, Bandi, E.K, Yoshida, A and
Ohtake, K 2013 'Aerodynamic and response charac-
teristics of super-tall buildings with various configu-
rations, keynotes', In: Proceedings of the Eighth Asia-
Pacific Conference on Wind Engineering, Chennai, In-
dia
Tominaga, Y, Mochida, A, Yoshie, R, Kataoka, H, Nozu,
T, Yoshikawa, M and Shirasawa, T 2008, 'AIJ guide-
lines for practical applications of CFD to pedestrian
wind environment around buildings', Journal of wind
engineering and industrial aerodynamics, 96(10), pp.
1749-1761.
Voogt, J. A and Grimmond, C. S. B 2000, 'Modeling sur-
face sensible heat flux using surface radiative tem-
peratures in a simple urban area', Journal of Applied
Meteorology, 39(10), pp. 1679-1699
Xia, JY and Leung, DYC 2001, 'pollutant dispersion in
urban street canopies', Atmospheric Environment,
2001;35, p. 2033–43
Xie, J 2014, 'Aerodynamic optimization of super-tall
buildings and its effectiveness assessment', Journal
of Wind Engineering and Industrial Aerodynamics
DESIGN TOOLS | Evaluations
Shifting Design Work from Production to Evaluation
An Evolutive Design Tool

Günter Barczik1 , Rolf Kruse2

1
Erfurt School of Architecture 2 Erfurt University of Applied Sciences, Department
of Computer Sciences
1,2
{guenter.barczik|rolf.kruse}@fh-erfurt.de

We are developing an evolutive design tool that seeks to facilitate a shift in the
focus of the process of designing architecture: away from the production of design
alternatives or options towards an evaluation of semi-automatically generated
ones. We work towards outsourcing the production of design alternatives in a
given design task to a CAD tool and thereby give human designers more time to
evaluate and discuss those alternatives and guide the tool in the production of
improved alternatives. The format of our work is an experimental student design
and research project where architects and computer scientists collaborate.
Though the project is in a very early stage, our aim is to ultimately shift the focus
of human designers' involvement from production of design options to the
evaluation of those, in order to give humans more time to think, discuss, find,
analyze and include many different points of view and make it easier for them to
be impartial in finding optimal solutions. We developed a design tool that uses
interactive evolutionary algorithms to support exploration of design options.

Keywords: Genetic Algorithm, Evolutive Design Strategy, Interactive

evolutionary computation

Background and Introduction
An important part of designing architecture consists
of producing and testing alternative possible solu-
tions for the given architectural design problems, and
then evaluating the test results to produce a next
generation of alternatives. Then repeating the pro-
cess until, with luck (and time running out before a
submission deadline) the process converges and one
alternative is chosen as an appropriate solution. Such
a production- and test-intensive process has proven
to be feasible because usually at the start of a de-
sign project, not all criteria necessary to recognize
a possible solution for the best one are known and
have to be found out as part of the design process.
The process has two large interconnected problems:
firstly, production of alternatives takes up a lot of
time: they have to be developed mentally and rep-
resented as drawings and, ideally, physical models.
Alas, the more time used for production, the less
available for evaluation. Together with the human
tendency for laziness, this secondly often seduces de-
signers to quickly narrow the production of alterna-

DESIGN TOOLS | Evaluations - Volume 2 - eCAADe 34 | 109

 tives to those resembling already known or person-
ally - instinctively - preferred ones. This is problem-
atic, because it in fact restricts the browsing of the
solution space without having any sort of proof that
the restriction does not rule out viable, indeed bet-
ter solutions than the ones studied. Our aim is to re-
move both problems by providing designers with a
tool that - partially - automates the production of de-
sign alternatives both in number and diversity. Thus,
designers would be able to use more time to eval-
uate solution candidates and would be presented
with new, previously unthought, options that could
both provide new paths of enquiry and better solu-
tions than the ones generated purely on the basis
of pre-existing experiences. Such a tool could help
to overcome personal limitations and biases and fur-
thermore free up time for controversial discussions of
provocative solution proposals, helping to incorpo-
rate valuable ideas.
The presently prevalent design process and the
interactive evolutionary process we propose is illus-
trated in Figure 1: Both processes are initiated by the
identification of needs and constraints regarding the
possible design solution. Based on these identifica-
tions, a - preferably large - number of possible solu-
tions are imagined. From those, promising ones are
selected and prototypes build. Those are tested and
evaluated. Based on those evaluations, the criteria
are refined and the process re-iterated from the step
of imagining possible solutions.
We propose to automate 3 of those steps: the
creation of possible solutions, the building of proto-
types, and the refinement of criteria.
As the creation of possible solutions to a design
problem is as a matter of course a rather extensive
undertaking, for our project we restrict it to the cre-
ation of alternative arrangements of the building's
mass (which can easily be derived from the functional
program). The building of prototypes means out-
putting those arrangments as 3D models in different
geometric interpretations. Figures 2-6 show exam-
ples of such automatically generated solution pro-
posals, both as a 'raw' voxel arrangments (Figures 3-5)
110 | eCAADe 34 - DESIGN TOOLS | Evaluations - Volume 2
and as an interpretation of those voxels in the form of
a smooth surface enveloping those voxels (Figure 2:
top view, Figure 6: perspective view).
The refinenemt of criteria may be automated as
a tracking of the users' personal preferences, or the
destillation of those from the choices made by the
users. For example, if the users repeatedly make
choices where the building's mass is concentrated
in a specific area, the system should be able to
recoginze this and automatically focus the search in
this area.
Experimental inter-disciplinary student de-
sign and research project
We conduct this research and development in a col-
laboration between architects and computer scien-
tists. Students from both disciplines discuss and
question the concept and develop strategies and
prototypes for the envisioned tool. Different styles
of thinking, working and expressing ideas are juxta-
posed, forcing architects to state clearly issues they
rarely discuss let alone think about (i.e. how to actu-
ally create a shape) and forcing computer scientists
to be much more inquisitive than they are used to in
order to tease out the aforementioned issues.
Design Tool: Subdividing Space and ar-
ranging the resultant parts over several
generations
The criteria for what actually makes a good design
option good are reason for endless quarrel between
practitioners and users of architecture. In order not
to get lost and be able to start producing a usable
tool, we greatly narrow down the tool's actual func-
tionality to the mere subdividing of space. Design-
ing any kind of architecture can be partially under-
stood as simply subdividing space. Or put differently:
to arrange a known amount of built volume (that is
capable of housing the desired functions of a build-
ing) within a larger given volume, the allowed build-
ing envelope. This envelope is the result of a given
site and the allowed height. Any architectural de-
sign therefore has to answer the question: exactly
how (in which shape and form) does the desired vol-
Figure 1
Presently Prevalent
(top) and proposed
(bottom)
evolutionary design
processes.

ume reside within the allowed enveloping volume ? Technical Implementation

Both volumes are relatively easy to determine from
the site, local building regulations and the desired
functional programme for the building. Our tool pro-
duces a user-defined number of possibilities for this
inclusion or residence. The users' task lies in studying
and evaluating the produced alternatives, and com-
municating the result to the tool by simply choosing
a number of alternatives. The tool will then produce
variants of the choices and the study, evaluation and
choice repeat. The new generation of variants is pro-
duced by cross-breeding of the choices as well as by
random mutation to prevent locking oneself into a
narrow part of the solution space.
To create the mass or volume we at the moment em-
ploy a very simple method: the allowed building en-
velope is divided into voxels, the resolution choos-
able by the users. These voxels are then filled one
after another, starting from a number of voxels in
the base plane. This number of starting voxels can
also be chosen by the users. Voxels are filled until
they together make up the desired volume. The first
prototype is built with web technologies (HTML, CSS,
Javascricpt and WebGL) to allow easy access just with
a modern web browser and without installation. One
can load a b/w-bitmap as definition for the building
site. Several parameters can be set for the model gen-

Figure 2
Top view of
automatically
generated massing
studies,
subsequently
manually
developed as
smooth surfaces
enveloping the
voxel arrangments
directly output from
the tool prototype
(also see Figure 6).

DESIGN TOOLS | Evaluations - Volume 2 - eCAADe 34 | 111

 Figure 3
Screenshot of Tool
Prototype showing
15 automatically
generated massing
options on the basis
of a site uploaded
as pixel image.

Figure 4
Screenshot of Tool
Prototype showing
100 automatically
generated massing
options.

112 | eCAADe 34 - DESIGN TOOLS | Evaluations - Volume 2

eration: maximum height of the building volume,
number of variants to be created, total volume of new
buildings, number of initial locations (germ cells), al-
gorithm for generation. The system then generates
several variants and displays them in an interactive
3D viewport for analysis and rating. Currently model
generation and storage is done within the browser.
This will be shifted to a server backend, which will fa-
cilitate to implement remote collaboration.
Semi-Automatic Protoarchitecture as a ba-
sis for further manual work
The current output of the tool as a matter of course
does not proclaim to be a piece of architecture. It is
merely a pile - or several piles - of voxels (Figure 5).
But it is also a possible form of the desired building
volume. As such, it can be taken as a basis for further
development towards architecture. The voxels may
for example inform a continuous curved surface (Fig-
ures 2 and 6). Such development options and pos-
sibilities and their possible automation by inclusion
into the tool's display and/or output are on our - ad-
mittedly long - list for future development.
Evolutive Aspects of the Tool
The choosing of automatically generated variants
from one generation to the next together with cross-
breeding and random mutation constitutes a genetic
algorithm, although one in which the fitness criteria
for survival are not known initially. They co-evolve
with the actual population: the population is created
by the tool, the fitness criteria emerge from the dis-
cussion amongst the designers who evaluate the op-
tions. This eliminates the problem of pre-defining the
fitness criteria, something which architects find no-
toriously impossible to do [1], as it constitutes defin-
ing architectural quality in general. Although devel-
oped independently, our ideas relate to the field of
Human-Based Genetic Algorithms [2, 3] and Inter-
active evolutionary computation [4]. Fitness criteria
may vary from one generation to the next as design-
ers learn more and more about the design problem
with each discussed generation. This effect turns out
to be startling to the computer scientists, but well-
known to the architects. In fact, in architectural com-
petitions, which in their selection process somewhat
mimic evolutionary algorithms, projects which have
been ruled out earlier are not rarely 'brought back'
into the discussion. Occasionally, a project that had
already been discarded in fact wins the competition,
examples being Jorn Utzon's Sydney Opera House or
Zaha Hadid's Hongkong Peak Proposal.
Designing without Prejudice
The initially random generation of design alterna-
tives means that designers' personal prejudices (i.e.
preferences in terms of formal style) can be avoided
or at least made apparent through detection of
trends(see below). This, we propose, facilitates a
wider search of the solution space and therefore the
finding of better design solutions.
Future development
We aim to equip our tool with the capability to detect
trends in the designers' choices and to point those
out to the designers so that they can choose wether
they want to follow and reinforce the trends or work
against them to avoid too early convergence of so-
lutions. Such 'trend detection' could furthermore be
coupled individually to different users so that the tool
can be personalized and, more importantly we find,
team discussions be facilitated. One important fea-
ture to implement next is the exclusion of unwanted
proposals. The parameter range of the remaining
suggestions will serve as input for the next iteration.
Outlook
We are very well aware that the project is at a very
early stage. However, communication between ar-
chitects and computer scientists is now established,
common language and goals found and develop-
ment work commenced. The next steps will be
the actual implementation of the features described
above. Eventually, we envision a change in design-
ers' day-to-day work: away from the production of
designs by single individuals to their discussion, anal-
ysis and evaluation by teams. At the start of every
work period, design alternatives previously automat-
DESIGN TOOLS | Evaluations - Volume 2 - eCAADe 34 | 113
 Figure 5
Automatically
generated massing
studies as raw voxel
arrangments
directly output from
the tool prototype.

Figure 6
Automatically
generated massing
studies manually
developed as
smooth surfaces
enveloping the
voxel arrangments
directly output from
the tool prototype.

114 | eCAADe 34 - DESIGN TOOLS | Evaluations - Volume 2

ically produced are assembled. The work period is
spend with discussion and ends in the selection of
only several alternatives deemed viable. Those are
used by the design tool as a basis for the automated
nightly production of the next generation, which is
presented to designers at the start of the next work
period, and so on, until the team of designers evalu-
ate one alternative as the end result.

REFERENCES
Kurth, W. and Barczik, G. 2007 'From Designing Objects
to Designing Processes', Proceedings of the eCAADe
2007, Frankfurt a. M.
[1] http://www.cddc.vt.edu/host/delanda/pages/algor
ithm.htm
[2] http://web.archive.org/web/20091027041228/http:
//geocities.com/alex+kosorukoff/hbga/hbga.htm
l
[3] https://en.wikipedia.org/wiki/Human-based_genet
ic_algorithm
[4] https://en.wikipedia.org/wiki/Interactive_evolu
tionary_computation

DESIGN TOOLS | Evaluations - Volume 2 - eCAADe 34 | 115

Integrated Spatial-Structural Optimization in the
Conceptual Design Stage of Project
A tool to generate and optimize design solutions aiding informed decision
making for Architects, Engineers and Stakeholders

Subhajit Das1 , Samaneh Zolfagharian2 , Mehdi Nourbakhsh3 ,

John Haymaker4
1,2
Georgia Tech 3 Autodesk 4 Perkins Will
1,2
{das|samanehz}@gatech.edu
3
mehdi.nourbakhsh@autodesk.com 4 john.haymaker@perkinswill.com

Healthcare design projects require the careful integration of spatial and

structural requirements. Today, design teams typically resolve these requirements
in two separate, largely sequential steps. In the first step, architects leverage their
experience and vision to develop space plans that address program and goals.
Next, based on the architect's recommended design, engineers generate and refine
a structural design to address structural requirements. This manual process
produces a very limited number of non optimal spatial and structural design
solutions with unclear decision rationale. This paper presents the Integrated
Spatial-Structural Optimization (ISSO) decision making methodology. ISSO
supports design teams by helping them generate, analyze, and manage a vast
number of integrated spatial and structural solutions. ISSO features a bi-level
optimization workflow that has been customized for spatial and structural design
of healthcare facilities. The paper describes implementation in the Dynamo
parametric modeling platform, and retrospective validation of the algorithm and
workflow on an industry case study to demonstrate how ISSO can help design
teams generate, analyze, and manage more conceptual design options.

Keywords: Spatial Design, Generative Design, Design Optimization, Facility

Planning, Design Tools, Design Automation

INTRODUCTION Sharma 2005, Drira et al. 2007). For instance, to de-

Facility layout problems, found in design domains in- sign the layout of a hospital, designers must inter-
cluding circuit boards, service centers, airports, and relate various organizational units (e.g. pharmacy,
hospital, seek to locate and interrelate objects to lab, patient room) to satisfy design requirements and
optimally meet requirements (Yeh 2006, Singh and minimize resource consumption (e.g. nurse travel

DESIGN TOOLS | Evaluations - Volume 2 - eCAADe 34 | 117

 time). Also known as space layout problem in build-
ing design, facility layout problems are NP-hard prob-
lems requiring heuristics to manage the vast de-
sign spaces. Healthcare facilities are extreme exam-
ples because they have extreme spatial complexities
wherein multiple operational variables are in process
to satisfy the design requirements of many stake-
holders. The challenge is how to structure a design
process that allows the stakeholders most systemat-
ically explore the design process to maximize stake-
holder value (Clevenger et al. 2013).
In hospital design, Hassan et al. (2014) explored
heuristic methods including hill climbing, simulated
annealing, and extended simulated annealing with
genetic algorithm-style operators to model pedes-
trian flow statistics and find feasible spatial layout de-
sign elements. Using exact methods such as New-
ton's differential equation, Lorenz et al. (2015) im-
proved the change management aspect of hospi-
tal layouts design by propagating local changes im-
mediately to the global space layout. However,
most space layout planning studies focus on single-
objective optimization, which are limited in practice.
For example, a hospital layout that is optimized for
space allocation may not have a feasible or affordable
structural layout. Holst et al. (2013) suggests that
researchers should solve the space layout problem
of hospitals as a multi-objective optimization prob-
lem, where various parties may have conflicting ob-
jectives. To this end, the objective of this study is to
develop an integrated spatial-structural optimization
(ISSO) approach to incorporate parametric structural
design into space layout designs of hospitals at the
conceptual design stage.
STATE OF THE KNOWLEDGE AND PRAC-
TICE
Hospitals have highly complex layout constraints and
requirements, making their design one of the most
critical facility layout problems. While architects are
typically responsible for aesthetic and effective spa-
tial designs, structural engineers focus on the struc-
tural functionalities of these layouts. Due to time
118 | eCAADe 34 - DESIGN TOOLS | Evaluations - Volume 2
constraint and lack of structural consultants in the
team in conceptual design stage, architects often cre-
ate designs based on architectural contexts and re-
quirements without explicitly considering structural
constraints and requirements. Next, structural engi-
neers develop structural designs based on the archi-
tectural design recommendations (Clevenger et al.
2013, Hassan et al. 2014). Because of the separa-
tion in workflows and goals, finding solutions that
optimize both architectural and structural require-
ments of a project is challenging. Thus, project teams
may benefit from a tool that automatically generates
and analyzes initial structural configurations (e.g. the
placement of columns) based on proposed spatial
design or schematic layout in the conceptual design
stage (Mora et al. 2006), and helps them understand
trade-offs, and communicate preferences and a com-
mon decision between architects and engineers.
Table 1
Summary of the
studies on
structural/spatial
design problem.
Reflecting the lack of integration between architec-
tural and structural design, most of the studies in
literature only focus on one or the other problem
in hospital design (Table 1). For instance, Elshafei
(1977) formulated a hospital layout as a quadratic
assignment problem to minimize the effort of pa-
tients walking from one department to another. Vos
et al. (2007) evaluated the flexibility and fit of an
architectural design for the operation of a hospital.
Hahn et al. (2010) proposed a method that gener-
ate the design of the departments of a multi-story
hospital design while accounting for the evacuation
plan of patients. Helber et al. (2016) proposed a
hierarchical planning approach, for large and com-
Figure 1
Overview of the
research
methodology.
plex datasets, to calculate the location of depart-
ments and wards. To solve the structural and spa-
tial optimization problem in non-healthcare projects,
Nimtawat and Nanakorn (2009) employed a grid-
based topology optimization method that automati-
cally generates beam-slab layouts. Mora et al. (2006)
developed a prototype called StAr which enables
engineers to find the best solutions from architec-
tural and structural designs. Delgado and Hofmeyer
(2013) proposed a virtual toolbox that generates op-
timized structural design solutions from spatial de-
signs through simulating the iterative interaction be-
tween spatial and structural designs. However, their
current implemented model only generates struc-
tural layouts for spatial designs without optimizing
its structural layout.
A major limitation of these studies is the lack of
an integrated environment for establishing paramet-
ric relations and dependency among architectural
and structural elements for design generation, analy-
sis, and decision making. This study aims to develop
ISSO, an integrated spatial-structural optimization
toolkit to help designers during the conceptual de-
sign stage to explore more design alternatives before
choosing optimal solutions. ISSO provides an inter-
face and process to bring architects, engineers, and
stakeholders together early on in the project cycle,
imparting the opportunity to define shared objec-
tives and generate spatially and structurally promis-
ing design solutions, understand the tradeoffs, and
make and communicate better-informed decisions.
METHODOLOGY
ISSO has four main phases (Figure 1). First, users de-
fine their spatial-structural requirements such as list
of program spaces, adjacencies, qualities, load cases,
and section types. Next, the spatial optimization ker-
nel optimizes the space layout problem, based on
the spatial requirements. Third, these spatial lay-
outs are optimized using structural optimization ker-
nel. Finally, users construct pareto-optimal solutions
and then select optimum solutions based on decision
theory and social network techniques. The following
sections describe the details of the process.
DESIGN TOOLS | Evaluations - Volume 2 - eCAADe 34 | 119
 Figure 2
Higher level process
map of the Spatial
Optimization
Kernel.

Figure 3
Design process of
parametric
structural design
optimization.

User Requirements
First, project design teams (including architects and
engineers) input contextual data into the system, in-
cluding template layouts as shown in Figure 4(a).
They represent layouts from industry best practices,
constraining spatial allocation of program elements
to industry tested layouts. Next, architects input fa-
cility program data, including program names, quan-
tity, area, dimension or spatial aspect ratio expecta-
tion, as well as priority values for each program ele-
ment, on a scale of one to 10. For example, an exam
room is likely to be more relevant for the design team
than staff toilet. After completing the spatial require-
ments, users input structural analysis data such as
building materials, a list of sections, and loads and
supports.
Spatial Optimization Kernel
The first stage of ISSO seeks to quickly generate and
analyze spatial design alternatives (Figure 2). To be-
gin, the architect develops a building program, con-
sisting of a list of spaces, their required dimen-

120 | eCAADe 34 - DESIGN TOOLS | Evaluations - Volume 2

 sions, and adjacencies. Next, the architectural de- sub-list of program spaces. The algorithm places the
signer creates any number of template layouts that input program spaces on one of the rectangle boxes
represent common and successfully accepted layout (selected randomly) by aligning its shorter edge with
Figure 4 strategies for a healthcare building. the shorter edge of the bigger rectangle box (Figure
(a) Test case 7(a)). This process is repeated till all the input rect-
template layouts for angle boxes are packed or till there is no more space
the exam room available in the rectangle boxes. The algorithm then
department (b) computes a fitness function based on adjacency be-
Sub-list making tween relevant spaces & leftover waste space to score
process of program and rank each alternative. The GA learns from the fit-
spaces. ness score after each round and steers the future al-
ternatives to deliver better design alternatives. Fig-
ure 5 describes the Spatial Optimization Component
in more detail.

Structural Optimization Kernel

The second stage of ISSO integrates a parametric
structural optimization process that automatically
generates structural layouts satisfying structural re-

Figure 5
Process Map
showing the Spatial
Optimization ISSO uses a custom Genetic Algorithm (GA) to auto-
component of ISSO. matically pack rectilinear spaces onto these template
layouts in order to generate and evaluate hundreds
of alternatives against design goals and objectives in
the form of a fitness function. The algorithm packs
programs in a wide range of ways to create distinct
individuals in the initial population, then randomly
selects one of the rectangles and starts to prepare
sub-lists of program spaces as shown in Figure 4(b).
First ISSO parses template layout provided and builds
a list of empty rectangle boxes. In every iteration,
ISSO randomly selects an empty rectangle box and
starts assigning programs to it. The system keeps a
check on the total area of selected program spaces,
which should not be more than the selected empty
rectangle box. If the area exceeds, then the program
quits this rectangle and moves on to the next rect-
angle box and this goes until all rectangles are se-
lected. At the next step, in order to pack the pro-
gram spaces inside the rectangles, the code iterates
through each of the rectangle boxes and selects its

DESIGN TOOLS | Evaluations - Volume 2 - eCAADe 34 | 121

 quirements by considering space boundaries and
requirements established in the first stage. Figure
3 shows the workflow of the parametric structural
design optimization, including four phases: space
boundary, structural layout, structural analysis, and
optimization.
Space Boundary: An algorithm developed in a
Python script node extracts details required for im-
plementing structural analysis such as coordination
of each corner of rooms, from spatial design alter-
natives (Figure 8(b)). Output of the algorithm is the
boundaries of spaces in which structural elements
will be placed next. The algorithm also determines
whether a wall is along EW or NS directions.
Structural Layout: The inputs from Optimo, a
Genetic Algorithm based optimizer in Autodesk Dy-
namo (Asl et al. 2015), identifies the number of axis
in EW and NS directions as well as profile sections of
beams and columns. Then the algorithm checks if
there is any wall within a determined distance which
is 0.1m to 1.5m here from an axis. If there is any wall
within this distance, the algorithm measures the to-
tal length of walls located at the same distance from
the axis and finds the location of the maximum total
length so as to move the axis to that location. After
placing axes, the algorithm places columns at each
intersection of axes, except those intersections that
fall within corridors or spaces. In addition, the algo-
rithm places beam between each two columns along
each axis.
Structural Analysis: In this step one of the node in
Dynamo, part of SAP package assigns analytical bars,
structural loads, materials, profile sections, and sup-
ports to the model generated in the previous step.
After that, one of the nodes in SAP transfers values to
Robot Structural Analysis for calculating and analyz-
ing the maximum and minimum stresses and struc-
tural weights. If the stress of a member to the maxi-
mum allowable stress is more than one, then the de-
sign will be automatically rejected.
Optimization: ISSO is based on discrete opti-
mization; however, since Optimo generates contin-
ues real inputs, a Structural-Optimization custom
122 | eCAADe 34 - DESIGN TOOLS | Evaluations - Volume 2
node has been developed to round real values to in-
teger, adopted from Asl et al. (2015). For example, if a
generated value in Optimo for number-of-axis is 4.3,
then the Structural-Optimization custom node will
round the value to the closest integer number which
is 4 in this case. Optimo will keep track of inputs
and outputs until finding variables that generate so-
lutions with an acceptable stress ratio and less struc-
tural weights (Figure 9). Then designers can compare
structural details of the potential space design alter-
natives to select the best spatial design alternatives
that satisfies requirements of both spatial and con-
ceptual structural designs.
Decision Making Kernel
The final stage of ISSO seeks to bring the spatial and
structural design spaces together into an integrated
framework for decision making. After generating a
design space of promising alternatives, the design
team is able to use a variety of tools to analyze this
data, understand trade-offs, and provide final prefer-
ence weighting to communicate a decision. For an-
alyzing the tradeoffs a number of free tools are be-
coming available. Ultimately the resolution of these
tradeoffs requires knowledge of the preferences of
the stakeholders. We, therefore, imported the data
for the most promising alternatives into the Wecision
decision making tool (2016). Multiple scoring and
weighting schemes are possible in Wecision, Figure
10 shows the application of the "Choosing by Ad-
vantages" methodology, where stakeholders collec-
tively determined the Importance of Advantages be-
tween Alternatives. Importance of each advantage is
summed to determine the Alternative with the great-
est Value.
The decision making kernel provides an inter-
face to select options not only by gauging users ob-
jective and subjective design choices. For example,
stakeholders can input on a scale of 0 to 100, be-
tween all the options, how effective they find the
circulation design for the said option. They can in-
put like or dislike reaction on a certain aspect of a
design option. Subjective inputs are translated to
Figure 6
Higher level process
flow of decision
making kernel.

Figure 7 quantitative values, then they are summed with ob-

(a) ISSO, spatial jective inputs, then normalized and visualized in the
optimization model interface. Wecision communicates each design op-
showing different tions total performance and individual performance
ways to place ratings against the set criteria, with respect to each
program elements other. This aids decision making by stakeholders as
in the input they are informed early on in the project cycle about
rectangle boxes. (b) the pros and cons of the design choices they make.
ISSO - Output space
plan layouts with
individual scores
CASE STUDY - PROTON CENTER
We tested ISSO on a Proton Therapy Center, an ex-
and overall space
isting project designed and built by an architectural
plan scores.
practice in the USA. ISSO implemented and testified
the process above to enable the design team to gen-
erate and explore a large number of potential alterna-
tives as design solutions. We intended to have ISSO
rapidly generate and evaluate large numbers of alter-
natives specifically focusing on the outpatient clinic
department comprising of nurse station units and
exam rooms. Table 2 describes the design space in-
vestigated in the case study.

DESIGN TOOLS | Evaluations - Volume 2 - eCAADe 34 | 123

 Define user requirements Table 2
For this prototype, we investigated on three template Details of spatial
layouts namely a pod, peninsula and ladder scheme. adjacency and
We evaluated the space plans with respect to the fol- structural layout
lowing metrics: optimizations.
Adjacency factor - the spatial proximity of exam
rooms to the nurse stations and the spatial together-
ness of all exam rooms in the hospital. Lower value
represents better fitness.
Structural factor - the weight of the structure,
inferred by generating and analyzing beams and
columns from the structural grid layout. Lower value
represents higher structural fitness.
Visibility factor - the amount of visibility the
nurse station gets for each spatial layout alternative
generated. Higher value represented better space
plan fitness.
Figure 8
Generate Spatial Layouts (a) ISSO - State of
After the design team defined the requirements, ISSO the Dynamo Graph
set about generating a design space of potential generating spatially
space plans, scored in terms of the metrics. ISSO optimized space
commenced the process with an initial population plan. (b) Set of
of randomly generated spatial configurations. Next, labelled space plan
it scored them for their ability to minimize the ad- options as output
jacency factor inclusive of distance between exam represented as
rooms, distance from nurse stations to exam rooms, polygons and poly
and leftover wasted space (Figure 7(b)). The system surfaces depicting
(Figure 8(a)) sorted the spatial layouts based on the program elements.
adjacency factor and prepared the next generation
of the population. Further, the system produced a
new generation by combining the best layouts from
previous iterations with new spatial layouts that are
generated randomly or cross-bred with two potential
spatial layouts from previous iterations. This process
iterated until the fitness values stabilized and gener-
ated potential design alternatives, as shown in Fig-
ure 8(b). In the last step, selected design alternatives
passed on to the structural optimization kernel.

124 | eCAADe 34 - DESIGN TOOLS | Evaluations - Volume 2

Figure 9
Visualization of
various alternatives
in Wecision,
showing
stakeholder
preferences
determining the
winning solution,
showing the
tradeoffs of various
options with
respect to
circulation,
visibility, adjacency,
structural weight
etc.

Generate Structural Solutions Make Decision

In the next step, for each spatial layout, the structural
optimization kernel fit a grid of beams and columns.
The optimization algorithm optimized the structural
layout based on the number of axis on the X and Y di-
rection, and the section of each structural member.
Columns that were not placed on the walls were ei-
ther moved or removed from the design. The struc-
tural optimization kernel minimized the weight of
structural members while satisfying the yield stress
requirements. Figure 9 shows the structural layout of
three selected spatial design alternatives.
Figure 10
A set of structural
design alternatives.
Finally, ISSO provides a collection of promising spa-
tial and structural designs to the decision making ker-
nel for final weighting and communication of the de-
cision. Figure 10 shows the resulting designs, with
their attributes and importance identified, and the
importance of the advantages weighed by the deci-
sion maker.
DISCUSSION AND CONCLUSION
ISSO seeks to assist architects in a new type of design
process - one in which they can simply input their de-
sign preferences, generate a broad variety of spatial
and structural designs, and find the best design so-
lution. We tested ISSO retrospectively on data from
the design process for the Nurse Station & Exam room
department for a cancer treatment facility. The de-
scription of the research process describes how ISSO
renders a number of novel design alternatives that
optimized either spatial goals or structural goals or
a trade-off between both, which can inform design-
ers of better design directions. The prototype test
on the Proton Therapy case study successfully deliv-
ered multiple space plan options optimized for spa-

DESIGN TOOLS | Evaluations - Volume 2 - eCAADe 34 | 125

 tial adjacency and structural cost principally. It was
successful to broaden the design team's overview of
the design space at hand, with the specific score for
each design. However, from the case study, we re-
alized that the optimization processes adopted were
agnostic of each other, limiting possibility to gener-
ate many more potential design options for the de-
sign team. Next, we look forward to implementing si-
multaneous multi-objective optimization of the pro-
cesses described in ISSO to deliver robust and more
realistic space plan layouts. ISSO highlights a work-
flow. Starting from existing template layouts, com-
bining industry best practices with automated sys-
tems are one of the salient features of ISSO. In future,
we envisage to develop ISSO develop its own tem-
plate layouts.
Future experiments will test the extent to which
ISSO can allow architectural designers to better sup-
port structural design requirements, decrease the
number of design modifications during structural de-
sign stage, reduce waste space, optimize spatial lay-
outs, structural layouts, initial dimensions of struc-
tural elements, and structural weight of many design
alternatives. ISSO seeks to increase the spatial and
structural performance of a design, decrease the cost
and time of projects, and enhance collaboration be-
tween architects and structural engineers.
ACKNOWLEDGEMENTS
The authors acknowledge Design Space Construc-
tion (DSC 2016) at Georgia Tech for creating the plat-
form for this collaborative research, and our DSC
project team members Khatereh Hadi, Adetania Par-
manik, Samual Shams and mentors Diana Davis and
Josh Emig from Perkins+Will, and Mohammed Asl
from Autodesk for helping scope and test this re-
search.
REFERENCES
Asl, MR, Stoupine, A, Zarrinmehr, S and Yan, W 2015
'Optimo: A BIM-based Multi-Objective Optimization
Tool Utilizing Visual Programming for High Perfor-
mance Building Design', Proceedings of the Confer-
ence of Education and Research in Computer Aided Ar-
chitectural Design in Europe (eCAADe), pp. 673-682
126 | eCAADe 34 - DESIGN TOOLS | Evaluations - Volume 2
Clevenger, CM, Haymaker, JR and Ehrich, A 2013, 'Design
exploration assessment methodology: Testing the
guidance of design processes', Journal of Engineer-
ing Design, 24(3), pp. 165-184
Delgado, JD and Hofmeyer, H 2013, 'Automated genera-
tion of structural solutions based on spatial designs',
Automation in Construction, 35, pp. 528-541
Drira, A, Pierreval, H and Hajri-Gabouj, S 2007, 'Facility
layout problems: A survey', Annual Reviews in Con-
trol, 31(2), pp. 255-267
Elshafei, AN 1977, 'Hospital layout as a quadratic assign-
ment problem', Operational Research Quarterly, pp.
167-179
Hahn, P, Smith, JM and Zhu, YR 2010, 'The multi-story
space assignment problem', Annals of Operations Re-
search, 179(1), pp. 77-103
Hassan, FH, Swift, S and Tucker, A 2014, 'Using Heuristic
Search with Pedestrian Simulation Statistics to Find
Feasible Spatial Layout Design Elements', Journal of
Algorithms, 2(4), pp. 86-104
Helber, S, Böhme, D, Oucherif, F, Lagershausen, S and
Kasper, S 2014, 'A hierarchical facility layout plan-
ning approach for large and complex hospitals', Flex-
ible Services and Manufacturing Journal, pp. 1-25
Holst, MK, Kirkegaard, PH and Christoffersen, L 2013
'Parametric Optimization of Hospital Design', Pro-
ceedings of World Academy of Science, Engineering
and Technology, p. 200
Lorenz, WE, Bicher, M and Wurzer, GX 2015, 'Adjacency
in hospital planning', IFAC-PapersOnLine, 48(1), pp.
862-867
Mora, R, Rivard, H and Bédard, C 2006, 'Computer rep-
resentation to support conceptual structural design
within a building architectural context', Journal of
Computing in Civil Engineering, 20(2), pp. 76-87
Nimtawat, A and Nanakorn, P 2009, 'Automated layout
design of beam-slab floors using a genetic algo-
rithm', Computers & Structures, 87(21), pp. 1308-1330
Singh, SP and Sharma, RR 2006, 'A review of different ap-
proaches to the facility layout problems', The Inter-
national Journal of Advanced Manufacturing Technol-
ogy, 30(5-6), pp. 425-433
Vos, L, Groothuis, S and van Merode, GG 2007, 'Evalu-
ating hospital design from an operations manage-
ment perspective', Health care management science,
10(4), pp. 357-364
Yeh, IC 2006, 'Architectural layout optimization using an-
nealed neural network', Automation in Construction,
15(4), pp. 531-539
[1] http://wecision.com
[2] http://designspaceconstruction.org
Tangible Grasshopper
A method to combine physical models with generative, parametric tools

Boris Plotnikov1 , Gerhard Schubert2 , Frank Petzold3

1,2,3
Technische Universität München / Chair for Architectural Informatics
1
pborisp@gmail.com 2,3 {schubert|petzold}@tum.de

The use of digital tools in the early, creative design process is the focus of an
interdisciplinary teaching and research project. Starting from the question of how
a seamless connection between physical and digital tools could be made possible,
the proposed method tries to bridge the gap between both methodologies and
provide intuitive, visual and collaborative design coupled with advanced, real
time computer simulations. A design platform has been developed which supports
a seamless connection between freely shaped physical models, GIS data and
Grasshopper3D. The environment combines the reconstructed physical models
with the digital one (surrounding buildings) and passes the information to a
custom Grasshopper3D plug-in which serves as a link to existing and custom
developed simulative tools. All simulations are performed and visualized in real
time to support the intuitive and iterative design process.

Keywords: urban design, tangible interface, grasshopper, sustainable design,

design decision support

INTRODUCTION requiring a long time to master. Such advantages

Physical models have always played a central role
throughout the design process for architects and ur-
ban planners. Such models help in analyzing and
understanding the context, intuitively convey de-
sign ideas and work collaboratively. In the last few
decades and with the popularization and enhance-
ments in the field of computation, computer tools
became prominent in the field of architecture, first
mainly as tools to enhance efficiency and produce
technical drawings, but in recent years also as deci-
sion support tools in a variety of fields such as cli-
mate, construction and mobility. As these tools grew
to be more advanced they usually also grew in com-
plexity, often resulting in a less intuitive process and
and limitations have been discussed in previous stud-
ies (Weytjens et al. 2012).
The proposed method tries to bridge the gap
between the physical and digital design methodolo-
gies and provide intuitive, visual and collaborative
design coupled with advanced, real time computer
simulations. A physical design platform has been de-
veloped which supports a seamless connection be-
tween freely shaped physical models, GIS data and
Grasshopper3D. The environment combines the re-
constructed physical models with the digital one and
passes the information to a custom Grasshopper3D
plug-in which serves as a link to existing and custom
developed simulative tools. All simulations are per-

DESIGN TOOLS | Evaluations - Volume 2 - eCAADe 34 | 127

 formed and visualized in real time to support the in-
tuitive and iterative design process.
BACKGROUND
As CAAD tools grew in popularity, their influence
and reach expanded to a wide array of spheres in-
cluding but not limited to structural, climate, en-
ergy use, acoustics and mobility analysis. The use
of these tools, which were hard to master and very
specific was usually limited only to specialists, result-
ing in a linear process where many evaluations were
conducted only after the design process for valida-
tion purposes and not assisting in informing the de-
sign. In recent years there have been multiple at-
tempts to bridge the gap between architects and
other professions involved in the building industry
by grouping and incorporating evaluation and simu-
lative tools in existing CAD and BIM tools, notably on
top of Rhinoceros and Grasshopper3D. Some of these
tools are Diva [1], a daylighting and energy modeling
plug-in and Ladybug and Honeybee (Sadeghipour-
Roudsari and Pak 2013), two open source environ-
mental plug-ins meant to help designers create an
environmentally-conscious architectural design. Re-
cently a plug-in called UMI (Reinhart et al. 2013) used
to evaluate the environmental performance of neigh-
borhoods and cities was also released. In contrast to
Diva, Ladybug and Honeybee, UMI focuses on the ur-
ban context, providing less detailed, larger scale re-
sults.
These and other tools simplify the evaluation of
designs by allowing the user to model and run sim-
ulations in a single program, eliminating the need to
master a variety of tools and minimizing the risk of
incompatibilities due to manually converted geom-
etry. While running some of the simulations inside
the plug-ins, more advanced simulations are being
run by verified tools in the background such as Radi-
ance (Ward and Shakespeare 2004) and Daysim (Rein-
hart and Walkenhorst 2001) for daylight simulations
and Energyplus (Crawley et al. 2000) and OpenStu-
dio (Guglielmetti et al. 2011) for energy and thermal
simulations. Although these tools help to inform the
128 | eCAADe 34 - DESIGN TOOLS | Evaluations - Volume 2
design from rather early stages, the process of set-
ting up the models and running the simulations is still
time consuming and the interaction with the model
is limited and requires technical expertise.
APPROACH
Typical design workflow is becoming increasingly in-
terlinked with digital tools, especially when address-
ing the neighborhood and the urban scale. As a di-
rect result, professionals involved in urban design
and planning require longer technical training, not
necessarily for their main expertise but in order to
keep up to date with current digital tools. The nature
of working in front of a screen does not promote col-
laboration, especially in the design realm, and peo-
ple from different fields without knowledge in the
used software are excluded from the process alto-
gether. As collaboration is at least at times manda-
tory, both for approval of projects and for discussions
between designers, consultants and decision mak-
ers, the design process is often paused and much
work is spent preparing a simplified and accessible
representation of the progress. At times such repre-
sentations are meaningful and concise, but more of-
ten they are simplified and partial and potentially not
representative of the bigger whole. Often the design
is paused and static representations such as images
and schemes are produced. A task which is time con-
suming but more importantly disconnected from the
actual design. One of the unfortunate byproducts is
that these collaborations are occurring seldom and
at later stages of the design, by which point fewer
changes are likely to occur.
Figure 1
Seamless
connection
between tangible
objects and
Grasshopper 3D.
The method proposed addresses these limitations
and suggests both a conceptual shift in common de-
sign practices and a practical solution which facili-
tates such a shift. By first minimizing configuration
and setup time by seamless GIS integration, a phys-
ical design platform and a set of predefined digital
simulations the process can start much faster and al-
lows the participants to focus on design rather than
on technical aspects. Once started, the process no
longer needs to be fragmented and disconnected.
The designer can work in an intuitive physical 3D en-
vironment by manipulating physical masses on top
of the digital data or in a digital environment tak-
ing advantage of commonly used and validated tools
and simulations, Rhinocorous3D and Grasshopper3D
serving as the digital environment as well as the
bridge for other software. Both processes comple-
ment each other as they are different ways of work-
ing with the same model, meaning no conversion or
synchronization is required. Such a setup is also more
convenient for showcasing and discussing designs in
larger and sometimes cross disciplinary groups. The
process does not need to be paused, represented
in another way for a different audience and then
converted back, rather one continuous iterative pro-
cess with an efficient feedback loop exists, harness-
ing both advanced digital tools and physical intuitive
design strategies.
By linking between digital tools with physical
models which require no configuration or program
specific knowledge, the proposed method addresses
the early stages of design, when very little apart from
the context is already known. As such the urban and
neighborhood scales are targeted and the method
provides quick and meaningful simulations which
evaluate design decisions such as massing possibil-
ities and their effects on surrounding buildings and
the public realm. By linking this tool with existing
design tools such as physical models, it is possible to
control and monitor simulations through already es-
tablished mechanisms. For example, changes made
to the physical model will have an immediate and di-
rect effect on the results of digital simulations and
analysis. This allows for a quick assessment and
comparison of the potentials and issues of design
variants, directed and determined by developments
made in physical models. It is also important to note
that more detailed and accurate simulations are to
be performed at the building scale at a later stage
when more about the design is known. Thus the pro-
posed method provides a complimentary and an in-
tuitive toolset for early design stages, supporting the
designer by providing real-time analyses and a qual-
itative comparison between alternatives.
Scenario
To illustrate the benefit of the proposed methodol-
ogy, a scenario describing the development process
of a new housing neighborhood was chosen. An ar-
chitectural firm has won a public competition orga-
nized by the city to design the master plan for the
site. There were already several preliminary meetings
in which the city officials expressed their main goals
and requirements and now the architects initiated a
meeting with the city officials and consultants in the
fields of transportation, mobility, climate, energy use
and acoustics in order to discuss several massing al-
ternatives and receive feedback. The participants are
gathered around a large multi-touch table and a big
screen, showing a plan view and a perspective of the
site respectively. Both surfaces show the latest dig-
ital model showing a massing proposal for the site.
As the architects describe the concepts they zoom
in and out using gestures on the surface of the ta-
ble. Other participants also engage and in addition
to walking around the digital model scale it to focus
on and discuss specific areas in the site in greater de-
tail.
The architects switch between several saved
variations to discuss design alternatives. At this
point, the climate consultants load a real time radia-
tion simulation to examine the amount of solar radia-
tion on the facades of the buildings to optimize orien-
tation, building heights and density. They identify an
area which seems to be problematic and immediately
remove several buildings and as they have a differ-
DESIGN TOOLS | Evaluations - Volume 2 - eCAADe 34 | 129
 ent idea regarding the placement of the massing in
that particular location, they take several Styrofoam
blocks available, trim them to fit the shapes they en-
vision and place them on the table. The digital model
and the simulation results are updated in real time to
reflect the changes. At this point the transportation
expert wishes to examine the distances to the loca-
tion of the planned school from all of the housing
blocks. She switches to that simulation and as she
is not satisfied with the results examines the effects
of moving the school to another location, reordering
some of the buildings and roads in the process. Now
the climate engineers run an outdoor comfort simu-
lation to examine the conditions in the park next to
the newly placed school.
That iterative process continues until the ex-
perts, city officials and architects are pleased with
the results. All of the participants were exposed to
the process and were able to contribute. In addi-
tion the experts in different fields were able to gather
together and work collectively to produce a syner-
getic design which is optimized not just in relation to
their area of expertise but as a complex system main-
taining a balance between many, sometimes oppos-
ing, requirements. Perhaps several alternatives were
saved during the process for future reference and as
catalysts for further development. At this point the
architects are able to proceed with the design, know-
ing that it fulfills the requirements from all of the
groups involved. There is no need for extra work to
integrate the conclusions from the meeting into the
design as all of the work has been done on the actual
model and the design process can continue without
interruptions, misinterpretations or time loss.
Related Work
As buildings grow in complexity and include an ex-
panding number of building related professions, so
does the need for better tools to improve communi-
cation and collaboration between those teams (Kalay
et al. 1998). While much of the research focused
on attempts to facilitate purely digital collaboration
by developing custom Information communication
130 | eCAADe 34 - DESIGN TOOLS | Evaluations - Volume 2
technology and software as described in the liter-
ary study conducted by Rahmawati et al. (2013), a
second, more physical approach advocating for end
users collaborating around a table has also been ex-
plored. URP (Underkoffler and Ishii 1999) is a sys-
tem for urban planning which runs simulations on
physical architectural models placed on a table sur-
face, including visualizing shade for any time of day,
throwing reflections off glass facade surfaces and vi-
sualizing 2D windflow. BUILD-IT (Fjeld et al. 2002)
an augmented reality (AR) system, loads digital 3D
models and allows users to select and manipulate
them by moving physical specialized bricks which
are used as interaction handlers. Offering users two
views of the model, the top view is projected on
the table and a side view is projected on a screen
nearby. Arthur (Penn et al. 2004) is an augmented re-
ality system which includes a head mounted display
and stereo cameras facilitating the 3D view of digi-
tal models. Placeholder objects and pointing devices
are used to interact with the virtual media in addition
to finger tracking and gesture recognition. More re-
cently and with the popularization of tangible user
interface object markers , programming languages
such as processing and parametric design environ-
ments such as Rhinoceros 3D and Grasshopper3D,
more direct and intuitive workflows became possi-
ble. (Salim et al. 2010) describe the design and de-
velopment of two such prototypes over a period of
four days. Tangible markers representing buildings
are placed on a special table, tracked and update a
virtual model. Participants manipulate the markers
which triggers a real time update of the virtual model.
Viola and Roudsari (2013) propose a workflow con-
sisting of an "observer" which tracks markers, an "in-
terpreter" which analyses and assesses the observed
model and a "visualizer" which displays the results
onto a design space. Markers represent predefined
buildings and can be freely manipulated on the de-
sign space. An incoming stream within Grasshop-
per3D interprets the data from the markers, recon-
structs a 3D digital model, performs environmental
analysis in near to real time and visualizes the results.
 The proposed method shares many similarities with
some of the described tools but breaks free from
the digital-physical constraints, allowing for free
form physical modeling which requires no pre-
configuration, does not limit the user regarding ma-
terials used in building the model, and provides ad-
vanced computer simulations.
SYSTEM SETUP
To implement the mentioned approach the system
setup includes two main parts (Figure 2):
• A Design environment (Figure 2 - A) called Col-
laborative Design Platform (CDP) (server ap-
plication) for urban planning based on a multi
touch table with additional 3D-Object recon-
struction to digitalize physical working mod-
els placed on the table in real-time (Schubert
et al. 2011) (Schubert et al. 2011) (Schubert
2014). The platform is designed to suite the
needs of an urban planning in scale 1:500.
• Custom Grasshopper3D component (client
application)(Figure 2 - B) , receiving 3D infor-
mation and user interactions from the design
platform. This component allows using the re-
constructed data (geometry) from the work-
ing models seamlessly in Grasshopper3D. This
connection works in real time and updates au-
tomatically.

Figure 2
System setup: (A)
Design Platform:
physical models
placed on the
Design Platform
table (Styrofoam
blocks) with
integrated 3D
object
reconstruction, (B)
Custom
Grasshopper3D
component
running and
visualizing
simulations.

DESIGN TOOLS | Evaluations - Volume 2 - eCAADe 34 | 131

 CDP - Collaborative Design Platform
The technological basis for the designed develop-
ment platform is a large-format multi-touch table
(Figure 3 - A) with an integrated digital 3D object re-
construction fun realtime realized with the help of a
3D depth camera on top (Figure 3 - B).
Using semantic GIS data as its basis, a digital plan
is displayed on the surface of the table at a scale
of 1:500. Physical objects forming design variants
can be placed on the multi-touch table and a digital
counterpart of them will be created automatically in
real-time - without the use of markers or any other
similar predefined methods. Both the digital plan of
the surroundings, as well as the reconstructed shapes
of the physical 3D objects can be seen in a perspec-
tive view on another, vertical mounted touchscreen
132 | eCAADe 34 - DESIGN TOOLS | Evaluations - Volume 2
(Figure 3 - C).This means that any physical objects
and their geometry will always have a digital equiv-
alent. These counterparts are created through a cou-
pled system. The footprint of the building is tracked
by the surface of the table and an additional on-top
mounted 3D depth camera scans the three dimen-
sional shape of the object as a point cloud. Two
construction algorithms were conceived to assist the
process and reconstruct the point cloud. One cap-
tures the footprint of the building and extrudes it.
The other records the free-form as a triangulated
mesh and connects it to the footprint. A more de-
tailed examination of this process can be found in the
paper (Schubert et al. 2013).
To support the creative design process, different
design approaches and the resulting use of differ-
Figure 3
System setup of the
CDP. A: Multitouch
table B: On-top
mounted 3D depth
camera C:
Perspective view of
the design scene D:
Flexible extension
via a network
protocol.
 ent tools, the hardware and the software of the table
were designed to be both flexible and extendable.
In terms of linking hardware, this is achieved using a
custom developed network protocol (Goldschwendt
et al. 2014)(Figure 3 - D). The CDP serves as a server-
based application and provides a client with all rel-
evant design information. This includes geometric
data from digital site models, physical working mod-
els and digitalised sketches. As a result external ap-
plications, such as Grasshopper, can access and use
the design information in real-time.
Grasshopper component
Figure 4 A custom component which facilitates the communi-
The Custom cation between the CDP platform and Rhinoceros3D,
Grasshopper3D including Grasshopper3D and all the other stimula-
plugin, receiving tive tools previously mentioned has been conceptu-
and reconstructing alized and developed with several key points in mind:
3D information over
TCP/UDP in real 1. Allow for a stable and persistent connection
time , making it between the CDP platform and Grasshop-
available for further per. This connection first establishes a link
manipulations in to the physical model and reconstructs it in
Grasshopper and in Grasshopper and then ensures that any up-
Rhino. dates to the physical model translate in real
time to Grasshopper and in turn to any fur-
ther logic which is defined. On top of cre-
ating a physical input stream, the compo-
nent extends upon the usual functionality of
Grasshopper by allowing for dynamically de-
fined input geometry which frees the user
from configuring input parameters and allows
them to focus on designing while providing
constant input from any running simulations,
maintaining a feedback loop which informs
the design.
2. Accurately translate existing physical geome-
try to the digital model. Rather than a position
of a generic or previously defined geometry,
the actual 3D representation of the model is
transmitted, processed and visualized by the
component. The system implementation sup-
ports a wide array of physical models, ranging
from extruded blocks (represented as planar
surfaces and heights) to free form shapes (rep-
resented as triangulated meshes). This opens
up an entirely new range of possibilities of in-
tegrating physical models in a digital platform
as users are not limited to any specific geom-
etry and have much more design freedom be-
ing able to build any shape from any material
and integrate it seamlessly into the digital set-
ting. The technology is not limiting the free-
dom of the design but rather adds additional
layers of information on top of it.
In order to accommodate for these requirements, to-
gether with the motivation to keep the configuration
as fast and easy to operate as possible, it was decided
to implement a single Grasshopper component. A
TCP (for reliably establishing connections) and UDP
(for streaming data once the connection has been
established) client which was described in the previ-
ous section has been adjusted and maintains the live
stream between the CDP platform and Grasshopper.
An additional program was developed in order to cre-
ate a bridge to a Grasshopper component (Figure 4)
whose input parameters consist of the connection
details, custom configuration settings and a Boolean
flag allowing to turn live streaming of data on and
off. The output parameters consist of the footprint
surfaces and heights for extrusions and mesh geom-
etry for free form shapes. A distinction is made be-
tween the physical and the digital geometry passed
from the CDP platform. The physical would often cor-
relate to the actual design while the digital to the sur-
roundings. That separation is particularly important
DESIGN TOOLS | Evaluations - Volume 2 - eCAADe 34 | 133

when running simulations where a clear distinction
between the context which is fixed and the design
which is subject to change.
Once data is received and processed by the
Grasshopper component, it is available for the vast
array of manipulations and simulations which exist
either natively in Grasshopper or in custom plugins.
Though most of the development focused on pro-
viding a generic solution to serve as a base for other
Grasshopper definitions, several workflows, mostly
in the field of climate simulations, have been devel-
oped further as test cases. These range from rather
straightforward simulations such as shading studies
up to radiation, energy use and outdoor comfort sim-
ulations. To better illustrate the proposed workflow,
an implementation of a solar rights envelope simu-
lation based on the method proposed by Capeluto
and Shaviv (2001) which determines the maximum
height of new development in a site while preserv-
ing direct sun access to surrounding buildings is pre-
sented (Figure 5).
134 | eCAADe 34 - DESIGN TOOLS | Evaluations - Volume 2
Figure 5
A custom Solar
rights envelope
simulation
component in
Grasshopper
showing an urban
context and a 3D
polysurface
representing
maximum heights
which new
development
should not exceed
in order to not
block direct sun
access.
The process begins as a digital model is loaded to
the CDP platform and is extended with physical ob-
jects. Once the Grasshopper plugin has been config-
ured to the IP address and port of the CDP platform
a live stream first constructs the combined model
in Grasshopper and then responds in real time to
changes made to the model. From this point data is
available for further manipulations in Grasshopper.
The solar rights plugin has several inputs- the ge-
ometry on which solar access is tested on, the geom-
etry which potentially blocks solar access, and con-
figuration settings such as grid size and sun positions
tested upon. The output is a polysurface represent-
ing the maximum heights of the new buildings which
would still preserve solar access to the surroundings.
The polygons of the surrounding buildings are
chosen and their facades are extracted and passed to
the solar rights plugin as is the geometry of the physi-
cal objects. The other settings are preconfigured and
can be optionally overwritten for advanced usage.
Once the simulation is set to run, a polysurface ap-
pears in a certain relation to the new development
and another component checks for an intersection
between the physical buildings and that polysurface.
If intersections exist, meaning there is a height devi-
ation, the new buildings that are harming the solar
access rights of the existing ones are colored red, al-
lowing for an intuitive way to identify the problem-
atic areas. Any time the physical objects are removed,
added or moved, the simulation is re-triggered and
the results are updated in real time, creating a con-
stant feedback loop which helps to inform and im-
prove the design in real time.
SUMMARY
The prototype presented in this paper allows for
direct control of digital geometry in Grasshopper
3D using tangible interfaces. Due to the seamless
connection of physical models and digital content,
physical models of urban design strategies can be
analyzed directly in Grasshopper. One of the no-
table advantages of the proposed method relates to
the tangibles, serving not only as placeholders in a
Grasshopper and a Rhino scene, but are free form
physical objects which are digitally reconstructed
and allow for further analysis and simulation tools
from Grasshopper to be directly used in a physical
model. In addition parametric relationships can be
programmed in grasshopper and be controlled by
the tangible interfaces. One of the challenges at
the moment concerns the high computation time of
some analyzes. Furthermore, in order to accurately
conduct some of the mentioned simulations, more
detailed information which is usually not known dur-
ing the early design stages must be supplied. Besides
urban scenarios other forms of usage for parametric
design in Grasshopper are conceivable, as any form
of placed objects on the table can be used directly as
a starting point for digital manipulations.
FUTURE DEVELOPMENT
One area of further research revolves around the de-
velopment and incorporation of tools to quickly eval-
uate thermal and visual outdoor comfort in the pub-
lic realm. Such research will focus on simulating mi-
croclimate and spatial comfort maps. An additional
area of development concerns visualization of simu-
lation results. Currently the visualized results are pre-
sented on screens as perspective, side views and a
juxtaposed top view incorporating the digital and the
physical model. As the method described in this pa-
per tries to simplify the design process and make it
more intuitive and tangible, two further visualization
techniques are being developed:
1. Projecting the results of the simulations on
the facades of the physical model in addition
to the 2D visualization.
2. Incorporating augmented reality techniques.
The proposed approach opens up new possibilities
for the early stages of design and decision making
process. This is achieved by offering an intuitive and
collaborative tool with which architects, urban plan-
ners, community leaders and policy makers are able
to simulate and evaluate implications of design de-
cisions in the urban and neighborhood scale in the
ongoing attempt to create healthier, sustainable and
livable cities.
REFERENCES
Capeluto, I.G and Shaviv, E 2001, 'On the Use of Solar Vol-
ume for Determining the Urban Fabric', Solar Envery,
70, pp. 275-280
Crawley, DB, Pedersen, CO, Lawrie, LK and Winkelmann,
FC 2000, 'EnergyPlus: Energy Simulation Program',
ASHRAE Journal, 42, pp. 49-56
Fjeld, M, Lauche, K, Bichsel, M, Voorhorst, F, Krueger, H
and Rauterrberg, M 2002, 'Physical and Virtual Tools:
Activity Theory Applied to the Design of Groupware',
Computer Supported Cooperative Work, 11, pp. 153-
180
Goldschwendt, T, Anthes, C, Schubert, G, Kranzlmüller,
D and Petzold, F 2014 '[DEMO] The Collaborative De-
sign Platform Protocol: A Protocol for a Mixed Reality
Installation for Improved Incorporation of Laypeo-
ple in Architecture', Proceedings of the International
Symposium on Mixed and Augmented Reality (ISMAR)
Guglielmetti, R, Macumber, D and Long, N 2011 'OPEN-
STUDIO: An open source integrated analysis plat-
form', Proceedings of the 12th conference of interna-
DESIGN TOOLS | Evaluations - Volume 2 - eCAADe 34 | 135
 tional building performance simulation association, China, pp. 210-221
Sydney,, pp. 442-449 Underkoffler, J and Ishii, H 1999, 'Urp: A Luminous-
Kalay, YE, Khemlani, L and Choi, J 1998, 'An integrated Tangible Workbench for Urban', proceedings of the
model to support distributed collaborative design SIGCHI conference on Human Factors in Computing
of buildings', Automation in Construction, 7, pp. 177- Systems, 1, pp. 386-393
188 Viola, A and Roudsari, MS 2013 'An innovative workflow
Penn, A, Mottram, C, genSchieck, AF, Wittkämper, M, for bridging the gap between design and environ-
Störring, M, Romell, O, Strothmann, A and Aish, F mental analysis', Proceedings of BS2013, Chambéry,
2004, 'Augmented Reality Meeting Table: a Novel pp. 1297-1304
Multi-User Interface for Architectural Design', in Ward, G and Shakespeare, R 2004, Rendering with RADI-
Leeuwen, Jos P. Van and Timmermans, Harry J.P. ANCE. The Art and Science of Lighting Visualization,
(eds) 2004, Recent Advances in Design and Decision Booksurge Llc
Support Systems in Architecture and Urban Planning, Weytjens, L, Macris, V and Verbeeck, G 2012 'User Pref-
Springer, pp. 213-231 erences for a Simple Energy Design Tool: Capturing
Rahmawati, Y, Utomo, C, Anwar, N, Setijanti, P and Nurc- information through focus groups with architects',
ahyo, CB 2013, 'An Empirical Model for Successful 28th PLEA Conference, Lima
Collaborative Design Towards Sustainable Project [1] http://diva4rhino.com
Development', Journal of Sustainable Development,
7, pp. 1-14
Reinhart, CF, Dogan, T, Jakubiec, JA, Rakha, T and Sang,
A 2013 'UMI- an urban simulation environment for
building energy use, daylighting and walkability',
Proceedings of BS2013, Chambéry
Reinhart, CF and Walkenhorst, O 2001, 'Dynamic
RADIANCE-based Daylight Simulations for a full-
scale Test Office with outer Venetian Blinds', Energy
and Buildings 33, 33, pp. 683-697
Roudsari, MS and Pak, M 2013 'Ladybug: a paramet-
ric environmental plugin for grasshopper to help
designers create an environmentally-conscious de-
sign', Proceedings of BS2013, Chambéry
Salim, FD, Mulder, H and Jaworski, P 2010 'Demonstra-
tion of an Open Platform for Tangible and Social
Interactions with Responsive Models', ACADIA 2010,
New York,, pp. 227-233
Schubert, G 2014, Interaction forms for digital design:
a concept and prototype for a computer-aided de-
sign platform for urban architectural design scenarios,
Ph.D. Thesis, Technical University of Munich
Schubert, G, Artinger, E, Petzold, F and Klinker, G 2011 a
'Bridging the Gap: A (Collaborative) Design Platform
for early design stages', Proceedings of eCAADe 2011,
Ljubljana , pp. 187-192
Schubert, G, Artinger, E, Petzold, F and Klinker, G 2011
b 'Tangible tools for architectural design: seamless
integration into the architectural workflow', Proceed-
ings of ACADIA 2011, Banff, pp. 252-259
Schubert, G, Riedel, S and Petzold, F 2013 'Seamfully con-
nected: Real working models as tangible interfaces
for architectural design', Proceedings of 15th Inter-
national Conference, CAAD Futures 2013, Shanghai,

136 | eCAADe 34 - DESIGN TOOLS | Evaluations - Volume 2

A Simple System for Complex Mass Housing Design
Collaborations
A system development framework

Tian Tian Lo1 , Marc Aurel Schnabel2 , Tane Moleta3

1,2,3
Victoria University of Wellington
1,2,3
{sky.lo|MarcAurel.Schnabel|Tane.Moleta}@vuw.ac.nz

Through the lens of participatory mass housing the paper explores the conference
theme of simplicity and complexity. A suitable home is a deep rooted desire in the
heart of people, and everyone has their own vision of what is a suitable home. Yet
the multi-faceted social needs of housing and how they are being designed and
developed in mass housing buildings appear too complex and appear too costly
that the process would involve direct user participations. The authors have
developed a Computer Aided Participatory Housing Design System (CAPHDS) to
allow end users (future occupants) become active stakeholders in the design
process with the aid of computational design instruments. These tools allow end
users to actively engage in the process. The paper describes how a mass housing
design process can be broken down into a set of simple tasks that encourage the
active engagement and joint development of end users and architects with the
proposed design.

Keywords: Participatory design, housing design system, computer-aided,

bottom-up

INTRODUCTION to enhance the architect-user relationship and gen-

'The idea of utopias not only courts criticism of being
'unrealistic' or 'idealised', it also has a whiff of revolu-
tionary fervour.' - John Wood, 2007.
Individual needs cannot remain constant in our
ever changing world. As each citizen plays their role
in society, they build their unique version of 'life'. It
becomes a significant challenge to provide equitably
for all and allow, as Thomas More (1966) puts it in
his definition of utopia, for each to lead a 'cheerful
life, free of anxieties.' Yet, we believe that Wood's call
for dreamers is necessary to elucidate possible ways
erate a built environment that matches the vision of
the user.
Wood (2007) argues that one way to achieve
'micro-utopia' is to rely more on a 'distributed mode
of action and responses,' to use a strategy of bottom-
up, rather than top-down navigation. In the field of
housing, this aligns with John Harbraken's concept
of 'open housing' (1972). According to Harbraken's
description to navigate from bottom-up, it is nec-
essary to attract useful feedback through an 'Open
Source' design that is transparent. The role of compu-

DESIGN TOOLS | Evaluations - Volume 2 - eCAADe 34 | 137

 tational media is thus seen less as an instrument to in-
crease the rate and quality of transactions, but rather
as one enhancing the efficiency of communication
and in turn generating an outcome that is closer to
the user's needs.
By envisaging user participation through effi-
cient communication as the key to the achievement
of a collaborative collective housing, we support
Wood's statement that 'we have become experts at
choosing and complaining but forgotten how to en-
visage what we really want' (2007 pg. 12). In sum-
mary, this paper is making a daring proposition by
proposing a possible solution within this complex
context: a framework for a Computer Aided Partic-
ipatory Housing Design System (CAPHDS) that pro-
vides a communicative and interactive platform for
end users (future occupants) to envision and commu-
nicate their housing desires within their capabilities.
By offering end users an instrument that enhances
their communication and expression of their dreams,
the possibility of user participatory mass housing is
explored.
AN APPROACH TO DEVELOPING CAPHDS
Diving straight to the participatory context in com-
puter aided design systems, the diverse and chang-
ing perceptions of the participants makes it increas-
ingly difficult to determine if a design proposal of a
mass housing building is capable of being generated
from such systems. The primary challenge faced by
this collaborative design system is shuffling between
a simplification of the design parameters and details
for end users with no architectural background and a
complexity for the architects who need a rich set of
profession specific details and data to ensure a build-
able and successful project delivery. The idea of an
ideal living space can be easier achieved if it is a stan-
dalone house, which don't have to fulfil overall collec-
tive needs in a mass housing building that is charac-
terised by a network of partly conflicting design pa-
rameters - without yet addressing individual needs or
requirements. Although this paper sets out to pro-
pose a possible solution, a clear design structure and
138 | eCAADe 34 - DESIGN TOOLS | Evaluations - Volume 2
defined assumptions have to be set up to backup this
theory.
The participatory design development is broken
down into three main stages; a) Formulation of the
collaborative design workflow, b) Generation of the
design collaboration and c) Evaluation of the design
outcome. This paper is concerned with the third
phase, the assessment of the outcome by applying
design analysis criteria and benchmarking it against
sustainable housing developments and Open Build-
ing Principles (OBP) as developed by Kendall et al.
(2010). Emphasis is placed on providing a good level
of communication flow and a system usability allow-
ing laypersons to interact with the system and the
proposed design intuitively.
The definition of John Wood's micro-utopia is
very similar to what CAPHDS is trying to achieve, the
difference is the scale of the condition. To facilitate
the development stages, close reference is also taken
from John Wood's (2007) five steps towards micro-
utopia: i) breaking through participants' psychologi-
cal barrier, ii) providing a means for co-sharing, iii) de-
termining the participants' needs versus desires, iv)
generate the outcome for visual observation and ex-
change, v) engaging the participants towards a prac-
tical outcome. From the steps, we can identify the
challenges for a better communication system to en-
courage more collaborative and participation in the
design process.
To demonstrate the possibilities of the proposed
system, on based on the criteria of CAPHDS an in-
strument, called ModRule, has been developed by the
author and is described in detail in Lo et al. (2015).
ModRule is designed to enable end users of a mass
housing building to communicate in the initial de-
sign phase their desires and wished to each other as
well as to to the architects by setting rules and pa-
rameters that are crucial to end users. The design
tool is able to work diversely to generate a wide va-
riety of design options for every individual occupant
as well as to negotiate conflicting interests and out-
comes. Subsequently it is an instrument that allows
architects to cooperate closely with potential inhabi-
Figure 1
Five major
components of
mass housing
design: spatial
driver, structure
frame, skin
modules, inner
partition system
and utility system.
tants of a mass housing building development and to
react to users' wishes and desires. The setup, there-
fore, is quite different from a typical design process.
This paper will explain how ModRule as a system, pro-
vides the capability to address these points and what
are the many challenges that remain to be resolved
in the design process.
FORMULATION
The simplification of design starts from the begin-
ning of the CAPHDS development process. As Gao
(2012) describes the formulation of a collaborative
design workflow can be initiated by breaking a down
mass housing design into five major components:
spatial driver, structure frame, skin modules, inner
partition system and utility system (Figure 1). This
subdivisions allow for design flexibility to occur and
for the architects to have control of every aspect of
the design. Using a Building Information Modelling
(BIM) methodology components are not separated
and rather linked together to correlate and maintain
information throughout the process. The five com-
ponents offer the possibility to control information
selectively which can be published as open source for
others to use or contribute to the libraries of informa-
tion.
Within these five components countless possi-
bilities to develop meaningful design outcomes can
be generated (Figure 1). The two components, skin
modules and inner partition system, allow sufficient
flexibility to design solutions that response to the de-
sign parameters as set by the end users. By opening
up these two components for user participation, the
architects will have to design the other three compo-
nents specifically to accommodate the design varia-
tions. In other words, the architects have two critical
operations; one, to construct the three components
which include creating the basic building form, de-
veloping the structure frame accordingly and setting
the public space and utilities. And two, to establish
the variation schema of the other two components
which include laying out a range of skin module and
working out the arrangement of the inner partition
system allowing types of input parameters such as
daylight, sky-view factor, accessibility, thermal radia-
tion, cost, etc. In this situation, a BIM model is neces-
sary to manage the data and connecting the data of
the dynamic open source model to the construction
documents without consuming extra resources. The
challenge for the architects is providing a BIM model
that can respond to the wide spectrum of data that
arrives from the various design options generated by
the end users.
With the design workflow structured properly,
ModRule facilitates the collaborative tasks of this
stage of the design development (Lo et al. 2014).
The basic design of the mass housing building is di-
vided into grids of various sizes depending on the
architects' preferences, and every grid space can be
assigned with parameters and variables. The archi-
tects first layout the building plan according to nec-
essary elements such as building core, access- and
exit routes, utilities, etc. (Figure 2). The grids have
parameters attached to them such as space require-
ments, budget, orientation, daylight preference, etc.
quantified and predefined by the architects. The end
DESIGN TOOLS | Evaluations - Volume 2 - eCAADe 34 | 139
 users then are provided with their individual user
profile that is created based on a profiling question-
naire that users complete prior to using ModRule.
The user profiles quantify the various needs and de-
sires of end users with respect to the set of parame-
ters established by the architects. This acts as a guide
for the end users while working with ModRule to de-
sign their desired way of living.
In principle the predefined grids in ModRule can
be designed in any manner as the architects requires
it. Taking reference of a Japanese tatami mat which
is has a standard dimension. The layout of Japanese
houses and apartments are based on the dimensions
of one tatami, which acts a base unit of space. The ar-
chitects set the dimensions of the base grid in Mod-
Rule (Figure 2) and users design their spaces based
using the grids fields as units by selecting the grid-
cells that are within the values sets of their user pro-
files.
After setting up the plan layout with its grid, the next
step is to define the skin module and the inner parti-
tions. The amount of variation can be controlled by
the architects to keep the design economically and
efficient. With these all defined and set up, the de-
sign is ready to be open up for end users to develop
and generate the design of the mass housing collab-
oratively.
140 | eCAADe 34 - DESIGN TOOLS | Evaluations - Volume 2
GENERATION
The generation of the design collaboration is the
next stage to enhance the design participation fur-
ther. Referring back to Wood's steps, the first step is
neither technological nor political but psychological.
Yet, to break this psychological barrier, technology is
necessary with the help of a strategy: Gamification.
Gamification functions both to promote partic-
ipation and simplify the process of communication
between various parties. Through the involvement
of end users, the complexity of the design process
is likely to increase especially in the context of mass
housing. Recognition of a layperson's interest in
the conceptual design stage necessitates immedi-
ate communication with the architects and opens up
the problem of dovetailing layperson demand, and
maintenance of the professional architects' quality
control (Schnabel et al. 2014).
SimCity and MineCraft are two computer games
frequently utilized in relation to architectural tasks
Figure 2
(Sanchez 2015). Although they are open world
Interface of
games where the players are required to design ev-
ModRule system
erything from 'tabula rasa', the players are confined
showing the fixed
and limited by rules and logics such as working in a
building core and
grid environment and using building elements pro-
parameters
vided by the system. In comparison, role-playing
(coloured) set by
computer games (RPG) such as Mass Effect and Un-
the architect.
til Dawn, which follow a particular story-line, require
players to make decisions that change the course of
the story resulting in individual and unique endings
for each participant. The players simply need to make
choices. They are not required to keep score or fol-
low complex rules. The story, however, will be ad-
justed according to the choices made by the players
resulting in different story endings. These tasks in the
game are stored automatically in a 'black box' in the
background freeing the player to focus on the game
story.
Similarly, it is possible to simplify the CAPHDS de-
sign interface by shifting the complex computational
processing into a black box so that end users partici-
pate in the design process with ease while architects
export the required information from the black box
 to guide their production of the models. In doing so,
the system could adjust the design outcomes accord-
ing to the choices made by the end users resulting in
design variation for each of them (Figure 3).
Figure 3
Characteristics
comparison
between computer
game interactions
and mass housing
In ModRule, the design collaboration with the users
design
starts with setting a goal system based on their per-
participation.
sonal profile. This goal system is like a checklist of
targets that end users shall follow (Figure 4). This
is taken in reference to 'objectives' in games where
players try to achieve while playing the game. A 'goal
bar' is available in ModRule interface that is interact-
ing with the parameters and variables input in each
grid. The goal bars fill up accordingly while choices
are being made by the end users, indicating whether
they are achieving what they declared they wanted.
Figure 4
Interface of
ModRule system
showing the fixed
spaces set by the
architects with the
layout ‘selected’ by
four end users
according to their
goals and profiles.
The role of the goal bar is for every individual to man-
age their desires. This is to ensure that users are not
designing a living space that is bigger than or over-
functioned of what they specified in their user pro-
files. The info of users' goal bars in available to ev-
ery participant to encourage sharing, understanding,
and social interaction. Ideally, the user transparency
will encourage every participant to help others to
achieve both, their personal and their common liv-
ing space since all participants will finally be living
together in their future building that they are jointly
developing.
Depending on the project it is almost impossi-
ble to fulfill all personal goals because of the need to
negotiate one's goals with the other end users, yet
the certainly aim and personal desire is to achieve as
many as possible. There might be occasions where
goals will have to be compromised to achieve others
or a solution that fits the common interests. These sit-
uations will all be negotiated by the end users them-
selves offering every participant a better understand-
ing of the overall process and a possible higher ac-
ceptance of the outcome.
Once everyone's goals are more or less achieved
up to a minimum threshold (i.e. above 80%), to a
point where everyone has agreed on the outcome
(Figure 4), the most important stage starts. This is
where the individual living spaces are synthesised to
generate a common overall architectural design. The
'ingredients' which were prepared at the earlier stage
will be applied where users select and design their
space with these ingredients At this stage, the ar-
chitects take the leading role. Although the aim is
to achieve an overall design through a bottom-up,
'democratic' process, the top-down is not negligible
especially in this context of mass housing. There are
too many building and architectural issues that need
the knowledge of professionals.
A CASE STUDY
To verify the proposed CAPHDS development stages,
a case study was carried out. This case study was an
initial test with ModRule. The main aim of this pa-
per is to analyse the design outcomes and to provide
an evaluation framework for the next development
of the system.
A few designers were selected to use ModRule
and set up the rules and parameters. After every de-
signer has set his or her design framework and pa-
rameters, discussion and a trial 'plays' were carried
out with each design to rate advantages and disad-
vantages in terms of flexibility and constraints of the
set parameters. One design (Figure 5) was subse-
DESIGN TOOLS | Evaluations - Volume 2 - eCAADe 34 | 141
 quently chosen for further development into a full
mass housing building. This design provides clear
parameters with sufficient restrictions to test a full
end user participatory design process.
Eight family profiles were developed to reflect a vari-
ety of end users' profiles (Figure 6). The available pa-
rameters for the profiles were set up based on some
key housing design parameters (Levitt 2009), such
as daylight, privacy, sky-view, sound, wind, height,
orientation, view, accessibility and needs for elderly.
Kazue (2005) states in her research that it is be very
difficult for end users to answer what they want, yet
it is much easier if they have to respond to ques-
tions about the qualities and problems do they find in
their current living spaces. In our case, the questions
asked took reference to Kazue's findings and the user
profiles contained questions about the participants'
142 | eCAADe 34 - DESIGN TOOLS | Evaluations - Volume 2
background characteristics such as their daily rou-
tines, habits, interests, likes and dislikes that are rel-
evant for the design process. User profiles that do re-
flect the real situations of the participants are neces-
sary to be taken into account from the beginning of
the design phase. Placing families together without Figure 5
consideration of these characteristics has proved to Chosen design with
create dissatisfaction among them (Kazue 2015). set parameters.
These profiles are then converted into a goal
system where ModRule will use to determine if the
choices made are simple desires or needs. Also, when
conflicts arise during the design collaboration pro-
cess, the information can be used as a basis for ne-
gotiation. During this case study, there were min-
imal conflicts and everyone was working together
to achieve a final design that best-fit everyone. The
transparency of information appeared to contribute
positively to the process.
As ModRule was still a prototype and not ma-
tured enough for a layperson to manipulate the
space intuitively, each family had one representative
designer to carry out the actual design according to
the profile and instructions. For example, a family of
four shall not be located beside a single family who
opposes children, an elderly family is preferably lo-
cated at the lower level and closer to the lift core to
have better access to the ground level, etc. In or-
der for every role to fulfill their desired living envi-
ronment, multiple rounds of discussion and negotia-
tion had to be performed. The parameters enhanced
the negotiation with a clearer understanding of each
other's needs and criteria. The designers will then
based on the finalised outcome (Figure 7), design the
units in detail as a whole.
EVALUATION
With this case study, we can summarize CAPHDS to
two main objectives: i) aiding end users to envision
their housing needs and ii) provide a platform for
sharing and communication between the architects
and end users during the design process. This ontol-
ogy would be used to ensure consistency in the eval-
uation of the design outcome. Since the system al-
Figure 6
Family profiles
based on
background
characteristics and
housing design
parameters.

lows the end users to set their own criteria and tar- Enhancing communication:
get values, it is possible that, although the targets
• Do architects find the inclusion of client ideas
are met, the end users may still not be satisfied when
possible using the system of parameters?
they compare their outcomes with those of others.
• What determine the effectiveness of the sys-
This brings us to the final stage of the CAPHDS
tem in achieving the CAPHDS criteria?
development; the evaluation of the design outcome.
• Does the provision of bonus points (dollar
To evaluate, we need to first define the specific ques-
reductions?) encourage collaboration and
tions. These questions are defined from the observa-
bonding of the user community?
tion of the case study to achieve the two main objec-
• Do they regard the process as useful and effi-
tives of CAPHDS.
cient in relation to cost and time?
Aiding end users:
Although the designers show great interest and sat-
• Do potential buyers find it helpful to use a
isfaction from the design outcome, it is very hard to
game system to investigate potential living
determine if the CAPHDS answers to the questions
spaces?
directly. This initial case study helps to determine a
• Is their discovery of limitations to fantasised
clearer evaluation framework that could be used for
options because of achievability concerns
a extended study with an enhanced ModRule system.
likely to frustrate them and lead to withdrawal
In addition, the above presented case study worked
from the process?
only with one model and one group of users which
• Does a colour scheme to inform users that
does not provide any basis for deeper observation
choices are achievable improve the outcome?
and analysis.
• Does the design parameters act as a medium
of negotiation between users to resolve con-
flicts? DISCUSSION
• What actually determines user satisfaction The case study provided a list of components that are
with outcomes? necessary ingredients for CAPHDS to be able for end
users to envision their living spaces.

DESIGN TOOLS | Evaluations - Volume 2 - eCAADe 34 | 143


The system has to be able to generate a design
flexible enough for open participation and permit a
layperson to engage in the design process intuitively.
The components listed are based on OBP as they pro-
vide a clear discrimination of various building com-
ponents, which dovetail with CAPHDS. The evalua-
tion includes a way of assessing OBP criteria to de-
termine the effectiveness of the system in achieving
the aims. This list can then be later used for further
development of ModRule:
• A clear understanding of responsibility in rela-
tion to community coherence and individual
freedom
• Flexibility of customization options that main-
tain optimized efficiency in production
• Flexibility to adjust to variation in family types
• Clear distinction of decision-making levels
with the aid of professional tools
• Interchangeability of building components
• Clear separation of different building subsys-
tems
• Assurance that building performance can be
considered and easily understood by end
users
144 | eCAADe 34 - DESIGN TOOLS | Evaluations - Volume 2
• Assurance of design responsivity to needs of Figure 7
end users after changes Final design by
• Responsivity of system operation to end users designers
without need for technological assistance collaboratively with
• Supervision to minimise conflict during col- the choices made
laboration by families.
• Setup of parameters to match end users pro-
file
Taken from this list, the key towards a CAPHDS re-
volves around design parameters. ModRule is de-
signed such that every grid space does not only
have one but multiple parameters and at least one
of the parameters is of negative nature. Going back
to game logic, this is to ensure every player under-
stands the conditions for their choices, considering
them carefully, and not blindly choosing the 'best'
spaces. This also shall prevent every player choos-
ing the same grid space resulting in unnecessary con-
flicts. In ModRule, bonus points are given to end
users when their choices benefit their surrounding
neighbours. This would encourage end users to col-
laborate and foster a bond among the community.
Simple parameters could also limit the imagina-
tions of end users to achievable options within the
building structure. Because this information may be
difficult for laypersons to grasp, the implementation
of a simple colour scheme shall inform the end users
whether their choices are practical.
If the parameters are set properly, they can pro-
vide more than just being design attributes. These
parameters can also act as a medium of exchange
between end users. This is very useful when con-
flicts occur between them during the participation
process.
In general, there are two key challenges to the
practicability of CAPHDS: i) The perceptions of the
participants are diverse to begin with but also change
in an interactive context, thus creating further com-
plexity; and ii) A client-architect collaboration re-
quires a structure that provides the simplification
that laypersons need and the depth and complexity
necessary for the architects. These two challenges
can be addressed with ModRule that allows for dif-
 ferent levels of negotiation and engagement.
Gamification is a key strategy in dealing with the
challenges. Excluding the high-quality graphics and
intense gameplay to captivate the interest of play-
ers, a game design also requires dealing with a broad
range of players. By choosing the right group of tar-
get players, the game mechanics and parameters are
set accordingly to allow constant or increasing en-
gagement of the players. Although housing design
does not involve any gameplay on the surface, the
complex management and algorithm are being used
to develop games such as SimCity. The intention is
to use the same strategy by referring back to these
games and determine how we can create a system to
engage users in a real design context.
FURTHER DEVELOPMENT
Although there are clear research questions, the first
case study only provides a groundwork for improve-
ment of ModRule and the setting up of parameters.
Subsequently a second study is planned to explore,
observe and analysis the design outcome. This sec-
ond study has to involve two entirely different mod-
els to provide comparison and reduce bias in review-
Table 1
Characteristics that
various parties take
into consideration.
ing the results.
The architects will develop a CAPHDS model
(with ModRule) and a Fully Designed (FD) model
(conventional 3-4 rooms types). A sample group of
new first-year architecture students will be assigned
the task of investigating a unit in a medium density
housing project using the CAPHDS model. The rea-
son for choosing this sample group is - besides the
easy access to the pool of participants by the research
team - that they can be regarded as layperson with
minimum architecture knowledge. They are also a
group of people who has a certain degree of desire
to design their own living space. This provides the
characteristics closest to the targeted end users that
this research is ultimately looking for.
The students will then try out the system: share
their ideas among themselves, collaborate with each
other to achieve their desired outcomes, negotiate to
reduce conflicts and develop the design results with
the architects. With the CAPHDS model developed,
similar groups of student will then be asked to com-
pare their design result with the architects' FD model.
Two questionnaires will be given to each student and
the architects, one for them to differentiate between
DESIGN TOOLS | Evaluations - Volume 2 - eCAADe 34 | 145
 the two models and the other to understand if the
system provides them with the capabilities and satis-
faction to generate the outcome.
To reduce biases, the two models will be opened
up for the general public to vote and examine. The
bias in this case mainly refers to the group of stu-
dents. There is a possibility that the time and effort
taken to develop the CAPHDS model is simply the
factor that they prefer it to the FD model. By engag-
ing the general public, this allows to counter the bias
and to receive feedback which design appears to be
more accepted, the CAPHDS or the FD model. This
could just be achieved by a voting for the preferred
design.
We developed a summary of the processes and
considerations for the evaluations as presented in Ta-
ble 1. There are five parties: the architects and end
users who engaged with the CAPHDS and the gen-
eral public, officials and professionals. They will eval-
uate the design and system accordingly. The ques-
tionnaires act as a checklist based on the CAPHDS cri-
teria to evaluate whether design and system met its
purpose. If these parties accept the outcome, a high
degree of micro-utopia has been achieved due to the
direct proportionality to the evaluation scale.
CONCLUSION
This paper presented CAPHDS as the next develop-
ment for mass housing design. ModRule is designed
based on CAPHDS to demonstrate the potential of
using such a digital system for participatory process.
A case study is presented to generate a detailed eval-
uation framework to determine if the system does
provide the necessary means to aid end users to
achieve their desired design. Although this paper
tries to provide a simple tool for the users to de-
sign their space freely, a complex but well-structured
system is definitely necessary to support the pro-
cess. Since the target audiences are public users, few
rounds of studies have to be carried out to fine tune
the system performance and if the users are engag-
ing the design collaboratively and not challenging
each other. The simplicity of the output, in this case,
146 | eCAADe 34 - DESIGN TOOLS | Evaluations - Volume 2
is directly proportionate to the complexity of the lo-
gistic. However, with the help of computation, this
could help allow architects to engage in participatory
design processes of mass housing design whereby
end users are active contributors of the design out-
come.
REFERENCES
Habraken, NJ 1999, Supports: An Alternative to Mass
Housing, Urban International Press, UK
Kazue, D 2005, Only one of my home in the world (in Chi-
nese translated from original Japanese text), Kodan-
sha, Japan
Kendall, S 2000, Residential Open Building, Spon Press,
London & New York
Levitt, D 2009, The Housing Design Handbook: A Guide to
Good Practice, Routledge, UK
Lo, TT, Schnabel, MA, Aydin, S and Shi, K 2014, 'Mod-
Rule: Using Gamification for Collaborative Mass-
Housing Design Process', in Madeo, F and Schn-
abel, MA (eds) 2014, ACROSS: Architectural Research
through to Practice, Architectural Science Associa-
tion (ANZAScA), Genoa, p. 733–743
Lo, TT, Schnabel, MA and Gao, Y 2015, 'ModRule, A User-
Centric Mass Housing Design Platform', in Celani, G,
Sperling, DM and Franco, JMS (eds) 2015, Computer-
Aided Architectural Design Futures. The Next City -
New Technologies and the Future of the Built Envi-
ronment, Springer Berlin Heidelberg, Sao Paolo, pp.
236-254
More, T 1966, Utopia (in Latin), Scolar Press Ltd, UK
Sanchez, J 2015 'Block'hood - Developing an Architec-
tural Simulation Video Game', Real Time - Proceedings
of the 33rd eCAADe Conference, Vienna, pp. 89-97
Schnabel, MA, Lo, TT and Aydin, S 2014 'Gamification and
Rule Based Design Strategies in Architecture Edu-
cation', DesignEd Asia Conference Proceedings, Hong
Kong, pp. 150-160
Wood, J 2007, Design for Micro-Utopias: Making the Un-
thinkable Possible, Gower, England
Evaluating Daylighting Analysis of Complex Parametric
Facades
Mohamed Gomaa1 , Wassim Jabi2
1
Tanta University 2 Cardiff University
1
msrgomaa@gmail.com 2 jabiw@cardiff.ac.uk

Lighting analysis tools have proven their ability in helping designers provide
functional lighting, increase comfort levels and reduce energy consumption in
buildings. Consequently, the number of lighting analysis software is increasing
and all are competing to provide credible and rigorous analysis. The rapid
adoption of parametric design in architecture, however, has resulted in complex
forms that make the evaluation of the accuracy of digital analysis more
challenging. This study aims to evaluate and compare the performance of
daylighting analysis in two industry standard software (Autodesk Revit and 3ds
Max) when analysing the daylighting of complex parametric façade patterns. The
study has shown that, generally, both Revit and 3ds Max underestimate
illuminance values when compared to physical scaled models. 3ds Max was
found to outperform Revit when simulating complex parametric patterns, while
Revit was found to outperform 3ds Max when simulating simple fenestration
geometries. As a general conclusion, the rapid progress of parametric modelling,
integrated with fabrication technologies, has made daylighting analysis of
complex geometries more challenging. There is a need for more sophisticated
algorithms that can handle the increased level of complexity as well as further
verification studies to evaluate the accuracy claims made by software vendors.

Keywords: Daylighting analysis evaluation, Parametric patterns, Revit, 3ds

Max, Complex façades

INTRODUCTION credible and rigorous analysis results, while, ironi-

The use of lighting analysis tools is becoming in-
creasingly important. These pieces of software have
proven their ability in helping designers provide
functional lighting, increase comfort levels in spaces
and reduce energy consumption through the use of
natural light (Maamari et al. 2006). Consequently,
the number of lighting analysis software is increasing
around the world and all are competing to provide
cally perhaps, the error margins of the results are still
high. Thus, identifying the lighting software's capa-
bilities and limitations as well as validation studies
play an important role.
Over the last two decades, several associations
have been conducting studies in the field of light-
ing software validation (e.g. CIE Technical Commit-
tee, BRE-IDMP and NRC) (Tsountani and Jabi 2014).

DESIGN TOOLS | Evaluations - Volume 2 - eCAADe 34 | 147

 These associations have produced a number of tech-
nical reports and test cases that have set clear guide-
lines for evaluating the accuracy of lighting analysis
tools (Maamari et al. 2006; Tsountani and Jabi 2014).
The rapid adoption of parametric design in architec-
ture, however, has made the evaluation of the light-
ing software's accuracy more challenging. The com-
plex shapes and geometries produced by many mod-
elling software have made evaluation using simple
technical reports inapplicable and unreliable.
HYPOTHESIS, AIMS AND OBJECTIVES
Digital design software for architecture seeks to in-
crease the efficiency of design process by integrat-
ing performance analysis tool with parametric mod-
elling. While the lighting analysis engines in Re-
vit and 3ds Max have been validated against other
industry-standard engines such as Radiance and
against physical scale models, there are no published
studies to date that test their capabilities when used
to simulate daylighting in complex façade geome-
tries. 3ds Max Design has been well validated and
has been proven to be highly reliable for perform-
ing daylighting analysis as concluded by a number
of authors such as Reinhart and Breton (2009) and
Tsountani and Jabi (2014). Yet, there are still some
concerns regarding its accuracy in lighting analysis
for complex geometries (Tsountani and Jabi 2014).
Several authors such as Carroll (1999), Bazjanac and
Kiviniemi (2007) and Tsountani and Jabi (2014) have
been emphasizing the importance of integrating per-
formance analysis with advanced parametric mod-
elling in the early design stage.
This study aims to assess the capabilities and ac-
curacy of the daylighting analysis for complex para-
metric façade geometries using two popular pieces
of software (Autodesk Revit and Autodesk 3ds Max).
Both products combine parametric design tools with
daylighting analysis tools and both claim to have the
ability to generate accurate analysis for complex ge-
ometries, which can be produced by the parametric
tools. The key difference between them is that Re-
vit is designed mainly for building information mod-
148 | eCAADe 34 - DESIGN TOOLS | Evaluations - Volume 2
elling (BIM), while 3ds Max is more suitable for para-
metric 3D modelling, visualization and animation.
METHODOLOGY
The methodological approach of the study is exper-
imental, based on four main test cases. Each test
case consists of a comparison among three types
of models; Revit 2015, 3ds Max 2015 and a physi-
cal scaled model. The limited time and resources
have prevented performing comparisons with actual
built cases. However, physical scaled models are con-
sidered a reliable validation method for the illumi-
nance prediction in real spaces by many lighting re-
searchers (Love and Navvab 1991; Mardaljevic 2001;
Thanachareonkit et al. 2005). As a consequence, it
has been also considered as a trusted technique for
producing benchmarks for validating lighting soft-
ware (Mardaljevic 2001). In this study, the scaled
model is used as a baseline case for comparison pur-
poses. While it is understood that the accuracy of
physical scaled models varies in relation to the 1:1
construction, they serve as a stand in and their inac-
curacy is taken into account in the reported results. It
is planned to use actual built cases in future studies.
These test cases form a set of four different mod-
elling and simulation challenges. Each test is con-
ducted in two different overcast sky conditions; mid-
summer and mid-winter. The chosen location is Lon-
don, United Kingdom. The tests use a simple rect-
angular room with unglazed opening at the south-
ern façade. This opening is where the different pat-
terns in each test are placed. The initial test (test 1)
simulates the room with simple unglazed opening.
The next set of tests (2, 3 and 4) use three types of
parametric patterns to form the façade geometries,
which approximate some of the commonly used
techniques in modern building envelopes. These
techniques are force-field, packing, and weaving re-
spectively.
The room model and the parametric patterns for
each test were built then daylighting analysis was
performed using the available tools in each software.
The assessment of results depends mainly on evalu-
Figure 1
The figure shows
examples of the
three chosen
patterns; the force
field in Metropol
Parasol Seville in
Spain, the packing
in packed pavilion
in Shanghai and the
weaving pattern ating the discrepancies in illuminance values as re-
geometry by Erwin ported by Revit and 3ds Max and by comparing those
Hauer [1; 3; 4]. to measurement of illuminance values in the physi-
cal scaled model. The physical scaled models were
produced by 3d printing and laser cutting processes.
In order to assure the accuracy of the comparisons,
identical boundary conditions and material proper-
ties were maintained among all models.
METHODOLOGY OF CHOOSING THE PARA-
METRIC PATTERNS
These patterns represent three different algorithmic
thinking techniques. They were chosen based on the
findings from two books; Elements of Parametric De-
sign by Woodbury (2010) and Parametric Design for
Architecture by Jabi (2013). These patterns represent
the main possible challenges when using parametric
design for producing façades' patterns or building's
envelope structure.
The force-field technique was chosen to rep-
resent the use of complex shading devices, which
formed as a grid of vertical and horizontal louvers.
These louvers have changing depths to simulate dif-
ferent needs of indoor daylighting conditions. The
packing technique offers an economic and efficient
way of using materials and spaces while combining
other parametric concepts such as tiling and subdi-
vision (Jabi 2013). The main challenge of using pack-
ing in daylighting simulation is in calculating the light
penetration through perforated screens. Finally, the
weaving technique represents a high level of com-
plexity. Formations by weaving contain a large num-
ber of curves while the interlacing of threads creates
complex double-skinned geometries. These factors
are considered as a highly challenging test for the
ability of any software to perform daylighting anal-
ysis accurately. Figure 1 shows examples of using
these three patterns in real projects.
MODELLING DETAILS OF TEST CASES
Test 1 was conducted using the room without adding
patterns to the southern façade. This test was con-
sidered as a baseline case to obtain a basic overview
of the calculations' accuracy within the virtual mod-
els as compared to the scale model measurements.
The second test used the force-field technique to pro-
duce the façade pattern to fill the opening at the
southern façade of the room. It was designed with
variable depths ranging from 15.0 to 60.0 cm in or-
der to simulate the most commonly used depths for
shading and louvers. The grid size of pattern was
set to (30.0*30.0) cm. The third test used the pack-
ing technique. The pattern contained multiple num-
bers of circles which have variable diameters ranging
between 5.0 and 50.0 cm. The thickness of the pat-
tern was set to 15.0 cm. The final test used the weav-
ing technique. The dimensions of a real sculpture by
Hauer (2004) were used in this study. The pattern
contained two main interlacing layers. The thickness
of each layer was set at 5.0 cm while the maximum
gap distance between them was set at 10.0 cm. This
resulted in an overall maximum thickness of 20.0 cm
for the overall pattern. The design of the pattern pro-
duced a number of small openings with a maximum
width and height of 40.0 cm and 20.0 cm respectively.
DESIGN TOOLS | Evaluations - Volume 2 - eCAADe 34 | 149
 Figure 2
Photos show the
process of
producing the
physical model
geometry. Source:
author.

Figure 3
The virtual and
physical models for
test 2, 3 and 4
(Surrounding
envelope removed
in images of virtual
model for clarity).
Source: author.

DEFINING MODEL SPECIFICATIONS AND
FABRICATION PROCESS
Test cases were conducted using a simple rectangu-
lar room with dimensions of (4*4*3) m. This room was
designed to be similar to the one used in other vali-
dation studies such as CIE 171:2006 technical report.
This room has an unglazed opening on the southern
façade with dimension of (3*2) m. The opening is
filled with a different parametric pattern in each test.
All inner surfaces were white painted with an average
reflection ratio of 91.0 %. The scale model envelope
was fabricated by a laser cutter machine using ply-
wood sheets. Using other materials to produce the
envelope would not affect the testing process as the
most important factor is the final reflectivity of the
surface material. The inner surfaces of the envelope
were painted with 5 layers of white spray with 30 min
time interval between each layer in order to achieve
consistent and homogeneous reflectivity. The model
scale is 1:10 in order to fit the overall scale of the test-
ing area and to fit the photometric sensors as well.

150 | eCAADe 34 - DESIGN TOOLS | Evaluations - Volume 2

Figure 4 The parametric patterns were fabricated using PLA
Testing physical material by an Ultimaker 2.0 3d printer. The Ulti-
models at the maker printing platform is limited to (20*20*18) cm.
artificial sky dome. Thus, it was necessary to divide the patterns geome-
Source: author. tries into smaller parts to make them fit into the print-
ing area. These parts were then assembled to form
the final geometry before painting. Patterns were
painted in white as well to produce high reflectivity.
All three patterns had an average reflectivity ratio of
85.0 %. The white paint finish was chosen to achieve
better light distribution inside the model (Tsountani
and Jabi 2014) while it is also closer to the real fin-
ishing used in building interiors. Figure 2 shows the
process of producing the physical model geometry,
while Figure 3 shows the virtual and physical models
for test 2, 3 and 4.
DEFINING BOUNDARY CONDITIONS FOR
THE TEST CASES
The boundary conditions for the testing process were
based on two periods throughout the year in order
to assess different types of sky illuminations. The
first period is a winter scenario measured on Jan-
uary 21st while the second period is a summer sce-
nario measured on June 21st. The simulation time is
12:00 pm in order to gain the highest altitude of the
sun. The sky condition for validation testing using a
scale model is recommended to be an overcast sky
(Mardaljevic 2001; Thanachareonkit et al. 2005). It is
also recommended for artificial sky testing as there is
a large discrepancy in illumination between a real sun
and an artificial sun and also to avoid the penetration
of direct sun rays into the model (Mardaljevic 1999).
The test location was set to London, UK. The bound-
ary condition of the scale model was produced by
using an artificial sky dome (see Figure 4). The mea-
sured lux level from the artificial sky was used for sky
illuminance input in both Revit and 3ds Max.
MONITORING AND EVALUATING TEST RE-
SULTS
The test target is to monitor and record the illumi-
nance values in Lux taken on a working plan with a
height of 0.75 cm, based on the LEED accreditation
requirements (au.autodesk.com 2016). Lux values
are recorded using sixteen sensor points distributed
on an orthogonal grid (1*1) m ( see Figure 5). An-
other four sensors were used on the top (for the scale
model) to measure the average of sky illumination
level. These illuminance levels were used in the anal-
ysis setting of Revit and 3ds Max to ensure identi-
cal boundary conditions. The comparisons used val-
ues from each column of sensors separately, start-
ing from column 1 and moving to column 4 while
also moving from row A to row D. Two main items
were assessed, the first was the consistency of soft-
ware analysis to the physical scale model measure-
ments (general trend of the charts) and the second
was the percentage of discrepancy between the vir-
tual and the physical scale model. 12 "Megatron Se-
lenium" photo-electric cells connected with the "Ag-
ilent 34970A Data Acquisition" control logger were
used to record the illuminance measurements. The
reflection ratios from physical model parts were mea-
sured using the Luminance meter 'Minolta LS-100'.
DESIGN TOOLS | Evaluations - Volume 2 - eCAADe 34 | 151

PREDICTION OF ERRORS
Interpreting the results was based on assessing the
amount of discrepancies. The eventual result will be
evaluated after checking the available error bands. A
number of researchers have outlined different causes
of errors when using scale models in validation
(Schiler 1987; Love and Navvab 1991; Thanachare-
onkit et al. 2005; Aghemo et al. 2008). These ex-
pected causes are as follow:
• Parasitic light penetration into scale models.
• Sensors levelling and placement in the scale
models.
• Surfaces reflectance.
• Lux meter calibration.
• photocells size (sensing aperture).
Different literature sources, including the ones men-
tioned previously, were checked to gather the final
error margins for this study. These sources have used
different pieces of software besides using scale mod-
els or real buildings for validation. The next table (Ta-
ble 1) shows the reported error bands by each litera-
ture source:
From Table 1, it is concluded that scale mod-
els mostly overestimate the results when compared
to real tests. This over-estimation is agreed to be
up to 50% and in extreme cases could reach 60%.
However, the average reported error of 40% is used
for this study. On the other hand, the reported er-
ror bands from comparing software to real cases or
benchmarks vary between 10% to 25%. Since this
study focuses on 3ds Max and Revit, the used error
band is based on the recommendations from Rein-
hart and Breton (2009) and Tsountani and Jabi (2014)
152 | eCAADe 34 - DESIGN TOOLS | Evaluations - Volume 2
Figure 5
Room’s plan,
section and
perspective
including the
sensors grid
dimension and
design. Source:
author.
who studied 3ds Max and set the error band between
10% and 20% of under-estimation. Thus, an average
of 15% under-estimation is used in this study. As a
consequence, there is an overall predicted error band
for this study of 55.0 % (40.0 % + 15.0 %). This is pre-
dicted as an underestimation of the lux levels in soft-
ware analysis when compared to the physical scaled
model.
For further detailed assessment, the discrepan-
cies were categorized into two stages of error bands:
the lower error band (20.0 %) and the higher error
band (55.0 %). The lower error band was chosen
to reflect a high accuracy analysis and to represent
the maximum error by Reinhart and Breton's study
(2009). The higher error band is the maximum limit
for error. Discrepancies which exceed the 55.0% re-
flect a lack of accuracy by the software.
DISCUSSION
Based on the findings from the four tests, it is gener-
ally noticed that both Revit and 3ds Max are under-
estimating the illuminance values as compared to
the physical scaled models. This under-estimation by
software was expected as discussed previously. In
test 1, where a simple opening model is used, Revit
has shown an overall better performance than 3ds
Max. However, the differences in discrepancies be-
tween Revit and 3ds Max were not high, as the ma-
jority of discrepancies in both pieces of software lo-
cated within the lower error band (20 %) while no dis-
crepancies exceeded the higher error band (55 %).
In addition, the discrepancies in both Revit and 3ds
Max were at their lowest values nearest to the open-
Table 1
Recorded error
bands by different
literature. Source:
author.

Figure 6 ing. Overall, test 1 has revealed that Revit is produc-

Graph of average ing more accurate daylight analysis than 3ds Max for
discrepancies simple geometries.
performance Starting from test 2, as the daylight analysis con-
throughout the four ditions became more complex by adding the para-
main test cases. metric patterns to the room opening; the perfor-
mance of Revit drops when compared to test 1. Al-
though both Revit and 3ds Max suffered a drop in
daylighting analysis performance in test 2, 3ds Max
has shown superiority over Revit. All discrepancies in
Figure 7 3ds Max located below the higher error band of 55 %
Graph of maximum in both scenarios, while in Revit they were 75.0 and
discrepancies 56.0 % in the same order. The superiority of 3ds Max
performance remained in tests 3 and 4 as well. Yet, it is noticed
throughout the four that the daylight analysis performance by both Revit
main test cases. and 3ds Max decreased gradually when moving from
test 1 to test 4. As shown in Figure 6 and 7, the dis-
crepancies trend in all scenarios increased after ap-
plying more complex geometries. Test 4 results have
the highest discrepancies in both software. The high-
est recorded discrepancy is 70.0 %, which is recorded
in Revit and 3ds Max in test 4. This exceeds the higher
error band by an extra 15.0%.

DESIGN TOOLS | Evaluations - Volume 2 - eCAADe 34 | 153


To summarise, the average of discrepancies from Fig-
ure 6 show satisfactory results, as Revit exceeds the
higher error band only in the winter case scenario in
test 4, with a small difference of 2.0%. On the other
hand, the maximum of discrepancies from Figure 7
shows that both Revit and 3ds Max exceed the higher
error limit in tests 3 and 4. However, 3ds Max has
shown that the majority of its discrepancies are lo-
cated below the higher error band with percentage
between 69 and up to 100 % of sensor points (see Ta-
ble 2).
154 | eCAADe 34 - DESIGN TOOLS | Evaluations - Volume 2
Table 2
Different recorded
discrepancies in
Revit and 3ds Max
in the four test
cases.
The last three columns in Table 2 show the percent-
age of points on the sensor grid that have discrepan-
cies within and above error bands per each test case.
For example, the first row of values in the table shows
that in Revit winter case, 94.0% of points are located
within the lower Error band. This means that 15 out
of 16 sensor points have discrepancies values that are
lower than 20.0%. The following graphs in Figure 8
outline the percentages of these points in each test
case.
Figure 8
The percentage of
points within and
above error bands
in each test case in
both winter and
summer scenarios.
CONCLUSION
The final conclusion regarding which software is per-
forming better daylighting analysis depends on two
main questions. The first question is what level of
complexity do the project's geometries have?, while the
second is question is what level of accuracy is needed?.
This study has found that the claims of Autodesk
about Revit's capabilities are correct only when simu-
lating simple geometries such as the model in test 1.
It is proven through test cases 2, 3 and 4 that Revit has
low performance of daylighting analysis when sim-
ulating complex patterns. In contrast, 3ds Max has
shown better performance when simulating complex
patterns.
Based on these findings, Revit is recommended
to provide a fast and credible daylighting analysis for
simple forms and only in the early stages of design
when high levels of accuracy are not required. Re-
vit also can be useful to obtain an initial insight re-
garding the daylighting distribution inside the model
even with the complex patterns. However, users
should expect an overall under-estimation of illu-
minance values. If more sophisticated daylighting
analysis is required, then 3ds Max is highly recom-
mended for the task. As a general conclusion, the
rapid progress of parametric modelling, integrated
with fabrication technologies, has made daylighting
analysis of complex geometries less credible. The ex-
isting daylighting simulation tools need validations
and urgent upgrades in order to narrow the increas-
ing gap between predictions and physical reality.
The followed methodology in this study has
helped in achieving the required aims. The main ob-
stacle for the study was the limitation in time and
facilities. However, there are several suggested im-
provements which could raise the efficiency of the
study's experimental part. These suggestions are as
follow:
• To use real sky conditions instead of using the
artificial sky in simulating the daylight system
in the physical scale models. This can mini-
mize the possible errors related to calibrations
of the artificial sky illumination or the mainte-
nance of hardware such as light bulbs, but will
produce its own challenges.
• To calibrate test 1, as baseline verification,
with an actual built situation as a starting
point including comparative measurements
to the physical model. This will increase the
relevance and significance of this research.
Further studies are recommended on evaluating the
daylighting analysis in Revit and 3ds Max using clear
sky conditions. These further tests shall include glaz-
ing in the façades, whether transparent or translu-
cent.
ACKNOWLEDGEMENT
We would like to thank the teaching and research
staff at Cardiff University; Don Alexander, Hue Jenk-
ins, Heba Elsharkawy and Dylan Dixon. We would
DESIGN TOOLS | Evaluations - Volume 2 - eCAADe 34 | 155
 also like to thank the Chevening program, this paper Reinhart, C and Breton, PF 2009, 'Experimental Valida-
would not have been done without their generous tion of Autodesk(R) 3ds Max(R)', Leukos, 6(1), pp. 7-
funding and support. 35
Schiler, M 1987, Simulating Daylight with Architectural
Models, Daylighting Network of North America, Los
REFERENCES Angeles, University of Southern California
Aghemo, C, Pellegrino, A and LoVerso, VRM 2008, 'The Thanachareonkit, A, Scartezzini, J.-L and Andersen, M
approach to daylighting by scale models and sun 2005, 'Comparing daylighting performance assess-
and sky simulators: A case study for different shad- ment of buildings in scale models and test modules',
ing systems', Building and Environment, 43(5), pp. Solar Energy, 79(2), pp. 168-182
917-927 Tsountani, A and Jabi, W 2014 'Evaluating lighting simu-
Bazjanac, V and Kiviniemi, A 2007, 'Reduction, simplifica- lations in Autodesk 3ds Max Design', Proceedings of
tion, translation and interpretation in the exchange IBPSA 2014, London
of model data', CIB W, 78, pp. 163-168 Woodbury, R 2010, Elements of Parametric Design, Taylor
Carroll, W 1999, Daylighting simulation: methods, algo- and Francis
rithms, and resources, Lawrence Berkeley National [1] http://www.fosterandpartners.com/projects/smith
Laboratory, Berkeley sonian-institution/
Galasiu, AD and Atif, MR 2002, 'Application of Day- [2] http://www.autodesk.com/products/lighting-analy
lighting Computer Modeling in Real Case Studies: sis-revit/overview
Comparison Between Measured and Simulated Day- [3] https://www.competitionline.com/en/projects/458
light Availabililty and Lighting Consumption', Build- 38
ing and Environment, 37(NRCC-42862), pp. 363-377 [4] http://erwinhauer.com/eh/installations/church-i
Hauer, E 2004, Continua-Architectural Walls and Screens, n-liesing-vienna-austria
Princeton Architectural Press
Jabi, W 2013, Parametric design for architecture, Laurence
King LTD
Jabi, W and Garber, R 2006, 'Control and Collaboration:
Strategies for Digital Manufacturing in Academia
and Practice', International Journal of Architectural
Computing (IJAC), 2(4), pp. 121-143
Kota, S, Haberl, JS, Clayton, MJ and Yan, W 2014, 'Build-
ing Information Modeling (BIM)-based daylighting
simulation and analysis', Energy and Buildings, 81, pp.
391-403
Love, J and Navvab, M 1991, 'Daylighting Estimation un-
der Real Skies: A Comparison of Full-Scale Photom-
etry, Model Photometry, and Computer Simulation',
Journal of the Illuminating Engineering Society, 20(1),
pp. 140-156
Maamari, F, Fontoynont, M and Adra, N 2006, 'Applica-
tion of the CIE test cases to assess the accuracy of
lighting computer programs', Energy and Buildings,
38(7), pp. 869-877
Mardaljevic, J 1999, Daylight simulation: validation, sky
models and daylight coefficients, Ph.D. Thesis, De
Montfort University
Mardaljevic, J 2001, 'The BRE-IDMP dataset: a new
benchmark for the validation of illuminance predic-
tion techniques', Lighting Research and Technology,
33(2), pp. 117-134

156 | eCAADe 34 - DESIGN TOOLS | Evaluations - Volume 2

DESIGN TOOLS | Applications
How Do Small and Medium Architectural Firms Deal with
Architectural Complexity? A Look Into Digital Practices
Adeline Stals1 , Sylvie Jancart2 , Catherine Elsen3
1
LNA-ULg - Faculté d'Architecture, Université de Liège ; FNRS, Fonds de la
Recherche Scientifique 2 LNA-ULg - Faculté d'Architecture, Université de Liège
3
LUCID-ULg, Faculté des Sciences Appliquées, Université de Liège
1,2,3
{adeline.stals|sylvie.jancart|catherine.elsen}@ulg.ac.be

Digital design tools are nowadays deeply rooted in most architectural design
processes. Either used as a way to communicate a designed artifact, or rather as
a medium to test innovative morphologies (lately referred to as non-standard
architecture), these design tools profoundly shape the way architects envision
complexity all along their day-to-day tasks. The large teams of "star-architects"
were able to develop dedicated and specific strategies and digital tools to
overcome some of the technological gaps they were faced to when using
commercial packages. But how are small and medium firms dealing with the
inherent complexities of such software? This contribution analyzes how those
firms (and more specifically the Walloon ones) deal with the growing complexity
of digital tools, both in terms of use and interdisciplinarity.

Keywords: Digital design support tools, architectural practice , non-standard

architecture, architectural complexity, small and medium architectural firms

INTRODUCTION master, the misunderstanding lied both in regard of

First introduced as a substitute for hand-drawing
tools, digital design tools are currently used as a way
to innovate, notably in terms of architectural mor-
phology. Enabling the development of new - and
more complex - morphologies, these design tools are
also partly responsible for the growing complexity of
the design process itself.
Historically architects have raised questions con-
cerning their design tools, mostly about their rel-
evance and impact at different stages of the de-
sign process. Regarding these new-generation de-
sign tools, issues first emerged because of the over-
all complexity of the software themselves: difficult to
their very underlying logic and their unfriendly user
interfaces. Architects therefore began to develop
strategies to go beyond these difficulties and slowly
recover consistency in their processes.
"Star-architects", with their large multidisci-
plinary teams, were of course able to develop ded-
icated and specific digital tools to overcome some
of the technological gaps (Shelden 2002): one au-
tomatically thinks about Franck Gehry and his "CA-
TIA teams" of software engineers, for instance. This
paper rather looks into the habits and strategies
adopted by small and medium architectural firms.
Which steps and mechanisms of the design process

DESIGN TOOLS | Applications - Volume 2 - eCAADe 34 | 159

 have been influenced by the use of digital tools?
What are the commonly used software packages,
and how do designers adapt to master them? How
are those small and medium firms dealing with the
inherent complexities of those commercial software?
What impact does digital design have on teams, and
what kind of digital expertise is expected nowadays?
The paper first discusses new forms of complex-
ity reflecting current architecture practice, and inves-
tigates how complexity takes roots through the digi-
tal medium. The contribution then describes and an-
alyzes the situation of small and medium Belgian of-
fices, reflecting particularly on the way architects im-
plement and take advantage of digital tools when de-
signing architecture.
STATE OF ART
The development of digital design tools has in-
troduced a big shift in architectural morphologi-
cal approaches, fostering the growth of what is to-
day known as "non-standard architecture" (Migayrou
2003). More than the digitization of traditional rep-
resentations, numerical modeling has transformed
both the formal vocabulary and architectural think-
ing by considering and modeling other factors influ-
encing the shape (spatial, societal, aesthetic), gener-
ating possibilities better suited to the renewed ex-
pectations of some architects. This digital turn was
illustrated by the "Non-standard Architecture" exhi-
bition organized by Migayrou in 2003 at the Pompi-
dou Center in France. He also curated another ex-
hibition, called "Naturalizing Architecture", milestone
event for the second digital turn, driven by theories of
complexity and by new digital technologies (Artemel
2015).
Non-standard architecture has left many projects
at their virtual, embryonic stage. The supposed free-
dom left to architects to design innovative shapes
would indeed, as a consequence, rather introduce a
complexity in terms of building them. However, built
projects such as the Guggenheim Museum in Bilbao,
for instance, pave the way for a renewed vision of
feasibility shadowing the evolution of materials and
160 | eCAADe 34 - DESIGN TOOLS | Applications - Volume 2
digital manufacturing possibilities. They displace the
traditional issues of complexity in terms of form and
structure and bridge the gap between the thinkable
and the buildable (Monier and al. 2012).
The meaning of complexity itself has changed
in regard of design methodologies used at different
times as well as in regard of available resources. To
understand these evolving levels of complexity in ar-
chitecture, it is interesting to look into the work of the
French philosopher Edgar Morin, who highlighted
the emergence of complexity in general as a new
paradigm, renewing our relation to the world. Con-
sidering this notion of complexity, we mention here
some studies that define the limits of this concept.
Corning (1998) defines the properties that are com-
monly associated to the term rather than defining
the term itself. Three attributes are often implied: (1)
a complex phenomenon consists of many parts (or
items, or units, or individuals); (2) there are many rela-
tionships/interactions among those parts; and (3) the
parts produce combined effects (synergies) that are
not easily predicted and may often be novel, unex-
pected, even surprising. To focus more deeply on ar-
chitectural field, a research by Chase and Murt (2000)
mentions two notions of complexity to take into ac-
count: the design complexity, based on visible fea-
tures of the designed object; and the CAD complex-
ity, based on the actual CAD embodiment of the de-
sign.
A THEORETICAL LOOK INTO COMPLEXITY
THROUGH HISTORY OF ARCHITECTURE
These theoretical contributions taken into account,
this section deepens the notion of complexity in ar-
chitecture in light of three different eras of architec-
tural history: first when the physical experimenta-
tion was considered the primary tool for dealing with
such complexity, second at the very beginning of
digital architecture and, third, in regard of the current
use of digital design tools.
Morphogenesis through experimentation
In the past, understanding and controlling morpho-
genesis of structures, often inspired by nature, usu-
ally relied on empirical, trial/error methodologies re-
quiring complex physical prototyping and experi-
mental settings. Architects and engineers such as
Antonio Gaudi, Heinz Isler or Frei Otto conducted
such studies in order to progressively refine their fu-
nicular, shell or lightweight tensile and membrane
structures (Stals and al. 2015). One can admit that
these architectures expressed a certain kind of sim-
plicity both on an aesthetic and structural level, given
the intrinsic coherency of their formation process.
However those experimental processes, while pro-
viding a visual simplicity and coherency, still gener-
ated complexity in terms of mathematical descrip-
tion of the shape and transfer to reality.
First step into the digital era
In the late eighties, computer aided design machines
speeded up the drafting and modifying process, leav-
ing the rest of design steps mostly unchanged. Then
a new generation of tools (such as parametric soft-
ware) started to impact more deeply the design pro-
cess. These design tools gave a wider number of
shape possibilities, leading to the non-standard ar-
chitecture introduced above.
The inventory of projects designed at that time
indicates that the complexity of these shapes did
put some distance between the ideas on screen and
their feasibility. This software, initially supposed to
simplify the design process, rather generated mul-
tiple levels of complexity. We sum up these levels
through three main ruptures partly explaining why
lots of these complex shapes remained at this virtual
stage (Picon 2010; Stals and al. 2015). The first rup-
ture takes place between morphology and structure:
the digital approach rather encourages erasing the
structure at the benefit of morphological audacity.
The second one appears at the interface of multidis-
ciplinary skills and knowledge of the design process,
while the third rupture operates at a scale and tec-
tonic level of the project. All these ruptures lead to
tensions between architectural desires and techno-
logical potentialities.
Digital architecture nowadays
Lately more and more of these complex shapes have
been built: projects such as the Pompidou Center
by Shigeru Ban, or the Louis Vuitton Foundation by
Frank Gehry do push forward the possibilities offered
by digital tools. The attitudes of architects are con-
sequently slowly evolving. Some of them realize that
the introduction of digital tools into the design pro-
cess enables more than simply processing informa-
tion: they are therefore ready to explore how these
tools might contribute to the development of inno-
vative morphologies, better adapted to their expec-
tations and creativity. By doing so, they still have to
address the above-mentioned complexities and rup-
tures, and they tend to develop protocols and strate-
gies. While large architectural firms have developed
their own research and development teams, and
even their own proprietary software, little is known
about the strategies adopted by small and medium
architectural firms.
A LOOK INTO BELGIAN PRACTICES
Digital tools are presently recognized for their po-
tential to develop new complex, non-standard ar-
chitecture. But how do small and medium architec-
tural offices deal with those digital tools? Do they
achieve such breakthrough as easily? The paper re-
ports findings from the current situation of Belgian
offices, more specifically about the challenges they
face in dealing with digital tools during their design
process, and about their perception of complexity all
along this process. Within this research, Wallonia (the
south part of the country) is a perfect case study be-
cause it is known to be dotted with quite small of-
fices.
Research gap
Following up existing literature, we understand that
a gap of knowledge exists when it comes to cur-
rent architectural practices in small and medium of-
fices. A significant amount of work is done about
large architectural firms, such as Shelden (2002) fo-
cusing on Gehry's architecture. More specifically in
DESIGN TOOLS | Applications - Volume 2 - eCAADe 34 | 161
 Belgium, the last study about the use of digital tools
by architects was conducted in 2008 (Weytjens and
al. 2008). The goal of this survey, mainly addressed
to the North part of the country, was to assess the im-
pact of different type of design support tools (DSTs)
through the decision making process. This research
was thus not specifically focusing on the role of dig-
ital tools in architectural practices. It rather classi-
fied six types of design tools according to the role
they played all along the design process: knowledge
- based tools, communication tools, modeling tools,
presentation tools, structuring tools and evaluation &
analysis tools. Researchers observed that the use of
DSTs was important in the architectural design pro-
cess, and indicated an increasing use in the future.
The results stressed the importance of a clearer un-
derstanding of users' needs, and notably pointed out
that CAD-software was merely used as a presentation
and modeling tool.
Considering this current state of knowledge, this
paper will therefore address three main research
questions:
• How do architects use digital tools nowadays
in Wallonia? Do they express interest for
new technologies and software, supposedly
helpful when designing non-standard archi-
tecture?
• Where does the day-to-day complexity hides,
according to Walloon actors of architectural
design? What are the factors and actors who
make their work more complex?
• What challenges do they face when using dig-
ital tools? Are those challenges considered as
building up current architectural complexity?
Methodology
Regarding the large amount of people to reach
(about 13.000 architects or architectural engineers),
we used an online-based survey strategy in order
to explore the previous research questions. The fol-
lowing sections aim at developing the methodology
used to rigorously build and analyze this survey.
162 | eCAADe 34 - DESIGN TOOLS | Applications - Volume 2
The questionnaire was built around three main sec-
tions. The first part began with collecting the par-
ticipants' demographic data in order to contextualize
each profile. Ten questions were formulated (1 open-
ended question, 7 semi-open questions and 2 closed-
ended questions) and mainly related to the partici-
pants' gender, age, background, expertise, main day-
to-day tasks and size of firm. The second and most
important section questioned designers' digital cul-
ture, the digital tools they use, their feelings about
those digital tools and the impact those digital tools
have on the architectural design process, from their
point of view. It also formulated specific questions
about complexity, such as "List here the 5 main fac-
tors that, from your point of view, make your cur-
rent practice more complex?" or "Do you think digi-
tal tools make your current practice more complex?".
This section contained 26 questions with 6 open-
ended questions, 10 semi-open questions and 10
closed-ended questions. The concluding sections in-
vestigated parametric design and tools.
The survey was tested with a first round of a
few participants, which enabled us to specify the
meaning of some questions, to adapt some fixed-
alternatives answers and to test how much time was
needed to complete the questionnaire rigorously.
If a completed survey fulfilled one of the next cri-
teria, it was considered unusable and therefore was
not included in the next steps of our research:
• The survey was completed far too quickly and
therefore could not been taken seriously. The
test-survey round demonstrated that the 15
minutes boundary was the right limit;
• Only the first section of the survey was com-
pleted, and therefore offered no data about
digital or parametric design/tools. This means
that some surveys, where only a few ques-
tions had been dismissed, were still consid-
ered as valuable (in that case, a "no answer"
- NA appears in regard to the few dismissed
questions);
• Regarding the size of the firm, we put aside
participants working in structures of more
 than 100 people. These people, the "back-
ground" and "main tasks" sections reveal, are
mostly architects working as academics or in-
cluded in larger, contractors structures.
Sample description
Figure 1 The survey was addressed to all Belgian architects
Size distribution of and architectural engineers, and has covered more
firms in Wallonia. specific topics, including parametric design and para-
metric tools, as previously stated. Given the spe-
cific scope of this paper, we will concentrate here on
the data provided by architects working in the Wal-
loon part of Belgium (where most of the small and
medium architectural firms are settled) and we will
more specifically focus on topics related to digital
tools and architects' perception of complexity. This
first exploration of the data mostly concentrates on
quantitative results basically treated in order to de-
lineate general trends, and supported by qualitative
data to more closely look at some of these trends.
After cleaning the data, 331 surveys were treated
for this research, representing 6.2% of architects reg-
istered to the French and German-speaking Archi-
tects Association. The female-male parity is close
to data already collected by this Association, in re-
gard of a previous survey conducted in 2013 (66%
male architects at that time, Tchinda 2013). In our
case, 73.4% of the surveys are answered by men
and 26.0% by women (while 0.6% did prefer not to
answer), indicating that the current sample is suf-
ficiently representative of the Walloon population.
49.8% of the participants are under 40 years old, con-
firming the relative youth of the population as al-
ready observed by the 2013 survey. 34.2% of the re-
spondents are practicing their main occupation for
more than 20 years and 23.5% are practicing it for
10 to 20 years. Regarding their main function, 56.3%
of the respondents are isolated, independent archi-
tects (working on their behalf ), 10.5% are indepen-
dent architects working for some collaborator, while
6.4% are architectural engineers and 4.7% are teach-
ers (other participants distribute among other occu-
pations). Throughout this paper we will refer to the
participants as "designers".
Figure 1 demonstrates the relevance of the Wal-
loon case, since 41% of the respondents are working
in a firm of only one or two people. Furthermore, 73%
of the participants are working in a structure smaller
than 10 people.
RESULTS
Use of digital design tools in Wallonia
Our results first show that 72.5% of the participants
indeed use digital tools during the design phase.
Figure 2 moreover shows that designers using de-
sign tools just for 2D drawing mainly use AutoCAD
(55.9%), followed by ArchiCAD (17.6%) and Vector-
works (15.1%). ArchiCAD is also used as a 3D sup-
port tool (24.5%) but Sketchup remains the refer-
ence for 3D modeling in architectural design, at least
for 50.3% of the users. Parametric software such as
Grasshopper, Generative Component, Vasari or Dig-
ital Project are either totally, or largely unknown by
the Walloon population. Revit, however, deserves
a closer look: our results show that it is generally
used in 3D rather than in 2D. Revit is mainly used
by designers under 45, with a peak for the range 31-
35 years. It is also more often referred to by offices
counting more than 20 employees, representing only
14.2% of the Walloons designers (Figure 1).
Looking at how designers integrate digital tools
to their everyday practices (the 102 "NA" answers
put aside, that is almost one third of the respon-
dents), our results show that 49.5% of them are auto-
didacts, learning thanks to tutorials or forums, asking
their better trained colleagues for some help. 31.5%
learned either at school, through additional training
or through on-the-job training. Another 9.7% hire ex-
perienced employees to whom they can either del-
egate some work, or ask for some help when they
DESIGN TOOLS | Applications - Volume 2 - eCAADe 34 | 163
 want to improve their own skills. Another interest-
ing trend is that around 9% of designers implement
working protocols (such as using layers, pre-defined
templates, ...) to streamline and standardize their
process. This strategy is usually implemented after
having completed training or thanks to the input of
newly hired, experienced employees.
The survey moreover offered the participants to com-
ment about the aspects and features of software they
appreciate or not. Some designers comment that
"the project can be quickly modelled in 3D" thanks
to digital design tools, particularly through data li-
braries, and that there are easy ways to "use only
one software to make the 2D and 3D simultaneously".
Another advantage of digital tools is that "changes
are easier to perform without much costs (time, en-
ergy)". The 3D model also enables them to "check
the impact of architectural choices" and "urban inte-
gration", while getting a better understanding of the
client and the administration. In this regard, the soft-
ware enables "faster exchanges" with partners, out-
side and inside the office (as we will discuss later).
While some Walloons highlight the advantages of
tools integrating BIM, others still see these tools as
the simple "extension of their drawing board".
Among the most common negative aspects
reported by participants, the fact that there are
"too many tools available" is considered as a chal-
lenge in terms of day-to-day workflow. The "ex-
pensive price of software", "too frequent updates"
that require "counterproductive adjustment periods"
164 | eCAADe 34 - DESIGN TOOLS | Applications - Volume 2
and the "growing burden" of purchasing new com-
puter equipment are moreover frequently referred
to. Some raise the fact they spend "too much time
working on it", one of the reason being that digital
tools push to "draw too precisely from the sketch on".
Another criteria considered important is that "com-
plex shapes are difficult to represent" (e.g., curves)
and that producing a "non-standard element is com-
plex", generating "less creativity". Figure 2
Knowledge levels
Complexity and the use of digital design and use of digital
tools tools in
The large proportion of designers using digital tools architecture.
(72.5%) first of all confirms that our research meets
the current day-to-day working realities. We still have
to underline that the use of digital technologies de-
creases as the age increases, 23.7% of the designers
aged 55 and more indeed declaring not using any
digital tool, comparing to the 5.5% younger than 55
years old who do not use digital tools.
84.6% of the respondents are satisfied with the
digital tools they use. 58.2% of them are not only
satisfied but consider that the digital tools do not
make their work more complex. That leaves 15.4%
of surveyed designers unsatisfied, while 23.4% find
that digital tools have made their work more complex
in general (other respondents do not have a strong
opinion). Digital tools are indeed among the top
six factors that make nowadays architectural practice
more complicated, side by side with administrative
formalities; regulations (and more specifically plan-
ning regulations), "PEB" certification (Belgian build-
ing energy efficiency certification), evolving building
techniques and customer requests. Some of these
criteria were already mentioned by the Flanders
study conducted in 2008 (Weytjens and al. 2008).
This study, that more specifically researched the vari-
ous parameters taken into account when making de-
sign decisions, underlined professional experience as
the most important factor (86%), closely followed by
the client's demands (76%). Almost 60% of the re-
spondents at that time said they made design deci-
sions based on regulations, while only 21% of them
 included design support tools as a decision-making time in particular thanks to 3D visualizations. A sim-
factor. Comparing both studies, we can therefore ilar effect is observed for people older than 60 years
suggest that the factors influencing decision-making (58.6%). As a side note, one of the participants com-
processes in design are partly similar to the factors mented: "All in all, the computer is magical for those
making the design process more complicated. who began drawing by hand".
Beyond this noticeable complexity, digital tools
Figure 3 have strongly increased the execution speed of
Growing projects, strongly facilitated exchanges with stake-
complexity of holders and also strongly facilitated the implemen-
architect’s work tation of projects (Figure 4, 62 No Answer hidden).
with the advent of Moreover, 78% of the contributors consider numer-
digital tools, in ical modeling as enriching the design of an architec-
regard with age. tural project.
The impact of digital tools on current Wallonia ar-
chitectural practices is moreover illustrated in Figure
Figure 4 5. Again, for clarification, the 92 respondents who
Influence of the skipped this question have been hidden (92 "NA").
digital tools on This chart resumes which pre-defined factors and
diverse pre-defined actors are considered as mostly influencing the de-
parameters. signed shape. Planning regulations and customer
demands are again considered important influenc-
ing criteria. 3D software is the 4th most influencing
factor when it comes to architectural morphology,
according to the surveyed designers.
To fully understand the influence of the digi-
tal factor, we add here some results looking at how
digital tools modify the architects' roles, from their
Figure 5
point of view. Designers seem first divided when it
Factors and actors
comes to the designer's intent, and how it might have
influencing the
been impacted by the digital era. They rather agree
produced and
(57.3%) that digital tools have modified their control
designed shape.
over the implementation of the project, and at the
same time did not improve control of building costs.
These trends are not influenced by the designers' age
or professional expertise, but are influenced by the
size of the firm. The designer's intention, control over
implementation and control over costs are indeed
proportionally considered as more deeply impacted
Figure 3 shows that the perception of complexity in
by digital tools as the size of the firm increases.
architecture globally increases with age. A genera-
tion effect is nevertheless observed for the age group
Complexity into interdisciplinarity
41-45. 62.8% of them indeed find that digital tools do
It is nowadays largely accepted that design processes
not make their work more complex: the main reason
do integrate more and more stakeholders. 54.8%
being that digital tools have allowed them to save

DESIGN TOOLS | Applications - Volume 2 - eCAADe 34 | 165

 of the respondents consider time as the main factor
pushing them to outsource tasks. The complexity of
the shape is the next reason to externalize some de-
sign tasks, at least for 24.4% of the designers. In fact
designers a larger proportion of designers (57.1%)
agree that interdisciplinarity is totally beneficial to
the project, but still assert that this interdisciplinar-
ity is difficult to manage it humanely speaking (for
51.1% of the respondents - 108 NA removed) and
technically speaking (50.2% - 108 NA removed). Wal-
loon designers consider that the digital tools have
strongly facilitated the exchange with stakeholders
(Figure 4). Looking more closely at these exchanges,
Figure 6 reveals that architects rarely outsource the
chore steps of ideation (the architect being the most
common internal "consultant"), while the building
engineer is the most requested external consultant.
3D modeling specialist is more frequently reached
out internally (i.e., part of the design team) than ex-
ternally, showing a close working proximity between
him/her and the architect. The graphic designer
is slightly more consulted externally than internally,
showing that the proximity between the two profes-
sions is less important. Moreover, 3D modeling spe-
cialists are approximately 50% more consulted inter-
nally than graphic designers. This trend is already
illustrated in the Figure 4: 3D modeling specialists
are considered as having more influence on the de-
signed shape than graphic designers. All these obser-
vations are supported by another result: 49.5% of the
architects consider numerical modeling as enriching
the design of an architectural project, even when this
task is not assigned to the person who primarily de-
signed the project. Qualitative information help us to
better understand this result: we have to distinguish
architects who consider 3D modelization as part of
the design process itself from architects considering
3D models useful only to produce "commercial" im-
ages. When outsourcing the task, architects from the
first group generally rely on 3D modeling specialists
considered as close colleagues to work with. Some of
them comment: "modeling is part of the design pro-
cess and evolves with it in an iterative process". Ar-
166 | eCAADe 34 - DESIGN TOOLS | Applications - Volume 2
chitects from the second group, on the other hand,
rather resort to graphic designers working remotely.
Some of them recognize that entrusting some exter-
nal consultant with the 3D modeling task might be
interesting: "different points of view, ideas or advices
are always welcome"; "architects cannot assume ev-
erything anymore" and have to learn to "delegate by
giving the good information" and through "regular
exchanges".
Figure 6
Comparison of
stakeholders
consulted internally
or externally
DISCUSSION
Our first research topic was concerned with the cur-
rent use and perception of digital tools in Wallonia,
where firms are mostly of small and medium size. Our
results underlined that architects are globally satis-
fied with their digital design tools since they mostly
consider them as a faster way to process a project,
compared to drawing it by hand. They do not cur-
rently work a lot with complex 3D or parametric tools,
and feel remote from these new design support tools
considered as designed for - and more adapted to -
larger offices working larger-scale projects. Yet, when
expressing some interest, one has to observe that ar-
chitects do not hesitate to slowly master software as
autodidacts, with the help of tutorials and close or
freshly hired colleagues. Few of them do implement
working strategies and protocols specifically related
to the use of digital tools.
While digital technologies have freed architec-
tural innovation for star-architects while producing
several layers of intricate morphological complexi-
ties, small and medium architectural firms deal with
other levels of day-to-day challenges, dependent of
the architect's level of expertise and size of firm. In
Wallonia, the complexity does not seem to lie in the
expression of the designed shape, but rather in the
down-to-earth use of digital tools. While 58.2% con-
sider that digital tools simplify their processes, the
multiplicity of available tools and the difficulty to re-
main up to speed with evolving software (both on fi-
nancial and timing aspects) distance them from fur-
ther possibilities these new generation design tools
have to offer.
When faced with difficulties, close to 25% of the
architects do not hesitate to outsource the 3D mod-
eling task, while being aware of both the positive
and negative effects such outsourcing might have on
the design process and output. Some of them con-
sider 3D modeling as part of the design process it-
self: those do prefer to keep the digital phase under
control, therefore entrusting the task to some close
colleague(s). Others see in 3D modeling nothing else
than a way to generate nice-looking pictures.
Our results eventually shed light on the signif-
icant gap existing between theoretical models of
complexity mentioned in literature review, and how
complexity is considered, day-to-day, on the field
of small and medium architectural firms. While
complexity may indeed lie in the design of inno-
vative morphologies for larger architectural firms, it
rather hides inside administrative procedures, regu-
lations, building techniques and customers requests
for small and medium ones.
Before unveiling other layers of design complex-
ity to small and medium architectural firms, one has
thus first to acknowledge down-to-earth, day-to-day
operational challenges. Only then, architects will be
able to afford renewed thinking about what digital
design tools have to offer in terms of architectural in-
novation.
FUTURE WORK AND CONCLUSION
This paper looked into the challenges architects
working in small and medium firms face when deal-
ing with digital tools during their design processes.
It underlines how their perception of complexity all
along their day-to-day practices diverge from the cur-
rent trends discussed in literature review, especially
in regard to larger architectural firms.
Future work will concentrate on two areas.
Firstly, more data has to be retrieved in order to test,
and eventually validate and amplify the trends al-
ready identified in this paper. This data will cover the
entire country and may refine results in regard of dig-
ital culture and size of architectural firms. Secondly,
the part of the survey concerning the use of para-
metric design and tools will be investigated. Thirdly,
we will deepen our understanding of current working
strategies by interviewing selected offices that an-
swered positively to the possibility of further contact
and on-field observation. This final phase will help us
raising awareness amongst small and medium offices
tabout new digital design tools, researching whether
we should help Belgian designers adapting their pro-
cesses or rather push for software adaptations.
REFERENCES
Artemel, A. J 2015 'Q&A: Mario Carpo on Archi-
tecture’s Digital Past and Present', Metropolis -
www.metropolismag.com
Blumenthal, M 1971 '« Introduction », Techniques et Ar-
chitecture', Special number Informatique et architec-
ture, no 4, 33e série, p. 33
Bourbonnais, S 2015, Sensibilités technologiques : expéri-
mentations et explorations en architecture numérique
1987-2010, Ph.D. Thesis, Université Paris-Est
Chase, S and Murt, P 2000 'Evaluating the complexity of
CAD models as a measure for student assessment',
ACADIA 2000: Eternity, Infinity ans Virtuality
Corning, P 1998, Complexity is just a Word ! In Technolog-
ical and Social Change, Elsevier
Migayrou, F and Zeynep, M 2003, Architectures non stan-
dard, Centre Pompidou, Paris
Monier, V, Duchanois, G and Bignon, J-C 2012 'Généra-
tion de formes non-standards au moyen d’éléments
natifs irréguliers en bois – Elaboration d’un outil
numérique de conception architecturale', SCAN 12 :
Complexité(s) des modèles de l’architecture numérique
Picon, A 2010, Culture numérique et architecture,
Birkhäuser, Bâle
DESIGN TOOLS | Applications - Volume 2 - eCAADe 34 | 167
 Shelden, D 2002, Digital Surface Representation and the
Constructibility of Gehry’s Architecture, Ph.D. Thesis,
Masachusetts Institute of Technology
Stals, A, Elsen, C and Jancart, J 2015 'Challenges in teach-
ing architectural morphogenesis', Proceedings of the
3rd International Conference for Design Education Re-
searchers, Chicago
Tchinda, C 2013 'Rapport d’enquête : conditions de tra-
vail des architectes', Unpublished report conducted by
French and German-speaking Architects Association
Weytjens, L, Verdonck, E and Verbeeck, G 2008 'Classifi-
cation and Use of Design Tools: The Roles of Tools
in the Architectural Design Process', Design Principles
and Practices: an International Journal

168 | eCAADe 34 - DESIGN TOOLS | Applications - Volume 2

Estimating Potential Event Occurrence Areas in Small
Space based on Semi-supervised Learning
Atsushi Takizawa1
1
Osaka City University
1
https://sites.google.com/site/drtakizawa/
1
takizawa@arch.eng.osaka-cu.ac.jp

We propose a method for relatively small space that can optimize the size and
shape of the neighborhood of an event occurrence spot on a grid space to
minimize the classification error using classification by aggregating emerging
patterns based on the concept of semi-supervised learning. We apply this method
to data of waiting people in the Umeda Underground Mall and show that the
proposed method can improve classification accuracy and understandability of
classification rules.

Keywords: Small space, spatial event, clustering, classification, mixed integer

quadratic programing

INTRODUCTION Extensions of Poisson regression and logistic regres-

Architects design spaces such as buildings and open
spaces that are relatively small compared to geo-
graphical or urban scale space, and various types
of events involving people are expected to occur in
such spaces. To validate such intentions, the space
syntax (Hilliar and Hanson 1984) community analyzes
the relationship between events and various types of
space such as office (Pen et al. 1999), shop (Min et al.
2012), and street space (Desyllas and Duxbury 2001).
However, the boundaries of such spaces are some-
times ambiguous, which makes spatial analysis dif-
ficult. Moreover, contemporary trends in spatial de-
sign such as continuous architecture, which involve
simple curved surfaces, makes clear segmentation of
space difficult and actively handles ambiguous space
boundaries. Such spatial ambiguousness can cause
difficulty in using space. Therefore, it is necessary for
us to know the relationship between spaces and hu-
man behaviors more than before deeply.
sion models used in geographically weighted gen-
eralized linear models are typically used to analyze
spatial events that occur in geographical scale space.
Those models define the area where the influence of
an observation point is the Euclidean distance from
the observation point. We refer to this distance as
bandwidth and optimize its size to maximize the like-
lihood of the model. Inevitably, the band becomes a
circle whose center is the event point. However, the
degree of freedom of a circle is insufficient to apply to
a small space because the discontinuous change of
spatial attributes is not negligible for a small space.
For example, visibility of space (Benedict 1979) is a
very important attribute for a small space, and this
tends to change drastically with a slight difference
in location. The author has previously investigated
the relationship between street crime and space (Tak-
izawa 2013). In that research, the author regarded the
crime occurrence location as a circle whose center is

DESIGN TOOLS | Applications - Volume 2 - eCAADe 34 | 169

 the actual crime occurrence point. The radius is opti-
mized to maximize classification accuracy. However,
some circle areas expand across roads, and this ex-
pansion was not adequate for our research purpose
because we generally feel that the opposite side of
the road belongs to different area.
From the above background, we propose a
method to optimize the potential event occurrence
area based on machine learning. In particular, we for-
malize the problem as a classification problem with
two classes, i.e., the occurrence and non-occurrence
of spatial events initiated by people. Here we regard
the neighborhood of event occurrence locations that
can become either class and optimize the shape of
the neighborhood to maximize classification accu-
racy under a certain constraint. Then, we apply this
method to the data of rendezvous spot in the Umeda
Underground Mall located in Osaka city. Matsuo et
al.(2013) researched waiting people in this mall. They
regarded the event occurrence area as a uniform cir-
cle with a radius of 1.2 m, which comes from the
distance of personal space. The proposed method
differs from their approach in that we optimize the
shape of a neighborhood with a higher degree of
freedom than that of a circle. Incidentally, cellular
automata and multi-agent based models that have
been frequently used in the architectural research
field are simulation models and essentially different
from our approach. Just neural networks that have
been also used in this field can be applied to a classi-
fier in the proposed method.
The proposed method can also be considered to
belong to semi-supervised learning (Zhu and Gold-
berg 2009) in which labeled and unlabeled data are
mixed. The proposed method regards the neighbor-
hood of event points as an unlabeled area and learns
a classifier without the unlabeled data. The proposed
method is also related to methods that optimize the
area shape to maximize the density of point data in
grid space, for example in spatial data mining (Mori-
moto et al. 2003). That approach is similar to the pro-
posed method in that both restrict the shape of area;
however, that approach is a type of kind of a cluster-
170 | eCAADe 34 - DESIGN TOOLS | Applications - Volume 2
ing task in unsupervised learning.
The remainder of this paper is organized as
follows. The next section presents the proposed
method. Then we validate the proposed method and
discuss the generality with other classifiers. Finally,
results and conclusions are presented.
PROPOSED METHOD
We explain the detail of the proposed method below.
Preliminaries
Throughout this paper, we deal with a classification
problem with two classes that are labeled Positive (P )
or Negative (N ) for event occurrence. We deal with a
two-dimensional plane and discretize it into square
cells, where the length of a side is denoted as cl. As
illustrated in Figure 1, we denote the set of all cells ex-
cept obstacle cells as C. We divide C into three sub-
sets, CP ⊂ C where events occur, CN ⊂ C where
events do not occur, and CU ⊂ C where both possi-
bilities for event occurrence exist. We refer to cells in
those subsets as event cells, non-event cells, and un-
labeled cells, respectively. Here CP ∪ CN ∪ CU =
C,CP ∩ CN = ∅, CP ∩ CU = ∅, CN ∩ CU = ∅.
We denote each cell as c ∈ C. Each cell has a spatial
attribute vector vc and a class label lc ∈ {P, N }. We
refer to a cell that is labeled P among unlabeled cells
through optimization as an estimated event cell.
Unlabeled cells are assumed to be within radius
r cells from any event cells, where r ∈ Z+ is a pa-
rameter (Figure 1). An unlabeled cell is classified as
either class P or N through optimization. Figure 1
illustrates two areas of unlabeled cells (red and blue
lines). The area outlined in red does not include an
obstacle cell. This is because an unlabeled area must
be a rectilinear ellipsoid, as described later. Gener-
ally, it can be considered that the neighborhood of
a spot has high spatial homogeneity. As mentioned
above, the problem with the geographical scale is
that it tends to use a circular area as the neighbor-
hood; however, the degree of freedom of a circular
area appears to be too small for the target problem,
i.e., relatively small scale.
Figure 1
Types of cells in the
grid space.
Figure 2
Area families in a
grid space.
Figure 3
Difference between
two areas: (a)
Rectilinear ellipsoid
area and (b)
non-rectilinear
ellipsoid area.
Therefore, we introduce a more flexible shape for the
area. For small space analysis, isovist and visibility
graph analysis, which are indices of the visibility of
space, are often used to examine the local property
of visibility from the perspective of a given point in
space. It is natural to limit the neighborhood area to
one within which the event occurrence spot is visible
from all points. Figure 2 illustrates some area families
on a grid space in order of the degree of freedom of
the shape. A rectilinear ellipsoid composed of rect-
angles where one corner cell is the event cell. A cell
in the connected area shares an edge or a vertex with
its adjacent cell. The area of a circle and rectangle are
co-visible, i.e., any two points in the area are visible
from each other. Rectilinear ellipsoids and isovists
are one-way-visible, i.e., only visibility from the cen-
ter point to any point in the area is guaranteed. In a
connected area, visibility between the event cell and
other cells is not guaranteed. Therefore, we immedi-
ately reject the connected area. Because the degree
of freedom of the shape is low, we also deselect the
circle and rectangle. An isovist is sometimes a spiky
shape because it is a set of visual lines and is also not
adequate. A rectilinear ellipsoid tends not to gener-
ate such over complex shape. Given these considera-
tions, we select a rectilinear ellipsoid as the shape of
the neighbor area.
A rectilinear ellipsoid is a set of rectangles whose
vertices on one side are shared by all rectangles.
Since a rectangle is convex, a rectilinear ellipsoid is
guaranteed visibility from the shared cell. Sets of four
rectangles are shown in Figure 3. Figure 3(a) illus-
trates a rectilinear ellipsoid and Figure 3(b) illustrates
a non- rectilinear ellipsoid because its right rectangle
slides in an upward direction and does not share the
event cell. There are many obstacles exist in the build
environment, and ordinal areas are affected by such
obstacles (red outlined area, Figure 1).
Classification Accuracy Evaluation
A confusion matrix, as shown in Table 1, is often used
to evaluate the performance of a classification prob-
lem. Table 1 summarizes the number of records ac-
cording to the success and failure of classification.
The most general criterion is accuracy
TP + TN
Accuracy = . Accuracy
TP + FP + FN + TN
values are high even if the number of the dataset
of each class is imbalanced, that is, the major class
tends to score a high precision value and that of
the minor class is low. Since spatial events occur
in a relatively narrow area of the whole space, the
dataset for narrow space analysis may have such an
imbalanced characteristic. In such cases, we should
use other indices such as balanced error rate (BER),
T P rate + T N rate
BER = 1− where T P rate =
2
TP TN
and T N rate = . The inter-
TP + FN TN + FP
val of TPrate and TNrate is both [0.0, 1.0], and the
higher value is better. BER evaluates both classes
evenly and is easy to minimize using mixed integer
DESIGN TOOLS | Applications - Volume 2 - eCAADe 34 | 171
 diagonal cells are c and pi , and S(c', pi ) denotes the Table 1
set of support cells for event cell pi within the neigh- Confusion matrix.
borhood of cell c' ⊂ R(c, pi ). A support cell is nec-
essary cell to satisfy the rectilinear ellipsoid constraint
for the target cell. A cell has a maximum of two sup-
programing in our problem setting. However, BER port cells adjacent to the cell. If either coordinate of
cannot distinguish the imbalanced combination of the cell is the same as that of an event cell, the cell
both values. Now let (T P rate, T N rate) denote the has one support cell, i.e., s1 (pi ) = s2 (pi ).
accuracy vector of a solution. The BER value of a solu-
tion with (0.8, 0.8) and another with (1.0, 0.6) is com- Problem 1: minimizexc ,c∈CU RSBER(AU , mp)2 ,
monly 0.2; however, we want to select the former so- (2)
lution because the balance of two accuracy values of subject to
the former is better.
Therefore, we propose another criterion, root (1 − T P rate)2 + (1 − T N rate)2
RSBER(AU , mp)2 = ,
squared balanced error rate (RSBER), 2
√ (3)
(1 − T P rate)2 + (1 − T N rate)2 ∑ ∑
c∈CU ac · xc
RSBER = . c∈CP ac +
2 T P rate = , (4)
(1) mp
∑ ∑
Since the RSBER value of the former solution is 0.2 c∈CN (1 − ac ) + c∈CU (1 − ac )(1 − xc )
T N rate = ,
and the latter is 0.28, this criterion can distinguish |C| − mp
such over imbalanced solutions to some extent. The (5)
square of RSBER also can be minimized quickly us-
∑
ing mixed integer quadratic programing (described xc = mp, (6)
in the next section) using recently developed mathe- c∈CU
matical programming solvers.
xc ∈ {0, 1}, (7)
Optimization problem
2xc' ≤ xs1 (p1 ) + xs2 (p1 ) or ... (8)
Let ac ∈ {0,
{ 1} denote the } class label of cell c ∈ C,
and A = a1 , a2 , . . . , a|C| denote the set of class
labels. Here ac = 1 if the (estimated) class is P , oth- or 2xc' ≤ xs1 (pn ) + xs2 (pn ) , ∀c' ∈ R(c, pi ),
erwise ac = 0. The class label of unlabeled cells is (9)
given by AU ⊂ A where AU is a set of class labels
of unlabeled cells that are predicted with classifier f . s1 (pi ), s2 (pi ) ∈ S(c', pi ), (10)
We also let xc ∈ {0, 1} denote whether an unla-
beled cell c ∈ CU belongs to an event cell (xc = 1) pi ∈ CP (c), ∀i ∈ {1, . . . , n}, (11)
or non-event cell (xc = 0). Let mp ∈ Z+ denote the
total number of event cells and supposed event cells. n = |CP (c)|. (12)
Obtaining the optimal xc (c ∈ CU ) that mini-
mizes RSBER for given AU and mp can be formalized Algorithm
as Problem 1 which is defined from equation (2) to Figure 4 illustrates the proposed algorithm and its
(10), where R(c, pi ) ⊂ CU denotes the set of un- outline. First, this algorithm learns a classifier with
labeled cells that are included in a rectangle whose the dataset excluding unlabeled cells (line 2). Then,

172 | eCAADe 34 - DESIGN TOOLS | Applications - Volume 2

Figure 4
Algorithm (upper
stand) and outline
of the optimization
process (lower
stand).
the algorithm predicts the label of unlabeled cells
with the learned classifier (line 6) and solves Problem
1 to maximize classification accuracy. Cells in the un-
labeled area are partitioned into estimated event cell
and non-event cell space (line 13). This procedure is
iterated while increasing mp by one, where mp de-
notes the size of the estimated event cells. Each cell
c ∈ C has a pair (vc , lc ) of attribute vector vc and
label lc . DP , DN ⊂ D denotes the datasets of CP
and CN , respectively, T (C) denotes the function
that generates a dataset of cells in C, CU P , CU N ⊂
CU denotes sets of cells of supposed event cells and
non-event cells in unlabeled cells respectively, and
imax denotes the upper limit of the iteration.
VALIDATION
We validate the proposed method with the data of
waiting people in the Umeda Underground Mall lo-
cated in Osaka, Japan.
Data and spatial attributes
Figure 5 shows the space of target area of the Umeda
Underground Mall. The entrance gates of two rail-
way companies and the entrances of two depart-
ment stores face this area. Therefore, this area is a
rendezvous spot for people. However, in this space
many columns and pathway connects from various
angles rather than orthogonally. The author con-
ducted a field survey on May 12, 2015, from 11:00 to
11:30. We regard a person who remains in the same
DESIGN TOOLS | Applications - Volume 2 - eCAADe 34 | 173
 and mpmax = 1000. We use CPLEX 12.6.3 to solve Figure 5
Problem 1. The specifications of the computer used Space of the target
for optimization are as follows: Intel Core i7 5960x area.
@4.2 GHz, 64 GB memory, Windows 8.1 Professional,
and C++ compiler was Visual Studio 2013 Ultimate.

Results
It took 33 minutes and 12 seconds to solve the prob-
lem. Figure 7 illustrates the precision of the opti-
mized dataset for each mp. The values change grad-
ually even for near mp. The best RSBER value is ob-
tained when mp = 466. We refer to the dataset at
this time as Opt. Table 2 lists the precision of Opt and
two different datasets for comparison. One has no
place for more than 30 seconds as waiting. Let C
supposed event cells (R=0) and in the other dataset
denote the set of cells in the target area (cell side,
cells are within a 1 m radius of each event cell and are
50 cm). We have 7926 cells in total and 137 cells in
uniformly regarded as supposed cells (R=1) approx-
which waiting or talking behaviors are observed. We
imating the personal space. Compared with those
assume that some visibility related spatial attributes
precisions values, the Opt result is remarkably high.
might affect waiting behavior and perform isovist
Thus, we can conclude that good classification accu-
analysis at the center of each cell c ∈ C. Accord-
racy cannot be obtained without adequate buffers.
ing to Batty (2001), we obtain the following attributes
Table 3 lists the top five CAEP itemsets that have
of isovist fields: the maximum length of axial lines
the highest value of contribution for Opt and R=0.
(dmax ), visible area (a), isovist compactness (Γ), and
Compared with those results, the overall tendency of
another index of isovist compactness (ψ). We also in-
the itemsets is similar. However, in the case of R=0,
troduce another spatial attribute of distance to the
some items such as dmax = H, a = M , and a = H
nearest obstacle (omin ). Figure 6 illustrates the dis-
appear in both classes. This is because we regard cells
tribution of those spatial attributes.
adjacent to event cells as non-event cells that have al-
most similar attributes to those of the adjacent event
Experimental settings
cell for R=0. As a result, the precision of R=0 is not
We use classification by aggregating emerging pat-
high, and the itemsets are somewhat complex and
terns (CAEP) (Dong et al. 1999) as a classifier because
more difficult to understand than Opt. In addition, let
CAEP can classify an imbalanced dataset quickly with
us consider the value of contribution. Except for the
high precision without any spatial treatment such as
top 1 and 2 itemsets, the contribution of the lower
a cost sensitive classification scheme. CAEP also out-
itemsets of Opt is higher than R=0. These results imp-
puts readable classification rules as itemsets. Note
that CAEP is described in detail in the appendix. In Table 2
order to apply CAEP, we discretize the numeric at- Comparison of
tributes to three levels evenly. We label the dis- precision of three
cretized intervals as L (low), M (middle), and H (high). datasets with CAEP.
The hyper parameters of CAEP are set as follows:
supmin = 0.1, grmin = 3, dimmin = 1 and
dimmax = 4. The parameters of the target problem
are: r = 3(= 150cm), imax = 10, mpmin = 152,

174 | eCAADe 34 - DESIGN TOOLS | Applications - Volume 2

Figure 6
Distribution of
spatial attributes.

Figure 7
Classification
precision with
optimized dataset
for each mp.

DESIGN TOOLS | Applications - Volume 2 - eCAADe 34 | 175


ly that, using the proposed method, the Opt dataset
is more clearly divided into two classes of sub
datasets.
Let us return to the Opt result in Table 3 and ex-
amine the itemsets in detail. In the case of class P,
we can often observe the combination of the long ax-
ial line, wide visible area, and closeness to the obsta-
176 | eCAADe 34 - DESIGN TOOLS | Applications - Volume 2
Table 3
Top five itemsets
with the highest
contribution (Opt
and R=0).
cle. On the other hand, in the case of class N, we can Figure 8
observe a single attribute such as distant from obsta- Distribution of the
cles and short axial line. These results are consistent estimated event
with commonsense. Most of the people in this area cells of Opt.
are waiting for other people. However, many pedes-
trians pass through this area from various directions.
Therefore, it is natural to assume that waiting people
tend to select positions where visibility is high and
that are out of the way of pedestrians.
Figure 8 illustrates the cluster distribution of Opt.
The area and shape of potential event cells differ ac-
cording to the location. The area of supposed event
cells tends not to spread in the inner area of the target
space. However, supposed event cells in the outer
area spread widely.
DISCUSSION
The proposed method can use any classifier; how-
ever, we used CAEP as the classifier. Since there is a
possibility that the derived dataset (Opt) is optimized
only for CAEP and not for other classifiers, we verify
the dataset with other classifiers. Here we use logistic
regression and a support vector machine (SVM) with
a linear kernel implemented in Weka 3.6.13, which
is often used for classification problems (Hall et al.
2009). Considering the imbalance in the number of
records in each class, the misclassification cost of mi-
Table 4
Precision with SVM
and Logistic
regression via Cost
Sensitive Classifier.
|DN |
nor class (class P) is set to times using CostSen-
|DP |
sitiveClassifier implemented in Weka. Those classifi-
cation precisions are listed in Table 4. In either clas-
sifier, the precision of Opt is remarkably high com-
pared with the dataset of R=0 and R=1. In addition,
the balance between TPrate and TNrate is higher
than that of CAEP. From those results, we can con-
clude that the dataset optimized by the proposed
method with CAEP has a certain amount of general-
ity for other classifiers. Incidentally, CAEP is a non-
linear classifier that combines multiple items. On the
other hand, logistic regression and SVMs are linear
classifiers. It is interesting that the precision of a lin-
ear classifier exceeds that of the nonlinear classifier.
Note that we discretized each numeric attribute to
three levels and this might give the effect of a non-
liner classifier to a linear classifier.
CONCLUSIONS
In this paper, we propose a method that can opti-
mize the shape of the neighborhood of an event cell
to minimize the classification error with CAEP. Then,
we applied the proposed method to data of waiting
people in the Umeda Underground Mall and found
the following.
Compared to the classification accuracy of the
proposed method and conventional ones, the pro-
posed method with CAEP shows the highest classi-
fication accuracies. Moreover, other linear classifiers
such as logistic regression and SVMs show the best
performance with the dataset optimized using the
pro-posed method. From this result, we can con-
clude that the proposed method does not depend on
the classifier.
The size of the optimized area of supposed event cells
varies by place, and they do not spread in a circle but
in a more complex shape. This result is consistent
with our intention.
The significant itemsets of CAEP for classification
reflect the itemsets of the original dataset and are
simpler and more understandable than later item-
sets. That is, the proposed method modifies the orig-
inal dataset slightly, such that the classification is sim-
pler.
The proposed approach and research methodol-
ogy have some limitations as follows.
The proposed method is a wrapper method that
does not limit classifiers. Since CAEP is not sensitive
to the selection of hyperparameters, we should per-
form additional experiments with other classifiers.
We need to assess the interpretation of the ob-
tained classification rules. As explained, the pro-
posed method simplifies classification rules; how-
ever, this might invalidate the true classification rules.
Nevertheless, since simplification is advantageous in
many cases, the proposed method can be used as-
suming conscious consideration of its benefits and
limitations.
We can obtain the optimized solution in approx-
imately 30 minutes for data whose radius is three
cells. However, obtaining a solution requires approx-
imately 12 hours the radius is increased to five cells
due to the rapid increase of the search tree in the
solver. More efficient algorithms and faster computer
are required to handle a large neighborhood.
ACKNOWLEDGEMENT
This study was supported by a Grant-in-Aid for Scien-
tific Research (C) (No.25420633).
APPENDIX
First, we explain the emerging pattern (EP), which
is the basis of CAEP. The EP is defined as an item-
set whose support increases significantly from one
dataset to another. Suppose an attribute A has two
distinct class levels C and C, and consider datasets
DC and DC obtained by partitioning the original
DESIGN TOOLS | Applications - Volume 2 - eCAADe 34 | 177
 dataset D according to the values of the attribute A.
Let t ∈ DC denote a record that belongs to C and
e ⊆ t denote an itemset of t. Support that represents
the ratio of e containing t in D can be expressed as
|t ∈ DC , e ⊆ t|
supC (e) = . (13)
|DC |
To maintain the generality of itemsets, we normally
use itemsets whose support is greater than or equal
to supmin ∈ [0, 1] ⊂ R+ . We only select itemsets
to those whose combination size of items is greater
than or equal to dimmin and less than or equal to
dimmax where dimmin , dimmax ∈ Z+ . The
growth rate of an itemset e from DC to DC , denoted
growthrateC (e), is
{ ing Emerging Patterns', Proceedings of the 2nd Inter-
supC (e)
supC (e) 6= 0
growthrateC (e) = supC (e) .
∞ supC (e) = 0
(14)
We only select itemsets whose growth rate is greater
than or equal to grmin ≥ 1 as EPs where grmin ∈
R+ . Since an EP with a higher growth rate is more
likely to be found in one particular class, such an EP
is thought to be effective for classification. CAEP is
based on this idea. The contribution of itemset e to
class C is
growthrateC (e)
contributionC (e) = supC (e).
growthrateC (e) + 1
(15)
Let E(C) denote the set of EPs for class C derived
from the training data. The aggregated score, which
represents the probability that an instance s belongs
to C, is
∑
score(s, C) = contributionC (e).
e⊆s,e∈E(C)
(16)
To compare aggregated scores between different
classes, we normalize the scores by the median of the
aggregated scores.
Finally, an instance s is classified into the class
with the higher normalized score.
178 | eCAADe 34 - DESIGN TOOLS | Applications - Volume 2
REFERENCES
Batty, M 2001, 'Exploring isovist fields: space and shape
in architectural and urban morphology', Environ-
ment and Planning B: Planning and Design, 28(1), pp.
123-150
Benedikt, M 1979, 'To take hold of space: isovists and iso-
vist fields', Environment and Planning B, 6, pp. 47-65
Chen, D, et al. 2004 'Efficient Algorithms for Approximat-
ing a Multi-dimensional Voxel Terrain by a Unimodal
Terrain', Proceedings of the 10th Computing and Com-
binatorics of Conference, pp. 238-248
Desyllas, J and Duxbury, E 2001 'Axial Maps and Visibil-
ity Graph Analysis - A comparison of their method-
ology and use in models of urban pedestrian move-
ment', Proceedings of the 3rd International Space Syn-
tax Symposium, pp. 27.1-27.13
Dong, G, et al. 1999 'CAEP: Classification by Aggregat-
national Conference on Discovery Science, pp. 30-42
Hall, M, et al. 2009 'The WEKA Data Mining Software: An
Update', SIGKDD Explorations
Hillier, B and Hanson, J 1984, The Social Logic of Space,
Cambridge University Press
Matsuso, K, et al. 2013 'Relationships between waiting
behavior of pedestrians and space structure of pub-
lic underground mall', Proceedings of the 22nd Japan
Conference of GISA, pp. D-5-3
Min, SY, et al. 2012 'The impacts of spatial configuration
and merchandising on the shopping behavior in the
complex commercial facilities', Proceedings of the 8th
International Space Syntax Symposium, pp. 8066.1-
8066.15
Morimoto, Y, et al. 2003 'Extracting Spatial Knowledge
from the Web', Proceedings of the International Sym-
posium on Applications and the Internet, pp. 326-333
Penn, A, et al. 1999, 'The space of innovation: Interaction
and communication in the work environment', Envi-
ronment and Planning B: Planning and Design, 26(2),
pp. 193-218
Takizawa, A 2013, 'Emerging Pattern Based Street Crime
Analysis - Street Level Spatial Analysis of Crime Lo-
cation Associated with Built Environment in Fushimi
Ward, Kyoto City -', Transactions of AIJ, 78(686), pp.
957-967
Zhu, X and Goldberg, AB 2009, Introduction to Semi-
Supervised Learning, Synthesis Lectures on Artificial In-
telligence and Machine Learning, Morgan and Clay-
pool Publishers
In Search of Design Parameters for Well-Being and
Creativity in Knowledge Work Environments
Piia Markkanen1 , Aulikki Herneoja2
1,2
University of Oulu
1
www.innostava.fi
1,2
{piia.markkanen|aulikki.herneoja}@oulu.fi

Contemporary knowledge working environments are rapidly evolving alongside

the digital and virtual technologies used for knowledge work. Contemporary
offices range from cellular offices with assigned desks to activity-based offices
with shared-desk policies and location independent blended working.
Furthermore, advanced technologies, such as intelligent and adaptive lighting,
are being implemented in our everyday surroundings, including working
environments. The changes require adaptation from both knowledge workers and
architectural elements of the environments. We propose that knowledge work
environments should be explored from user-centric point of view and we aim to
elucidate how the physical design of the workplace enhances well-being,
creativity and innovation of their users. The various parameters of knowledge
work environments should be specified through architectural design process but
also through ethnographic methods, which enable us to retrieve the parameters of
knowledge work environments users deem important. Using both spatial design
and functional design approach we will challenge the multidimensional problem
field of designing collaborative knowledge work environments.

Keywords: knowledge work, creativity, well-being, collaborative knowledge

sharing, activity-based office

BACKGROUND tivities while maintaining communication landscape

Simply put, knowledge work requires surroundings
that support both concentration and communication.
However, combining those two features in the same
environment has proven to be extremely difficult as,
for example, low noise levels suitable for concen-
tration differ from the level that is generated from
face-to-face communication and collaboration. Find-
ing solutions to combine the complex needs of vari-
ous knowledge work character types and multiple ac-
that supports collaboration is critical for the next
generation knowledge work environment design.
Also, understanding personal needs for environmen-
tal factors that support individuals' outcome. Holis-
tic understanding of knowledge work environment
is critical because perceived lack of privacy, acous-
tic problems and decreased confidentiality may af-
fect users' satisfaction negatively and thus, instead
of promoting well-being, creativity and collaborative

DESIGN TOOLS | Applications - Volume 2 - eCAADe 34 | 179

 knowledge creation, perceived productivity and effi-
ciency is decreased (Vishcer 2008; Van Yperen 2014)
Sharing creative ideas and implicit knowledge is
often instrumental for generating successful innova-
tions in organizations and enterprises. This, in turn,
is dependent on well-being and sense of comfort
the workspace generates for its users. Interaction,
teamwork, creative goals and recognition of creative
ideas are part of the social-organizational arrange-
ment of knowledge work environment. Therefore,
rather than designing a space for workstations; archi-
tects and organization heads should aim to construct
environments that serve as implicit knowledge medi-
ums. The physical settings of the space can encour-
age exploration, collaboration and discussion. Alter-
natively, the space can also carry unspoken message
of silence, which may be critical enabler for concen-
tration intense working. The space for communica-
tion and collaboration itself should support listening,
sharing, presenting and comprehending knowledge,
thus supporting the flow of knowledge through the
space and its users (Aznavoorian and Doherty 2011;
Turner et al. 2013). In this article, we aim to approach
the features of contemporary knowledge work en-
vironments from the point of view of well-being
and creativity, which, in our opinion, are the criti-
cal enablers for innovation and sustainably success-
ful organizations varying from small scale growth-
oriented companies to large, established enterprises.
CONTEMPORARY KNOWLEDGE WORK EN-
VIRONMENT TYPOLOGIES
One of the current paradigms of knowledge work en-
vironments is their mobilization alongside with the
state-of-the-art virtual and digital mobile technolo-
gies. We are now free to move with our laptops
and tablets into any nearby café, or choose to work
at home. Contemporary offices take into account
the blended working (i.e. time-independent and
location-independent working) practices through
non-assigned desk policies and activity-based work-
station configurations (Van Yperen et al. 2014). Inter-
estingly, and in total contrast to increased individual
180 | eCAADe 34 - DESIGN TOOLS | Applications - Volume 2
freedom of choosing our desk location either on-site
or off-site, several organizations value teamwork and
collaboration. The teamwork practices typically in-
volve different work phases requiring both collabora-
tion and individual, high-concentration demanding
tasks. The individual users' needs for collaborative
environment and more private, distraction free en-
vironment has been critical paradigm work experts
from various backgrounds have tried to solve.
The characterization of four different knowledge
worker characters by Greene and Myerson (2011)
describes contemporary knowledge work culture
through knowledge workers' different need for their
organizations' physical environment. The work tasks
require different levels of mobility, communication
and autonomy and thus, office environment should
be able to respond to that. Shortly, Greene and My-
ers categorized four characters (presented in Figure
1): Anchor, Connector, Gatherer and Navigator. It is
important to understand that all the four characters
require workstation in the office either on daily ba-
sis or only occasionally. Therefore, creating a shared
environment is a complex organizational and archi-
tectural design task.
• Anchor - iconic office worker who is con-
stantly present in the office. Low mobility and
high need for comfort. High interaction. An-
chor has vital role in knowledge transform in
organization. Tasks often require concentra-
tion.
• Connector - spends only part of his days at his
desk and interacts with people within the or-
ganization a lot. Work environment changes
a lot during the day. Connectors often needs
more visual tools to support their work and its
creative processes than the clean-desk policy
permits.
• Gatherer - interacts with people outside the
office and brings back information into or-
ganization. Depends on off-site working
through mobile and wireless technologies.
Uses office for concentration and face-to-
face collaboration. Gatherers' presence in of-
 fice requires good shared-desk opportunities,
otherwise they rely on off-site blended work-
ing.
• Navigator - Rarely in the office, but highly ex-
perienced and beneficial for the organization.
For organization and team to benefit from
Navigators' presence requires good shared-
desk opportunities and perceived feeling of
being welcome.
To respond to the varying needs of collaborative
knowledge sharing and teamwork, one knowledge
work environment solution is the multi-space office
typology, developed by Boutellier et al., which offers
versatile selection of spaces, such as teamwork area,
quiet zones, meeting rooms and informal hang-out
areas (Boutellier et al. 2008). In essence, one con-
cept provides high diversity of workstations for dif-
ferent activities that support transfer of tacit knowl-

Figure 1
The four worker
character types
work on-site and
off-site with
different mobility
levels: Anchor (A),
Connector (C),
Gatherer (G) and
Navigator (N)
Modified from
Greene and Meyers,
2011.

DESIGN TOOLS | Applications - Volume 2 - eCAADe 34 | 181

 edge by enabling more face-to-face interactions be- into account while designing new working environ-
tween researchers. Importantly, users have also the ments. There is very limited amount of research of
opportunity to withdraw into quiet areas for high- how collaboration technologies are used in knowl-
concentration tasks or alternatively feel free to collab- edge work environment. Virtual status updates, in-
orate and discuss without disturbing others (Boutel- formation sharing, brainstorming and presentations
lier et al. 2008). Further variations are office envi- have different needs for both communication and in-
ronments with non-dedicated desks, such as activity- formation sharing (Marlow et al. 2016).
based offices (Appel-Meulenbroek et al. 2011) or flex-
offices (Bodin Danielsson et al. 2014). Fundamen- WELL-BEING IN KNOWLEDGE WORK ENVI-
tally, activity-based and flex-office types aim for same
RONMENT
workstation variety as multi-space offices in order to
In knowledge work environments, sharing creative
support different and changing work tasks. Further-
ideas and implicit knowledge is often instrumen-
more, these two types of offices respond better to
tal for success of the organization. This, in turn, is
more mobile knowledge worker types, such as gath-
linked to the well-being and sense of comfort the
erers and navigators.
workspace generates for its users. Well-being and
The most recent layer in contemporary knowl-
perceived job satisfaction in work environments can
edge working environment is use of various mobile
be assessed through physical environment and so-
and digital technologies. Several technologies have
cial environment, which includes both psychologi-
for long been routinely used in knowledge work,
cal and organizational aspects. Our personal envi-
including collaboration technologies that enable on-
ronment, which we experience in our own unique
site video-mediated distributed meetings and mo-
manner, is comprised of and affected by both so-
bile technologies that enable off-site blended work
cial and physical environment (Jaakkola 1998). There
(Van Yperen, 2014). On-site collaboration technolo-
are several indoor conditions that have a direct ef-
gies usually require various sized screens for virtual
fect on our health, comfort and productivity. It is
face-to-face interaction. In addition to interactive,
essential to remove possible external physical and
collaboration supporting, context, the collaboration
psychosocial stressors from working environment to
technologies support online sharing, analysis and
support users' well-being (Bluyssen et al. 2011). Fur-
documentation (Issa et al. 2006). One of the emerg-
thermore, improving the conditions in a manner that
ing research fields related to knowledge sharing and
users' satisfaction for their environment is improved
collaborative work is distributed user interfaces. In this
supports well-being and productivity in working en-
kind of user interface architecture, the components
vironments. Understanding how ambient conditions
of knowledge or other material are distributed across
(noise, lighting, air quality, thermal comfort), fur-
different hardware devices in space and time. To sup-
niture layout and ergonomics (workstations, offices
port various peer-to-peer communication situations,
and shared amenities), and process issues affect com-
Fisher et al. (2014) have envisioned scenarios of, for
fort and satisfaction of the knowledge worker is im-
example, a display wall of multiple screens and mul-
portant in studying and designing new knowledge
tiple computers interacting with digital media, or al-
work environments.
ternatively, envision a situation where information
is transferred from private device to a shared pub-
lic display (Fischer et al. 2014). Subsequently, col-
laboration technologies set different needs for pri-
vacy in shared working environments and spaces
and their current and future needs should be taken

182 | eCAADe 34 - DESIGN TOOLS | Applications - Volume 2

Figure 2 Functional and psychological comfort in
Environmental knowledge work environment supports
comfort model well-being
presents how users' Users perceive their feelings of personal work envi-
satisfaction and ronment through feelings, such as sense of territory,
well-being can be ownership and belonging (Vischer 2008). Discom-
divided to different fort affects negatively on satisfaction levels and de-
levels of comfort. At creases perceived productivity and knowledge shar-
the psychological ing. Vischer (2008) defined in her environmental com-
comfort level fort model of workspace quality different levels of oc-
occupants' cupant satisfaction and well-being (presented in Fig-
satisfaction to their ure 2) and their effect on working: In a working en-
environment is vironment, where the physical conditions meet the
highest and thus level of functional comfort, users are able to con-
increases serve their attentions and energy for their work tasks.
well-being and Elevating the user satisfaction and sense of control
efficiency. Modified in their own environment (e.g. adjusting the level
from Vischer 2008 of lighting), users reach psychological comfort level,
and interpretations which has positive effect to the knowledge sharing
of comfort for and creativity through decreased territoriality (Vis-
lighting in the cher 2008).
environment We consider ambient factors, such as noise, pri-
modified from vacy and lighting, important parameters in designing
Markkanen 2014. knowledge work environments that promote users'
well-being. For example, lighting has central effect on
the atmosphere of the space. During the office hours,
the prevailing lighting conditions are typically an
outcome of both natural daylight and artificial light.
Daylight is typically perceived as a positive health-
promoting factor in the knowledge working environ-
ments (Leslie 2003). The daylight varies in both color
and intensity and thus, its dynamic nature positively
influences on mood and stimulation (Van Bommel
and Van den Beld 2004). When artificial lighting is
considered through environmental comfort model,
well designed lighting can fulfill both functional and
psychological comfort levels and through increased
aesthetic appreciation and satisfaction to the envi-
ronment, lighting can increase workers' engagement
to work and productivity (Veitch et al. 2013, Vis-
cher 2008). When different parameters of lighting de-
sign are placed in the environmental comfort model
(Figure 2) (Markkanen 2014), we can argue that rec-

DESIGN TOOLS | Applications - Volume 2 - eCAADe 34 | 183

 ommended lighting levels of 500 lux (Van Bommel important because it affects interaction and subse-
and Van den Beld 2004) ensures that working envi- quently knowledge sharing through peoples' move-
ronment provides physical comfort for its user. Fur- ment and interaction in the space and through their
thermore, favorable lighting for individual users af- co-presence. Cellular offices provide private spaces
fects positively works structure, complex cognitive that support knowledge work that requires high lev-
appraisal and through work engagement, users' mo- els of concentration in a distraction free environment.
tivation towards work (Veitch et al. 2013). However, cellular office do not support interaction as
efficiently as open-plan offices, where same space is
CREATIVITY AND INNOVATION IN KNOW- shared by larger group. On the other hand, open-
plan office does not support concentration (Rashid et
LEDGE WORK ENVIRONMENT
al. 2005). Even though communication and collab-
Knowledge work tasks range from mundane tasks,
orative problem solving is important for ideation, in-
such as storing and retrieving information, to tasks
novation and product development in organizations,
such as planning, analyzing, processing information
knowledge workers often need privacy to think, to
and developing and designing products, which re-
analyze and to reflect in order to build upon their ex-
quire high cognition level (Heerwagen et al. 2004).
isting knowledge (Heerwagen et al. 2004; van Sprang
Dynamic knowledge and creation of new knowledge
2012). The interaction has been shown to be impor-
are continuous processes of increasing and updat-
tant also in different stages of product development.
ing the existing, personal or organizational, knowl-
Weak ties created in an interdisciplinary multi-space
edge base and problem solving (Tyagi et al. 2015).
office environment through chance encounters was
Knowledge creation processes are indispensable for in-
shown to promote creative thinking and idea pro-
novation and creation of new or improved variations
duction and increasing productivity in early stages
of products and services are essential for sustainable
of discovery. Subsequently, strong ties formed by
competitive advantage of organizations (Esterhuizen
frequent encounters and communications increased
et al. 2012). It is important to understand that creativ-
tacit knowledge transfer in a dedicated workspace
ity has different levels. Eminent creativity is relatively
devoted for a team work during later phases of prod-
rare and the products of it have a major impact. Ev-
uct development (Zoller and Boutellier, 2013)
eryday creativity is daily problem solving and individ-
uals' ability to adapt to change. Construction of per-
sonal knowledge and understanding is also defined
Knowledge sharing and creation of new
as a creative process (Hennessey and Amabile 2010).
knowledge
SECI knowledge creation model is linked to both
Innovation, on the other hand, can be defined as the
organizational structure and spatial architecture of
successful implementation of creative ideas and the
knowledge work environment (Nonaka and Konno
number and frequency of creative ideas and inno-
1998, Boutellier 2008). SECI is an abbreviation of
vations can be seen as the outcome of knowledge
following four phases of knowledge transformation
work. Communication is considered highly impor-
from tacit to explicit knowledge: socialization, exter-
tant action for creativity and innovation, which are
nalization, combination and internalization (Nonaka
eminently linked with each other: creativity is essen-
and Konno 1998). When situated in multi-space envi-
tially idea production, which is a crucial stage preced-
ronment, the steps of knowledge conversion can be
ing innovation. Innovation can be seen as a process,
considered as follows:
where the idea is implemented into product, service
or a problem solution that is valuable for individual
1. Socialization and face-to-face interactions
or a larger social group (Zhou and Hoever, 2014).
enable sharing tacit knowledge between in-
The layout of the work environment is highly
dividuals. This occurs during interactions in

184 | eCAADe 34 - DESIGN TOOLS | Applications - Volume 2

 spaces such as teamwork and break areas.
2. Externalization of the knowledge requires
converting tacit knowledge into explicit to
make it transferrable on peer-to-peer level in
meeting rooms.
3. Combination of knowledge occurs when the
explicit information is shared on organiza-
tional level.
4. Internalization of the new explicit knowl-
edge into tacit knowledge occurs as the final
phase of knowledge conversion, and this of-
ten requires concentration and silent working
environment.
Sharing the knowledge appears to be vital for inno-
vation and thus, it is important that architecture and
atmosphere of working environments support inter-
action. Integral part of SECI model is also the con-
cept 'Ba', which refers to specific time-space nexus
that can occur in physical, virtual or mental space
during interaction when new knowledge is created
or bridges gaps in information (Nonaka and Konno
1998; Tyagi 2015). 'Ba' lays foundation for all four SECI
modes for informal, simultaneous and dialectical di-
alogues between individuals and among groups. For
new knowledge creation to occur, it is vital to com-
mit to spend time and energy on activities and inter-
actions in 'Ba'.
PERSPECTIVES OF KNOWLEDGE WORK
ENVIRONMENT DESIGN PROCESSES AND
EVALUATION
Copious information on knowledge work envi-
ronments has been published from the point of
view of organizational knowledge creation and em-
ployee satisfaction, such as large self-reported re-
search studies or review analyses of existing studies
(Bluyssen et al. 2016; Wohlers and Hertel, 2016). De-
signing a workplace that enhances the performance
and creativity is challenging. Understanding the re-
quirements of workspace for creative thinking and
collaboration gives designers more tools to aid the
workspace design process. Nevertheless, from a de-
signer's perspective, the existing research does not
give direct answers how to design a working environ-
ment that promotes both well-being and creativity,
thus, providing a landscape that promotes privacy,
concentration, communication and control over per-
sonal environment. Although we are not able to
answer such complex design problem within this
publication, we aim to elucidate tools to design in-
novation supporting knowledge work environments
in growth-orientated companies within our research
project InnoStaVa [1].
In our project we will use user-centric ap-
proaches to generate information of knowledge
work processes and user experiences in their current
workspaces. This information will be analyzed and
design goals are defined to design new concepts of
knowledge work environments, which better sup-
port creativity and well-being. These concepts will
be tested 'in-the-wild' by constructing pilots in the
premises of startup companies to evaluate the user
experiences. The gathered information will be im-
plemented in subsequent rounds of concept design
and piloting. Through this interventionist manner,
we will bring the test-bed of enhanced knowledge
work environment to users.
Alongside the research-by-design approach for
generating new concepts for creativity and well-
being supportive work environments, we intend to
generate a network of workspace parameters in or-
der to understand how individual factors and their
combinations affect to the knowledge work envi-
ronments and their functional use. These parame-
ters will be generated during the design phase of
workspace concepts and also from the user experi-
ence data generated from the pre-analysis and eval-
uation of the current workspaces and pilot test envi-
ronments of local startup companies.
WHAT CAN WE ACHIEVE THROUGH PARA-
METRIC ANALYSIS OF KNOWLEDGE IN-
TENSE WORK ENVIRONMENT?
In the beginning of this paper we simplified the re-
quirements of knowledge work environment into
DESIGN TOOLS | Applications - Volume 2 - eCAADe 34 | 185

two words: concentration and communication. We
also discussed briefly of well-being in knowledge
work environment, creativity and knowledge shar-
ing. In our research project we approach de-
sign of knowledge work environments in a holistic
manned while acknowledging the impact of individ-
ual users' personal preferences regarding knowledge
work supporting factors. One view of network of
key parameters that affect well-being and creativity
is presented in Figure 3.
In order to use contemporary knowledge work
environment most efficiently and in a manner that
supports both well-being and creativity, their users
should demonstrate more choice-making, control
and active workstation switching according to task-
requirements. Users should also be able to adjust
186 | eCAADe 34 - DESIGN TOOLS | Applications - Volume 2
Figure 3
Network of
parameters for
well-being and
creativity
supporting
knowledge work
environment.
ambient factors in the environment (lighting, level
of privacy, noise) to their individual preferences and
be able to work in distraction free environment when
needed.
Activity-based offices with desk-sharing and
multispace offices provide opportunities to switch
workstation to meet the requirements of the work
task. Surprisingly, not all knowledge workers use
the opportunity to switch the workstation and thus
their satisfaction to their environment is decreased.
However, the individuals who switch several times
a day have significantly higher satisfaction to their
work environment. (Hoendervanger et al. 2016). Fur-
thermore, regardless the desk-sharing policy, people
tend to claim workstation with personal items while
not using those (Appel-Meulenbroek et al. 2015). It
appears that negative effects of open-office environ-
ments with activity-based workstations could be re-
duced by informing how to use them in a manner
that supports the working.
We propose here that analyzing how individual
office environments are being used and understand-
ing the users' personal and task-related needs could
be used to organize the desk choosing and switch-
ing to support both concentration and communica-
tion and consequently increase both well-being and
creativity in knowledge work environment. Knowl-
edge workers use different digital applications daily
to document their work tasks and meetings. Daily
activities, such as meetings, are typically planned in
digital calendars. Knowledge workers would bene-
fit from an application, smart office grid, that would
understand the requirements of different tasks and
would propose a specific workstation. In the most
simplified manner, the smart grid could propose a
silent area for concentration intense work, meeting
room for meetings and teamwork area for collabora-
tion, similar to what a user could choose in a multi-
space office. Even this level of guiding might ben-
efit individuals who do not switch workstations ac-
cording to task and increase their satisfaction to their
environment. Also, this would benefit knowledge
workers whose activities are location-independent
and who use blended working environments - an ap-
plication would ensure an available workstation with
good collaboration opportunities with other team
members.
Organization might have certain perception on
individuals' worker character type that does not ap-
ply to the person and assigned workstation or en-
abled working does not necessarily support individ-
ual's performance (Greene and Myers, 2011). Hence,
extending the application to contain personal infor-
mation, such as self-perceived character type, per-
sonal preferences on level of privacy for concentra-
tion or collaboration intense work tasks and envi-
ronmental factors, such as lighting, noise, physical
proximity and accessibility. More advanced level of
intelligence to smart office grid would obtain feed-
back from the users of the space in relation to work
task and interactions. The application would en-
able designers and heads of organizations to plan
prior building or acquiring new facilities. In larger
organizations with activity-based work environment
with shared-desk policy is a risk of team members
scattering outside communication range. Assigned
desk-policy, on the other hand, promotes less chance
encounters. Matching interactions, setting up new
collaboration opportunities and following up posi-
tive communication and knowledge sharing expe-
riences would enhance interaction and innovation
landscape in organizations. Furthermore, the most
mobile knowledge work characters, such as gather-
ers and navigators, would effortlessly find suitable
workstation while working on-site, but they would
also find communication opportunities with people
they need to interact with. All things considered,
we propose that organizations and designers should
generate a location- and user-dependent database
of task-related and interaction-network supporting
parameters to enhance well-being through personal
environment and organizational creativity through
communication and collaboration.
AKNOWLEDGEMENTS
Presented work is supported by European Regional
Development Fund.
REFERENCES
Appel-Meulenbroek, R, Kemperman, A, Kleijn, M and
Hendriks, E 2015, 'To use or not to use: which type
of property should you choose?', Journal of Property
Investment & Finance, 33, pp. 320-336
Aznavoorian, L and Doherty, P 2011, 'Forces Driving
Change', in Coles, D (eds) 2011, Work on the Move,
IFMA foundation
Bluyssen, PM, Janssen, S, van den Brink, LH and de
Kluizenaar, Y 2011, 'Assessment of wellbeing in an
indoor office environment.', Building and environ-
ment, 46, pp. 2632-2640
Bluyssen, PM, Roda, C, Mandin, C, Fossati, P, de
Kluizenaar, Y, Mihucz, VG, de Oliverira Fernandes, E
and Bartzis, J 2016, 'Self-reported health and com-
fort in ‘modern’ office buildings: first results from
DESIGN TOOLS | Applications - Volume 2 - eCAADe 34 | 187
 the European OFFICAIR study', Indoor Air, 26, pp.
298-317
Van Bommel, W and Van den Beld, G 2004, 'Lighting for
work: a review of visual and biological effects', Light-
ing Research and Technology, 36, pp. 255-266
Boutellier, R, Ullman, F, Schreiber, J and Naef, R 2008,
'Impact of office layout on communication in a sci-
ence�driven business', R&D Management, 38, pp.
372-391
Bodin Danielsson, C, Chungkham, HS, Wulff, C and West-
erlund, HS 2014, 'Office design', Ergonomics, 57, pp.
139-147
Esterhuizen, D, Schutte, CS and Du Toit, ASA 2012,
'Knowledge creation processes as critical enablers
for innovation.', International Journal of Information
Management, 32, pp. 354-364
Fisher, ER, Badam, SK and Elmqvist, N 2014, 'Design-
ing peer-to-peer distributed user interfaces: Case
studies on building distributed applications', Inter-
national Journal of Human-Computer Studies, 72, pp.
100-110
Greene, C and Myerson, J 2011, 'Space for thought: de-
signing for knowledge workers', Facilities, 29, pp. 19-
30
Heerwagen, JH, Kampschroer, K, Powell, KM and Loft-
ness, V 2004, 'Collaborative knowledge work envi-
ronments', Building Research & Information, 32, pp.
569-598
Hennessey, BA and Amabile, TM 2010, 'Creativity', Annual
Review of Psychology, 61, pp. 569-598
Hoendervanger, JG, De Been, I, Van Yperen, NW and
Albers, CJ 2016, 'Flexibility in use: Switching be-
haviour and satisfaction in activity-based work en-
vironments', Journal of Corporate Real Estate, 18, pp.
48-62
Issa, M, Rankin, J and Christian, A 2006 'A framework
to assess the collaborative decision making process
in interactive workspaces', Proceedings of the Joint
International Conference on Computing and Decision
Making in Civil and Building Engineering, p. 14
Jaakkola, JJK 1998, 'The office environment model: a
conceptual analysis of the sick building syndrome',
Indoor Air, 8, p. 7,16
Leslie, RP 2003, 'Capturing the daylight dividend in
buildings: why and how?', Building and Environment,
38, pp. 381-385
Markkanen, P 2014 'Knowledge Work in Campus
Environment-Opportunities of New Technologies
in Working and Learning Spaces.', Proceedings of the
6th Annual Architectural Research Symposium in Fin-
land 2014, Oulu, pp. 255-267
188 | eCAADe 34 - DESIGN TOOLS | Applications - Volume 2
Marlow, J, Carter, S, Good, N and Chen, JW 2016 'Beyond
Talking Heads: Multimedia Artifact Creation, Use,
and Sharing in Distributed Meetings', Proceedings of
the 19th ACM Conference on Computer-Supported Co-
operative Work & Social Computing, pp. 1703-1715
Nonaka, I and Konno, N 1998, '"The concept of
"Ba": Building a foundation for knowledge cre-
ation', Knowledge management: critical perspectives
on business and management, 40, p. 40
Rashid, M, Zimring, C, Wineman, J, Flaningam, T, Nubani,
L and Hammash, R 2005 'The effects of spatial be-
haviors and layout attributes on individuals' percep-
tion of psychosocial constructs in offices', Proceed-
ings of the Fifth International Space Syntax Confer-
ence, p. 71
van Sprang, H 2012 'The effects of workplace design
on knowledge worker’s perceived productivity in
Dutch Universities of Applied Science', 11th EuroFM
Research Symposium
Turner, A, Welch, B and Reynolds, S 2013, 'Learning
Spaces in Academic Libraries–A Review of the Evolv-
ing Trends', Australian Academic & Research Libraries,
44, pp. 226-234
Tyagi, S, Cai, X, Yang, K and Chambers, T 2015, 'Lean tools
and methods to support efficient knowledge cre-
ation.', International Journal of Information Manage-
ment, 35, pp. 204-214
Veitch, JA, Stokkermans, MG and Newsham, GR 2013,
'Linking lighting appraisals to work behaviors', Envi-
ronment and Behavior, 45, pp. 198-214
Vischer, JC 2008, 'Towards an environmental psychology
of workspace: how people are affected by environ-
ments for work', Architectural Science Review, 51, pp.
97-108
Wohlers, C and Hertel, G 2016, 'Choosing Where to
Work at Work–Towards a Theoretical Model of Ben-
efits and Risks of Activity-based Flexible Offices',
Ergonomics, DOI: 10.1080/00140139.2016.1188220,
pp. 1-52
Van Yperen, NW, Rietzschel, EF and De Jonge, KMM 2014,
'Blended working: For whom it may (not) work', PloS
One, 9, p. e102921
Zhou, J and Hoever, IJ 2014, 'Research on workplace cre-
ativity: A review and redirection', Annu. Rev. Organ.
Psychol. Organ. Behav., 1, pp. 333-359
[1] http://www.innostava.fi
Manifold Façades
A grammar-based approach for the adaptation of office buildings into
housing

Camilla Guerritore1 , José Pinto Duarte2

1
ABC Politecnico di Milano/DCG FA ULisboa 2 Pennsylvania State University
1
camilla.guerritore@polimi.it 2 jxp400@psu.edu

This article focuses on the use of shape grammars in rehabilitation processes to

transform existing, obsolete building stocks into required building types. It is
described how a grammar-based transformation methodology can lead to the
development of a design tool that enables the exploration of preliminary design
solutions and the evaluation of their impact in terms of massing, functional
programme and, eventually, cost and energetic behaviour. The goal is to assess
the capacity of an existing building to be adapted to a different use. The article is
focused on the transformation grammar. In particular, it is investigated the
transformation of "office building types" into "residential building types", aiming
at defining a quicker and more informed decision-making process. Future work
will be concerned with evaluating the performance of the solutions generated by
the grammar.

Keywords: Rehabilitation, office buildings, adaptive reuse, addition strategy,

shape grammars

INTRODUCTION lowed by construction, which have high energetic

The overall goal of this research is to support rehabil-
itation processes based on adaptation, that is, on the
transformation of existing, obsolete building stocks,
into required building types.
According to the Intergovernmental Panel on
Climate Change (IPCC) Fourth Assessment Report
(2007), "Over the whole building stock, the largest
portion of carbon savings by 2030 is in retrofitting
existing buildings and replacing energy using equip-
ment due to the slow turnover of the stock." Such
rehabilitation processes represent, in fact, a viable
alternative to strategies that rely on demolition fol-
and environmental impacts. Within the built environ-
ment, commercial buildings are quite important in
this regard, given the number of obsolete buildings
of this kind that exist on the market and the possibil-
ity of rehabilitating them using an "across-use adap-
tation strategy" (Wilkinson et al. 2014), which con-
sists in the re-use of existing buildings that still have
a potential lifespan, by adapting them to a new func-
tion, thereby extending their lifecycle.
Moreover, transforming obsolete vacant office
buildings into housing, will respond simultaneously
to the crescent structural vacancy rate of offices,

DESIGN TOOLS | Applications - Volume 2 - eCAADe 34 | 189

 which will be unlikely reabsorbed in the market due
to multiple factors, as well as to the high demand
for new affordable dwellings, a situation that resulted
from the economic crisis of the past years. The reha-
bilitation process of an existing building to suit the
new conditions and requirements takes into consid-
eration functional, behavioural and structural charac-
teristics. It is therefore needed the development of a
redesign strategy, defined as the "process of modify-
ing an existing built design based on additional cri-
teria from another domain" (Chase and Liew 2001).
Shape grammars are the formalism explored in this
paper to systematize the desired rehabilitation strat-
egy, by identifying and encoding in the grammar the
design principles and rules for such an adaptation.
ADAPTING BUILDINGS THROUGH A less the functional programme or the future devel-
GRAMMAR-BASED METHODOLOGY
Shape grammars (Stiny 1980) are considered one
of the earliest methodologies of visual computa-
tion using "algorithmic systems for creating and un-
derstanding designs directly through computations
with shapes, rather than indirectly through computa-
tions with text or symbols" (Knight 2000). In the ap-
proach proposed in this paper, the shape grammar
formalism is not used to create new designs, but it
is adopted as an instrument to encode the principles
of a methodology for the rehabilitation of existing
buildings. In this sense, it follows the footsteps of pre-
vious works on design grammars for the customiza-
tion of mass housing (Duarte 2005) and on trans-
formation grammars for housing rehabilitation (Eloy
2012), which stem from Knight's transformations in
design concept (Knight 1994).
The aim is to encode in a rigorous way the rules to
adapt an existing building type to a new use, taking
into consideration specific contextual features. The
goal is twofold: (i) exploring different scenarios of con-
version for a certain building type, whether given or
not a preliminary functional programme; and (ii) sup-
porting the decision-making process by providing in-
formation for the evaluation of the different adapta-
tion scenarios.
190 | eCAADe 34 - DESIGN TOOLS | Applications - Volume 2
The rehabilitation methodology is organised in two
main parts: the first includes the definition of a gram-
mar for transforming buildings to a new function and
the second outlines a framework for assessing the
performance of the generated solutions. This paper
is focused on the former, while the elaboration of the
latter will be developed in a subsequent phase of the
research.
The grammar for adapting offices into hous-
ing can be defined simultaneously as transforma-
tive, because it is applied in the transformation of
existing parametric building types and generic, be-
cause it aspires to develop an "ontological clas-
sification of context-and-style independent design
moves" (Beirão et al. 2011) for the adaptation of a
"generic office building type" into housing, regard-
oper or user.
However, at this stage of the research, it is still
not clear whether it is possible to identify a "generic
office building type," even within a certain cultural
and temporal context, and to represent the overall of-
fice stock by a single generic grammar. To overcome
these difficulties, first we chose a particular building
as a case study, which we know to be archetypic of
a well-known office type widespread in the region
that requires the urgent development of specific re-
habilitation strategies, and then we used the selected
building to devise a transformation grammar that
can be applied to all instances of the type.
CONTEXT OF THE RESEARCH
Field of application: from a specific case to
a broader application
The urban region of Milan has been taken as the fo-
cus of this research due to the massive presence of
obsolete office buildings whose use needs to be re-
defined. A survey on vacant and abandoned offices
in this region has been developed (different sources
have been taken into account for the survey on vacant
and abandoned offices in the Milanese region, between
them: Municipality of Milan, Map of unused private
buildings available at [1] on May 31 2016) with the
Figure 1
From the original
case building to a
parameterized
schema of its
building type: left –
original floorplan;
middle – floorplan
after demolition of
unnecessary
elements; right – purpose of identifying the localization and the typo-
parametric schema logical, morphological, constructive, and performa-
encoding the tive behaviour of these buildings. The data collected
building type. in this survey suggested the coherent grouping of
the analysed sample into building types, character-
ized by substantial common features. Nevertheless,
within the same building type, there is considerable
variation in terms of dimensions and spatial relations
between the constitutive elements of buildings, in-
cluding external perimeter, structural elements, and
the position and configuration of the circulation core,
which enriches the corpus under consideration. Each
type has been described through parametric mod-
els as abstract representations of entire categories. In
this way, it is possible to work with families of shapes,
characterized by similar topologies but different pa-
rameters, and it is feasible to propose refurbishment
actions applicable to all the buildings with the same
features, that is, of the same type. As mentioned
above, to develop and test the proposed shape gram-
mar methodology a specific building that was con-
sidered an archetypical instance of the selected type
was chosen as case study (Figure 1).
Rehabilitation strategy: incremental addi-
tion
The main intervention strategy considered for the
rehabilitation is "incremental addition", that is, the
introduction of a new multidimensional envelope,
which can be 60 to 250 cm deep. By adding this en-
velope, the floor plan can be expanded both with
new interior spaces (e.g. new service areas or exten-
sions of the already existing spaces) and new exterior
areas, usually missing in a traditional office building
(e.g. loggias or balconies). The addition of this enve-
lope also facilitates the upgrading of technical instal-
lations.
In short, the advantage of this strategy is three-
fold. First, it maximizes the expected benefits of
refurbishment, namely "additional floor area, addi-
tional equipment, implemented technical installa-
tion, improved envelope performance" (Gaspari et al.
2013), by concentrating in the volumetric additions
the majority of the interventions. Secondly, it min-
imizes the operations inside buildings, which tend
to be more complex and expensive. And thirdly, it
permits a complete renewal of obsolete buildings
from several viewpoints: qualitative, by requalifying
neglected buildings and activating urban regenera-
tion processes of the surrounding context; spatial/-
functional, by adapting and equipping buildings with
adequate spaces for a comfortable living standard;
structural, by intervening in buildings that are near
to the end of their life cycle (about 50 years) and
need seismic improvements; and energetic, by im-
proving the envelope, reducing energy consumption
and enabling energy production through renewable
sources.
A FRAMEWORK FOR THE GRAMMAR-
BASED REHABILITATION METHODOLOGY
The proposed methodology dwells on the idea that
an algorithmic design tool for generating and test-
ing new design configurations from given contextual
data, could be useful for the preliminary evaluation
of the efficacy and repercussions of different refur-
bishment scenarios. The methodology takes into ac-
DESIGN TOOLS | Applications - Volume 2 - eCAADe 34 | 191
 count: (1) boundary conditions; (2) urbanistic con-
straints; (3) construction constraints, according to na-
tional and local regulations; and (4) information de-
scribing the specific building type, including struc-
tural, spatial, and functional characteristics.
The main steps of the proposed methodology
are organised in four phases and rely on "user inputs"
and "generated outputs" as described in following.
The fourth phase, the Evaluation, is here mentioned
only for the sake of completeness, but it will be de-
veloped in future work.
-Diagnosis (Input): it concerns the analysis of the
existing building and its context to determine admis-
sible incremental additions scenarios. In this analysis
are considered aspects related to (i) dimensions; (ii)
constructability; and (iii) constraints.
-Problem setting (divided into Brief Finding and
Design Strategy): it deals with the definition of
the preferred functional programme (percentage of
dwelling types and mix of incremental additions) and
the identification of the main design strategies, both
for the division into dwellings and for the allocation
of incremental additions, to use in the generation
of scenarios within the grammar. The result of this
phase is the definition of the "ideal rehabilitation pat-
tern" for the adaptation of the building type.
-Design process (Formulation and Generation): it
consists in the application of the ideal rehabilitation
pattern to the real building. Starting from the out-
lined brief, the final functional programme will be for-
mulated as a combination of the "ideal programme"
and the building capacity identified in the Diagnosis
phase. As a result, it will be generated a system of so-
lutions for adapting the building, able to merge the
set of constraints, given by the original building and
its context, and the admissible incremental additions.
-Evaluation: it aims to establish the "transforma-
bility" of a specific building by obtaining transforma-
tion indicators that express its aptitude to reuse.
192 | eCAADe 34 - DESIGN TOOLS | Applications - Volume 2
THE OFFICE INTO HOUSING ADAPTATION
GRAMMAR
Methodology
The development of the grammar rules evolved it-
eratively taking into account normative constraints,
features of the dwelling types, and characteristics of
incremental additions. In each iteration, rules were
tested by applying them manually to a specific case
and then refined to get a more satisfying result. The
resulting grammar uses different representation de-
vices, including: 1) lines representing walls combined
on a plane to generate the floor plan; 2) planar sur-
faces representing dwelling at the building level and
rooms at the dwelling level 3) labels to express at-
tributes of shapes and sub-shapes, to represent con-
textual conditions, and to indicate where subsequent
shapes can be placed; 4) graphs, made by points and
lines, to represent required and de facto topological
relations between the different components of floor
plans; 5) weights, to indicate properties of the shapes
(for example constructive features) and to identify
appropriate areas for the allocation of specific func-
tions.
In order to have a process of generating solutions
amenable to use by designers, it was necessary to
define compositional principles. These principles are
encoded into a "design matrix" that represents both
the features of the given building type and admis-
sible boundaries of final design solutions. The gen-
eration of solutions by the grammar works by "pop-
ulating" the grid with 2D shapes that represent the
dwellings and their respective incremental additions.
Structuring the grammar: the levels
The structure of the adaptation grammar is organ-
ised in different levels, each subdivided in stages and
steps that correspond to the various and often inde-
pendent phases of the design process (Figure 2):
Level 0_Preparation of the Floor Plan: it deals with
the definition of a representational system that iden-
tifies the building elements and the characteristics
of its context. The main properties that have been
defined and play a specific role in the application of
Figure 2
General structure of
the Office into
Housing
Adaptation
Grammar organised
into six levels.
the grammar rules comprise: (1) contextual features;
(2) building functions; (3) constructive features. This
level will be extended with rules for making demoli-
tions. In fact, before design can proceed it is neces-
sary to remove from the floor plan all the elements
that are not required for conversion into housing,
such as internal partitions, specific office appliances,
and additional stairs and elevators.
Level 1_Design Matrix Definition: it consists in the
definition of a system of axes (structural grid, central
core and circulation loop) and a system of bound-
aries (admissible perimeters for the allocation of ad-
ditions) related to the initial building features and to
the range of transformation possibilities, constrain-
ing the alignment of walls and the assignment of
functions on the design matrix. This derives from the
superimposition of axes and boundaries and it pro-
vides the metrics for the composition of new inte-
rior and exterior spaces, defining modules and sub-
modules to use in the organization of space.
Level 2_Division into Dwellings: it is meant to ex-
plore the ability of the floor plan to accommodate
a specific or variable functional programme and it
will be described in detail in the following sections.
It has to be considered together with level 3, which
deals with the assignment of incremental additions,
because they are strongly interdependent.
Level 3_Volumetric Additions: it regulates the spa-
tial configurations deriving from the allocation of in-
cremental additions. Together with level 2, it per-
mits one to explore the range of possible solutions
within the framework defined by the grammar rules,
thereby permitting to find a solution that suits the
purpose of the refurbishment, both in terms of func-
tional and performative behaviour.
Level 4_Internal Organization: it will be used
only to verify if the division into dwellings deriving
from the previous phases is able to generate inte-
rior spaces in an adequate manner. The minimum
dimension required for each room has been estab-
lished according to existing regulations and three
main internal zones have been identified: day area,
night area and service area. Once allocated the main
zones, the subsequent subdivision into rooms will
generate detailed organization of the dwelling.
Level 5_Facade Grammar: from the combina-
tion of the previous levels, it is possible to gener-
ate the corresponding elevations and a 3D model,
thereby allowing one to understand the formal con-
sequences of the different refurbishment strategies.
Level 6_Allocating Technical Installations this level
will be added in future work.
DESIGN TOOLS | Applications - Volume 2 - eCAADe 34 | 193
 A LEVEL IN DETAIL: LEVEL 2 "DIVISION
INTO DWELLINGS"
A detailed description of the level "Division into
Dwellings" is provided below. The main design ac-
tions that are performed during the execution of the
stages and steps of this level are: (1) the customiza-
tion of the design matrix, (2) the choice of a circula-
tion scheme, and (3) the allocation of dwelling types
in the programme, which runs in parallel with the
allocation of the Volumetric Additions (Level 3). The
main purpose of this level is to allocate dwellings
that match given conditions. A dwelling is defined in
the grammar as a 2D rectangular shape characterized
by width (w), length (l) and area (a) which can vary
within a pre-determined range of square meters. The
different dwelling types considered in the functional
programme were established based on the analysis
of the competitions for the design of new social hous-
ing in the municipality of Milan in the last ten years
(Living in Milan 1 (2002-2005), Living in Milan 2 (2005-
2009), Milano Housing Sociale "Bando 8 aree" (2008), In-
ternational Design Competition of Social Housing "Una
comunità per crescere" (2009-2010), Housing Contest,
(2010)). These include t0: studio flat, t1: 1-bedroom
apartment, t2: 2-bedroom apartment, t3: 3-bedroom
apartment and they can be allocated in different po-
sitions within the building boundaries, depending
on their dimensions. The two main positions are:
the corner position, that is, a location where the new
dwelling has two sides adjacent to external walls, and
the adjacent position, meaning a location where the
new dwelling has one side adjacent to an allocated
194 | eCAADe 34 - DESIGN TOOLS | Applications - Volume 2
dwelling and one side adjacent to an external wall.
In order to be an acceptable solution, the gen-
erated shape has to comply with four main condi-
tions: (1) the minimum area is within the predefined
range; (2) the width and length are within certain pre-
defined proportions; (3) lighting and ventilation are
guaranteed; and (4) a connection to the circulation
loop is guaranteed. The grammar rules include spe-
cific instructions that limit the ways in which they can
be applied so that a dwelling matching the above
conditions is generated. In addition to conditions in-
trinsically related to the dwelling shape, conditions
related to the context need also to be taken into ac-
count and these are expressed in the shape part of
the rule using labels and in the conditional part us-
ing if/then relations. Figure 3 shows the schematic
partial generation of a floorplan matching a specific
program following the grammar rules.
Stages and steps
The derivation of designs consists of five stages. Each
stage is divided into steps that are accomplished
through the application of the shape grammar rules
(Table 1). The consecution of stages in each level is se-
quential as each stage has to be fully finished before
the following can be initiated. Nevertheless, more
than one rule can be applied at a given stage in a
non-deterministic order and the execution of the var-
ious levels of the grammar often takes place in par-
allel. Level 2 concerns the allocation of dwellings in
the housing programme. In Stage 1, it is customised
the Design Matrix taking into account the context of
Table 1
Stages and steps of
Level 2_Division
into Dwellings,
which have to be
followed in the
adaptation of a
floorplan.
Figure 3
Simplified shape
rules (a) and partial
derivation of a
design for
formulated
functional
programme (b).
There are rules for
allocating
rectangles (A, B and
L), for dissecting
(N), connecting (M),
extending (H) and
reducing (I)
rectangles; rules for
assigning (C) and
modifying (D)
functions and rules
for allocating
shades (E). Legend:
T4 – 3-bedroom
apartment, T3 –
2-bedroom
apartments, T1 –
studio flat.

the building under consideration (Step 1.1) and ap-
plied the design strategies defined in the previous
phases (Step 1.2). In Stage 2, it is selected the cir-
culation strategy that is more adequate, considering
the functional programme and the features of the
specific building (Steps 2.1 and 2.2). The circulation
schema is the starting schemata for the subsequent
application of rules. Stage 3 includes the allocation
of dwelling types and it starts with the allocation of
the bigger dwellings in the corner position (Step 3.1)
and proceeds with the positioning of the remaining
dwellings in the functional programme following the
chosen compositional principles (Step 3.2). In paral-
lel with the execution of the Step 3, takes place the al-
location of the incremental additions of the dwellings
being allocated (Level 3_Volumetric Additions). In
Step 3.3 it is verified the compliance of the dwellings
generated in Step 3.2, after allocating incremental
additions, with area requirements for each dwelling
type, readapting their dimensions if necessary to sat-
isfy them. In Stage 4, the access to dwellings is guar-
anteed (Steps 4.1 and 4.2) by connecting them to

DESIGN TOOLS | Applications - Volume 2 - eCAADe 34 | 195

 the central core, either directly or through a corridor.
Rule application terminates once all the dwellings in
the programme are allocated (Stage 5).
Rules
The definition of the set of rules comprises: design
moves, requirements, conditions, and constraints.
The rules encode the design strategies and principles
used in the adaptation methodology, so that it can
be applied to a wide range of situations and adjusted
or modified according to the necessities that emerge
from their application to a concrete case in practise.
A detailed rule is shown in Figure 4 to facilitate the
understanding of the logic behind the development
and structuring of the grammar rules. It defines the
allocation of a dwelling in the corner position and it
aims at embedding the instructions for the definition
of any type of dwelling in the programme. Each rule
is organised in five parts: a shape part; a label part; a
conditional part, which imposes restrictions on func-
tional, dimensional and regulatory aspects; a descrip-
tion part, and an additional part that specifies the ini-
tial general conditions of the rule applications pro-
cess. These parts are described in further detail be-
low.
General conditions. They include: the set of
dwellings included in the design brief (D); the set
of dwellings allocated so far in the derivation (D'); the
set of functions in the floor plan when rules begun
to be applied, namely, from inside out, core area, cir-
culation loop area, habitable area, and incremental
additions area; (E), the possible different dwelling
types (t(dn)) and their dimensional requirements.
Moreover, it also includes the main contextual fea-
tures that affect the allocation of dwellings: (i) solar
orientation; (ii) contextual location of the different
elevations of the original building; (iii) position of the
building with respect to the lot boundaries; and (iv)
views types and quality.
Shape part. In the shape part, the rule is visually ex-
plained, indicating, on the left side, the shape be-
fore the application of the rule and, on the right side,
the result of the transformation. It also includes the
196 | eCAADe 34 - DESIGN TOOLS | Applications - Volume 2
design matrix, which constrains rule application. Fi-
nally, in articulation with the description part, it in-
cludes a graph, depicting the spaces and functions
involved in the transformation and the topological
relations between them.
Conditional part. The conditional part of the rule
comprises a section describing the dimensional re-
quirements and a section describing the functions.
The dimensional section specifies the range of pos-
sible areas for the different dwelling types. Alloca-
tion takes place only if the area available for the al-
location of the dwelling permits the introduction of
a shape within the specified dimensional range. In
the functional part, "F" represents the general func-
tion of "habitable area". "Ft" and "Fl" are exterior walls
(Wexn), and "Fb" and "Fr" are other generic habitable
areas. The conditional part also indicates that any
dwelling type can be allocated in the corner position,
starting from the bigger one and then considering
others in descendant order.
Description part. The abbreviated general descrip-
tion of an allocation rule has the following format:
Ri <Dn:Wexn ,{hba };F;D';E> → <Dn:Wexn ,{hba };d(tn );
D'+{d(tn )};E-{hba },E+{d(tn )}>
The left part of the description indicates: the stage in
the derivation to which the rule can be applied (Dn);
the functions of the surrounding shapes, namely ex-
terior walls (Wexn) and the inhabitable area (hba));
the function of the shape to which the rule can be ap-
plied (F); the set of dwelling types in the programme
already allocated in the floorplan (D'); and the set of
current functions in the floorplan (E). The right part
of the description includes: the functions of the sur-
rounding shapes (Wexn, hba); the function to allo-
cate, that is, the generic dwelling type d(tn), which is
added to the set of dwelling types in the programme
already allocated in the floorplan (D'); the inhabitable
area (hba) is deleted from the set of current functions
in the floorplan (E); and the allocated function d(tn) is
added to the set E. Each time the rules is applied and
a dwelling is allocated, it is deleted from the set D of
Figure 4
Example of detailed
rule for the
allocation of a
dwelling type in the
corner position,
with general
conditions, shape
part, condition part
and abbreviated
description.

the dwellings in the programme to allocate, to pre-
vent further allocation of the same dwelling in subse-
quent steps of generation, and it is added to the set
D' of the dwelling types in the programme already al-
located in the floorplan.
A TOOL FOR FACILITATING THE DECISION
MAKING PROCESS
The main goal of the ongoing research is the cre-
ation of a "scenario-making" tool that encompasses
all the principles behind the adaptation of obso-
lete office buildings into housing. This tool could
have a significant role in implementing transforma-
tion policies for existing building stocks as it could be
used in finding the most suitable strategy for adapt-
ing specific buildings in accordance with such poli-
cies. For this purpose, it is rather important to com-
plement software implementing the grammar with
software for evaluating solutions from various view-
points, which could provide performance indicators
and inform the generation of solutions matching de-
sired criteria. The final objective of the proposed tool
is not to generate optimized solutions but to provide
a universe of adequate solutions with known perfor-
mances thereby making it possible to understand the
trade-offs among different design alternatives and
the relation between geometry and performance or,
in other words, between form and function. In this
way, designers and developers could gain an ade-
quate understanding regarding the effects of their

DESIGN TOOLS | Applications - Volume 2 - eCAADe 34 | 197

 design decisions, since early stages of the design pro-
cess, while they still have the possibility of modifying
the solutions to attain specific design goals.
DISCUSSION AND CONCLUSIONS
The initial motivation for this research was the ne-
cessity to satisfy sustainable requirements for the
built environment, while responding to the growing
demand for new affordable dwellings. This paper
presents a grammar-based methodology for this pur-
pose. The proposed grammar enables the adapta-
tion of existing building types with a specific func-
tion, "office building," into a new converted function,
"residential building," through the application of a se-
quence of predefined and systematic operations, de-
fined as steps to be performed by rules in the gram-
mar.
An appropriate testing of this general framework
is still to be done. So far, the rule set was devel-
oped for the Level 2, Division into Dwellings, and
Level 3, Volumetric Additions. Nevertheless, it al-
ready takes into account functional, technical, and
construction requirements, thereby giving an indica-
tion of the feasibility and appropriateness of the pro-
posed methodology.
In the current stage of the work, the grammar
addresses the generation of floor plans in two di-
mensions. However, to fully evaluate design scenar-
ios, it will be necessary to develop the façade gram-
mar, foreseen in Level 5, which should articulate the
floor plan with the elevations, thereby encompassing
three dimensions.
Future work will be concerned with the com-
pletion, refinement, and computer implementation
of the grammar. This objective is crucial to achieve
some of the most important promises of the pro-
posed grammar-based approach, namely the fast
generation of design alternatives that are adequate
to particular design contexts.
The complete methodology for the rehabilita-
tion of existing building stocks also includes the eval-
uation framework for assessing the performance of
the generated solutions, which is yet to be devel-
oped.
198 | eCAADe 34 - DESIGN TOOLS | Applications - Volume 2
ACKNOWLEDGEMENTS
The grammar was developed at the DCG FA/ULisboa
and is part of a PhD research being developed at
the Department Architecture, Built Environment and
Construction Engineering, at the Politecnico di Mi-
lano.
REFERENCES
Beirão, JN, Duarte, JP and Stouffs, R 2011, 'Creating
generic grammars from specific grammars: towards
flexible urban design', Nexus Network Journal, 13(1),
pp. 73-111
Intergovernmental Panel on Climate Change, (IPCC)
2007, 'Fourth Assessment Report', [online], available
at [2] on May 31 2016, p. electronic
Chase, SC and Liew, P 2001 'A Systematic Method for
Redesign. Using function, behaviour and structure
to facilitate grammar transformation', Proceedings of
eCAADe 2001, Helsinki, pp. 18-24
Duarte, JP 2005, 'A Discursive Grammar for Customizing
Mass Housing: the case of Siza', Automation in Con-
struction, 14(2), pp. 265-275
Eloy, S 2012, A Transformation grammar-based methodol-
ogy for housing rehabilitation: meeting contemporary
functional and ICT requirements, Ph.D. Thesis, IST, TU
Lisbon
Gaspari, J, Antonini, E and Boeri, A 2013 'Volumetric Ad-
ditions for Sustainable Refurbishment of Residential
Buildings: From Theory to Practice', Central Europe
towards Sustainable Building 2013: Sustainable Build-
ing and Refurbishment for Next Generations, GRADA
Publishing, Prague
Knight, TW 1994, Transformations in design. A formal ap-
proach to stylistic change and innovation in the visual
arts, Cambridge University Press
Knight, TW 2000, 'Shape Grammars in education and
practice: history and prospects', [online] Mas-
sachusetts: MIT, available at [3], p. electronic
Stiny, G 1980, 'Introduction to shape and shape gram-
mars', Environment and Planning B, 7, pp. 343-351
Wilkinson, SJ, Remøy, HT and Langston, C 2014, Sustain-
able Building Adaptation: Innovations in Decision-
making, Wiley-Blackwell
[1] http://www.comune.milano.it/wps/portal/ist/it/s
ervizi/territorio/Monitoraggio_edifici_aree_s
tato_di_degrado
[2] http://www.ipcc.ch/publications_and_data/ar4/wg
3/en/ch6-ens6-es.html
[3] http://web. mit.edu/tknight/www/IJDC
Assisting Early Architectural Planning Using a Geometry-
Based Graph Search
Torsten Thurow1 , Christoph Langenhan2 , Frank Petzold3
1,2,3
TU München, Chair for Architectural Informatics
1,2,3
{torsten.thurow|langenhan|petzold}@tum.de

In early design phases of architecture ideas exist mostly on a vague level

concerning the expectations for the building plan and the respective design
parameters. One established method is to examine and develop ideas through
existing designs, and to use these to clarify design parameters and be further
inspired. Thus, the aim is a computer-based system like sketch-based query
approach to show similar floor plans using semantic building fingerprints.During
the search floor plans are compared in form of graphs, which means that the
sketch-based floor plans are converted to graphs together with the existing floor
plans. Herewith, a gradual condensation of the request is possible. The entry is
condensed continuously through the repetitive process of entry and search. The
challenges with this approach lie in the following mathematical model behind
similar floor plans, Queries that satisfy complexity of the data and optimal way
for the user to engage in search process.

Keywords: Semantic fingerprints, early architectural planning, geometry-based

graph search, adjustment theory

INTRODUCTION edge must be made accessible and processed into a

Against the backdrop of growth in global construc-
tion projects with soaring complexity levels, the de-
velopment and transfer of architectural knowledge
is becoming essential. The use of powerful IT infras-
tructure, as well as design and planning processes
that are based on semantic models (e.g. BIM/IFC and
GIS/CityGML), unleashes enormous potential - but
also increasingly generates large quantities of digi-
tal data that are more and more frequently stored on
the Internet in cloud solutions, model servers and the
like. This stored data contains "inherent" knowledge
which can be used as a "pool of ideas/inspiration"
for future projects. To do so, the "inherent" knowl-
searchable format. Information query capabilities are
a prerequisite for developing knowledge. To that end
there is a need for knowledge management systems
in which information can be mapped and harnessed
through knowledge representation and engineering
methods.
In early design phases, architects utilize refer-
ences; this is a recognized design method employed
to investigate and develop ideas through existing
plans, clarify design parameters and find new inspi-
ration. Alongside mainstream media such as books
and magazines, there is also growing use of digital
catalogs in this regard, although searches mostly in-

DESIGN TOOLS | Applications - Volume 2 - eCAADe 34 | 199

 volve key words or fixed categories. The objective of • Adjacency relationships between rooms
the approach described in the paper is to develop in- • Size ratios
novative search methods to support design activities • The geometry of room boundaries
during the early stages, the concept phase.
In early design phases, concepts - usually in the form
of sketches representing spatial arrangements and
Metis research project and the significance
configurations with textual annotations through to
of fingerprints
sketched scale drawings - are used as a means of
The aim of the research project METIS (Langenhan,
reflecting, visualizing and investigating mental de-
Weber, Liwicki, Petzold and Dengel 2013) (supported
sign ideas. During the design process, more concrete
by the DFG) is to develop innovative IT-based search
specification is given to the spatial model, i.e. the
methods to support design activities in the concept
rooms and their relationships to each other. The level
phase. The iterative nature of the design process re-
of abstraction is reduced, with geometric and topo-
quires continuous switching between creative, ana-
logical details fleshed out. Sketched graphical rep-
lytical and investigative steps in order to choose the
resentation is used as an input metaphor in the re-
best design approach by selecting from the most
search project. From there, "query graphs" are for-
promising alternatives. We therefore recommend
malized in accordance with the defined fingerprints.
utilizing semantic fingerprints to formalize aspects of
Search queries are submitted with the help of these
creative and analytical issues. Thus formalized, the is-
graphs, and the results obtained serve either as in-
sues can be interpreted by computer and related so-
spiration for (or more concrete specification of ) de-
lutions can be displayed as references. Buildings and
sign concepts, or as feedback for changing the search
plans serve as a knowledge base, encompassing not
queries.
only spatial situations but also solutions for specific
The approach used in the METIS project can
architectural forms. Reference usage is an efficient
be regarded as a multi-stage formalization of con-
method in both design work and downstream activ-
ceptual thought. The conceptual thoughts will be
ity fields.
sketched out and topological relationships between
In order to facilitate the comparability of floor
these sketches will be formalized in computer-based
plan solutions (full or partial) and aspects thereof,
graphs.
consideration is paid to specific floor plan character-
From a user's perspective, the process is as such:
istics, such as spatial topology properties, which are
a basic layout will be sketched out with the proper-
called semantic fingerprints (Figure 1). Semantic fin-
ties of spaces and connections such as geometry, the
gerprints should be understood as separate levels of
function of spaces, accessibility, special properties
abstraction containing various aspects or details of a
such as the orientation of the windows, etc. Different
spatial arrangement. A number of different finger-
types of fingerprints will be derived from this infor-
prints are defined for a reference in the form of a
mation and examined on the basis of these datasets.
constructed or planned building, and derived using
The types of fingerprints vary widely from room
computer-based processes. These fingerprints are
functions, throughway graphs, orientation relative
represented mathematically in graph form, allowing
to the cardinal directions, to geometrical properties.
floor plan solutions - graphs - to be investigated with
In the following, specific geometrical properties will
regard to their similarity. Examples of fingerprints in-
now be specifically considered as a subset of finger-
clude:
prints.
• The number of rooms on a floor plan
• Accessibility and non-accessibility between
rooms

200 | eCAADe 34 - DESIGN TOOLS | Applications - Volume 2

Figure 1
METIS approach:
The use of
fingerprints.
Geometrical fingerprints
In order to support a more finely or coarsely granu-
lated exploration, a new approach to interaction is
proposed below. In addition to classic input regard-
ing the shape of rooms and the definition of the rela-
tionship between rooms such as adjacency and circu-
lation, the entered room geometry can be modified
semi-automatically. Thus the room proportions or
the spatial arrangement can be modified by entering
the distances and angles within spatial boundaries,
i.e. the incremental generation of a scaled drawing
from sketched input. Since the corner points of the
room do not need to be individually selected and dis-
placed by the user, an increase in efficiency can be
assumed.
From the sketch to the scaled drawing as a
basis for deriving geometrical fingerprints
An algorithm was developed for the search applica-
tion, which supports the incremental generation of
a scaled drawing from sketched input. As such, the
rooms may not be to scale and can be entered us-
ing approximate room cubature (sketch) as well as
the topological relationships between them. With
this input, the user is already indirectly providing di-
verse information such as the topology of the out-
line of rooms, edges and outlines that extend parallel
to one another, etc. The geometrical dependencies
that are later explicitly entered by the user, such as
linear dimension, perpendicularity, parallelism, etc.,
supplement this information. In the event of incon-
sistencies, these will be meaningfully resolved. The
algorithm "adjusts" the layout in accordance with the
input values and, in addition to the direct dimen-
sional input, also the shows the user the dimensions
that are automatically determined. A scaled draw-
ing is incrementally generated from sketched input
by indicating dimensions. This drawing can be subse-
quently modified, concretized and prioritized at any
time (see Figure 2).
In the case of sketch-based input, the geometry
is concretized by inputting concrete measurements
such as linear dimension, diagonal dimension, per-
pendicularity and parallelism, including pre-defined
accuracies or those that can be defined on the user
side, as appropriate. The chapter, "Concretization of
the search query with the aid of geodetic adjustment
computation", shows the internal numerical imple-
mentation of this functional principle.
A "geometrical" fingerprint is derived from this
input, which implicitly uses the adjusted room ge-
ometry, and which explicitly uses the dimensions for
search and similarity measures.
Comparison on the basis of geometrical fin-
gerprints
The comparison of geometries is characterized by
various problems. Humans divide image informa-
tion in various ways and compare it against memory.
Humans specifically manage geometric comparisons
impressively well. Currently, these skills can only be
transferred to the computer to a limited degree. One
DESIGN TOOLS | Applications - Volume 2 - eCAADe 34 | 201

of the main issues is the mathematical definition of
similarity. Humans filter out a lot of detail differences
and thus are able to correlate geometric objects that
are roughly similar. Following this approach, a com-
parison algorithm is presented, which can recognize
similar geometries even in the case of locally insignif-
icant, topological differences. Humans not only dis-
tinguish between the same and different, but also
have a "feel" for greater or lesser degrees of correla-
tion. This approach has also been transmitted in the
form of a similarity measure.
Thus one problem is the recognition of correla-
tions when geometries are not oriented identically,
i.e., they are skewed relative to one another. Addi-
tionally, geometries often have symmetries, for ex-
ample in the case of a square or octagon. In this case,
the different alignment options need to be tested.
202 | eCAADe 34 - DESIGN TOOLS | Applications - Volume 2
Figure 2
Screenshot
Prototyp.
The chapter, "Comparison with the database",
presents a possible implementation of a correspond-
ing similarity comparison. It is based on the specific
allocation of the geometrical topography of a sketch
entered by the user (query) to rooms in a database.
Geometrical dependencies that are directly defined
by the user are tested against the database. Other
approaches are also possible, and a free approach
is presented in the outlook. Regardless of the im-
plementation, this fingerprint considers the similar-
ity of the three-dimensional shape as a criterion of
the overall similarity comparison. For example, the
fingerprints can be combined for the throughways,
room type and the similarity of the room shape.
 CONCRETIZATION OF THE SEARCH QUERY nates therefore allows said coordinates to be used
WITH THE HELP OF GEODETIC ADJUST- as approximate values for X 0 . In addition, in the
MENT COMPUTATION case of the incremental creation of the sketch, a re-
The method used for the geometrical concretization duction of the unknowns is already carried out. To
of the search queries will now be explained on the this end, pseudo-unknowns are introduced. Pseudo-
basis of this introduction. To this end, as described unknowns may be used as unknowns, however they
here, we will expand here on a model of geodetic are not variables but instead, reference functions,
adjustment computation. Adjustment computation which in turn reference unknowns and / or additional
is a mathematical method of optimization in order functions. Pseudovectors are constructed on the ba-
to determine the unknown parameters of the geo- sis thereof, the components of which (the entry of the
metrical, physical model for an array of measurement vector, for example the individual coordinates x, y or
data, or the parameters of a predetermined function. z in the case of position vectors) pseudo-unknowns.
In the approach described here, the sketch-based These references may continue recursively (Figure 3).
geometry incrementally adjusts according to the in-
Figure 3
put by the user, and connects this geometry with
Pseudo-unknowns
different geometrical abstractions and specifications.
and pseudovectors.
The Gauss-Markov model is used as a basis therefore.
Consequently, the modifications are traced in the se-
quence of the individual steps of the model. In the
The simplest and most frequent case is a multiple ref-
interest of simplification, the mathematical notation
erence to unknowns in the case of perpendicular ge-
was selected such that it is analogous to the standard
ometries (Figure 4).
for geodetic adjustment computation.
Figure 4 The primary goal of the Gauss-Markov is to cal-
Reduction with culate the adjusted unknowns X̂. The explanation
multiple reference. of the developed method is explained at the end of
the adjustment computation, in other words accord-
ing to the following points:

• Determining the approximate values

• Linearization of the observation equations
• Determining the functional and the stochastic
model
• Determining the system of equations, regular-
ization and solution
• Testing

These individual steps are then incorporated into the

method of the discrete observation groups.
[ ] [ ] [ ] [ ]
x1 x2 x2 x1
Determination of the approximate values P1 = , P2 = , P3 = , P1 =
x3 x3 x4 x4
and pseudo-unknowns (Note: xi always represents an unknown i according
The geometry of the layout is defined to a large ex- to geodetic notation)
tent by corner points such as those of the individual Thus the number of unknowns is reduced and
rooms. The sketch-based input of the point coordi-

DESIGN TOOLS | Applications - Volume 2 - eCAADe 34 | 203

 at the same time, hard constraints are modelled.
Other hard constraints must be modelled more
complexly. Here is an example [ of a direction
] Thus the matrices A and P and the shortened ob-
f1 (x1 , x2 , α)
vector r~1 as a pseudovector =
f2 (x1 , x2 , α)
[ ]
cos (α)x1 − sin (α)x2
, which is constructed on
cos (α)x1 + sin (α)x2
the basis of the unknown x1 , x2 and the scalar α.
From an information-technological point of view, the
result is function blocks that can be flexibly intercon-
nected. They have standardized inputs and outputs,
for example as transformation in R2 with an input
and an output vector. By solving the references, it
is ultimately possible to determine the actual quan-
tity of the true unknowns of a pipeline, which form
all pseudo-unknowns.
Adding observations and linearization
Geometric values such as distances are introduced
as observation and observation equation. By way
of simplification, these are regarded as uncorrelated.
The weight matrix P can therefore be easily formed
from the dialogue of precision. The model ma-
trix A is formed through linearization on the ba-
∆f
sis of the central difference quotient [1]
∆xi
f (x1 , . . . , xi + , . . .) − f (x1 , . . . , xi − , . . .)
. In
2
the case of a meaningful selection from  (see Holzer
2002), no disadvantages were identified as compared
to analytic differentiation but rather, a higher degree
of robustness was found (extreme value digits etc.).
The author used values such as those 2−20 , which
are coded in machine-internal floating-point repre-
sentation 2−20 · 1. The calculation times are negli-
gible in the overall process. The complexity is in lin-
ear proportion to the number of equations of con-
ditions since an equation of condition always uses
only a small number of constants. The approach also
solves the problem of linearization of the pseudo-
unknowns. These are solved by the quantity of un-
knowns formed thereby.
204 | eCAADe 34 - DESIGN TOOLS | Applications - Volume 2
Regularization and solution of the system
of equations
servation vector l arise in a linear calculation period.
A is extremely sparsely populated, P a diagonal ma-
trix, which allows for very efficient calculation N =
AT P A und n = AT P l .Though the approach of
first creating a sketch, and then supplementing said
sketch with geometrical values, there are, as a rule,
far more unknowns available than observations, and
thus N is not regular. In order to achieve rapid regu-
larization, continuous regularization ||N x − n||2 +
α||x||2 = min was used, since this yields the gener-
−
alized inverse Nα = (N + αI)−1 , and this simply
needs to be added α to each element of the diago-
nals of N . This approach improves the condition of
the matrix in the normal case, however is ill suited for
singular matrices with many iterative solvers, while
direct solvers have excessive calculation times and
produce fill in. An appropriately selected Krylow sub-
space method also produced good results with good
parallelization.
Testing and discreet observation groups
Both poor approximate values in the numeric sense
=
at startup and the rounding error of the iterative
solver determine even more the necessity of testing,
without which the system would quickly destabilize
during the startup process or would oscillate toward
the end with increasing precision. Like the overall
process, testing must also occur in a short period of
time, and thus must remain linear in complexity. The
mean unit of weight error m0 cannot be used since,
as described, u > n generally arises. Instead, Ω =
∑n
vT P v = (vi vi Pi,i ) is directly drawn on for the
i
assessment. A reweighting of the improvement vec-
ˆ
tor x' has proven effective in countering the issues
ˆ
initially described. Different α for x' = αx̂ are tested,
both for α > 1, α < 1 and α = 1. The calculation of
Ω can thus be easily parallelized; all vi vi Pi,i can be
calculated in parallel.
In the case of inconsistencies in the geometrical
 input conditions, there are different conceivable pos-
sible solutions. Discrete observation groups were in-
troduced in order to support monitoring by the user
during solution. As such, observations will be allo-
cated by the user-managed groups. A group is rep-
resented by a level. The adjusted observations L̂ of
a superordinate group are thereby attributed to the
next subordinate group as equations of condition.
In Figure 5, delta is symbolized as a gray rectangle.
In this way, quantities of (vector) components
arise, which are combined with each other. Any
quantity is thereby assigned to an individual vari-
able, which represents the pendant to the pseudo-
unknowns in the chapter, "Concretization of the
search query with the aid of geodetic adjustment
computation".

Figure 5
Reduction.
COMPARISON WITH THE DATABASE
As described in the chapter, "Introduction," the com-
parison of search query and database must account
for imprecisions. As explained, these are both ge-
ometrical and topological. The steps for the com-
parisons with each entry in the database are shown
below. Insofar as it is not clear whether the entry
matches, said entry will be referred to as a candidate
in the following. The steps are performed for each
new candidate.

Prior transformations
The prior transformation corresponds to the classic,
known method referred to as the Helmert transfor-
mation (Luhmann 2000, Niemeier 2008, Wolf 1968),
which prepares a candidate for the subsequent steps.
Following this step, a candidate that is oriented
roughly parallel to the coordinate axis should be P1x = P8x = x1
available. P1y = P2y = x2
P2x = P3x = P4x = P5x = x3
(1)
Reduction and consolidation equation P3y = P4y = P5y = P6y = x4
Now the topological description of the geometry of P6x = P7x = x5
the candidates will be simplified at the same time, in P7y = P8y = x6
that local manifestations are removed and the num- From a purely topological standpoint, points are thus
ber of geometrical parameters are reduced in a man- recognized, which can be consolidated. The identi-
ner analogous to the chapter, "Concretization of the fication only takes into account the consolidation of
search query with the aid of geodetic adjustment the (vector) components. If all components are con-
computation". The method is directly derived from solidated, this means that it is a reduction down to
the determination of the pseudo-unknowns. A con- just one single point. In Figure 5, P3x = P4x = P5x
formance test is performed for all (vector) compo- and P3y = P4y = P5y are thus reduced to a sin-
nents of point coordinates within a delta defined by gle point. In this way, the loops are reduced. After
the user. The equivalent value of the sketch input reduction, they are therefore freed from small local
(capture) can be used as an initial value for delta, characteristics (Figure 6).
however it may be modified by the user as desired. What is essential in this strategy is the defini-

DESIGN TOOLS | Applications - Volume 2 - eCAADe 34 | 205

 tion of the consolidation equation. This equation de-
scribes how, as a rule, similar but non-identical val-
ues for a quantity of (vector) components can be al-
located to the initial value of their substitution vari-
ables. Approaches such as the sole average of the
values fail to take into account the different influence
of the components on the overall geometry. More
suitable are variants, which take into account the el-
ement of a characterized environment such as the el-
ement ratios of the length of the adjacent edges as
weight, for example. Thus P2x and P3x have sub-
stantially more influence on the initial value of x3
than P4x and P5x and P5y and P6y have substan-
tially more influence on x4 than P3y and P4y .
P1x = P6x = x1
P1y = P2y = x2
P2x = P3x = x3
(2)
P3y = P4y = x4
P4x = P5x = x5
P5y = P6y = x6
Variant test
After the reduction, only those candidates with a cor-
responding topological description are considered.
This refers to the number of points in the loops
(Samet 2006) that form the geometry and the re-
duction of the point elements by means of com-
mon variables. The more symmetrical a geometry is
206 | eCAADe 34 - DESIGN TOOLS | Applications - Volume 2
in the topological sense, the more variants that can
be found for a candidate that correspond with the
query. In the case of Figure 6, the conditions accord-
ing to the block of formulas (2) are met by all rota-
tions and different definitions of the starting point in
the loop. These variants are now further tested and
reduced.
The first test radically reduces the number of vari-
ants and is structured very simply. In addition, each
corner point is examined from its "perspective" to de-
termine whether the further edge contour now ex-
tends to the left (sin (Pi−1 , Pi , Pi+1 ) > 0) or right
(sin (Pi−1 , Pi , Pi+1 ) < 0). If the change in direc-
tion is very small, it is flagged as such so that it is
not taken into consideration in the comparison. The
example yields the allocation P1 -> R, P2 -> R, P3 - Figure 6
> R, P4 -> R, P5 -> L, P6 -> R. In this way, the vari- Candidate after
ants are immediately reduced to 1, albeit mirrored reduction.
geometries are immediately seen as equal to 2.
Figure 7
Excemple of finger
print.
Figure 7 shows a sample query. In this case, the allo-
cation is PA -> R, PB -> L, PC -> R, PD -> R, PE -> R,
PF -> R. Thus it is immediately clear: PA -> P4 , PB
-> P5 , PC -> P6 , PD -> P1 , PE -> P2 , PF -> P3 .
PAx = PBx = xa
PAy = PFy = xb
PBy = PCy = xc
(3)
PCx = PDx = xd
PDy = PEy = xe
PEx = PFx = xf
The allocation thus gives rise to the allocation xa -> article, a distinction was therefore already made be-
x5 , xb -> x4 , xc -> x6 , xd -> x1 , xe -> x2 und xf -> tween points and corner points in the mathematical
x3 . description.
Once the reduced variables of the query and
variants have been allocated, the geometrical de- REFERENCES
pendencies and values entered by the user can be Holzer, S. M. 2002, Numerische Differentiation und Dif-
directly transferred. Figure 7 shows a distance of 6 ferenzenverfahren, Universität der Bundeswehr
m between √ the points PD and PE . This is modelled München, Institut für Mathematik und Bauinfor-
( )2 matik, https://dokumente.unibw.de/pub/bscw.cgi
as 6m = (PEx − PDx )2 + PEy − PDy = /d1419763/Differenzenverfahren.pdf
√
(xf − xd ) + (xe − xe ) . Thus ... yields the allo-
2 2 Langenhan, C., Weber, M., Liwicki, M., Petzold, F. and
√ Dengel, A. 2013, 'Graph-based retrieval of build-
cation in the variant l = (x3 − x1 )2 + (x2 − x2 )2 . ing information models for supporting the early de-
sign stages', Advanced Engineering Informatics, 27, p.
As a rule, l will not precisely correspond to the
413–426
desired 6 m. After the weighting, which was indi- Luhmann, T. 2000, Nahbereichsphotogrammetrie: Grund-
cated in the P matrix in the chapter, "Concretiza- lagen, Methoden und Anwendungen, Wichmann Ver-
tion of the search query with the aid of geodetic lag
adjustment computation", the difference is therefore Niemeier, W. 2008, Ausgleichungsrechnung, de Gruyter
weighted. This is done using all of the geometrical Lehrbücher
Samet, H. 2006, Foundations of multidimensional and
dependencies that have been directly or indirectly in-
metric data structures, Morgan Kaufmann Publ., Am-
dicated by the user, which are modelled as observa- sterdam
tions. The overall dimension of the deviations is de- Thurow, T. 2007, 'Bestandserfassung: Ein mitwach-
termined on the basis of the weighted differences. sendes Geometriemodell für planungsrelevante
Bauaufnahmen', in Riedel, A., Heine, K. and Henze, F.
(eds) 2007, Von Handaufmass bis High Tech II, Verlag
OUTLOOK Phillip von Zabern in Wissenschaftliche Buchge-
The variant of geometric comparison presented here sellschaft, pp. 85-92
looks for geometrical dependencies that are directly Wolf, H. 1968, Ausgleichungsrechnung nach der Methode
and indirectly fixed by the user, on the basis of corre- der kleinsten Quadrate, Dümmler Verlag
lating geographic topologies. Other approaches al-
low more freedom in the definition of geometric sim-
ilarity. As such, the approach pursued is similar to
hash values. Different topological and geometrical
properties are projected as a floating point number in
a function.This function is defined in such a way that
similar geometries also yield similar floating-point
numbers. Although statistically rare, roughly similar
geometries may have similar values, however, due to
the loss of information.
In this article, all examples for geometric proper-
ties that are directly entered by the user are consis-
tently considered in direct relation to corner points.
But it is also possible to introduce points as pure ge-
ometric reference points similar to construction aids.
To this end, additional steps must be added. In this

DESIGN TOOLS | Applications - Volume 2 - eCAADe 34 | 207

ColorTracker
Stefanie Holzheu1 , Sang Lee2
1
Aleatorix 2 Delft University of Technology & Aleatorix
1
aleatorix.info
1
s.holzheu@aleatorix.info 2 s.lee@tudelft.nl

With the work-in-progress research project ColorTracker we explore color as a

formal design tool. This project-based paper describes a novel software
application that processes color composition of a place and transcribes the data
into three-dimensional geometries for architectural design. The research
comprises two parallel trajectories: a theoretical survey and the software
application design. The theoretical survey presents the historical background of
color. The project-based research seeks to develop digital methods and
techniques that analyze the color compositions of the environment. Subsequently
the objective is a novel application software for smart mobile devices in order to
demonstrate the potentials of examining the color composition and chromatic
parameters of a given environment and how it can contribute to the design.

Keywords: color, architecture, urbanism, tracking, form-generation

As the cities become denser, competition for visibil-
ity between buildings and businesses increase. With
densification, we also find more and more spaces that
fall out of the desirable visual perception. Another
compelling problem stems from the leftover or in-
between urban places where no one cares to go. On
paper, these places may not be seen problematic.
But from the pedestrians' point of view, they may
stand out as dark spots that appear undesirable and
thus avoided. This research is motivated by the de-
sire to use new encoding technology to translate the
contextual color combinations for architectural and
urban design. ColorTracker (hereafter CT) envisions
a unique tool to augment the simulation and devel-
opment of chromatic formal composition for archi-
tectural design.
Digital technologies have greatly augmented ar-
chitectural work. They help increase quality and pre-
cision, as well as efficiency and performance. With
the computer technologies in architecture and de-
sign becoming increasingly sophisticated, the disci-
pline has transitioned from simply emphasizing pro-
ductivity toward generative design experimentation
and simulation. Now we use digital tools and tech-
niques to simulate, explore, organize, and solve prob-
lems with enhanced visual or structural complexity.
With the ongoing parametric design research project
ColorTracker we aim to explore one specific element
of environmental variables in architectural and urban
design: Color.
With the preceding points in mind, this on-
going research project consists of two continuous
phases: the theory-based research and the project-
based one. While the theory-based research aims
to establish the conceptual framework, the project-
based research, based on the scientific yet theoreti-
cal premises of the first phase, directly engages the
advanced inquiries in software development and de-

DESIGN TOOLS | Applications - Volume 2 - eCAADe 34 | 209

 sign application. In the following section, though not
a comprehensive summary, we will start with illus-
trating the complexity of the historical precedents on
color research.
Theory
Two people might agree about the color of the ob-
jects surrounding them but the perception of color
is actually a highly subjective experience. Physio-
logical as well as cultural factors play an important
role. The potential disjunction between the human
perception of color and the physical composition of
an object that exhibits color has been argued since
the ancient Greek philosophy. In "Color for the Sci-
ences," Jan J. Koenderink explains the dispute as fol-
lows: "The pessimistic notion that colors are 'mere
mental paint' and have no relation to the physical and
chemical constitution of things at all is popular in sci-
ence and (especially) in philosophy, but it has no ba-
sis in fact" (Quoted in Kuehni 2013:3). Thus, color con-
cerns a large body of research ranging from natural
sciences, to engineering and to metaphysics.
Definition. Color: Attribute of visual perception con-
sisting of any combination of chromatic and achro-
matic content. This attribute can be described by chro-
matic color names such as yellow or brown, red, pink,
green, blue purple, etc. or by achromatic color names
such as white, gray, black, etc., and qualified by bright,
dim, light, dark, etc., or by combinations of such names
(Kuehni 2013:3).
Color Values. The modern understanding of color
began with Isaac Newton. His observation of light
passing through a prism led to the discovery that
white light is a mixture of distinct color rays. New-
ton also states, "For the rays, to speak properly, are not
colored. In them there is nothing else than a certain
power and disposition to stir up a sensation of this or
that color" (Quoted in Kuehni 2013:3).
Thomas Young (1773-1829) most notably con-
tributed the wave theory of light and brought in-
sight into the nature of light. He related color to
wavelength and calculated the approximate wave-
lengths of the seven colors Newton recognized.  Fur-
210 | eCAADe 34 - DESIGN TOOLS | Applications - Volume 2
thermore, Young and Hermann von Helmholtz (1821-
1894) postulated that the human eye has three dif-
ferent types of color receptors with different spec-
tral sensitivities, specifying that those photorecep-
tors (cones) roughly correspond to red, green and
blue (RGB). Based on this trichromatic vision theory
the RGB color space was developed in the 1920. Sub-
sequent experiments proved that the RGB primaries
could indeed match all visual colors within a certain
range of color spectrum. To this day we use the RGB
color-space model and represent colors in terms of
the tri-stimulus values.
Kurt Nassau (1927-2010) contributed to how
matter alters the composition of light and causes di-
verse effects of color. He describes fourteen cate-
gory of causes, "[...] four dealing with geometry and
physical objects, and those remaining dealing with var-
ious effects involving electrons in atoms or molecules of
materials and causing absorption or emission of light
at selected wavebands" (Kuehni 2013:4). Nassau also
states that " [...] perceived color is merely the eyes'
measure and the brain's interpretation of the domi-
nant wavelength or frequency or energy of a light wave"
(Byrne & Hilbert 1997:3).
Cultural Preferences and Color Connotations.
Color conveys information and affects our perception
and response to our environment and what we see.
Depending on the environmental circumstances one
color may convey different meaning from one place
to the other. A color seen alone might hold a differ-
ent meaning than when it is seen along with another
color. When working with color we also have to rec-
ognize the cultural, personal and emotional disposi-
tion colors carry. Colors have different meaning in
different cultures for example: in Russia the color red
is associated with communism while in South Africa
it is the color of mourning. Likewise, the color blue in
India is associated with Krishna while in the Western
culture in general it is seen as a cool, masculine color
(Livingston 2014:133).
Goethe conducted one of the earliest formal ex-
plorations of color theory. Derived by the poet's in-
tuition (later corroborated in part by science) he pro-
vided the first catalogue of how color is perceived un-
der different circumstances. Goethe observes:
"Green: If yellow and blue, which we consider as
the most fundamental and simple colors, are united as
they first appear, in the first state of their action, the
color which we call green is the result. The eye experi-
ences a distinctly grateful impression from this color. If
the two elementary colors are mixed in perfect equal-
ity so that neither predominates, the eye and the mind
repose on the result of this junction as upon a simple
color. The beholder has neither the wish nor the power
to imagine a state beyond it. Hence for rooms to live in
constantly, the green color is most generally selected"
(Goethe 1840:316).
Following Goethe's lead, Johannes Itten further
explored the psychological aspects of color. Itten's
The Art of Color is a synopsis of his studies and teach-
ings where he describes among other aspects his
subjective association of particular colors with spe-
cific emotions. Itten "represented colors as expressive
moods and descriptions over the importance of differ-
ent types of color contrast: hue, light-dark, cold-warm,
complementary, simultaneously, extension and satura-
tion" (Opara & Cantwell 2014:172). In his writings It-
ten describes the impact of colors on the viewer: "Col-
ors are forces, radiant energies that affect us positively
or negatively, whether we are aware of it or not" (Itten
1970:12). Most of all, "Color is life; for a world without
color appears to us as dead... Colors are the children of
light, and light of their mother" (Itten 1997:8).
Project
How do we embody the cultural significance of color
connotations and include the scientific parameters of
visible light in architectural design? CT tackles this
question through geometry. CT grows out of the re-
search work on apparatus-centricity (Lee 2016). One
facet of the theory posits that the cultural construct
is largely driven by implementation and use of per-
ceptual and cognitive apparatuses that are designed
to augment and sometimes to replace human capaci-
ties. The CT research aims to develop an architectural
and environmental instrument to analyze, map and
transcribe the color palette of a given urban place
at the eye level for architectural analysis and form-
giving. CT approaches the topic of colorscape from
a numerical point of view. (Here the term colorscape
indicates a combination of colors that characterize a
specified environment. It will also be referred to in
video recording as creating the sensation of experi-
encing the colors of a particular environment or as
compositions created using the found color ensem-
ble of a particular environment, either exclusively or
in combination with visual performances.) It keeps
track of chromatic parameters such as brightness,
contrast, hue and saturation that can be organized
as data, not the perceptual and emotional interpre-
tation.
Within this framework, CT aims to develop a way
to track color compositions of cities and landscapes,
and utilize the information as a form-making tool. In
addition, the extended capabilities and application
may very well include the potential as an analysis tool
to diagnose and evaluate the color palette, satura-
tion and balance of a given architectural and/or ur-
ban environment. Therefore the application may be
used in mapping and studying the color composition
of cities and its in-between spaces. CT makes a tool
for both data visualization and form-making.
Iteration 1. In the first series of exercises, CT was
used to analyze colorscapes and map them in 2D in
the vocabulary of lines, shapes and materials. This
process defines spatial relations and expressions that
correspond to different arrangements of color in the
formal vocabulary.
Iteration 2. During the second series of exercises, the
colorscapes are treated as a design tool where the 3D
shapes generated represent the colorscapes. While
architects usually describe spatial relations in terms
of form, CT allows to describe the relations in terms of
the color qualities, which can also inform about light
or material. The CT app outputs a series of 3-d data-
forms in the .obj file format that can be easily used by
common modeling and design applications. The .obj
file from the CT application can then be modified and
elaborated on using various additional parameters.
DESIGN TOOLS | Applications - Volume 2 - eCAADe 34 | 211
 CT.1.4. In the current iteration CT.1.4 consists of
a custom application that can read from real-time
video camera feed or a prerecorded video file from
a video camera or mobile device. Here we apply digi-
tal programming methods to create a new design ap-
plication. The resulting custom CT application trans-
lates the color composition and chromatic parame-
ters of a given video input to inform and manipulate
a three-dimensional geometry and surface.
As mentioned before algorithmic processes are
used in the actual color analysis and form making
process. The program seen in Figure 1 was written
in Max/Msp/Jitter. Max/Msp/Jitter has been used by
performers, artists, and composers extensively in or-
der to customize applications for use in computer
music, interactive media and sound design. It is com-
parable to Grasshopper for Rhino in architecture as
it also follows a graphic approach rather than the
more traditional coding. It offers a unique and di-
verse collection of tools so called objects that are
visual boxes containing singular instructions to per-
form specific functions. Some create noises, some
other make video effects or perform simple calcula-
tions. These objects are added to a visual canvas and
connected with patch cords (virtual wires). By com-
bining objects, one can design an interactive, unique
software application.
The patch consists of three segments. The first seg-
ments contains the main components for the color
212 | eCAADe 34 - DESIGN TOOLS | Applications - Volume 2
tracking and the shape generator. The second sec-
tion are the sub-components that are necessary to
execute each function, such as the timer and the four
individual shape components for each color range.
The third section finally consists of the components
necessary to generate, save and export a shape as an
.obj file.
Analyzing the contents of the video input. Even
though the basic algorithms for tracking motion and
color have been around for a while (such as those in
animated movies), the dynamic nature of the tech-
nology has been hardly put to use in architecture.
Therefore CT takes advantage of such dynamic, real-
time algorithms in the analysis and signification of
colorscapes. This part of the application is designed
to define and track a particular color range in a video
as its position changes from frame to frame. This
technique helps locate a particular color within a
given scene. First the video is captured and displayed
in a separate window. It is possible to adjust upfront
the image quality, saturation, hue, contrast, and so
forth within the image control function of the pro-
gram. The Max MSP Jitter object suckah is positioned
over the window of the video and will output, after
simply clicking on the display window, the color of
choice as RGB color value. This value is used to al-
ter the video to a preferred color range. In the sec-
ond step, the altered video is channeled through a
Figure 1
jit.chromakey object. This object takes two inputs:
Custom
the video on the left and a reference color, in this
Max/Msp/Jitter
case, black, on the right input. This object measures
color tracking
the chromatic distance between each of the left in-
software; Aleatorix.
put's cells (pixels) and the reference black. This step
was performed to achieve a "green screening" effect.
A second suckah object over the new picture is used
to define the final color value for the tracking func-
tion. The output value is then routed through a small
sub-patch that defines a specific color range. Us-
ing only one definite value would not allow any light
ergo color changes in the input live video and there-
fore not deliver accurate data.
The object used to locate a particular color in an
image is jit.findbounds. It tracks the previously de-
 fined color range in the video using the min and max
attributes. When jit.findbounds receives a video in-
put, it looks through the entire input for values that
fall within the specified range. It then outputs the cell
indices (coordinates) that describe the region where
it located the designated values. In effect, it outputs
the indices of the bounding region within which the
values fall. The bounding region is a rectangle, so
jit.findbounds will output the indices for the left-top
and bottom-right cells of the region in which it found
the specified values.
The process described here is executed for each
color range individually and the coordinates are send
to the corresponding part of the form generator as
well as the sound output component for further use.
Figure 2
NURBS display
window; Aleatorix.
Figure 3
NURBS with four
control points as
flat plan; Aleatorix.
Figure 4
coordinate system;
Aleatorix.
Sound Output (alternative way to express the col-
lected data). This section transforms the received
values into audible sound. A makenote object out-
puts MIDI sound information paired with a velocity
value which can be adjusted. With the pgmaout ob-
ject a specific color is assigned to a specific sound
type. This component executes four times in total,
one for each color range. The result is a unique sound
montage based on the urban colorscape samples.
Form Generator. Starting point of the actual form
generation process is the jit.gl.nurbs object, which
renders a NURBS surface. Displaying the NURBS re-
quires a jit.gl.render object that renders the shape.
In total there are four NURBS displayed as there are
four different color ranges in the current iteration of
the application and each color range generates its
own shape as part of the final composition (Figure 2).
Through various control points defined in a jit.matrix,
the shape of the NURBS surfaces can be altered. In
the specific case here, the jit.matrix object specifies
the amount of control points to 40 x 40 points, which
describes the maximum amount of points in X and
Y direction. In the dimension component, the or-
der and initial position of these control points is pre-
defined to a regular grid. Several attributes can be
added to the jit.gl.nurbs object. On one hand there
are attributes equal for all four NURBS surfaces such
as position, rotation, predefined viewports and scale.
On the other hand there are the attributes specific
to each individual NURBS surface such as appearance
and color. At the start of the run-cycle the grid is set
to display four control points as a flat plane (Figure 3).
DESIGN TOOLS | Applications - Volume 2 - eCAADe 34 | 213

The values obtained from the color tracking process
are then used to alter the location of these control
points and therefore the geometry of the NURBS sur-
214 | eCAADe 34 - DESIGN TOOLS | Applications - Volume 2
face (Figure 4). Figure 5
To manipulate the location of a single control NURBS point and
point, 3 values X,Y and Z are needed. Figure 5 shows X-Y-Z coordinates;
which points are altered and how the values for these Aleatorix
points are registered to X, Y and Z coordinates. In this
case the point in column 0 (first column) and row 0
(first row) is altered. The initial location was X= 1.0,
Y= 0.0, Z= 0.0 which was altered to X= 16.0, Y= 19.75, Figure 6
Z= 14.48 (Figure 6). Y and Z coordinates are defined NURBS with four
through the color tracking process. To the X coordi- altered control
nate equals duration, a time length set by the user. points; Aleatorix.
The X coordinate therefore changes as the user chan-
ges the time settings. Every time a sequence is com-
pleted the column value will increase as there will be
one more control point added to each row, while the
row value itself stays constant, set to 2 rows. This
technique can be describes as a mono-directional (or
linear) growth process. At a given time, 2 control
points will be manipulated simultaneously.
The values of a given set of control points are
stored in the sub-matrix. In the next sequence the
consequent set of control points will be transformed.
This entire process can be repeated at least 4 times
but beyond that as many times as needed (Figure 7).
Once the video input has ended it is possible to Figure 7
save the shape in the .obj file format that for exam- shape example;
ple can be read in AutoCAD or Rhino (Figure 8). The Aleatorix
shape will be exported in a rectangular grid mesh.
Therefore the amount of control points must be a
multiple of 4. A sub-patch monitors the amount of
control points. If the user chooses to save the shape
it will check if the value is in multiples of 4. If it mis-
matches, it will set the amount of control points to
the next smaller multiple of four.
Calibration. The visual performance of the camera
sensor presents a series of complexity in calibration.
The settings to produce equal images throughout
different cameras vary from one to the next and do
not remain consistent throughout. For the sake of
clarity, we will adopt the zone system and color tem-
perature based on the existing conventions of pho-
tography. The zones indicate the brightness range of
light and dark, and the color temperatures, the light
Figure 8
input/ output
example; Aleatorix.
source that directly affects the color rendition. The
zone 0 indicates pure white and the zone 10 pure
black when printed on white paper. The zone gra-
dients are measured using the neutral photo grey
(80% grey). The color temperatures refer to three
main reference points: 2700K (incandescent light),
3200K (tungsten photo light) and 5500-6500K (sun-
light). The white balance on the video camera in-
put is calibrated according to the three color tem-
perature references. The exception to the color tem-
perature references include fluorescent lights (green-
ish tint) and sodium vapor street lights (orange). For
these two kinds of lights, we will rely on the auto-
matic white balance function of the camera.
CT Mobile App (Work in Progress)
The CT Mobile App will take advantage of the photo-
and video-capturing capabilities of today's mobile
devices. The app is aimed at an architectural and ur-
ban use for color-to-geometry conversion with a mo-
bile device. The CT Mobile App utilize hand-held de-
vices such as smartphones or tablets equipped with
a camera for site analysis and design. Here the app
is intended to work as location-based mobile appli-
cation that translates color disposition of architec-
tural and urban places immediately at the scene into
three-dimensional data-geometry. Such hand-held
mobile devices may also be networked; exchange the
visualized color data; and superimpose the informa-
tion on different locations.
Application & Use
CT can be used to assess and evaluate colorscapes
and identify the perceptual black holes as well as to
evaluate how they can be improved in order to re-
store the chromatic balance and composition of the
DESIGN TOOLS | Applications - Volume 2 - eCAADe 34 | 215
 place. Furthermore CT enables its users to compare Livingston, J 2014, Designing With Light: The Art, Science
the color palettes of different architectural and urban and Practice of Architectural Lighting Design, John Wi-
places. ley & Sons
Opara, E and Cantwell, J 2014, Best Practices for Graphic
Designers, Color Works, Rockport
Complexity behind the Making of Simplicity
The application that drives CT has a simple interface
design to deliver pure forms that are free of the cul-
tural and enculturating baggage human constructs
always carry. But those forms are derived from the
complex combination of hardware and software. A
kit of hardware (e.g. smartphone, webcam, com-
puter, etc.) requires a complex string of existing,
mass-produced parts and assemblies. A piece of soft-
ware is often a behemoth of conflicting algorithms
that have accumulated and become too complex to
fix but too expensive to be abandoned. In essence,
CT embodies such contradictions of science versus
culture and a never-ending pursuit of creating some-
thing simple from a warehouse-full of complex bits
and pieces of technology.

In Closing
The significance and contribution of the overall re-
search can be summarized as the development of an
easy-to-use, portable application. It enables archi-
tectural and landscape professionals to analyze the
color disposition of a place and experiment with the
formal expression by means of color composition.
The value of the research output can be summarized
as a tool for documenting and visualizing the com-
plex urban conditions in terms of color and provid-
ing the effective means to express the combination
of urban colors in 3D-shapes.

REFERENCES
Byrne, A and Hilbert, DR (eds) 1997, Readings on Color:
The science of color, Band 2, MIT Press
von Goethe, JW 1840, Theory of Colors, MIT Press
Itten, J 1970, The Elements of Color, John Wiley & Sons
Itten, J 1997, The Art of Color, John Wiley & Sons
Koenderink, JJ 2010, Color for the Sciences, MIT Press
Kuehni, RG 2013, Color: An Introduction to Practice and
Principles, Wiley
Lee, S 2016, Architecture in the Age of Apparatus-Centric
Culture, Unpublished Manuscript

216 | eCAADe 34 - DESIGN TOOLS | Applications - Volume 2

Automated Service Core Generator in Autodesk Dynamo
Embedded Design Intelligence aiding rapid generation of design options

Subhajit Das1 , Colin Day2 , Michael Dewberry3 ,

Varvara Toulkeridou4 , Anthony Hauck5
1
Autodesk, Georgia Tech 2,3,4,5 Autodesk
1
das@gatech.edu
2,3,4,5
{colin.day|michael.dewberry|varvara.toulkeridou|anthony.hauck}@autodesk.com

Building design entails an intuitive and informative exploration of an architect

involving iterative refinement of design ideas till client objectives, and priorities
are satisfied. Similarly, service cores in a building are designed through the
exploration of multifarious design options each with different performative
metrics regarding accessibility, efficiency, cost, feasibility, etc. As the current
process is labor-intensive, manual & dependent on the expertise of the architect,
the search space leading to the selection of an optimal design alternative is very
limited. This paper describes Service Core Generator (SCG) library in Autodesk
Dynamo enabling automated generation of service core models for varied
building shell geometry types (limited to orthogonal profiles). The tool described
encodes explicit and implicit domain knowledge into the system facilitating
service core models for buildings across varied scale with use type's including
offices, hotels or residential buildings.

Keywords: Design Alternatives, Geometry Analysis, Parametric Modelling,

Design Tools, Design Automation,

INTRODUCTION knowledge from building codes, fire safety regula-

A typical building is a combination of a building shell
(BS) and a service core (SC). BS is a container of inhab-
itable spaces satisfying the client's program require-
ments and project philosophies (See Figure 1(a)). SC
enables vertical circulation and utility delivery for its
users, comprising of elevators, stairwells, restrooms,
mechanical rooms and other project-specific utility
spaces (electrical rooms, laundry, etc.) Many of these
components of SC are designed with prescriptive
tions, structural design guidelines, and other indus-
try standards (Trabucco, 2010). The intent of the re-
search summarized in this paper is to formalize ex-
pert knowledge from experts and design guidelines
in the form of a computational construct reusable for
multiple projects, aiding rapid generation of multi-
ple design alternatives based on set objectives and
constraints described by the architect. Multiple de-
sign options increase the solution space to the design

DESIGN TOOLS | Applications - Volume 2 - eCAADe 34 | 217

 team enabling better decision making early on (Shea,
et al., 2005). It has a substantial impact on over-
all building functionality and cost, as compared to
changes inflicted on the project at the later stages, as
early decisions define the context and sets the tone
for further improvisation (Eastman, et al., 2008). We
intend to remodel the existing workflow, by aiding
architects to brainstorm technical challenges early on
in the project. Automating some aspect of the work-
flow by embedding design intelligence in the com-
puter allows modeling reusable parametric objects
(Kat Park, 2010), which is repetitive across projects
or design options and requires highly technical ex-
pertise. This allows instantiation of SC modules over
and over again for any input BS geometry. This helps
tackle key technical challenges early on, validates
proposed design scheme and reduces project cost in-
curred due to changes inflicted later on in the project.
The idea presented depends on the hypothesis that
some aspects of the explicit (from building codes,
guidelines, etc.) and implicit (from experience and
expertise) knowledge base of an architect can be au-
tomated by embedding the same as a series of algo-
rithms enhancing the architect's capability to analyze
and appraise hundreds of design alternatives based
on set goals and metrics (See Figure 1(b)).
RELATED WORKS
Marcelo et al. presented a top-down approach,
modeling nested assemblies of a service core model
based on custom user defined functions, facilitating
multiple instantiation and reusability of parametric
models. They not only elaborated a successful ser-
vice core modeler but also an analyzer gauging the
efficacy and success of each design option. On the
218 | eCAADe 34 - DESIGN TOOLS | Applications - Volume 2
downside their model uses multiple software plat-
forms involving multiple points of data exchanges,
rendering the prototype less user-friendly and inter-
active (Marcelo Bernal, 2011). In the world of game
development, Fernando Marson et al. explained an
approach to deliver automated floor plan layouts by
compartmentalizing programs of residential apart-
ments into group able zones, then hierarchically as-
signed a spatial location to them. They implemented
squarified treemap algorithm by slicing spatial zones
into sub-compartments (Fernando Marson, 2010).
We researched on service cores focusing on both
mid-rise and high-rise buildings as height increases
complexity and constraints on building design and
service core layouts. Lijie et al. presented an in-depth
study on service core design for skyscrapers. They
outlined that buildings 150m or taller, need to have
elevators arranged in groups, or zones, called eleva-
tor banks. They emphasized the importance of verti-
cal shafts to house M&E pipelines, fire exit staircases
and optimized restroom allocation for better plumb-
ing design (Xie, et al., 2012).
SHELL AND CORE SYSTEM
This paper describes Service Core Generator (SCG), a
library in Autodesk's Dynamo software, providing de-
tailed 3d model of SC for buildings with use type's of-
Figure 1
fices, hotels and residential towers having orthogo-
(a) Building shells
nal BS geometry as an user input to the system. SCG
(BS) and service
can work with low, mid and high-rise building typolo-
cores (SC) two main
gies.
components
defining design
Building shell
proposals at
At the conceptual design stage architects develop
conceptual design
BS representing proposed building's shape, volume,
stage. (b) Service
and location on the site satisfying client's vision and
core generator
objectives. Architects also integrate BS with exact
includes explicit
size, number, and position of required number of
and implicit
SC satisfying local building codes and safety guide-
knowledge base
lines. Architects explore multiple design solutions
realized by design
by developing multiple BS in quick succession, each
automation.
having a schematic SC model. Such quick changes
in BS requires manually updating SC, which is time-
 consuming and tedious. A typical SC includes mul-
tiple sub-components, the design of which requires
a high level of technical expertise and experience
Figure 2 in the field. At the initial stage, design teams of-
Shows the main ten struggle to deliver and translate required level of
components of a technical competence into the model due to lack of
typical service core manpower and enough time to meet project dead-
system. lines. This impedes design and representation of a
working SC model, leading to the selection of design
alternatives which might need substantial changes
later on, to meet building codes requirements.
Service core design principles
Architects at times do not take the design of SC se-
riously at the initial stage of the project not only
because it is hidden inside the BS but also due to
the level of technical expertise needed to design a
working SC model (Trabucco, 2008). However, SC is
an extra cost to the building for developers as they
cannot lease it to tenants. Likewise, the industry
prefers compact and lean SC design (Hennessy & Wal-
lace, 2012). This is ever so more relevant in modern
skyscrapers, where the need to build taller and slen-
der is rising due to the rising price of land. In the ini-
tial design stage, not only size but the placement of
SC is extremely crucial for building performance. It
affects mechanical and electrical systems (M&E) dis-
tribution routes, structural shear wall position, and
structural cost, significantly impacting building ven-
tilation and vertical circulation. Financially, the cost
of an SC for the high rise is around 38 percent of the
total structural cost and around 4-5 percent of the
development cost (Yeang, 2000). Therefore, SCG is
embedded with design intelligence catered to min-
imize space and volume of SC, while satisfying stake-
holder's objectives. For example, minimizing 50 mm
for all walls in a concrete walled core can reduce cost
by 30-40 percentage of the total structural cost. NGR
is the ratio between net saleable or rentable area to
the total gross area of the building, and it is one of
the metrics in SCG to appraise the performance of
a generated design. SCG knows higher NGR means
better SC design. Evidently, more time spent on SC
design makes the design proposal more efficient and
sustainable (Ali & Armstrong, 2008).
Service core components
SC contains an Elevator Bank Unit (EBU), Technical
Room Unit (TRU) and Stair Halls with Restroom Units
(SRU) (See Figure 2).
1. EBU is the principal component in SC, espe-
cially in high rise buildings, which houses banks of
elevators stacked next to each other (Le Messurier,
1986). Fire safety guidelines define the arrangement
of elevators in EBU, thereby determining the overall
size of it. Other components TRU and SRU are placed
around EBU and their volume and size also gets af-
fected by the fire safety regulations.
2. TRU principally houses M&E rooms, duct risers,
mechanical pipelines, shafts transferring pipelines
for electrical and communications cabling. Good SC
design places the M&E programs away from the shear
walls, maintaining the stiffness of SC (Xie, et al., 2012).
3. SRU houses general stairways along with fire
stairways and restroom areas. As staircases cannot
pass through the refuge or mechanical floors, SCG
makes sure that stairs, especially fire exit stairs get
vertical continuity across different zones as BS rises
higher. SCG determines the number of stairways,
(min. two for fire safety), the number of male and fe-
male toilets and their sizes. SRU when placed along
the external periphery of SC, gets the advantage of
external ventilation access.
Service core placement strategies.
Placement of SC in BS significantly affects design per-
formance and efficiency. SCG generates design op-
tions by placing cores in the following subtypes: Cen-
tral, Split, End and Atrium (See Figure 3(a)). For each
DESIGN TOOLS | Applications - Volume 2 - eCAADe 34 | 219
 of these strategies, SCG computes distance from the
floor plate outline to the edge of the core. SCG ac-
cepts this distance to be within a set range, say 6m-
12m for a floor height range from 3.8m to 5m. Other-
wise, SCG places another SC, to maintain the distance
to the floor outline.
Service core configuration strategies
SCG can place cores by employing two modes each
depending on the way elevators are designed to
reach respective floors (See Figure 3(b)). In high-rise
buildings, traffic loads are balanced across the eleva-
tor banks to access various levels and thus all eleva-
tors do not access every level. These strategies are
explained as below:
Strategy 1. This strategy entails all elevators start
from the ground level having a high density of ele-
vators at the lower level compared to upper levels,
arranged in separate banks. The number of banks is
directly proportional to the number of vertical sec-
tions or zones of the building, which is set depend-
ing on the height of the building. Elevators banks
are dropped as we go higher in the BS. The sequence
of dropping of elevator banks depends on the load
balancing of the traffic amongst each bank, meaning
core layout is reconfigured at each level, enlarging
usable floor area (UFA). UFA is the area available to
be leased out to tenants. At lower levels where ex-
press elevators do not stop, elevator lobbies are used
for restrooms or other spaces from TRU. Similarly for
upper levels, where elevator banks are dropped, the
empty space retrieved is either returned to the floor
as leasable space or assigned to either TRU or SRU
(Marcelo Bernal, 2011).
220 | eCAADe 34 - DESIGN TOOLS | Applications - Volume 2
Strategy 2. This strategy also balances traffic load
amongst different elevator banks, but unlike strategy
1, all elevators do not start from ground level. Dif-
ferent vertical sections of the building have different
elevator banks assigned for them. Elevators in each
bank travel from the starting level to the ending level
of the each vertical section. People interchange be-
Figure 3
tween different zones via common floor levels where
(a) Service core
elevators across different zones overlap. These floor
placement
levels are called sky lounge or lobby.
strategies. (b)
Service core and Building Shell dependency Service core
SCG has direct linkage and dependency between the internal
BS and SC, meaning that updating BS, triggers SCG to configuration
recompute SC in real time by calculating total build- strategies
ing footprint, the requirement of a number of ele-
vators, computing elevator banks, key floor levels
where the elevator banks start and stop, etc. This
workflow of changing BS in quick succession to test
ideas in the initial stage of the project is very fre-
quent amongst architects in the conceptual design
stage. The automatic option generation of SC on
change of BS geometry eliminates the manual and
laborious task of re-modeling and re-computing ser-
vice cores allows architects to focus on other aspect
of the project.
Rapid option development with SCG
The whole precedent of having SCG automatically
generate design options is to widen the solution
space of possible service core layouts which the
proposed BS can integrate within, embedded with
nested sub-components as described above. Con-
sequently, SCG validates the feasibility of the design
or BS, by testing if it can fit the necessary SC satisfy-
ing building codes and design guidelines. SCG not
only aids architects to develop better design solu-
tions for the clients but also saves time and budget
of the project, by encouraging and facilitating them
to tackle the technical aspect of the project early on,
which reduces cost of the project. (Perino, et al.,
2015).
Figure 4 nested assemblies satisfying building code require-
Shows state of the ments and safety guidelines capable of being inte-
Dynamo graph grated into the supplied BS is the principal user re-
implementing SCG quirement from SCG.User inputs for SCG include a file
under the hood of path to a '.stl' file as the empty BS (in case the shell
python based is modeled in a different software). SCG is capable
custom nodes. to model a BS in place, based on user inputs like site
position, length, width, height of the bounding box,
type of shell (there are total four types allowed, as
mentioned below), number of floors, habitable floor
height, mechanical floor height, mechanical floor fre-
quency etc. Other inputs consist of elevator lobby
Figure 5 width, area to be satisfied per elevator, elevator width
Shows key and length or area per elevator; maximum allowed
processes in SCG distance to the nearest SC, maximum elevators per
and user inputs for elevator bank, the number of vertical sections of the
each of them. BS, aspect ratio and area for rooms in TRU and SRU,
etc (See Figure 5).

Figure 6
(a) Diagram
showing four types
of shell outline
METHODOLOGY
This section explains the system design of SCG (See
tested (b) Shows
Figure 4) modeled as a collection of Python (Michael
the SCG adding
T. Goodrich, 2013) based 'Custom Nodes' in Au-
service core to the
todesk's Dynamo platform. Here 'custom nodes' and
input shell outline.
'packages' refer to reusable blocks of code which can
be instantiated repetitively by users across multiple Shell Maker (SM):
projects and teams for varied building typology and Shell Maker(SM) employs two building shell gener-
use types. The workflow is that the user inputs an ation methods (See Figure 8(a). One allows users
empty BS initiating SCG, to generate plausible SC de- to supply a predesigned BS to the system as a '.stl'
signs supporting the supplied BS. Following sections file. SCG can read four distinct types of the shell in-
describe the major components of the system, start- cluding 'Rectangle, L, U and Courtyard' schema (See
ing with an explanation of the user requirements and Figure 6). SM reads the .stl model, then generates
inputs to the system, then the Shell Maker generating floor outlines based on input floor heights. Next, the
BS, then the Service Core Maker, an Analyzer high- algorithm subdivides each floor outlines into a col-
lighting analytics and metric score for the generated lection of rectangles to simplify necessary computa-
SC design and finally a Core Geometry Maker, render- tions downstream. Based on input building height
ing SC 3d model in Dynamo. of the BS bounding box, habitable floor height, me-
chanical floor height, and frequency of mechanical
User Requirements and Inputs: floor level, the floor outlines were placed at the as-
Successful delivery of all of the aforementioned com- signed height level.
ponents of an SC, namely, EBU, TRU and SRU as The other process allows users to model the shell

DESIGN TOOLS | Applications - Volume 2 - eCAADe 34 | 221

 in place proceeding from a combination of user pa-
rameters and random variables to investigate differ-
ent kinds of shell geometry feasible within the four
kinds of shell typologies. Essentially a 'shape gener-
ator' custom method was written forming floor out-
lines within the confines of the bounding box, whose
length, width and height are provided to the system
as user inputs. These shapes are of four aforemen-
tioned types. Next SM places floor outlines at the as-
signed height level based on the height list formed
as described. (See Figure 7)
Service Core Maker (SCM)
Based on floor outlines from SM, Service Core Maker
(SCM) node of SCG, enables multiple alternatives of
SC based on different ways of SC placement and in-
ternal configuration of its components (See Figure
8(b)). First, SCM computes the total built area and as-
certains the number of passenger elevators and ser-
vice elevators needed. This is computed based on the
user input of 'area each elevator needs to satisfy' in
SCG.
SM provides rectangular regions in the floor out-
lines. For each rectangular region, SCM divides the
longer dimension with the user input of 'maximum
egress distance allowed to a service core'. The pro-
cess arrives at a quantity of SC to be positioned in
each rectangular region. Subsequently, SCM places
SC in various positions within BS, satisfying the maxi-
mum egress distance allowed constraint. Positioning
possibilities include aligning SC along central, longi-
tudinal or transversal axes, placing cores along exter-
nal walls or combining the previous two options. For
the rectangle shell type, the placement of the service
core is straightforward, but for other shell types, if
222 | eCAADe 34 - DESIGN TOOLS | Applications - Volume 2
SCM finds two SC placed within close vicinity (a pa-
rameter), SCG pushes them apart till there is space
available in the BS (See Figure 10(b)). Next, SCM dis-
tributes the spatial allocation of needed amount of
EBU, TRU and SRU among the placed cores. Here
the process can either distribute the space allocation
evenly or through reference to a specified propor-
tion, i.e. 50: 30: 20 percent, etc. Notably, the spatial
allocation of the cores directly affects the respective
core's size and proportion. SCM disallow cores with
unrealistic proportions.
Post placement of cores, the algorithm space
plans each of them (See Figure 9(a) and 10(a)). For Figure 7
allocating program spaces, the algorithm employs a In place Building
custom squarified tree map process, using a space shell modelled in
subdivision method that keeps in consideration the SCG with varied
aspect ratio of generated subspaces. This approach heights.
employs recursive space division based on a prede-
fined space hierarchy, placing the EBU first in the
SC space container. Based on the number of verti-
cal sections of the BS, number of elevators needed,
and number of elevators allowed per bank (a user
input parameter), SCM makes lists of elevators rep-
resenting elevator banks. Using squarified tree map
technique, SCM allocates elevator banks in EBU and
provides space for elevator lobbies to ensure acces-
sibility to the elevators (See Figure 9(b)). SCM can
demarcate elevators as passenger or service eleva-
tors, each type having a unique aspect ratio or di-
mensions based on user requirements. Each elevator
bank must satisfy the requirements of a specific num-
ber of floors in the assigned vertical section of the BS
and thus SCM provides each elevator bank a starting
floor level and a stopping floor level also referred to
as key floor level (KFL). The concept of KFL is inspired
from the previous research of Marcelo et al. providing
a description of KFL computation to maximize SC effi-
ciency. KFL for each elevator bank is computed based
on the target area to be satisfied by the elevator bank.
Target area is the area of the vertical section of the BS
minus the area consumed by SC per floor. Complex-
ity lies in computation of target area because area of
SC is informed by target area and vice versa.
Figure 8
(a) Process map
showing workflow
of Shell maker. (b)
Process flow
showing workflow
of Service core
maker.

vator bank (Marcelo Bernal, 2011). This is repeated

for all elevator banks such that SCM knows KFL for
Figure 9
every elevator bank in the EBU. Dropping elevator
(a) Floor layout
banks makes the SC compact and efficient as it rises
showing EBU, SRU
through the BS. SCG relocates restrooms and techni-
and TRU
cal rooms spatially in the area of the dropped eleva-
component of the
tor bank, making the core slender and leaner higher
service core. (b)
up (See Figure 11). To implement core configuration
Generated service
strategy 2, SCM computes starting floor level (SFL)
core for a twisted
First target area for every elevator bank is computed for each elevator bank. Basically KFL for previous ele-
tower on a
by dividing the total building area (TBA) by number vator bank becomes the SFL for the current elevator
rectangle shell type
of vertical sections. Then usable floor area (UFA) for bank and KFL for current bank is computed by com-
showing EBU, SRU,
each floor is computed by subtracting area of TRU puting served area starting from SFL.
TRU and Elevator
lobbies. and SRU from total floor area. Every elevator bank
maintains area served as the elevators in it goes up
Figure 10
the BS level by level. The algorithm iterates through
(a) Floor layout of
each floor and if area served by the elevator bank has
the generated
not reached target area, then the elevator bank area
Service core for U
is subtracted from the UFA and the value obtained
Shaped shell type.
is added to the area served variable. As soon as the
(b)Service core
area served value for the elevator bank hits the target
models populated
area, the elevator banks are dropped and the floor
in an input U
level where it occurs, is stored as the KFL for that ele-
Shaped Shell
topology

DESIGN TOOLS | Applications - Volume 2 - eCAADe 34 | 223

 tained spaces for each floor level. The built volume of Figure 11
each component of SC can be obtained by extruding Shows restrooms in
the polygons in Autodesk Dynamo as poly-surfaces. SRU, relocated to
These can be directly exported to Autodesk Revit or vacant spaces from
other 3d software for further use in design develop- dropped elevator
ment. CGM also exports point cloud information of banks as the core
the output geometry as a .csv file for further space rises up.
plan analysis, visualization and computation. CGM
deploys Boolean operations to construct SC corridors
and elevator lobbies for each floor level. Figure 12
Analyzer
IMPLEMENTATION : SCG LIBRARY component of SCG
SCG is delivered as a collection of four custom nodes showing, expected
in Autodesk's Dynamo, using inbuilt Python pro- number of
grammable nodes (See Figure 12). The first two cus- elevators, number
tom nodes build the BS using two processes de- of elevators
scribed above. The next custom node implements provided, target
SCM with analytics information to appraise the gen- area for each
erated design option. The final custom node renders elevator, total
the geometry as a set of polylines and poly surfaces served area by each
representing components of SC. Following the suc- elevator bank,
cess of the python based prototype, work is in the usable area left
process to develop SCG further as Zero Touch Dy- each floor level etc.
Analyzer
namo Nodes (delivered as a C sharp based .dll library).
The analyzer informs the architect about the per-
This helps scale and optimize the library further en-
formance and efficiency of each SC design option.
hancing its functionality and preciseness.
As the architect can change the number of elevator
banks and served area per elevator, the analyzer can Figure 13
inform the designer if the elevators placed in the core Screenshot
are working below or above their nominal capacity. showing the
Further, the analyzer displays the percentage area custom nodes
of SC with respect to the total shell area, indicating placed in Autodesk
the marketable area or NGR, a key success metric for Dynamo Studio.
building design. Additionally, based on the analytics
from the analyzer, architects can recommend certain
core configurations suitable for certain shell typolo-
gies. Other analytics reported include expected ele-
vators per bank, provided elevators per bank, UFA per
floor, area served per elevator and male or female toi- DISCUSSION & RESULTS
let area and configuration (See Figure 11). The objective of the described parametric model in
Dynamo was to enable architects or design teams
Core Geometry Maker (CGM) create reusable assemblies of SC model. One straight
The Core Geometry Maker (CGM) outputs geome- forward use case of SCG is to generate detailed SC
tries by providing labeled polygonal outlines of con-

224 | eCAADe 34 - DESIGN TOOLS | Applications - Volume 2

Figure 14
Screenshots
showing generated
service core design
options from SCG.

models on different shell typologies supplied. Owing tates appraisal of each design option based on space
to the user's input of contextual data and prescrip- usage, efficiency, safety, structural load, cost, etc. It
tive knowledge, SCG is designed to find potential de- encourages reusable, modular and goal-driven de-
sign alternatives which are reasonably distinct and sign practice among architects by automating a sig-
architecturally rational, broadening architects design nificant amount of work which is repetitive, time-
space. consuming and requires significant amount of tech-
nical knowledge. SCG can benefit from Genetic Al-
Figure 15 gorithm based optimizers, making it iteratively learn
(a) Screenshot of from the generated options and find solution sets
Akaba, a web based which approach described goals as closely as possi-
space planning tool ble.
showing a service
core (the grey box)
and the egress
CONCLUSION
Our prototype demonstrates the feasibility to embed
route (green spline). Another use case of SCG was its integration with design intelligence into a parametric model, deliver-
(b) Shows service Akaba, a web-based application developed by Au- ing reusable content. The research proves the no-
core generator todesk's generative design team. Akaba, a collabo- tion that repetitive tasks needing prescribed tech-
integrated with rative 3D space planning tool generates optimal pro- nical expertise are instantiable, allowing architects
Akaba placing gram stacking options to achieve desired goals and to invest time in other aspects of the project en-
detailed service objectives set by the user. SCG was used to deploy hancing team productivity. We realize that the cur-
core models. service cores in the Akaba framework. (See Figure rent system is limited to orthogonal geometry with
15(a) and 15(b)).SCG successfully evaluates and facili-

DESIGN TOOLS | Applications - Volume 2 - eCAADe 34 | 225

 predefined shell typologies with a limited degree of
freedom for the user. But we intend to grow the
system further including angular or curvilinear ty-
pologies. One of the salient features of SCG is the
real time update and feedback of the service core
model, upon any changes made to the building shell,
enabling the designer to validate design proposals
by testing if supplied building shell is fit to inhabit
necessary service cores to satisfy building code re-
quirements. Visual programming tools like Dynamo
was efficient to handle this level of complex com-
putation, allowing the ability to build the system
in one interactive platform for the user, eliminating
the burden of managing multiple tools. SCG fol-
lows a top-down approach beginning with comput-
ing number of cores and their volume to assigning
detailed nested assemblies of relevant components
of the core like EBU, TRU and SRU. SCG delivers an in-
tegrated 3d model of the service core with its compo-
nents nested within, enabling easy instantiation and
quick addition to the supplied building shell making
it scalable and reusable across different project gen-
res. Including SCG in the current workflow at the ini-
tial stage, technical challenges will be dealt early on
by the design team, saving investment of time and
manpower by not having to make multiple changes
to the project downstream. SCG exemplifies the de-
velopment parametric objects driven by encoded ex-
plicit and some extent of implicit knowledge, sat-
isfying technical requirements for any project, re-
sulting in automation of key tasks for an architect
which are repetitive across projects or various stages
in projects. Future development of SCG includes the
provision to input building use type (office, hotel, res-
idential, hospital, etc.) such that SCG can recommend
design options by implementing strategies from best
practices tailored for specific types of buildings. We
also plan to include floor usage e.g. single, shared or
multi-tenant as a user input to have SCG place and
configure service cores with higher efficiency. Cur-
rent system is based on the assumption that the pro-
posed design will have uniform distribution of resi-
dents across floors, but in future we can address non
226 | eCAADe 34 - DESIGN TOOLS | Applications - Volume 2
homogeneous density of people.
REFERENCES
Ali, M and Armstrong, P 2008 'Overview of sustain-
able design factors in high-rise buildings.', CTBUH 8th
World Congress, Dubai
Bernal, M and Eastman, C 2011 'Top-down Approach
to Embed Design Expertise in Parametric Objects
for the Automatic Generation of a Building Service
Core', CAAD Futures, pp. 149-164
Eastman, C, Tiecholz, P, Sacks, R and Liston, K 2008, BIM
for architects and engineers. BIM Handbook : A guide
to building information modeling for owners, man-
agers,designers, engineers and contractors, John Wi-
ley & Sons Inc, Hoboken, New Jersey
Michael T Goodrich, RT 2013, Data Structures and Algo-
rithms in Python, Wiley; 1 edition
Hennessy, M and Wallace, D 2012 'Slender towers on
compact sites', CTBUH 9th World Congress, p. 876
Fernando Marson, SR 2010, 'Automatic Real-Time Gen-
eration of Floor Plans Based on Squarified Treemaps
Algorithm', International Journal of Computer Games
Technology, 2010(7), p. 10
Le Messurier, W 1986, 'Designing High-Rise Buildings.
In Design for High-Intensity Development.', no title
given
Kat Park, NH 2010, 'Parametric Design Process of a Com-
plex Building in Practice Using Programmed Code As
Master Model', International Journal of Architectural
Computing, 08(03), pp. 359-376
Perino, M, Corrado, V, Tian, ZC, Chen, WQ, Tang, P,
Wang, JG and Shi, X 2015 'Building Energy Optimiza-
tion Tools and Their Applicability in Architectural
Conceptual Design Stage', 6th International Building
Physics Conference, IBPC, pp. 2572-2577
Shea, K, Aish, R and Gourtovaia, M 2005, 'Towards inte-
grated performance-driven generative design tools',
Automation in Construction, 14(2), pp. 253-264
Trabucco, D 2008, 'An analysis of the relationship be-
tween service cores and the embodied/running en-
ergy of tall buildings', The Structural Design of Tall and
Special Buildings CTBUH 2nd Annual Special Edition:
Tall Sustainability, 17(5), pp. 941-952
Trabucco, D 2010, 'Historical Evolution of the Service
Core', CTBUH Journal(1)
Xie, L, Ma, L and Xu, Q 2012 'Core design in skyscrapers',
CTBUH 9th World Congress Proceedings, p. 810
Yeang, K 2000, Service cores, Wiley-Academy
DESIGN TOOLS | Explorations
SpaceBook
A Case Study of Social Network Analysis in Adjacency Graphs

Gabriel Wurzer1 , Wolfgang E. Lorenz2

1,2
TU Wien
1,2
{gabriel.wurzer|wolfgang.lorenz}@tuwien.ac.at

In this paper, we have adopted methods from Social Network Analysis in order to
analyze adjacency graphs. Our intent was to uncover as much hidden structures
as possible so as to improve adjacency requirements before they are used further
on during the design process. To that end, we have conducted a case study using
two readily available software packages (Gephi, Pajek), concluding that these
could benefit from being more transparent about the underlying algorithms and
more geared towards the problem domain 'adjacency analysis' when it comes to
data entry and visualization. As a matter of fact, we produced an open-source
prototype called SpaceBook, which customizes computation and visualization in
the aforementioned spirit.

Keywords: Adjacency Graph, Social Network Analysis

INTRODUCTION analyze that graph itself! "Would it be possible to gain

In architecture, it is usual to prescribe functional re-
lationships in linguistic terms, e.g. 'dining should be
close to cooking, but cooking should be far from sleep-
ing'. More formally, each of these statements can
be seen as a pair of functions or spaces having a
prescribed relation (e.g. mandatory, desirable, neu-
tral or negative; interaction or no interaction) which
can be mapped to numbers (mandatory=1, desir-
able=0.5, neutral=0, negative=-1; interaction=1, no
interaction=0). The set of all relationships establishes
an (undirected, weighted) adjacency graph in which
each node stands for a space or function and each
edge for the adjacency relation between the nodes it
connects.
Numerous authors have shown that such graphs
can be used to drive space layout. Surprisingly, one
thing that that has not yet been covered is how to
insight into the prescribed adjacency relations them-
selves, using off-the-shelf methods from Social Network
analysis?" was thus our motivation for conducting
this case study. In more detail,
1. we have looked into several metrics for an-
alyzing an adjacency graph theoretically, all
of which are reported in the 'Background and
Related Work'. After doing that, we
2. went on to try to apply this knowledge using
two of the most prominent software packages
used for SNA (see 'Case Study'), only to fail in
several aspects having to do with practicabil-
ity for daily work (see 'Discussion'). Instead of
dismissing the subject as being 'beyond archi-
tectural practice', we
3. implemented a quick mash-up between Excel
and NetLogo (Wilensky 1999), which is easy to

DESIGN TOOLS | Explorations - Volume 2 - eCAADe 34 | 229

 use and comprehend even when not coming
from SNA (see 'SpaceBook'), available for free
at [3].
Beyond conducting the case study, we would sum-
marize our contribution as follows: Our occupa-
tion shows that there needs to be a mapping be-
tween adjacency relationships (used by architects)
and weights (used by SNA/graph analysis). This map-
ping is not always the same, it needs to be adapted
as stated in the section 'SpaceBook', in order to feed
the used algorithms. Secondly, we find that there
is disagreement between the notions of importance
(more correctly: of centralities) that the SNA commu-
nity itself uses, and thus design computing needs to
have access to the underlying algorithms in order to
justify what is being computed [which, in our case, is
done via the use of an open-source paradigm].
BACKGROUND AND RELATED WORK
Centrality refers to the importance of each node in
a graph when compared to every other node. Cen-
trality measures have been investigated first as part
of graph theory and later in network analysis. In ar-
chitecture and urban planning, the most prominent
example of their use lies in the Space Syntax model
(Hillier and Hanson 1984) which thereby determines
the prominence of each element within a set of in-
tersecting line segments (axial lines) which are trans-
formed into a graph (justified graph; see Wang and
Liao 2007 for a detailed description and implementa-
tion). Space syntax commonly uses betweenness and
closeness centrality (for others, see e.g. by Porta et al.
2006; Law et al. 2012):
Betweenness centrality. The shortest path
from each node to every other node is computed
(Freeman 1977). For each node lying within the
shortest path (i.e. all nodes between start and end
node), we add 1 to the node weight. Nodes with a
high total weight are thus highly prominent from the
standpoint of transitioning between nodes.
Closeness centrality. For every node, we com-
pute all shortest paths to every other node. For each
shortest path, we add the number of nodes in that
230 | eCAADe 34 - DESIGN TOOLS | Explorations - Volume 2
path to the starting node's weight. This measure
gives the "farness" of every node, of which the recip-
rocal value is the "closeness" we are after. Nodes with
high closeness are central hubs from which it is easy
to reach other nodes.
In case that edges signify "roads", we might add
the edge length instead of 1 in the above computa-
tion. Other centrality measures include:
Degree centrality. Each node counts its number
of edges, giving a measure of "connectedness".
Eigenvector centrality. An extension to degree
centrality which assigns a high weight to a node if it
is connected to nodes having a high weight them-
selves. In more detail (Bonacich 1972; Bonacich
2007), (1.) a score of 1 is attributed to every node, (2.)
the scores of each node is recomputed as a weighted
sum of centralities of all nodes its neighborhood, (3.)
the score is normalized by dividing it by the overall
maximum score, then (4.) steps 3 and 4 are repeated
until the score stop changing.
This is a selection of the most common centrality
measures which is far from complete. Further exam-
ples include Random Walk Centrality (Noh and Rieger
2004; see Fidler and Hanna 2015 for an application
in space syntax) and Information Centrality (Stephen-
son and Zelen 1989). The study of networks not solely
from the viewpoint of the graph being analyzed but
with a view towards the position of each individual
actor's position within that network has furthermore
led to the umbrella term Social Network Analysis (SNA;
Borgatti et al. 2009), which furthermore includes the
following measures:
Density. A ratio defined by a node's degree ver-
sus the number of possible edges (these are n(n −
1)/2 in a complete, undirected graph).
Clique analysis. Identifies fully connected sub-
graphs (cliques), i.e. groups of nodes that are highly
dependent of each other.
SNA also seeks to characterize a graph by its ten-
dency to cluster (Holland and Leinhardt 1971; Watts
and Strogatz 1998), to be measured by:
Local clustering coefficient. A ratio defined by
the number of edges between a node's neighbors
 versus the total number of possible edges between
its neighbors. This measure gives an indication of
embeddedness of a node.
Average local clustering coefficient. Measures
the overall clustering of a graph by calculating the
mean of each node's local clustering coefficient.
Global clustering coefficient. Overage cluster-
ing of a graph based on the determination of triplets:
For each node, one examines each pair of neighbors.
If that is connected, we have a closed triplet, if not,
an open triplet. Overall clustering is then simple the
number of closed triplets divided by the total num-
ber of triplets.
A high clustering coefficient is typical for small-
world graphs, where every node can be reached from
every other node by a small amount of hops (e.g. 6
degrees of separation in real-world social networks).
Figure 1
Adjacency matrix of
a design studio
from White 1986,
pp. 130-144.
CASE STUDY
Past work on adjacency graphs in architecture and
urban planning is diverse and does not follow a
common methodology. One exception is the work
by White (1983; 1986), who captures the essence
of functional planning using adjacency graphs as a
driver in great detail. We have taken the example of
an adjacency matrix printed in his book (see Figure
1) as starting point for occupation within this paper,
and show how the different measures apply when
this is transformed into a graph (note that an adja-
cency matrix is equivalent to an adjacency graph - ev-
ery row of the [half-]matrix is transformed into a node
of the same name; every colored entry into an edge
between the rows at whose intersection it lies]).
Gephi
Having established the data, we needed to find a soft-
ware with which architects and urban planners can
immediately determine the sought graph measures.
We firstly selected Gephi (Bastian et al. 2009 [1]; avail-
able from [2]), an open-source tool that is available for
all platforms, for our initial test-drive. We noticed that
there were two downsides:
1. The tool supports the entry of negative edge
weights in principle, however, some algo-
rithms and the visualization component do
not take these into account. Thus, we left
them away for the time being.
2. Furthermore, we also found hierarchical adja-
cencies ("interior" contains all interior spaces
and "exterior" all exterior ones) in the graph
which could not be accounted for. This seems
to be commonplace, as further software pack-
ages (i.e. Pajek, which we have also look into)
would not allow for that either. SNA, it seems,
is not using hierarchies but rather concen-
trates on the 'flat case' - each space or function
is deemed equal. Hierarchical adjacencies
could, however, be constructed recursively, in
the case that there were only relationships be-
tween the nested spaces or functions and the
parent space or function would be a sufficient
DESIGN TOOLS | Explorations - Volume 2 - eCAADe 34 | 231
 Figure 2
SNA measures in
Gephi. (a)
Betweenness, (b)
Closeness, (c)
Eigenvector
centrality, (d)
Cliques.

232 | eCAADe 34 - DESIGN TOOLS | Explorations - Volume 2

 aggregation of all underlying spaces (the term
"sufficient" demands that a space or function
having sub-spaces/functions does in fact, on
its level, represent all the underlying relations;
if there are relations between different hierar-
chical levels, this will in general not hold).
Figure 2 shows the overall results of using Gephi,
which we now want to examine in full detail:
Betweenness (Figure 2a) labels the 'Design Stu-
dio' as most central space. It is no wonder that this
space is singled out, since it is the core function of
the building and behaves as meeting space.
Closeness centrality (Figure 2b) is somewhat
surprising: It labels 'Mechanical', 'Service area' and
'Lounge' as being closest to every other spaces,
which might point to a planning error: Is 'Mechanical'
really required to be a hub space with short distance
to every other space? We think that this should not
necessarily be the case. On the other hand, 'Design
studio' (see middle of graph) seems to be quite dislo-
cated (potentially long distances to other spaces).
Eigenvector centrality (Figure 2c) establishes
an ordering in the relative importance of spaces, not
from the standpoint of transitioning but from the
standpoint of connectedness. In that view, 'Design
Studio' (having the most degree!) would be the most
important space, then come the offices and meet-
ing/conference rooms. This is consistent with the
overall function of the building.
Clique analysis (Figure 2d) uncovers four clus-
ters of spaces, namely entrance areas (top part), core
spaces (center), secretary spaces (lower-right part)
and, interestingly, a kitchen/lounge area (right part).
From our study of Gephi, we concluded that the
mentioned metrics can indeed be used to show (1.)
different purposes of a building [e.g. through cliques
as shown in Figure 2d] and (2.) uncover potential
planning errors regarding adjacency without requir-
ing any extra data [e.g. through analysis of connected
components, not shown in Figure 2]. However, as
the tool failed to include negative adjacencies, we
clearly felt it as necessary to look at a different means
of conducting our study. The choice fell onto Pa-
jek (Batagelj and Mrvar 2003), which is commonly
used because it is a software package that is free [but
closed source!].
Pajek
Pajek itself was harder to handle as opposed to
Gephi, since its visualization abilities are rather lim-
ited (mostly, researchers export graphs into a vector
format and then work on beautify them in a differ-
ent program) and there is need for a statistical/ana-
lytical background from the begin on since its user
interface demands that one knows about the differ-
ence between a graph and the categories of mea-
sures taken on it (i.e. vectors on each node, clusters,
permutations etc.). The centralities reported (see Fig-
ure 3) were partly in agreement with what Gephi had
produced (Figure 3a: betweenness singles out De-
sign Studio; Figure 3b - closeness, does not identify
an overall top node as in the case of Gephi; Figure 3c:
Eigenvector centralities do not exist per se; instead,
there is a "hubs and authorities" model that seems
to use Eigenvector centralities underneath, therefore
sharing the same graph measures that are hard to
perceive due to the lack in visualization capabilities
of Pajek; Figure 3d shows Clique analysis, rather hard
to perceive again [even though correctly labelling
out Design Studio-Principals Office as clique, others
being subordinate; certainly, the image could have
been perfected by a proficient Pajek user using a
different layout algorithm but this is what it really
looks like for someone that is just trying to make
the software 'show something interesting', in terms
of adjacency analysis, without further insights what
the statistical backgrounds were {and this is clearly
refutable, to be sure! It is, nevertheless, the reality in
terms of usability}].
DISCUSSION
There is no clear statement as to "thus, something
needs to be different here" from seeing these two
programs used for SNA. Certainly, Gephi excels at its
visualization capabilities and Pajek does at its statis-
tical options, which - as must be said - are far beyond
DESIGN TOOLS | Explorations - Volume 2 - eCAADe 34 | 233

being comprehensible for an architectural office as it
stands. What truly lies at the heart of the problem
is that both are not accessible, in architectural terms.
For this paper, we tried to make that crossing-over
between both worlds more easily viable, using our
own implementation which we coined "SpaceBook".
SPACEBOOK
SpaceBook is a mash-up between Excel and NetLogo,
utilizing the ease in which it is possible to input adja-
234 | eCAADe 34 - DESIGN TOOLS | Explorations - Volume 2
Figure 3
SNA measures in
Pajek. (a)
Betweenness, (b)
Closeness, (c)
Eigenvector
centrality, (d)
Cliques.
cency relationships in Excel (Figure 4a) and the ana-
lytical capabilities of the NetLogo Network Extension.
On a technical level, we use a self-written extension
called NetLogoExcelBridge which allows NetLogo to
be run from within Excel. Upon being started, NetL-
ogo establishes a channel through which it can read
and write data from Excel, which we then use to get
adjacencies given as half-matrix (Figure 4a) and ana-
lyze/visualize that on a graph level (Figure 4b).
Figure 4
SNA in SpaceBook.
(a) Input of an
adjacencies in Excel,
(b) computation of
graph measures in
NetLogo.

Customizing graph analysis algorithms for
adjacency relationships
Being open-source, NetLogo makes it easy to under-
stand and extend the graph measures as well as the
graph visualizations which are the key to the discus-
sion we want to raise with this paper. As an example,
we could easily re-implement closeness and eigen-
vector centrality so as to take negative relationships
into account. In that process, we noticed that rela-
tionships need to be mapped to edge weights, since
centrality algorithms have different semantics and
thus expect different input (also see Table 1):
• For closeness and betweenness, the re-
lations (negative=-1, desirable=0.5, manda-
tory=1) need to be mapped to positive
weights since the underlying algorithm ex-
pects distances between nodes as input. If
the relation is negative, we map this to a high
number that simulates "infinity" (1000000 in
our case). Desirable relations are assigned
a positive weight (1), mandatory ones are
mapped to 0 (i.e. there is "no cost" when cross-
ing between the linked spaces or functions).
• Eigenvector centrality determines the im-

Table 1
Overview of
examined graph
measures, the
necessary mapping
from relations to
weight as well as
purpose for
adjacency analysis.

DESIGN TOOLS | Explorations - Volume 2 - eCAADe 34 | 235

 portance of each node, based on the weight Figure 5
of the surrounding edges. Therefore, negative Visualizations in
relations must map to zero (unimportant), all SpaceBook. (a)
other should be positive (important, 1 and 2 Circular layout, (b)
in our case). radial layout
• Degree centrality, density, clustering coef- showing
ficient and clique analysis do not use edge hierarchies, (c)
weights but examine the number of edges for cluster layout.
a node. Thus, no mapping needs to be done in Original sceenshots
that case. However, we need to remove edges from within
which have a negative relation in order to be SpaceBook - black
able to compute these measures. background has
been turned into
After performing the input mapping and tailoring the transparent in order
algorithms such that they can use negative weights to conserve some
and disregard isolated nodes without edges, we were ink.
able to repeat our experiments using the Design Stu-
dio data from White (1986; pp. 130-144). Since graph
measures per se are mathematical tools not targeted
at adjacency analysis in architectural terms, we tried
to narrow down their specific purpose in the context
of adjacency analysis (see again Table 1):

• Proximity and circulation are addressed by

closeness and betweenness centrality.
• Importance of each space or function within
the set of all functions can be examined using
eigenvector centrality.
• Planners tend to attribute adjacencies only
when they see them as being "important".
Thus, it makes sense to consider how "well-
specified" each node is in that regard, which
can be done using either degree or density
centrality.
• Spaces or functions tend to form groups. The
two measures that can be used to measure
that tendency are the clustering coefficient
and clique analysis. Clique analysis reports
subsets of the graph that are fully connected
- i.e. spaces or functions which are highly in-
terdependent, while the clustering coefficient
tells us about the connectedness of the neigh-
bors of a node (in SNA terms, this would be:
"Do my friends know each other, too?").

236 | eCAADe 34 - DESIGN TOOLS | Explorations - Volume 2

Customizing the visualization of results
In general SNA software, the layout is applied in a
post-step. However, as this approach specifically
deals with adjacency analysis, we thought it reason-
able to merge the computation and visualization into
one step, since we already know what we are look-
ing for when calling up the computation of a graph
measure. Netlogo already offers some layouting op-
tions (e.g. layout-spring based on Fruchterman and
Reingold 1991), and these can be extended either in
the back-end (Java) or in the NetLogo language it-
self (which is what we did). We used three types of
layouts (1.) circular layout (Figure 5a) in which each
node's importance is also encoded into node size and
color, a hierarchical layout which is radially centered
around the most prominent node (Figure 5b) and a
custom-written clustering layout, which lays out all
supplied clusters supplied by clique analysis in a grid
(Figure 5c). The latter was a bit more complex, since
a node can belong to many clusters. Merging com-
putation and visualization into one step is a good ap-
proach for exploring the data. The results can also be
written in tabular form into Excel for further process-
ing.
CONCLUSION AND OUTLOOK
This paper has presented a case study which uses
graph measures from Social Network Analysis in or-
der to analyze adjacency relations. From examining
two software packages (Gephi, Pajek) on adjacencies
of a design studio taken from literature, two main ob-
servations were made: (1.) There is a slight variation
in the computation of the graph measures used, ow-
ing to different semantics and the fact that negative
relationships (commonplace in adjacency analysis)
might not be taken into account. An added issue in
that context is that the underlying algorithms cannot
always be scrutinized, since such "end-user" pack-
ages may not be open-source. (2.) The same goes for
the graph layout/visualization features, which may
additionally be too general to be useful for the prob-
lem domain at hand.
In order to address both of these problems, we
have conceived SpaceBook, a mash-up between Ex-
cel and Netlogo that is able to analyze and visualize
adjacency in a meaningful way, with the ease of a
click in Excel. We are positive that this approach is
well-integrated into the workflow of early-stage de-
sign, which requires rapid turnaround between data
entry and analysis. As a difference to classical Social
Network Analysis, we want to note that the calcu-
lated measures need to be interpreted qualitatively
rather than quantitatively (e.g. for indication of rela-
tive rank instead of 'exact' centralities for each node),
since data at that stage is still conceptual. It can nev-
ertheless be nswer question such as "are there any
ill-specified spaces with no relationships" and "what
are the most prominent nodes in the adjacency net-
work". As outlook, we want to conduct the same case
study with a much larger adjacency graph such as
in the hospital domain, which seems interesting be-
cause there are also hierarchical relationships (sub-
graphs contained in a node).
ACKNOWLEDGEMENTS
The authors wish to acknowledge the valuable in-
sights into SNA given by FASresearch (especially Har-
ald Katzmair and Christian Gulas). Furthermore, we
wish to acknowledge the important role of the Center
for Computationally Complex Systems at TU Wien in
bringing disciplines together (especially Niki Popper
and Felix Breitenecker), as well as Institute for Manag-
ing Sustainability (especially Markus Hametner, Asya
Dimitrova and André Martinuzzi) for bringing us into
the field of SNA in project terms.
REFERENCES
Bastian, M, Heymann, S and Jacomy, M 2009 'Gephi: an
open source software for exploring and manipulat-
ing networks', International AAAI Conference on We-
blogs and Social Media
Batagelj, V and Mrvar, A 2003, 'Pajek - Analysis and Visual-
ization of Large Networks', in Jünger, M and Mutzel,
P (eds) 2003, Graph Drawing Software, Springer,
Berlin, pp. 77-103
Bonacich, P 1972, 'Techniques for analyzing overlapping
memberships', Sociological Methodology, 4, pp. 176-
DESIGN TOOLS | Explorations - Volume 2 - eCAADe 34 | 237
 185 [3] http://www.iemar.tuwien.ac.at/processviz/spaceb
Bonacich, P 2007, 'Some unique properties of eigenvec- ook
tor centrality', Social Networks, 29, pp. 555-564
Borgatti, S, Mehra, A, Brass, D and Labianca, G 2009, 'Net-
work Analysis in the Social Sciences', Science, 323,
pp. 892-895
Fidler, D and Hanna, S 2015 'Introducing random walk
measures to space syntax', Proceedings of the 10th
Space Syntax Symposium, pp. 145:1-145:9
Freeman, LC 1977, 'A set of measures of centrality based
upon betweenness', Sociometry, 40, pp. 35-41
Fruchterman, TMJ and Reingold, EM 1991, 'Graph Draw-
ing by Force-Directed Placement', Software - Practice
and Experience, 21(11), pp. 1129-1164
Hillier, B and Hanson, J 1984, The Social Logic of Space,
Cambridge University Press
Holland, PW and Leinhardt, S 1971, 'Transitivity in struc-
tural models of small groups', Comparative Group
Studies, 2, pp. 107-124
Law, S, Chiaradia, A and Schwander, C 2012 ''Towards
a multi-modal space syntax analysis. A case study
of the London street and underground network',
Proceedings of the 8th Space Syntax Symposium, pp.
8021:1-8021:20
Noh, JD and Rieger, H 2004, 'Random Walks on Com-
plex Networks', Physical Review Letters, 92(11), pp.
118701:1-118701:4
Porta, S, Crucitti, P and Latora, V 2006, 'The Network Anal-
ysis of Urban Streets: A Primal Approach', Environ-
ment and Planning B, 33(5), pp. 705-725
Stephenson, K and Zelen, M 1989, 'Rethinking centrality:
methods and examples', Social Networks, 11, pp. 1-
37
Wang, W-C and Liao, H-J 2007 'Implementing space syn-
tax in an open-source GIS: GRASS GIS approach',
Proceedings of the 6th Space Syntax Symposium, pp.
100:01-100:14
Watts, DJ and Strogatz, S 1998, 'Collective dynamics
of "small-world" networks', Nature, 393(6684), p.
440–442
White, ET 1983, Site Analysis, Architectural Media
White, ET 1986, Space Adjacency Analysis, Architectural
Media
Wilensky, U 1999, NetLogo (Software), ccl.northwestern.edu/netlogo/,
Center for Connected Learning and Computer-
Based Modeling, Northwestern University,
Evanston, IL
[1] https://gephi.org/publications/gephi-bastian-fe
b09.pdf
[2] https://gephi.org/

238 | eCAADe 34 - DESIGN TOOLS | Explorations - Volume 2

Surveying Design Spaces with Performance Maps
A Multivariate Visualization Method for Parametric Design and
Architectural Design Optimization

Thomas Wortmann1
1
Singapore University of Technology and Design
1
thomas_wortmann@mymail.sutd.edu.sg

This paper presents a novel method to visualize high dimensional parametric

design spaces with applications in computational design space exploration.
Specifically, the visualization method presented here supports the understanding
of design problems in architectural design optimization by allowing designers to
move between a high dimensional design space and a low dimensional
"performance map". This performance map displays the characteristics of the
fitness landscape, develops designers' intuitions about the relationships between
design parameters and performance, allows designers to examine promising
design variants and delineates promising areas for further design exploration.

Keywords: Fitness Landscape, Design Space Exploration, Multivariate

Visualization, Optimization, Star Coordinates

A NEED FOR DESIGN SPACE CARTOGRA-
PHY
The challenge of representing high-dimensional data
in the two or three dimensions that are visually
understandable by humans appears in many disci-
plines, and -with the advent of Big Data in the last
decade- has emerged as a field of study in its own
right. This paper presents a novel method to visualize
high dimensional parametric design spaces and their
performance - such as efficiency in energy and ma-
terial consumption- with applications in computa-
tional design space exploration. Specifically, the visu-
alization method presented here supports the under-
standing of design problems in architectural design
optimization (ADO) by allowing designers to move
back and forth between a high dimensional design
space and a low dimensional "Performance Map".
Architectural design space exploration and architec-
tural design optimization rely on the notion of design
spaces to organize design variants and processes
(Woodbury and Burrow 2006). Since these spaces of-
ten have more than two or three dimensions, they
are difficult to visualize. In parametric design, the
number of dimensions of the design space equals the
number of design parameters. The difficulty of rep-
resenting such abstract spaces motives the need for
multivariate visualizations, which, for example, helps
designers understand the relationships between pa-
rameters and the similarities characterizing group-
ings of design variants.
ADO combines the notion of a parametrically de-
fined design space with one or more numerically ex-
pressed performance criteria. In ADO, performance
criteria typically pertain to reducing a building de-

DESIGN TOOLS | Explorations - Volume 2 - eCAADe 34 | 239

 sign's resource or energy consumption. For exam-
ple, the dome of the Louvre Abu Dhabi -with a struc-
tural span of 165 meters- has been optimized both in
terms of its structural and daylighting performance
(Imbert et al. 2013). When such performance crite-
ria are included in a design space, the literature often
speaks of a fitness landscape. This paper focuses on
representing fitness landscapes with only one perfor-
mance criterion. (A possible extension to multiple cri-
teria is discussed in the conclusion.) In a fitness land-
scape, the height of a coordinate in the fitness land-
scape corresponds to the performance value of a de-
sign variant, and the remaining coordinates specify
the design itself. As an extension of a design space,
such a fitness landscape is often defined by more
than two parameters, which makes it impossible to
represent as a literal landscape.
Recently, ADO has received new understandings
both as a generative design tool that provides start-
ing points for further design exploration (Bradner et
al. 2014) and as a representational tool that aids
the understanding of design problems (Wortmann
et al. 2015). Chen et al. (2015) attempt to group
large numbers of design variants -found with a ge-
netic algorithm- with a clustering method to bet-
ter understand the relationship between design fea-
tures and performance. Their effort is symptomatic of
the need for human-understandable representations
of fitness landscapes. In other words, ADO requires
simple (low dimensional) representations of complex
(high dimensional) relationships. This paper presents
a novel, low-dimensional representation of a high di-
mensional fitness landscape - the Performance Map-
that has at least four applications: It provides insights
into the characteristics of the fitness landscape, and
thus into the nature of the optimization problem
(1), develops designers' intuitions about the relation-
ships between design parameters and performance
(2), allows designers to examine promising design
variants (3) and delineates promising areas for further
-manual or automatic- design exploration (4). This
paper shows how to construct such a performance
map, followed by an example and comparison with
240 | eCAADe 34 - DESIGN TOOLS | Explorations - Volume 2
an existing technique and a more extensive discus-
sion of potential applications. The next section dis-
cusses multivariate visualizations relevant to design
space exploration and ADO and addresses the re-
versibility of these methods, a property that is nec-
essary for turning them into exploratory, interactive
design tools.
MULTIVARIATE VISUALIZATIONS
Multivariate visualizations represent data that have
more than two or three dimensions and thus are
difficult to display on print-outs and screens. They
serve a range of purposes, for example summarizing
data, identifying patterns in or similarities between
data, and displaying correlations between parame-
ters. Hoffman and Grinstein (2002) survey multivari-
ate visualization types. According to them, multivari-
ate visualizations harness humans' perceptual abili-
ties to "look for structure, features, patterns, trends,
anomalies, and relationships in data" (p. 21). De
Oliveira and Levkowitz (2003) describe how "[mul-
tivariate] visual mapping techniques are now be-
ing used both to convey results of [data] mining al-
gorithms in a manner more understandable to end
users and to help them understand how an algorithm
works." They suggest that multivariate visualization
makes data mining methods more transparent and
interactive. This paper proposes that design space
exploration and ADO can benefit from multivariate
visualization in a similar manner, by, instead of pre-
senting only one or a small selection of design vari-
ants, providing the designer with an overview of the
design space or fitness landscape.
Parallel and Radial Visualizations
A straightforward method to represent data with sev-
eral parameters is Parallel Coordinates. Parallel Coor-
dinates introduces a set of parallel -usually vertical-
axes equal to the number of parameters for the data.
To display a datum, one marks the value of each pa-
rameter on the corresponding axis and connects the
resulting points with a ployline (see Figure 1). A vari-
ation of this method uses radial axes, which allow the
Figure 1
Parallel Coordinates
represents each of
the 362 explored
designs of the
example
optimization
problem as a
polyline, with the
best-performing
design variants on
top.
representation of data as closed polylines. Although
such visualizations are easy to construct and under-
stand, they become hard to read when representing
many data, since the polylines tend to overlap. Natu-
rally, visualization methods become harder to under-
stand as number of parameters, i.e. coordinate axes,
increases.
The visualization method applied in this paper -
Star Coordinates- also introduces one coordinate axis
for each parameter, with the axes typically arranged
radially (see Figure 2). In contrast to Parallel Coor-
dinates, however, Star Coordinates displays a datum
not as a closed polyline but as a single point. As
a point-based representation, Star Coordinates can
represent many more data than Parallel Coordinates
without overlaps, and even display a continuous field
to represent the space of the data, although at the
price of making individual parameters less readable.
Note that, although in theory Star Coordinates might
represent several data with an identical point, the
Performance Map circumvents this limitation by pre-
ferring better performing design variants.
RadViz - a method closely related to Star
Coordinates- places data with a spring system, with
each parameter represented by a spring. However,
Rubio-Sanchez et al. (2016) conclude in a compar-
ison of the two methods that RadViz "introduces
non-linear distortions, can encumber outlier detec-
tion, prevents associating the plots with useful linear
mappings, and impedes estimating original data at-
tributes accurately."
An important advantage of parallel and radial vi-
sualizations is that their coordinate axes allow the es-
timation of numerical parameters. This advantage
contrasts with other multivariate visualization meth-
ods such as Self-Organizing Maps (Kohonen 1982),
which cannot directly represent numerical relation-
ships. The ordering of coordinate axes can influence
the insightfulness of parallel and radial representa-
tions. The figures presented here order the coordi-
nate axes according to the order of the parameters in
the underlying optimization problem.
Visualizing Optimization Results
In the optimization field, fitness landscapes are usu-
ally represented with a Matrix of Contour Plots. Such
a matrix consists of n2 − n two-dimensional plots,
with each plot representing the relationship between
a pair of parameters. The plots on the diagonal
of the matrix are left blank or show the relation-
ship of the performance objective to a single vari-
able (see Figure 4). In other words, this visualiza-
tion consists of two-dimensional "sections" or "con-
tours" through the higher dimensional space. This
architecturally-inspired metaphor clarifies an impor-
tant limitation: Although the number of contour
DESIGN TOOLS | Explorations - Volume 2 - eCAADe 34 | 241
 plots increases quadratically with the number of pa-
rameters, these plots display only a very small portion
of the space, since, to draw a single plot for two pa-
rameters, all other parameters must be kept constant.
In terms of the section metaphor, the sections are
drawn through a single "base point" in the higher di-
mensional space. This limitation is usually mitigated
by basing the visualization on the best set of param-
eters found during optimization. Van Wijk and van
Liere (1993) propose an interactive Matrix of Contour
Plots that allows the user to navigate the space by
changing the base point of the visualization.
Nevertheless, contour plots are unable to pro-
vide an overview of the design space as a whole and,
due to their exponentially growing number, become
increasingly difficult to understand even for relatively
small numbers of parameters. Additionally, there
is no guarantee that the selection of design vari-
ants displayed is relevant for understanding the de-
sign space at hand, since the contour plots only cap-
ture interactions between pairs of parameters and
thus miss other, potentially more important interac-
tions. This inability to represent non-pairwise, non-
linear interactions adds to the cognitive difficulty of
integrating information from an exponentially large
number of contour plots.
Pareto Fronts are a type of visualization often em-
ployed in multi-objective optimization. Such fronts
are two- or three dimensional plots of the relation-
ship between two or three performance criteria. Al-
though Pareto Fronts support designers in under-
standing tradeoffs between different criteria (Rad-
ford and Gero 1980), they are of limited use for design
space exploration, since they cannot visualize the re-
lationship between performance criteria and design
parameters.
Reversing Multivariate Visualizations
An important issue for design space exploration is the
reversibility of multivariate visualizations. In order to
support designers with interactive tools that visual-
ize design spaces and fitness landscapes, one should
not only be able to map high dimensional designs
242 | eCAADe 34 - DESIGN TOOLS | Explorations - Volume 2
into a low dimensional "performance map" for visu-
alization, but also to map locations on the "perfor-
mance map" back into the high dimensional design
space. Ideally, such a mapping would bijective. In
other words, every design in the higher dimensional
design space would map uniquely onto a lower di-
mensional representation, and every lower dimen-
sional representation would map onto exactly one
design in the higher dimensional space. Many mul-
tivariate visualization methods don't support such
a "reverse" mapping from low to high dimensions,
and none support a bijective mapping. Matrices of
Contour plots allow reverse mappings, but since this
method only takes (relatively arbitrary) slices of the
higher dimensional space, contour plots are not bi-
jective. Instead, they are injective, with every loca-
tion in the two-dimensional representation mapping
onto a unique design variant, but with many design
variants not representable at all.
This paper proposes a novel extension to Star Co-
ordinates that allows designer to go back and forth
between the design space and its representation -the
Performance Map- using a surjective mapping. A sur-
jective mapping, while falling short of the bijective
ideal, is an improvement over injective mappings in
that all designs from the higher dimensional space
can potentially be represented, but with some design
variants overlapping in the representation. The next
section details this surjective mapping.
METHOD
The method presented in this section extends the
metaphor of the fitness landscape into higher dimen-
sions through a multivariate visualization that rep-
resents performance not as elevation but as color.
A novel extension of Star Coordinates, the method
allows mappings from the two-dimensional Perfor-
mance Map to the high-dimensional design space.
The method achieves this reversible, surjective map-
ping in four steps: Project the evaluated design vari-
ants (i.e. design variants whose performance has
been simulated) onto two dimensions using Star Co-
ordinates (1), compute a Delauney triangulation for
Figure 2
The 362 explored
designs of the
example
optimization
problem
represented as
two-dimensional
points with Star
Coordinates, and
triangulated with
the Delauney
algorithm.
the projected points in the two-dimensional map (2)
and approximate the performance of unexplored de-
sign variants (i.e. design variants with unknown per-
formance values) by interpolating their performance
values based on the corner points of the triangula-
tion's triangles (3) or by estimating their performance
values via a surrogate model (4). Note that, since the
Performance Map is surjective and based on interpo-
lating between evaluated designs, it does not repre-
sent the whole design space, but rather a collection
of design variants that is relatively similar to already
evaluated designs. This similarity, however, increases
the likelihood of performance estimates to be accu-
rate. In other words, it is of little use to represent por-
tions of the design space that have not been previ-
ously explored at all.
Star Coordinates
The method projects evaluated design variants us-
ing Star Coordinates. Although other arrangements
are possible, the coordinate axes are typically spaced
equally around a circle, which is also true for the ex-
ample presented here. Note that, to avoid skew-
ing the visualization due to differences in scale be-
tween parameter values, it is advisable to normalize
the parameters to the same range, i.e. [0, 1]. Star
Coordinates determines the two-dimensional posi-
tion p of a n-dimensional design by multiplying the
value of each parameter with the vector represent-
ing the corresponding coordinate axis, and by then
adding the resulting scaled vectors. In other words,
the two-dimensional embedding p ∈ R2 of the n-
dimensional design xn ∈ Rn is a linear combination
of n coordinate vectors v:
p = x1 v1 + x2 v2 + . . . + xn vn
A potential complication for Star Coordinates is
that two designs with different parameters can map
onto an identical point. For example, all designs
which have the same value for all parameters (all
zero, all one, etc.), map onto the origin of the co-
ordinate axes. Since the two-dimensional mapping
consists of linear combinations of more than two,
pairwise linearly independent vectors in the plane,
a two-dimensional point maps to a theoretically in-
finite number of parameter sets in the higher dimen-
sional space. This mapping to more than one pa-
rameter set implies that Star Coordinates is not bijec-
tive, but surjective: Every parameter set maps onto
a unique point, but this uniqueness is not true in re-
verse. In terms of creating the Performance Map, this
complication is resolved by preferring the better per-
forming design variant when several variants map to
the same point. (In practice variants rarely overlap).
As a result, the Performance Map is slightly biased to-
wards better performing designs.
DESIGN TOOLS | Explorations - Volume 2 - eCAADe 34 | 243
 Triangulating a Surjective Mapping
The Delauney algorithm (de Berg et al. 2008) pro-
vides a quick (O(mlog m)) and unique triangulation
for m points in the plane. The triangulated points are
termed vertices, the connections between the ver-
tices edges, and the triangles formed by the edges
faces. Since, in the Delauney triangulation, every set
of 2D coordinates is located in exactly one face (with
coordinates coinciding with edges or vertices a spe-
cial case), the triangulation provides a unique set of
three vertices (i.e. three mappings of evaluated de-
sign variants) for every set of 2D coordinates in terms
of the face's vertices (see Figure 2). Delauney trian-
gulations have the attractive property that every ver-
tex is connected to its nearest neighbor by an edge.
As a consequence, a point in the Performance Map
will always be close -though not necessarily closest-
to the three vertices of its face, i.e. it will always relate
closely to the three evaluated design variants that
its parameters are interpolated between. An alterna-
tive is to find the three nearest neighbours for every
point, which, however, is more computationally ex-
pensive and less visually intuitive .
Interpolating Performance Values
The method approximates the performance of unex-
plored design variants by linearly interpolating be-
tween either the performance values or the design
parameters of the three evaluated designs associated
with an unexplored variant through the Delauney tri-
angulation. In the latter case, the method obtains an
estimate from a surrogate model using interpolated
design parameters. The linear interpolation repre-
sents each evaluated design as a three-dimensional
point v, where the first two coordinates are given
by Star Coordinates and the third coordinate is the
performance value. The approximated performance
pz of the unexplored variant is given by the inter-
section of a vertical line through the 2D coordinates
of the variant p and the plane of the three three-
dimensional points v of the evaluated designs form-
ing the corresponding face of the triangulation (2).
The plane is defined by calculating its normal n (1):
244 | eCAADe 34 - DESIGN TOOLS | Explorations - Volume 2
n = (v1 − v0 ) × (v2 − v0 ) (1)
−nx · px − ny · py + n · v1
pz = (2)
nz
Estimating Performance Values
A more sophisticated method to approximate the
performance of unexplored design variants is the em-
ployment of a surrogate model. Surrogate mod-
els approximate unexplored fitness landscapes from
evaluated designs by using machine learning tech-
niques. They interactively approximate performance
values and speed up optimization processes by ei-
ther supplementing or completely supplanting time-
intensive simulations of design variants (Koziel and
Leifsson 2013). The surjective mapping presented
here employs the linear interpolation described
above to interpolate not between the performance
values, but between the design parameters of eval-
uated designs. Using the interpolated parameters, a
surrogate model quickly generates a performance es-
timate that is more accurate than the linear interpo-
lation described above. In the example presented in
the next section, a model-based optimization algo-
rithm has iteratively (re-)constructed the surrogate
model during the optimization process to decide
which design variant to evaluate next (Wortmann et
al. 2015).
VISUALIZING AN EXAMPLE PROBLEM
This section presents a Performance Map created
with the method described above. The Perfor-
mance Map visualizes a standard problem from struc-
tural engineering that concerns a small transmission
tower to be optimized in terms of weight while meet-
ing stress and deflection constraints. The tower con-
sists of 25 structural members, which are categorized
into eight beam groups. One of thirty cross sections
can be assigned to each beam group. The problem
thus has eight variables that can take one of thirty
integer values. Consequently, there are 308 design
variants. The visualizations presented here represent
Figure 3
The Performance
Map represents the
design space in
terms of the
estimated
performance values
of unexplored
design variants and
indicates explored
variants. Note the
groupings of
well-performing
variants in the
upper left corner
and near the origin.

the optimization results and surrogate model derived
from a ten-minute run with RBFOpt, a model-based
optimization algorithm. During ten minutes, 362 de-
signs were evaluated through a structural simulation,
with the lightest design weighing 240 kilograms. (For
details on the benchmark problem and optimization
method see (Wortmann and Nannicini 2016).)
Performance Map
In Figures 2 and 3, every evaluated design is repre-
sented as a colored circle, with the position of each
circle corresponding to the design's parameters, and
the circle's fill color corresponding to its weight. Due
to the large weight difference between the best and
the worst evaluated designs (240 kg and 1940 kg)
-which also includes penalties for exceeding stress
and deflection constraints- the colors are applied on
a logarithmic scale. In this way, more detail is visi-
ble for the better, lighter designs. The colors follow a
palette proposed by Niccoli and Lynch (2012) that is
more perceptually adequate than conventional rain-
bow palettes due to its linearly increasing luminance.
The color palette and scale are identical for all all fig-
ures.
Figure 2 indicates only the colored points of the
evaluated designs on the left, with the Delauney tri-
angulation used to interpolated between them on
the right. Figure 3 shows the approximated Perfor-
mance Map derived from the surrogate model. Note
how in Figure 3, the visualization indicates two sepa-

DESIGN TOOLS | Explorations - Volume 2 - eCAADe 34 | 245

 Figure 4
Matrix Contour
Plots representing
the estimated
performance of
unexplored design
variants in terms of
pair-wise
interactions
between
parameters. The
circle represents the
best solution found,
which is the base
point for the
contours. The
diagonal from
lower left to upper
right represents the
eight parameters
individually. The
matrix is
symmetrical across
the diagonal.

rate areas of good, i.e. light designs: One closer to the
center, and one in the upper left corner. These group-
ings indicate two different types of solutions, with
each type displaying similar parameters. The visual-
ization also shows that the central group has been ex-
plored extensively, while the peripheral group con-
sists only of a small number of explored points. The
visualized surrogate model indicates that this area
might contain very high-performing design variants
and thus is promising for future exploration.
Comparison with Matrix of Contour Plots
The contour plots in Figure 4 represent the identical
optimization results and surrogate model as Figure
3. Note that, in this visualization, only the best op-
timization result can be indicated because the con-
tours plots do not contain the remaining evaluated
points. Also note that the two different areas of good
designs visible in Figure 3 are not identifiable in Fig-
ure 4. This comparison between Star Coordinates
and a Matrix of Contour Plots confirms the disadvan-
tages of contour plots identified above, such as the
fragmentation of the design space into a quadratic
number of contours and the inability to identify non-
linearities between more than two parameters. The
next section discusses potential applications for the
Performance Map, with an emphasis on interactive
design space exploration and optimization.

246 | eCAADe 34 - DESIGN TOOLS | Explorations - Volume 2

APPLICATIONS
The proposed visualization method not only sup-
ports the designer's understanding of the fitness
landscape, but also promises to improve the interac-
tivity of ADO tools. Performance maps provide in-
sights into the nature of the optimization problem
(1), represent the predicted performance of unex-
plored designs (2), display designs that are estimated
to perform well (3), and identify promising areas for
further exploration (4). Beyond ADO, applications ex-
tend to higher dimensional design space exploration
more broadly. For example, the mapping can repre-
sent similarities between parametrically defined de-
signs or display trajectories of design decisions in a
parametric design space.
Identifying the Optimization Problem
Performance Maps capture salient characteristics of
the fitness landscape. For instance, a fitness land-
scape is "smooth" when the relationships between
design parameters and the performance criterion
are relatively linear and without discontinues, and
"rugged" when they are not. Another important dis-
tinction is between "convex" fitness landscapes with
only a single peak (i.e. optimum) and "non-convex"
landscapes with additional peaks (i.e. sub-optima).
Such characteristics, which define the difficulty of an
optimization problem and inform the selection of an
appropriate optimization algorithm, can be identi-
fied via a Performance Map. For example, the perfor-
mance map in Figure 3 shows multiple sub-optima
and a rugged fitness landscape where design vari-
ants with similar parameters can have very different
performance values. This ruggedness is expected for
a structural optimization problem, where the light-
est, feasible designs are similar to failing ones.
Relating Parameters and Performance
Both the second and third possibilities imply linking
the performance map to the underlying parametric
model in real-time. Designers can adjust the para-
metric model according to design intentions not cap-
tured by a numerical performance value and simulta-
neously obtain the estimated performance of the ad-
justed design as a location on the Performance Map.
They can also identify similarities between the new
design variant and evaluated ones in terms of their
locations in the two-dimensional parameter space of
the map. If the new design variant is promising, they
can evaluate its performance and add this informa-
tion to the Performance Map and underlying surro-
gate model.
Examining Promising Designs
Conversely, the designer can select a promising loca-
tion on the Performance Map and simultaneously see
its representation in the original parametric model.
In this way, the designer can appreciate the values
for the design parameters that correspond to a loca-
tion in the performance map, and what kind of de-
sign these parameters imply. A simple but impor-
tant part of fostering this appreciation is to facilitate
zooming and panning over the Performance Map.
Guiding Automated Design Exploration
Finally, based on the understating gained from the
Performance Map, the designer can guide further de-
sign exploration by limiting the range of the design
parameters. This limiting can be achieved by select-
ing a promising area on the Performance Map and re-
stricting future exploration to the parameter ranges
in the selected area. By selecting a specific parame-
ter range, a designer can point the optimization al-
gorithm in a direction that supports her design in-
tentions while ensuring a more efficient optimization
process due to the smaller design space to be ex-
plored. In short, Performance Maps allow designers
to interact manually with an otherwise automated
optimization process.
Conclusion and Future Research
Beyond presenting a larger number of examples and
implementing the possibilities described in the pre-
vious section, several directions present themselves
for future research. Assuming that a surrogate model
of the performance dimension is available, one can
statistically analyse the model's performance values
to identify correlations between parameters and the
DESIGN TOOLS | Explorations - Volume 2 - eCAADe 34 | 247
 sensitivity of the parameters in terms of performance.
Such analyses can increase the relevance of visualiza-
tions by suggesting orderings for the coordinate axes
or a reduced selection of parameters to be displayed.
Another direction is the visualization of multiple
performance criteria. Heinrich and Ayres (2016) have
overlaid different color palettes to visualize multiple
performance criteria using contour plots. Their ap-
proach implies creating a performance map for each
criterion individually, which from a mathematical op-
timization perspective is much easier than trying to
address multiple criteria at once. Overlaying the re-
sults from several single criteria in a visual repre-
sentation therefore not only promises greater under-
standing, but also better designs found in a shorter
amount of time. Combining their approach of over-
laying color palettes with the Performance Map pre-
sented here should yield insightful visualizations that
represent the design space in its relationship to mul-
tiple performance criteria.
The Performance Map presented here fits into
a wider context of framing ADO not as a method-
ology to find the "best" performing solution, but to
better understand the relationships between design
parameters and one or more numerical performance
criteria. It demonstrates the importance of multivari-
ate visualization for fostering better understandings
of design spaces, both with and without numerical
performance criteria.
Acknowledgment
I thank Prof. Nannicini (SUTD) for his constant ad-
vice, and especially the inclusion of features in RB-
FOpt (projects.coin-or.org/RBFOpt/wiki) that facili-
tated this research.
REFERENCES
de Berg, M, Cheong, O, van Kreveld, M and Overmars, M
2008, Computational Geometry, Springer
Bradner, E, Iorio, F and Davis, M 2014 'Parameters tell the
design story: ideation and abstraction in design op-
timization', Proceedings of SIMAUD
Chen, KW, Janssen, P and Schlüter, A 2015 'Analysing
Populations of Design Variants Using Clustering and
248 | eCAADe 34 - DESIGN TOOLS | Explorations - Volume 2
Archetypal Analysis', Real Time - Proceedings of the
33rd eCAADe conference, Vienna, AUT, pp. 251-260
Heinrich, MK and Ayres, P 2016 'For Time-Continuous
Optimisation', Living Systems and Micro-Utopias: To-
wards Continuous Designing, Proceedings of the 21st
CAADRIA, Hong Kong, pp. 373-382
Hoffman, P and Grinstein, G 2002, 'Introduction to Data
Visualization', in Fayyad, U, Grinstein, GG and Wierse,
A (eds) 2002, Information Visualization in Data Min-
ing and Knowledge Discovery, Academic Press, Lon-
don, UK, pp. 21-45
Imbert, F, Frost, KS, Fisher, A, Witt, A, Tourre, V and Ko-
ren, B 2013, 'Concurrent Geometric, Structural and
Environmental Design: Louvre Abu Dhabi', in Hes-
selgren, L, Sharma, S, Wallner, J, Baldassini, N, Bom-
pas, P and Raynaud, J (eds) 2013, Advances in Archi-
tectural Geometry 2012, Springer Vienna, pp. 77-90
Kohonen, T 1982, 'Self-organized formation of topolog-
ically correct feature maps', Biological Cybernetics,
43(1), pp. 59-69
Koziel, S and Leifsson, L (eds) 2013, Surrogate-Based Mod-
eling and Optimization, Springer, New York, NY
Niccoli, M and Lynch, S 2012 'A more perceptual color
palette for structure maps', CSEG GeoConvention
2012:Vision
De Oliveira, MCF and Levkowitz, H 2003, 'From visual
data exploration to visual data mining: a survey', IEEE
Transactions on Visualization and Computer Graphics,
9(3), pp. 378-394
Radford, AD and Gero, JS 1980, 'On Optimization in Com-
puter Aided Architectural Design', Building and Envi-
ronment, 15, pp. 73-80
Rubio-Sanchez, M, Raya, L, Diaz, F and Sanchez, A
2016, 'A comparative study between RadViz and Star
Coordinates', IEEE Transactions on Visualization and
Computer Graphics, 22(1), pp. 619-628
van Wijk, JJ and van Liere, R 1993 'HyperSlice: Visualiza-
tion of Scalar Functions of Many Variables', Proceed-
ings of the 4th Conference on Visualization '93, Wash-
ington, DC, USA, pp. 119-125
Woodbury, RF and Burrow, AL 2006, 'Whither design
space?', AIEDAM, 20(2), pp. 63-82
Wortmann, T, Costa, A, Nannicini, G and Schroepfer, T
2015, 'Advantages of Surrogate Models for Architec-
tural Design Opt.', AIEDAM, 30(4), pp. 471-481
Wortmann, T and Nannicini, G 2016 'Black-box opt.
for architectural design', Living Systems and Micro-
Utopias: Towards Continuous Designing, Proceedings
of the 21st CAADRIA, Hong Kong, pp. 177-186
Spatial Optimisations
Merging depthmapX, spatial graph networks and evolutionary design in
Grasshopper

Reinhard Koenig1 , Tasos Varoudis2

1
ETH Zurich and Bauhaus-University Weimar 2 University College London, Bartlett
School of Architecture
1
www.uni-weimar.de/computational-architecture
2
http://www.bartlett.ucl.ac.uk/people/?upi=TVARO74
1
reinhard.koenig@arch.ethz.ch 2 t.varoudis@ucl.ac.uk

In the Space Syntax community, the standard tool for computing all kinds of
spatial graph network measures is depthmapX (Varoudis, 2012). The process of
evaluating many design variants of networks is relatively complicated, since they
need to be drawn in a separated CAD system, exported and imported in
depthmapX via dxf file format. This procedure disables a continuous integration
into a design process. Furthermore, the standalone character of depthmapX
makes it impossible to use its network centrality calculation for optimization
processes. To overcome this limitations, we present in this paper the first steps of
experimenting with a Grasshopper component (Varoudis, 2016) that can access
the functions of depthmapX and integrate them into Grasshopper/Rhino3D. Here
the component is implemented in a way that it can be used directly for an
evolutionary algorithm (EA) implemented in a Python scripting component in
Grasshopper.

Keywords: Space Syntax, Evolutionary Algorithm, Grasshopper, Python,

DepthmapX, Optimization

CONCEPT sponding street network or a new part of it is com-

The guiding principle behind the presented network
optimization is that if we understand the idea of var-
ious centrality measures, we can apply it for a back-
casting approach. The main idea of backcasting is
that we define a certain goal in the beginning and ask
what actions must be taken to get there. In our con-
text this means that we define the requested central-
ity for certain areas or streets in a city and the corre-
puted. To achieve this, we implement an EA to find
automatically the optimal street network for a certain
goal. The EA concept follows an approach presented
by Koenig, Treyer, and Schmitt (2013).
According to Weber et al. (2009), the synthesis
of urban structures consists of a sequence of several
processes: the creation of a road network, the defi-
nition of land use and parcelling, and building place-

DESIGN TOOLS | Explorations - Volume 2 - eCAADe 34 | 249

 ment. We focus in the following on the first step. Sys-
tems have been developed for the procedural cre-
ation of road networks based on L-systems (Parish
and Müller 2001). In particular, the system CityEngine
by ESRI facilitates the three-dimensional, rule-based
modelling of cities and urban structures to the level
of building details (Gool et al. 2006; Weber et al.
2009). In all these examples, the rules for the creation
of an urban design solution have to be specified a pri-
ori in detail. The rules of generative or procedural al-
gorithms are also very technical, abstract and not re-
lated to a planning problem. More importantly, they
are not combined seamlessly with evaluation models
and optimization methods. With these methods we
therefore "have a model that can generate designs
but has no means of establishing whether those de-
signs are any good" (Radford and Gero 1988, 20).
To achieve more advanced and more meaningful
street network synthesis, we therefore need to find a
representation that is able to create realistic network
and can incorporate a performance measures (objec-
tive functions) that can be defined by a designer. This
information should make it possible for the synthesis
system to generate a correspondingly large amount
of possible design solutions.
EVOLUTIONARY OPTIMIZATION
The basic technique we use for synthesizing geome-
try is evolutionary algorithms (EA) due to their flexi-
bility with regard to problem representation as well
as their robustness. This allows us to flexibly experi-
ment with how we technically encode a design prob-
lem in the knowledge that the EA still work in an ac-
ceptable way even if we have a poor technical im-
plementation. EA can be applied on various scales
(Coates and Derix 2008; Derix 2009), for example to
synthesize layout design (Koenig and Knecht 2014),
building volume arrangement (Koenig 2015), urban
district planning (Knecht and Koenig 2012), or net-
work development (Koenig et al. 2013; Schaffranek
and Vasku 2013). The EA may be supplemented by a
number of local search strategies in order to optimize
its calculation speed (Koenig and Schneider 2012).
250 | eCAADe 34 - DESIGN TOOLS | Explorations - Volume 2
During the computer-supported design process,
planners obtain immediate feedback in the form of
at least one design solution that fulfill the formu-
lated design requirements as well as possible. The
presented system for synthesizing designs offers the
possibility to experiment with various restrictions
and objectives for a street network design. This is
an important feature since the definition of a design
problem can be considered as a main step towards its
solution (Rittel and Webber 1973).
TECHNICAL DESCRIPTION
The experimental Grasshopper component for net-
work analysis (Varoudis, 2016) accesses depthmapX
via a remote network call. Therefore, you need to run
depthmapX (currently depthmapXnet 0.30; Varoudis,
2012) in the background. The component is de-
signed by (omitted) for being used directly as an iter-
able (realtime parametric solver) network analyser
and adapted here as fitness function in the EA for the
perpose of the workshop (Varoudis, 2015; together
with Koenig, R., Koutsolampros, P. and Papalexopou-
los, V.). In principle can be used independently of the
optimization if we send input data to it and receive
the results from it by using a custom Python script
as shown in Figure 1 on the left. On the right side of
Figure 1 the results of an exemplary network analysis
are shown. The advantage of using the depthmapX
Grasshopper component is that if we manipulate the
street network in Rhino and the centrality measures
are immediately updated. So far it is possible to use
an axial map representation and steps as distance
measures. This means it is not useful to submit street
segments and distances are computed in steps to ar-
rive from one place to the other, whereas to pass an
axial line is counted as one step. Other options may
follow in next versions of the component.
The code in the Python component (Figure 1) to
read the results from the depthmapX Grasshopper
component (dX_acial_Fnc in Figure 1) is the follow-
ing:
import rhinoscriptsyntax as rs
lines = []
Figure 1
Left: The new
network analysis
component
(dX_acial_Fnc)
connecting to
depthmapX and the
Python script
component to
control it. Right:
Result of the
network analysis
showing global
Integration for an
area in the city of
London. Red
print(len(mylines)) The parameter rad_list allows to define the radiuses
represents high and
for i in range(len(mylines)): that shall be analyzed. The results from depthmapX
green low values. lines.append(mylines[i]) can be acced in the for loop by this line:

rad_list = "R, -1"#, 10" if str2d [0][i] == "Choice [Line Length

do_choice = 1 ,→ Wgt ]":
results = fkt(lines , rad_list ,
,→ do_choice) whereas you can define the type of result that shall be
spRS = results.splitlines () used by the corresponding string. Two other options
str2d = [] are out-commented.
for i in range(len(spRS)): The depthmapX Grasshopper component re-
str2d.append(spRS[i]. split(',')) turns a function, which is used to communicate with
in from the Python script:
place = -1
results = fkt(lines , rad_list ,
for i in range(len(str2d [0])): ,→ do_choice)
#if str2d [0][i] == "Integration [
,→ HH] R10": This function is used in the loop of the EA. This im-
#if str2d [0][i] == "Integration [ plementation has the advantage that we don't need
,→ HH]": any additional looping components in Grasshopper
if str2d [0][i] == "Choice [Line which are usually difficult to handle and would make
,→ Length Wgt ]": the optimization process slower.
place = i
How the EA is implemented is described in the
print(place)
following example. In Figure 2 on the left the ini-
out = [] tial street network is shown. The yellow marked line
for i in range(len(str2d)): serve as reference line for computing the fitness val-
out.append(str2d[i][ place ]) ues for our design variants. The objective is to max-
imize the integration value of the reference line by
strout = ",". join(out) placing 6 additional lines. The small red points are
results = strout the pre-defined references for start and end points

DESIGN TOOLS | Explorations - Volume 2 - eCAADe 34 | 251

 of the new lines. Figure 2 in the middle illustrates the ,→ fitness calculation
Python optimization component. Its inputs are the if refFitLine is not None:
reference points (refPoints), the existing network (en- idxRefLine = network.index
vironm), the reference line (refFitLine), the number of ,→ (refFitLine)
childFitnessValues .append(
lines that shall be added (linesToAdd), the number of
,→ tempFitness[idxRefLine
iterations of the EA (generations), and the population
,→ ])
size of the EA (populatSize). The outputs of the com- else:
ponent are the resulting fitness values for the gener- #childFitnessValues .append
ated variants, and the best solution that was found. ,→ (sum(tempFitness)) #
On the right side of Figure 2 an exemplary result is ,→ we use the sum of the
shown that was found with the settings of Figure 2, ,→ fitness value
middle. childFitnessValues .append(
The function from the depthmapX component, ,→ max(tempFitness)) #
is used for the evaluation operator of the EA. Since ,→ we use maximization!
the code for the whole EA is too much for this article, The access to the depthmapX component works very
we only show the evaluate method: similar than in our first example. The centrality mea-
def Evaluate (): sures are used to define the fitness value of a variant
rad_list = "R, -1, 10" in the last line:
do_choice = 1
childFitnessValues .append(max(
del childFitnessValues [:]
,→ tempFitness))
for i in range(len(
,→ childChromosomes)):
Of cause it's very easy now to change this function to
network = copy.copy(
,→ childChromosomes[i])
use others than the max value of the Integration [HH]
if environm is not None: # measure.
,→ test if there is a
,→ specified environment CASE STUDY
for k in range(len( The Grasshopper component as well as the Python
,→ environm)): EA were tested in the context of a workshop at the
network.append(
(excluded for review) conference. The workshop was
,→ environm[k])
for both beginners and intermediate users. The idea
stringResults = fitnessFkt( of the workshop was to explore spatial design vari-
,→ network , rad_list , ants through computational and parametric design
,→ do_choice) thinking. We used Rhino5, Grasshopper and the
if isinstance(stringResults , Python scripting language to introduce concepts be-
,→ basestring): hind depthmapX, Space Syntax and parametrism. By
tempFitness = ReadResults( the new Grasshopper component we first demon-
,→ stringResults , " strated the advantages of real-time analysis within
,→ Integration [HH]")
Rhino as well as the new depthmapX[net] that is the
else:
tempFitness =
brain behind the new component.
,→ stringResults In a next step the participants were introduced
to the basics of EA. By providing the source code of
#index of the reference line , our Python EA implementation, we showed them the
,→ which shall be used for most important parts of the code to quickly adapt it

252 | eCAADe 34 - DESIGN TOOLS | Explorations - Volume 2

Figure 2
Left: Initial street
network with the
area to be filled, the
red reference
points, and the
yellow reference
line for the fitness
function. Red
represents high and
green low values.
Middle: The Python
optimization
to their own ideas. For this purpose, a complete un- tom gens, selectors and fitness functions for the rep-
component with
derstanding of the EA was not necessary. Already by resentation of individual design problems.
input parameters
basic changes in the code many possibilities to de-
and the depthmapX
fine own customized fitness functions result. For ex- REFERENCES
components that is
ample, the participants experimented with minimize Coates, P and Derix, C 2008 'Smart Solutions for Spa-
used to compute
or maximize certain centrality measures and with a tial Planning: A Design Support System for Urban
the fitness function.
combination of various objectives in a summed fit- Generative Design', Architecture 'in computro': 26th
Right: Result of the eCAADe, Antwerpen, Belgium, pp. 231-238
ness function.
optimization Derix, C 2009, 'In-Between Architecture Computation In-
process that added Between Architecture Computation', International
6 lines to maximize CONCLUSIONS Journal of Architectural Computing (IJAC), 07(04), pp.
the integration We introduced the first attempt of a Grasshopper 565-586
value of the network analysis component (Varoudis, 2016), which Gool, P, Müller, P, Wonka, P, Haegler, S, Ulmer, A and Van
allows to compute spatial measures from the well Gool, L 2006 'Procedural Modeling of Buildings', ACM
reference line. SIGGRAPH, Boston, pp. 614-623
established analysis tool, depthmapX. This enables
Knecht, K and Koenig, R 2012 'Automatische Grund-
real-time analysis within Rhino and the usage of the stücksumlegung mithilfe von Unterteilungsalgo-
analysis for, among others, an EA as presented here. rithmen und typenbasierte Generierung von Stadt-
In our tests the implemented components run sta- strukturen', Arbeitspapiere Informatik in der Architek-
ble and can be adapted to various planning scenar- tur, Weimar, pp. 1-21
ios. The parametric approach and the integration Koenig, R 2015, 'Urban Design Synthesis for Building
Layouts Urban Design Synthesis for Building Lay-
into Grasshopper allows the usage of the results of
outs based on Evolutionary Many-Criteria Optimiza-
the network analysis or the EA optimization process tion', International Journal of Architectural Comput-
for other even more complex scenarios. Current ob- ing, 13(3+4), pp. 257-270
jectives of our work focus on research of synthesis of Koenig, R and Knecht, K 2014, 'Comparing two evolu-
urban structures from the street network and the re- tionary algorithm based methods for layout genera-
sulting street blocks, parcels and buildings using this tion: Dense packing versus subdivision', Artificial In-
telligence for Engineering Design, Analysis and Manu-
set of new tools.
facturing, 28(03), pp. 285-299
In a next step we plan to extend the depthmapX Koenig, R and Schneider, S 2012, 'Hierarchical structur-
functions that are accessible via the Grasshopper ing of layout problems in an interactive evolutionary
component. Furthermore, we intend to develop the layout system', AIEDAM: Artificial Intelligence for En-
EA Python optimization component to provide it as a gineering Design, Analysis and Manufacturing, 26(2),
ready-made component which can be feed with cus- pp. 129-142

DESIGN TOOLS | Explorations - Volume 2 - eCAADe 34 | 253

 Koenig, R, Treyer, L and Schmitt, G 2013 'Graphical
smalltalk with my optimization system for urban
planning tasks', Proceedings of the 31st eCAADe Con-
ference – Volume 2, Delft, Netherlands, pp. 195-203
Parish, Y and Müller, P 2001 'Procedural Modeling of
Cities', SIGGRAPH, Los Angeles, CA, pp. 301-308
Radford, A and Gero, JS 1988, Design by optimization in
architecture, building and construction, Van Nostrand
Reinhold, New York and Wokingham
Rittel, H and Webber, M 1973, 'Dilemmas in a General
Theory of Planning', Policy Sciences, 4, pp. 155-169
Schaffranek, R and Vasku, M 2013 'Space Synatax for
Generative Design: On the application of a new tool',
Ninth International Space Syntax Symposium, Seoul,
p. 12
Turner, A 2004, Depthmap 4: a researcher's handbook,
Bartlett School of Graduate Studies, University Col-
lege London, London
Varoudis, T 2012 'depthmapX: Multi-Platform Spa-
tial Network Analysis Software', Space Syntax Lab,
Bartlett, UCL
Varoudis, T 2015 'Introduction to architectural and spa-
tial computation', 10th Space Syntax Symposium, Uni-
versity College London, UK
Varoudis, T 2016 'Grasshopper remote for analyser for
depthmapX version 0.11, Software platoform.', Space
Syntax Lab, Bartlett, UCL
Weber, B, Müller, P, Wonka, P and Gross, M 2009 'Interac-
tive geometric simulation of 4D cities', Eurographics,
pp. 481-492

254 | eCAADe 34 - DESIGN TOOLS | Explorations - Volume 2

Structural Optimisation Methods as a New Toolset for
Architects
Sebastian Białkowski1
1
Lodz University of Technology
1
sebastian.bialkowski@p.lodz.pl

The paper focuses on possibilities of already known engineering procedures such

as Finite Element Method or Topology Optimisation for effective implementation
in architectural design process. The existing attempts of complex engineering
algorithms implementation, as a form finding approach will be discussed. The
review of architectural approaches utilising engineering methods will be
supplemented by the author's own solution for that particular problem. By
intersecting architectural form evaluation with engineering analysis
complemented by optimisation algorithms, the new quality of contemporary
architecture design process may appears.

Keywords: topology optimization, design support tools, complex geometries,

finite element method, CUDA

INTRODUCTION this studies the emphasis is put on structural or ma-

A continuous development of Computer Aided De- terial efficiency optimisation.
sign tools complement by computational design
methods highly influence architectural design devel- BACKGROUND
opment. Architects are looking for new approaches
Methods
which may support their form finding process with
Topology Optimisation of structures can be divided
new and unique solutions for particular problem.
on two main subfields: Layout Optimization (LO) and
As an extraordinary tool in the architectural design
General Shape Optimisation (GSO) (Kutyłowski 2004).
practice, Topology Optimisation methods might be
The first approach concerns grid-like or trust-like
pointed. That algorithm is very widely used in the
structures having very low volume fraction. Opposed
industrial product design such as aerospace and au-
to LO method, GSO is concerning higher volume frac-
tomotive where mechanical parts efficiency and its
tions, optimising the topology and shape of material
material usage is crucial. Also civil engineering is not
continuum. The three subtypes of GSO method can
lagging in this field (Guest and Moen 2010). Many sci-
be pointed, as shown on Figure 1, based on material
entific discourse and researches have been made, im-
properties applied as a ground structure: Isotropic
plementing Topology Optimisation for various pur-
Solid-Empty (ISE), Anisotropic Solid-Empty (ASE) and
poses. In majority of them, only separate structural
Isotropic Solid-Empty Porous (ISEP). The force distri-
elements are taken into account such as beam, col-
bution in the material is the main criteria for that
umn or truss layout (Amir and Bogomolny 2011). In
classification. GSO can be also categorized based on

DESIGN TOOLS | Explorations - Volume 2 - eCAADe 34 | 255


optimisation algorithm implemented on. Two most
popular approaches are Solid Isotropic Microstruc-
ture with Penalization (SIMP) developed in 1989 and
Evolutionary Structure Optimisation (ESO) firstly ap-
plied in 1992. The principle of SIMP methodology
is material density distribution by applying heuris-
tic optimization algorithm such as Optimal Criteria or
Methods of Moving Asymptotes (Bendsøe and Sig-
mund 2004). Indecently which algorithm will be cho-
sen, the optimisation process is based on design sen-
sitivities. The ESO approach introduced firstly by Xie
and Steven in 1992 were detail described in paper
from 1997. Through time, method derived into dif-
ferent approaches such as:
• ESO - base method, which only remove mate-
rial based on stress criteria. Algorithm works
until maximum stress is achieved.
• AESO - Additive ESO - procedure starts with
small amount of material, and adds it near the
points with high stress until maximum stress
256 | eCAADe 34 - DESIGN TOOLS | Explorations - Volume 2
Figure 1
Typology of
Topology
Optimisation.
Diagram based on
Kutyłowski, 2004.
criteria is met.
• BESO - Bi-directional ESO - this method com-
bine principles of basic ESO and AESO ap-
proaches. Algorithm is able to add or remove
material in every point in space during any
step of optimisation. This methodology de-
crease the possibility of hitting local optimum
which can led to wrong solutions.
Both algorithms are based on Finite Element toolset.
Topology Optimisation in architectural de-
sign
Topology Optimisation as an engineering tool is
rarely applied in the architectural design process.
Commonly it is caused by a complex and time tak-
ing process to achieve results which would satisfy a
designer. However, that obstacle should not discour-
age some architects from experimenting with these
tools in the field of architecture. The first practical
application of structural optimisation methods was
 made in 2004 in the project named Akutagawa River by the application of 3D Extended method of ESO.
Side in Takatsuki in Japan. Two adjacent walls in the Despite its iconic and unique shape (see Figure 4),
four story office building nearby Takatsuki Japanese due to budgetary constraints, the project has never
Railway station were optimised by two dimensional been build. Nevertheless, that approach has been
implementation of ESO method (see Figure 2). Dur- the first comprehensive application of topology opti-
ing the whole process multiple types of load were misation algorithms towards architectural form find-
taken into account. Except from typical dead and live ing process.
load, dynamic earthquake loads were also applied
Figure 2 (Ohmori 2010).
Iterations of ESO Related approach, but in a bigger scale was pre-
algorithm for sented by Stromberg (Stromberg et al. 2011) towards
Akutagawa Project, a high raised building. This theoretical application
Japan. of Simple Isotropic Microstructure with Penalisation
(SIMP) method developed by Bendsøe and Sigmund
(Bendsøe and Sigmund 2004) is exploring design for
constructability of high-raised building. As authors
pointed, "The optimality comes from the idea of un-
derstanding how the forces are ''moving'' through
the structure to the foundation and embrace this flow
with the structural members", the research are con-
ducted on external skin of designed building with
compliance minimization as an objective function
(see Figure 3).

Arata Izosaki - the pioneer of Structural Op-

timization methods applied in Architecture
All previous approaches implementing Structural
Optimisation methods were focused on structural as-
pects of this tool diminishing their spatial appear-
Figure 3 ance. One of the first attempts to use shapes, ob-
Application of tained from Topology Optimisation algorithms as an
Topology architectural form, was proposed by Japanese archi-
Optimisation tect Arata Isozaki with collaboration of a structural
algorithm in engineer Matsuro Sasaki. They designed a multifunc-
skyscraper design. tional object for culture and tourism, called Illa de
Blanes at the seaside of Blanes (Costa Brava, Spain)
developed in the years 1998-2002 (Januszkiewicz
2013).
This enormous complex, covering 75.000 square
meters of mixed functions is supported by a large
structure generated by ESO algorithm. In combina-
tion with tree like, organic shaped columns, build-
ing was equipped with double curved roof created

DESIGN TOOLS | Explorations - Volume 2 - eCAADe 34 | 257

 One year after Illa de Blanes project, Arata Izosaki to- Figure 4
gether with Matsuro Sasaki took part in an interna- Project of a
tional architectural competition for the largest train multispace cultural
station in Italy - Santa Maria Novella in Florence. and tourism
Japanese designers proposed a huge structure gen- structure called Illa
erated by Topology Optimisation algorithms (see Fig- de Blanes.
ure 5), namely 3D Extended Evolutionary Structure
Optimisation(EESO) (Januszkiewicz 2013). A 400 me-
ters long and 42 meters wide flat roof, designed
as a land strip for lightweight aircrafts, elevated 20 Figure 5
meters above the ground, was supported by mas- Process of
sive columns in a few points. On the ground, those generating form by
columns had only four main roots to grow from, EESO algorithm.
which reminiscent of trees branches (see Figure 6).
Unfortunately, the project won the second prize, and
it has never been build.
The vison of application of engineering tools,
such as Topology Optimisation in architectural de-
sign, finally came true in 2008. The Qatar National
Convention Centre (QNCC) in Doha became a chance
for Arata Isosaki to implement his innovative vision
Figure 6
of architecture driven by engineering computational
Design of Santa
methods. In cooperation with Buro Happold he de-
Maria Novella Train
signed a tree-like structure(see Figure 7), which sup-
Station in Florence.
ported 250 meter long and 110 m wide lobby roof
(see Figure 8). As in a previous approach, Izosaki im-
plemented iterative EESO algorithms. Until today, it
is the biggest structure created by generative tools Figure 7
based on Topology Optimisation (Zwierzycki 2013). 3D model of one
branch of structure
Optimisation outcome and manufacturing in Doha Convention
process. Centre.
A result of Topology Optimisation is typically rep- Figure 8
resented as a spatial boundary of material distribu- Convention Center
tion. Three dimensional versions of algorithms re- in Qatar.
turn complex geometries, highly difficult to manu-
facture by a common production process. Especially,
when the result of optimisation is not intended for
mass production, but is unique for particular prob-
lem. Some solutions for solving that problem is de-
scribed in researches done by Arup. They imple-
mented an additive manufacturing process named
SLS for producing irregularly shaped structural ele-

258 | eCAADe 34 - DESIGN TOOLS | Explorations - Volume 2

 ments made from steel for tensegrity structure (see
Figure 9). Generally shaped element (by authors
named AM Node) were optimised by SIMP method
and as authors claimed "The topology optimization
resulted in an organic form with less material while
the original functions as cable connectors are still en-
sured." (Block et al. 2015). Compared to traditional
design, presented approach not only provide optimi-
sation of material, but also caused increase of node
stiffness. That proves usefulness of Structural Optimi-
sation Methods such as Topology Optimisation, but
cost of production based on 3D printing technology
is still too high.
Figure 9
Optimised
structural element
prototyped by
utilizing SLS
3D-printing
technology.
NEW TOOLSET
tOpos - new Topology Optimisation ap-
proach towards architecture
Figure 10 Existing architectural adaptation of Topology Optimi-
General formulation sation are the result of deep research and time con-
of minimum suming experiments often aided by specialists in civil
compliance engineering. Furthermore, a majority of engineering
approach. software which are implementing Topology Optimi-
sation algorithm, needs high abilities and knowledge
to operate on. Their explicit user interface and, more-
over, plethora of options and decisions which user
has to make, put it as a highly specified software ded-
icated to the limited range of users. That limitation
decrease a possibility of usage Topology Optimisa-
tion by architects for enriching their design process.
All presented arguments affirm the author about the
need for developing a new tool for designers. Topol-
ogy Optimization as a material distribution method
based on numerical approach can be successfully en-
hanced by contemporary computing tools. The aim
of the author research is to create a form finding
real-time tool for architects based on the engineer-
ing method of Topology Optimization. The project,
named "tOpos", unlike the complicated and expen-
sive commercial tools for engineers such as Altari or
Abacus software's is planned to be easy to use for ar-
chitects, fast and efficient. None special knowledge
about continuum mechanics will be needed to use it.
By developing this tool, it is highly intended to give
the opportunity to variety of architects and design-
ers to use the exceedingly complex and compound
process to improve their designs. What is more, for a
fast and effective design process, the designed soft-
ware needs to provide an immediate feedback to a
designer.
Structure of the algorithm
To obtain high performance optimisation process it
is essential to understand basis of this algorithm.
Topology Optimization is a mathematical approach
that optimises material layout within a given design
space, for a given set of loads and boundary con-
ditions, such that the resulting layout meets a pre-
scribed set of performance targets. The mathemat-
ical formulation of the optimization problem can be
state as follows:
where c is the compliance of the design, U and F are
the global displacement and force vectors, respec-
tively, K is the global stiffness matrix, ue is the ele-
ment displacement vector, k0 is the element stiffness
matrix for an element with unit Young's modulus, x is
the vector of design variables (i.e. the element densi-
DESIGN TOOLS | Explorations - Volume 2 - eCAADe 34 | 259
 ties), N is the number of elements used to discretize Figure 11
the design domain, V (x) and V0 are the material vol- Structure of
ume and design domain volume, respectively, and f Topology
is the prescribed volume fraction. To solve equitation Optimisation
KU = F finite element methods (FEM) (Zienkiewicz algorithm.
and Taylor 2005), referred to as finite element analysis
(FEA) by some authors, has to implemented.
From the algorithmic point of view, the whole
optimisation process can be divide into 3 phases pre-
sented on Figure 11. Prepocess is the first stage,
where Finite Element model is build and initial de-
sign variables are defined. User need to define de-
sign domain, all loads and supports conditions and
target volume fraction. The next step is named Pro-
cessing, where the main optimisation process is per-
formed. This step is iteration loop where specified
subroutines are executed as follow:
• Based on current distribution of densities
compute by FEA subroutine the design result-
ing displacement.
• Compute the compliance and sensitivities of
the design.
• Check the improvement over last design. If
only marginal improve achieved, terminate
the loop. Else continue.
• Update the densities by chosen algorithm
(Optimality Criteria or MMA)
• Repeat the iteration loop.
The last part of the algorithm is Post processing. The
outcome densities can be represent as a 3D spatial
model of optimised material boundary. As shown
in Figure 11, if proper and efficient FEA subroutine
is available, only few functions need to be add, to
perform Topology Optimisation. The performance of
analysing subroutines is fundamental for whole opti-
misation process.

260 | eCAADe 34 - DESIGN TOOLS | Explorations - Volume 2

Figure 12 Code implementation
Base ground Based on scientific researches (Schmidt and Schulz
structure model – 2011) authors came up with an idea of developing
Stanford Bunny. and implementing own Topology Optimization algo-
rithm enhanced with the GPU acceleration. Contrary
to current scientific implementation of Topology Op-
timisation presented by Schmidt and Schulz(2011)
or Liu and Tovar (2014), tOpos has possibility to de-
fine irregularly shaped ground structure. The mod-
ule responsible for creation of Finite Element model,
Figure 13 creates elements and nodes only inside user defined
FE models with boundary represents as a mesh (look Figure 12). The
different density. basic environment for tOpos is Rhinoceros3D soft-
Divisions count is ware with Rhinocommon libraries. McNeel software
performed on X axis is used as 3D engine for data display and what is
as follow: a) 20; more, as a modelling environment necessary for the
b)40; c) 60; d) 80. algorithm to create input data. tOpos also utilises
Rhinocommon library to process meshes and vec-
tors.
The core of tOpos is based on the host code (per-
formed on CPU) implemented on DotNet C# with ac-
celeration kernels based on CUDA API (Sanders et al.
2012), the extension of C language. To execute real
time Finite Element Method subroutine, an algorithm
needs to be parallelized on many threads. For easy
mathematical operations, which need to be repeated
many times, GPU demonstrates higher performance
by speeding computation dozen of times. GPGPU
Figure 14 (General Purpose computing on Graphics Processing
a) Output Units) allows to exploit parallel architecture pipeline
displacement to bring a new approach to a real time applica-
vector filed. b) tion. On the other hand, GPU has major disadvan-
Boundary tage which need to be concern during algorithm de-
conditions applied sign. Memory capacity embedded on Graphic Cards
for benchmark. is smaller than amount of CPU memory installed in
modern computer. This limitation has a big impact
on data structure and it management.
Based on Martínez-Frutos and Herrero-Pérez
(2015) researches as a main FEM solver iterative Con-
jugate Gradient (CG) algorithm has been chosen. Op-
posed to direct solvers such a Cholesky or QR factor-
ization, iterative solvers are highly adaptable for spe-
cific needs such a memory consumption or time re-

DESIGN TOOLS | Explorations - Volume 2 - eCAADe 34 | 261

 served for solving particular problem. During tOpos
developing process, two CG solvers were design con-
cerning memory usage on certain graphic card. The
main difference is storage model of global stiffness
matrix. The first solver, called PreAssembly, cre-
ates stiffness matrix [K] in CRS format before itera-
tive process will run. This approach reuse once com-
puted matrix, therefore algorithms execute faster,
but needs additional memory resources on GPU. The
second solver, named MatrixFree calculate Matrix-
Vector product in each iteration and never stores
global stiffness matrix. That cause slower code per-
formance, but saves many resources on graphic card
(see Section: Numerical Results).
Numerical Benchmarks
High performance of Topology Optimisation can be
achieved only by usage of well profiled and opti-
mised FEM solver code. The numerical benchmark
were performed to check Conjugate Gradient solvers
262 | eCAADe 34 - DESIGN TOOLS | Explorations - Volume 2
efficiency, both on CPU and GPU. The benchmark was
run on 4-core Intel i5-2500 3.3 Ghz CPU with 8 GB of
RAM and nVidia GTX 970 as a GPU (1664 CUDA cores).
To verify algorithm correctness, the Stanford Bunny
model is used as a general design boundary for FEM
(see figure 12). The support and load condition are
applied as shown on Figure 14.
The performance evaluation was based on four
benchmarks performed on numerus mesh density
presented in Table 1 and visualised on Figure 13. Dur-
ing benchmarks CPU was fully overloaded. GPU ker-
nels reported 89 % of computational occupancy. It is
caused by adaptation of FEM on irregular shapes.
The results are presented in three group of graph.
The Figure 15 represents the general execution time
need to solve particular problem. It is clearly visi- Table 1
ble the huge divergence of the results for CPU and Benchmark setup.
GPU implementation. Enlargement of mesh densi-
ties caused rapid decrease of CPU performance. The
GPU PreAssemble implementation shows the effi-
ciency up to 110 times over CPU MatrixFree imple- Figure 15
mentation. The detailed information about bench- Efficiency of all
marks components are presented on Figure 16. Each solvers depend on
diagram represent one benchmark for all solvers, di- FE model quality.
vided by main subroutines, to evaluate it execution Less is better.
time. The considerable disproportion in execution
time between CPU and GPU solvers, effected the
graphs to be illegible. For appropriate data represen-
tation, supplementary charts were generated only
for GPU solvers (see Figure 17).
Properly scaled charts for PreAssemble GPU
solver revealed high disproportion in execution time
between creation of FE model and solving the eq-
uitation. That behaviour was unnoticed in other
solvers, independently of the FE model density. That
result points the weak part of TO implementation and
highlight the parts of code which need more atten-
tion during further application development. This is-
sue is clear guideline for the next steps for code op-
timisation. Another important factor is memory con-
sumption by the solvers. Depends on GPU architec-
ture and memory resources available for the com-
putation, proper solver need to be chosen. The last
Figure 16
All benchmark
scenarios with
additional division
on basic
components.

Figure 17
Detailed data for
PreAssemble (top)
and MatrixFree
Solver (bottom).

Figure 18
Memory usage
depend of solver
type and FE mesh
density.

DESIGN TOOLS | Explorations - Volume 2 - eCAADe 34 | 263

 graph (see Figure 18) shows the memory consump-
tion regarding to the solver type or density of the
FE model. As visible, matrix free implantation of CG
algorithm require up to 10 times less GPU memory
then PreAssemble solver.
FURTHER WORK
At the time the paper is written, an intensive work is
proceed on next steps of algorithm to archive satis-
factory performance of Topology Optimisation. The
GPU implementation of Optimality Criteria (Bendsøe
and Sigmund 2004) or Method of Moving Asymp-
totes provided by Svanberg (Svanberg 2007) is un-
der develop. It is predicted to achieve the final im-
plementation of TO before ECAADE 2016 conference
will run.
CONCLUSION
Fast spreading of generative design techniques al-
lows to extend existing form finding processes with
engineering solutions. Based on author's research
potentiality to adhibit structural optimisation meth-
ods towards architecture is noticeable. In conjunc-
tion with high performance of Finite Element model
and assumed easiness of tool, Topology Optimisation
might be indispensable part of form finding process
in architecture practise. Spatial structures, as an out-
come of those algorithms, can bring new and undis-
covered forms to contemporary architecture. Addi-
tionally, in collaboration with additive manufactur-
ing known as 3d printing, it moves a design process
on the next level of freedom.
Guest, JK and Moen, CD 2010 'Reinforced Concrete De-
sign with Topology Optimization', 19th Analysis and
Computation Specialty Conference, pp. 445-454
Januszkiewicz, K 2013 'Evolutionary digital tools in de-
signing nonlinear Shaping of concrete structures in
current architecture', Concrete structures in urban ar-
eas, Wroclaw, pp. 1-6
Kutylowski, R 2004, Optymalizacja topologii kontin-
uum materialnego, Oficyna Wydawnicza Politechniki
Wroclawskiej, Wroclaw
Martinez-Frutos, J and Herrero-Perez, D 2015, 'Efficient
matrix-free GPU implementation of Fixed Grid Finite
Element Analysis', Finite Elements in Analysis and De-
sign, 104, pp. 61-71
Ohmori, H 2010, 'Computational Morphogenesis: Its
Current State and Possibility for the Future', Interna-
tional Journal of Space Structures, 25(2), pp. 75-82
Stromberg, LL, Beghini, A, Baker, WF and Paulino, GH
2011, 'Application of layout and topology optimiza-
tion using pattern gradation for the conceptual de-
sign of buildings', Structural and Multidisciplinary Op-
timization, 43(2), pp. 165-180
Svanberg, K 2007 'MMA and GCMMA -- two methods for
nonlinear optimization', Optimization and Systems
Theory, KTH, Stockholm, Stockholm, Sweden.
Zienkiewicz, OC and Taylor, RL 2005, The finite element
method for solid and structural mechanics, Elsevier
Butterworth-Heinemann, Amsterdam and Boston
Zwierzycki, M 2013, 'Ewolucyjne narzedzia cyfrowe
w formowaniu struktur przestrzennych', Archivolta,
58(2), pp. 54-61

REFERENCES
Oded Amir, MB 2011 'Topology Optimization for Con-
ceptual Design of Reinforced Concrete Structures',
EUROMECH 522
Bendsoe, MP and Sigmund, O 2004, Topology Optimiza-
tion: Theory, Methods, and Applications, Springer
Berlin Heidelberg, Berlin, Heidelberg
Block, P, Knippers, J, Mitra, NJ and Wang, W (eds) 2015,
Advances in Architectural Geometry 2014, Springer In-
ternational Publishing, Cham

264 | eCAADe 34 - DESIGN TOOLS | Explorations - Volume 2

Soft Modelling
Open source Java application for flexible structural systems

Manuel Jimenez Garcia1

1
University College of London - The Bartlett School of Architecture
1
manuel.j@madmdesign.com

Contemporary advanced simulation software allow for a higher accuracy in the

understanding of material behaviour. The increase in computational power is
enabling designers to get much closer to real time physical simulations, which
facilitates the inheritance of those tools in their design workflows.However, the
use of those tools is normally limited to a series of specific steps within the entire
workflow, rather than a feature integrated in the design process
itself.Softmodelling is an open source Java application which aims to bridge this
gap by seamlessly integrating physical simulations in every step of the design
process, giving designers the ability to not only test structural behaviours of a
given output, but also allow them to design while taking both structural stability
and material behaviour into account at every stage.This paper will discuss the
design and evolution of the software, as well as showcase physical prototypes
which explore the possibilities of such design methods. These projects are
fundamental in materialising the evolution of Softmodelling, towards becoming
an application that does not only enable the design of flexible elements, but also
facilitates their manufacturing and assembly into large scale structures.

Keywords: Particle-spring systems, Dynamic relaxation, Physics Simulation,

Flexible materials, Discrete Computation, Open source, Design Software

INTRODUCTION Contemporary advanced simulation software allow

Simulation software for digital design. "Digital
tools are a powerful ally of design by making, be-
cause digital simulations can make and break in no
time more models than a physical craftsman could
in a lifetime, thus making intuitive, heuristic form-
finding by trial and error a viable design strategy.
And when a model works, either a physical model or
its digital equivalent, there may be no need to know
or tell why." (Carpo 2012)
for a higher accuracy in the understanding of mate-
rial behaviour. Software modules such as Maya dy-
namics, and Plugins for design software such as Kan-
garoo for Rhino/Grasshopper allow for innovative de-
sign methods, enabling users to understand mate-
rial performance at different stages of the design pro-
cess.
In addition, the increase in computational power
is enabling designers to get much closer to real time

DESIGN TOOLS | Explorations - Volume 2 - eCAADe 34 | 265

 physical simulations, which facilitates the inheritance
of those tools in their design workflows.
The increasing accessibility to those tools, as well
as the proliferation of "easy to learn" simulation pack-
ages, has changed the nature of "flexible" structure
design, establishing simulation-driven design as a
new territory within CAD (Computer-aided design)
software.
RESEARCH CONTEXT
From physical to digital form-finding. Since the in-
troduction of simulation software in architectural de-
sign, they have been most commonly used as a dig-
ital form finding technique through dynamic relax-
ation (Day 1965).
Form-finding methods are those in which struc-
tures define their own shape under applied loads.
The most commonly known examples deploying
these methods in Architecture are:
• The hanging chain models used by Anto-
nio Gaudi for Colonia Guell, wherein loads
were distributed on the structure as weighted
hanging chains. The pure tensile hanging ge-
ometry was then inverted, forming a com-
pression structure optimized for that particu-
lar load case (Xie 2005).
• Frei Otto's experiments on tensile structures,
which then materialised into large scale
projects. The Munich Olympic Park for the
1972 Summer Olympics, or the Institute for
Lightweight Structures, completed in 1967 at
University of Stuttgart in Vaihingen are exam-
ples of these methods.
In these examples, membranes are solely working
under tension, suspended from a multitude of verti-
cal masts. Both given examples were developed in
an analogue domain. Performing these experiments
digitally allows for much faster and more accurate
workflows, since "digital simulations can make and
break in no time more models than a physical crafts-
man could in a lifetime" (Carpo 2012).
However, in architectural design, digital simula-
tions often become a second step for previously com-
266 | eCAADe 34 - DESIGN TOOLS | Explorations - Volume 2
pleted schemes, where a modelled structure is trans-
formed into a "physically active" object. The topology
of that structure needs to be therefore fixed prior to
its conversion into an active membrane. This takes
digital form-finding slightly far from the playfulness
of the design processes followed by Frei Otto or An-
tonio Gaudi, where local modifications to the struc-
tural setup can occur simultaneously to their physical
performance. This results in singular processes com-
piled into a series of operations, rather than a feature
integrated in the design process itself. The workflow
progresses as follows: the designer models a struc-
ture, then applies physical forces, remodels it in case
of any undesired results. After the physical simula-
tion is completed, the user is unable to intervene in
the resulting design, since any topological modifica-
tions to the object will need to be followed by a new
physical simulation.
Limitations of physical simulation in computa-
tional design. Some software packages and simula-
tion plugins allow for further interaction with the ob-
ject after becoming "physically active", allowing the
addition of new elements to the "physical scene".
This is the case for Autodesk Maya NCloth and
Grasshopper Kangaroo. However, the dynamic trans-
formation of the scene is limited to the level of the
object as a whole rather than to its specific topology.
Thus, users can´t perform any modelling operations,
such as extruding or subdividing faces. Addition and
subtraction of new objects to the scene are the lim-
ited operations possible.
The closest solution available is the reduction of
the user interaction to merely particles and springs,
interpreting those as objects. This allows for the ad-
dition of new members to a structure, as well as their
dynamic subtraction. The structure can therefore
be actively modified, growing or shrinking locally, or
even breaking previously created springs. However,
this limits the manipulation of the element to its low
level structure, configuring itself as a tedious (non-
intuitive) process. In turn, making the creation of
faces a difficult procedure, which is crucial for design
processes involving solid objects.
Figure 1
Softmodelling
interface developed
in Processing.
Render mode with
active conversion
from faces to linear
elements.
Figure 2
Softmodelling -
Graphical map of
selection modes
and
mesh-manipulation
types.
Poly-modelling physics. Softmodelling (e.g. Figure
1) is an open source Java application which aims to
bridge this gap between poly-modelling and simula-
tion, by seamlessly integrating physical simulations
in every step of the design process, giving design-
ers the ability to not only test structural behaviours
of a given output, but instead allow them to design
while taking both structural stability and material be-
haviour into account at every stage; The software de-
veloped by Manuel Jimenez Garcia creates a contin-
uous feedback between poly modelling and physical
simulations, it incorporates most basic modelling fea-
tures present in off-the-shell design packages, con-
necting them to physical simulations at all stages.
Its strength relies on the design and manipulation of
tensile structures, giving the user the ability to mod-
ify the topology of the object, while actively respond-
ing to physical forces.
The software redefines the use of particle-spring sys-
tems, establishing them as a step-by step transforma-
tive process. As a result, this allows the local manipu-
lation of physical elements, through the use of com-
mon modelling techniques. The software seamlessly
transforms mesh vertices into particles, and edges
into springs; bringing forward an output that is both
physically and geometrically precise.
FLEXIBLE COMPUTATION
Design software for membrane structures. When
working on specific structural types, the tendency is
to use specialised software and plugins. In the case
of membrane structures, software packages such as
Membrane NDN, K-3 tent or Formfinder, are exam-
ples of such programs. However, those packages are
closer to the development of a finished product, fol-
lowing an engineering approach, this makes them
less practical for their use in early stages of the design
process.
During those stages, modelling software, such as Au-
todesk Maya or 3DS Max, allow for more playful de-
sign methods, establishing themselves as the perfect
DESIGN TOOLS | Explorations - Volume 2 - eCAADe 34 | 267
 tool to respond to the changing nature of a design in
its conception stage.
Softmodelling situates itself as a design tool, in
which playfulness and flexibility of poly modelling
software packages are seamlessly integrated. It of-
fers accuracy in structural form finding methods, of-
fered by membrane engineering packages. This in-
tends to offer a more extensive approach in regards
to the manipulation of the structure. As a result,
one is able to create different tensile morphologies
beyond the commonly used Hyperbolic paraboloid,
conic and barrel vault shapes. (e.g. Figure 2)
Open source in design and simulation. "The free-
dom to study how the program works and change it
so it does your computation as you wish [...] by doing
this you can give the whole community a chance to
benefit from your changes." (Stallman 2001)
Modelling software are commonly established
as commercial packages, and therefore their source
code is not accessible for users. Due to their commer-
cial use, and the large number of users within design
disciplines who depend on such software, the pack-
ages need guaranteed robustness in relation to their
performance.
On the other hand, there are a vast variety of
small applets and libraries which allow for concrete
design operations. The open source character of
those tools allow for the development of custom
made applications, in which designers can focus on
the integration of specific features, rather than hav-
ing to develop a full package from scratch.
Softmodelling emerges from the integration of
some of those libraries, into a software that intends
to bring a more complete modelling experience to
those interested in the design of tensile structures. Its
Open Source nature allows users to create their own
version of the software, as well as contribute to its
development, enabling the creation of more specific
tools when required by the user´s project.
The software is developed using Processing, de-
veloped by Casey Reas and Ben Fry, as a framework,
and it makes used of a variety of open source libraries,
such as Toxiclibs by Karsten Schmidt for physics sim-
268 | eCAADe 34 - DESIGN TOOLS | Explorations - Volume 2
ulations, HEMesh by Frederik Vanhoutte for half'adge
meshes, Peasycam by Jonathan Feinberg for three
dimensional navigation, and ControlP5 by Andreas
Schlegel for the GUI integration.
Responsive particle-spring systems. In commonly
used 3d modelling packages, such as Autodesk Maya
or 3Ds Max, physical simulations are considered to be
post-design processes. This is a result of the recom-
putation of the mesh structure when any modelling
operation is performed. Vertices and edges' indexes
are recurrently recomposed. This constant transfor-
mation proposes a difficulty while matching the in-
dex of the particles to vertices and springs to edges.
Therefore, the parity of both systems can't be guar-
anteed.
In order to avoid the constant negotiation be-
tween these two systems, most modelling soft-
ware programs, with integrated simulation mod-
ules, fix all index numbers corresponding to the
mesh components prior to the transformation of
the object into a particle-spring system. Hence,
any subsequent alteration in the meshes' topology is
not taken into account in the physical simulation.
When faces are extruded or subdivided, the new ver-
tices do not become particles, and the new edges do
not become springs. This produces a mismatch be-
tween the data structure of the mesh and its phys-
ical system, hence a deviation between the visual
representation of the object and its performance in
the simulation. Therefore, if any further adjustments
should be performed to the morphology of the three
dimensional mesh, the user is forced to return to a
previous stage and re-run the simulation after those
adjustments are completed.
Softmodelling develops a strategy, to overcome
the technical challenges regarding the most com-
mon poly-modelling operations. First, all index num-
bers of the particles and springs are relocated af-
ter any topological transformation is performed, this
maintains the parity between the particle-spring sys-
tem and their associated vertices and edges. Con-
secutively, a detailed analysis of the mesh iden-
tifies which elements have been affected by any
Figure 3
Softmodelling 1.0
launch event - Left:
Installation for
London Clerkenwell
Design Week 2014
at the Actiu London
Showroom. Right:
Catalog of three
dimensional
models created
with the software.

topological transformation. The isolation of those
areas allows for an increase in the system's effi-
ciency, thus avoiding the use of computationally ex-
pensive processes in any other area of the object.
Once the local data matching process is completed,
particles and springs get reconnected to the new as-
sociated vertices and edges. This enables the object's
physical properties to be updated with a new topo-
logical structure.
TOWARDS A DESIGN AND FABRICATION
WORKFLOW
Data to matter and matter to data.. Softmodelling
was first tested during Resonate 2014 in Belgrade,
and was officially launched during London Clerken-
well Design Week (May 2014) at the Actiu London
Showroom, where the software included its first
multi-tactile interface. This allowed participants to
create their own structures. The exhibition also in-
cluded a series of 3D printed models designed by
MadMDesign, which aimed to test the efficiency of
the software as a design tool. The models were a re-
sponse to the challenge set by Actiu London Show-
room to create 24 different objects in only eight
hours (e.g. Figure 3). The feedback collected dur-
ing the event was fundamental for the progression
towards the next stages of the software.
Since then, Softmodelling has been tested in sev-
eral workshops and installations, aiming to resolve
medium scale structures through the use of flexible
materials. Alongside this process, the contribution
of workshop participants has resulted in the creation
of a variety of pavilions and installations, aiming to
measure both material behaviour and the efficiency

DESIGN TOOLS | Explorations - Volume 2 - eCAADe 34 | 269

 Figure 4
Trans-
Computational
Pavilion 1.0. -
Textile prototype
created during the
AA Visiting School
Madrid 2014.
Exhibited during
the X Week of
Architecture in
Madrid at COAM.
Design team:
Manuel Jimenez
Garcia, Roberto
Garcia, Antonio
Guijarro, Juan
Rodriguez Martín,
Jose Luis E. Penelas,
and AAVSMadrid
2014 students.

of the software as a design/fabrication tool.
The material experiments and physical studies
that took place during these several events became
extremely instrumental to the development of the
software. This paper will focus on two installations,
which correspond to the two stages of the software's
development: membrane structures manipulation,
and flexible tubular elements which perform in a
structural symbiosis with the membrane itself.
HYBRID FLEXIBLE STRUCTURES:
Softmodelling 1.0. The prototypes developed dur-
ing the first stage, make use of the force density
methods (Schek 1974) integrated in the software to
create inhabitable structures. These structures, are
driven by basic principles of rapid reoccupation sys-
tems, which offer a variety of flexible spaces. The pro-
totypes were conceived from the idea of utilising low
cost materials with a high degree of flexibility. Inter-
weaving these flexible elements together, in order to
create localised stiffness in areas of the structure with
higher stress levels, results in increasing the overall
stability of the structure.
A series of pavilions are developed as an out-
put of Softmodelling, through an assemblage of two
types of flexible materials: PVC pipes and stretch-
able textiles. While the PVC pipes bundle into struc-
tural objects, the textile pieces get stitched together
to generate continuous surfaces. Thus the multiple
membranes generate compression forces to encour-
age the PVC pipes to stay connected, while the linear
structure simultaneously pushes the membranes to
maintain their state of tension. This results in struc-
tural balance producing stiffness within the architec-
tural element.
Computational membranes. Prior to the official
launch of the software, a primitive version of Soft-
modelling was used for the design of the first large

270 | eCAADe 34 - DESIGN TOOLS | Explorations - Volume 2

Figure 5
nstallation for
London Clerkenwell
Design Week 2015
at the Actiu London
Showroom. Flexible
structures - Design
team: Manuel
Jimenez Garcia,
Roberto Garcia,
Miguel A. Jimenez,
Ignacio Viguera.

scale pavilion in this research, executed during the creation of linear structures that work in conjunction
AA Visiting School Madrid 2013 (e.g. Figure 4). Due to with the general topology.
the technical limitations of the application in its early
Figure 6
stages, the construction was created as a result of an
Trans-
intuitive process, rather than as a product of digital
Computational
fabrication workflows. Regardless of the lack of con-
Pavilion 2.0. -
tinuity between the digital model and the physical
Flexible structure
prototype, this testing process allowed for the first
based on linear PVC
establishment of structural elements. This, as a result
elements.
materialised a link between material density, connec-
Photography taken
tivity and stretching.
prior to the
Later prototypes created with this system in-
addition of the
clude the second iteration of the AA Visiting School
textile membrane.
Pavilion, in 2014 (e.g. Figure 5), as well as the in-
stallation for London Clerkenwell Design Week 2015 Although this feature is not yet fully completed, Soft-
(e.g. Figure 6-7). Both projects offered a smoother Modelling 2.0 does offer a conversion from faces to
workflow between the software and the fabrication linear elements. This aims to create new workflows in
method. However, the fact that the software only of- which designers can not only control a tensile struc-
fered surface manipulation and dismissed the inte- ture with external anchor points and frames, but also
gration of the frame, became an apparent limitation. create the frame itself. This is based on a combination
This created a clear necessity for SoftModelling to in- of multiple bended linear elements which connect to
corporate a new feature which would allow for the create a structure in perfect equilibrium.

DESIGN TOOLS | Explorations - Volume 2 - eCAADe 34 | 271


The modularisation of the linear elements; which
bend differently to be assembled into larger objects,
brings a higher degree of control to the structure.
The fixed lengths of the pieces not only simplifies
their computation, but also the manufacturing pro-
cess, which as a result allows for the use of simpler
digital fabrication methods.
DISCRETE BENDING :
Softmodelling v2.0. "Digital materials consist of a fi-
nite set of parts that have discrete connections and
occupy discrete space." (Ward 2010)
As previously mentioned, the second stage of
the software development focuses on the flexible lin-
ear elements rather than on the membrane itself. The
aim is to generate a series of discrete flexible pieces
that maintain the membrane in a state of tension.
However, in order to conserve the intuitive nature
of Softmodelling as a design tool, the distribution of
those pieces must be linked to the topology of the
mesh. The most recent version of the software of-
fers a conversion from faces to linear elements. This
aims to create new workflows in which designers can
not only control a tensile structure with external an-
chor points and frames, but also create the frame it-
self (e.g. Figure 8).
This feature was tested in several installations
where the bendability and connectivity of flexible el-
ements were further explored. The projects corre-
sponding to this second stage of research, aim to
isolate the structural integrity to discrete flexible el-
ements. Therefore, the integrity of the structure is
more dependent on the contribution of these afore-
272 | eCAADe 34 - DESIGN TOOLS | Explorations - Volume 2
mentioned elements rather than the on the mem- Figure 7
brane itself. Trans-
Computational
Discrete computation for flexible linear elements.
Pavilion 2.0. -
The installation at Dezact, entitled Offshore Bezier
Completed flexible
(e.g. Figure 9), is the largest prototype exploring this
prototype, created
combinatorial arrangement of linear elements. One
during the AA
of its primary aims is it to explore the material be-
Visiting School
haviour of linear elements in a large scale. In this oc-
Madrid 2015.
casion, a vernacular material was explored and vastly
Exhibited during
available in the area: bamboo. Even though bamboo,
the XI Week of
in terms of material performance is stiffer than the
Architecture in
PVC pipes used in previous installations, the geomet-
Madrid at Roca
rical logic for their arrangement remains the same.
Madrid Gallery.
Despite early attempts at digitally manufacturing the
Design team:
nodes between the elements to achieve higher con-
Manuel Jimenez
trol over the structure in relation to its digital replica,
Garcia, Roberto
the use of SoftModelling was limited to an approxi-
Garcia, Antonio
mation of the structural arrangement rather than as
Guijarro, Maria
a fabrication tool. The software however, proved
Olmos, Miguel A.
that this arrangement can be linked to stress anal-
Jimenez, Ignacio
ysis and therefore the distribution and connectivity
Viguera, Jorge
of discrete linear elements can be controlled in real-
Cerdá, Marina
time during the modelling stage.
Rodriguez, Vicente
CONCLUSION Soler, Silvia Rivera,
The aforementioned projects not only configure a Maria Olmeda, Jose
showcase of the possibilities of such design meth- Real, Jose Luis E.
ods, but also contribute to generating data which in- Penelas, and
forms the development of the software. The evolu- AAVSMadrid 2015
tion of the software towards digitally controlling the students.
assemblage of flexible elements, facilitates the ability
to achieve higher degrees of complexity in the gen-
erated structures.
This configures a new territory in the domain of
flexible structure design. Maintaining the high level
of control of such structures through simple mod-
elling operations, configures a unique feature of Soft-
modelling. However, controlling both membrane
and frame while maintaining the playful character of
the software, becomes a challenge yet to be resolved.
The next stages of the research and develop-
ment of SoftModelling will tackle the refinement of
the linear elements in the digital realm, introduc-
ing physical testing for their bending moments and
Figure 8
Softmodelling 2.0. -
Application
screenshot -
Isolation of bezier
piping structure
linked to mesh
faces.

Figure 9
OffShore Bezier. -
Installation
produced during
Dezact 2015 (Extra
Fabrica) at Shin
Chien University,
Taipei. Hanging
structure
configured as an
assemblage of
bamboo
pipes. (Photography
by Dezact). Design
team: Manuel
Jimenez Garcia,
Christina Dahdaleh,
and Dezact 2015
students.

DESIGN TOOLS | Explorations - Volume 2 - eCAADe 34 | 273

 their structural performance. This will be achieved
through redefining the relationship between the lin-
ear elements and the faces of the object. By estab-
lishing these elements as separate entities indepen-
dent of each other, the digital model will resemble a
closer configuration to the reality of the built struc-
tures, while maintaining the intuitive nature of Soft-
modeling as a design tool.
Furthermore, the development of a robust fab-
rication module becomes essential for the construc-
tion of future large scale prototypes. The module
should enable the smooth translation of digital data
from Softmodelling into robotic instructions. This
should not only enable the fabrication of discrete el-
ements, but also simplify the assembly process. The
installation for Dezact Taipei 2016, will mark the first
attempt at testing this process. Tool paths for two
Universal Robots (UR) working collaboratively are be-
ing developed. While these two robotic arms per-
form bending operations on the linear elements, a
third UR, equipped with a custom-made heating end-
effector, moves along the previously bended ele-
ment to keep its shape after completion of the pro-
cess.
These steps attempt to fill the gaps in the design
to fabrication workflow, intending to offer a contin-
uous experience from the conception stages of flex-
ible structures to their materialisation. This will be a
fundamental milestone in the development of Soft-
modeling. The Dezact project is to be completed in
August 2016.
REFERENCES
Carpo, MC (eds) 2012, Log: Digital Darwinism: Mass Col-
laboration, Form-Finding, and The Dissolution of Au-
thorship, Anyone Corporation
Day, ASD 1965, 'An Introduction to Dynamic Relaxation',
The Engineer, 219, p. 218–221
Schek, HJS 1974, 'Force Density Methods for Form Find-
ing and Computation of General Networks', Com-
puter Methods in Applied Mechanics and Engineering,
3, pp. 115-134
Ward, JR 2010, Additive Assembly of Digital Materials,
Ph.D. Thesis, Massachusetts Institute of Technology
[1] https://www.gnu.org/events/rms-nyu-2001-transcr
ipt.txt

274 | eCAADe 34 - DESIGN TOOLS | Explorations - Volume 2

Esquis'sons ! Sound Sketch : A Parametric Tool to Design
Sustainable Soundscapes
How to apprehend environmental complexity in a simple tool for
architectural design

Théo Marchal1 , Nicolas Remy2 , Grégoire Chelkoff3 ,

Jean-Luc Bardyn4 , Noha Gamal5 , Hengameh Pirhosseinloo6
1,2,3,4,5,6
CRESSON
1,3
{theo.marchal|gregoire.chelkoff}@grenoble.archi.fr
2,5,6
{nicola.remy|n.gamalsaid|hp.amini}@gmail.com
4
jlbardyn@sfr.fr

Since the 80s, several researches have developed the theoretical notions of sound
effects, sound proxemy, city sound identities, sound comfort, architectural sound
prototypes which were meant to help designers consider sound in their projects.
Nevertheless, taking care of the inherent sound dimensions in architectural urban
projects remains an unresolved challenge. The researches of the last 30 years
have shown how the sound environment qualities are forgotten in favour of visual
qualities. This article presents a new method dedicated to generating simple
sound sketches for architectural conception while preserving the complexity of
acoustic simulation. This paper argues that the Esquis'sons! sound sketch tool
reconfigures architectural design by considering an innovative view its the
temporality, allowed by numeric designing tools able to intervene and offer a
continuous feedback regarding sound environment.

Keywords: sound environment, sound effects, sketch, parametric design,

architecture, didactic, grasshopper

CONTEXT AND ISSUES, SOUND QUALITY ing care of the inherent sound dimensions in archi-
PREDICTION IN URBAN SPACES  tectural urban projects remains an unresolved chal-
Since the 80s, several researches have developed the lenge. The researches of the last 30 years have shown
theoretical notions of sound effects, sound proxemy, how the sound environment qualities are forgotten
city sound identities, sound comfort, architectural in favour of visual qualities. Yet today we are aware
sound prototypes (Augoyard et al. 1982; Chelkoff of the importance of soundscape and its place in the
and Balaÿ 1987) which were meant to help designers perception of spaces as well as visual cues. It appears
consider sound in their projects. Nevertheless, tak- important to consider this dimension of space with

DESIGN TOOLS | Explorations - Volume 2 - eCAADe 34 | 275

 a qualitative eye and not only quantitative preoccu-
pation (Rémy 2005). In addition, standardization as-
sociated to sonic space concerns tends to forget the
characters (Luis Bento Coelho 2015) and variability of
the places that are a real design challenge.
Regarding architecture, tools and ways of think-
ing have yet to be developed in the light of new
prospects. From this perspective, the aim would be
to "integrate" the soundscapes in the spatial design
process based on the potential of tools and instru-
ments proposed by "digital" architectural design.
Sound and "vibrations" in space, in their usual
meaning, are often considered as qualifying ele-
ments for objects or their materiality. Here, we
seek to understand "sounds" as factors, interacting
with the rest of space component (Casati and Dokic
1994). Thus, they do not qualify emitting elements,
but instead, they become, by their spatial and sen-
sitive presence, real events constituting the << at-
mosphere>> or ambiance. This work on the quality
of sound in dense urban areas questioned the con-
frontation between the different elements in a per-
ceived sound environment, and thus, in the "sound
scenes", specific for each element of the space.
From "room acoustic" to "sound quality de-
sign" tool
Today, the only tools developed to "predict" sound
qualities for urban and architectural planning are
dedicated to room acoustics, therefore they are not
relevant for outdoors and complex sound environ-
ments. In addition, the complexity of their interfaces
make them hard to use and to understand. As a mat-
ter of fact, to use these tools (for both indoor and
outdoor acoustic simulation) every parameter has to
be designed and characterized by the user (volumes,
materials, etc.) to assess the design. The logic of the
process is closer to correction than design. Facing
this lack of tools to design the project and its sound
"qualities" conjointly, this article presents new meth-
ods that allow simple sound sketching for architec-
ture while preserving the complexity of acoustic sim-
ulation.
276 | eCAADe 34 - DESIGN TOOLS | Explorations - Volume 2
If the current sketch tools give the possibility - in a
very intuitive way - to see and tests formal hypothesis
in the early stages of the design, they don't offer any-
thing to hear, and the new perspectives given by the
recent development of parametric tools for architec-
tural design finally allow us to think about intuitive
and simple tools to produce sketches of the sound
space complexity.
THE CHALLENGES OF SENSITIVE SKETCH
FOR THE DESIGNER : SIMPLICITY WITH
COMPLEXITY, NEW PROCESS AND PRAC-
TICES IN DESIGN METHODOLOGIES
Considering the architectural design process not as a
time line, but as a cyclical dynamic "where the object
assumes a place in a continuum by variation" (Deleuze
2006), means being aware of the impact this transfor-
mation necessarily has on design practices. Indeed,
our proposal addresses the architectural design as
a tool continuum. Thus the design tool, whether it
is material, intellectual or virtual exists by itself and
tends to detach from the temporal anchor that for-
merly characterized him. The architect or designer
works on the different stages of the process, all articu-
lated and linked to one another, without ever having
to query the "temporal relevance" of the tool. Mod-
elization, or building a three-dimensional prototype
becomes part of the continuous protocol and design
crystallizes itself through action. The activity trans-
forms the object, but "in doing [transforms] the sub-
ject itself"(Vergnaud 2006).
This methodological shift results in a poten-
tial for the instrument itself, which could be appre-
hended as a sound space conception tool. Therefore,
it seems appropriate to think of technical tools not as
simple "means" or unconcealment (Heidegger 1980)
for designed spaces, but as a "part" of them, the in-
formant and the listener nourishing both process and
method. A sound sketch tool allows both representa-
tion and conception activities.
Thus, this work seizes contemporary technical
tools to exploit their adaptive capacities and the as-
sociated thought processes. It aims at articulating
those tools with the creative process and at providing
an alternative to normalisation and to digital automa-
tion. Indeed, despite considerable progress in terms
of knowledge and experiences regarding sound en-
vironment, the current works are still struggling to
make concrete proposals likely to help designers to
create new "variable" sound environments that offer
a variety of sound affordances (Gibson 2011) .
These questions, which attempt to couple a re-
flection on the contemporary changes in architec-
tural design process and a special attention to the
quality of perceived areas raise questions about new
dynamics and "concept-in-action" (Vergnaud 2009):
it comes to seize the digital tools in this perspective
in order to develop the potential for the field of spa-
tial design.
Didactic potential of programming tools for
design
We present here a potentiality linked to design meth-
ods associated with digital programming tools that
engage "cognitive" processed related to the assem-
bly and error logic found in works based on models
or sketches. Unlike other types of tools, the different
elements of a script are built gradually and conjointly
to the creative process: The elements in the script in-
terchange, feed, cancel themselves and even canni-
balize some part of it in the manner of a sketch that
transforms itself under the designer's impulses and
the previous layers which produce the shape or "sil-
houette" of space.
There is a didactic potential of the scripting and
programming (visual or not) to which is associated
an aesthetic of the "system" that approaches the "in-
tuitive" process of models and sketches. These new
tools have a great potential for a continuous design
process, which needs to be mobilized in design and
building spaces.
These considerations lead us to consider us-
ing digital design tools by summarizing the known
process of sketching and model through a visual
programming tool : Esquis'Sons! then tries to fit
within this process by allowing the technical articula-
tion and a sound sensitive and interactive variability.
Seen this way, the tools allow to develop unfrozen
systems, intuitive, variables and parametric and so to
not mobilize artefacts to design but instruments of
design. Indeed, the designer takes the artifact and
personalises it through a process of instrumentaliza-
tion, at the same time it creates a set of schemes
of use through an instrumentation process (Rabardel
2002). Esquis'sons! then offers to take over these pro-
cesses through a sketch tool that unlike a validation
tool is not intended to be a frozen artifact but articu-
late and facilitate this personalization by incorporat-
ing it into the design process.
The sketch as a physical and mental design
tool
Working with numeric simulation and evaluation to-
day appears to add an unwanted complexity to the
designing process. Yet, with the ever evolving power
of new tools and computers, we start to think of tools
capable to change this complexity of data in an un-
derstandable way without leading to an oversimplifi-
cation and loss of information. The "sensitive sketch"
developed with these tools, able to adapt to these
new design paradigms (Saggio 2013) should allow to
experience sensitive issues within the architectural or
urban project.
This work seeks to exploit new cyclical processes
that emanate from the contemporary configuration
of architectural design. This means switching from
a linear process consisting in a logical sequence
where the sketch creates the project and its design,
which is then validated or invalidated by simulating
some phenomena; to a more complex articulation
between the different phases that feed each other
and a interact to continuously actualize the process.
The various tools related to these phases end up ex-
isting by themselves, breaking away from their tem-
poral position in the process. The paradigm behind
the development of this tool emphasizes the need for
a simultaneity of sketching and assessing within de-
signing processes.
By questioning the design of environments we
DESIGN TOOLS | Explorations - Volume 2 - eCAADe 34 | 277
 consider the posture of sketching more relevant and
versatile in comparison with a simulation and valida-
tion posture which is defined through its timing in
the design. Indeed, the relationship between project
and sketches is not a causal relationship but a con-
stitution, or an evolution of the same "effect" con-
stituted of a single essence. In the way of the sen-
sitive sound environment, we do not try to catego-
rize items according to a specific sound character but
to bring out a constituent assemblage (Deleuze and
Guattari 2005) of the sensitive soundscape.
In its first meaning, sketching includes both a
preparatory tool for design work and an equalizer for
the final project which the nature must match the
main ideas promoted by the sketch. This is actually
a "first" act of representation, but also of design.
The sketch, by evolution, builds itself around
elements and events that transform the process of
sketching, and requires a number of conditions to be-
come a physical and mental design tool, stimulating
the imagination. Then, more than an evolution, the
sketch becomes "revolution" in a rotating way, exac-
erbating the complexity of its design. The sketch as a
design tool is the initiator of the project, it is both the
essence and the project itself.
The simplicity of the sketch illustrates the stakes
of its use in the era of digital design tools. It has to
stay easily usable and understandable, but also capa-
ble to manage and to fill the gaps left by unknown
information. The architectural project today is no
longer defined by the sequence of events and project
items aimed at a final achievement, but constitutes
a whole, reacting as a system of interconnected pro-
cesses ; therefore, it involves a reflection on continu-
ous and non-linear systems which have to be taken
into account from the beginning of the process.
The architectural sketch needs to be integrated
in the design process through this new perspective
of timing in the architectural project and as a part
of an on-going multiplicity. Thereby, we constructed
an understandable tool which can manage and trans-
late the complexity of an environment by sketching
acoustic spaces and phenomena, but also include in-
teraction to change and impact the whole project.
The second issue is the media that will figure the
sketch and its role in the constitution of it, through
the degree of abstraction, and liberty or rigor it pro-
poses. Thus "the type of dialogue between the user
and the system"(Ciblac et al. 2005) plays a key role
since it is essential to question how the tool interacts
with the user and offers solutions to the data high-
lighted by the designer. Sketch methods then deter-
mines the creativity and imagination of the designer;
and the tool becomes a condition to elements the
designer has to "coerce" through a set of parameters
decided previously or during the formalization of the
sketch.
ESQUIS'SONS ! THE INTERFACE AS A CASE
STUDY, ASSUMING AN URBAN SOUND
QUALITY
The dense habitat today partially solves the urban
sprawl problem, but it still runs some challenges in
raising the question of the quality of life through the
social link it offers, the concept of privacy, and in-
teraction with the surrounding context. Residents
could accept the density more easily, setting aside
"the ideal of the house with garden", if their accom-
modation can offer some extensions to the outside
such as deep balconies, large covered terraces, pa-
tios, courtyards, gardens, etc. Whatever their use,
these intermediate spaces are at the interface be-
tween private and public spaces. From an acoustic
point of view, these interfaces can be characterised
by the qualities of the environment, and by the uses
they receive, which are also interfering with the qual-
ities of the public spaces.
The question that arises is the following: how
to create outdoor spaces attached to the apartment
that can offer enough insulation to enjoy privacy, but
also remain connected to the outdoor sonic environ-
ment ? Interfaces as a typology of spaces can re-
sume the majority of the acoustic issues studied at
the neighbourhood scale. The research proposes to
study it at an architectural scale, in another words at
the human body in movement scale (Chelkoff 2005)

278 | eCAADe 34 - DESIGN TOOLS | Explorations - Volume 2

 where sound qualities have meanings for the inhabi-
tants.
Methodology and case studies
In its first stage the software Esquis'Sons! is dedicated
to sustainable neighbourhood design and offers the
possibilities to sketch and listen sound qualities that
might be experimented both within building facades
(form and material used for balconies, terraces and
loggias) and in public spaces (form of city blocks,
with or without vegetation, etc.). To achieve this
goal, a one-year in situ study assessed sound qual-
ities in a large spectrum of sustainable neighbour-
hoods (France, Sweden, Germany and Spain), the
analysis of the recordings collected in these neigh-
bourhoods was used to create and integrate several
scenarios to the tool.
Figure 1
Balconies in Paris,
Trapèze de l’Île
Seguin (up),
Grenoble
Vigny-Musset
(down).
Assuming that interfaces are spatial arrangements,
classified in 4 types (balconies, terraces, loggias and
circulations or BLTC), the goal was to focus on their
dual role : both creator and receiver of urban sound-
scapes. As they are one of the "aesthetic" outdoor
expression for architecture, these BLTC are in specifi-
cally developed in terms of spatial types and surfaces
in recent dense eco-districts. As these housing ex-
tensions meet a strong demand from the inhabitants
and add value to the uses inside this density, it seems
appropriate to assess and to design them according
to their potential uses and sounds in different urban
and climate contexts.
The tool "Esquis'sons!" as a design support to sus-
tainable soundscapes, strives to present a set of re-
alities observed in the field and organized into cat-
egories according to the work of Gregoire Chelkoff
distinguishing "form", "forming" and " formality "of
a space (Chelkoff 2011). Thus, the categorization
"form" allows firstly to objectify a system: it is consid-
ered through its dimensions, materiality, orientation,
degree of openness towards outside, etc. The aim is
to reveal the type of planning designed as presented
formally to the surrounding elements
The "forming" category is more specifically re-
lated to the modes according to which the system
will form over sound. We will consider the degree of
form in relation to the use and listening and the char-
acteristics of the sound space to apprehend the sensi-
tive variables of the system (hermetism and "exhaust
degree", or special sound effects for example).
Finally, the category of "formalities" specifically
reports "best" or singular uses occurring in the sys-
tem and characterized by it ("forms" and "formings").
Especially, we are interested in the practices and their
temporalities pertaining to "sensible uses".
This research aimed at working on a broad panel
of fieldworks that could bring out remarkable and in-
novative architectural forms, and reflect morpholog-
ical evolutions that occur on interfaces (see Figure 1).
Furthermore, the international perspective of this ty-
pology was necessary to take into account cultural,
climatic and geographic variations in the uses of sim-
ilar BLTCs. That way, this work tries to consider a reci-
procity between technical / productive and contex-
tual / projective methods (Leatherbarrow 2009), join-
ing measurements and theoretical situations to an at-
tention to adaptability and context.
To illustrate these differences, we worked in 6
fieldworks located more or less on a north-south axis
and in 4 different European countries, as described
below :
1. The Caserne de Bonne (CB) district in Greno-
ble, France
DESIGN TOOLS | Explorations - Volume 2 - eCAADe 34 | 279
 2. The Sarriguren "EcoCiudad" (EC), Metropoli-
tan area of Pamplona, Spain
3. Französisch Viertel (FV), Tübingen, Germany
4. Hammarby Sjöstad (HS), Stockholm, Sweden
5. Trapèze de l'Île-Seguin (TIS), Boulogne-
Billancourt, France
6. Vigny-Musset (VM), Grenoble, France
On this methodological basis, the research results in
a catalog of situations studied and based in particular
on the upstream study of sound recordings through
a series of "cross listenings" (Remy et al. 2015)
These analyses, especially through "comparative
listening", allowed us to update a number of ques-
tions / scenarios related to architectural design for
this kind of spaces and to architectural gestures and
constructive choices or material that will impact on
the quality of sound - produced and listened. Those
are simple gestures or scenarios, but they act as an
architectural formulation for listening assumptions.
These "development" stages will then guide us to
representation and auralization questions which are
the main topic here according to our goal to bring a
simple way to design by listening to sounds. What
would be the consequences for listening if : "I am no
longer on a balcony but on a loggia" / "the material-
ity of our designed space changes" / "my balcony is
double-height" / "I am facing a street or not" etc., are
the kind of questions the tool is able to answer.
FROM IN SITU ANALYSES TO A DIGITAL
TOOL FOR ARCHITECTURAL DESIGN
The software interaction between sound and space
parametric modelling is incarnated in a modelled 3D
sketch in Grasshopper and an acoustic sketch in a
standalone application connected with each other
interactively. Thus, the program is developed as a de-
sign aid for the auralization (Kleiner, Dalenbäck, and
Svensson 1993), that is to say a spatial visualization in
interconnected and interrelated ways. It seeks to in-
troduce the digital tool as an instrument which could
articulate sounds and spatial dimensions as a whole
and translate the complexity of acoustic calculations
280 | eCAADe 34 - DESIGN TOOLS | Explorations - Volume 2
in auralizations. That way, we are keeping most of the
complex information from sound in space, but ren-
dered in an acoustic auralization which allows the de-
signer to hear it in real time.
Figure 2
Esquis'Sons!
interface
visualization [2].
Soundsketch tool, the basics
The Esquis'Sons! module offers listening points on
a project for pre-hearing, with the same level of ac-
curacy than the architectural visual sketch, various
sound scenarios that the designer might want to test.
This stage also includes a formulation of architectural
questions that, at that sketch "level", could affect the
sound quality: formal, material and functional quali-
ties of urban voids - which will focus on the effects on
the propagation of sound, the organization and the
distribution of activities, on the aesthetic dimensions
of the project that may influence the sound qualities
by future users.
The ambition has been to engage a 3D model-
ing software (Rhinoceros3D) with a module capable
of broadcasting and modulate the sound (see Figure
2), the aim was to link these two elements by exploit-
ing the interactivity and the potential of parametric
tools. We chose to use Grasshopper3D and MaxMSP
that, focusing respectively on formal modeling and
acoustic modeling, were able to adapt to each other
according to determined parameters.
The main idea of linking these two parametric
softwares comes from the fact that they both use the
 principle of settings (variable and mainly digital) as
background information. Therefore, a normal user
using Rhino during his design process can download
and install the Max / MSP module (compiled as an
application) that will dialogue with the spatial model
trough the rhino plugin Grasshopper.Designers have
to declare some basic parameters of geometry, and
they can hear the sound live, informed directly by the
digital sketch: it then produces a "sound sketch".
The interactivity principle between the two soft-
wares, one that generates spatial forms and the other
that generates sound events, is managed by a UDP
communication protocol (User Datagram Protocol
(Postel 1980)), preferred for its simplicity, versatility
and fastness of use (see Figure 3). The idea of having
these two softwares communicate is based on the
parametric principle (variables and math) they both
use as their basis information to generate forms or
sounds. We then use the information from one to an-
other.
Figure 3
User Datagram
Protocol in
Grasshopper –
gHowl plugin.
Soundsketch tool Principles
The software solution considers a "BLTC" attached to
the listener: a cuboid with adjustable dimensions ac-
companies the declared points (see Figure 4). This
volume is characterized by a degree of closure for
which the following principles were applied:
1. The cuboid around the listening point can fit
the dimensions of a front "BLTC".
2. The porosity of each of the 6 faces can be ad-
justed with an individual slider : the cover of
the BLTC, the possible closure of the side walls,
the floor and the wall overlooking the ground
(partial for a railing or or total for a double
skin) is expressed by a closure percentage on
each face.Esquis'Sons! on this basis, allows to
compare two listening points. A stereo listen-
ing to these two points selected by the user is
then used as a tool to update (to hear) differ-
ences that are actually sound translations of
space designer questions.
It is possible to assign an absorption coefficient on
the declared "BLTC" to reflect the capacity and quality
of materials. That coefficient is comprised between
0 and 1 and is relative to the entire "BLTC".One to
four buildings can be declared, for which the length,
width and height can be adjusted. They also can be
rotated to adjust the opening or closing of the block.
The designer can also import one or more of his own
Rhinoceros/Grasshopper modelled volumes.
The user then declares 4 "cardinal" soundscapes
around the block (in the North, East, West and
South), and their "distance". He could assign pre-
recorded tracks (neutral and loop) to these environ-
ments or import his own sounds. Finally, it is pos-
sible to declare localized sources as fountains (dif-
ferent types), playground, ball games, school (court-
yard, retracted or classes), cafes terraces, shops, bells,
public spaces call (languages), sounds steps on differ-
ent soils, sounds of wildlife and fauna (birds, wind in
the leaves), electroacoustic sounds (radio, TV, music),
mobile phone ringtone + voice, passage 2 wheels,
bus pass - tram - truck, boat passage, etc. A simple
algorithm should also introduce random elements in
sound environments: a passing car, a motorcycle, a
pedestrian conversation, etc. An urban rumour - like
a sound environment - from beyond the area and
composed with identity sounds (such as bells) could
be heard according to the block degree of opening
and the height of the listening point (distant sounds)
is also present.
Using the same process, we define three situated
sources with "Esquis'Sons", which are permeable re-
garding position, height and size in Rhinoceros /
Grasshopper (see Figure 4).
DESIGN TOOLS | Explorations - Volume 2 - eCAADe 34 | 281

Finally, Esquis'Sons! includes a "block scale reverb"
(developed as a parametric reverb) to make the dif-
ference from closed spaces with mineral surfaces and
wide open spaces with vegetation.The entire script
operation, starting with the sources positions and
two receivers, consists in a calculation of filters paired
with a mix made from geometric data only (dis-
tances) calculated and exported from the 3D model
in Rhinoceros, through Grasshopper.
Such specifications allow "to sketch a sound an-
swer to the questions listed above". The easy and in-
tuitive handling of these criteria then enables the de-
signer to validate the choices and / or to seek other
alternatives. The posture of a sketch is more relevant
than a simulation tool because of its versatility: it al-
lows a "first" act of representation, but also of design
and space sound design through an introduction to
experiencing the changing parameters and some re-
lationships between space, uses and sounds.
VALIDATION AND PROSPECTIVES
Different stages of tests led to a development and an
improvement of the tool and continues today, espe-
cially through educational workshops and with the
282 | eCAADe 34 - DESIGN TOOLS | Explorations - Volume 2
community of architects and researchers gathered Figure 4
around these teachings. The tool has also been vali- Situated sources
dated and improved by a return of use and field work assignation and
of the research members that has enabled its devel- listening points
opment and improvement. In this sense, through the with their “attached
whole process of construction and validation that we BLTC” in the 3D
have implemented, we tested with other users the re- model.
alism of the sound scenes produced. We also noted
user/designers desires in Esquis'Sons! about the op-
tions available when they were testing. These desires
are considered like architectural gestures to integrate
into the tool.
It seems important now to make the tool evolve,
in particular through the optimization of the module
and a better flexibility using Python, and to consider
a larger field of use for this tool (Generative potential,
inside sketches, etc.). We are also listening to partici-
patory impulses and perspectives related to the open
source status and to the "grasshopper community".
This tool - which is not a modelization or simu-
lation tool - is primarily intended to evoke a sound
situation. As a sketch, it escapes the question of fi-
delity of the model and metrological "reality". Consis-
tency of sonic results is validated by listening and by
the "realism" of sound productions thus generated.
This question of the sketch tool thus developed in the
course of the research "esquis'Sons! "And focused on
the interface between housing and outdoor amounts
to a new way of designing space. We're leaving the
post-project approval process and proposing solu-
tions, which become inherent in the design process.
These "technics" raise the issue of the use of the tool
in these processes including learning and design "by
doing" for the architectural discipline.
At the moment, we are exploring future uses and
issues that arise while using the tool, whether in tech-
nical or theoretical dimension so as to integrate the
tool to a continuous design process to allow it to con-
sider variable, adaptive and qualitative sound envi-
ronment, which lead to considering occurrences (Pi-
con 2011) rather than elements for the design of lived
space.
Figure 5
Split-Screen of the
tool during a
validation step : In
situ sound (top left)
, Sketch sound in
Esquis’Sons! (top
right), Model
definition in the
Grasshopper
module (bottom
right), Model
visualisation in
Rhinoceros 3D
(bottom left) –
Video : [2].

DESIGN TOOLS | Explorations - Volume 2 - eCAADe 34 | 283

 REFERENCES Saggio, A 2013, The It Revolution in Architecture. Thoughts
Augoyard, JF, Belle, O, Chelkoff, G and Balaÿ, O 1982, on a Paradigm Shift, lulu.com, New York
Sonorité, sociabilité, urbanité, CRESSON, Grenoble Vergnaud, G 2006, 'Rabardel Pierre \& Pastré Pierre (dir.).
Casati, R and Dokic, J 1994, Philosophy of sound, Editions Modèles du sujet pour la conception : dialectiques,
J. Chambon, Nîmes activités, développement', Revue française de péda-
Chelkoff, G 2005 'Architectural sound prototypes', Build- gogie. Recherches en éducation, 1(154), pp. 219-222
ing with sounds : proceedings 17th-18th march 2005, Vergnaud, G 2009, 'The Theory of Conceptual Fields', Hu-
Paris, France, pp. 20-24 man Development, 52(2), pp. 83-94
Chelkoff, G and Balaÿ, O 1987, Conception et usage de [1] https://tools.ietf.org/html/rfc768
l'habitat, CRESSON, Grenoble [2] www.esquissons.fr
Chelkoff, G 2011, 'Forme, formants et formalités : caté-
gories d', in Grosjean, M and Thibaud, JP (eds) 2011,
L, Parenthèses, Marseille, pp. 101-126
Ciblac, T, Guéna, F and Untersteller, LP 2005, 'De
l'esquisse d'architecture au modèle numérique', in
Ciblac, T (eds) 2005, Journées SCAN (Séminaire
de Conception Architecturale Numérique): Rôle de
l'esquisse architecturale dans le monde numérique,
Ecole Nationale Supérieure d'architecture de Paris-
Val-de-Seine, Charenton (94)
Luis Bento Coelho, J 2015, 'Approaches to Urban Sound-
scape Management, Planning, and Design', in Kang,
J and Schulte-Fortkamp, B (eds) 2015, Soundscape
and the Built Environment, CRC Press, pp. 197-214
Deleuze, G (eds) 2006, The Fold: Leibniz and the Baroque,
Bloomsbury Academic, London
Deleuze, G and Guattari, F 2005, What is Philosophy ?, Les
Editions de Minuit, Paris
Gibson, JJ 2011, The ecological approach to visual percep-
tion, Psychology Press, New York
Heidegger, M 1980, Essais et Conférences, Gallimard, Paris
Kleiner, M, Dalenbäck, BI and Svensson, P 1993,
'Auralization-An Overview', Journal of the Audio En-
gineering Society, 41(11), pp. 861-875
Leatherbarrow, D (eds) 2009, Architecture Oriented Oth-
erwise, Princeton Architectural Press, New York
Picon, A 2011, 'Architecture as performative art', in Grob-
man, YJ and Neumann, E (eds) 2011, Performalism:
Form and Performance in Digital Architecture, Rout-
ledge, London ; New York, pp. 15-19
Rabardel, P 2002, people and technology, université paris
8
Remy, N 2005 'Sound quality : a definition for a sonic
architecture', Twelth International Congress on Sound
and Vibration, Proceedings, Lisbon
Remy, N, Chelkoff, G, Marchal, T, Gamal Said, N, Bar-
dyn, JL and Amini, H 2015, Esquis’sons�! Out-
ils d’aide à la conception d’environnements sonores
durables,, CRESSON/ADEME Direction Villes et terri-
toires durables, Grenoble

284 | eCAADe 34 - DESIGN TOOLS | Explorations - Volume 2

SPATIAL REASONING AND
ONTOLOGIES
Abstract Object in the World of Data
A Play of Books - Symbolicity of Information as Potential for Architectural
Articulation

Miro Roman1
1
Chair for CAAD - Institute for Technology in Architecture – Swiss Federal Insti-
tute of Technology (ETH)
1
www.caad.arch.ethz.ch
1
roman@arch.ethz.ch

We are beyond representation; our abstract objects are symbolic; figures, fugues,
faces, masks, atoms, elements, characters, avatars, indexes. It is about infusing,
narrating, doping, context, information and masterful articulations. Concepts
become spectrums; they live like the memory or traces of things that have been;
they are not documents - they are animate. In this context interest of this paper is
to see how can an image of book in the world of data be different than Kevin
Lynch's systemic image of the city (Lynch 1960). Lynch has abstracted from the
physical city. He is representing cities on the level of text, grammars and
structures. What if one takes his exercise seriously and starts to play with text on
the level on information, data, lists and indexes? What is an image of a book, its
character, mood, how many faces does it have? This is a going to be a drama.

Keywords: information, text, image, abstract object, generic

PROLOGUE the city which are described as the most appropri-

"A clear and comprehensive image of the entire
metropolitan region is a fundamental requirement for
the future. If it can be developed, it will raise the expe-
rience of a city to a new level, a level commensurate
with the contemporary functional unit." (Lynch 1960).
In the context of cybernetics and systems theory,
capturing complex phenomena on an image is not
an unusual gesture. In his book The Image Of The
City, Lynch is trying to pin down the city on a map
by classifying five different elements: paths, edges,
districts, nodes and landmarks, and by using two
methods - expert opinion and interviews with local
people. Lynch finds stability in explicit features of
ate for the observed kind. A simple question that
one might ask Lynch is how could he have been so
sure that precisely these five features and these two
methods could produce a valid representation of the
city? Why not use more or less parameters and meth-
ods? Within networks, the representation becomes
of a different kind. Relation with the origin becomes
just one amongst many. What would happen if one
plays with relations on an algebraic level? Is there a
way of looking at things without explicating parame-
ters and categories but letting them play like a fugue
plays a certain theme? What if a play of books is con-
ceived as a fugue? It can be played as a spectrum.

SPATIAL REASONING AND ONTOLOGIES - Volume 2 - eCAADe 34 | 287


There is a certain objectivity - a theme. This theme is
being repeated, always in a slightly different way. It
is not mimetically tied to its representation, it is it's
offspring. A fugue inhabits the algebraic symbolic
space. In a similar manner, it would be challenging
to see how a figure of a book in the world of data
could be different than a systemic Image of the City.
As we have already mentioned, Lynch has abstracted
from the physical city. He is representing cities on
the level of text, grammars and structures. If we take
his exercise seriously, as a theme for a fugue, what do
we get? What happens if we want to play with text
on the level of information, data, lists and indexes?
What are figures and fugues of books? What is their
character, mood, how many faces do they have?
DRAMA
This is going to be a play. A fugue is going to be
staged almost like an ancient Greek comedy with its
actors, a stage and a stage play. This drama does
not have a moral. Its actors are computational - their
288 | eCAADe 34 - SPATIAL REASONING AND ONTOLOGIES - Volume 2
Figure 1
These are drawings
of books, or better
to say these are
faces of our actors.
Each cell is one
actor. Each actor
wants to distinguish
himself by showing
his most prominent
concepts (red and
yellow highlights).
character is algebraic. Although it is ambiguous what
a computational drama actually is, its stage script is
about engendering potentials and possibilities of ar-
ticulating abstract objects in the world of data. It
looks promising. In this play, books are going to be
abstract objects, finite objects in terms of letters and
words, but infinite in any other way: interpretation,
reading, meaning, translating, discussing... To be pre-
cise, books are going to be our Actors. Let us play
with text on the level of information.
Orthography - Actors - Pre-Specific Objects
Book, actors, libraries. Books are our actors. They are
getting ready for the stage. They are never alone,
they are always a part of some collections, libraries,
datasets, ensembles; they are moving in the move-
ment of the others. Actors are not specific - they are
of a symbolic kind. One can say that they have an
algebraic character which makes them opportunistic
to a meaning. Their mood is dependent on the library
they inhabit and their face changes when looked at.
Like in quantum physics, "when electrons (or light) are
measured using one kind of apparatus, they are waves;
if they are measured in a complementary way, they are
particles" (Barad 2012). Materiality itself becomes
undetermined. Double slit experiment challenges
the notion of predefined categorizations, which were
crucial for the development of Lynch's Image of the
City. New space is opened and we can finally pose the
question: How to describe an object with as many in-
dexes as one can? With no predefined categories, no
key parameters, no explicit grammars, and no rules,
one has to learn how to orient himself within the rel-
ative. In this constellation, each word or letter be-
comes an index. There are no more exceptions; ev-
ery singularity can be seen as a case. By doing so, we
are inverting notions of grammars, rules and param-
eters. Instead of a specific normality, indexes inhabit
a probabilistic universe. In this scenario, notions of
neighborhood and context become very important;
libraries that our actors populate, shelves that accom-
modate them, their friends and neighbors, all this be-
comes a part of their moods and manners. Our actors
act differently in different environments. The library
is the data that informs the faces of our actors. Books
are the actors and the library is their "good neighbor-
hood". They are entering the stage.
Scenography - Stage - The Generic Ground
The actors are entering the stage. Their faces are
motionless and without expressions. The play has
not started yet. In order to perform a play in a dig-
ital environment, they need to get dressed, put on
their masks, and learn how to articulate their infor-
mational faces. The stage is an apparatus, an in-
strument, a generic machine, a scenography. It is
a generic ground that is able to present actors in
multiple vector spaces. The stage is equipped with
available generic knowledge: Stanford CoreNLP - Nat-
ural Language Processing software (Manning et al.
2014), WordNet - lexical database of English language
(Princeton University, n.d.), RiTa - toolkit for experi-
ments in natural language and generative literature
(Howe 2009), Self-Organizing Map - a data process-
SPATIAL REASONING AND ONTOLOGIES - Volume 2 - eCAADe 34 | 289
ing algorithm (Kohonen 1982), implemented in Java
programming language. Libraries, books and ac-
tors are on one side of the double bind. The stage
is its complementary part, the infrastructure for ac-
tors; it measures them, curates their behavior, me-
diates their expressions, provides clothes and lights,
sounds and smells. Actors and stage, libraries and the
generic machine both serve as a double articulation
that presents informational faces of our actors. This is
a generic process of populating the stage with differ-
entiated instances of actors. Their faces can express
a multiplicity of realities, objectivities and interpre-
tations. They are not scientific, nor humanistic, but
computational and symbolic. An informational face
is not a map; it is a spectrum of different expressions
of the same face. It does not have specific features,
but it is indexed by as many indexes as one can find.
By doing this, each index relates to the whole library,
while each face becomes an expression of a book in
a library - a face in its ambient milieu. The subject of
the book is never explicitly in the book but in its in-
terplay with other books. Accordingly, a face of the
book is never determined, but always a part of a noisy
dramatic environment open to negotiations, consid-
erations, talks... In other words an actor can have dif-
ferent informational faces by indexing the text in dif-
ferent ways. These are the main indexing processes:
1. Curation is a process of normalization. e.g. to-
kenization, extraction of words, extraction of
letters, etc.
2. Mediation is a process of indexing the cu-
rated text in multiple measurable ways. e.g.
letter frequency, word frequency, bigram
frequency, trigram frequency, synonym fre-
quency, antonyms frequency, etc.
3. Measuring is a process of counting and relat-
ing of mediated indexes. e.g. frequency of a
word in a book in relation to all the books in
the library.
Here are some extracts taken from the examples of
informational faces. They measure how many times
a specific index appears in each book. The index is at

the same time a reflection of the whole library and a
measure of a specific book:
• Word frequency vector: building 10 24 0 45 82
39 0 609 18 15 21 50 29 60 1 0 85 11 249 40 33
49 16 233 72 53 31 2 0 26 1 75 179 121 1...
• Letter frequency vector: c 64758 86137 23964
10744 33271 26169 41685 32659 32708 27677
49217 18427 57030 21832 6937 6964...
• Bigram frequency vector: new_york 0 84 11 4
0 46 0 5 1 2 21 0 62 6 0 1 3 13 5 2 7 120 1 30 34
218 0 13 0 0 0 176 36 36 0 5 81 0 0 51 0 97...
Kinds of metrics one can apply to a text depend on
the richness of the stage. There are many of these
and they do not have to belong to the same kind.
Since informational faces are algebraic, they can be
transformed from words and letters, to synonyms
and antonyms, to Kindle and Google ratings, to taste
and moods of any intensity. Our probability space is
relative to that. It changes according to the libraries
we would like to read from, and according to the way
290 | eCAADe 34 - SPATIAL REASONING AND ONTOLOGIES - Volume 2
Figure 2
Galaxy of concepts -
In the background
of each character or
each face of a book,
is the context,
vocabulary of all
the books. Similar
indexes (words)
groupe together,
they start to form
concepts. This is a
conceptual
alphabet of our
library.
we read them. Knowledge and information become
relative to the way we look at them, just like in quan-
tum physics. Contradicting pictures do not exclude
each other anymore. On the contrary, they develop a
different picture all together (Barad 2012). We are be-
yond objective and subjective, practical and poetic.
As Eco would say, we are lost in the infinity of lists,
but we are ready for our stage play.
Ichnography - Play - Articulation
Act 1 - Abstraction 1a - Concepts. Actors are on the
stage. A play begins. What is happening on the stage
is not straightforward and intuitive. As Serres would
put it: "Objects, in the distance, change their skins,
they send one another kisses." (Serres 2000). There is
one process that appears consistent even from a dis-
tance: similar indexes attract each other; similar in-
dexes group together. The whole library is in motion;
concepts are negotiating and being negotiated. Self-
Organizing Map is an algorithm that articulates this
interplay of indexes. This stage play is a double ar-
 ticulation of actors and the stage, libraries and the
generic machine, data and models. This drama is a
comedy of appearances, and while images that we
are taking are snapshots of facial expressions of ac-
tors in an intense play (Figure 1). They are articulating
themselves by forming concepts - new abstract let-
ters of an alphabet - a cypher (Figure 2). How many
letters this alphabet has, what its resolution ends up
to be, is entirely up to us. Its consistency is provided
by the stage play. The newly formed letters are tem-
poral and can express any quality or concept what-
soever. This quality is never explicit and predeter-
mined; it is always a void in a cloud of indexes. On
the other hand, each letter has a specific explicit nu-
merical value for a given galaxy. It is a number and
a letter, a wave and a particle. So, what are the let-
ters from our library that are concerned with images
and text? If we look at the Concept 1160, we will no-
tice that it acquires the same interest (Figure 3). Let
us look at its Indexical cloud:
Figure 3
Zoom to concept
1160. To get a
subtle notion of
what this concept
might be about,
one has to get to
know its
neighbouring
concepts. It is an
image in a poetic
context.
• 1160e13_image, imagination, intimate, psy-
chologist, commonplace, exaggeration, fore-
word, underline, boredom, manor, psycho-
logically, smallness
and numerical vector:
• 0.0, 7.706139E-4, 2.568714E-4, 0.00480554,
5.137426194E-4, 0.00256304, 0.0, 0.0066784,
0.0048055... (one dimension for each book)
SPATIAL REASONING AND ONTOLOGIES - Volume 2 - eCAADe 34 | 291
To get a subtle notion of what this concept might be
about, one has to get to know its neighbouring con-
cepts which are indexd by:
• 1159e23_poetic, shell, nest, phenomenolog-
ical, daydream..., 1224e1_poem, 1225e2_-
greek, poet, 1161e3_memory, immemorial,
pedagogical, 1096e6_fragment, geometrical,
inflection..., 1097e16_surface, page, transla-
tion, vision...
Not far away from poetry but far enough, there is a
different mood for the notion of text indexed by:
• 771e13_text, signify, metaphor, articulation,
mute, structuralism, semiology... 707e8_ori-
gin, trace, criticize, detour, factual, ferdinand,
semiological..., 708e7_language, sign, indis-
pensable, rigorous, genealogy..., 834e10_lin-
guistics, script, logo, efface, anthropologist...
All this is a mood of a concept where image and po-
etry mix in a delicate way. It is interesting that there
was no predefined explicit grammar, no structure, no
rules or parameters, and these indexes grouped to-
gether and formed differentiated refined notions of
text and images. There are many moods that can be
inhabited by text. If we change the instrument of
looking, we can get a different spectrum.
Images and texts, paintings and poems, maps
and linguistics. If one looks at this specific constel-
lation of concepts and tries to find actors or books
whose faces are articulated by the same concepts,
one finds himself immersed in an amusing discus-
sion. There are three actors that distinguish them-
selves by their interest in this particular constellation
(Figure 4). Each one comes with its own attitude.
Let us give them space and listen to what they have
to say: Blanchot's The Book to Come is opening up
the scene with some personal concerns: "This first
of all: that there is no longer a limit of reference. The
world and the book eternally and infinitely send back
their reflected images. This indefinite power of mirror-
ing, this sparkling and limitless multiplication - which is
the labyrinth of light and nothing else besides - will then
 be all that we will find, dizzily, at the bottom of our de-
sire to understand." (Blanchot 1959).
Cache's Earth Moves The Furnishing Of Territories
replies by seeing Blanchot's concerns of limitless ref-
erencing of continuous transformations as a main
strength of his Objectile: "Just as Leibniz had con-
ceived it, texts, information, images, and sounds are
now all the object of numerical manipulation, so much
292 | eCAADe 34 - SPATIAL REASONING AND ONTOLOGIES - Volume 2
so that the electronic parts that make up the functional
core of the modern object no longer have anything to do
with the visual or auditory restitution that realizes their
concrete function." (Cache 1995). Deleuze's Difference Figure 4
And Repetition circles around the stage several times Faces of three
in smaller and greater distances from the three while books concerned
trying to make sense of Blanchot's concerns: "More with notions of
profoundly, however, the true Platonic distinction lies image and text.
elsewhere: it is of another nature, not between the orig- Their main focus is
inal and the image but between two kinds of images almost at the same
[idoles], of which copies [icones] are only the first kind, spot, but it is
the other being simulacra [phantasmes]. The model articulated in a
copy distinction is there only in order to found and ap- different manner.
ply the copy simulacra distinction, since the copies are
selected, justified and saved in the name of the identity
of the model and owing to their internal resemblance
to this ideal model." (Deleuze 1968). Curtain goes
down but the discussion continues... These are the
faces, their expressions, highlights, differences in col-
ors (Figure 4). Each actor wants to distinguish himself
by showing his most prominent concepts. Staged in
this artificial way, they provide interesting insights in
the relation between text and images. But there is a
trick. We can easily influence this discussion. If a book
is added or removed, the whole galaxy changes. If
we change the instrument of looking, or tune our
generic machine in a different way, the galaxy finds
itself in a new constellation and the actors change
their mood. Their faces become different, and we are
always taking new and different snapshots of them.
We slowly learn how they behave.
Act 2 - Abstraction 1b - Spectrums. Alongside with
negotiation between concepts, as the other part of
the double bind, actors themselves start to gather
and form galaxies of discourses. Since each actor
presents himself in a multiplicity of ways, he is a
part of multiple discourse galaxies. The same pro-
cess and same algorithm (SOM) are being used, yet
in an inverted manner. Mathematicians would say it
is a transposed matrix. Instead of indexing words by
books (Abstraction 1A), one is indexing books with
newly developed concepts (Abstraction 1B).
• Abstraction 1A (word; book1, book2,
 book3...) morphem 0.0078349, 0.0069416, Instead of words, books are those who are grouping;
0.001685852, 8.92502979E-4, 0.0086275... they are trying to find their place in the library. This
• Abstraction 1B (book; concept1, concept2, time they are arranged in a line in which the library
concept3...) Blanchot_The Book to Come becomes a spectrum (Figure 5).
3.182003E-4, 3.08664844E-4, 4.848412E-4... Act 3 - Abstraction 2a - Characters. Spectrums pro-
vided a new ground. This is a new level of abstrac-
Figure 5 tion. Actors can now abstract from concepts and fre-
Library arranged in quencies of indexes, given that their new faces are
spectrum. If one composed out of multiple analyses. Each analysis is
just scans through a specific view on the library - a spectrum. By relat-
this list, there is a ing spectrums, we are composing more abstract in-
consistency in the formational faces. Instead of describing a book by
way books inhabit 15000 values of word frequencies, we are describing
the line. Only by it with a combination of any chosen number of values
counting the words provided by the analysis. The way we look and how
and using a generic many instruments we use is not arbitrary, it is up to
algorithm, we have us to decide.
arranged the
• antonymsFrequency, antonymsFrequency,
library. It is
bigramFrequency, holonymsFrequency, hy-
articulated in its
pernymsFrequency, hypernymsFrequency,
own terms.
hyponymsFrequency, letterFrequency, syn-
onymsFrequency, synonymsFrequency,
wordFrequency, wordFrequency, wordFre-
quency, wordFrequency

Books are not vectors of concepts anymore, but vec-

tors of analysis.

• Abstraction 2 (book; Analysis1, Analysis2,

Analysis3, etc.) Blanchot_The_Book_to_-
Come 0.285 0.349 0.243 0.862 0.221 0.194
0.498 0.662 0.248 0.222 0.218 0.441 0.228
0.521

The same process repeats, yet it becomes more ab-

stract. Actors are choosing shelves according to their
mood; it is almost comparable to "making friends"
while agreeing on finding a specific place in the li-
brary - the one that suits them best. Now books, their
facial expressions, are indexing shelves. Each shelf
contains many books. The shelf is the new actor. So
how many shelves do we need for our current library
of 148 books? Not too many, not too few. Let us an-
swer 12 and detect what happens. Who are the new

SPATIAL REASONING AND ONTOLOGIES - Volume 2 - eCAADe 34 | 293

 actors? The aforementioned 12 actors describe and
index themselves according to the books they like to
read. It is a spectrum that goes from Swift and Asi-
mov, through Aureli and Cache, to Eco and Zlamea.
Nevertheless, we need to keep in mind that this is not
294 | eCAADe 34 - SPATIAL REASONING AND ONTOLOGIES - Volume 2
a library of all the books. It is the library of personal
interests and current preoccupations. Since this pa-
per is about images and text in the context of ar-
chitecture, we will kindly ask the four literary actors
from the library top - shelves 0, 1, 6, 7 to withdraw
and take their books with them, but to leave Shake-
speare's books since this is a play and his books might Figure 6
come in handy (Figure 6). Four actors have left the Library organized in
stage, but four new ones have entered it. There are 27 shelves. The shelf is
books less. The library is challenged each time a book the new actor.
is added or removed from it. The whole vocabulary
changes, concepts shift, indexes rearrange. New ac-
tors replace the old ones, which results in a more in-
tense and focused atmosphere. This is the appropri-
ate moment to ask questions and set the stage. Here
is a simple question:
• Which shelf offers a temporary home for this
alien text?
It is a fiction, an avatar, an alien book in the library.
The question serves as an index, as a new book. The
moment we ask this question, the whole library re-
arranges according to it. So what is the new milieu?
How have our actors arranged themselves? The shelf
two is its temporary home (Figure 7):
• 2 Cache Projectiles, Innis Empire And Commu-
nications, Blanchot The Book to Come, Play of
Books, McLuhan The Guttenberg Galaxy
These kinds of explorations become more like a mas-
terful playing of a violin than an objective analysis.
This is a self-referential process. Both the persona
and its neighbors look familiar. This constellation
is by all means a comfortable and challenging one.
McLuhan and Innis are pointing to the shifts in liter-
acy while anticipating coding as a new kind of liter-
acy (McLuhan 1962, Innis 1950). Blanchot is an enig-
matic driver of the discussion: "But the essence of lit-
erature is precisely to escape any essential determina-
tion, any assertion that stabilizes it or even realizes it:
it is never already there, it always has to be rediscov-
ered or reinvented." (Blanchot 1959), and Cache is the
one who relates the whole story back to the generic
 ground: "And this is how the new digital montages are The stage play does not provide one objective per-
created: no longer is a given sound coupled to a given spective; its semantics are not explicit, grammars not
image, as in the good old days of cinematography; in- visible; it is beyond dichotomies such as subjective -
stead, sounds are visualized or images heard in a chias- objective, nature - culture, art - technology. It is any
mus of perceptions." (Cache 2011). library and any method. It is up to the one that is us-
ing the library and his world-views to choose. It is a
Figure 7 symbolic compression to temporary atom-letters, an
Which shelf offers a endless process of articulations, motivated, personal,
temporary home yet operational. It is an operational stance on Eco's
for this alien text (A lists: a double articulation between a process and a
Play of Books)? The system, content and expression, books and instru-
shelf two is its ments. Instead of providing a solution or a generic
temporary home projection of certain data, one might be closer to a
for this alien text. personal algebraic projection of a certain discourse.

EPILOGUE
Coding as literacy. We are beyond representation;
our abstract objects are symbolic; figures, fugues,
faces, masks, atoms, elements, characters, avatars, in-
dexes. It is about infusing, narrating, doping, con-
text, information and masterful articulations. Con-
cepts become spectrums; they live like the memory
or traces of things that have been; they are not docu-
ments - they are animate. They do not have individ-
ual faces - they define zones of probability. It is a mul-
tiplicity of ciphering that makes them possible in ev-
ery sense and direction (Deleuze and Guattari 1980).
Similar examples can be found all around our world.
Brands inhabit symbolical spaces of myths; simulacra
are expressing a different environment populated by
differences which are not copies of a model (Massumi
1987). They do not merely represent, they have lives
of their own. Mathematics, especially algebra, does
not emphasize representation but rather the symbol-
ization of abstract concepts. It is not natural, but a
part of a specific nature. Programing languages show
us a nature different from natural languages. This
novel language is a language of noise and entropy.
It has left the archive and dwells on the Internet. The
question is not anymore how to classify the archive,
but how to articulate the generic notion of the web.
Noise and entropy are not peripheral any more, they
are the generic ground. We have once again inverted

SPATIAL REASONING AND ONTOLOGIES - Volume 2 - eCAADe 34 | 295

 the world. Instead of trying to find the basic laws of
nature, we are on a quest to discover pockets of life in
the entropic, that is, how to articulate out of the white
noise when it is observed to have a flat spectrum over
the range of probabilities relevant to the context. We
are in a search of most universal and abstract ingre-
dients in a given situation. We are not comparing
or deconstructing in the entropic, since everything is
on its way towards achieving a balance. In a world
where everything is connected, in which each actor
has many roles and can be rendered in many ways
(Figure 8), we are left with the question how to find
stability, how to make masterful articulations?
296 | eCAADe 34 - SPATIAL REASONING AND ONTOLOGIES - Volume 2
REFERENCES
Barad, K 2012, Karen Barad: What Is the Measure of Noth-
ingness: Infinity, Virtuality, Justice: 100 Notes, 100
Thoughts: Documenta Series 099, Hatje Cantz, Ost-
fildern
Blanchot, M 1959, The Book to Come, Stanford University
Press
Bühlmann, V 2014, 'Articulating a thing entirely in
its own terms Or: what can we understand by
the notion of «engendering» ?', in Hovestadt, L
and Bühlmann, V (eds) 2014, Eigenarchitecture,
Birkhauser Architecture, pp. 69-127
Cache, B 1995, Earth Moves: The Furnishing of Territories,
MIT Press
Cache, B 2011, Projectiles, Architectural Association
Deleuze, G 1968, Difference and Repetition, A\&C Black
Deleuze, G and Guattari, F 1980, A Thousand Plateaus, Figure 8
Bloomsbury Academic, London Multiple faces of
Eco, U 2009, The Infinity of Lists: An Illustrated Essay, Riz- Kevin Lynch's The
zoli, New York Image of the City.
Howe, DC 2009 'RiTa: creativity support for computa-
tional literature', C&C, p. 205 Same book in the
Innis, HA 1950, Empire and Communications, Oxford Uni- context of multiple
versity Press libraries, staged
Kohonen, T 1982, 'Self-organized formation of topolog- (measured) in
ically correct feature maps', Biological Cybernetics, multiple ways
43(1), pp. 59-69 shows us many
Lynch, K 1960, The Image of the City, MIT Press
Manning, CD, Surdeanu, M, Bauer, J, Finkel, J, Bethard, different facial
SJ and McClosky, D 2014 'The Stanford CoreNLP expressions.
Natural Language Processing Toolkit', Association for
Computational Linguistics (ACL) System Demonstra-
tions, pp. 55-60
Massumi, B 1987, 'Realer than real', Copyright, no.1, pp.
90-97
McLuhan, M 1962, The Gutenberg Galaxy, University of
Toronto Press
Moosavi, V 2014, 'Computing With Contex-
tual Numbers', CoRR, abs/1408.0889, p.
http://arxiv.org/abs/1408.0889
Serres, M 2000, The Birth of Physics, Clinamen Press
Architecture of Intermodal Complex
Sang Lee1
1
Delft University of Technology
1
s.lee@tudelft.nl

This paper focuses on the conception and design of architecture as the work of
producing media about buildings and other environmental artifacts. I approach
the questions regarding simplicity and complexity through "interdependence" and
"intermodality." I believe the two concepts offer more precise frames of relations
and contexts involving simplicity and complexity. I will first discuss the
complexity as a condition of interdependences and how today's interdependences
may provide a framework to understand complexity. I will then propose that
intermodality adds to interdependence a notion that specifically pertains to
today's media-driven culture and its complexity. I will next discuss how
dependences and modalities are interconnected at various levels and eventually
producing a new kind of semiosis that results from the disjunction between the
medium and the content. I will in conclusion propose a new concept
"apparatization" driven by interdependence and intermodality and how it
changes shape and remain fluid, rather than scaling between simplicity and
complexity, without a specific physical locus.

Keywords: apparatus, interdependence, intermodality, media, pervasive

computing

We may at first view the simplicity vs. complex-
ity contrast as a matter of degree and progression:
something starts as a simple, singular entity and pro-
gresses into a complex one, gaining multitude of
qualifications and variations. We can also take for ex-
ample the pace and scope of such progression and
how fast and pervasive it may eventually become. As
often, we break down a complex object or situation
into simple constituents so that we can understand
it clearly. In a reticulate formation, complexity in
essence indicates a context of relationship and con-
nectedness. Especially since the emergence of the In-
ternet, not only the expanse but also the frequency at
which the reticulate configuration operates has come
to dominate human affairs. In short, the questions
regarding simplicity and complexity hinge on how
rapidly and pervasively the reticulate configuration
operates.
Against the backdrop of reticulation, interdepen-
dence includes the performance-critical, contextual
elements, the prerequisites, that are necessary for an
object or an event to take place. Certain prerequi-
sites bring about radical changes that propel the de-
velopment of human culture at an astonishing rate.
In the context of today's Internet-driven cultural mi-
lieu, Java, the W3 standards, fiber optic infrastructure
and the smart, networked mobile devices for exam-
ple stand out as such prerequisites. Each of them

SPATIAL REASONING AND ONTOLOGIES - Volume 2 - eCAADe 34 | 297

 in turn carries its own prerequisites. Such prereq-
uisites - the so-called "general purpose technology"
(Brynjolfsson and McAfee 2014: 75-76) regardless of
simplicity or complexity in itself - brings about pro-
found changes to a vast swath of human culture,
thereby greatly increasing the complexity of artifacts
and knowledge systems. A general purpose technol-
ogy includes: previously existing technologies; activ-
ities and processes involved in the production and
accumulation of artifacts; and contextual knowledge
systems (Bijker et al. 2012: xli-xliii). Each event of
general purpose technology indicates a critical mile-
stone that substantially elevates the performance of
human material culture. In this sense, it can also be
called the performance-critical technology. Due to its
transformative influence and near-universal deploy-
ment, performance-critical technologies - such as in-
ternal combustion engine, electricity, transistors, and
so forth - have become critical to the ontology of hu-
man constructs. Without them, the human culture as
we know it may cease to exist. We can add the digi-
tal algorithm and computer technology, especially in
combination with the Internet and the W3 standards
[1], to the performance-critical category.
Performance-critical technology prompts devel-
opment and production of other technologies, and
brings about tangible, very often fundamental and
transformative, effects in the cultural production pro-
cess. Computation exemplifies the performance-
critical dimension. (For the purpose of this paper,
I will sketch "computation" as the combined func-
tion of: the hardware as an instrumental assemblage;
the software as aggregation of algorithmic encod-
ing languages and expressions; and the activities and
processes involved in making, distributing and con-
necting the hardware-software functionalities.) Such
general purpose, performance-critical technologies
always become pervasive and almost universal. In
addition, they are almost always recombinant of the
prerequisites. They indicate a creative way of under-
standing and combining existing technologies in or-
der to tackle and solve problems and so as to en-
gender subsequent technological innovations that
298 | eCAADe 34 - SPATIAL REASONING AND ONTOLOGIES - Volume 2
impact human living and its environment. The in-
terdependencies among various performance critical
technologies define the nature of complexity in hu-
man culture.
In the meantime, intermodality has come to mir-
ror interdependence and consists of interchangeable
modes of media and agency, for example, that may
flow from a piece of paper to a computer screen, to
a printer, and back to paper. It indicates the media-
devices that can hold and present numerous types
of contents, and are used to access them, ranging in
scale from a wrist watch to a smartphone, to a tablet
to a computer, to a billboard and to an entire build-
ing façade. Each media-device presents a particular
modality because of its own distinctive mode of exis-
tence and operation in hardware (the machine) and
software (the encoding) combination. Creating con-
tents that can seamlessly traverse from a wrist watch
to an electronic building façade, with all the interme-
diate scales of display and reception, represents the
quintessential state of intermodality.
According to Mark Weiser, who proposed
"ubiquitous computing" and "embedded virtuality"
(Weiser 1991), computing should be as common as
paper in order to achieve its full potential. Along the
way, we have reached a point where simplicity in fact
represents - or symptomizes - incredibly complex in-
terdependent systems that are indeed pervasive. We
have come quite close to Weiser's ubiquitous "tabs,
pads and boards" (Weiser 1991:98). As a result, in
order to make data available (almost) everywhere,
intermodality has become absolutely crucial: it is no
longer a matter of choice. No one any longer cre-
ates the kind of content dedicated to one medium
(mode) of presentation and pragmatics. For exam-
ple, Nokia's and subsequently Blackberry's demise
demonstrates today's intermodal technological mi-
lieu. Both companies are no longer viable because
they chose to ignore the wave of new intermodal ap-
paratuses initiated by Apple with its iPhone. Around
the time of the W3 standards, America Online (AOL)
used to be the most dominant Internet service. It em-
ulated a walled-in community where AOL controlled
the communication between its members as well as
the media content and distribution according to its
own codification. Now AOL is no longer viable be-
cause it was still modeled after the protected, closed
model of television programs and movies. In today's
pervasive intermodality model, Facebook is open to
various modes of media ranging from text to video
on demand with minimal control and enforcement
of membership and participation. Facebook can very
well monetize its thumbs-up icon (U+1F44D) or a
birthday cake icon (U+1F382), intermodality par ex-
cellence, without the need to collect membership
fees.
From a causal perspective, simplicity, beyond su-
perficial appearance, almost always carries a com-
plex matrix of dependences. For example, a simple-
looking terracotta bowl implicates a series of prereq-
uisites: the availability of clay in a reachable area; the
tools and labor for collecting and transporting the
clay; the knowledge of suitable sizes and shapes (e.g.
a cup, a bowl or a jug); the knowledge and skills of
tooling and shaping the mud into such a shape; the
ability to generate and calibrate heat for baking; and
the like (Hodder 2012: 17-18). In turn, each prereq-
uisite also carries a series of requirements for it to
take place. In this matrix of prerequisites, each ele-
ment is crucial to the ontology of a given object, be
it a bowl, a building, a computer or an airplane. To
cite a more recent example, beneath the intermodal-
ities of graphic user interface environment is a long
series of interdependences that folds into itself in a
web of disparate prerequisites ranging from encod-
ing languages to lithium. The semiosis of technologi-
cal codification arises from a complex of Babelian lan-
guages and material conditions. Through the codifi-
cation of isolating, tagging and transposing individ-
ual elements, the encoding languages connect with
one another and help propagate semiotic contents in
highly affective, preferably alluring, form.
The matrix of prerequisites and dependences is
derived from the techné in the classical sense of the
term of Four Causalities in material, form, intent and
purpose (Aristotle 1999). The question of simplic-
SPATIAL REASONING AND ONTOLOGIES - Volume 2 - eCAADe 34 | 299
ity vs. complexity in the causal techné depends on
how tightly or loosely the four causalities rely on
one another. Each individual causality also embod-
ies its own set of dependences that aggregate into
an ever-increasing snowball that fluctuates in shape
and structure. In today's context of computational
technologies, interface exemplifies the complexity of
intermodality that facilitates the transition between
different sets of interdependences. While we may
consider it simple enough to use a laptop computer
or a smartphone, the seeming simplicity and the ease
of use are made possible by the interface. The func-
tion of interface depends on the culturally accepted
semiotic understanding of a given task in relation to
the sign that represents it. However, the underlying
causalities of simple - often culturally specific skeuo-
morphic - interface is nothing but simple from both
iconographic and engineering points of view.
From a teleological perspective, any given de-
sign must conform to the way it is understood and
used: the interface based on existing, widely recog-
nized sign-systems becomes crucial. In architecture,
we speak of the "duck" versus the "decorated shed"
(Venturi et al. 1972: 88-91). A "duck" object is an icon
in itself of which ontology is limited to its iconic role.
It has no other modality beyond the iconic presen-
tation of its uniqueness regardless of its functional
purpose. A decorated shed is a building that carries
a sign "Duck" but not necessarily unique in itself in
relation to what the sign signifies. In this case, the
functionality of the building assumes a new mean-
ing by virtue of the sign, the interface, that is applied
to the building. The building as a functional object
becomes recessive and may underlie other modalities
depending on the interface, the sign system, that is
expressive. We can add the third category to the duck
and the decorated shed. Since the advent of digital
image making, the interface as sign-system has be-
come and must be intermodal: one should be able to
understand the cognitive assemblage in a consistent
way regardless of the environmental context and its
variables. Such intermodality requires simplicity in
expressing the operative logic of a given construct in
 order to facilitate efficient and effective semiosis. We
may call this condition the "decorated ducks" (Fos-
ter 2001: 15). This category of objects includes not
only the iconic semblance (e.g. a duck, a fish or some
thing that looks like crumpled paper or a bundle of di-
sheveled spaghetti), but also the embellishment that
reinforces the icon.
The causal and teleological perspectives have di-
verged from each other: the medium is no longer
the message and the message is no longer specific to
the medium. More importantly, the medium and the
content have become separate: they are no longer
bound by the historical conventions of congruity we
often summarily call "media." To paraphrase Friedrich
Kittler, medium is irrelevant: the medium-specific ex-
pression neither exists nor appears viable any longer
(Kittler 1999: 2).
The separation between content and its
(re)presentation or (re)presence, the algorithm-
driven media disjoin the content from its means of
presentation, and even from presentability, thereby
rendering the eventual pragmatics of media-content
problematic. The development of encoding lan-
guages such as Java, HTML and XML augmented the
pervasive deployment of the Internet and the W3
standards, as well as the kind of application software
that runs on any hardware-system software combi-
nation. They explicitly emphasize intermodality by
facilitating various types of information to appear
in a consistent manner across various hardware and
software environments. Today's digital media are
distinctive in their capacity to span different types of
devices, change shapes, and shift places, depending
on how the content is encoded, codified and pre-
sented.
The new encoding languages makes it possible
to assemble a large amount of information in a way
that is not determined by temporal and geographi-
cal bounds. The encoded and codified database has
led to a new construct of knowledge that is based
on the values based on exchangeability. Thus, the
digital media make the one-to-one correlation be-
tween the physical presence and its content highly
300 | eCAADe 34 - SPATIAL REASONING AND ONTOLOGIES - Volume 2
problematic, even impossible. Yet, the systemic dis-
junctions and unforeseen slippages (or bugs) inher-
ent in digital media also make it possible to work with
and manipulate various kinds of content in ways that
would have been impossible without the capability
to separate content and presentation. Our daily in-
teractions with and through digital devices consist
of modulations between our explicit action and the
contingencies that occupy the disjunctions of such
modulations.
The pure form and the configuration of the for-
mal and programmatic semiosis that can be tran-
scribed and extrapolated endlessly on-demand reign
supreme. The separation of the content and the
medium has also brought about fragmentation, re-
contextualization and reconfiguration, which render
the notions of originality and authenticity by and
large moot and irrelevant. The separation in ef-
fect produces "organs without a body" (Zizek 2004:
172-173). Alluring infonemes (minimally meaningful
units of distinctive sounds, images and texts) are har-
vested, embellished and hustled in a way not unlike
the fetishized pornographic images of silicon-filled
breasts, a glatt-depilated vagina or a bleached anus
at the expense of the body as a whole. The aggrega-
tion of excessive infonemes reinforces and tribalizes
authority and power, contrary to the idealism that
technology usurps the status quo.
In the twentieth century industrial machines
augmented and to a great degree replaced physical
human labor. With the proliferation of computing,
our cognitive capacities are also apparatized, out of
our body. We are tethered to various cognitive ap-
paratuses in order to augment our capacity to deal
with complexities and to relieve our mind and body
of labor. The extent of such extra-cognitive appa-
ratization determines the degree of simplicity and
complexity. The more extra-cognitive we become,
the more complex our environment is. We have also
come to equate extra-cognitive capabilities with em-
powerment and to a large extent freedom: the more
extra-cognitive, the more power to freedom. In ad-
dition, the extra-cognitive capacities connect with
one another and have become pervasive, the me-
diated socialization. Here I adopt the archaeologist
Lewis Binford's classification of "technomic" (direct
extension and augmentation of human body; e.g.
a hammer extending human arm and fist), "socio-
technic" (a tool becomes common to a group of peo-
ple; e.g. everyone has a hammer), and "ideo-technic"
(a tool that loses its original function; represents a
specific ideological or belief system; and is used to
reinforce such a system; e.g. the hammer-and-sickle
that represents Communism) (Binford 1962). The in-
termodal, extra-cognitive technology turns "socio-
technic" and produce apparatus-regimes that em-
powers the "ideo-technic." In this case, the notion
of simplicity and complexity depends on the degree
to which a given knowledge system is apparatized
by the dominant power; how the apparatus-regime
fabricates semiosis through ideo-technics; and how
such semiosis is embodied (or actualized) by the sub-
ject.
Within the apparatization process, while certain
individuals may (appear to) exercise certain freedom
of various personal choices, such freedom is defined
in diagrammatic, combinational, and configurational
ways. It depends entirely on the potentialities and
more importantly on the exclusionary processes that
an apparatus-regime chooses to exercise and en-
force. The apparatization process has caused a dis-
ruptive shift in architecture as a discipline in the cul-
ture of pervasive algorithm and computing.
First, the disciplinary apparatization indicates
that autonomy and authenticity become irrelevant.
An apparatus-regime by way of its codification sys-
tem define the apparatus-centricity and its genera-
tive capabilities. The codification system molds and
shapes the functioning of procedures and protocols,
and thus determines the operativity of a given disci-
pline's constituent agents and contingencies.
Second, apparatization assumes incremental de-
velopment, in which variation and combination
emerge as its primary operative modes. This in turn
intensifies the decentralization and fragmentation of
parts and production, and makes the re-combination
SPATIAL REASONING AND ONTOLOGIES - Volume 2 - eCAADe 34 | 301
and re-versioning the most crucial aspects of compo-
sition. This is primarily represented as the flexibility of
design and renewal, the "flexible accumulation" (Har-
vey 1989: 147).
Third, the apparatization neutralizes media
specificity by means of underlying codification sys-
tems (algorithms+hardware+networks), and there-
fore becomes fluid, more horizontally distributed,
and intermodal. What used to be known as medium
in the historical sense, the substrate, no longer re-
quires, or at least no longer assumes, material, physi-
cal actualization. One kind of content may - is in fact
required to - be easily transcribed and transformed
into another kind.
Apparatization superimposes its own operative
logic and discipline, affecting the work regardless of
the authorial intent. Such apparatization anticipates
particular functions or tasks within certain means in
order to address a need and serve a purpose. In ar-
chitecture for centuries, the operative logic has been
projective geometry: drawings are constructed by
projection as plans, sections, elevations, and per-
spectives. With the advent and proliferation of digi-
tal algorithmic apparatuses, architecture has become
a thoroughly apparatized, extra-cognitive practice.
Its disciplinary field has come to include an increas-
ingly expansive array of elements, modalities, and at-
tributes to the extent that the historical autonomist
view of the discipline appears no longer viable. Archi-
tects have become a new class of ideo-technical pro-
fessionals whose work involves dealing with the com-
plexities of the extra-cognitive apparatuses and the
types of content that algorithmic apparatuses help
produce and promote. Architecture is also increas-
ingly regarded as an apparatus-driven image-making
practice on behalf of the dominant regime of one
kind or another.
Architects' excessive optimism and reliance on
extra-cognitive, intermodal apparatuses often result
in a bipolar disorder: the simplex of formal euphoria
mated with the complex of overt techno-optimism.
One the one hand, the complexity and density of
apparatuses have increased exponentially, ranging
 from robotic construction to turning lights on and
off. On the other, the very same apparatuses re-
quire and are limited to efficient simplicity in cul-
tural and economic semiosis. No one knows for sure
what lies behind the kind of simple, almost reflex-
ive use of our apparatuses. The excessive enthusi-
asm for and over-reliance on the apparatus disregard
the technological apparatuses' complex automatic-
ity and codification. The necessity of generative, for-
mal rationality depends largely on the exclusive codi-
fication by means of software-hardware-network ap-
plications and systems. Such codification is designed
to perform and produce a specific set of affects imple-
mented by yet another kind of author, the program-
mers, and what they choose to address and make
visible. The apparatization has come to determine
the fate of the actual performance, be it music or
architecture. The emphasis is on fluency in specific
skills, vis-à-vis specific apparatus-centric rationality
and knowledge. Additionally, in the conception and
production of architecture today we see a new class
of experts whose central role is dedicated to the cod-
ification and operation of the apparatus. If we reflect
on recent tendencies of both architecture and music,
specifically since the appearance of the first purely
electronic sound generation in the 1950s, we also see
a process of codification, in the sense that the envi-
ronment in which the sound occurs is constructed
on the basis of rigorous technical operations. But
the performance is detached from the presupposi-
tion of substantive narratives, or from mundane and
serendipitous moments.
The roles of extra-somatic augmentation and
extension of human sensory capacities define the
modes of mediatized cultural formation. We may also
consider the extra-somatic formation of aura. In Wal-
ter Benjamin's notion of mechanical reproducibility
(Reproduzierbarkeit), the machine overshadows the
artist (Benjamin 2010). Ultimately, the kind of ma-
chines involved in the (re)production of the works of
art determines and embodies the aura. Such a me-
chanical aura represents the industrial-capitalist cul-
ture that consists of identical, mass-produced, and
302 | eCAADe 34 - SPATIAL REASONING AND ONTOLOGIES - Volume 2
flawless objects that are inexpensive and available
to everyone at an appropriate price. We endow
machine-made objects with an aura of seeming per-
fection and equity. Whether they are reproductions
or original work does not matter much: machines
made them, and the machines cannot waver. The
sense of enchantment and magic of technology and
machines still puts us under a spell. If we were to sup-
pose what an aura may indicate in the digital age, the
question is not only whether or not it is present in
the extra-somatic digital apparatus. But it pertains
more crucially to how it displaces and replaces the
existing codification of cultural discourse and forma-
tion. We find comfort in the cultural discourse of the
digital apparatus and codification in the way it masks
the messiness of reality. The aura of artistic work val-
orized by today's cultural dispositif stems from the
pervasive and ubiquitous presence of the digital ap-
paratus. We can paraphrase Benjamin that today the
aura surrounds the sense of a cult that is so intimate,
yet remains distant and invisible. If the machine age
was thought to have rid art of cultic and ritualistic
valorization, the digital age has resurrected it with
vengeance.
The cult of the digital may also celebrate disso-
nance and alterity by capturing and making them vis-
ible, lifting them to the surface of perception. The cel-
ebration of dissonance and alterity is embedded in
the tradition of technological avant-garde that hails
novel inventions as the progenitors of creative ag-
gression. But as soon as they are captured, the cult-
novelty withers away as quickly as it was fabricated.
The mediatization process no longer involves physi-
cal heft. It is no longer a question of vor- or zuhanden,
to borrow from Heidegger. Encoding and codifying a
given work determines its viability and eventual on-
tology. Regardless of the content, the surface ap-
pearance determines the cult-value of a given work,
and whether or not the work may be allowed to fit in
the striations of the virtual space.
Whether or not any new codification system may
afford yet another promise of liberation, empower-
ment and freedom is highly questionable and re-
mains to be seen. At the same time, such techno-
logical automaticity provides architecture with the
potential that is unprecedented in terms of its vast
heterogeneous spectrum that now touches almost
all aspects of human culture. Inserting the idea
of various local contingencies and noise into com-
putational aesthetic work process will help over-
come hollow manifestos and naïve panacea that of-
ten result only in the thoroughly forgettable images
of pornographic contrivances or chewed-up bubble
gums. Counteracting the dispositif , contingency, dis-
sonance and alterity should form intrinsic part of any
apparatus-centric aesthetic work.
We may find it assuring and comfortable to be
able to mold our imagination into tangible shape.
Along with the new form of empowerment leading
imagination directly to factory, the combination of
technologically encoded aesthetics and algorithmic
conception and composition dematerializes works of
architecture as their conception, design and appre-
ciation are apparatized in an extra-cognitive way. In
this process, engrossed by the seeming simplicity at
the receiving end, architecture loses the political, so-
cial and cultural potency and criticality it is expected
to embody. What remains is the impression of ex-
perience that turns into yet another agent for com-
moditizing excesses. The algorithmic media afford
the new capability possible only through apparatiza-
tion and codification. We may criticize and accuse ap-
paratization of pulverizing the inviolable subjectivity
of the architect-as-author, or even of the author at
large. We may also mourn the demise of such au-
thor and attribute it to the new apparatus and cod-
ification regime, of which purpose is thoroughly cap-
tured by the economic and programmatic maximiza-
tion of what may be viewed as "cognitive capitalism"
(Moulier-Boutang 2011: 50-59). However, the poten-
tial for dissonant, unsettling alterity to rise above the
surface of our pervasively augmented and sanitized
consciousness is also as compelling as the power-
authority to capture and objectify it. The capabilities
gained from the assemblage of new apparatuses and
codifications should be liberating in its very potential
SPATIAL REASONING AND ONTOLOGIES - Volume 2 - eCAADe 34 | 303
to expand and intensify alterity, in resistance to the
ideo-technic of power-authority that is deluding and
oppressive.
REFERENCES
Aristotle, . 1999, Nicomachean Ethics, Hackett Publishing
Co
Benjamin, W. 2010, The Work of Art in the Age of Mechan-
ical Reproduction, Prism Key Press
Bijker, W.E., Hughes, T.P. and Pinch, T. (eds) 2012, The So-
cial Construction of Technological Systems (Kindle Edi-
tion), MIT Press
Brynjolfsson, E. and McAfee, A. 2014, The Second Machine
Age: Work, Progress, and Prosperity in a Time of Bril-
liant Technologies, W.W. Norton
Foster, H. 2013, The Art-Architecture Complex, Verso
Hodder, I. 2012, Entangled: An Archaeology of the Rela-
tionships between Humans and Things, John Wiley &
Sons
Kittler, F.A. 1999, Gramophone, Film, Typewriter (trans. Ge-
offrey Winthrop-Young and Michael Wutz), Stanford
University Press
Moulier-Boutang, Y. 2011, Cognitive Capitalism (trans. Ed
Emery), Polity Press
Venturi, R., Scott Brown, D. and Izenour, S. 1972, Learning
from Las Vegas, MIT Press
Weiser, M. September 1991, 'The Computer for the 21st
Century', Scientific American, 265(3), pp. 94-104
Zizek, S. 2004, Organs without Bodies: On Deleuze and
Consequences, Routledge
[1] www.w3.org
Ontologies and Shape Grammars
A Relational Overview Towards Semantic Design Systems

Rui de Klerk1 , José Beirão2

1,2
Faculty of Architecture - University of Lisbon
1,2
{ruideklerk|jnb}@fa.ulisboa.pt

This paper provides an overview on the relation between computational

ontologies and shape grammars regarding the development and production of
multi-purpose Semantic Design Systems. The objective of the author's ongoing
research is to assist the creation of generative design systems, applicable to
design processes in general. Shape grammar rules and ontologies in these
systems will be focusing on abstract, generic rules and generic descriptions.
When combined through contextually specified relations, these assume semantic
expressions and should be able to produce meaningful results.We collect here a
short state of the art of the research developed in the fields of architecture,
urbanism and computer science in the past ten years regarding the use of
knowledge bases (ontologies) combined with generative design systems (with a
particular focus on shape grammars). We expect to provide both insight about
architectural and urban typologies and the production of meaningful designs
using automated generative design systems.

Keywords: Ontologies, Shape Grammars, Semantic Design Systems,

Architectural Design, Urban Planning

INTRODUCTION use of ontologies and shape grammars.

The main objective of the research presented in this
paper is the production of multi-purpose Semantic
Design Systems that may be applicable to design pro-
cesses in general. The research aims at defining a
generative system that, using simple basic shapes
and generic grammar rules, when combined with on-
tologies (adding specific meanings and relations to
shapes) may be able to generate meaningful designs
in the particular design domain defined by the ontol-
ogy. In this paper we show the preliminary stage of
surveying the present knowledge on the combined
Generative design systems, namely shape gram-
mars (Stiny and Gips 1971), allow the semi-automatic
creation of a large quantity of varied designs. These
designs are generated in a procedural fashion,
through changes in parameters or the iterative ap-
plication of a limited set of simple production rules,
resulting in varied set of designs which define the
grammar's design space.
Despite being able to generate large quantities
of syntactically correct designs, generative design
systems do not guarantee the designs are semanti-

SPATIAL REASONING AND ONTOLOGIES - Volume 2 - eCAADe 34 | 305

 cally correct. This leads to the production of unfeasi-
ble or undesired solutions.
Semantic accuracy can be achieved through the
combination of generative systems with knowledge
bases, such as computational ontologies (Gruber
1993). According to Grobler et al. (2008), "the combi-
natory nature of the design rules captured by shape
grammar and ontology offers the possibility that de-
sign knowledge can be explicitly represented, main-
tained and processed."
In this paper, we will begin by looking at mecha-
nisms introduced over time to control shape gram-
mar derivations, especially those attempting to in-
troduce semantic control over shape rule applica-
tion. Over time, these mechanisms have been com-
plementing the shape grammar formalism shifting
this type of generative systems towards more com-
plex compound grammars. (Knight 2003)
Later, we will move towards the combination of
generative design systems based on shape gram-
mars with ontologies. We will be looking at six ap-
proaches by different authors regarding both archi-
tecture (physical or virtual) and urbanism, over a pe-
riod of time that spans from 2006 to 2015:
• Shape rules as descriptions for designs, repli-
cated in natural language to create an ontol-
ogy for a type of buildings (Andaroodi et al.
2006);
• Multi-level projection ontology design ap-
proach to create 3D models of Chinese An-
cient Architectures (Liu et al. 2006; 2008), us-
ing approximation theory of granular com-
puting to enhance its accuracy and perfor-
mance (Liu et al. 2010). These approaches
were later formalised as the Onto-Draw frame-
work (Liu et al. 2012);
• Ontologies working in parallel with shape
grammars, informing shape rule application
through a lightweight BIM, to characterize ex-
isting buildings (Grobler et al. 2008; Aksamija
et al. 2010);
• Ontologies defining the relevant concepts for
a generative process, shape grammars as de-
306 | eCAADe 34 - SPATIAL REASONING AND ONTOLOGIES - Volume 2
sign generators, heuristics guiding the gen-
eration process and validations as a mecha-
nism to test and evaluate designs (Trescak et
al. 2010);
• Ontologies as a common representation
protocol between different design modules
(Beirão 2009; 2012);
• Ontologies as meta-models for generative de-
sign, using description rules to guide shape
rule application (Stouffs and Tunçer 2015).
Also worth mentioning is the work of Tutenel et al.
(2011), a framework to procedurally create buildings
for computer games with consistent interrelated in-
teriors and exteriors. In this case, the knowledge
base is represented and stored in a purpose-built re-
lational database, not an ontology.
CONTROLLING SHAPE GRAMMARS
Generative design systems based on shape gram-
mars allow designers to work with large quantities of
solutions within a design language, testing multiple
outputs in search for the one(s) that meet the desired
criteria (Stiny and Gips 1971).
Shape grammars are a formal rewriting mecha-
nism which applies shape rules (α → β) algorithmi-
cally replacing a predicate condition, an Euclidean
transformation of a shape (t(α)), found in a design
(C), by a consequent shape defined under the same
transformation (t(β)), so that: C' = C - t(α) + t(β).
One issue that arises when working with shape
grammars is related with the system's ability to pro-
duce valid solutions. This is a particularly important
feature in automated systems, where they may pro-
duce large quantities of syntactically correct deriva-
tions of a shape grammar but which are, in the
end, semantically incorrect and therefore unfeasi-
ble or undesirable. Therefore, providing generative
systems with semantic support tools able to distin-
guish the semantically robust from the semantically
meaningless derivations is a mandatory procedure
on any implementation aiming at developing a se-
mantically robust generative design system. The
semantic difficulties in relation to shape grammars
have been expressively pointed by Fleisher (1992)
even after some of the semantic formalisms devel-
oped for shape grammars had already been devel-
oped.
To regulate shape grammar derivations, some con-
trol mechanisms were introduced over time. From
the very beginning, with the definition of shape
grammars, labels were introduced to control the iter-
ative application of shape rules (Stiny and Gips 1971).
Labels can be used, for instance, as symbols associ-
ated with shapes on the left-hand side (α) of a shape
rule (α → β), breaking the symmetry group of (α)
and, therefore, conditioning the number of possible
solutions (comparing with unlabelled shapes) (Stiny
1980). In this case, labels are used as spatial relation
markers. Labelled shapes can be used to reduce am-
biguity in shape rule application or distinguish shape
rules in the construction of different parts of a de-
sign. In some situations, labels add a particular classi-
fication to a shape. For instance, in Duarte (2001) la-
bels are used to distinguish types of space according
to their functions, a "kitchen" as something different
from a "bedroom" even though they might be both
represented by a rectangle. In this case labels work
as classifiers.
Description grammars followed, operating in
parallel with shape grammars and constraining the
application of shapes and symbols with description
rules (Stiny 1981). After that came colours (Knight
1989), weights (Stiny 1992) and discursive grammars
(Duarte 2001). Discursive grammars combined shape
grammars, description grammars and heuristics in
order to give a meaning to the derivation of solutions
based on a design brief.
Later, Haldane Liew (2003) proposed a Shape
Grammar Meta-Language (SGML), creating layers of
abstraction in the shape grammar formalism and de-
veloping descriptions that can organize and control
the application of rules. The rule application pro-
cess is expanded into six phases: control, context,
transformations, parameters, descriptors and applica-
tion. Particularly interesting in the SGML approach
are the descriptor, the context and the control phases.
SPATIAL REASONING AND ONTOLOGIES - Volume 2 - eCAADe 34 | 307
These three phases allow the conditioning of rule ap-
plication to shapes that are only composed of maxi-
mal lines and/or have void spaces (descriptor), for in-
stance, filtering specific labelled shapes or focusing
on a particular shape when applying a rule (context),
by linking a series of rules together forming macros
or by determining the availability of a set of rules at
any point in the grammar derivation (control).
Graph grammars may also be used as a control
mechanism, working in parallel with shape gram-
mars to ensure the maintenance of topological re-
lations between shapes (Grasl and Economou 2010;
2013). Graphs can also be used to "group rules ac-
cording to the topological transformations involved,
without considering specific shapes". (Eloy and
Duarte 2011)
When the rules in the abovementioned gram-
mars are linked between them, constraining each
other's application, they become compound rules
and the grammar can be designated as a compound
grammar. (Knight 2003)
These control mechanisms impose restrictions to
the application of rules in a shape grammar, being
able to provide some semantic control - by associat-
ing qualitative labels with descriptions, for instance.
By using such mechanisms, shape grammars con-
tain knowledge about families of designs and how
these can be generated. This knowledge is not, how-
ever, explicit or easy to work with, extend or maintain.
(Duarte 2001; Beirão 2012).
As we will see in the next section, knowledge re-
garding families of designs (namely building types
and typology) is better represented using knowledge
bases such as ontologies.
Ontologies can be described as "an explicit spec-
ification of a conceptualization" (Gruber 1993), struc-
turing information using taxonomy (classes and sub-
classes: "is a") and meronomy ("part of ") relations.
This allows ontologies, through the use of reasoners,
to infer relations which were not explicitly stated be-
tween its elements, according to their object or data
properties. Also known as knowledge-based mod-
els, ontologies are perfectly suited to contain infor-
 mation about architectural types, building designs or
the significant relations among different types of ob-
jects and features found in urban space (Beirão 2012).
Ontologies are particularly suited for well-known
and well defined historical architectural types, and
may contain within it many different types of infor-
mation: either regarding its physical (and geometri-
cal) characteristics, conceptual definitions about the
type, etc. The development of an ontology is meant
to be a collaborative and adaptive process, chang-
ing or redesigning the ontology as new knowledge
about a subject is introduced, until consensus about
the subject is reached.
Furthermore Stouffs and Krishnamurti (2001)
propose a formalism called sorts for representing dif-
ferent data types. Conceptually, elementary data
types are used to define primitive sorts and these
combined can be used to define composite sorts un-
der formal compositional algebraic operations. As al-
gebraic forms, sorts are particularly fit to define and
combine grammars of different types as well as es-
tablish explicit relationships between sorts provid-
ing therefore semantic representations of concepts
in any representation domain.
In the next section we will look into works that
combine ontologies with shape grammars.
KNOWLEDGE, MEANING AND THE GENER-
ATION OF DESIGNS
In the previous section, we introduced shape gram-
mars as a tool to represent families of designs
and generate instances of designs algorithmically,
through the iterative application of rules. We have
also looked into mechanisms that, working in par-
allel with shape grammars, can control the gram-
mars' derivation process in order to produce desir-
able results. In this section, we will look into six ap-
proaches by different authors to the interaction be-
tween shape grammars and ontologies, over a period
of time spanning from 2006 to 2015.
308 | eCAADe 34 - SPATIAL REASONING AND ONTOLOGIES - Volume 2
Inductive System: Shape Grammars inform-
ing Ontologies
Andaroodi et al. (2006) developed a shape grammar
schema to classify Iranian Safavid and Ghajar open,
on-route caravanserais of the Silk Roads, and extract
attributes of spatial organization from it.
After representing the shape grammar rules in
a CAD drawing, these were redefined in natural lan-
guage in order to describe the design functions and
clarify the topology of shapes involved in the genera-
tion of these cultural heritage relics. Shape grammar
rules hold information regarding the spatial organi-
zation of its entities (geometry, topology, shape of
plan and dimension parameters) and provide a sys-
tematic recognition of the parts of the design that
could then be formalized, through descriptions, as an
ontology.
In this case the goal was not to generate new
designs within the type, but to use this schema to
extract computer-based semantics of the rules in or-
der to provide a cultural heritage database with en-
hanced documentation, using ontologies.
Extensive Semantics System
Grobler et al. (2008) identified certain similarities be-
tween knowledge-based models and shape gram-
mars: both of them contain design rules, though the
nature of the rules varies. In their case, ontologies
were used in parallel with shape grammars as a way
to inform shape rule application, giving it contextual
information specifically focused on existing architec-
tural types by capturing knowledge about the archi-
tectural type.
In this case, ontologies work as a knowledge rep-
resentation about a subject, describing individuals as
basic objects and organizing collections or types of
objects under classes. They also define properties
and characteristics and, more importantly, relations
between objects.
Based on the requirements, a lightweight Build-
ing Information Modelling (BIM) was constructed,
containing extensive information about the building.
This information (data) was then used as initial input
for the shape rules.
An ontology of architectural design drivers was
then used to provide additional data to the shape
grammar rules. Including specific information about
the building (type, location, culture, environment,
structural system and context) the ontology de-
scribed the relations between its different elements,
such as size and dimensions, form, structure, circula-
tion and movement patterns.
Collaborative Multi-Expert Semantics Sys-
tem
In the digital heritage project for ancient architec-
tures in southeast China, Liu et al. (2006; 2008) were
required to generate large quantities of 3D models
of ancient chinese buildings. Their approach focused
on a semiautomatic modelling process at semantic
level by using accumulated domain knowledge ex-
tracted from the buildings' architectures. This al-
lows the user to focus on the semantic relations be-
tween different components rather than focusing on
their geometric details. The modelling system had to
be capable of distinguishing different elements and
styles among multiple buildings. The system should
also be able to generate building parts in the style
or structure based on the semantic knowledge ex-
tracted from existing buildings. The modelling pro-
cess in this system can be described as designing
an ontology for the mentioned architectures, which
could then be used to interpret its styles and struc-
ture.
The authors consider a multi-level projection on-
tology design approach, with three participants in-
volved in ontology design: machine, programmer
and architect. Different ontologies are required to
accommodate the differences in domain knowledge
between the participants. Though the three partici-
pants are all represented with the same domain con-
cepts of architecture models, each has their own vo-
cabularies and relationships.
Since all of the rules of the grammar were man-
ually extracted by architects, most of them are em-
pirical and imprecise. This might lead to a highly re-
SPATIAL REASONING AND ONTOLOGIES - Volume 2 - eCAADe 34 | 309
dundant knowledge library, with the following draw-
backs: generation of unreasonable buildings from
incorrect rules; generation of hybrids from multiple
styles due to rule sharing between different styles;
and a decrease in the performance of the modelling
system due to redundant and incorrect rules.
To enhance the accuracy and performance of
their system, Liu et al. (2010) combined the ontology-
design method with a granular computing method.
Granular computing simplified the design of the
auto-modelling system and increased the hit ra-
tio (which describes the performance of the input
knowledge) when generating complex architectures
with a similar style structure.
Because the selection of appropriate rules from
a large knowledge library can be time consum-
ing, a Domain-Knowledge-based-Heuristic-Selection
(DKHS) algorithm was used, pruning the spatial
search tree. DKHS was considered to be an effec-
tive knowledge-selection algorithm, used to com-
pute the best knowledge subset from another knowl-
edge set.
A style-check module evaluates the generated
architecture instance to confirm if it belongs to the
desired ontology domain. Since a house generated
by the system is produced by the grammar, each
house can be represented individually as a sequence
of grammar terms - reducing the judging of the style
of the house to grammar checking. Components and
control rules are described in XML [1] and later veri-
fied with a technique based on Document-Type Def-
inition (DTD).
The methods described above were later for-
malised as the Onto-Draw framework. The Onto-
Draw engine is a unified platform and interface which
supports that multiple participants design the on-
tology of a complex problem in visual collabora-
tion. Onto-Draw combines varied domain knowl-
edge from multiple experts from different fields and
provides visual feedback from their designs. (Liu et
al. 2012)
 Automated Semantically Responsive Sys-
tem
Trescak, Esteva and Rodriguez (Trescak et al. 2010)
define a Virtual World Grammar (VWG) to automat-
ically generate, in real time, virtual auction houses
with semantic content. The Virtual World Grammar is
a compound grammar consisting of two shape gram-
mars (one for the institution building and its inner ac-
tivities; the other creates a separate building for each
activity), an ontology, validations and a set of heuris-
tics.
The ontology contains concepts related to (a) the
description of the activities that will take place in the
Virtual World and (b) concepts defining the proper-
ties of the Virtual World. When combined with the
shape grammar, descriptions of the activities (a) will
determine the layout of the Virtual World, while the
second (b) will define the properties of the Virtual
World elements.
The execution of a shape grammar is tested and
evaluated using validations. During the generation
process, the validations can be evaluated either af-
ter each generation step (step validations), prevent-
ing the selection of invalid paths of execution of a
shape grammar, or at the end of the generation pro-
cess (final validations), evaluating the final design.
Heuristics are used to guide the generation of
the Virtual Worlds. They decide the order in which to
process the elements from the specification and find-
ing possible execution nodes in the execution tree
for the selected specification element. Information
regarding the generation process is stored in a tree
structure, holding execution states which are either
defined by a rule or a shape.
Definition and execution of the Virtual World
Grammars is done via the Virtual World Builder
Toolkit (VWBT), through visual interfaces. The Virtual
World Builder Toolkit allows designers to explore dif-
ferent designs based on a shape grammar. The de-
signer of the grammar can browse possible designs
or fine tune parts of the grammar to obtain desirable
results. (Trescak et al. 2010; 2012)
One interesting aspect of this system is its abil-
310 | eCAADe 34 - SPATIAL REASONING AND ONTOLOGIES - Volume 2
ity to dynamically react to changes in the specifica-
tion - that is, changes in the ontology -, and automat-
ically regenerate the 3D environment incorporating
the new definitions, at runtime.
The workflow is divided into three main parts:
1. Preliminary definition: where the user defines
the ontology and the shape grammar;
2. Instance definition: loading the specification,
creating and defining all specification and
shape grammar elements, as well as specify-
ing mappings between them. Heuristics and
evaluations are also introduced here;
3. Execution: browsing random designs and
modifying instance parameters to reproduce
2D drafts, followed by its transformation into
3D.
Purely Semantic System
Stouffs and Tunçer (2015) take a different approach,
using typological descriptions as generative guides
for historical architecture. The typology of classical
period Ottoman mosques is described by an ontol-
ogy and description grammars are used as the mech-
anism to generate an instance of the typology from
the ontology. The ontology, represented as an XML
document, defines a meta-model that is able to cre-
ate a model within a domain.
As in previous researches, the physical architec-
tural object is decomposed in a hierarchical manner
and its components structured as an ontology. Struc-
tural and organizational aspects of the mosques are
classified under the "physical" class of the ontology.
Perceptual, spatial, philosophical and contextual as-
pects are classified under the "conceptual" class.
In their case, the ontology prescribes the design
descriptions the authors want to arrive at, at least in
their hierarchical decomposition. It also drives the
rule application process using a compound grammar,
constraining the selection of rules that can be applied
at any step in the process.
Their compound grammar consists of a descrip-
tion grammar and a shape grammar, with rules each
consisting of one or more description rule compo-
nents and one or more shape rule components. By
combining the description grammar with shape pro-
duction rules, the compound grammar is able to gen-
erate instances of the typology both as an ontologi-
cal description and as a geometric model.
Stouffs and Tunçer (2015) suggest the use of
the term shape grammar to denote their compound
grammar, in reference to Stiny's (1992) definition,
considering their grammar to apply over an algebra
that is composed of both shape algebras (extended
with labels and/or weights, following its definition),
and one or more description algebras.
Generative with Semantic Control System
Regarding urban design, an ontology describing
street systems was used as a common representation
protocol between three interrelated modules of an
urban design tool developed in the context of the
City Induction project: a formulation module, a gen-
eration module and an evaluation module. (Duarte
et al. 2012; Beirão 2012; Beirão et al. 2012)
In this tool, the representation of new urban
plans is generated by a compound grammar com-
posed of several discursive grammars, each one us-
ing in their shape set one of the object classes found
in the city ontology. The city ontology describes sev-
eral complementary representations of a city: the city
seen as a street system; the city as a built system;
the city as a property system and so on. Each view
of the city is described by a taxonomy that hierar-
chically describes its system branching from object
classes describing generic or large concepts and pro-
gressively detailing each concept into sub-classes. In
a way, each system can be seen as a sub-ontology of
the larger ontology describing the city.
An instance in a class is composed of a shape part
and a semantic part (a label, a description, or any type
of classifier). Depending on the class the shape part
or the semantic part may be an empty set.
The compound grammar is built out of smaller
grammars in the form of discursive grammars that
replicate common urban design operations and are
SPATIAL REASONING AND ONTOLOGIES - Volume 2 - eCAADe 34 | 311
therefore called urban design patterns - they inte-
grate simultaneously the design pattern concept de-
veloped by Gamma et al. (1995) and the idea devel-
oped by Alexander et al. (1977) of defining languages
of design by repeating well known and time tested
design operations.
DISCUSSION
In the previous sections we looked into six systems
for semantic design: Inductive, Extensive Semantics,
Collaborative Multi-Expert Semantics, Automated Se-
mantically Responsive, Purely Semantic and Generative
with Semantic control.
These relate generative design systems (shape
grammars) with knowledge bases (ontologies), in-
troducing semantic control in the design generation
process. The Inductive System's sole objective is to
extract knowledge, as an ontology, from the gener-
ative rules of a shape grammar that represents an
historical architectural type. The remaining systems
take different approaches to the semantic generation
of designs, producing meaningful results within a de-
sign language. Apart from the Inductive System, all
systems share characteristics between one another;
their identification tries to reflect their most promi-
nent characteristics.
The systems reveal the complexity of the subject
of semantic design systems and the difficulties inher-
ent to its computational implementation.
Table 1 presents an overview of these systems,
looking into some characteristics that may be used as
a reference for future development of semantic de-
sign systems. It highlights specific characteristics in
each system, such as the ontology language used or
the final output of the system.
One interesting thing to notice is a slight prefer-
ence for the XML language to represent the ontolo-
gies, over OWL [2] and/or RDF/S [3][4]. The use of
heuristics and validations in the derivation process
is widespread in all systems. Heuristics appear to be
fundamental to coordinate the semantic application
of shape rules, using validations to confirm the se-
mantic accuracy of the derivation.

The Virtual World Builder Toolkit interface, in the
Automated Semantically Responsive System, is inte-
grated in their own Shape Grammar Interpreter (SGI),
using 3D transformation engine jMonkey Engine [5] -
a free, open source game engine.
Most systems rely heavily on knowledge descrip-
tions, which are then used as a control mechanism
for shape grammar derivation. Knowledge in the
ontologies is also used to restrict parameters in the
shape grammars.
Regarding computer implementation, only the
Generative with Semantic Control System uses an in-
terface which is familiar in architectural design: the
Grasshopper 3D [6] plugin for Rhinoceros [7].
One important topic for future developments
of semantic design systems would be the agree-
312 | eCAADe 34 - SPATIAL REASONING AND ONTOLOGIES - Volume 2
Table 1
Overview of the six
Semantic Design
Systems.
ment upon the structure of ontologies representing
building or urban types and typology. The works
presented here already pave the way for (universal)
building or urban knowledge descriptions.
CONCLUSION
This paper presented an overview of approaches to
Semantic Design Systems in the fields of architec-
ture and urbanism (for both real and virtual environ-
ments), more specifically regarding the combination
of computational ontologies and shape grammars.
A review of control mechanisms to regulate
shape grammar derivation was presented, suggest-
ing the need for more flexible and adaptive knowl-
edge representations than just descriptions. Knowl-
edge bases such as ontologies are a suitable solu-
tion to introduce semantic controls in shape gram-
mar derivations.
Following this, six strategies were presented
proposing different approaches to the same prob-
lem: how generative systems and knowledge-bases
may be combined to produce feasible, meaningful
results within a design language.
Inversely, we also saw how a generative system
(shape grammar) may be used to inform a knowledge
base (ontology) about a type of building, helping to
its classification. This illustrates the symbiotic rela-
tion between knowledge bases and generative de-
sign systems.
Pairing ontologies with shape grammars can be
used for the (collaborative) creation of structured
knowledge about families of designs, as well as a
(semi-)automatic mechanism to generate meaning-
ful solutions within a family of designs - the underly-
ing goal of the research.
None of the examples presented in this survey
offers a complete solution for the creation of seman-
tic design systems but suggest partial mechanisms to
further explore.
ACKNOWLEDGEMENTS
The authors are thankful for the financial sup-
port given by Portuguese Foundation for Science
and Technology (FCT). This work was partially sup-
ported by national funds through FCT with reference
TECTON-3D PTDC/EEI-SII/3154/2012.
REFERENCES
Aksamija, A., Yue, K., Kim, H., Grobler, F. and Krish-
namurti, R. 2010, 'Integration of knowledge-based
and generative systems for building characteriza-
tion and prediction', AI EDAM, 24, pp. 3-16
Alexander, C., Ishikawa, S., Silverstein, M., Jacobson, M.,
Fiksdahl-King, I. and Angel, S. 1977, A pattern lan-
guage, Oxford Univ. Pr.
Andaroodi, E., Andres, F., Einifar, A., Lebigre, P. and
Kando, N. 2006, 'Ontology-based shape-grammar
schema for classification of caravanserais: a specific
corpus of Iranian Safavid and Ghajar open, on-route
SPATIAL REASONING AND ONTOLOGIES - Volume 2 - eCAADe 34 | 313
samples', Journal of Cultural Heritage, 7, pp. 312-328
Beirão, J. 2012, CItyMaker: Designing Grammars for Urban
Design, Ph.D. Thesis, TU Delft
Beirão, J., Duarte, J. and Stouffs, R. 2009 'Grammars of de-
signs and grammars for designing - grammar-based
patterns for urban design', CAAD Futures 2009
Beirão, J., Duarte, J., Stouffs, R. and Bekkering, H.
2012, 'Designing with urban induction patterns: a
methodological approach', Environment and Plan-
ning B: Planning and Design, 39(4), pp. 665-682
Duarte, J. P. 2001, Customizing mass housing: a discursive
grammar for Siza’s Malagueira houses, Ph.D. Thesis,
MIT
Duarte, J. P., Beirão, J. N., Montenegro, N. and Gil, J.
2012, 'City Induction: A Model for Formulating, Gen-
erating, and Evaluating Urban Designs', in Müller
Arisona, S., Aschwanden, G., Halatsch, J. and Wonka,
P. (eds) 2012, Digital Urban Modelling and Simula-
tion, Springer Berlin Heidelberg, pp. 73-98
Eloy, S. and Duarte, J. P. 2011, 'A Transfformation Gram-
mar for Housing Rehabilitation', Nexus Network Jour-
nal, 13(1), pp. 49-71
Fleisher, A. 1992, 'Grammatical Architecture?', Environ-
ment and Planning B: Planning and Design, 19(2), pp.
221-226
Gamma, E., Helm, R., Johnson, R. and Vlissides, J. 1995,
Design patterns: elements of reusable object-oriented
software, Addison-Wesley
Grasl, T. and Economou, A. 2010 'Palladian Graphs : Using
a graph grammar to automate the Palladian gram-
mar', Proceedings of eCAADe 2010, pp. 275-283
Grasl, T. and Economou, A. 2013, 'From topologies to
shapes: parametric shape grammars implemented
by graphs', Environment and Planning B: Planning
and Design, 40(5), pp. 905-922
Grobler, F., Aksamija, A., Kim, H., Krishnamurti, R., Yue,
K. and Hickerson, C. 2008 'Ontologies and Shape
Grammars: Communication between Knowledge-
Based and Generative Systems', DCC'08, pp. 23-40
Gruber, T. R. 1993, 'Toward principles for the design of
ontologies used for knowledge sharing?', Interna-
tional Journal of Human-Computer Studies, 43, pp.
907-928
Knight, T. 1989, 'Color grammars: designing with lines
and colors', Environment and Planning B: Planning
and Design, 16(4), pp. 417-449
Knight, T. 2003, 'Computing with emergence', Environ-
ment and Planning B: Planning and Design, 30(1), pp.
125-155
Liew, H. 2003 'SGML - A Shape Grammar Meta-Language',
Proceedings of eCAADe 2003, pp. 639-647
 Liew, H. 2004, SGML: a meta-language for shape gram- [2] https://www.w3.org/OWL/
mars, Ph.D. Thesis, MIT [3] https://www.w3.org/RDF/
Liu, Y., Jiang, Y. and Huang, L. 2010, 'Modeling Complex [4] https://www.w3.org/TR/rdf-schema/
Architectures Based on Granular Computing on On- [5] http://jmonkeyengine.org/
tology', IEEE Transactions on Fuzzy Systems, 18(3), pp. [6] http://www.grasshopper3d.com/
585-598 [7] http://www.rhino3d.com/
Liu, Y., Xu, C., Zhang, Q. and Pan, Y. 2006 'Ontology
based semantic modeling for chinese ancient archi-
tectures', Proceedings of the National Conference on
Artificial Intelligence, pp. 1808-1813
Liu, Y., Xu, C., Zhang, Q. and Pan, Y. 2008, 'The Smart Ar-
chitect: Scalable Ontology-Based Modeling of An-
cient Chinese Architectures', IEEE Intelligent Systems,
1, pp. 49-56
Liu, Y., Zhang, M., Tang, F., Jiang, Y., Pan, Z., Liu, G. and
Shen, H. 2012, 'Constructing the virtual Jing-Hang
Grand Canal with onto-draw', Expert Systems with Ap-
plication, 39(15), pp. 12071-12084
Stiny, G. 1980, 'Kindergarten grammars: designing with
Froebel’s building gifts', Environment and Planning B:
Planning and Design, 7(4), pp. 409-462
Stiny, G. 1981, 'A note on the description of designs', En-
vironment and Planning B: Planning and Design, 8(3),
pp. 257-267
Stiny, G. 1992, 'Weights', Environment and Planning B:
Planning and Design, 19(4), pp. 413-430
Stiny, G. and Gips, J. 1971 'Shape Grammars and the Gen-
erative Specification of Painting and Sculpture', IFIP
Congress (2)
Stouffs, R. and Tunçer, B. 2015, 'Typological Descrip-
tions as Generative Guides for Historical Architec-
ture', Nexus Network Journal, (2015), pp. 785-805
Trescak, T., Esteva, M. and Rodriguez, I. 2010, 'A Virtual
World Grammar for automatic generation of virtual
worlds', The Visual Computer, 26, pp. 521-531
Trescak, T., Esteva, M. and Rodriguez, I. 2012, 'A shape
grammar interpreter for rectilinear forms', Computer-
Aided Design, 44(7), pp. 657-670
Tunçer, B. 2013, 'A digital precedent library for classi-
cal period Ottoman mosques', Journal of Design Re-
search, 11(4), pp. 351-371
Tutenel, T., Smelik, R. M., Lopes, R., De Kraker, K. J. and
Bidarra, R. 2011, 'Generating Consistent Buildings: A
Semantic Approach for Integrating Procedural Tech-
niques', IEEE Transactions on Computational Intelli-
gence and AI in Games, 3(3), pp. 274-288
Şener, S. M. and Görgül, E. 2008, 'A shape grammar al-
gorithm and educational software to analyze classic
Ottoman mosques', A|Z ITU Journal of the Faculty of
Architecture, 5(1), pp. 12-30
[1] https://www.w3.org/XML/

314 | eCAADe 34 - SPATIAL REASONING AND ONTOLOGIES - Volume 2

Human Behaviour Simulation to Enhance Workspace
Wellbeing and Productivity
A BIM and Ontologies implementation path

Armando Trento1 , Antonio Fioravanti2

1,2
Sapienza, Università di Roma
1,2
{armando.trento|antonio.fioravanti}@uniroma1.it

Three-quarters of the production value are generated during activities that

involve thinking, conducting relational and brainstorming activities. Most of the
European office buildings today have been designed on more than fifty year old
architectural and psychosocial concepts. To improve wellbeing and productivity,
design innovation focuses on human's use-process, evolving individual workspace
to flexible and specialized ones, according to the users tasks - activity-based.
BIM supports sophisticated behaviors simulation such as energy, acoustics,
although the state of the art, this paradigm is not able to manage space
use-processes. Compared to current research on simulation systems, the proposed
method links spaces to user's Behavioral Knowledge including formalization of
Personality Typologies and profiled behavioral patterns. A hybrid approach for
computational technique has been identified, combining (big) data-driven
algorithm with ontology-based context reasoning, in order to achieve both, the
best performance from intensive data-driven methods, and the finest adaptation
for ontological context awareness (including unexplored context capabilities and
objects adaptations).

Keywords: Event Ontology, Design Knowledge Representation and

Management, Human Behaviour, BIM

WORKSPACES TO ENHANCE WELLBEING
AND PRODUCTIVITY
In order to improve the quality of an architectural
artefact, a central task for multidisciplinary design
teams is to test the tentative design solutions and see
how well they work 'in practice' before, during and af-
ter the construction: from digital model to real world.
The building quality does not involve only aes-
thetic and economical aspects but also wellbeing
and productivity of employees.
Nowadays companies not only recognize the
connection between employee productivity and
mindfulness, but also they are looking at good data
to support it.
The fact is, there is a major shift underway in
the workspace from employee wellness to wellbe-

SPATIAL REASONING AND ONTOLOGIES - Volume 2 - eCAADe 34 | 315

 ing. Almost all (96 %) of the 6,500 organizations sur-
veyed by WorldatWork [1] offer elements of a wellbe-
ing program, and three-quarters of them are increas-
ing their offerings in the next two years with the main
objectives of improving employee health, impacting
healthcare costs and increasing productivity.
While the two terms "wellness" and "wellbeing"
might sound much the same, in the workspace con-
text they mean very different things.
Most of the European buildings, specifically facil-
ities used as working places and offices, used by com-
panies today are designed and built on more than
fifty years old architectural and psychosocial con-
cepts. This is reflected on the workspace use-process
that does not respond to the needs and way of work-
ing required nowadays. The problem of human dis-
engagement from the cultural and working context
generates a loss of productivity and excessive operat-
ing costs that frequently are reflected on a reduction
of competitiveness.
Given these premises, emerges the potential and
strategic importance of rethinking the workplace, to
be considered not only as an element of cost reduc-
tion related to unused spaces but above all as a lever
to increase the percentage of workspace user en-
gagement.
In fact, according to Gensler (2011), to be able to
improve by 10% the productivity of human resources
can be reflected on an economic advantage 20 times
higher than the one obtained with the 10% reduction
of the spaces.
Workspaces should be redesigned by focusing
on the wellbeing of people and considering the value
generated by activities carried out by the latter. Ac-
cording JLL global [2], three-quarters of the com-
pany production value are generated during activi-
ties that involve thinking, conducting relational and
brainstorming activities. The urgent problem is that
nowadays workspaces, for how they are structured,
they are not designed to facilitate these activities, nor
are they designed to increase the level of welfare and
"engagement".
Design innovation, focused on considering the
316 | eCAADe 34 - SPATIAL REASONING AND ONTOLOGIES - Volume 2
humans at the center of the use-process, is currently
oriented to an evolution of the individual workspace
to flexible and specialized workspaces according to
the tasks that users must play - activity-based.
Productive workspace elicit collaboration, cre-
ativity and foster new work experience integrating
processes, actors and products.
Involvement, open connections and peering,
multidisciplinary sharing, interoperability starting
from comfort and wellbeing for the most important
assets of any companies: people.
In order to provide different spaces and suitable
for the various activities, it is important to review the
design process starting from the analysis of the oc-
cupants, the space use-process, employees and their
prevalent psychological profiles.
To allow coexistence and availability to all work-
ers of different types of environment it is necessary
to define and adopt a smart and flexible working
model-process so that everyone can benefit from the
most suitable and comfortable space.
Design by means of simulations
A long debate has arisen, especially in Europe, from
the possibility of re-using and giving new functions
to existing buildings.
On one side, there is the necessity and the oppor-
tunity of giving new life to these environments while,
on the other side, especially in big cities' historical
centers, where still higher is the density of office fa-
cilities, several problems arise from the preservation
of the built heritage artifact and its actual capabilities
of hosting new functions, users and use scenarios.
At present, the task of predicting and assessing if
and how an existing environment will effectively host
new uses and/or users is still unsupported and com-
pletely left to designers' expertise and imagination.
To enhance the control on the final design prod-
uct and use process quality is a key element for boost-
ing work space users' wellbeing and productivity.
Property managers, architects, structural engi-
neers, energy engineers and all the actors involved
in companies' asset re-functionalization, know very
 well the importance of how the building and the ac- His 'Reciprocal Determinism' theory, defines Be-
tivities are accurately modelled in order to better de- haviour as function of the Person and his/her (physi-
sign and manage the use-process. cal/social) Environment, where Person, Environment,
Researchers are urgently called at develop- and Behavior influence one another in a dynamic way
ing methods and tools to support the decisions- (Figure 1). In this way, he started to develop a sys-
making processes related to workplaces refurbish- temic view of environmental behavior.
ment, change of use, conservation and management.
Figure 1
On this basis, few current research projects
Behaviour as
(Tabak 2009; Zimmerman 2006; Shen 2012) have
function of
been involved in in the development of conceptual
Personality,
modeling approaches in order to enhance simulation
Environment and
of existing artifacts in relation with their potential re-
their relations.
uses.
Specifically, the problem addressed by the au-
thors' research group since the last decads is the
The environment does not only cause behavior, but
conceptual development and validation of a model-
is also influenced by behavior:
ing and simulation platform to test different use sce-
Persons ---> MODIFY ---> Environment
narios in order to understand and evaluate the re-
Persons actively search for situations that fit their
lationship between humans users and the designed
aims and personality:
workspace, namely if and how the re-designed envi-
Persons ---> SELECT ---> Environments
ronment will be able to host its intended new func-
Personality, attitudes, expectancies, goals, and
tions, users and use processes.
competencies are influenced by the social and nat-
Actual research framework outlined in this pa-
ural environment.
per starts from occupancy analysis (users profiling,
Authors perspective starting from Lewin theory
context and reciprocal interaction) in order to col-
and extending the concepts of Persons and Environ-
lect Behavioral Knowledge. A suitable structure for
ment, is that a reciprocal influence exist between
formalizing project-process semantics and incremen-
users and context, where the first affects the latter,
tally populate it have been studied.
and viceversa (Gargaro 2015).
The core of the modelling task is to ensure an ef-
Users <---> MODIFY<---> Context
ficient connection between building process, prod-
Users <---> SELECT<---> Context
uct, context and users, in order to support design
Since Personality makes users unique and differ-
and evaluation, by means of both static simulation
ent from each other, starting from the late XIX cen-
(Data-driven and Ontology-based) of a specific arte-
tury numerous structured psychological approaches
fact, and dynamic simulation of a contextualized use
have been proposed such as e.g. Analytical and deep
process (Agents+AI).
Psychology (Freud), Humanistic Psychology (Rogers;
The environmental behaviour Maslow), Trait approaches towards personality and
According to Kurt Lewin's founder of the experi- Personality Typologies.
mental Social and Environmental Psychology (1936), Anyway, in today practice, it can be observed a
Topological psychology (or "Field Theory"), the field diffuse pragmatic classification approach adopted by
of social and environmental forces influences be- commercial goods and services industry, based on
haviour. opportunistically oriented and effective Personality
LEWIN'S BEHAVIOUR FORMULA: B = f (P, E) Typologies definition.
'Consumer' typology and 'life-style' typology use

SPATIAL REASONING AND ONTOLOGIES - Volume 2 - eCAADe 34 | 317

 psychological and other features to describe a group
of persons: these are not personality theories or traits
in the classical psychological sense, e.g. not stable
over lifetime.
Such typologies often include consumption pat-
tern and behaviours as a basis for classification. Of
course it is required a clear distinction between char-
acteristics used for classification, and related behav-
ioral characteristics, e.g. different variables can be
focused on the same or different typologies: demo-
graphics, environmental knowledge, environmen-
tal concern, norms, activism, shopping motivations,
shopping behavior etc.
In order to collect formalized knowledge related
to users' profiles, in order to model phenomenon and
process simulations we classified two main method-
ological categories.
On one side, more traditionally, experts work on
the defining space users Personality Typologies by
means of structured surveys for outlining differences
in the same classes of users approaching the same ac-
tivities (preferences, value orientation, expectancies,
attitudes, etc.).
On the other side, the knowledge source origi-
nates by a sort of reverse engineering process, cap-
turing data, information and knowledge from real
world monitoring, by means of different media tech-
nologies (temperature revelator, camera, RFID, In-
ternet of Things, etc.) according to the following
process: REALITY -> BIG DATA COLLECTION -> DATA
DRIVEN PROCESSING -> ONTOLOGY RECOGNITION -
> ONTOLOGY POPULATION
Expected capability, implemented action
and oriented behaviour instances
Designers are used to define function as what could
be done in every space, and the action as an actual in-
stantiation of a function (i.e. a function that was put
into effect).
In order to clearly state a basic taxonomy of Be-
havioral Knowledge, authors, according to Wurzer
(2009) state two notions:
• Capability: the ability to perform a certain ac-
318 | eCAADe 34 - SPATIAL REASONING AND ONTOLOGIES - Volume 2
tion in a space (e.g. capability can sit and can
park bed in a hospital's waiting area).
• Action: the implementation of a capability in
space and time (e.g. action wait in the same
waiting area that either uses can sit for pa-
tients that can walk or can park bed for re-
clined patients).
A substantial part of present scientific research is fo-
cused on developing techniques to model, learn, rec-
ognize, and predict what users are doing in the envi-
ronment, so that the system is able to make decisions
about how to assist them and designers.
Usually, the literature calls what users are doing
human behavior or human activity interchangeably
(Remagnino et al. 2005; Ros et al. 2013).
These terms usually mean a sequence of hu-
man actions that can be tagged with a label, that is,
the corresponding activity/behavior. These authors
agree to define human action as the simplest unit in
the human activity, and it is usually associated with a
sensor event.
From the authors' point of view, as long as new
semantic approaches are being developed (Chen and
Nugent 2009), new abstraction levels appear in the
system.
For this reason, from the authors' perspective, a
difference should be made between the terms hu-
man activity and human behavior to separate the
concepts of what the user is really doing in the en-
vironment (activity, intended as the implementation
of space functional capability), which is inferred from
sensor data and machine-learning techniques, and
the purpose or meaning it could have (behavior, in-
tended as a contextualized and oriented system of
users' actions).
Despite the great advances produced in the last
decade, the complexity and the quantity of possible
complex activities, the temporal interdependences
among actions (Ros et al. 2013), the relevance of the
semantics associated with a behavior (Chen and Nu-
gent 2009), and the existence and interaction of sev-
eral actors in the same environment/context, make
learning and recognition of human behavior relevant
and bring up clear challenges in present research.
DESIGN TOOLS
The advent of BIM technologies and their perva-
sive diffusion in the professional design studios is
introducing an interesting modification of designer
habits, extending their capacity to foresee building-
related problems and conflict managements, which
are typical of the subsequence phases: construction,
maintenance, re-use or demolition.
This happened because the multidisciplinary de-
cision making process, complex and, for some as-
pects highly recursive, relies on the way product-
related knowledge is modelled in the actual CAAD
design tool.
BIM and Users
Studying the most common standard in the BIM field,
such as Industry Foundation Classes (IFC) [3], we can
observe that CAAD tools have been developed by
means of a space-components product approach,
successful in terms of data exchange and information
interoperability between programs. Notwithstand-
ing that, these tools lack of semantics that in turn is
reflected in the modelled buildings, especially when
required to simulate its behaviour in terms of usage,
safety and comfort.
More specifically, predicting human behaviour
in a building during its usage, by means of the cur-
rent standards, tools and technologies is an urgent
open problem, which has challenged knowledge en-
gineers and building designers for long time. It also
involves a lot of resources in terms of industrial re-
search and development in the fields of military and
videogames.
BIM paradigm supports sophisticated design ap-
proaches to simulate complex behaviors such as en-
ergy, acoustics, lighting, although at the state of the
art, it is not able to manage both user's activities and
space use-process.
On the other side, in recent decades, compa-
nies typically centered wellness programs only on the
physical aspects of health, such as blood pressure,
SPATIAL REASONING AND ONTOLOGIES - Volume 2 - eCAADe 34 | 319
blood sugar and so on.
By contrast, the actual trend is oriented to well-
being, a more holistic concept - focused on 'being'
- acknowledging the connection between the body
and mind, as well as the workplace environment and
experience itself [4].
In order to evaluate wellbeing in workspaces,
simulation models need to manage complex seman-
tics, including both, BIM based environmental phys-
ical parameters, and formalized user's personal as-
pects, like personality typologies, traits profiling and
expected behavioral patterns.
As new approaches for task modeling include se-
mantics to simulate human activity, context aware-
ness techniques in terms of space and time become a
more central part of next future systems.
BIM models must be integrated with more ab-
stract sematic levels, e.g. characterizing Spaces not
only by physical parameters related to environmen-
tal comfort but also with space-time functional as-
pects, including, in terms of Capabilities the effec-
tive distinction introduced by Tabak (2004) between
two different classes of Activities: Skeleton Activities,
referred to companies planned functional processes
and Intermediate Activities, intended as recurrent oc-
curences, only partially predictable in term of pro-
cess.
Once collected and formalized, Knowledge can
be computed to process simulations.
Use-process simulation tools
As agreed by the most scientific literature in the field,
a simulative model is based on two main compo-
nents:
• A static component, representing a specific
and unique system status based on all for-
malized entities, including all the instances
present in the instant T0,
• A dynamic component, able to perform the
changing of the entities state from the system
status T0 to T1.
 In the last years, agent-based modeling approaches
have been introduced in this research field, aiming
at simulating users' behavior in built environments
by developing a series of autonomous entities - the
agents - each of which interacts in an autonomous
way with the other users and with the environment
surrounding it.
This kind of simulation approaches, such as "nar-
rative approach" doesn't allow a prediction but the
pre-defined scenario visualization.
According to Kalay (2013), agent-based models
has shown to be highly requiring in terms of compu-
tational resources and not enough expressive in the
simulation of events in which the users-agents have
to make context dependent decisions and behave in
an interleaved way.
The simulation model here presented integrates
two main modules:
• Use Process Knowledge which structure has
been presented in previous papers (Trento
2013), linking, in a homogeneous computa-
tional environment, BIM to higher level se-
mantics;
• - Simulation engines to perform and visualize
the effects of the model status change.
Based on this kind of model, a hybrid Agents based
simulation model is investigated.
Agent are associated to AI resources, that reside
not only in the Actors' Knowledge Bases, but oppor-
tunistically in other entities (Context, Product, Pro-
cess), reducing computational loads and enabling in-
ference engines to process rules-based reasoning.
Techniques for representing Human Be-
haviour
Current trends show that tracking and monitoring
people is becoming an integral part of everyday life
so that user behaviours can be captured and stored.
This way we have the user actual actions per-
formed with their specific movements, characteris-
tics, peculiarities. We can distinguish an activity, the
planned action for every space; the undertaken ac-
320 | eCAADe 34 - SPATIAL REASONING AND ONTOLOGIES - Volume 2
tion that starts in a space; the concrete action per-
formed that depends on the personality of each per-
son. An activity can be planned for certain spaces for
instance -checking passport, but the subsequent ac-
tions are determined by circumstances and each per-
son has his/her own habit to do his/her job.
Data-based approaches (i.d. Bayesian networks,
decision trees, etc.) appear to stand out while
knowledge-based techniques are back. The latter
include, among others, information indexing and
retrieval, hierarchical knowledge sources (i.d. tax-
onomies, ontologies, encyclopedias, dictionaries),
representation languages, distributed knowledge,
and especially logical tools.
According to Villalon et al. (2010), the broader
the ontology is, the more situations are possible to
be modelled, in order to assist the users in their daily
activities; and the less usable the ontology is in order
to achieve a specific goal. However, the more spe-
cific the ontology is, the fewer possibilities exist to be
reused, and the more efficient the ontology is for that
field.
The size of the problem, specific domain, and
concrete task are key elements when selecting an on-
tology. However, what can be appreciated from the
Rodriguez (2015) survey is that most of the works
require a data-intensive-driven first approach to ro-
bustly identify the most basic-level actions or activi-
ties.
Context Awareness
Context consists of not only physical space like envi-
ronment, but also cultural, psychological, technical,
etc. (Gargaro 2013) or any information that can be
used to characterize the state of an entity (Dey and
Abowd 2000). Entities can include a person, an ob-
ject, an environment, an application, or a device that
interacts between them and the user.
Proposals to model context can be integrated
with human activity models provided with seman-
tics.
According to Rodriguez (2015), as time is not a
feature inherently treated in knowledge-driven ap-
proaches such as logic-based systems, having hy-
brid methods with a first data-driven preprocess-
ing stage appears to be the right direction to bene-
fit from both data- and knowledge-driven comput-
ing paradigms. As ontological reasoning can be
computationally expensive, this type of combina-
tion would achieve the best performance and effi-
ciency from (time-dependent) data-driven methods
(that can be more efficiently computed) and obtain
the best adaptation for context awareness in each
case.
It is important to note that Web Ontology Lan-
guage 2 (OWL2) is powerful for expressing knowl-
edge entities, context entities and relations among
entities. However, OWL2 is insufficient to model con-
text relations and rules with the form of cyclic rela-
tions
Therefore, the ontologies discussed require an
integration with a rule language - such as Semantic
Web Rule Language (SWRL) [5] or SPARQL Inference
Notation (SPIN) [6] - in order to express more complex
and real-life ontext rules.
The combination of Data language with Rule-
based language improves the reasoning capabilities.
Rule-based languages enable definition of con-
sistency rules, reducing ambiguity in the context
information and thus maintaining and improving
the information quality. For instance, SWRL ex-
tends the semantics of OWL and defines antecedent-
consequent rules and built-in operators (math, com-
parisons, string and time).
As a concluding remark, we can clearly point out
that the integration of different methodologies (i.e.
data-driven and knowledge-based ones) could help
overcome current limitations in scenarios with sev-
eral actors, providing semantics to social activities,
user identification (according to behavior semantics),
and so forth.
Current hybrid approaches such as Gomez-
Romero et al. (2011) have shown that these types
of combinations can enhance the response of data-
driven approaches as the environment complexity
and the context awareness needs increase.
SPATIAL REASONING AND ONTOLOGIES - Volume 2 - eCAADe 34 | 321
A FRAMEWORK TO FORMALIZE SPACES
AND HUMAN BEHAVIOUR
The model for simulation of the interaction between
humans and environment presented here outlines
an implementation pattern for integrating BIM space
and physical characteristics by behavioural knowl-
edge, intended as a semantic structure capable
of representing the entities and their relationships
between activities-actors-context and places (Fiora-
vanti 2010)
Compared to current research in the field, the
present approach to represent Behavioural Knowl-
edge includes among others formalization of user
personal aspects, like Personality Typologies, classes
profiling and expected behavioural patterns.
A hybrid approach for computational technique
has been preferred, combining (big) data-driven al-
gorithm with ontology-based context reasoning, in
order to obtain both, the best performance and effi-
ciency from intensive data-driven methods, and the
best adaptation for ontological context awareness
(including unexplored context capabilities and ob-
jects adaptations).
This section reports on a theoretical framework
and some early implementation patterns developed
in the general framework of an on-going research
aimed at the definition of a new approach to support
assistive systems for management and performance
simulations.
To model and test knowledge related to the user
behaviour in a building environment we need:
1. Spaces that are characterized by physical pa-
rameters related to environmental comfort
but also with space-time Functional aspects,
Potentialities or Behaviours, Capability or Ac-
tion (Wurzer 2010);
2. A Use Process Knowledge structure (Trento
2013) that includes Skeleton Activities and In-
termediate Activities (Tabak 2004);
3. An Agent-based simulation, enhanced by as-
sociating agents with AI resources (upper on-
tology level), that reside not only in the Actors'
Knowledge-based systems, but also in other
 systems (Context, Product, Process).
Planner traditional approach conceives at the begin-
ning of planned processes expertise, technical regu-
lations, best practices, etc. in an architectural schema
(Wurzer 2009; Wurzer 2010). However, those pro-
cesses are correctly performed only if the planner can
rightly anticipate and inform the usage of the build-
ing by different building users.
Evidently this is a risky task, relying on designers
limited and implicit knowledge, since the human be-
haviour in a building is highly non-deterministic and
not a-priori definable.
To support this kind of operation, we rely on
a general structure of knowledge representation
known in the scientific community developed by
DaaD group that works to extend its application
fields (Fioravanti et al. 2011; Fioravanti et al. 2012).
As a starting point we used the already de-
fined "Meaning-Properties-Rules" (M-P-R) knowledge
structure (Carrara et al. 2009) able to describe the
product entities and their relations by means of a
general template (freezing the control on the other
knowledge realms, as context, actor and process).
Behavioural simulation innovative ap-
proach. How to afford the mismatching
"squared peg in a round hole problem"
The innovation is to take into account misused
spaces or inappropriate user's behaviour to modify
one of them or both to reach the activity goal.
The human history - especially the science one -
is plenty of overcoming (apparently) impossible goal.
The "trial and error" paradigm both in technology
and in theory, very often intertwined, as a means to
adapt circumstances, tools and goals has been the
normality not an exception at different abstraction
layers with different granularity.
The key concept to resolve deadlocks has fre-
quently been the "adaptation" mechanism.
For instance, the classical problem of "squared
peg in a round hole problem" - extending this con-
cept to human satisfaction of workspace - can be
solved by cutting the peg, or enlarging the hole, or
322 | eCAADe 34 - SPATIAL REASONING AND ONTOLOGIES - Volume 2
both.
Going back to the peg: How can it fit in the hole?
Can it be cut? What is the stiffness of material?
Are there tools to modify it in a right manner? What
is the cost? How much time should be spent for mod-
ifying the initial planned work? What phase is it, the
design or the construction one?
Situational and contextual factors need to be
taken into account for understanding human be-
haviour.
Apart from personality traits, further interper-
sonal differences are important for explaining and
predicting environmental behaviour, e.g. compe-
tencies and knowledge, expectancies, value orien-
tations, (environmental) attitudes, personal norms,
psychological states (e.g. tiredness, stress, state-
anxiety).
Explaining environmental behaviour contem-
porarily relies, to very large extent, on different con-
cepts than clinical and personality psychology. As
value-orientations, attitudes, knowledge and com-
petencies are usually understood as less stable (less
deeply rooted in personality) as compared to person-
ality traits, considering these factors opens up possi-
bilities for changing environmental behaviours.
The new approach we present here, differently
from (Trento 2013) does not completely renounces
at the task of modelling the actor's personality profile,
and focuses on defining a set of occurrences, that dy-
namically happen in the workspace and, organized in
a proper way, represents a scenario of experience of
the building use.
An 'event' represents a set of activities, per-
formed by a determined group of actors in a deter-
mined space of the building.
If we use simulations in a virtual world to simu-
late future events of the real world, we have to reli-
ably represent it, but not be limited by the real world
rules. So, while in the real world only people have the
capabilities to think, evaluate the environment and
control their behaviour, in the virtual world this task
can be assigned to entities, both representing physi-
cal or abstract objects.
 This is helpful for our approach because it offers
the possibility to build event- agents and give them
the task of controlling other 'passive' objects (actors,
furniture, building components, facilities), their be-
haviours and their interactions.
In this modelling approach a scenario of build-
ing usage can be represented as an organized,
time-depended and context-depended structure of
events by means of a suitable formalism.
Events and Relations among them are extremely
influenced by the status of the built environment and
by what is happening in it, so we can consider the
event-environment system as a stochastic system in
which there is a high level of uncertainty and varia-
tion. While in the real world events emerge from peo-
ple behaviour, in the proposed model events are pre-
defined in order to build a usage scenario but this
doesn't mean that the structure is fixed. In fact, to
build a reliable structure of events, the model deals
with two different components during the simula-
tion: narrative events sequence and events structure
adaptation. The first is necessary because, to build
a scenario, it's necessary to build a general plot of
the occurrences that are going to happen in environ-
ment; this can be derived by the pattern of activities
related to the functional aims of the designed build-
ing and by the initial building program. But this is not
sufficient to reliably represent the usage scenario be-
cause we need to adapt this plot to the status of the
on-going built and activity context to simulate the ef-
fective sequence of events. It is not a one-way system
comparing initial event structure of the model with
event actual status, but a bidirectional one because
at each instant the system has to look at the actual
event status - the behaviour, evaluate it and adapt
the events structure on the basis of this status.
Use and Behaviour Knowledge
'Use Process Knowledge' is represented by means of
Use Process Ontology, a structure based on Use Pro-
cess Entities, qualified by a system of Use Process
Rules (Trento 2013). On one hand these process rules
govern activities planning and on the other hand
SPATIAL REASONING AND ONTOLOGIES - Volume 2 - eCAADe 34 | 323
they control the relationship with the rest of knowl-
edge realms: who does what, where, when and how.
Use Process Knowledge can be described by
means of process classes, at different levels of aggre-
gation:
• Use Process Actions: elementary class entities
structuring the Use Process Ontology. They
represent the process based on user's mini-
mum ergonomic function.
• Use Process Activities: a set of Use Process Ac-
tions structured in time and space, oriented
by the functional programme. They qual-
ify the relation between users and building
(spaces, components, facilities, equipment,
etc.).
• Use Process Rationale: aggregation of Use
Process Activities. The importance of repre-
sentation for use rationale has been recog-
nized but it is a more complex issue that ex-
tends beyond artefact function. It is function
of social-economical-environmental sustain-
ability.
• Events: particular process entities, "mile-
stones" that occur in the dynamics of the ac-
tivities. Emergencies necessary to structure
the causal and dependency relationship be-
tween Use Process entities".
IMPLEMENTATION PROCESS AND EX-
PECTED RESULTS
The research project is going on, according to the fol-
lowing steps:
1. Study of the knowledge domain related to the
re-use of built heritage and identification of
the knowledge to be modelled for the pur-
pose of the platform;
2. Definition of a knowledge-based model for
the formalization of the knowledge related to
the heritage artifact and its intended use and
users;
3. Formalization of this knowledge;
 4. Selection and definition of simulative ap-
proaches and models for the predictions of
the use processes and of the related building
performances (space functionality, energy be-
havior, structural behavior, fire egress, etc.);
5. Integration of the Knowledge modelling en-
vironment and the Simulations environment
within the platform;
6. Selection of some case studies and experi-
mental application of the platform (to be per-
formed recursively for the platform calibra-
tion);
7. Verification, validation and critical analysis of
the platform and its functioning.
For implementing this theoretical model, we are us-
ing ontologies plus agents (upper ontology level) in
order to model, the design use events entities, phys-
ical or abstract, and their space-time relationships
structured by means of M-P-R (Carrara et al. 2009)
meanings, properties (defining their state) and rules
(relations, reasoning rules, consistency, best prac-
tices).
The new challenge now is to represent by means
of the same knowledge structure M-P-R, the right
entities and their relations, in order to manage in
a CAAD environment, human behaviour simulation
needs to be taken into account including psycholog-
ical aspects.
Analysis, checking, evaluation and control of
concepts associated to specific entities are per-
formed by means of inferential engine demons, with
deductive 'If-Then' type procedures.
A system of engines will work on a deductive
layer overlapped at the actual BIM level, allowing the
designers to use in a coherent manner different levels
of abstraction.
The implementation steps are:
1. Represent Design Knowledge of Event Ontol-
ogy (e.g. expressed in OWL language);
2. Connect Event Ontology with actual BIM, or
IFC (by means of API, or using Beetz (2009)
transcription of IFC in OWL);
324 | eCAADe 34 - SPATIAL REASONING AND ONTOLOGIES - Volume 2
3. Connect BIM+ Ontologies with a Narrative en-
vironment (e.g. BPM, Virtools, etc.)
4. Find out in the Semantic Web community
or build an inference engine to perform the
user's behaviour.
The dynamic and semantically-specific representa-
tion together with Inference and Simulation Engines
are able to predict human behaviour, so coherent/-
favourable situations will be evaluated by means of a
set of constraints, and will be highlighted and man-
aged in real time.
At the same time, this model allows actors to as-
sess alternatives, more consciously reflecting on the
consequences of their intents.
By this way the impact of a networked ontology
makes actors more aware of overall design problems
and allows them to operate more participative and
shared choices.
CONCLUSIONS
This paper reports on theoretical contents and some
early implementation patterns developed in the gen-
eral framework of an on-going research aimed at the
definition of a new approach for modelling and test-
ing knowledge related to the users behaviour in a
building environment, oriented to support assistive
systems for management and performance simula-
tions.
The research agrees on introducing hybrid ap-
proaches to take advantage of each technique's best
strengths. Combining ontology-based context rea-
soning with data-driven algorithms has shown to be
a promising path to be explored.
This paper proposes how - in a very general case
study - data-driven techniques, ontology taxonomy
and reasoning agents can be an efficient resource
for assisting designers in decision making processes
along recursive building design sessions, performing
event based simulation of human behaviour in a de-
fined building environment.
At present the proposed general framework has
been defined a framework implementation that can
count on a limited number of building entities for-
malized by means of current ontology editing sys-
tems in order to be used for design reasoning, using
the large family of ready-built inference engines and
information extraction tools.
The specific field of application will be the design
of building renovations - such as the ones belonging
to Italian real estate public assets "Agenzia del De-
manio" - in accordance with a predefined Use Func-
tional Program and/or modifying it, by rescheduling
activities and behaviours.
REFERENCES
Carrara, G., Fioravanti, A., Loffreda, G. and Trento, A.
2009 'An Ontology-Based Knowledge Representa-
tion Model for Cross-disciplinary Building Design -
A General Template', Proceedings of eCAADe 2009, Is-
tanbul, pp. 367-373
Chastain, T., Kalay, Y.E. and Peri, C. 2002, 'Square peg in
a round hole or horseless carriage? Reflections on
the use of computing in architecture', Automation in
Construction, 11 (2), pp. 237-248
Chen, L. and Nugent, C.D. 2009, 'Ontology-based ac-
tivity recognition in intelligent pervasive environ-
ments', International Journal of Web Information Sys-
tems (IJWIS), 5, 4, pp. 410-430
Fioravanti, A., Loffreda, G. and Trento, A. 2011, 'An in-
novative comprehensive knowledge model of archi-
tectural design process', International Journal of De-
sign Sciences & Technology, 11 (2), pp. 1-18
Gargaro, S. and Fioravanti, A. 2013 'A Context-
Knowledge Model for Architectural Design A holis-
tic approach by means of artificial intelligence tech-
niques', Proceedings of eCAADe 2013, Delft, pp. 81-90
Ioannidis, D., Tzovaras, D. and Malavazos, C. 2012 'Oc-
cupancy and Business Modelling', Proceedings of Eu-
ropean Conference on Product and Process Modelling
(ECPPM) - 3rd Workshop on eeBDM, eeBIM, Reykjavik
Remagnino, P., Hagras, H., Monekosso, N. and Velastin,
S. 2005, 'Ambient intelligence', in Remagnino, P.,
Foresti, G. and Ellis, T. (eds) 2005, Ambient intelli-
gence, Springer, New York, pp. 1-14
Diaz Rodrigues, N.A. 2015, Semantic and Fuzzy Modelling
for Human Behaviour Recognition in Smart Spaces. A
Case Study on Ambient Assisted Living, Ph.D. Thesis,
TUCS - Turku Centre for Computer Science
Ros, M., Cuellar, M.P., Delgado, M. and Vila, A. 2013, 'On-
line recognition of human activities and adaptation
to habit changes by means of learning automata
SPATIAL REASONING AND ONTOLOGIES - Volume 2 - eCAADe 34 | 325
and fuzzy temporal windows', Information Sciences,
220, 20 , p. 86–101
Tabak, V. 2009, User Simulation of Space Utilisation. Sys-
tem for Office Building Usage Simulation, Ph.D. Thesis,
Technische Universiteit Eindhoven
Trento, A. and Fioravanti, A. 2014 'Human Activity Mod-
elling Performed by Means of Use Process Ontolo-
gies', Proceedings of eCAADe 2014, Newcastle, pp.
385-394
Trento, A., Fioravanti, A. and Simeone, D. 2013 'Building-
Use Knowledge Representation for Architectural
Design. An ontology-based implementation', Pro-
ceedings of eCAADe 2013, Prague, pp. 683-689
Wurzer, G. 2010, 'Schematic Systems - Constraining
Functions Through Processes (and Vice Versa)', Inter-
national Journal of Architectural Computing, 8 (2), pp.
201-217
Wurzer, G., Fioravanti, A., Loffreda, G. and Trento, A.
2010 'Function & Action: Verifying a functional pro-
gram in a game-oriented environment', Proceedings
of eCAADe 2010, Zurich, pp. 389-394
Zimmermann, G. 2006 'Multi-Agent Model to Multi-
Process Transformation - A housing market case
study.', Proceedings of 8th International Design and
Decision Support Systems Conference, Einhoven, pp.
203-219
[1] https://www.worldatwork.org/adimLink?id=78679
[2] http://www.us.jll.com/united-states/en-us/Docum
ents/Workplace/JLL-workplace-fit-for-future.p
df
[3] http://www.buildingsmart-tech.org/specification
s/ifc-overview
[4] http://www.uns.ethz.ch/edu/teach/masters/spring
/ebcdm/scripts/ebcd_ind_behav_I_FS_2015.pdf
[5] http://www.w3.org/Submission/SWRL/
[6] http://spinrdf.org/
SHAPE, FORM AND GEOMETRY |
Grammars and Concepts
An Algebraic Approach to Implementing a Shape Grammar
Interpreter
Rudi Stouffs1
1
National University of Singapore; Delft University of Technology
1
stouffs@nus.edu.sg

Shape grammars come in a variety of forms. Algebras of shapes have been

defined for spatial elements of different kinds, as well as for shapes augmented
with varying attributes, allowing for grammar forms to be expressed in terms of a
direct product of basic algebras. This algebraic approach is extended here to the
algebraic derivation of combinations of basic shape algebras with attribute
algebras. This algebraic abstraction at the same time serves as a procedural
abstraction, giving insights into the modular implementation of a general shape
grammar interpreter for different grammar forms.

Keywords: shape grammars, shape algebras, parallel grammars, compound

shapes, implementation

INTRODUCTION Grammar formalisms for design come in a large va-

A shape grammar is a formal rewriting system for pro-
ducing languages of shapes (Stiny and Gips 1972;
Stiny 1980). A shape grammar is typically specified
to consist of a set of productions, or shape rules, op-
erating over a vocabulary of (terminal) spatial ele-
ments and a vocabulary of (non-terminal) symbols or
markers, e.g., labels, and to include an initial shape as
the starting point in the productive (generative) pro-
cess (Stiny 1980; Yue and Krishnamurti 2013). Then,
a shape is defined as any composition of spatial ele-
ments and, optionally, symbols from the respective
vocabularies. A shape rule is commonly expressed
in the form a → b, with both a and b constituting
shapes, such that the application of the rule to a
shape s under a (similarity) transformation t yields the
shape s − t(a) + t(b), with the condition that t(a) ≤ s.
The language defined by a shape grammar is the set
of shapes generated by the grammar that do not con-
tain any (non-terminal) symbols.
riety, requiring different representations of the en-
tities being generated, and different interpretative
mechanisms for this generation. Shape grammars
also come in a variety of forms, even if less broadly.
Most examples of shape grammars rely on labeled
shapes, a combination of line segments and labeled
points (in two dimensions) (Stiny 1980). Stiny (1992)
proposes numeric weights as attributes to denote
line thicknesses or surface tones. Knight (1989; 1993)
considers a variety of qualitative aspects of design,
such as color, as shape attributes. Stiny (1981) also
proposes to augment a shape grammar with a de-
scription function in order to enable the construction
of (intended) descriptions of designs.
Implementing a shape grammar interpreter requires
implementing the part relationship for shapes—with
or without attributes—, the operations of sum and
difference on shapes, and solving the matching prob-
lem, that is, identifying under which transformation a

SHAPE, FORM AND GEOMETRY | Grammars and Concepts - Volume 2 - eCAADe 34 | 329
 rule may apply to a given shape. Beirão (2012, 228-
236) offers a survey of (implementations of ) shape
grammar interpreters so far, and must conclude that
they have common limitations. Many of them com-
pute only on two-dimensional shapes; and most do
not apply subshape detection and therefore do not
support emergence. Also, "very few shape grammar
interpreters allow for the implementation of rules op-
erating with symbols" (Beirão 2012, 235), never mind
other attributes, or a description function.
In this paper, we will address the problem of
developing an implementation of a shape grammar
interpreter supporting varying shape grammar for-
malisms, by focusing on the implementation of par-
allel and compound shape grammars. We will re-
view the literature on parallel and compound shape
grammars and propose an algebraic treatment facili-
tating a modular approach, based on a similarity be-
tween algebraic abstraction and procedural abstrac-
tion (Frank 1999).
PARALLEL AND COMPOUND SHAPE
GRAMMARS
Originally, a parallel shape grammar was defined by
Stiny (1975, 37) and Gips (1975, 7) as a shape gram-
mar intended to be used in the parallel generation
of shapes, that is, "whenever a shape rule is used, it
is applied simultaneously to every part of the shape
to which it is applicable" (Gips 1975, 7; Stiny 1975,
37). This is in contrast to the more common serial
application of shape rules, where at each step of the
generation a shape rule is applied to only one part of
a shape. However, more recently, the term parallel
(shape) grammar has been adopted in the context of
parallel computations on multiple descriptions.
Li (1999) defines a parallel grammar as a gram-
mar operating on different descriptions with the ob-
jective to resolve parametric dependencies. For ex-
ample, consider a shape grammar generating a plan
where the number of rooms is depended on the se-
lection of rules applied. In this process, it is almost
impossible to constrain the boundaries of the plan at
the same time, as this would require the room sizes
330 | eCAADe 34 - SHAPE, FORM AND GEOMETRY | Grammars and Concepts - Volume 2
to be made dependent on the total number of rooms,
which is only known at the end of the production pro-
cess. Instead, Li (1999) suggests adopting (at least)
two descriptions, the first one a diagram with the
number of rooms as independent parameter(s), and
the second one the plan with the room sizes as de-
pendent parameters (of the number of rooms). By
staging the production of the diagram before the
production of the plan, the assignment of values to
the dependent parameters can be postponed until
after the assignment of values to the independent
parameters.
Duarte (2001) defines a parallel grammar as sep-
arating different representations or aspects of a de-
sign into different computations that interact with
each other. Specifically, Duarte (2001; 2005) con-
siders a discursive (parallel) grammar incorporat-
ing a shape grammar and a (textual) description
grammar−as well as a set of heuristics, where the lat-
ter is intended to constrain the rules that are appli-
cable at each step of the design generation. While
the shape grammar operates on shapes and the de-
scription grammar on textual (including numeric) de-
scriptions, their rules are commonly coupled, with
the description rule part constraining the shape rule
part. This combination of a shape grammar and a
description grammar follows Stiny's (1981) definition
of a description function to augment a shape gram-
mar in order to construct design descriptions. Where
Stiny (1981) considers a description function as made
up of functions, with each function assigned to a
shape rule and computing in parallel to the shape
rule, Duarte instead denotes the functions as descrip-
tion rules and the description function as a descrip-
tion grammar. Otherwise, these operate in exactly
the same way and with the same intention. In fact, al-
though Stiny nowhere adopts the term parallel gram-
mar in this sense, Knight (1999; 2003a) explicitly at-
tributes the concept to Stiny (1981). Knight (2003a)
also offers a definition of a parallel grammar as "a net-
work of two or more grammars that operate simulta-
neously."
Admittedly, Li (1999) and Knight's (2003a; also,
Duarte 2001) interpretation are not unequivocal.
Following Stiny's (1990) definition of a design as
"an element in an n-ary relation among drawings,
other kinds of descriptions, and correlative devices
as needed," Li (2001; 2004) considers seven drawings
(from plan diagram to plan, section and elevation)
and nine descriptions (specifying measures of width,
depth, and height, among others), in his specifica-
tion of a shape grammar for (teaching) the architec-
tural style of the Yingzao fashi. However, he consid-
ers only four grammar components to define his par-
allel grammar, corresponding to four stages in the
production (Li 1999). However, in Knight's (2003a)
parallel grammar interpretation of Stiny's (1981) ex-
ample, the two grammar components of the parallel
grammar−the shape grammar and the description
function−apply hand in hand: "the rules of a paral-
lel grammar may be linked so that the application
of a rule in one grammar triggers the application of
one or more rules in other grammars" (Knight 2003a).
In fact, Knight (2003a) suggests the same interpreta-
tion for Li's (2001; 2004) grammar, with each drawing
and description specifying a component grammar in
the parallel grammar, for each stage. Duarte (2001)
takes a similar position when describing his discur-
sive grammar applied to the houses designed by the
architect Alvaro Siza at Malagueira. Even though,
strictly speaking, he only refers to two grammars−a
shape grammar and a description grammar−, he
specifically acknowledges that both grammars in-
clude several sub-grammars. "These sub-grammars
correspond to viewpoints in the shape grammar (e.g.
first floor plan), and to features in the description
grammar (e.g. morphology)" (Duarte 2001). View-
points define separate drawings (sketches, plans, el-
evation, envelope, etc.) and features individual de-
scriptions.
When the rules of a parallel grammar are linked,
the linked rules can be expressed as one compound
rule (Knight 2003a). Therefore, some authors con-
sider compound grammars as an alternative term
to parallel grammars, though others adopt the term
compound grammars also to denote compositions
SHAPE, FORM AND GEOMETRY | Grammars and Concepts - Volume 2 - eCAADe 34 | 331
of grammars that do not operate in parallel. For ex-
ample, Beirão (2012) uses the term compound gram-
mar to denote a composition of several discursive
grammars, where each discursive grammar formal-
izes a so-called Urban Induction Pattern (UIP), en-
coding a typical urban design operation or design
move. Knight (2004), instead, suggests a distinction
between synchronized and a-synchronized parallel
grammars, where the latter allows for sequential pro-
duction stages as proposed by Li (1999; 2001).
SHAPE ALGEBRAS
While Stiny avoids the term parallel grammar when
referring to parallel computations, he does empha-
size parallel computations in multiple algebras. For
example, when a rule applies to a shape composed
of points and line segments, though the rule may
require both one or more points and one or more
line segments to be present within a prescribed spa-
tial relationship, the rule computes with points and
line segments in parallel. The overall shape rule com-
putation actually combines two computations−one
with shapes of line segments and one with shapes of
points−"that are carried out in parallel and influence
one another mutually" (Stiny 1990).
Points and line segments, and by extension other
spatial elements, can be considered to adhere to an
algebra (specifically, a generalized Boolean algebra
(Krstic 1999)), that is ordered by a part relation ('≤')
and closed under the operations of sum ('+'), prod-
uct ('·'), and difference ('−'), as well as relevant trans-
formations. For example, points may belong to the
algebra U02 and line segments to the algebra U12
(in two dimensions) (Stiny 1992); Uij denotes the al-
gebra of spatial elements of dimension i, e.g., 0 for
points, 1 for line segments, 2 for plane segments, em-
bedded in a space with dimension j, e.g., 1-D, 2-D,
3-D. Stiny (1992) extends the notion of algebras to
shapes with attributes: labeled points belong to the
algebra V02 , while weighted line segments belong
to the algebra W12 . Then, shapes of line segments
and labeled points can be said to belong to an alge-
bra that is the direct product of the algebras U12 and
 V02 , U12 ×V02 . Consequently, a shape rule applying
to shapes of line segments and labeled points can be
considered to combine two shape rules, one in U12
applying to line segments and one in V02 applying
to labeled points.
Any selection of shape algebras, including la-
beled and weighted shape algebras, can be com-
bined using the direct product into a compound al-
gebra of compound shapes that are made up of a
mix of various spatial elements, and optionally aug-
mented with labels or weights. Chase (1999) notes
that "this is common in maps, as map features may
be distinguished by different element types (for ex-
ample, lines representing roads, points representing
cities), or labels used to distinguish elements with
the same basic geometry but different semantics (for
example, lines can represent roads and rivers)." But,
examples in architectural representation abound as
well. Compound shapes may be expressed in unions
of the sets that form shapes from different algebras,
with the understanding that basic and augmented
spatial elements only interact if they are of the same
kind (same dimension and attribute kind, if any), and
are independent otherwise (Stiny 1992).
This notion of compound shapes, as resulting
from compound algebras defined by the direct prod-
uct of basic shape algebras, is conducive to a modu-
lar (or procedural) implementation of a shape gram-
mar interpreter (e.g., Frank 1999). Each basic shape
algebra can define a single module (or a class in an
object-oriented programming paradigm), and mod-
ules can be combined to define compound algebras,
facilitating a variety of shape grammar formalisms.
The algebra's signature specifies the operations
of the algebra, at a minimum, the operations of sum
('+'), product ('·'), and difference ('−'). The allowable
transformations can be considered external to the al-
gebra, as part of the signature or as part of the al-
gebra's carrier, i.e., the set of elements of the alge-
bra (Krstic 1999; 2012). The part relation ('≤') is not
an operation of the algebra but can be expressed in
terms of the operation of product: t(a) ≤ s ⇔ t(a) · s =
t(a). From a modular implementation point of view,
332 | eCAADe 34 - SHAPE, FORM AND GEOMETRY | Grammars and Concepts - Volume 2
the algebra's signature contributes to the module's
interface (the class methods in object-oriented pro-
gramming), but the interface can be extended to in-
clude, among others, the part relation. Nevertheless,
having a (even partially) uniform interface, i.e., shar-
ing the same class methods, greatly eases the imple-
mentation of a general shape grammar interpreter.
Unfortunately, as Knight (2003b) acknowledges,
"the algebras that designers use, informally or for-
mally, are many." Beyond labels and weights, shapes
can be augmented with attributes of any kind: "aes-
thetic, formal, functional, structural, and so on. For
example, points can have diameters, lines can have
thicknesses, planes can have colors, solids can have
materials" (Knight 2003b). "The only condition is that
the operators of any shape algebra are defined on
all its elementary objects, are recursively applicable,
and are closed" (Yue and Krishnamurti 2013). Fortu-
nately, the notion of an algebra as derived from exist-
ing algebras can be extended to augmented shapes,
e.g., of labeled points (Stouffs 2008). Defining an al-
gebra for labels is straightforward. Similar to points,
a label can be said to be part of another label only
if these are identical. Then, the operations of sum,
product, and difference correspond to the set oper-
ations of union, intersection and difference. Labels
do not exhibit any allowable transformations, unless
we consider case transformations. However, the op-
eration of direct product on algebras will not support
an attribute behavior, and it is not straightforward to
consider an alternative operation on algebras. Krstic
(1999) offers an unintentional hint.
ALGEBRAIC ABSTRACTIONS
Krstic (1999) notes the difference between an alge-
bra of (maximal) spatial elements and an algebra of
shapes. The former is a partial algebra as the opera-
tions of sum, product, and difference are not closed.
The sum of two spatial elements is a spatial element
only if the two elements overlap or are both part of
another spatial element and have boundaries that
overlap. In general, the operations of sum, product,
and difference on spatial elements are only defined
if the spatial elements exist in the same subspace,
where the dimension of the subspace equals the di-
mension of the spatial elements. For example, two
line segments must be part of the same infinite line,
two plane segments of the same infinite plane, and
two volumes of the same 3D hyperplane. This sub-
space is denoted the carrier of the spatial element,
but in order to avoid any confusion with the carrier of
an algebra, we will instead refer to the co-descriptor of
the subspace. Then, the operations of sum, product,
and difference on spatial elements are defined only
if the spatial elements have the same co-descriptor,
that is, they are co-equal. Note from above that, in
and of itself, this is not a sufficient condition for the
operations to be closed.
Having established the difference, Krstic (1999)
goes on to focus solely on shape algebras. How-
ever, it is possible to derive shape algebras from par-
tial algebras of spatial or other elements in a general
way. First, let us assume a two-sorted partial alge-
bra with carrier {A, P(A)} and signature including the
operations of combine, common and complement on
members of A. We consider a two-sorted algebra be-
cause we want the operations of combine, common,
and complement to extend upon the respective op-
erations of sum, product, and difference, and at the
same time be closed for co-equal (spatial) elements.
For example, the combine of two co-equal spatial el-
ements that do not have overlapping boundaries will
be the set of the two spatial elements. Thus, each
of the operations of combine, common, and comple-
ment takes as argument two elements of A and return
an element of P(A), the set of all subsets of A.
Deriving Algebras from Two-sorted Partial
Algebras
To derive a shape algebra from this two-sorted partial
algebra, we need to distinguish the desired behavior.
Each behavior will result in a different derivation. For-
tunately, we can reuse behaviors for different kinds of
spatial or other elements. The simplest behavior is a
discrete behavior, applying to both points and labels.
An algebra with carrier P(A) and signature in-
SHAPE, FORM AND GEOMETRY | Grammars and Concepts - Volume 2 - eCAADe 34 | 333
cluding sum ('+'), product ('·'), difference ('−'), and re-
duce ('r') can be defined for a discrete behavior as fol-
lows:
∀ X, Y ∈ P(A) ⇒
X +Y =X ∪Y
X ·Y =X ∩Y (1)
X −Y = X \Y
r(X) = X
In a discrete behavior, the operations of sum,
product, and difference correspond to the normal set
operations of union, intersection and difference. The
reduce operation reduces any set to a set of maxi-
mal elements. Under a discrete behavior, any set is
maximal because any duplicates are automatically re-
moved. The algebras U0 (shapes of points; we omit
the dimension of the embedding space for simplicity)
and L, sets of labels, can be defined in this way. Note
that descriptions, from an algebraic point of view, be-
have exactly as labels and, thus, the algebra D of sets
of descriptions can be defined in this way as well.
Before we address other spatial elements, let us
first consider a behavior for weights (e.g., line thick-
nesses or surface tones), as is apparent from drawings
on paper−a single line drawn multiple times, each
time with a different thickness, appears as if it were
drawn once with the largest thickness, even though
it assumes the same line with other thicknesses (Stiny
1992).
An algebra with carrier P1 (A), the set of all sin-
gleton subsets of A, and signature including sum ('+'),
product ('·'), difference ('−'), and reduce ('r') can be
defined for an ordinal behavior, applying to weights,
in terms of the two-sorted partial algebra with car-
rier {A, P(A)} and signature including the operations
of combine, common and complement, as follows:
∀ {x}, {y} ∈ P1 (A) ⇒
{x} + {y} = combine(x, y)
{x} · {y} = common(x, y) (2)
{x} − {y} = complement(x, y)
r({x}) = {x}
For weights, we know that the result of the op-
erations of combine, common, and complement on
 two singleton weights is always a singleton weight.
We use this knowledge to define the operations of
sum, product, and difference in terms of the opera-
tions of combine, common, and complement. Again,
the reduce operation results in the argument (single-
ton) set itself. The algebra N of singletons of numeric
weights can be defined in this way.
Deriving a shape algebra for spatial elements
other than points from a two-sorted partial algebra
is a little bit more complicated because of the need
to consider co-equal shapes. We take a two-step ap-
proach. First, we derive a sub-algebra for co-equal
shapes of spatial elements, next we define a shape
algebra for a single type of spatial elements from this
sub-algebra.
A sub-algebra with carrier P(A) and signature in-
cluding co-combine, co-common, co-complement,
and co-reduce can be defined for an areal behavior
as follows:
∀ X, Y ∈ P(A) :
∀ x ∈ X, ∀ y ∈ Y , co(x) = co(y) ⇒
co-combine(X,
 Y) =  ments with respect to another co-shape can be com-
∪  outside(b(X), Y )
construct  outside(b(Y ), X) 

same-side(b(X), b(Y ))
co-common(X, Y) =  only serve as abstractions of the actual procedures,
∪  outside(b(X), Y )
construct inside(b(Y ), X)  (3)

same-side(b(X), b(Y ))
co-compliment(X,
  Y)=  Then, a shape algebra with carrier P(A) and sig-
∪ outside(b(X), Y )
construct inside(b(Y ), X)  nature including sum ('+'), product ('·'), difference

opposite-side(b(X), b(Y ))
co-reduce(X)
{ =
co-combine({x}, co-reduce(X \ x)) ∃ x ∈ X
∅ otherwise
The operations of co-combine, co-common, co-
complement, and co-reduce only apply to co-equal
shapes. Instead of comparing the shapes for co-
equality, it is checked that all spatial elements have
the same co-descriptor ('co'). Then, the operations
can be expressed in terms of the boundaries ('b')
of each co-equal shape (Krishnamurti and Stouffs
2004; Stouffs and Krishnamurti 2006), here termed
co-shape. Specifically, 'outside(b(X), Y)' returns the
collection of boundaries of co-shape X that lie out-
side of co-shape Y. Similarly, 'inside(b(Y), X)' returns
334 | eCAADe 34 - SHAPE, FORM AND GEOMETRY | Grammars and Concepts - Volume 2
the collection of boundaries of co-shape Y that lie
outside of co-shape X. 'same-side(b(X), b(Y))' denotes
the collection of boundaries of both co-shapes X and
Y where the interiors of the respective co-shapes lie
on the same side of the boundary, and 'opposite-
side(b(X), b(Y))' the collection of boundaries of both
co-shapes X and Y where the interiors of the respec-
tive co-shapes lie on opposite sides of the bound-
ary. Then, the boundary of the co-shape result-
ing from the co-combine operation is formed by
the 'outside(b(X), Y)', 'outside(b(Y), X)', and 'same-
side(b(X), b(Y))' collections, and the co-shape can be
constructed from the union of these collections. The
co-common and co-complement operations are sim-
ilarly defined. In the case of the co-reduce oper-
ation, each spatial element in the co-shape is co-
combined with the co-reduced remainder of the co-
shape. From an implementation point of view, this
recursive definition of co-reduce may not be the most
efficient; actually, the same can be said about the
other operations−the classification of boundary seg-
puted once for all of the classes inside, outside, same-
side and opposite-side. Obviously, these definitions
we refer to Stouffs and Krishnamurti (2006) for actual,
and efficient, algorithms.
('−'), and reduce ('r') can be defined for an areal be-
havior, applying to line segments, plane segments
and volumes, in terms of the sub-algebra with car-
rier P(A) and signature co-combine, co-common, co-
complement, and co-reduce, as follows:
∀X, Y ∈ P(A) ⇒
X + Y = r(X ∪ Y ) ( )
∪ {x ∈ X : co(x) = c},
X · Y = c co-common
{y ∈ Y : co(y) = c}
X − Y ∪= (4)
∪  c {x ∈ X : co(x) =(c ∧ ¬∃y ∈ Y : co(y) =)c}
∪ {x ∈ X : co(x) = c},
 c co-complement {y ∈ Y : co(y) = c}
∪
r(X) = c co-reduce({x ∈ X : co(x) = c})
The operations of product, difference and reduce
are expressed directly in terms of the operations of
co-common, co-complement and co-reduce on the
respective co-shapes. In the case of complement, co-
shapes from X for which there exists no (co-equal) co-
shape in Y also form part of the result. A similar ap-
proach can be taken for the operation of sum, how-
ever, for simplicity, we prefer to express the opera-
tion of sum in terms of the operation of reduce on
the combined sets of spatial elements. Note that the
algebras U1 (shapes of line segments), U2 (shapes of
plane segments) and U3 (shapes of volumes) can all
be defined in this way.
We should note that while an areal behavior ap-
plies to shapes of line segments as well, from an im-
plementation point of view, it would be more effi-
cient to define an interval behavior for shapes of line
segments. Additionally, while these behaviors cover
shapes of different kinds of spatial elements, and sets
of labels and singletons of weights, other behaviors
can be identified to apply to other kinds of attributes,
for example, for material rankings (Knight 1993). In-
stead, we will now continue to derive compositions
of algebras under the direct product.
Deriving Algebras with the Direct Product
The direct product applies to all algebras that share
the same signature, specifically, the algebras U0 , U1 ,
U2 , U3 , L, D, and N , we have previously defined.
We will address some implications of this in the dis-
cussion below. Here, we define a shape algebra with
carrier P(A) × P(B) and signature including sum ('+'),
product ('·'), difference ('−'), and reduce ('r') in terms
of the shape algebras with carriers P(A) and P(B) and
identical signatures, as follows:
∀ (A1 , B1 ), (A2 , B2 ) ∈ P(A) × P(B) ⇒
(A1 , B1 ) + (A2 , B2 ) = (A1 + A2 , B1 + B2 )
(A1 , B1 ) · (A2 , B2 ) = (A1 · A2 , B1 · B2 ) (5)
(A1 , B1 ) − (A2 , B2 ) = (A1 − A2 , B1 − B2 )
r(A1 , B1 ) = (r(A1 ), r(B1 ))
An algebra of shapes of points and line seg-
ments, U0 × U1 , and an algebra of shapes of line
segments and sets of descriptions, U1 × D, among
others, can be defined in this way.
SHAPE, FORM AND GEOMETRY | Grammars and Concepts - Volume 2 - eCAADe 34 | 335
Deriving Augmented Shape Algebras
Next, we can tackle the issue of an algebra for aug-
mented shapes. Rather than referring to a shape al-
gebra and an attribute algebra (e.g., L or N ), we will
instead refer to the partial algebra for the spatial ele-
ments, just as we define the shape algebra from the
partial algebra of its spatial elements. The behav-
ior of an augmented shape algebra, after all, mimics
the behavior of the underlying shape algebra, with a
few differences. For example, consider two overlap-
ping line segments. Without attributes, these com-
bine. With attributes, they combine only if they share
the same attributes or, otherwise, of the segments
are identical. Otherwise, different segments (or parts
thereof ) will necessarily have different attributes and
will need to be represented separately. The behav-
ior of the attribute shapes, for each of the different
line segments (or parts thereof ), however remains
the same.
An algebra with carrier P(A × P(B)) and signa-
ture including sum ('+'), product ('·'), difference ('−'),
and reduce ('r') can be defined for a discrete behavior,
in terms of the (attribute) algebra P(B) with signature
including sum ('+'), product ('·'), difference ('−'), and
reduce ('r'), as follows:
∀ X, Y ∈ P(A × P(B)) ⇒
X + Y =
∪ {(x, Bx + By ) : (x, Bx ) ∈ X ∧ (x, By ) ∈ Y }
{(x, Bx ) : (x, Bx ) ∈ X ∧ ¬∃(x, By ) ∈ Y }

{(x, By ) : (x, By ) ∈ Y ∧ ¬∃ (x, Bx ) ∈ X}
X ·Y =
{(x, Bx · By ) : (x, Bx ) ∈ X ∧ (x, By ) ∈ Y } (6)
X −{ Y =
∪ {(x, Bx − By ) : (x, Bx ) ∈ X ∧ (x, By ) ∈ Y }
{(x, Bx ) : (x, Bx ) ∈ X ∧ ¬∃ (x, By ) ∈ Y }
r(X)
{ =
{(x, Bx )} + r(X \ (x, Bx )) ∃ (x, Bx ) ∈ X
∅ otherwise
Comparing this to the discrete behavior for a
non-augmented shape algebra, we observe that for
the operations of sum, product and difference, if a
spatial element is shared between both (augmented)
shapes, we combine both attributes of the spatial
element under the same operation. In the case of
the operations of sum and difference, we may need
to add spatial elements, with their original attribute,
 that belong to one augmented shape but not the entirety. Instead, we need to apply the classifica-
other. We express the operation of reduce in terms tion and construction to each pair of spatial elements
of the operation of sum for the same algebra. An al- from the respective co-shapes, in order to be able to
gebra of shapes of labeled points, V0 , or shapes of assign the appropriate attributes−as defined by the
weighted points, W0 , with the weights representing operation of sum, product or difference applied to
grey-scales or diameters, can be defined in this way. the respective attributes of the spatial elements. This
Having established an augmented shape algebra definition relies on the specification of helper func-
for points, a demonstration of an augmented shape tions 'common-sum', 'common-product', 'common-
algebra for other spatial elements, using the areal be- difference', 'complement', and 'co-r', defined below.
havior, remains. In fact, where the shape algebra for The first three helper functions return the common
an areal behavior is expressed in terms of the sub- shape of two co-equal spatial elements, with as at-
algebra for co-equal shapes, we can define a sub- tribute, respectively, the sum, product and difference
algebra for co-equal augmented shapes and retain of the respective attributes. The 'complement' re-
the definition for the shape algebra as a definition for turns the complement shape of a spatial element
the augmented shape algebra as well. The only diffi- with respect to a co-equal shape, with as attribute
culty is the use of the co-descriptor function ('co') on the original attribute of the spatial element. Finally,
elements of the shape algebra. However, if we over- 'co-r' is a variant of the operation of co-reduce that
load the co-descriptor function to accept augmented assumes that none of the spatial elements overlap,
spatial elements, i.e., elements with attributes, then though they may share boundaries. Two spatial ele-
there is no issue. ments that share boundary may be combined if they
A sub-algebra with carrier P(A × P(B)) and also share the same attribute(s).
signature including co-combine, co-common, co- common-sum(X,
 Y)=  
complement, and co-reduce can be defined for an ∪ ∪   (x, Bx ) ∈ X∧ 
y (z, B x + B y ) : (y, By ) ∈ Y ∧
areal behavior, in terms of the two-sorted partial al- x  
z ∈ co-common(x, y)
gebra with carrier {A, P(A)} and signature including common-product(X,
 Y ) = 
∪ ∪   (x, Bx ) ∈ X∧ 
the operations of combine, common and comple- (y, By ) ∈ Y ∧
x y (z, Bx · By ) :  
ment, and the (attribute) algebra P(B) with signature z ∈ co-common(x, y)
common-difference(X, Y ) =
including sum ('+'), product ('·'), difference ('−'), and  
∪ ∪   (x, Bx ) ∈ X∧ 
reduce ('r'), as follows: (z, B − B ) : (y, B ) ∈ Y ∧
x y x y

y
z ∈ co-common(x, y)
 (8)
∀ X, Y ∈ P(A × P(B)) :
∀(x, Bx ) ∈ X, ∀(y, By ) ∈ Y , co(x) = co(y) ⇒ complement(X,
{ Y{ ) = }
co-combine(X, ∪ (x, Bx ) ∈ X∧
  Y)=  x (z, Bx ) : z ∈ difference({x}, Y )
∪ common-sum(X, Y )
co-r complement(X, Y )  co-r(X) =
  
complement(Y, X) 
    
 (x, Bx ) ∈ X∧
co-common(X, Y ) = 
 {(z,  (y, Bx ) ∈ X∧

 B x )}∪ 
(7) 
co-r(common-product(X, Y )) co-r X\  x 6= y∧
co-complement(X, Y)= 

( { ) 
 {(x, Bx ), (y, Bx )} 
 {z} =
∪ common-difference(X, Y ) 
 
co-r 
 co-combine(x, y)
complement(X, Y ) 
∅ otherwise
co-reduce(X)
 =( )
 {(x, Bx )},
co-combine ∃ (x, Bx ) ∈ X An algebra of shapes of weighted line segments,
co-reduce(X\(x, Bx ))

∅ otherwise W1 , or shapes of labeled plane segments, V2 , among
We cannot simply determine a resulting co- others, can be defined in this way.
shape−from one of the operations of co-combine,
co-common, and co-complement−from the classifi-
cation of the boundaries of both co-shapes in their

336 | eCAADe 34 - SHAPE, FORM AND GEOMETRY | Grammars and Concepts - Volume 2
Discussion
We have adopted a constructive, algebraic approach
to defining shape algebras and augmented shape al-
gebras. This allows for a variety of algebras to be de-
fined from an array of basic (partial) algebras of spa-
tial and other elements. For example, we can de-
fine an algebra of shapes of line segments and la-
beled points, U1 × V0 , from basic partial algebras
of points, line segments and labels. Other opportu-
nities arise. Rather than restricting ourselves to la-
bels and weights as attributes, we can consider other
attributes, such as color, in the same way. Consider
a basic algebra of sets of colors, C, and assume we
express the combination of a shape algebra with an
attribute algebra with the operator '∧', termed attri-
bution, then we can define an algebra of shapes of
colored plane segments as U2 ∧ C. Similarly, we
can then write V0 = U0 ∧ L and W1 = U1 ∧
N . Additionally, we can define an attribute algebra
as a direct product of two basic attribute algebras,
e.g., U0 ∧ (L × N ) defines an algebra of shapes of
points with both labels and weights as attributes. We
may consider the operation of direct product to dis-
tribute over the operation of attribution, resulting in
(U0 ∧ L) × (U0 ∧ N ) = V0 × W0 .
While we commonly consider algebras of shapes
augmented with algebras of (sets of ) non-spatial
elements, and the direct product of (augmented)
shape algebras, other combinations are possible as
well. Consider the algebra D of sets of descriptions.
Though it is a non-spatial algebra, most authors con-
sider descriptions parallel to drawings, not as at-
tributes thereof. Nevertheless, Beirão (2012) consid-
ers descriptions as attributes to spatial objects, allow-
ing for objects to refer to alternate, though similar,
descriptions. Thus, U1 × D is as valid as U1 ∧ D. Of
course, some limitations in the way we combine alge-
bras may still be preferable, reflecting on the kinds of
design descriptions (whether drawings or other de-
scriptions) we tend to use.
Knight (2003b; see also, Krstic 2012) also con-
siders a sum of algebras that requires algebras to be
specified in the same space. For example, in the case
SHAPE, FORM AND GEOMETRY | Grammars and Concepts - Volume 2 - eCAADe 34 | 337
of a drawing of line segments and labeled points,
the segments and points operate in the same (2D or
3D) space and transformations of shapes of line seg-
ments and shapes of labeled points necessarily need
to go hand in hand. Instead, when shape algebras
operate in a different space (or drawing), transfor-
mations may differ. The sum of algebras applies in
the first case, the (direct) product in the second case.
However, as we left transformations out of the pic-
ture, the direct product of algebras as we defined it
is applicable to both cases. We refer to Krstic (2012)
for a treatment of transformations in the context of
shape algebras.
CONCLUSION
We presented an algebraic approach to describing
compound shapes that includes the definition of
non-spatial algebras and the combination of shape
algebras and non-spatial algebras under an opera-
tion of attribution. This algebraic abstraction serves
as a procedural abstraction for the modular imple-
mentation of a general shape grammar interpreter.
REFERENCES
Beirão, JN 2012, CItyMaker: designing grammars for urban
design, Ph.D. Thesis, Delft University of Technology
Chase, SC 1999, 'Supporting emergence in geographic
information systems', Environment and Planning B:
Planning and Design, 26(1), p. 33–44
Duarte, JP 2001, Customizing mass housing: a discursive
grammar for Siza's Malagueira houses, Ph.D. Thesis,
MIT
Duarte, JP 2005, 'Towards the mass customization
of housing: the grammar of Siza's houses at
Malagueira', Environment and Planning B: Planning
and Design, 32(3), p. 347–380
Frank, AU 1999, 'One step up the abstraction ladder:
combining algebras', in Freksa, C and Mark, DM (eds)
1999, COSIT’99, LNCS 1661, Springer, Berlin, pp. 95-
108
Gips, J 1975, Shape Grammars and their Uses: Artificial Per-
ception, Shape Generation and Computer Aesthetics,
Birkhäuser, Basel
Knight, TW 1989, 'Color grammars: designing with lines
and colors', Environment and Planning B: Planning
and Design, 16(4), p. 417–449
 Knight, TW 1993, 'Color grammars: the representation matics, 22(1), pp. 71-89
of form and color in design', Leonardo, 26(2), p. Stouffs, R and Krishnamurti, R 2006, 'Algorithms for clas-
117–124 sifying and constructing the boundary of a shape',
Knight, TW 1999 'Applications in architectural design, Journal of Design Research, 5(1), pp. 54-95
and education and practice', NSF/MIT Workshop on Yue, K and Krishnamurti, R 2013, 'Tractable shape gram-
Shape Computation, report, Cambridge, MA mars', Environment and Planning B: Planning and De-
Knight, TW 2003a, 'Computing with emergence', Envi- sign, 40(4), p. 576–594
ronment and Planning B: Planning and Design, 30(1),
p. 125–155
Knight, TW 2003b, 'Computing with ambiguity', Environ-
ment and Planning B: Planning and Design, 30(2), p.
165–180
Knight, TW 2004 'Interaction in visual design computing',
Visual and Spatial Reasoning in Design III, presenta-
tion, Cambridge, MA
Krishnamurti, R and Stouffs, R 2004, 'The boundary of
a shape and its classification', Journal of Design Re-
search, 4(1)
Krstic, D 1999, 'Constructing algebras of design', Environ-
ment and Planning B: Planning and Design, 26(1), pp.
45-57
Krstic, D 2012, 'Algebras of shapes revisited', in Gero,
JS (eds) 2012, Design Computing and Cognition ’12,
Springer, Dordrecht, p. 361–376
Li, AI 1999 'Expressing parametric dependence in shape
grammars, with an example from traditional Chi-
nese architecture', Proceedings of CAADRIA 1999,
Shanghai, p. 265–274
Li, AI 2001, A shape grammar for teaching the architectural
style of the Yingzao fashi, Ph.D. Thesis, MIT
Li, AI 2004, 'Styles, grammars, authors, and users', in Gero,
JS (eds) 2004, Design Computing and Cognition ‘04,
Kluwer, Dordrecht, p. 197–215
Stiny, G 1975, Pictorial and Formal Aspects of Shape and
Shape Grammars, Springer, Basel
Stiny, G 1980, 'Introduction to shape and shape gram-
mars', Environment and Planning B: Planning and De-
sign, 7(3), pp. 343-351
Stiny, G 1981, 'A note on the description of designs', En-
vironment and Planning B: Planning and Design, 8(3),
p. 257–267
Stiny, G 1990, 'What is a design?', Environment and Plan-
ning B: Planning and Design, 17(1), p. 97–103
Stiny, G 1992, 'Weights', Environment and Planning B:
Planning and Design, 19(4), p. 413–430
Stiny, G and Gips, J 1972, 'Shape grammars and the gen-
erative specification of painting and sculpture', in
Freiman, CV (eds) 1972, Information Processing 71,
North-Holland, Amsterdam, p. 1460–1465
Stouffs, R 2008, 'Constructing design representations us-
ing a sortal approach', Advanced Engineering Infor-

338 | eCAADe 34 - SHAPE, FORM AND GEOMETRY | Grammars and Concepts - Volume 2
Style and Type in a Generic Shape Grammar
The Case of Multipurpose Chairs

Garcia Sara1 , Romão Luís2

1,2
Faculty of Architecture of the University of Lisbon
1,2
{sgarcia|lromao}@fa.ulisboa.pt

Generic grammars are a means to describe and generate designs in different

domains with different abstraction levels. Simple mechanisms can generate a
complex variety of solutions within different types, among a coherence of similar
styles. This paper describes the process of defining a generic grammar for wood
frame multipurpose chair design from two specific grammars that define
individual styles of two particular designers. This research is part of a larger one
that comprehends the development of a multipurpose chair design tool. The
combination of the two specific grammars generates designs from two specific
styles, a commons style, a hybrid style and new styles. Results also comprise the
characterization of the styles by the analysis of rules and parameters.

Keywords: Generic shape grammars, Style, Type, Multipurpose chair design

INTRODUCTION implementation and (4) evaluation of an interactive

Generic shape grammars characterize one generic
class in terms of their inherent diversity of types and
meanwhile characterize specific styles. This paper
presents a generic shape grammar for the multipur-
pose chair class. The grammar is customized into two
specific styles of two particular designers, by the re-
striction of rules and parametric ranges of the more
generic grammar. The purpose of this grammar is to
provide a useful tool for the designers at the early
conceptual stage of the design process. In this stage
the exploration and combinations of existing chairs,
and thus the generation of hybrid styles, is likely to
occur. "The reuse of design styles is regarded as a
significant strategy of design innovation in product
design" (Ahmad and Chase 2012).
This grammar is part of a larger research that
comprises the phases of (1) analysis, (2) design, (3)
design tool. The second phase was subdivided into
four stages, which correspond to the evolutionary
development of a system which comprised a shape
grammar and an ontology. This paper refers to the
second stage, in a bottom-up approach from a spe-
cific grammar developed in the first stage to a generic
grammar. The third stage of the design comprised a
more generic grammar, its implementation and eval-
uation (Garcia and Romão 2015). The fourth stage
detailed the incorporation in the generic tool of a
specific grammar (Garcia and Barros 2015).
Shape grammars are ruled-based formalisms
that describe and prescribe designs in one language.
A language is a collection of designs that share some
recognizable compositional principles and formal co-
herence. Those designs belong to a class which may
represent a type, a style, a product class, and so forth

SHAPE, FORM AND GEOMETRY | Grammars and Concepts - Volume 2 - eCAADe 34 | 339
 (Stiny 1980). Shape grammars can analyze existing
languages (analytic grammars, e.g. Stiny and Mitchell
1978), generate new coherent ones (synthetic gram-
mars, e.g. Stiny 1980), or transform existing de-
signs (transformation grammars, e.g. Eloy and Duarte
2012). On another hand, shape grammars can con-
template one language (specific grammars) or more
than one language of designs (generic grammars).
Style in analytic grammars is seen as an explanatory
tool, used as post hoc classification, while in synthetic
grammars style is seen as a generative tool, driver for
the exploration of new designs.
Shape grammars largely explore the concepts of
(a) style and (b) type. Style and type both charac-
terize a group of objects that share similar features,
but the first one is context-dependent and the sec-
ond one is context-independent. (a) According to
Chan (2000), a style can be related to one person
(individual style - e.g.: Hepplewhite style in Knight
1980), group of persons (group style - e.g. Thonet
style in Barros et al. 2011), place (regional style) or pe-
riod of time (period style - e.g.: Queen Anne style in
Flemming 1987). One may also include an art move-
ment as a combination of all features (e.g. De Stijl
in Knight 1989). (b) According to Ahmad and Chase
(2012), types may characterize functions (e.g.: office
chairs) or morphologies (e.g.: cantilever chair). Flem-
ming (1987) noticed that style disregards differences
of type (e.g. Thonet style is found in chairs and ta-
bles) and type disregards differences of style (e.g. a
four-legged chair can be a Thonet or Hepplewhite).
Moreover, the same style may be found across differ-
ent design domains (e.g. De Stijl is an artistic style, ar-
chitectural style, and design style). Styles and types
can be subdivided (e.g. the De Stijl style contains dif-
ferent individual styles, each one with different stylis-
tic periods, which were compared in Knight 1989).
Among style, other specific languages are character-
ized by shape grammars. It is the case of the brand
identity (e.g. Pugliese and Cagan 2002) and prod-
ucts families or collections (Castro e Costa and Duarte
2014).
340 | eCAADe 34 - SHAPE, FORM AND GEOMETRY | Grammars and Concepts - Volume 2
SPECIFIC SHAPE GRAMMARS
Shape grammars can analyse and describe (1) one
singular language in specific grammars or (2) a mul-
tiplicity of languages in generic grammars.
Firstly, a large number of shape grammars are
from the analytic (specific) type, intending to cap-
ture one historical style and generate new designs
within the same style. A style is identified by com-
mon features that appear repeatedly in a set of ob-
jects. Those features may refer to the structure,
shape, size, colour, texture, ornaments, materials, etc.
We considered three key issues about the style defi-
nition in specific shape grammars: the style identifi-
cation, inference and evaluation. (a) For the identifi-
cation of the features that define a style there are two
main sources: written descriptions or rules, given by
art or architecture historians in manuals or treatises
(e.g. Yingzao Fashi in Li 2001) or guidelines given
by a brand identity (e.g. Harley-Davidson brand in
Pugliese and Cagan 2002), and a corpus of designs
that are instances of the style. Notice that the first
source may not exist, if style is not explicitly prede-
fined. (b) The inference of the style is made by encod-
ing the written features and the patterns in designs
(given by stylistic analysis) into formal rules and pa-
rameters, into what is considered to be the core of the
style. (c) The style evaluation is made by three tests
defined by Stiny and Mitchell (1978): the grammar
has to (T1) recreate the designs of the corpus, (T2)
recreate other existing designs of the same style and
(T3) generate new designs within the style. The new
generated designs must be plausible in appearance,
function and symbolism (Stiny and Mitchell 1980).
The application of the three tests is exemplified in
Orsborn et al. (2006). The formal definition of the
style given by specific rules and parametric ranges
may or not confirm the preliminary understanding.
Several issues arise with this definition of style
by shape grammars. Ahmad and Chase (2012) argue
that a language may contain designs that not belong
to a style, may not contain all designs from a style,
and omit style aesthetic qualities. The style definition
is frequently incomplete, either the definition of his-
 torians (thus some features have to be intuitively de- of 'genericness' would be one single design, which
fined by the developer) either the definition encoded is part of one specific grammar, which is part of a
in the grammar, as the corpus rarely contains all the generic grammar, which is part of a more generic
individuals of the style, and the representation of the grammar. The greatest level would be a fully generic
designs in the grammar includes only a small part of grammar, which could generate all the designs of a
the features (e.g. in Stiny and Mitchell 1978 only the domain. Notice that if the designs added to a gram-
plan is represented, disregarding proportions, col- mar are very close to the ones previously generated
ors, textures, materials, etc.). Moreover, style is al- by the grammar, the degree of 'genericness' will not
ways an interpretation, whether made by historians be significantly increased. The choice of the degree
or by grammar developers. On another hand, one of 'genericness' is crucial: "if the rules are too specific
can assure that the completed designs generated by to a problem, then the system is limited to certain
a grammar belong to its language, but we cannot en- range of designs. If the rules are too general, then the
sure that they belong to the style. In most cases, the system does not have the specific kind of knowledge
style evaluation is made only by the grammar devel- for the users to benefit from." (Lee and Tang 2004,
oper; in few examples it is also made by the author(s) p.12). As such, the flexibility of generic grammars in
of the corpus (Duarte 2001), or by the costumers of terms of extension and restriction is a relevant fea-
the product (Pugliese and Cagan 2002). ture. In Figure 1, the specific grammar results from
Secondly, a shape grammar can encompass a restriction of the generic grammar (by deleting
the definition of more than one specific language one rule and by adding labels), and thus generates a
through a more abstract generic grammar (Li 2001). smaller solution space and one single style (it consid-
Generic grammars capture one generic class, and can ers square designs instead of the square/round de-
be customized into specific grammars that define a signs of the generic).
more specific subclass, through restrictions in rules,
Figure 1 parametric ranges and/or labels. Therefore, generic
Example of a grammars can characterize context-independent de-
generic grammar sign types within a product class, while specific
and a specific grammars can characterize context-dependent de-
grammar. sign styles. The concept was later formalized by
Duarte and Beirão (2011) as an ontology-based gram-
mar that characterizes abstract designs and design
patterns within a design domain. This later definition
is a more generic one and thus characterizes a larger
domain. The underlying goal of generic grammars
is to provide a generative design tool, which can be
applied either to generic or specific contexts, consid-
ering the idiosyncrasies of a particular design prob-
lem or a particular designer. Both generic and spe-
cific grammars can be (a) extended to become more
general, by adding and redesigning rules and param-
eters, or (b) restricted to become more specific, by re-
stricting rules and parametric ranges. Thus, generic
grammars can have different degrees of 'genericness'
(Castro e Costa and Duarte 2014). The smallest level

SHAPE, FORM AND GEOMETRY | Grammars and Concepts - Volume 2 - eCAADe 34 | 341
 Generic grammars can be designed through several Figure 2
steps in different ways (Figure 2): (1) Start the pro- Specific (light grey)
cess either by (1a) select a corpus of existing designs and generic (dark
from one specific language, and then jump to step grey) grammar
2 or by (1b) select an existing corpus from a more processes.
generic language, and then jump to step 4. (2) De-
sign a specific grammar, which is able to generate:
(1a) the corpus; (2a) new designs from the same lan-
guage of the corpus, and (2b) other existing designs
from the same language of the corpus. If the spe-
cific grammar cannot generate (2b), jump to step 3.
Otherwise, jump to step 4. (3) Design another spe-
cific grammar or extend the previous one by the pro-
cess of transformation. This new specific grammar
will generate (2b) and other new designs. One may
now jump to step 4. (4) A generic grammar may be
inferred from two specific grammars that generate
two different specific languages (Benrós et al. 2012,
Castro e Costa and Duarte 2014) or by designs of a
more generic language (Garcia and Romão 2015). A
generic grammar is able to be customized into the
previous existing specific grammar (2), a new specific
grammar which generates a new specific language
(4a), and generate new designs of new specific lan-
guages (4b).
In product design there are several examples of OVERLAPPING SPECIFIC GRAMMARS
generic grammars that characterize product classes. Two specific grammars may be combined in order
The coffeemaker grammar (Agarwal and Cagan 1998) to generate a more generic grammar. Figure 3 illus-
focus on the generation of four brands (although re- trates two overlapping sets: set A and set B. Their
strictions are not specified). The motorcycle gram- combination creates four extra sets: set A-B (distinc-
mar (Pugliese and Cagan 2002) generates products tive set A) collects the distinctive features in A but not
from one brand, specifying rule and parameter range in B, and results from the difference of A and B; set B-A
restriction. The crossover vehicles grammar (Orsborn (distinctive set B) collects the distinctive features in B
et al. 2006) generates designs in three existing vehi- but not in A; and results from the difference of B and
cle subclasses, one existing hybrid class (which com- A; set A∩B (common set) collects the common fea-
bines features of two or more classes) and designs tures in A and B, and results from the intersection of
in new vehicle subclasses, defining specific rules and A and B; and set A∪B (hybrid set) collects all the fea-
specific parametric ranges (represented in areas in tures in A and B, and results from the union of A and B.
range charts). The tableware grammar (Castro e We can also consider the universal set U (generic set)
Costa and Duarte 2014) is the first so-called generic that contains more than the A and B sets, and there-
grammar in product design, and is able to create six fore the set U-(A∪B) (complement set) that collects
different collections. all the features that are neither in set A not in set B.
These sets are represented in Figure 3, where the sets
showed as areas do not contain subsets and the sets

342 | eCAADe 34 - SHAPE, FORM AND GEOMETRY | Grammars and Concepts - Volume 2
 showed as perimeters contain all the subsets that are
inside it.
Figure 3
Venn diagram of
two overlapping
sets.
Consider these two sets as being two specific gram-
mars, grammar A and grammar B. If grammar A and
grammar B have some overlapping features, their
combination will derivate into more three grammars:
the Common Grammar - which characterizes an am-
biguous language that can either be A or B, the Hy-
brid Grammar - which characterizes a language that
mixtures features of A and B, and the Generic Gram-
mar - which characterizes a more generic grammar
that contemplates more than both A and B. Notice
that we not considered the Distinctive Grammars, as
they cannot generate complete designs.
If one considers the rules as the elements of
those sets, we came out with four types of rules. (R1)
Specific rules of A are used to generate designs in the
language A but not in the language B. (R2) Specific
rules of B are used to generate designs in the lan-
guage B but not in the language A. (R3) Common or
unrestricted rules are applicable to all the specific lan-
guages, and thus can be used to generate designs in
both A and B languages. (R4) New rules are not used
to recreate any specific language.
Table 1
Restrictions in rules
and parameters of a
generic grammar.
SHAPE, FORM AND GEOMETRY | Grammars and Concepts - Volume 2 - eCAADe 34 | 343
Table 1 summarizes the application of these rules
and parametric ranges in the different types of gram-
mars. (1) Grammar A uses R1 and/or R3 rules with
parametric ranges restricted to A [a1, a2]. (2) Gram-
mar B uses R2 and/or R3 rules, with parametric ranges
restricted to B [b1, b2]. (3) The Common grammar
uses R3 rules and parametric ranges common to ei-
ther A and B, resulting from the intersection of [a1,a2]
and [b1,b2] (notice that if [a1,a2] and [b1,b2] are dis-
joint sets, the intersection will be an empty set). (4)
The Hybrid grammar uses rules R1, R2 and R3 with
the union of the parametric ranges of A and B. At this
point, a grammar that combines n specific grammars
can generate n+2 grammars. (5) The Generic gram-
mar is an extension of the Hybrid grammar to new
rules, new parameters, and/or the relaxation of para-
metric ranges (for e.g. into an infinite interval).
The application in this process will be detailed in
the next section, using two specific grammars as an
example. The clarification of the differentiation be-
tween specific languages can help in the develop-
ment of new designs. This reasoning for the gener-
alization of two specific grammars can be extrapo-
lated to the generalization of n specific grammars to
a generic grammar (introducing more sets) and also
of n designs to one specific grammar (when each set
represents a design).
MULTIPURPOSE CHAIR GRAMMAR
A parametric generic grammar for multipurpose
chair design was developed in four stages, listed in
Table 2. The stage 1.1 comprised the development
of an analytic grammar for multipurpose chairs of
one designer (Daciano da Costa). The stage 1.2 em-
braced the extension of the previous grammar in or-
der to incorporate the generation of multipurpose

chairs of a second designer (Jasper Morrison), orig-
inating a more generic grammar that characterizes
multipurpose chairs with wooden frame. The stage
2.1 involved the development of a generic gram-
mar from the analysis of twenty-six chairs of different
types. This stage emphasized types to the detriment
of styles, and thus is able to generate a significantly
larger number of solutions. The stage 2.2 consid-
ered the incorporation of an existing specific gram-
mar (Barros et al. 2011) to the grammar developed
in the last stage. The first two stages characterize
two individual styles, the third different types of de-
signs, and the fourth a product family (the variation
among the designs occurs only in the inner frame of
the backrest). An underlying process was common to
all the stages: the corpus selection, the development
of an ontology of parts, and the development of the
grammar. The last two stages also comprised an im-
plementation of the grammar and its evaluation.
This paper discusses the generic grammar of the
stage 1.2. This grammar includes two specific gram-
mars, each characterizing an individual style. Individ-
ual style arises from choices of the designer among
alternatives in design, which can be motivated by
individual knowledge, principles, preferences, and
procedures on handling domain-specific tasks (Chan
2000). The grammar was developed through the
process of transformation (Knight 1989) of the previ-
ous analytic grammar, by rule addition, rule change,
parameter addition and parameter change (relax-
ation of parametric ranges). As mentioned in Cas-
tro e Costa and Duarte (2014), the transformation
also involves the reformulation of the original rules,
where rules were added (by decomposition of rules
344 | eCAADe 34 - SHAPE, FORM AND GEOMETRY | Grammars and Concepts - Volume 2
Table 2
Evolutionary stages
of the multipurpose
chair grammar.
into simpler rules) and deleted (by generalization of
equivalent rules). For instance, notice that the Da-
ciano Grammar on stage 1.2 has a bigger number of
rules than in the stage 1.1. The grammar in stage 1.1
comprises a corpus of one designer and includes 23
rules, while in the stage 1.2 comprises a corpus of
two designers and contains 44 rules. The develop-
ment of the ontology followed the development of
the grammar, where parts were added and the se-
mantics readjusted.
Figure 4
Domain
decomposition.
Corpus
The criteria for the corpus selection followed a hierar-
chy of sets (Figure 4), where each set contains all the
sets below it. Each level corresponds to a category,
which may contain more than one set. The product
design domain is the more generic level, and was de-
composed into more specific levels until the bottom
level of individual chairs. Some levels could be or-
dered differently, other levels could be added, and
 some sets can be inferred from existing ones. Fig-
ure 4 distinguishes between categories of type (light
grey) and categories of style (dark grey). Notice that
types refer to nouns with an initial lower letter (com-
mon noun) and styles refer to nouns in an initial cap-
ital letter (proper noun), therefore one may consider
types as being more generic than styles.
The corpus of the grammar is composed by ten
chairs, five from the Portuguese designer Daciano da
Costa (1930-2005) and five from the English designer
Jasper Morrison (1959-). The chairs of Daciano (Mar-
tins 2001) date from 1966-1977 and were designed
for the interiors of four Portuguese hotels, being all
produced by a Portuguese company (Móveis Sousa
Braga). The chairs of Morrison (Morrison 2015) date
from 1988-2011 and were produced by three com-
panies - Cappelini (Italy), Murani (Japan) and Vitra
(Switzerland). The chairs are all from the multipur-
pose type and have wooden frame. The chairs have
different typologies, such as the number of legs, the
presence or lack of arms, the number of stretchers,
the seat shape, and so forth. Table 3 lists the chairs
of the corpus, chronologically ordered, plus two ex-
tra chairs from the control group. For the purpose
of the grammar, the corpus was simplified into lines
(that represent the frame) and planes (that represent
the seat and back pieces). The solid shape gener-
ated from the structure was introduced latter. Figure
6 shows the juxtaposition of the chairs in the corpus
and one chair of each style.
Table 3
Restrictions in rules
and parametric
ranges.
SHAPE, FORM AND GEOMETRY | Grammars and Concepts - Volume 2 - eCAADe 34 | 345
Rules
The initial shape of the grammar represents the seat
and back planes and the bounding box underneath
the seat where the legs are placed (see the left-hand
side of the rules in Figure 5). The initial shape is pa-
rameterized accordingly to the anthropometric stan-
dards for chairs. The rules are successively applied
to the initial shape, by adding parameterized parts
of the chair, until a final solution is reached. Those
parts are arranged in six groups, which correspond
to different functional regions of the chair: Legs, Seat,
Back, Stretchers, Base and Arms. Rules only draw the
left side of the designs, as they are bilaterally sym-
metrical. Figure 5 shows the two rules of the Legs,
respectively the Front Leg and the Back Leg. Each
leg has its endpoints parameterized; four parame-
ters correspond to the width and depth position in
the seat plane and the splay and rake angles. Table
4 shows the application of those rules and parame-
ters to the grammars (Daciano, Morrison, Common,
Hybrid and Generic), following the process described
before for the overlapping of two specific grammars.
The generic parametric ranges were defined in the
stage 2.1. Legs include another rule that adds cur-
vature (used in three Morrison chairs), but this was
deleted in the stage 2.1, as only curvatures in corners
were considered. In this case, the more generic is the
grammar, the less detail their rules tend to have.
 son Style, Common Style, Hybrid Style, and Generic Table 4
'Style'). The designs generated by the first two spe- Sample (corpus and
cific grammars were conceived accordingly to the control group).
three evaluation tests for the validation of an ana-
lytic grammar. The evaluation of the new styles com-
prised three additional tests.
Test 1: the grammar generates the ten chairs of
the corpus. Figure 6 shows the juxtaposition of the
chairs in the corpus, plus a chair from each style (D4
and M2).
Test 2: the grammar generates other existing de-
signs of the same style. Figure 6 shows the genera-
tion of the two chairs of the control group (D6 and
M6). These chairs could not be fully generated by
the specific grammars. The chair D6 uses a stretcher
which any of the chairs of Daciano's corpus use. The
chair M6 has three features outside the style: the back
leg uses a negative angle (inwards the chair); uses a
base rail; and the seat and back panels are composed Figure 5
by slats. Rules of legs.

Figure 6
Designs generated
by the grammar.

Designs
Figure 6 shows several designs generated by the
grammar, grouped within the five different gram-
mars discussed in this paper (Daciano style, Morri-

346 | eCAADe 34 - SHAPE, FORM AND GEOMETRY | Grammars and Concepts - Volume 2
Test 3: the grammar generates new designs that be-
long to the specific language. Figure 6 depicts two
new designs for each style (D7 and D8, M7 and M8).
Test 4: the grammar generates new designs from
the Ambiguous style (A1 and A2). This is a very re-
stricted style, with the smallest solution space from
the ones analyzed in this paper.
Test 5: the grammar generates new designs from
the Hybrid style (H1 and H2). The chair H1 is the clos-
est to the chair Thonet No. 14 (Barros et al. 2011) that
this grammar can generate.
Test 6: the grammar generates new designs in
the Generic style. As show in Figure 6, this includes
new types of chairs as monolithic base chairs (G1) and
cantilever chairs (G2). G3 represents the chair Thonet
No. 14 (Barros et al. 2011) in its generic form.
A qualitative characterization of the two de-
signer styles was made by comparison of the spe-
cific rules and the specific parameter ranges, for each
group of parts. (1) Daciano style (D) uses four and
three-legged chairs, while Morrison style (M) always
uses four legs; none of them uses one or two-legged
chair. D legs are vertical while M legs have slight an-
gles and curvatures. (2) The seat front radius in D is
0 while in M varies between [0,60]; and the seat rear
radius in D is {0,100} while in M varies between [0,80].
Both D and M uses all types of seat rails except the X-
shaped ones. (3) Both vary between an open back
and a solid back, using back outer frame elements
but not inner frame elements. (4) Only one chair in
M style uses stretchers. (5) Only one chair in D style
uses base rails (a side rail). (5) In D style, arms are sup-
ported by the front or/and back leg extension; while
in the only example of an M armchair, arms are sup-
ported by the back leg extension. As in seat rails, the
radius of the armrest arc is in D half of its length, while
in M this proportion varies. One can conclude that
the Daciano style uses more straight rigid lines and
proportions than the Morrison style.
DISCUSSION
The same style may describe different types, and the
same type may describe different styles. The first
SHAPE, FORM AND GEOMETRY | Grammars and Concepts - Volume 2 - eCAADe 34 | 347
statement is illustrated in Figure 6 by the chair D4 and
the chair D8, which belong to the same style (Daciano
style) but represent two different types of chairs (side
chair and armchair). Thus, one can identify products
of the same family from existing designs, for instance,
the chairs D2 and D5 from the corpus. On another
hand, the second statement is illustrated in Figure 6
by the chairs D4 and M2, which belong to different
authors but have similar configurations, thus may in-
dicate related influences.
Each individual style is defined by the restric-
tion of rules and parametric ranges. The compari-
son of two individual styles was made by comparison
of rules and parametric ranges, defining specific fea-
tures used by each one, common features, and hybrid
features. However, these characteristics do not con-
clusively describe one designer's style, only by the
analysis of few products and few design aspects. The
test 2 failed (the grammar could not generate other
designs of the same style), which means that either
(a) five designs are not enough to define a designer's
style or (b) the designer does not have a consistent
style. Moreover, the identification of style in chairs
with a high degree of abstraction may be over sim-
plifying. Future work should include the evaluation
of the style in more detailed chairs by other persons
beyond the grammar developer, for instance by the
authors of the corpus that are still alive (in the present
case only Jasper Morrison) and by their collaborators
in the design studio, so that one can conclude if the
new generated designs are actually in the style and
what is their degree of style (Chan 2000).
Comparisons can also be made between two
methods for designing generic grammars that cor-
respond to the stage 1.2 and the stage 2.1 (Table 2).
The first method is the one presented in this paper
and comprises an extension of a specific grammar to
a generic one. The second method is the one pre-
sented in Garcia and Romão (2015) and comprises
design of a generic grammar from the analysis of a
corpus of designs with different types. Although the
second method covers a greater variety of types of
designs (as one can observe by the chairs H1 and G3
 in Figure 6), and thus has a wider solution space, is
does not ensure that it can predict all the possible
types. In fact, there are two rules in the stage 1.2 that
are not applicable in the stage 2.1. The combination
of existing specific grammars of the same class would
be an interesting field to explore in the future.
REFERENCES
Agarwal, M and Cagan, J 1998, 'A blend of different
tastes: the language of coffeemakers', Environment
and Planning B: Planning and Design, 25, pp. 205-226
Ahmad, S and Chase, S 2012, 'Style representation in de-
sign grammars', Environment and Planning B: Plan-
ning and Design, 39, pp. 486-500
Barros, M, Duarte, J and Chaparro, B 2011 'Thonet Chair
Design Grammar: A Step towards the Mass Cus-
tomization of Furniture', CAAD Futures 2011: Design-
ing Together, pp. 181-200
Benrós, D, Hanna, S and Duarte, JP 2012 'A generic shape
grammar for the Palladian Villa, Malagueira House
and Prairie House', Design Computing and Cognition,
Texas, USA, pp. 321-340
Chan, CS 2000, 'Can style be measured?', Design Studies,
21, pp. 277-291
Castro e Costa, E and Duarte, JP 2014 'Generic Shape
Grammars for Mass Customization of Ceramic Table-
ware', Design Computation and Cognition DCC, pp.
437-454
Duarte, JP 2001, Customizing Mass Housing: A Discursive
Grammar for Siza's Malagueira Houses, Ph.D. Thesis,
Massachusetts Institute of Technology
Duarte, JP and Beirão, JN 2011, 'Towards a Method-
ology for Flexible Urban Design: Designing with
Urban Patterns and Shape Grammars', Environment
and Planning B: Planning and Design, 38(5), pp. 879-
902
Eloy, S and Duarte, JP 2012 'A transformation grammar-
based methodology for housing rehabilitation', De-
sign Computing and Cognition '12, pp. 301-320
Flemming, U 1987, 'More than the sum of parts: the
grammar of Queen Anne houses', Environment and
Planning B: Planning and Design, 14, pp. 323-350
Garcia, S and Barros, M 2015 'A Grammar-Based Sys-
tem for Chair Design: From Generic to Specific
Shape Grammars', Real Time - Proceedings of the 33rd
eCAADe Conference, Vienna University of Technol-
ogy, pp. 427-436
Garcia, S and Romão, L 2015, 'A Design Tool for Generic
Multipurpose Chair Design', in Celani, G, Sperling, D
348 | eCAADe 34 - SHAPE, FORM AND GEOMETRY | Grammars and Concepts - Volume 2
and Franco, J (eds) 2015, The next city - New technolo-
gies and the future of the built environment [16th In-
ternational Conference CAAD Futures 2015. Sao Paulo,
July 8-10, 2015], Springer, pp. 600-619
Knight, T 1980, 'The generation of Hepplewhite-style
chair-back designs', Environment and Planning B:
Planning and Design, 7, pp. 227-238
Knight, T 1989, 'Transformations of De Stijl art: the paint-
ings of Georges Vantongerloo and Fritz Glarner', En-
vironment and Planning B: Planning and Design, 16,
pp. 51-98
Lee, HC and Tang, MX 2004 'Evolutionary Shape Gram-
mars for Product Design', Seventh International Con-
ference on Generative Art, Politecnico di Milano Uni-
versity, pp. 309-322
Li, A 2001, A shape grammar for teaching the architectural
style of the Yingzao fashi, Ph.D. Thesis, Massachusetts
Institute of Technology
Martins, JP (eds) 2001, Daciano da Costa: designer, Fun-
dação Calouste Gulbenkian, Lisboa
Morrison, J 2015, A book of things, Lars Müller, Zürich
Orsborn, S, Cagan, J, Pawlicki, R and Smith, RC 2006, 'Cre-
ating cross-over vehicles: Defining and combining
vehicle classes using shape grammars', Artificial In-
telligence for Engineering Design, Analysis and Manu-
facturing, 20, pp. 217-246
Pugliese, M and Cagan, J 2002, 'Capturing a rebel: mod-
eling the Harley-Davidson brand through a motor-
cycle shape grammar', Research in Engineering De-
sign, 13, pp. 139-156
Stiny, G 1980, 'Kindergarten grammars: designing with
Frobel's building gifts', Environment and Planning B:
Planning and Design, 7, pp. 409-462
Stiny, G and Mitchell, W 1978, 'The Palladian gramar', En-
vironment and Planning B: Planning and Design, 5,
pp. 5-18
Stiny, G and Mitchell, W 1980, 'The grammar of paradise:
on the generation of Mughul gardens', Environment
and Planning B, 7, pp. 209-226
Wood Mass-Customized Housing
A dual computer implementation design strategy

Krystian Kwiecinski1 , Filipe Santos2 , Ana de Almeida3 ,

Bruno Taborda4 , Sara Eloy5
1
Warsaw University of Technology (WAPW) 2,5 Instituto Universitário de Lisboa
(ISCTE-IUL, ISTAR-IUL) 3 Instituto Universitário de Lisboa (ISCTE-IUL, CISUC)
4
Instituto Universitário de Lisboa
1
krystian.kwiecinski@arch.pw.edu.pl
2,3,4,5
{filipe.santos|ana.almeida|bruno_taborda|sara.eloy}@iscte.pt

This paper reports our current research on automatic generation of houses

layouts according to future inhabitant's requirements. For that generation we
propose the use of a design method based on shape grammars that encodes light
wood frame construction guidelines. Two different implementations for the design
system are presently under development. One based on shape grammars
supplemented with procedural knowledge and another using a genetic algorithm.
Both implementations allow the generation of house layouts that fulfill both the
user requirements and the design language.

Keywords: shape grammar, genetic algorithm, computer implementation

INTRODUCTION To make the system available for future inhabitants

This paper will focus on an undergoing research
that aims at developing a system that allows hous-
ing future inhabitants to participate in the design
of their house by using mass-customization design
(Kwieciński and Slyk 2014). The proposed system is
based on shape grammars and will allow future own-
ers to acquire houses that fit their needs while simul-
taneously complying to a language of design (Stiny
1980). Such a system will enable to deliver design
solutions with quality (Eloy et al. s.d.) and at afford-
able prices improving satisfaction of clients. The sys-
tem encodes knowledge on housing design princi-
ples (types and sizes of rooms and preferable connec-
tions) as well as in the timber construction technique
that underlies the shape composition principles.
a computer implementation is being developed so
that it could be made available online or at the hous-
ing construction company. The design tool will: i) de-
liver design solutions answering the client's' require-
ments based on a design grammar therefore main-
taining a language of design, ii) be feasible to be used
by non designers (e.g. clients, sales staff of a con-
struction company). With the aim of finding a good
technical solution to satisfy users' needs two strate-
gies of implementation are being developed simul-
taneously and in close collaboration. One is based in
shape grammars supplemented with processes and
the other based in genetic algorithms.
Research in mass-customized design using
shape grammars have been developed over the years

SHAPE, FORM AND GEOMETRY | Grammars and Concepts - Volume 2 - eCAADe 34 | 349
 and several shape grammars have addressed the
housing problem and the need to make available de-
sign systems that respond to the inhabitants needs
(Duarte 2001; Colakoglu 2005; Benros et al. 2011;
Eloy 2012; Coimbra and Romão 2013). The main dif-
ficulty to fully use these systems in design practice
is the delay of the computer implementation of ar-
chitectural design processes, namely by the use of
shape grammar logics. Computerized design tools
that offer design alternatives supporting the design
ways of thinking and working are still rare.
Current systems allow some aspects of genera-
tion of shape and are based on a built-in drawing ed-
itor, e.g. Rabo-de-Bacalhau grammar tool (Strobbe
2015) or in links to a CAD software, e.g. Grappa (Grasl
2012) for Revit (McKay et al. 2012). These systems
are essentially academic based and not yet aiming at
being used in the real design scenario. Design ap-
plications like Flemming (1987) Queen Anne Houses
tool, Li (2002) Yingzao fashi grammar, Grasl (2012)
Grappa tool and Ligler and Economou (2015) Ent-
elechy Grammar are systems that allow to develop
designs according to languages of past styles. Those
tools are used specially in an historical and design un-
derstanding perspective and not as tools to design
new solutions for architecture problems. Other im-
plementations like Strobbe (2014) Rabo-de-Bacalhau
grammar tool allow to generate new design solutions
that can then be used by the house owners to do their
refurbishment. The goal of the presented research is
to deliver such a tool that can deliver designs able to
be immediately used as a design project to build a
house.
Catalogues of house designs are very popular in
several countries. Future owners are provided a cat-
alogue of standardized solutions for them to choose
but those solutions are restricted to a small number
of changes and the act of choosing is based on look-
ing to all solutions and not to the ones that may re-
spond to the clients need. In this research we aim
at looking for systems that allow users to play with
a house design tool allowing them to find the best fit
house to their own personal wishes.
350 | eCAADe 34 - SHAPE, FORM AND GEOMETRY | Grammars and Concepts - Volume 2
WOOD MASS-CUSTOMIZED HOUSING
Choice of the house construction system has been
determined by its popularity in some european coun-
tries as well as the possibility of its prefabrication off
site and partial automation of manufacturing pro-
cess. In order to facilitate mass-customization cho-
sen construction system have to enable construction
flexibility and permit just-in-time production. The
use of industrialized methods of production allows to
save time and permit cheaper costs and quality con-
trol. For these reasons a light wood frame construc-
tion was chosen as the main structure of the building.
In consequence developed shape grammar reflects
the capabilities and constraints of the chosen build-
ing system.
Chosen building technique uses vertical and hor-
izontal structural members, studs, which provide a
stable frame to which interior and exterior wall cov-
erings are attached. The structures are strengthened
with rigid panels of oriented strand board (OSB) used
to form wall sections. Sections of the walls can be
prefabricated off-site therefore the whole building
could be erected in just a few weeks.
Wall framing in light frame construction system
includes the vertical and horizontal members of ex-
terior walls and interior partitions (see Figure 1). The
walls are build with the use of standardized studs
which are placed in the axial spacing of 60 cm. This
structural guideline is encoded in the shape gram-
mar rules. Therefore the shape rules are operating
on rooms which dimensions are a multiplication of
60x60 cm module. Moreover the maximum permis-
sible span of the ceiling, up to 6 m, affects the choice
of the house typology, therefore also the localization
of load bearing walls as well as the width of the build-
ing in general and rooms in particular.
The possible design solutions include detached
houses that can have either one or two floors, a com-
pact rectangular floor plan, where only the garage
sticks out of the rectangular perimeter. The houses
have a fixed width, a center axis with a right and left
side of the same width, and the house length may
vary. The position of the houses may be parallel or
 perpendicular to the access route. For the purpose of SHAPE-GRAMMAR
the computer implementations presented in this pa- The proposed grammar consists of a vocabulary of
per the scope of possible design solutions was nar- shapes and shape rules and a generation process re-
rowed down to single floor houses located perpen- sponsible for the generation of the layout solutions.
dicular to the access route. Shape rules have a shape part and a condition part
which enable the application of rules preventing the
Figure 1
application if some conditions are not fulfill. Rules
Rooms modularity
are grouped in order to differentiate 4 design stages
resulting from light
(see Figure 2): generation of the the grid, generation
wood frame
of rooms belonging to the entrance zone, genera-
structural guidlines.
tion of rooms belonging to the semi-public zone and
generation of rooms belonging to the private zone.
Each stage ends after all the required rooms of that
stage are generated leading the generation to the
next stage. The whole process is finalised with the
generation of a layout where all the required rooms
are present and all the conditions are fulfilled. Future
owners may then see the possible design solutions
for their requirements and decide which they want
to opt for.
The generation process initiates by collecting in-
formation from the future inhabitant in order to for-
mulate the house design brief. At this stage of the
development the users are asked to specify the num-
ber of future inhabitants and the quality level of
the house (minimum, recommended or optimum).
Based on that information the size of the house is
calculated and the functional program of the build-
ing is specified which can be further modified by
the user. Besides the layout design that is focused
in this paper, the future inhabitant will be able to
modify all other aspects of architecture as where to
build, what construction system and materials to use,
among others.
In the first stage of the house derivation a grid of
60x60cm cells is generated, which reflects the cho-
sen size of the house. In the next stages the rooms
belonging to different building zones are placed. The
placement of rooms occurs in relation to external fac-
tors (e.g. next to the front wall or next to the back
wall) and internal context (e.g a room is placed next
to another one). Conditions for grouping the rooms
constrain the layout configurations and guarantees

SHAPE, FORM AND GEOMETRY | Grammars and Concepts - Volume 2 - eCAADe 34 | 351
 Figure 2
Shape rules.

352 | eCAADe 34 - SHAPE, FORM AND GEOMETRY | Grammars and Concepts - Volume 2
 the correct placement of rooms according to housing
design standards. In order to preserve characteristics
as privacy, conditions for placing rooms are defined
which suggests e.g. placing the social areas closer
to the entrance and the private ones deeper in the
house.
The developed grammar uses the algebra U12
with bi-dimensional, linear base, floor plan of the
building. Automatically generated house layout so-
lutions based on client's inputs and a set of design
rules enable clients to participate in the design pro-
cess by acquiring only the designs that meet their
Figure 3 goals and requirements. Proposed grammar by pro-
Basic shapes. viding several design solutions, allows future inhabi-
tants to search for the ones that fits their needs. Gen-
erated solutions can be compared with each other or
might lead to the decision on changing the design
brief and generating new solutions fulfilling modified
user requirements.
IMPLEMENTATION
Two different implementations for the design sys-
tem are presently under development. One based
on shape grammars supplemented with processes
detaches shape knowledge from procedural knowl-
edge. The exponential nature of the search space led
Figure 4 to the need of employing different effective search
Positioning cell paradigms such as genetic algorithms which are used
within a new in a second implementation.
coordinate system.
Shape Grammar Implementation
The shape grammar implementation is based on
(Santos and Esmerado 2015) where differences with
respect to conventional shape grammars consist on
the adoption of any primitive vocabulary of shapes
and a complete detachment between shapes and
processes. The shape part is conceived to meet the
practical objectives of the design system and uses
a simplified form of sub-shape detection. Shapes
are represented by identifiers associated with pic-
tures and positioned in a coordinate system refer-
ring additionally to the applied Euclidian transforma-
tion. Shape compositions are represented by sets
of shapes positioned in the same coordinate system.
Processes use the usual procedural notions for cap-
turing sequences, alternatives and property tests for
the application of shape rules and conditional shape
rules during the design process.
To give an idea of the computational model used
in this shape grammar implementation, for produc-
ing houses layout design solutions, the following ba-
sic shapes (see Figure 3) are considered to represent
front and back markers and 0.6x0.6 m cells with dif-
ferent room occupation:
Each identifier represents the picture positioned in
the respective coordinate system xOy.
Positioning a shape defined in this way within a
different coordinate system x'O'y' is represented by a
pair (shape, transformation) specifying the transfor-
mation (translation, rotation, scale, etc.) required for
positioning the coordinate system xOy associated to
the defined shape into the new coordinate system
x'O'y'.
For instance, in Figure 4, cell is positioned into the co-
ordinate system x'O'y' by a scale S(c,d) and rotation
R(q) followed by a translation T(a,b) and thus is rep-
resented by (cell, T(a,b)R(q)S(c,d))).
Using these basic shape representations, the fol-
lowing shape compositions (Figure 5)

SHAPE, FORM AND GEOMETRY | Grammars and Concepts - Volume 2 - eCAADe 34 | 353

are represented respectively by:
basement3x3 = {(cell ,Id), (cell ,T
,→ (0.6 ,0)), (cell ,T(1.2 ,0)), (cell ,T
,→ (0 ,0.6)), (cell ,T(0.6 ,0.6)), (cell
,→ ,T(1.2 ,0.6)), (cell ,T(0 ,1.2)), (
,→ cell ,T(0.6 ,1.2)), (cell ,T(1.2 ,1.2)
,→ )}, vestibule3x3 = {(cves ,Id), (
,→ cves ,T(0.6 ,0)), (cves ,T(1.2 ,0)), (
,→ cves ,T(0 ,0.6)), (cves ,T(0.6 ,0.6)),
,→ (cves ,T(1.2 ,0.6)), (cves ,T(0 ,1.2)
,→ ), (cves ,T(0.6 ,1.2)), (cves ,T
,→ (1.2 ,1.2))}.
Conditional Shape Grammar Rules are pairs of shape
compositions plus a condition allowing it applica-
tion. The following shape grammar rule (see Fig-
ure 6) is represented by placeVestible with pre-
condition (next(matched(basement3x3),fron), base-
ment3x3, vestibule3x3).
Using identifiers for shape grammar rules allow its
reference in processes describing the design devel-
opment. Processes may also have names allowing it
reuse in other processes. The following processes de-
scribe the steps for producing houses layout design
solutions:
home = placeBasement; placeEntrance;
,→ placeSemipublic; placePrivate ;
,→ verify(homeRestrictions).
placeEntrance = placeVestibule;
,→ placeGarage;
,→ placeTechRoomandToilet .
placeSemipublic = placeHomeOffice and
,→ placeKitchen and placeDiningRoom
,→ and placeLiving Room.
placePrivate = placeDoubleBedroom and
,→ placeSingleBedrooms and
,→ placeBathroom.
placeSingleBedRooms =
,→ placeSingleBedRoom1 or
354 | eCAADe 34 - SHAPE, FORM AND GEOMETRY | Grammars and Concepts - Volume 2
,→ placeSingleBedRoom2 or ( Figure 5
,→ placeSingleBedRoom1 and An exemplary
,→ placeSingleBedRoom2 ). shape composition.
homeRestrictions = (next(vestibule ,
,→ garage) && next(toilet ,garage) &&
,→ next(technicalRoom ,garage) && (
,→ next(fron ,homeOfice) || next(fron ,
,→ homeOfice) || next(fron ,homeOfice)
,→ ) && (next(diningRoom ,fron) ||
,→ next(dinningRoom ,Kitchen)) && (
,→ next(homeOfice ,fron) || next(
,→ homeOfice ,livingRoom)) && (next(
,→ bathRoom ,doubleBedroom) || next(
,→ bathRoom ,singleBedroom)) && some(
,→ doubleBedroom ,next(doubleBedroom ,
,→ back)) && some(singleBedroom ,next(
,→ singleBedroom ,back)) && (next(
,→ kitchen ,fron) || next(kitchen ,
,→ dinningRoom) || next(kitchen ,
,→ homeOfice)) && (next(livingRoom ,
,→ fron) || next(livingRoom ,kitchen)
,→ || next(livingRoom ,dinningRoom)).
The shape grammar rules and the alternative
compositions of the process offer the possibility to Figure 6
generate a shape composition among different al- An exemplary
ternatives and thus a non-deterministic perspective conditional shape
is followed in the design process. Operationally, the rule.
design process applies repeatedly shape rules to the
intermediate shapes obtained at each step accord-
ing to the sequence order established in the process.
Due to non-determinism, each application of the pro-
cess to an initial shape composition may produce dif-
ferent designs by forward chaining using some op-
erational preference in the choice of the alternatives.
Each time a test process fails or a shape grammar rule
fails to apply, the system backwards trying to build a
different solution.
The tree of Figure 7 exemplifies the generation
process of houses layout design included in circles
fulfilling the design rules present so far for a 17x14
module house with the following rooms and dimen-
sions: 3x3 vestibule, 3x3 technical room, 9x6 garage,
3x2 toilet, 3x3 storage, 2x7 kitchen, 4x7 dining room,
5x7 living room, 4x7 double bedroom, 4x7 single
bedroom, 4x7 single bedroom with hall and 3x5
Figure 7
An example of the
generation process
of house layout
design.

SHAPE, FORM AND GEOMETRY | Grammars and Concepts - Volume 2 - eCAADe 34 | 355

bathroom.
A workbench "Alternative Shaper" supporting
this approach has been developed for exploring di-
verse case studies and is actually adapted to imple-
ment the referred design system (see Figure 8).
This first tool was implemented in SWI-prolog in
order to explore a depth first search strategy of the
design solution space.
Not only non-determinism of the procedural part
is intensively used to explore the design solution
space but also is used to describe sketchy and incom-
plete design processes on the specification level. As a
result, this approach allows also design specifications
emphasizing design properties instead of processes.
Genetic Algorithms Implementation
Evolutionary algorithms have been positioning as
flexible and useful tools within diversified research
areas. Based on the Theory of Evolution, more exactly
on the notion of the survival of the fittest, Genetic Al-
gorithms (GA) are able to search for floorplan design
solutions, evaluate them and continue the search for
the better fitted floor plans based upon that evalu-
ation. We propose an approach where the evalua-
tion criteria are the representation of the generative
shape rules considering the general positioning of
the house required rooms. A prototype tool, Shaper-
GA, using a genetic algorithm to generate several
356 | eCAADe 34 - SHAPE, FORM AND GEOMETRY | Grammars and Concepts - Volume 2
Figure 8
Alternative Shaper
Workbench.
floor plan design alternatives is presented.
In GA terms, a solution to a problem is an indi-
vidual or chromosome, the existent solutions at each
stage are grouped as a population, and the creation
of a new population is a generation. Each individ-
ual, representing a house layout, is a group (array)
of binary matrices, the genes, each of these repre-
senting a house division and its actual positioning in
the grid. The fitness of an individual corresponds to
the value of the objective function that is evaluated
in terms of the floor plan positioning rule's compli-
ance. The three classical genetic operators - selective
reproduction, crossover and mutation - are used for
design evolution. The GA starts with a random gener-
ation of an initial population, evaluates the individual
designs created, applies genetic operators to (some
of ) these individuals, producing a new generation of
floor plans, and loops until termination. Such a tool
enables effective automatic exploration of the design
space, evolving floor plans already found according
to the predefined criteria towards the desired goal.
Evolutionary computation as a technique is not
new to architectural design, specially for creative
generative design tools (Caldas and Norford 2002;
Steadman 2008; Knecht and König 2010; Pasternak
and Kwiecinski 2015). The GA is here employed as
a generative tool with evolution guided by a fitness
function that embodies the shape grammar rules,
Figure 9
Genetic Algorithm
Implementation.
thus measuring the aptitude of a generated layout in
terms of rules' fulfillments. The areas and rooms rel-
ative positioning rules are incorporated into the fit-
ness function and the violation of any of them results
in a fitness penalization. The algorithm also needs
the specific user information on the floor plan to be
generated, namely, which rooms and side dimen-
sions are to be inserted to begin the generative pro-
cess and design evolution towards feasible layouts.
This approach, Shaper-GA, begins by randomly
generating individuals (layouts) where each com-
prises the predefined needed rooms. However, be-
ing random, most of these layouts will not be fea-
sible because of overlaps or rules' failure of compli-
ance. Through the referred incorporation of penal-
ties in the fitness function and usage of this knowl-
edge in parental selection and for reproduction, the
best fitted individuals (houses' layouts that best fol-
low the design rules) are selected and their genes
propagated to the offspring. From the overall set of
parents and their offspring, the better fitted houses
are chosen for the next generation, Thus, the re-
peated process evolves the better layouts until being
able to generate completely feasible floor plans.
Shaper-GA has undertaken the necessary trial ex-
periments with fixed specifications much alike the ex-
ample described in the previous section: all the re-
quired rooms have fixed dimensions, and thus also
SHAPE, FORM AND GEOMETRY | Grammars and Concepts - Volume 2 - eCAADe 34 | 357
fixed orientations. The experimentation results are
very favorable, with Shaper-GA evolving to feasible
floor plans (two examples on the left-hand side of
Figure 9) within reasonable computational time. The
visual interface will display all of the feasible layouts
(left and right arrows at the bottom of Figure 9). Also
any of the remaining (lesser) fitted solutions found
in the final evolutionary generation can be visual-
ized (example on the right-hand side of Figure 9).
The interface is prepared to show product layouts
where some of the relative rooms' positioning rules
have been relaxed thus allowing for the emergence
of other designs where the user might see fit.
RESULT AND DISCUSSION
Both mentioned implementations are actually sup-
porting test experiments for generating house lay-
outs that comply with a language of design and re-
spond to user requirements. Up to now, for small
houses' layouts, the process shape grammar imple-
mentation produces solutions in just a few seconds,
faster than the genetic algorithm shape grammar im-
plementation. However, the number of potential so-
lutions is combinatorial on the number of divisions
and their side dimensions. Thus, when the number of
divisions increase and their sizes change, we are ex-
pecting that the process shape grammar implemen-
tation will take a considerable more amount of time
 just to produce a solution. That's where the genetic
algorithm counterpart plays an important role.
Shaper-GA implementation is being upgraded
to allow for evolutionary readjustment of the rooms
dimensions, according to predefined interval range
values. Meanwhile, the best way to integrate the
connections between rooms into the evolutionary
generation is also being studied. More research will
be done in order to develop fully efficient prototype
software for both approaches. For that we'll be using
heuristics to find paths of promising generating de-
signs and speeding up the algorithm so that several
optima are quickly found. One additional approach
will also be explored for the cases where there are
non-optimal solutions, i.e. there are no solutions sat-
isfying all the shape conditions. For that situation the
genetic algorithm approach will be used to integrate
a multi-criteria representation of customer's prefer-
ences.
The results found so far, make us believe that
these generative tools are promising ones that satisfy
the users' needs and can be used by non designers in
order to explore housing design solutions.
ACKNOWLEDGEMENTS
This work was partially supported by Center For
Advanced Studies Warsaw University of Technology
through Outgoing Scholarship No. CAS/37/POKL.
REFERENCES
Benros, D, Granadero, V, Duarte, JP and Knight, T 2011,
'Automated Design and Delivery of Relief Housing:
The Case of post-Earthquake Haiti', CAAD Futures:
Designing Together, pp. 247-263
Caldas, LG and Norford, LK 2002, 'A design optimization
tool based on a genetic algorithm', Automation in
Construction, 11(2), pp. 173-184
Coimbra, E and Romao, L 2013 'The Rehabilitation De-
sign Process of the Bourgeois House of Oporto:
Shape Grammar Simplification', eCAADe 2013: Com-
putation and Performance
Colakoglu, B 2005, 'Design by grammar: an interpreta-
tion and generation of vernacular hayat houses in
contemporary context', Environment and Planning B:
Planning and Design, 32(1), pp. 141-149
358 | eCAADe 34 - SHAPE, FORM AND GEOMETRY | Grammars and Concepts - Volume 2
Duarte, JP 2001, Customizing mass housing: a discursive
grammar for Siza, Ph.D. Thesis, Massachusetts Insti-
tute of Technology
Eloy, S 2012, A transformation grammar-based methodol-
ogy for housing rehabilitation, Ph.D. Thesis, Universi-
dade Técnica de Lisboa
Eloy, S, Vermaas, P and Andrade, M (s.d.), 'The Quality of
Designs by Shape Grammar Systems and Architects:
A Comparative Test on Refurbishing Lisbon', Journal
of Architecture and Planning Research, (forthcoming)
Flemming, U 1987, 'More than the sum of parts: the
grammar of Queen Anne houses', Environment and
Planning B: Planning and Design, 14(3), pp. 323-350
Grasl, T 2012, 'Transformational Palladians', Environment
and Planning B: Planning and Design, 39(1), pp. 83-95
Knecht, K and Konig, R 2010, 'Generating Floor Plan Lay-
outs with K-d Trees and Evolutionary Algorithms',
Generative Art Conference, , pp. 239-253
Kwiecinski, K and Slyk, J 2014 'System for customer par-
ticipation in the design process of mass-customized
houses', Fusion: Data Integration at its Best, Vol 2, pp.
207-215
I-kang Li, A 2002 'A prototype interactive simulated
shape grammar', Design education: Connecting the
Real and the Virtual, pp. 314-317
Ligler, H and Economou, A 2015 'Entelechy I: Towards
a Formal Specification of John Portman's Domestic
Architecture', Real Time - Extending the Reach of Com-
putation, Volume 1, pp. 445-452
McKay, A, Chase, S, Shea, K and Chau, HH 2012, 'Spa-
tial grammar implementation: From theory to use-
able software', Artificial Intelligence for Engineering
Design, Analysis and Manufacturing, 26(02), pp. 143-
159
Pasternak, A and Kwiecinski, K 2015 'High-rise Building
Optimization A Design Studio Curriculum', Real Time
- Extending the Reach of Computation, Volume 1, pp.
305-314
Santos, F and Esmerado, J 2015, 'A different shape gram-
mar approach for automatic design generation', In-
ternational Journal of Advances in Computer Science
& Its Applications, 1, pp. 90-97
Steadman, P 2008, The Evolution of Designs: Biological
Analogy in Architecture and the Applied Arts, Rout-
ledge
Stiny, G 1980, 'Introduction to shape and shape gram-
mars', Environment and Planning B: Planning and De-
sign, 7, pp. 343-351
Strobbe, T 2015, Computer-Aided Exploration of Architec-
tural Design Spaces: a Digital Sketchbook, Ph.D. The-
sis, University of Gent
Utilizing Gradient Analysis within Interactive Genetic
Algorithms
Matthias Kulcke1 , Wolfgang E. Lorenz2
1
Hamburg University of Technology/HafenCity University Hamburg 2 Vienna Uni-
versity of Technology
1
matthias.kulcke@tu-harburg.de 2 lorenz@iemar.tuwien.ac.at

The paper describes and discusses the possible integration of gradient analysis,
as a method and tool for architects and designers to analyze the degree of
proportion-complexity of a design, into the process of designing an object
utilizing interactive genetic algorithms (IGA). A VBA implementation for
AutoCAD has been developed by the authors, enabling to test the usability of
genetic algorithms (GA) for minimizing the angle-redundancy and
length-redundancy quotient. The gradient analysis itself has been developed on
the basic assumption that the complexity of an objects appearance is reduced by
redundancy, which can be measured focussing on different levels of comparison;
among others e. g. variety of material, colour-combinations and proportion. The
latter comes under scrutiny if the method of gradient analysis is applied.

Keywords: Gradient Analysis, Interactive Genetic Algorithm, Design

Complexity, Redundancy, Spatial Analysis, Form and Geometry, Proportion

THE ALGORITHM
The complexity of a given architectural object or ob-
ject arrangement can be measured on the level of
gradient relations, the gradients itself being the rep-
resentations of the proportion of each rectangle that
is defined by each pair of geometric points. Using
the gradient analysis means regarding the frequency
of repetition of gradients, comparing every signifi-
cant point with all the other points in a 2D CAD rep-
resentation of the object one by one and succes-
sively listing pairs of points and their gradients (Kul-
cke and Lorenz 2015). In addition the distance be-
tween points is taken into account; the gradient anal-
ysis returns an angle quotient and a length quotient,
each of which can be used as a fitness-value within
a genetic algorithm. This is tested by the authors on
simple facade elements with fixed main outlines and
two openings with varying corner-points. The co-
ordinates of the corner-points are translated into bi-
nary DNA (1st step: encoding). After using crossover
(2nd step; with or without fitness check beforehand)
and mutation (3rd step) on a population of n-parents
(chromosomes), angle quotient and length quotient
are used as fitness-values (4th step) to determine the
segment-size on Goldbergs weighted roulette wheel
(see Coates 2010, König 2010). By digitally spin-
ning the roulette wheel, members for the succeed-
ing breeding population are picked to be proposed
(element of chance); out of these the user choses
1 or 2 favourites. Taking this choice into account
the next breeding population goes through the algo-
rithm (see flowchart, Figure 1).

SHAPE, FORM AND GEOMETRY | Grammars and Concepts - Volume 2 - eCAADe 34 | 359
 Figure 1
Flowchart
representing the
interactive genetic
algorithm.

360 | eCAADe 34 - SHAPE, FORM AND GEOMETRY | Grammars and Concepts - Volume 2
Figure 2 posed the application of an interactive genetic al-
Invalid results. gorithm in design tasks, that allows users to influ-
ence the choice of candidates out of each succeeding
population (Poirson et al. 2010). They have applied
this interactive genetic algorithm e. g. integrating
customers in the process of a car dashboard design
by giving them an automatically generated selection
of visualisations of individuals of each generation to
chose 1 or 2 favourites from. In their case the user's
choice influences the segment size of the chosen in-
Figure 3 dividual on the roulette wheel which is spun after-
Using an alternative wards.
crossover mask to
achieve valid
Using Different Crossover Masks Depend-
results.
ing on the Results
The algorithm applies crossover and mutation in the
automated first part. If the crossover of two parents
fails to produce a valid child, e. g. a set of coordinates
that creates two window openings with overlap or
other flaws (Figure 2), then the binary description
of these coordinates is mutated. Another approach
could be to use a different mask for the crossover if
invalid solutions are produced (Figure 3).

THE AUTOMATED PART OF THE ALGO-

Alternative Use of the Roulette Wheel
Different orders of the steps for reaching the next
population are also possible. Poirson et al. have pro-
RITHM
Steps 1 to 4 of the algorithm (for the time being
still limited to a fixed rectangle, defining the main
outline and two openings) make up the automated
part without possibility for user intervention. Both,
step 2 (crossover) and step 3 (mutation) are for the
time being applied to each single element (of the
pool of possibilities) influenced by probability values.
Crossover masks have been applied in two ways: ei-
ther the binary code is cut into two pieces at any po-
sition or just in the middle. The latter results in sim-
ply switching the coordinates for the second window,
since the program currently takes only two windows
into account (each consisting of 4 values: the x- and
y-coordinates of the bottom left corner and the x-
and y-coordinates of the top right corner). The au-
thors utilized different versions of the program to ad-
just the automated part of the IGA with adjustments

SHAPE, FORM AND GEOMETRY | Grammars and Concepts - Volume 2 - eCAADe 34 | 361
 Figure 4
Development of the
facade through 12
generations.

362 | eCAADe 34 - SHAPE, FORM AND GEOMETRY | Grammars and Concepts - Volume 2
 still ongoing. The macro program used in AutoCAD
returns a set of designs starting with the initial in-
dividuals in the left column, with fitness decreasing
from top to bottom row. From left to right differ-
ent time steps are visualised with a finishing result
on the right. Differences between elements of the
last column are minimised as a (local) optimum is
reached (Figure 4).The initial aim to reach a design
that produces a minimal angle-redundancy quotient
and a minimal length-redundancy quotient is ques-
tioned by the results. Left unchecked by rules and
constraints this aim produces openings decreasing in
size, which is the logical consequence if the only goal
is to escape complexity in proportion as much as pos-
sible (Figure 5).
Figure 5
Openings
decreasing in
size/tending toward
extreme
proportions.
THE FITNESS FUNCTION AND USER
CHOICE INTEGRATION
The integration of gradient analysis in an interactive
genetic algorithm aims rather at didactic and pro-
fessional application as opposed to customer-use,
therefore it should be possible to optionally add the
fitness-values of each individual of the population
the user can choose from in text. This is done in a
two step process; first, the favourites are chosen by
visual preference, than the fitness-value representing
the degree of complexity is made to appear and the
user can alter his decision, which is now influenced
by intuition, perception and analysis-output. To al-
low for comfortable interactivity and easy access the
SHAPE, FORM AND GEOMETRY | Grammars and Concepts - Volume 2 - eCAADe 34 | 363
algorithm is rewritten in PHP and using the <canvas>
element in HTML5 (Figure 6).
Left uninfluenced in the individual evolutionary
steps, the genetic algorithm aiming for the lowest
angle-quotient possible applying the gradient anal-
ysis as a fitness value shows certain tendencies that
might not be wished for in a real building environ-
ment, e.g. window/door-openings tend to go for ex-
treme i.e. slim proportions resp. decrease in the size
of their area. In a fully automated context this would
call for corrective rules and constraints, in the case
of an interactive genetic algorithm avoiding extreme
proportions may be left to the user, if he/she wishes
to avoid them. In addition the degree of complexity
reduction could also be chosen by the user adjusting
the angle-quotient; an ideal angle-quotient could be
set by the user, either beforehand or in the course of
searching for the facade wished for. Functions could
be added accordingly to the proposed IGA.
DISCUSSION AND OUTLOOK
The current results are promising as the fitness of the
results increases after only a few steps. However, fur-
ther program adjustments have to be be examined
in detail. Designs allowing for more windows will be
tested as well, in order to gain experience regarding
the performance. The integration of user choices is
of main focus as the authors intend the tool to be uti-
lized as a cognitive and analytic aid during the design
process and not as an automatic design generator.
Currently the determination of the best position of
user intervention within the GA, thus making it inter-
active, is still ongoing. An advantage of using inter-
active genetic algorithms instead of fully automated
ones is that fewer iterations are needed to come to
a conclusion, but more importantly a dialogue be-
tween the algorithm and the user is introduced. With
the integration of Gestalt related methods like the
gradient analysis, interactive genetic algorithms be-
come didactic instruments on the aesthetic and cog-
nitive level, combining theory and design practise.
To further develop an asynchrone dialogue be-
tween designers and users in a form that conse-

quently upholds the initial idea of user-involvement
through an IGA that still allows for professionally re-
sponsible solutions, a next step will be to list and for-
mulate necessary rules and constraints that can be
used for an accompanying or closing analysis of the
(final) instances that are created by the genetic algo-
rithm in collaboration with the users' choices.
REFERENCES
Coates, P 2010, The Programming of Architecture, Rout-
ledge, New York
Kulcke, M and Lorenz, W 2015 'Gradient-Analysis:
Method and Software to Compare Different Degrees
364 | eCAADe 34 - SHAPE, FORM AND GEOMETRY | Grammars and Concepts - Volume 2
Figure 6
The algorithm in
PHP and HTML5.
of Complexity in the Design of Architecture and
Designobjects', Martens, Bob; Wurzer, Gabriel; Grasl,
Thomas; Lorenz, Wolfgang; Schaffranek, Richard. Real
Time – Proceedings of the 33rd International Confer-
ence on Education and Research in Computer Aided
Architectural Design in Europe, Wien, pp. 415-424
König, R 2010, Simulation und Visualisierung der Dy-
namik räumlicher Prozesse, VS Verlag für Sozialwis-
senschaften, Wiesbaden
Poirson, E, Petiot, J, Aliouat, E, Boivin, L and Blumenthal,
D 2010 'Study of the convergence of Interactive Ge-
netic Algorithm in iterative user’s tests: application
to car dashboard design', Proceedings of IDMME - Vir-
tual Concept 2010, Bordeaux
3D Reconstruction Survey of Complex Informal
Settlements
Towards an understanding of the genesis of form

Debora Verniz1 , Luis Mateus2 , José Pinto Duarte3 ,

Victor Ferreira4
1,2,4
Faculdade de Arquitetura da Universidade de Lisboa 3 Penn State University
1,2,4
{20143014|lmmateus|victor}@fa.ulisboa.pt
3
jxp400@psu.edu

The Brazilian favelas are a kind of informal settlements characterized by steep

topography and a maze-like structure. Like many other settlements of its kind that
are prevalent in developing countries today, they are often considered a problem
rather than a solution. This paper is part of a larger research that taking a
realistic stand aims to understand the formal structure of this city-shaping force
and capture it into a rule-based, computational model. The goal is to develop
appropriate requalifying procedures to intervene in informal settlements and
guidelines for designing formal settlements in similar sites. The paper is focused
on the use of digital technologies to reconstruct in 3D a favela used as a case
study, a preliminary step to understand its formal structure and create a
computational model.

Keywords: 3D reconstruction, informal settlement, Santa Marta, SFM, 3D Point

Cloud

INTRODUCTION cess that occurred in that decade when public opin-

Informal settlements (also called slums or favelas)
are strictly related with urban growth. According to
Davis (Davis 2006), these phenomena were first iden-
tified in France, North America, and India, in the 19th
century. The first Brazilian slum was founded in the
1880s in Rio de Janeiro, with the occupation of a hill
called "Morro da Providência" by poor people (Davis
2006) (Compans 2007).
Brazilian slums were not considered part of the
city by municipalities until the 1980s. The change in
viewpoint took place with the democratization pro-
ion started to reject extreme policies, including infor-
mal settlements removal. The possibility of requali-
fying and urbanizing such informal settlements was
then considered for the first time in Brazilian history
(Compans 2007). Today such settlements are part of
the cityscape in many cities, including Rio, where to-
pography has given them distinctive features.
This paper is part of a larger research that aims to
understand the formal structure of such informal set-
tlements, define guidelines for their requalification,
find compositional rules that help to explain their

SHAPE, FORM AND GEOMETRY | Grammars and Concepts - Volume 2 - eCAADe 34 | 365
 growing process, and ultimately, develop guidelines
for the design of planned settlements in similar to-
pographic conditions. The research uses the favela
Dona Marta as a case study and encompasses the
methodological steps described in the next section.
There were several reasons for choosing favela
Dona Marta. It is located in Rio de Janeiro and it has
had a big media exposure since the 1990s. It ap-
peared in two video clips (They don't care about us,
from Michael Jackson and Put in a love song, from Be-
yoncé and Alicia Keys) and two movies (the Brazilian
Elite Squad, the enemy within and the North Ameri-
can Fast Five.) This exposure to the public makes of
favela Dona Marta a paradigm of the image that peo-
ple have formed regarding Brazilian favelas. More-
over, it is located in the south part of the city, near
a prime area and close to Christ Redeemer, an im-
portant Brazilian monument. Not least important,
this favela has been pacified, which enabled us to es-
tablish links with the Residents Association and, ulti-
mately, have access to the favela.
STATE OF ART
The ultimate goal of the research is to build a com-
putational model of favela Dona Marta. The idea is
to capture the rules underlying the genesis of the
form into a shape grammar, while understanding the
relations between such form and physical site fea-
tures, socio-economic conditions, and technological
aspects, including the availability of materials. The
idea of using grammars to capture the generation
of urban form has its roots on previous work on the
development of a shape grammar for the Marrakech
Medina (Duarte et al. 2007). Like favelas, the Mar-
rakech Medina has a complex shape that grew in-
crementally over time without a clear plan; it was
the result of local rules expressing social conventions
and physical conditions, rather than the outcome of a
top-down approach. However, the steep topography
associated with favelas, makes it a unique case. More
recent work has used grammars to describe Mozam-
bican slums (Barros et al. 2013) and even Brazilian
favelas (Dias 2014). The former case has a signifi-
cantly different structure, partly because sites are al-
366 | eCAADe 34 - SHAPE, FORM AND GEOMETRY | Grammars and Concepts - Volume 2
most flat. The latter case addressed Rocinha, one of
the largest favelas in Rio that shares the steep site
features of favela Dona Marta. However, due to its
large size, Rocinha does not have a homogeneous ur-
ban fabric and still poses security challenges, which
makes it more difficult to study. This meant that the
study could not rely on an accurate survey of the
site. To avoid this limitation, the current study uses
cutting-edge 3D reconstruction technology to survey
the pacified favela, which is the focus of this paper.
The automated image-based 3D reconstruction of
sites and buildings has its roots in the work by Noha
Snavely, which resulted in the open source Bundler
software (Snavely 2006). This software implements
the structure-from-motion approach, in which large
collections of images can be oriented in a single co-
ordinate frame to produce a sparse 3d point cloud
model. Then the bundle of images can be used to
produce a dense point cloud model. One of the first
tools for this final procedure was the patched-based
multi-view stereo (PMVS) and the clustering views for
multi view stereo (CMVS) software packages devel-
oped by Furukawa et al. (Furukawa 2014). After this
work, other tools have been developed that take ad-
vantage of the possibility of using the graphics pro-
cessing unit (GPU) [1] and multi-core processing (Wu
2011).
METHODOLOGY
The first methodological step is concerned with gath-
ering and creating accurate representations of the
settlement, including 2D drawings and 3D models,
which will permit to study the relationships between
site features, particularly topography, and the com-
plex shape of the built structures. The second step
is to develop a shape grammar aimed at explaining
the genesis and structure of the form. This paper is
focused on the 3D reconstruction of the case study,
that is, on the development of a 3D digital model
using various computer-based techniques. This step
consists of two sub-steps, one that uses Google Earth
images, described below, and another that uses im-
ages gathered directly in the real environment, the
latter to be addressed in a future article.
 The workflow to produce the first exploratory three- construction, where image focal distance is refined
dimensional model was as follows: and radial distortion is calculated, following a self-
calibration approach that is then used to perform a
• image gathering from Google Earth;
sparse 3D reconstruction, in which the relative orien-
• 3D sparse reconstruction from such images
tations of the images are recovered.
using a structure-from-motion approach
Figure 1 (SFM);
Sparse • 3D dense reconstruction using a multi-view-
reconstruction of stereo approach (MVS);
favela Dona Marta • triangulated mesh generation using Poisson
surface reconstruction technique;
• texturing the triangulated mesh via image
projection.
The software Google Earth Pro shows many places re-
constructed in 3D and it was used to obtain site im-
ages from favela Dona Marta. Given limitations in vis-
iting the site, it was considered the hypothesis that a
sufficiently good 3D reconstruction was possible ex-
tracting views from the existing Google's 3D mod-
els, and processing them using an SFM approach. In
fact, this software allows one to save images of the
screen like a camera. To do this work, 417 images
Figure 2 were taken with resolution set to 1064 x 659 pix-
Dense els. The images were taken from different viewpoints
reconstruction of surrounding the case study, as suggested by Kout-
favela Dona Marta soudis (Koutsoudis 2013) and with some approxima-
tion to capture as much details as possible. This sur-
vey helped us to gain some basic understanding of
the site, permitting us to plan short but effective site
visits for information-gathering in the real environ-
ment, needed to build more accurate models.
The Google Earth images were processed in a
notebook with an intel i5 processor 1.60GHz with
6GB of RAM memory and an Nvidia GeForceTM 740M
graphics card with 2GB of dedicated memory. To
generate the 3D model it was used the software Vi-
sual SFM. The steps to obtain a sparse 3D reconstruc-
tion were the following: i) scale invariant features
transform (SIFT), where the software looks for charac-
teristic elements in the images, ii) compute matches,
where the software looks for correlated features in After the Sparse Reconstruction was completed us-
all possible image pairs and shows a matrix graphic ing Visual SFM (Figure1), it was necessary to do a 3D-
where the matching probability between image pairs dense reconstruction (Figure 2). For this purpose it
is shown by means of a color code, iii) sparse re- was used the CMVS software (Furukawa 2014), which

SHAPE, FORM AND GEOMETRY | Grammars and Concepts - Volume 2 - eCAADe 34 | 367
 can run within the Visual SFM environment, CMVS Figure 3
takes as input the Visual SFM output (Furukawa et al Final 3D mesh
2010). Both in the sparse and in the dense recon- model of favela
structions there are several parameters that can be Dona Marta
manipulated and can influence the quality and res-
olution of the final point cloud models as well as the
processing time. These parameters also have to be
managed taking into account hardware capabilities.
CMVS creates a point cloud that has to be
meshed and refined to become a more realistic
model (Bartos et al 2014). This was done using Mesh-
Lab. Cleaning the dense point cloud is an impor-
tant step to have successful mesh generation, be-
cause it removes off spurious points generated by
the dense reconstruction process. After this proce-
dure, it was created a triangulated surface using a
Poisson surface reconstruction. Several reconstruc-
tion depths (octree depth) were tried. By increasing
the octree depth, more dense and detailed meshes Figure 4
can be obtained up to the limit of the point cloud res- Detail from the final
olution.The following step is coloring the model. This model of favela
can be done in two different ways: i) transfer the color Dona Marta
from the point cloud to the mesh vertices (Figure 3),
and ii) projecting the color from the oriented images
to the mesh surface (Figure 4). The second approach
gave better results because color is stored as a texture
map rather than in a color per vertex format.
Although it was possible to incorporate the scale In addition to the final 3D model, a 2D site plan
and orientation of the model at the stage of the was created using a topographic charter and images
sparse reconstruction using known coordinates of from Google Earth Pro. Pathways were drew over the
reference points, we used Helmert transformation topography, in those areas where it was clearly visible
to recover scale, position, and orientation after the in pictures, taking also into consideration the align-
mesh model was finished. The transformation pa- ment of buildings and looking for logical aggrega-
rameters were calculated using JAG3D software [2] tions, for instance squares, streets, alleys, and so on.
and the implementation of the transformation was Then, public and green spaces, big topographic gaps,
done in Meshlab. and public transportation pathways were identified
(Figure 5).

368 | eCAADe 34 - SHAPE, FORM AND GEOMETRY | Grammars and Concepts - Volume 2
Figure 5
2D site plan of
favela Dona Marta
Finally, the 3D model and the site plan obtained were
represented in the same coordinate system to guar-
antee that spatial information was aggregated, en-
abling cross-referencing. The process of taking im-
ages and creating the 3D model strictly followed
the steps recommended by reference authors (Kout-
soudis 2013). The 3D model obtained as a result en-
ables one to see the volumes of the built structures
and how they go over the hill, being barred only by
stones or higher slant, as initially intended. However,
because data was not taken directly from the real en-
vironment, the level of definition is low and limited
by the resolution offered by Google Earth.
CONCLUSION
The creation of the slum's map was important to un-
derstand the process by which public spaces were
formed. Preliminary results suggest that in these
kinds of unplanned, informal settlements, pathways
are defined by building aggregation and, in a smaller
degree, by the need for guaranteeing the continu-
ity of movement through the urban space. Although
SHAPE, FORM AND GEOMETRY | Grammars and Concepts - Volume 2 - eCAADe 34 | 369
the Dona Marta favela seems to be densely built, a
more accurate analysis permitted by the developed
3D model shows that there are open public and pri-
vate spaces, as well as green areas. These preliminary
results suggest that the informal city replicates, to a
certain extent, the typologies of the formal city.
The process of creating the slum's 3D model and
site plan from Google Earth was a preliminary step
that was important to understand the basic structure
of the environment. It also was useful to identify ade-
quate ways of collecting images directly from reality
so to obtain better results, that is, more accurate and
higher definition representations. Results show that
the data gathering process in the real environment
will be very demanding due to the complexity of the
form. Future work will be concerned with the devel-
opment of a 3D model and site plan, using such data.
ACKNOWLEDGEMENTS
The work of Debora Verniz was funded by grant
207028/2014-1 from the Science without Borders
Program administered by the Brazilian National
Council for Scientific and Technological Develop-
ment (CNPq).
REFERENCES
Barros, P, Beirão, J and Duarte, JP 2013 'The Language of
Mozambican Slums', Proceedings of the 31st eCAADe
Conference 2013, Delft, pp. 715-724
Bartos, P, Sobavá, W, Pukanská, K and Weiss, G 2014,
'Analysis of low cost photogrammetric procedures
in the process of historical objects survey', Interna-
tional journal of science commerce and humanities,
2(3), pp. 203-210
Compans, R. 2007, 'A cidade contra a favela - A nova
ameaça ambiental', Revista Brasileira de Estudos Ur-
banos e Regionais, 9(1), pp. 83-99
Davis, M. 2006, Planet of Slums, Verso, New York
Dias, MA 2014, 'Informal Settlements: a shape Grammar
approach', Journal of Civil Engineering and Architec-
ture, 8(11), pp. 1389-1395
Duarte, JP, Rocha, JM and Soares, GD 2007, 'Unveiling
the structure of the Marrakesh Medina: a shape
grammar and an interpreter for generating urban
form', Artificial intelligence for Engineering Design,
Anallysis and Manufacturing, 21, pp. 317-349
 Furukawa, Y, Curless, B, Seitz, B and Szeliski, R 2014 'To-
wards Internetscale Multi-view Stereo', Proceedings
of IEEE CVPR 2014, Columbus, pp. 1434-1441
Furukawa, Y and Ponce, J 2010, 'Accurate, Dense, and
Robust Multi-View Stereopsis', IEEE Transactions on
Pattern Analysis and Machine Intelligence, 32(8), pp.
1362-1376
Koutsoudis, A, Vidmar, V and Arnaoutoglou, F 2013,
'Performance evaluation of a multi-image 3D recon-
struction software on a low-feature artefact', Journal
of Archaeological Science, 40, pp. 4450-4456
Snavely, N, Seitz, SM and Szeliski, R 2006 'Photo Tourism:
Exploring image collections in 3D', Proceedings of
SIGGRAPH 2006, Boston
Wu, C, Agarwal, S, Curless, B and Seitz, SM 2011 'Mul-
ticore Bundle Adjustment', Proceedings of IEEE CVPR
2011, Colorado Springs
[1] http://www.cs.unc.edu/˜ccwu/siftgpu/
[2] http://javagraticule3d.sourceforge.net/

370 | eCAADe 34 - SHAPE, FORM AND GEOMETRY | Grammars and Concepts - Volume 2
Optimization of the Building in Relation to the Insolation
Conditions of Premises in Adjacent Buildings
Agata Pasternak1
1
Warsaw University of Technology, Faculty of Architecture
1
agata.pasternak.me
1
agata.pasternak@arch.pw.edu.pl

When designing buildings in dense city centers, a very important step is to study
the nearest surroundings of the plot, in order to enable the best possible
placement of the building. Many aspects must be taken into account including
urban, legislative, environmental and aesthetic factors. These factors, in
particular the legislative ones, depend on local conditions. This article describes
a methodology for the analysis of insolation of surrounding buildings in the
context of local building regulations. A procedure is described that significantly
accelerates this phase of the design process and permits integrating it with an
optimization process concerning the location of the building on the plot.
Examples of application of that procedure are presented and their limitations and
capabilities are discussed.

Keywords: optimization, genetic algorithm, insolation, raytracing, sunlight

analysis

INTRODUCTION buildings. Discussed plot is not covered by any lo-

During the design process of buildings in dense city
centers, a very important step is to study the near-
est surroundings of the plot to permit the best in-
tegration of the building into its location. There
are many aspects of the context that need to be
taken into account and carefully analyzed, such as ur-
ban, legislative, environmental and aesthetic factors.
These factors, in particular the legislative ones, de-
pend strongly on local conditions and specific regu-
lations.
The following study was carried out in the down-
town of Warsaw, near the central station. In this area
of the city there are buildings diversified in terms of
height and function, there is a lot of high-rise office
buildings, as well as tenement houses or apartment
cal spatial development plan, therefore the form of
its development will depend on the received build-
ing conditions. This document is issued on request
for the specific plot on the basis of an attached design
proposal together with a description of its character-
istics, such as the supply of utilities, the size and use
of the buildings and plot development specificity.
The subjected plot of land has an area of 11,000
square meters and has been intended for office de-
velopment. Because of its large surface area and di-
verse architectural and urban context, the manner of
its development may differ greatly, depending on the
vision of the architect and specific conditions result-
ing from the analysis of the context. One of the regu-
lations that may significantly affect the possibility of

SHAPE, FORM AND GEOMETRY | Grammars and Concepts - Volume 2 - eCAADe 34 | 371
 the plot development in dense city centers is the pro-
vision on the insolation of apartments. The planned
investment cannot significantly reduce the quality of
life of other users of the city. The analysis of the im-
pact of the planned investment on the environment
is an important step of its planning process.
Local Building Regulations: Insolation
According to the Minister of Infrastructure Regula-
tion on the technical conditions to be met by build-
ings and their location guidelines on lighting, insola-
tion and occlusion of windows are specified. In this
article only aspects related to insolation of premises
are discussed and the process of their analysis is de-
veloped in detail.
In the above-mentioned provisions, the issue of
insolation of premises is expressed in the number of
hours for which each window should be illuminated.
Those provisions concern only apartment buildings,
schools, kindergartens and nurseries. The insolation
time shall not be less then 3 hours, measured on the
spring and autumn equinoxes, between 8 am and 4
pm for buildings of educational and care functions
and between 7 am and 5 pm for apartment buildings.
In a multi-room apartment it is allowed to limit
the requirements, referred to above, to at least one
room. In complementary buildings located in down-
town areas it is permitted to limit the time of required
insolation to 1.5 hours, and with regard to the one-
room apartments in such location the required inso-
lation time is not specified.
Described regulations should be met for all win-
dows, which insolation conditions are subject to
change after construction of a new building. De-
pending on the scale of the planned investment, the
scope of insolation analyzes can vary greatly and
the number of windows requiring analysis can reach
hundreds.
Sundial Method
Currently, the method most commonly used in
Poland for insolation analysis of windows was pro-
posed and disseminated by Mieczyslaw Twarowski
(1996). This method involves using a kind of horizon-
tal sundial designated for the moment of equinox. It
illustrates the lengths and angles of the shadow cast
on the full hours of local solar time by a bar with a
nominal height and its multiples. This tool, printed
on tracing paper, as it was inserted in the book "Sun
in Architecture", is used by superimposing it on the
analyzed building plan with the surrounding build-
ings drown to scale.
The sundial can be determined for certain loca-
tion, but usually it was drawn for each city only once
and it was not taking into account differences in lo-
cations within the whole area of the city (Figure 1).
The resulted in the need for some amendments for
each analyzed location or taking large safety margins
(Lose 2015). In densely built-up locations it is not al-
ways possible to accept ranges of safety beyond the
minimum required insolation duration, specified in
regulations. For this reason it is necessary to apply
tools that enable to conduct precise insolation anal-
ysis through the use of computer models.
Figure 1
Sundial for Warsaw
(Twarowski 1996).
Digital Insolation Analysis Methods
On the market there is a number of software for au-
tomated analysis of lighting conditions. These pro-
grams include Google SketchUp with a Shadow Anal-
ysis plugin, Autodesk Revit, Rhinoceros 3d with its
graphical algorithm editor Grasshopper and addi-
tional plugins for weather and energy analysis, such
as Ladybug and Honeybee that connects it with Ener-

372 | eCAADe 34 - SHAPE, FORM AND GEOMETRY | Grammars and Concepts - Volume 2
 gyPlus, Radiance, Daysim and OpenStudio software.
Listed computer programs enable precise determi-
nation of sunlight vectors on a given date and time,
but does not give the possibility to check whether
a particular window or apartment meets the con-
ditions described in the building regulations, men-
tioned above. In order to verify whether a window
is provided with the right amount of light, a method
for counting the time in which this window is illumi-
nated needs to be established.
Figure 2
Maximum Building
Volume determined
by the PRC Analysis
software.
Figure 3
Designation of the
impact area of the
designed building.
Range of shadow
cast on the spring
and autumn
equinoxes of an
extruded area of
the plot. There is also a software dedicated to this type of anal-
ysis of local lighting regulations. PRC Analysis created
by Jacek Markusiewicz (2014) enables the analysis of
windows and whole apartments for compliance with
the local laws of insolation and occlusion. After a pre-
liminary analysis it allows to trim the initial body of
the building so that all of the windows comply with
the regulations. It allows to determine a Maximum
Building Volume (MBV), but the calculation that in-
cludes all of the potentially affected windows is very
time consuming and involves preliminary windows
selection in order to permit the software to deter-
mine the solution (Figure 2). Because of this it is nec-
essary to conduct a series of preliminary analysis us-
SHAPE, FORM AND GEOMETRY | Grammars and Concepts - Volume 2 - eCAADe 34 | 373
ing other software (Kwiecinski and Pasternak 2016).
Repeated changing of software during the de-
sign process is cumbersome (Pasternak and Kwiecin-
ski 2015), therefore a new analysis procedure was cre-
ated in the software that was used for the initial anal-
ysis and determination of the MBV in the PRC Anal-
ysis. It allows to reduce the need to move from one
program to another and at the same time overcome
its the limitations in terms of possible amount of an-
alyzed windows.
The creation of this procedure in a parametric
environment allows also to create a wide variety of
detailed optimization procedures depending on the
needs of particular location and the configuration of
analyzed windows. For example, for a small plot with
a permitted high allowable building area, it may be
justified to create a single maximum building vol-
ume. For a large plot, which enables typologically di-
verse development, the optimized form will look dif-
ferently, depending on the input form and selected
typology.
METHODOLOGY
Within the framework of the developed method of in-
solation analysis of premisses in adjacent buildings,
three stages can be distinguished:
• Initial analysis of insolation of windows in the
existing state;
• Analysis of the impact of the proposed form
on the surrounding buildings;
• Determination of a Maximum Building Vol-
ume or conduct of a custom optimization pro-
cedure.
Insolation analysis
The first step of this process is to prepare a three-
dimensional model of the building surrounding,
which includes the whole impact area of the pro-
posed building form. The upper dimensions of the
building should be established in order to check how
far the shadow of the building during the spring and
autumn equinoxes reach (Figure 3), because it will
 determine the windows which insolation conditions
are potentially a subject to change.
The next step is to determine which windows
belong to apartments or care and education institu-
tions, because their lighting conditions are particu-
larly described in local building regulations. Other
windows are not evaluated in this procedure. It is
only necessary to model the surrounding buildings
to take into account their impact on the lighting con-
ditions that are subject to evaluation (Figure 4).
To satisfy the requirements of building regulations
sufficiently, but not restrict the building capabilities
of the plot too much, it is necessary to precisely es-
tablish the analysis outlines. Since the building reg-
ulations specify that the insolation requirements de-
pend on the size of the apartment, it is necessary to
discernment which windows belong to which apart-
ment and group them accordingly. In multi-room
apartments it is sufficient to meet the requirements
for only one room and one-room apartments do not
have to meet them at all. For this reason, it is neces-
sary to determine an internal functional arrangement
of residential buildings located within the building's
impact zone.
Because the regulations do not clarify to which
extent the windows should be illuminated, there is
a large scope for interpretation of these provisions.
It can be assumed that illumination should be pro-
vided on the entire plane of the window or only on
its selected points or edges. Depending upon which
interpretation is accepted, it is possible to get com-
pletely different boundary conditions for the process
of optimization and hence, obtain different results of
possible spatial building form.
374 | eCAADe 34 - SHAPE, FORM AND GEOMETRY | Grammars and Concepts - Volume 2
The procedure was created in a parametric en-
vironment. Energy Plus weather data file is used to
determine the direction of sun rays for the analyzed
time periods. The intersections of sun rays with build-
ing blocks are checked. Rays illuminating the win-
dow are grouped into two to create time intervals.
Depending the adopted accuracy of the calculations,
they can amount to a minute or more. The number of
intervals is counted and compared with the value of
the required insolation time specified in the regula-
tion.
At first the number of intervals are calculated Figure 4
for the current state, having regard to the existing Shadows cast on
buildings, but without taking into consideration the the analysed plot
planned building. This allows to determine whether by the surrounding
the windows are properly illuminated, before it is buildings.
checked how the new building is affecting their con-
ditions. If the windows are not correctly exposed
they cannot be shaded any more. These windows
designate certain surfaces, which the new building
cannot exceed in no case.
In the second step, the same windows are
checked again, but this time taking into account the
proposed building. The intersections of rays, that
until now were not obstructed by any building, are
checked if they intersect with the new building vol-
ume. The windows, that were previously correctly il-
luminated are checked whether their sun exposure
is not deteriorated to an incorrect level. If it is true,
they will be included in the optimization procedure
as boundary conditions.
The procedure allows to increase the accuracy of
window analysis by adding more control points to
the line located on the sill level on the inner face of
the wall. Depending on the computational capabil-
ities, it is possible to analyze only the center point
of each line or include additional points such as end
points and additional points between these.
The result of the insolation analysis process is the
division of windows into three groups:
• Group 1 - windows that have correct insola-
tion level and are not influenced by the de-
signed building;
 • Group 2 - windows that are insufficiently il-
luminated in the current state, without tak-
ing into account the designed building. There
shall be no sun rays obscured by its volume.
These windows define surfaces that shall to be
exceeded by the designed building;
• Group 3 - windows that were correctly illumi-
nated in the current state and the designed
building affects their level of insolation. Sun
rays obscured by the new buildings are in-
volved in the optimization procedure.
Maximum Building Volume
The procedure allows the determination of a Maxi-
mum Building Volume. The outline of the plot ex-
truded to a maximum height is trimmed so as to
meet the insolation regulations of premisses in ad-
jacent buildings. This procedure takes place in two
stages. For windows belonging to Group 2, all of the
extruded sun rays blocked by the new building are
subtracted from the volume (Figure 5).
Figure 5
MBV delimination -
stage 1.
In the next step sun rays, which are needed for a min-
imum illumination of the windows from the Group 3,
are selected in such a way that the volume subtracted
from the input form is minimized (Figure 6).
This procedure for determining the MBV is deter-
ministic, because its objective is to obtain the largest
possible volume. From the initial form the smallest
total volume designated by extruded sun rays is sub-
tracted. It does not allow, however, to control which
parts of the form are more valuable then other and
protect them from being deleted. For the designer
it is crucial to be able to control the shape and loca-
SHAPE, FORM AND GEOMETRY | Grammars and Concepts - Volume 2 - eCAADe 34 | 375
tion of the building at every stage of the design pro-
cess and at the same time meet all the requirements
arising from the context and regulations. The ability
to use the described analytical method as part of the
form finding process with the use of the optimization
procedure gives the designer more freedom to shape
the form and still meet the legal requirements.
Optimization procedures
Another possibility is to develop a bottom up ap-
proach, which will aim at finding the right configu-
ration of building volumes and their locations as a re-
sult of an optimization process. The objective is that
the proposed forms are not worsening the insolation
conditions of any windows below the legally accept-
able level. The final form depends on the configura-
tion of input parameters and their ranges. Optimiza-
tion procedures are prepared with the use of genetic
algorithms.
RESULTS
As an example of the principles of the proposed
method, a plot located in the center of Warsaw was
selected. In the impact area of the designed build-
ing there are located eighteen residential buildings
with the total number of windows of 1587. These
windows belong to all types of apartments and com-
munication spaces. One-room apartments should be
omitted and from multi-room apartments only one
room, which is to meet the conditions set out in the
regulations, should be selected.
The optimization procedure is designed in such a
way that in order to illuminate a particular window in
the corresponding period of time, the entire obscur-
ing volume is cut above the plane of the sun ray inci-
dent on the windowsill of analyzed window. There-
fore it is only necessary to analyze the windows lo-
cated at the lower floors of the buildings, assuming
that the functional layout is the same on each floor.
If some floor has a different functional arrangement
or other internal division of apartments, then these
windows should also be included in the analysis.
After removal of the windows that are not re-

quired in the analysis, the remaining 107 windows,
were analyzed in detail in terms of meeting the inso-
lation conditions. For parcels located in downtown
areas required insolation time amounts to 1.5 hours.
The insolation analysis shows that there are 53
windows that are properly illuminated in the present
state and after the construction of a new building lo-
cated anywhere on the plot (windows belonging to
the Group 1). Their insolation level does not fall be-
low the required minimum specified in the regulation
for any newly designed form.
There are 27 windows belonging to Group 2.
These are the windows which level of insolation is not
currently in compliance with regulations, therefore,
the designed building cannot worsen their lighting
conditions any more. These windows define the vol-
umes which are entirely subtracted from the initial
building form. Subtracted volumes are created by ex-
truding a surface defined by the two adjacent rays in-
cident on one of the control points located on the sill
line.
Remaining windows, and there are 27 of them,
belong to Group 3. These windows were properly il-
luminated and this condition worsened after the in-
troduction of the newly designed volume. After ver-
ifying how many minutes of the illumination every
window is missing and how many minutes of illumi-
nation is obscured by the designed building, vectors
that are to take part in the optimization process are
determined (Figure 7). Depending on how much in-
solation time is missing, from the set of vectors those
that reduce the Maximum Building Volume as little as
possible are selected.
376 | eCAADe 34 - SHAPE, FORM AND GEOMETRY | Grammars and Concepts - Volume 2
Figure 6
MBV delimination -
stage 2.
Figure 7
Vectors of sun rays
obscured by the
designed building
falling on the
windows, which
sun exposure has
been degraded.
Addressed procedure allows to add additional con-
trol points located on the line of each windowsill.
The above-described results were based on analysis
of two control points located on the end points of
the line. Analysis of three or more control points,
of which at least two are located at the ends of the
line gives the same results. This means that using
these 2 points is the most optimal and gives reliable
results. Perhaps in a more complex spatial arrange-
ment the ability to add more control points would be
more useful, because it would give the opportunity
to catch the sun's rays from among closely located
buildings. If, however, the analysis will take into ac-
count only the midpoint of a line, then less favorable
results will be obtained: Group 1 includes then 34
windows, Group 2 - 47 and Group 3 - 26. We can see
that the number of windows, which satisfy the pro-
visions significantly reduced, from 53, this number
dropped to 34. For 3 control points placed in 0.25, 0.5
and 0.75 length of the line number of windows that
meet the regulations equals 49.
The described results were calculated with an
accuracy of 5 minutes. If the accuracy of the cal-
 culations is reduced to 10 minutes, the number of
windows that meet the requirements decreases to
49, and the number of underexposed windows in-
creases to 33. This is due to adopted rounding of
values. If a calculation is conducted with an accu-
racy of one minute, the results will be the most reli-
able, but the time of calculations will increase signif-
icantly. For such precision of calculations the num-
ber of windows illuminated properly equals to 56, the
number of underexposed windows - 25 and windows
demanding optimization - 26.
The end result of this procedure is the determina-
tion of the Maximum Building Volume, which is rep-
resented by the impassable building envelope, set
within the framework of building regulations con-
cerning insolation of premisses in adjacent buildings
(Figure 8). In order for this volume to treated as final,
other legal conditions and the ones resulting from
the architectural and urban context should be also
taken into account.
Figure 8
Maximum Building
Volume obtained
after the insolation
analysis process.
The resulting Maximum Building Volume is heavily
cut, despite the fact that in the process of creating
its goal was to reduce the volume as little as possi-
ble. The time intervals, which influence on the initial
building volume was the smallest, were selected to
be subtracted from it. However, that this not the only
possible configuration of sun rays needed to illumi-
nate the windows. It is possible to designate a num-
ber of other configurations that also comply with the
regulations, but their arrangement will allow the lo-
cation of other volumes on the site , that does not
fall within the designated maximum volume. Plac-
ing the building on the entire surface of the plot is
SHAPE, FORM AND GEOMETRY | Grammars and Concepts - Volume 2 - eCAADe 34 | 377
not permitted in the building regulations, therefore
the Maximum Building Volume that is not taking into
account other provisions and local factors is useful in
the design process but not categorically conclusive.
It is possible to locate other forms, beyond the MBV
outline determined with the use of individual opti-
mization procedures based on the results of insola-
tion analyzes.
One of the possibilities of implementation of the
optimization procedure is the search for such an ar-
rangement size and location of a rectangular prism
on the plot in order to maximize its cubic capacity,
when there are no underexposed windows (Figure
9). The number of sun rays obscured by the body
of the building falling on initially underexposed win-
dows plus the amount of time needed to illuminate
the initially correctly isolated windows to the mini-
mum required level ofter adding the designed build-
ing are minimized. This procedure may also involve
more than one a rectangular prism or address an ar-
rangement of any other geometric shapes or search
for proper parameters of on arbitrary building shape.
The initial shape of the building is the entire surface
area of the plot extruded to a height of 82 meters. The
result of this optimization process is the body of the
building, which is 30 percent of the input volume and
all of the analyzed windows comply with the building
regulations.
Taking into account a greater number of build-
ing volumes and maximizing the accumulated cu-
bic capacity it is possible to approximate the Maxi-
mum Building Volume. Studying typologically differ-
ent geometries can lead to obtaining forms well fit-
ted to the context and fulfilling building rules.
CONCLUSIONS
The paper presents a method for finding an optimal
building volume and its location on the plot in rela-
tion to the insolation conditions of buildings which
are located in the impact area of the designed build-
ing. The proposed method refers to the national rules
governing the conditions of insolation of rooms and
apartments in adjacent buildings. These regulations
 Figure 9
Building volume
optimisation - one
rectangular prism
located on the plot.

define precisely how many hours of insolation time Markusiewicz, J 2014, 'PRC Analysis. Automating the
should each window that belong to an apartment, shadow analysis', Architectural and Town Planning
school, kindergarten or a nursery have during the au- Qarterly, 2/2014, pp. 57-58
Pasternak, A and Kwieciński, K 2015 'High-rise Build-
tumn and spring equinoxes between certain hours.
ing Optimization - A Design Studio Curriculum', Real
Depending on how tall the building is planned to Time - Proceedings of the 33rd eCAADe Conference -
be and in how densely built-up location, the num- Volume 1, Vienna, pp. 305-314
ber of windows that should be analyzed in terms of Twarowski, M 1996, Słońce w architekturze, Arkady
their potentially affected insolation conditions can
be very large. The proposed method allows simul-
taneous analysis of a large amount of windows and
optimization of the building volume under the given
boundary conditions. The first part of the process is
based on a parametric model, which analyzes data
from the Energy Plus weather data file, in the sec-
ond section the Maximum Building Volume id deter-
mined or optimization procedures based on genetic
algorithm are conducted.
Various optimization procedures were tested
with typologically different input forms. Depending
on the parameter settings, the optimization proce-
dure always resulted in an optimized and well fit-
ted to the requirements of insolation building vol-
ume.The fitness function of the genetic algorithm
minimized the number of underexposed windows
and maximized the volume of the building.

REFERENCES
Kwieciński, K and Pasternak, A 2016 'The Impact of Inso-
lation Regulations on the Determination of the Max-
imum Building Volumes at an Early Design Stage',
Proceedings of ASK.the.Conference 2016, Warsaw
Lose, M 2015, Nasłonecznienie mieszkań. Przepisy, prak-
tyka i rzeczywistość, Cursiva

378 | eCAADe 34 - SHAPE, FORM AND GEOMETRY | Grammars and Concepts - Volume 2
Daylighting Based Parametric Design Exploration of 3D
Facade Patterns
Amartuvshin Narangerel1 , Ji-Hyun Lee2 , Rudi Stouffs3
1,2
KAIST 3 National University of Singapore
1,2
{amartuvshin|jihyunlee}@kaist.ac.kr 3 stouffs@nus.edu.sg

A building façade plays an important role of reducing artificial lighting by

introducing natural light into the interior space. A majority of research and
current technology heavily focuses on the optimization of window properties such
as the size, location, and glazing with the consideration of external shading
device as well as the building wall in order to obtain appropriate natural lit
space. In the present work, we propose a 3-dimensional approach that can
explore the trade-offs between two objectives, daylight performance and
electricity generation, by means of paramedic modeling and multi-objective
optimization algorithm. The case study was simulated under the environmental
setting of the geographical location of Incheon, Korea without any urban context.
Using the proposed methods, 50 pareto-front optimal solutions were derived and
investigated based on the achieved daylighting and generated electricity.

Keywords: Parametric design, façade design, daylight performance,

building-integrated photovoltaics, multi-objective optimization

INTRODUCTION façade, or both, to further improve the sustainability

Daylight is considered the best source of light that
most closely matches human needs (Li and Tsang
2008). Building fenestration is responsible for intro-
ducing daylight into the indoor space, and when a
façade is designed properly, it can reduce the need
for artificial light significantly (Nabil and Mardaljevic
2005; Krarti et al. 2005). Commonly, the façade of
a high-rise office building is considered as a verti-
cally extruded glass envelope that consists of a num-
ber of transparent and opaque glazing layers. In
addition, shading elements may be attached in or-
der to protect the indoor from direct solar radia-
tion for improved indoor comfort. Furthermore, Pho-
tovoltaic (PV) panels or Building-Integrated Photo-
voltaic (BIPV) can be added to the building roof,
factor. All these elements need to be considered si-
multaneously within a sophisticated design method
to achieve better design in terms of indoor comfort
and sustainability.
Not only size and location of the window and
the external shading device have a significant ef-
fect on the level of daylight in a given space. A
vast number of studies have been carried out by re-
searchers, considering these as well as additional fac-
tors such as wall thickness, glazing properties and
the integration of external shading device (David et
al. 2011; González and Fiorito 2015; Sheikh and Ger-
ber 2011; Gadelhak 2013) as well as BIPV (Mandalaki
et al. 2012). However, very few studies take in ac-
count all these factors together, and almost no stud-

SHAPE, FORM AND GEOMETRY | Grammars and Concepts - Volume 2 - eCAADe 34 | 379
 ies can be found of a 3D façade replacing the con-
ventional building façade. Nevertheless, the interest
of more complex shapes and patterns applied to the
building façade is growing significantly in contem-
porary architecture due to technological and fabrica-
tion advancements (Rahimzadeh et al. 2013). In addi-
tion, while buildings which do feature a complex 3D
façade have been erected in some urban areas, most
of these buildings are very experimental. Therefore, a
systematic exploration of 3D façades and an investi-
gation of the benefits of these emerging façade pat-
terns are highly significant.
The aim of this paper is to suggest a methodol-
ogy to generate an enclosed 3D façade unit, which
is near optimal in terms of daylighting. The gener-
ated 3D facade unit consists of mainly three compo-
nents: a transparent window allows sunlight to pene-
trate into the indoor space; an opaque wall functions
as a shading device, and BIPV harvests solar energy.
To achieve this goal, 3D façade units are generated
in two phases: first a basic 2D shape is generated
and, next, it is expanded into a 3D façade unit. Sub-
sequently, materials are applied for daylight simula-
tion. By performing all assignments parametrically,
an evaluation of daylight simulation can take place
in order to suggest an optimal façade.
LITERATURE REVIEW
A vast number of peer-reviewed studies could be
found regarding the building façade. In this section,
we have categorized them into three main parts as
a generation, performance assessment, and the opti-
mization.
Parametric façade design
Techniques for generating building façades have
been investigated by a number of authors. One of
the most common methods is parametric modelling,
which is highly effective to automate the generation
of a large set of architectural design instances by the
combination of pre-defined design parameters (Tur-
rin et al. 2011). In designing a building façade, de-
signers and researchers are highly concerned by the
380 | eCAADe 34 - SHAPE, FORM AND GEOMETRY | Grammars and Concepts - Volume 2
placement of a rectangular fenestration on a planar
building envelope. These particular facades yield a
range of parameters such as the size, number, and
the distribution of the windows as well as the thick-
ness and the material of the wall with external shad-
ing devices (Hassaan et al. 2016; Echenagucia et al.
2015). For example, Echenagucia et al. (2015) stud-
ied the exterior wall of an open space office's plan-
ner in an urban and non-urban context at four dif-
ferent locations in Europe. In contrast to parametric
modelling, a new approach suggested by J. Wright et
al. (2014) generates façade patterns by dividing the
surface into small equal rectangular cells and deter-
mines the optimal number of windows and distribu-
tion through multi-objective optimization based on
energy performance and capital cost.
Assessment of a building façade perfor-
mance
In this study, we measured the amount of natural
light using the "Useful daylight luminance" (UDI) pre-
dictive method. The UDI method is first coined by
Nabil and Mardaljevic (2005), and divides annual day-
light illuminance at the workplace into three bins.
The first bin includes areas that receive under 100 lux,
which is not suitable and thus demands additional ar-
tificial lightning; the second bin corresponds to the
range of 100 to 2000 lux, which is suitable for work
activity; the third bin includes illuminance that ex-
ceeds 2000 lux and which results in potentially vi-
sual discomfort (Nabil and Mardaljevic 2005). This
method is more realistic than the conventional "day-
light factor approach" which only considers a single
overcast sky. When natural lighting cannot supply a
sufficient amount of light into the indoor space, ar-
tificial lighting would be required in the space. To
decrease this electrical demand, building integrated
photovoltaic panels could be attached at the outer
side of the façade for electricity harvesting. This prac-
tice is one of the sustainable features in the building
domain which could potentially cover more than half
of the daily energy needs (Berkel et al. 2014). Man-
dalaki et al. (2012) examine the thirteen most com-
Figure 1
Facade 3D unit
generation and
optimization
system diagram.
monly used types of fixed shading devices as a PV
panel for an office building. Among them, the single
inclined canopy showed the most efficiency when
comparing the area of PV with the generated elec-
tricity. Vartiainen et al. (2000) analyzed the optimal
size and orientation of a single rectangular fenestra-
tion in a fully covered PV integrated building façade
unit. A very low percentage of window area, ten to fif-
teen percent of the whole façade, proved to be ideal
when considering energy harvesting through PV and
the replacement of artificial lighting by daylight in
the specific location of Europe.
Façade Optimization
In order to achieve a better performance with re-
spect to daylighting, the building façade needs to
be optimized. There is a large pool of variables
that controls the design and the overall performance
of the building façade, which could be effectively
controlled parametrically to yield a number of alter-
SHAPE, FORM AND GEOMETRY | Grammars and Concepts - Volume 2 - eCAADe 34 | 381
natives for performance assessments. The optimal
building façade design can be achieved effectively
by means of building performance simulation cou-
pled with an evolutionary algorithm tool (Turrin et al.
2012; Evins et al. 2011) (Figure 1). Especially, the tech-
nique of multi-objective optimization is highly prac-
tical (Wang et al. 2005) in that it provides visual infor-
mation of the trade-offs between contrasting design
objectives (Mela et al. 2012). Several studies have
focused on the window-to-wall ratio (WWR) and en-
ergy performance (Goia et al. 2013; Echenagucia et
al. 2015); while other researchers optimize window
size and external shading types by the means of ge-
netic algorithms (Torres and Sakamoto 2007). Yi and
Malkawi (2009) investigated the overall building form
controlled by a hierarchical relation of geometry and
form optimization and the method was able to find
a particularly complex shape rather than the simple
boxy one.
 PROPOSED APPROACH Figure 2
The proposed approach section describes the sug- Facade geometry
gested methodology specifically, interpreting 3D generation.
façade geometry generation that followed by the op-
timization part in two sections.

Generation of the façade geometry

The façade generation process is divided into two
parts. The first part divides the façade and the second
part formalizes the unit. The workflow of second part
is explicitly explained in Figure 1. The division of the
building façade is based on a 2D tessellation (Figure
2, A and B); hereto, we only consider the equilateral
triangle, the square and the regular hexagon (Figure
2). This restriction is inspired by the regular tessel-
lations of the plane, though, obviously, other non-
regular polygons can also be considered to cover the
plane with a single element. In fact, the exploration
can be easily extended to other unit shapes by aug-
menting the number of parameters considered.
Once the façade base surface is divided, extra
nodes are parametrically added to the façade unit
(Figure 2 C). The location of these point(s) is limited
by the unit's perimeter, and lays either inside or on
the perimeter as defined by the edges and vertices.
The number of the additional points and the location
of these points is to be determined by the designer,
in order to give the designer more control over the
basic pattern of the façade. After placing the addi-
tional points on the façade unit, the additional nodes
are extruded into a direction perpendicular to the ini-
tial façade plane. The extrusion length serves as one
of the parameters for the fitness function. The façade
unit's vertices and additional points are clustered and
connected to each other by means of a Delaunay tri-
angulation (Figure 2 D). The Delaunay triangulation is
a commonly used method in the computational de-
sign domain to maximize all the angles in the gen-
erated triangles. We have chosen this method to re-
duce very thin and sharp fractured surfaces that are
not ideal for the fabrication and manufacturing pro-
cess in the façade design.

382 | eCAADe 34 - SHAPE, FORM AND GEOMETRY | Grammars and Concepts - Volume 2
 Subsequently, materials are applied to the spatial
space frame wires resulting from the Delaunay trian-
gulation (Figure 2 F).
Three main material types are being considered:
glazing, wall, and a totally opaque PV panel. For the
daylight simulation, these three materials can be cat-
egorized based on two general properties, whether
opaque (the PV panel and the wall) or transparent
(the glazing). This binary surface option serves as an-
other parameter for the optimization fitness function.
This method can be applied to any type of geom-
etry with the appropriate base 2D tessellation and a
sufficient number of additional vertices on each fa-
cade unit. In the case of a conventional façade de-
sign, with a traditional rectangular window within
the façade unit, it is sufficient to consider four addi-
tional points with a zero extrusion length.
Parametric modeling and Multi-objective
optimization
We used the Grasshopper parametric modeling tool
as a platform for the entire process, including both
design exploration and simulation.
Figure 3
Testing room is in
the middle of a 3 x 3
units. In this
picture, planar
façade with
rectangular unit is
simulated.
The 3D façade units were made parametrically; the
plug-ins Ladybug and Honeybee are adopted to
perform the dayligting simulations using Radiance
SHAPE, FORM AND GEOMETRY | Grammars and Concepts - Volume 2 - eCAADe 34 | 383
(Roudsari et al. 2013).
An adequate amount of daylighting requires an
appropriate window-to-wall ratio, while solar energy
harvesting increases when the PV surface area ex-
pands. These two characteristics are highly depen-
dent on the direction of the façade and the loca-
tion of the building. Furthermore, these two objec-
tives contrast with one another: when designers set a
goal to maximize the amount of electricity from BIPV,
it will affect the size of the window, consequently
deteriorating the daylighting potential. Therefore,
we employed evolutionary computation for multi-
objective optimization, using the Octupus plug in.
Optimization strategy and Fitness function
The objective function maximizes the area which is
correspondent to the range of UDI100-2000 in the
given space, while also maximizing the amount of an-
nual energy which obtained by BIPV on the façade
unit.
The evolutionary algorithm inputs are classified
into two main categories. The first category of inputs
are the extrusion lengths of the additional points and
the binary material selection of the triangular faces
that are generated from the Delaunay triangulation.
In our case study, four additional points make ten
triangular façade geometries in 3D space, thus ten
combinations of façade material and four extrusion
lengths, or a total of fourteen input genes for the op-
timizing algorithms.
In the multi-objective search, HypE mutation
and reduction method (Bader and Zitzler 2011) was
adopted to reduce the evaluation time of the multi-
objective optimization.
IMPLEMENTATION
In this section, the simulation environment, such as
the location and the material properties of test room
is briefly explained. Furthermore, the implementa-
tion of two case studies is presented.
 Simulation environment
A square division is selected for the basic tessella-
tion of the building façade, and we added identical
one-person office rooms to each square unit. The
building is located in Incheon, Korea, and the test-
ing façade is facing the south side. A typical office
room is selected for the daylight simulation as a case
study. The room dimension is 6m by 4.2m, depth
and width respectively, and 3.2m in height (floor to
ceiling). The reflectiveness of the materials taken
from Nabil and Mardaljevic (2005)'s experiment en-
vironment are wall 0.7, ceiling 0.8, floor 0.2 and the
window transmittance is 0.76 (Nabil and Mardalje-
vic 2005). The office model is generated in a non-
contextual environment. However, to take into ac-
count the shading from the remainder of the façade,
we applied the same façade geometry surrounding
the case room as shown in Figure 3. According to
the basic tessellation, the additional eight units are
located one at the east and on at the west side of
the case room and three positioned above as well
as three at the lower level. Consequently, nine units
having the identical façade geometry are generated
3 x 3 where the center of the middle level is present-
ing the testing room.
1012.5 kWh annual electricity was generated from
the opaque façade area.
For the second case study, we created a 3D
façade through extrusion of the optimization strat-
egy and the fitness function. The population size was
set to 50 and 25 generations have been conducted
for the optimization. General settings of parameters
are reported in Table 1.
Table 1
Multi objective
optimization
settings.
A constraint was introduced with each evaluation
and generation of the façade unit that surrounding
unit shapes are identical to the case room façade de-
sign. And extrusion lengths are limited to maximum
2.0 m. Other factors not considered is the façade con-
struction, e.t. size and type of mullions.
RESULTS

Case study Figure 4

In order to compare the daylighting ability results Pareto front
from conventional façade with our 3D façade, we ex- solutions.
ecuted two sets of case studies. For the first case
study, we simulated daylighting of a conventional
façade, and for the second case study, simulated day-
lighting of our 3D façade created through optimiza-
tion.
In the first case study, four additional points are
placed on the case room façade for the basic 2D tes-
sellation unit. These four points are co-planar to the
flat façade placed to create a rectangular window
in the horizontal center with the dimension of 2.8
meters by 2 meters, and 1.2 meters above the test-
ing floor (See Figure 3). In this simulation environ-
ment, the façade had 46.5% window-to-wall ratio, In Figure 4, the dark points connected by a black poly-
UDI (100-2000) covered 74.3% of the floor area, and line represent the 50 pareto-front solutions or the

384 | eCAADe 34 - SHAPE, FORM AND GEOMETRY | Grammars and Concepts - Volume 2
 optimized design alternatives, after 25 generations. concave part. It is likely that the combination of con-
The total surface area of each non-dominant designs cave and convex surfaces offers better solutions alter-
were 1.6 to 1.9 times larger than the initial office unit nating opaque and transparent surfaces.
area (17.64 m2). Thus no completely planar solution
Figure 5 has been suggested from these particular case. To il-
Achieved UDI. lustrate how each design alternatives performed, we
chose three cases for example, two at the extremes
(points A and C), and one in the middle (point B) as in-
dicated in Figure 7. The design alternative at point A
is the one that provides most daylight distribution in
UDI100-2000, where type C generates the most elec-
tricity using its BIPV surfaces annually. More specif-
ically, the design alternative at point A performed
the best in terms of daylighting covering 82.7 % of
the UDI within the range 100-2000 lux with the high-
est windows-to-surface ratio (40.54%). And the de-
sign alternative at point C had the largest total sur-
face area and electricity generation annually (2199.21
kWh).
Figure 6 In order to better understand the relationships
BIPV efficiency. between the variables, we graphed the relationship
between the window-to-opaque area and the aver-
age UDI achieved (Figure 5), and the relationship be-
tween the total electricity generation and the BIPV
(opaque area)-to-total surface ratio (Figure 6). As can
be seen, the result shows that when the window-
to-surface area increases the achieved UDI also in-
creases (Figure 5) even though having a concave
topology in most of the pareto-fronts.
Overall electricity generated in per meter square
does not seem to increase as the PV surface increase
(Figure 6). This mainly because of the shading impact
from surrounding unit geometries. DISCUSSION AND CONCLUSION
A finding that is worth mentioning is the materi- This paper suggests a novel approach of designing
als applied for the Delaunay space frames. The upper a 3D shaped building façade that replaces conven-
space frames mostly remained opaque as indicated tional design methods. The suggested method adds
in color black in Figure 8 which includes the repre- extra points on the rectangular building façade units
sentative design alternatives at points A, B, and C. A and connects them with the corner of the vertices of
possible reason for this frame material might be be- a facade unit. The number of additional points and
cause the top two additional points' locations are at the location of those are predetermined by the ar-
the ceiling level where the ceiling blocks the daylight chitect as designer and decision maker. Our method
significantly. Furthermore, the overall shape of the suggests the optimal extrusion lengths at the given
façade at the pareto-front is most often includes a points and the combination of materials for the sur-
faces (space frames) which are generated from the

SHAPE, FORM AND GEOMETRY | Grammars and Concepts - Volume 2 - eCAADe 34 | 385
 Figure 7
Pareto front
solutions of the
most UDI achieved
A, The most
electricity
generated C, and in
between B.

386 | eCAADe 34 - SHAPE, FORM AND GEOMETRY | Grammars and Concepts - Volume 2
Delaunay triangulation by means of multi-objective
optimization.
When the optimization method we created was
applied, we found that our pareto-front solutions
have significantly higher value in both daylighting
and electricity generation in comparison to the con-
ventional flat façade where the achieved UDI in-
creased by 9% in the best case of daylighting per-
formance and a growth in electricity harvesting of as
much as 50% in the best case electricity generation.
The advantage of implementing of multi-
objective optimization in this particular case were
twofold. First the number of unique solutions yielded
in pareto-front. Even though the performance is al-
most identical, significantly different design alterna-
tives could provide important information to design-
ers in early design stage. The second adventage is,
the comprehensive feedback on the performance of
the optimal solutions. The Designers can achieve
their set goals for their façade designs by consider-
ing the best trade-offs between the achieved UDI
and the generated electricity not completely disre-
garding one of the trade-offs.
Furthermore, the suggested method shows
great flexibility and compatibility of generating any
façade design with the same process. The method
could be used effectively in the early stages of a de-
sign of an office building and assess the daylighting
performance and renewable energy generation pre-
diction in a given location.
3D shaped building façades showed an advan-
tage of energy harvesting and daylighting perfor-
mances, however energy consumption was not ad-
dressed in this research. Future works will extend the
scope of this study by adding energy efficiency pa-
rameters into the façade generation method.
REFERENCES
Bader, J and Zitzler, E 2011, 'HypE: an algorithm for fast
hypervolume-based many-objective optimization.',
Evolutionary computation, 19(1), pp. 45-76
Van Berkel, T, Minderhoud, T, Piber, A and Gijzen, G
2014 'DESIGN INNOVATION FROM PV-MODULE TO
BUILDING ENVELOPE: ARCHITECTURAL LAYERING
SHAPE, FORM AND GEOMETRY | Grammars and Concepts - Volume 2 - eCAADe 34 | 387
AND NON APPARENT REPETITION', 29th European
Photovoltaic Solar Energy Conference and Exhibition,
pp. 366-372
David, M, Donn, M, Garde, F and Lenoir, A 2011, 'Assess-
ment of the thermal and visual efficiency of solar
shades', Building and Environment, 46(7), pp. 1489-
1496
Echenagucia, TM, Capozzoli, A, Cascone, Y and Sassone,
M 2015, 'The early design stage of a building enve-
lope: Multi-objective search through heating, cool-
ing and lighting energy performance analysis', Ap-
plied Energy, 154, pp. 577-591
Evins, R, Pointer, P and Vaidyanathan, R 2011 'Multi-
objective optimisation of the configuration and
control of a double-skin facade', Proceedings of Build-
ing Simulation 2011: 12th Conference of Interna-
tional Building Performance Simulation Association,
pp. 1343-1350
Goia, F, Haase, M and Perino, M 2013, 'Optimizing the
configuration of a facade module for office buildings
by means of integrated thermal and lighting simu-
lations in a total energy perspective', Applied Energy,
108, pp. 515-527
Gonzalez, J and Fiorito, F 2015, 'Daylight Design of Office
Buildings: Optimisation of External Solar Shadings
by Using Combined Simulation Methods', Buildings,
5(2), pp. 560-580
Hassaan, A, Mahmoud, A and Elghazi, Y 2016,
'Parametric-based designs for kinetic facades to op-
timize daylight performance : Comparing rotation
and translation kinetic motion for hexagonal facade
patterns', Solar Energy, 126, pp. 111-127
Krarti, M, Erickson, PM and Hillman, TC 2005, 'A simplified
method to estimate energy savings of artificial light-
ing use from daylighting', Building and Environment,
40(6), pp. 747-754
Li, D.H. and Tsang, E.K. 2008, 'An analysis of daylight-
ing performance for office buildings in Hong Kong',
Building and Environment, 43(9), pp. 1446-1458
Mandalaki, M, Zervas, K, Tsoutsos, T and Vazakas, A
2012, 'Assessment of fixed shading devices with in-
tegrated PV for efficient energy use', Solar Energy,
86(9), pp. 2561-2575
Mela, K, Tiainen, T and Heinisuo, M 2012, 'Comparative
study of multiple criteria decision making methods
for building design', Advanced Engineering Informat-
ics, 26(4), pp. 716-726
Nabil, A and Mardaljevic, J 2005, 'Useful daylight illu-
minance: a new paradigm for assessing daylight in
buildings', Lighting Research and Technology, 37(1),
pp. 41-59
 Peippo, K, Lund, P and Vartiainen, E 1999, 'Multivariate
optimization of design trade-offs for solar low en-
ergy buildings', Energy and Buildings, 29(2), pp. 189-
205
Rahimzadeh, S.D, Garcia-Hansen, V, Drogemuller, R and
Isoardi, G 2013 'Parametric Modelling for the effi-
cient daylight strategies with complex geometries',
The 47th International Conference of the Architectural
Science Association
Roudsari, M.S., Pak, M. and Gill, G 2013 'Ladybug: a
parametric environmental plugin for grasshopper to
help designers create an environmentally-conscious
design', Proceedings of the 13th International IBPSA
Conference, Lyon
Sheikh, ME and Gerber, DDJ 2011 'Building Skin Intel-
ligence A PARAMETRIC AND ALGORITHMIC TOOL
FOR DAYLIGHTING PERFORMANCE DESIGN INTE-
GRATION', Proceedings of the annual conference of the
Association of Computer Aided Design in Architecture
ACADIA, pp. 170-177
Torres, S.L. and Sakamoto, Y. 2007 'Facade design op-
timization for daylight with a simple genetic algo-
rithm', Proceedings of Building Simulation, Beijin
Turrin, M, Von Buelow, P, Kilian, A and Stouffs, R 2012,
'Performative skins for passive climatic comfort: A
parametric design process', Automation in Construc-
tion, 22, pp. 36-50
Turrin, M, Von Buelow, P and Stouffs, R 2011, 'Design ex-
plorations of performance driven geometry in archi-
tectural design using parametric modeling and ge-
netic algorithms', Advanced Engineering Informatics,
25(4), pp. 656-675
Wright, JA, Brownlee, A, Mourshed, MM and Wang, M
2014, 'Multi-objective optimization of cellular fen-
estration by an evolutionary algorithm', Journal of
Building Performance Simulation, 7(1), pp. 33-51

388 | eCAADe 34 - SHAPE, FORM AND GEOMETRY | Grammars and Concepts - Volume 2
SHAPE, FORM AND GEOMETRY |
Applications
Geometric Versatility of Abeille Vault
A Stereotomic Topological Interlocking Assembly

Irina Miodragovic Vella1 , Toni Kotnik2

1
University of Malta 2 Aalto University
1
irina.miodragovic-vella@um.edu.mt 2 toni.kotnik@aalto.fi

The Abeille flat vault, patented at the end of 17th century, consists of identical
ashlars arranged in a woven-like pattern that generates their interlocking mutual
support. In recent years, the availability of digital design and fabrication tools
has caused new interest in the Abeille vault. Several studies investigate the
interlocking principles through their application onto non-planar assemblies.
This paper is a is a more systematic exploration into the underlying geometric
interdependencies behind interlocking principles. It approaches the Abeille vault
as a topological interlocking assembly (TIA), an assembly where basic identical
elements of a special shape are arranged in such a way that the whole structure
can be held together by boundary constraint, while locally the elements are kept
in place by kinematic constrains imposed through the shape and mutual
arrangement of the elements. The paper looks at the full potential of the Abeille
vault application and studies the relation between the surface geometry and TIA

Keywords: stereotomy, Abeille vault, topological interlocking

Introduction of complex geometries, as well as change in cultural

Stereotomy is a discipline that accumulates theoret-
ical and practical knowledge of stone cutting and
stone construction. Since, as Jacques Heyman ex-
plains: "the key to understanding of masonry is to be
found in a correct understanding of geometry" (1995,
154), stereotomy is based on essential geometric re-
lationships embedded within masonry. It started
in the Gothic period, had its peak during the En-
lightenment in the examples such as the undulating
vaulted space of the Arles City Hall vestibule by Jules
Hardouin-Mansart, completed in 1637. The main rea-
sons for the later decline of stereotomic techniques
were the introduction of steel and concrete that were
less labour intensive and did not require knowledge
mind set where the rationalist tectonics did not ap-
prove of "deformed, showy tricks" (Evans 2000).
Stereotomic architecture is primarily based on
vaulted systems to cover large spans. Such material
systems ideally are shaped as an arch, as it was dis-
covered by Robert Hooke in 1675. Hooke described
the relationship between a hanging chain and the in-
ner force flow in an arch, as "ut pendet continuum
flexile sic stabit contiguum rigidum inversum" (Hey-
man 1998). The flat vault or voûte plate (Figure 1) in-
vented by Josph Abeille in 1699, thus, has to be seen
as an exceptional stereotomic solution that provides
"all in one a ceiling for the lower storey, and a pave-
ment for the upper storey" (Fleury 2009). In an Abeille

SHAPE, FORM AND GEOMETRY | Applications - Volume 2 - eCAADe 34 | 391

 vault the shape of each ashlar or voussoir is identi-
cal: a polyhedron with axial sections in the shape of
an isosceles trapezium. The unique ashlar geometry
and the rotation of neighbouring ashlars by ninety
degrees allows for a simple method of their mutual
arrangement: each ashlar is carried by two neighbor-
ing ones, and at the same provides support for two
other ashlars. "With this arrangement, each voussoir
is carried on two others through its protruding cuts,
and at the same time carries two others on its sloped
cuts [...], this being reciprocal in all the vault's area, it
supports itself at level" (Fleury 2009).
The woven-like organizational pattern of the Abeille
vault resembles the organizational pattern of recipro-
cal timber frame structures of Villard de Honnecout
(Lassus 1858) or Sebastiano Serlio (Hart and Hicks
1996). Both structures overcome the problem of find-
ing a constructive solution to cover a space with a flat
floor consisting of small discrete elements.
Structurally, however, the Abeille vault behaves
differently than a reciprocal frame structure, namely
as topological interlocking assembly (TIA), a notion
introduced by Yuri Estrin and Arcady Dyskin in mate-
rial science in 2001 (Glinckman 1984 and Estrin et al.
2011). TIA are defined by basic identical elements of
a special shape which are arranged in such a way that
the whole structure can be held together by bound-
ary constraint, while locally the elements are kept in
place by kinematic constrains imposed through the
shape and mutual arrangement of the elements. TIA
are scale and material independent, they are a con-
392 | eCAADe 34 - SHAPE, FORM AND GEOMETRY | Applications - Volume 2
figurational condition based on geometry. Research
into TIA is currently focused on planar configuration
comparable to the Abeille vault, looking primarily
into configurational properties and aspects like the
bearing capacity, stiffness, deformation, indentation
or sound absorption (Carlesso et al. 2012, Dyskin et
al. 2003a, Dyskin et al. 2012, and Dyskin et al. 2003b).
The Abeille vault, thus, can be seen as a historic exam-
ple for TIA, and understood not so much as a vault,
but rather as network of interacting elements.
Historically, the high amount of horizontal thrust
within the Abeille vault was resolved by boundary
constraints like buttresses or massive walls (Brocato
2012). Due to the effort needed to support Abeille Figure 1
vaults it is no surprise that they were rarely used Abeille Vault.
throughout history despite being "one of the most (Gallon 1735).
interesting technical and stylistic investigations into
art of stonecutting applied to building construction"
(Cities of Stone 2006).
In recent years, the availability of digital de-
sign and fabrication tools has caused new interest in
stereotomy, its geometric rules, and the Abeille vault
as specific case study (Etlin et al. 2008). The topologi-
cal nature of the interlocking principles of the Abeille
vault has been used by Giuseppe Fallacara to apply
the construction principle onto non-planar vaulted
structures and domes with focus on the woven aes-
thetics expressed in the construction principle. Oliver
Tessmann (2013) has started to investigate the po-
tential of variation within the ashlar itself and its gen-
erative effect on the shape of the assembly, as well as
the potential for architectural differentiation of sur-
face qualities. From both investigations it is appar-
ent that the non-planar assemblies can be achieved
through rotation of 'protruding' and 'sloped' ashlar
faces. The ashlar configuration is directly dependant
on the surface curvature, specifically the surface cur-
vature in an underlying grid system directions.
This paper is a more systematic exploration of the
full potential of the application of the Abeille vault
to non-planar assemblies and studies the relation be-
tween the surface geometry and TIA. A geometric
construction method is defined based on underlying
Figure 2
Abeille-based TIA
geometric
construction
method.
stereotomic rules and interdependencies of the orig-
inal Abeille vault.
Abeille-based TIA
The starting point of the geometrical construction
method is formulation of a point grid on a given sur-
face. For each grid field defined by vertices ABCD four
points at vertices mid-span are derived: Pt0, Pt1, Pt2
and Pt2. Points Pt0 and Pt2, and Pt1 and Pt3 laying
at opposite sides of the field define vectors in U and
V direction respectively. Thus, each mid-point Pt that
does not lie at the surface boundary defines a pair of
vectors in opposite directions, Vu+ and Vu-, or Vv+
and Vv-.
A plane Pl is defined at each mid-point Pt based
on the vector between the mid-point and adjacent
vertex, and the sum of the vector pairs at that mid-
point, referred to as Vδu or Vδv (Figure 2a). The mag-
nitude of vectors Vδu and Vδv reflects the surface cur-
vature in U and V directions respectively at the given
mid-point. Each Pl plane is rotated around the first
axes for an angle α. The rotation direction is opposite
to the rotation direction of the adjacent plane (Fig-
ure 2b). The intersection of the four rotated planes
defines a tetrahedron (Figure 2c).
Finally, pairs of trimming planes are defined.
SHAPE, FORM AND GEOMETRY | Applications - Volume 2 - eCAADe 34 | 393
Each trimming plane is an offset of the plane defined
by vectors Vu+ and Vv+ at the centroid Cn of each
field (Figure 2d).
The Abeille-based TIA geometric construction
method is, thus, based on the four main variables:
the point grid distribution of the initial surface, curva-
ture of the initial surface in U and V directions, angle
of plane rotation, and the position of the trimming
planes. Through establishing the interdependences
between the variables the construction method is
further parametrized, and the interlocking properties
assessed.
First geometrical interdependency established is
between the curvature of the initial surface in U and
V directions , and the plane rotation angles. The val-
ues of these variables are inversely proportional: the
higher the curvature, the lower is the planes rotation
angle, and vice versa. In this way the elements within
a planar assembly are keyed in, while elements within
an area with a high curvature degree do not have
awkward configurations (Figure 3).
Second geometrical interdependency estab-
lished is between the curvature of the initial sur-
face in the U and Vi directions, and the point grid
density, and is directly proportional. Through this
parametrization high curvature defines high grid
 point density and smaller ashlars that in turn ap-
proximate closer the steep curvature (Figure 3).
When the initial surface is non-planar and/or
point grid distribution non-rectangular, the non-
adjacent elements with the same orientation inter-
sect i.e. only in the case of the TIA of regular tetrahe-
dra that is based on a planar surface with rectangular
point grid, like the original Abeille vault, no tetrahe-
dra intersect. The trimming planes modify and rectify
elements' configuration by removing elements inter-
section for every other situation. The domain of pos-
sible trimming plane positions is, thus, determined
by both the curvature of the initial surface and its
point grid distribution.
The Abeille-based TIA geometric construction
method can generate a vast number of different as-
sembly configurations. Due to the geometric defi-
nition of the individual ashlars an Abeille-based TIA
is not limited to vaulted shapes but can be defined
along more general formed surfaces. From the con-
struction sequence shown above, it is clear that the
interlocking mechanism is a system based not so
394 | eCAADe 34 - SHAPE, FORM AND GEOMETRY | Applications - Volume 2
much on precise geometry of the intersection sur-
face of two adjacent ashlars but rather on the ori-
entation alternation of the intersection surface along
the ashlar. It is this alternation that directs the force Figure 3
flow within the overall system and results in the re- Curvature, point
ciprocal behaviour and the interlocking of the ash- grid distribution
lars. Abeille vault, thus, represents a basic example and plane rotation
of this system that conceptually lends itself for further interdependancy.
complexification. The complexification of the system
can be done by either increasing the number of poly-
gon sides that make up the point grid, or increasing
the number of alternations per ashlar face. In this
way the Abeille-based TIA defines assemblies start-
ing from TIA of tetrahedra, towards TIA of more com-
plex Platonic solids, and ultimately free form geome-
tries (Figure 4). Figure 4
Diagram of
Hexagonal Grid, Truchet Vaults and Osteo- Abeille-based TIA
morphic Blocks complexification .
Abeille-type configurations, are easily applied to
other gridded systems, like hexagonal grids, and
transformed into more fluid interlocking systems, like
Truchet vaults or osteomorphic blocks. The Abeille
vault's concept of alternating plane rotations applied
on a hexagonal grid generates TIA of cuboids (Fig-
ures 5a & 5b). Further, when the pairs of trimming
planes are introduced the ashlar configuration be-
comes that of irregular octahedron (Figure 5c).
Truchet flat vault is based on the same under-
lying logic like the Abeille vault, with the difference
that the alternating plane rotation is morphed into
alternating undulation of the ashlar intersecting sur-
face (Figure 6). The grid coincides with the vault ex-
trados, and defines the ashlar extrados as squares,
while the intrados are made up of alternating semi-
circular protrusion and indentations. The Truchet
was developed soon after the Abeille vault as its im-
provement, since its interlocking ashlars leave no
void on either side of the surface (Fallacara 2007). Al-
though the solution was considered "truly ingenious"
at the time it was developed the alternation of con-
cave and convex surfaces was found to difficult to ac-
tually execute (Etlin et al. 2008) until the introduction
Figure 5
The Abeille-type
configuration
applied on a
hexagonal grid.

of CAD/CAM processes. in a mastery way in his planar tessellations (Figure 9)

The osteomorphic block is based on the same (Haak 1976).
concept of Truchet's alternating concave and con-
Figure 6 vex surfaces, with the difference that the alternation
Truchet Vault. is developed along the same ashlar face (Figure 7).
The interlocking is achieved only along one axes and
since the system was invented for a wall and column
construction, ostemorphic block is not based on a
grid, but a running bond (Dyskin et al. 2003c).
Further complexifications investigated in the in-
troduction of perforations within the Abeille-type
system. Perforations can be achieved by manipu-
lating the tetrahedral ashlar configuration. Tessman
subtracts volumes form the tetrahedral ashlar to in-
vestigate the variations of the intersecting surface
Figure 7 contact zones. The resulting cropped ashlars define
TIA of ostemorphic perforations while maintaining the interlocking prin-
blocks. ciple (Tessman 2013). Perforations can be achieved
also through variation of the trimming plane posi-
tions, specifically when both trimming planes are
placed either below or above the initial grid (Fig-
ure 8). To ensure the interlocking within the system,
further parametrization is established between the
ashlar thickness and distance from the grid between
neighbouring pairs trimming planes.
Looking at the examples shown, it can be con-
cluded that for defining the interlocking ashlar ge-
ometry, it is enough to consider the alternation only
along the parameter of the grid field i.e. an 'inward'
rotation has its identical, corresponding 'outward' ro-
tation on the other side of the same ashlar. This is
the same principle used in two dimensions by Escher

SHAPE, FORM AND GEOMETRY | Applications - Volume 2 - eCAADe 34 | 395

 Conclusion
The discussion shows that the Abeille-like configu- Figure 8
ration can be seen as generic example of TIA. Other Variation in
examples like the Truchet vault or osteomorphic trimming planes
blocks and their parametric variation can be under- position in a
stood as simple variation of the principles that de- Abeille-based TIA.
fine the Abeille-vault. Such a unifying approach al-
lows a more systematic study of the structural be-
haviour of TIA and the limitation of the material sys-
tem with respect to the geometry of the underlying
surface. TIA are characterized by a systemic trans-
fer of loads based on the interaction of neighbour-
ing ashlars. This way a cyclic distribution of loads
appears within the material system that resembles
the non-hierarchical load pattern in reciprocal frame
structures (Kohlhammer et al. 2011). A more detailed
Figure 9
analysis is required, however, to understand in more
Escher's tessellation
detail the influence of curvature on the structural be-
based on
haviour of TIA.
hexagonal grid .
The local interaction of the ashlars is not only
providing a more even distribution of loads within
the structure itself but at the same time providing a
local stabilization of the configuration that has pos-
itive effects for the assembly. First test that have
been conducted with an Abeille-like pavilion struc-
ture show that the interlocking of the ashlars enables
a significant reduction of falsework that is typically
needed for the construction of masonry vaults. This
means that TIA have the potential to increase again
the interest in vaulting structures that is in a building
typology that had not been used very much in devel-
oped countries since the late 1970s due to the high
amount of labour that is needed for the preparation
of falsework.
This study on Abeille-like configurations there-
fore demonstrates clearly how the combination of
digital design tools and digital fabrication enables
the reanimation of historic construction techniques
for the benefit of contemporary architecture. In ad-
dition, the geometric nature of the Abeille-like con-
struction allows for a more holistic understanding of
the interplay of geometric relationship within the as-
sembly, specifically surface curvature and the config-

396 | eCAADe 34 - SHAPE, FORM AND GEOMETRY | Applications - Volume 2

uration of individual elements, and enables the trans-
fer of structural principles between different con-
struction typologies which in turn helps to delimit
system boundaries and opens up the possibility of
a more fluid understanding of structures in architec-
ture.
REFERENCES
Brocato, M and Mondardini, L 2012, 'A new type of stone
dome based on Abeille’s bond', International Journal
of Solids and Structures, 49(13), pp. 1786-1801
Carlesso, M, Molotnikov, A, Krause, T, Tushtev, K, Kroll,
S, Rezwan, K and Estrin, Y 2012, 'Enhancement of
sound absorption properties using topologically in-
terlocked elements', Scripta Materialia, 66(7), pp.
483-486
Dyskin, A, Estrin, Y, Kanel-Belov, A and Pasternak, E 2001,
'A new concept in design of materials and structures:
assemblies of interlocked tetrahedron-shaped ele-
ments', Scripta Materialia, 44(12), pp. 2689-2694
Dyskin, A, Estrin, Y, Kanel-Belov, A and Pasternak, E 2003,
'Topological interlocking of platonic solids: a way to
new materials and structures', Philosophical maga-
zine letters, 83(3), pp. 197-203
Dyskin, AV, Estrin, Y, Pasternak, E, Khor, HC and Kanel
Belov, AJ 2003, 'Fracture Resistant Structures Based
on Topological Interlocking with Non�planar Con-
tacts', Advanced Engineering Materials, 5(3), pp. 116-
119
Dyskin, AV, Pasternak, E and Estrin, Y 2012, 'Mortarless
structures based on topological interlocking', Fron-
tiers of Structural and Civil Engineering, pp. 1-10
Estrin, Y, Dyskin, A and Pasternak, E 2011, 'Topological
interlocking as a material design concept', Materials
Science and Engineering: C, 31(6), pp. 1189-1194
Etlin, R, Fallacara, G and Tamborero, L 2008, Stereotomy,
Acrobatic Stone Vaulting Past and Future, Aracne ed-
itrice S.r.l., Rome
Evans, R 2000, The Projective Cast, The MIT Press, Cam-
brige, Massachusetts; London, England
Fallacara, G 2007, Towards a Stereotomic Design, Aracne
editrice S.r.l., Roma
Fallacara, G 2009 'Toward a Stereotomic Design: Exper-
imental Construction and Didactic Experiences', no
source given
Fleury, F 2009 'Evaluation of the Perpendicular Flat Vault
Inventor’s Intuitions through. Large Scale Instru-
mented Testing', Proceedings of the Third Interna-
tional Congress on Construction History, Cottbus
SHAPE, FORM AND GEOMETRY | Applications - Volume 2 - eCAADe 34 | 397
Gallon, J. 1735-1777, Machines et inventions approuvées
par l'Academie Royale des Sciences depuis son étab-
lissement jusqu'à présent, Académie des Sciences,
Paris
Glickman, M 1984 'The G-block system of vertically inter-
locking paving', Proceedings of the 2nd International
Conference on Concrete Block Paving, Delft, Nether-
lands, April, pp. 10-12
Guerrieri, CD (eds) 2006, Cities of Stone, Fondazione La
Biennale, venezia
Haak, S 1976, 'Transformation Geometry and the Art-
work of M.C. Escher', Mathematics Teacher(69), pp.
647-652
Hart, V, Hicks, P and Serlio, S 1996, Sebastiano Serlio on
architecture, Yale University Press
Heyman, J 1995, The Stone Skeleton, Strucural Engineering
of Masonry Architecture, Cambrige University Press,
Cambrige
Heyman, J 1998, Structural analysis: a historical approach,
Cambridge University Press
Kanel-Belov, A, Dyskin, A, Estrin, Y, Pasternak, E and
Ivanov-Pogodaev, IA 2010, 'Interlocking of convex
polyhedra: towards a geometric theory of frag-
mented solids', Moscow Mathematical Journal, 10(2),
pp. 337-342
Kohlhammer, T and Kotnik, T 2011, 'Systemic behaviour
of plane reciprocal frame structures', Structural Engi-
neering International, 21(1), pp. 80-86
Lassus, JBA and Darcel, A 1858, Album de Villard de Hon-
necourt, architecte du XIIIe siècle: manuscrit publié en
fac-simile, annoté, précédé de considérations sur la re-
naissance de l'art français au XIXe siècle et suivi d'un
glossaire, Imprimerie impériale
Tessmann, O 2013, Interlocking Manifold Kinematically
Constrained Multi-material Systems, Springer
Origami Explorations
A Generative Parametric Technique For kinetic cellular façade to optimize
Daylight Performance

Yomna Saad ElGhazi1 , Ayman Hassaan Ahmed Mahmoud2

1
Arab Academy For Science, Technology & Maritime Transport, Cairo, Egypt
2
Department of Architecture Engineering, Faculty of Engineering, Cairo Univer-
sity, Giza, Egypt
1
yomnaelghazi@yahoo.com 2 arch.ayman@yahoo.co.uk

At present the kinetics is basic, but there is no doubt that research into the field of
responsive building facades will continue, to find more sophisticated design and
technical solutions. This research explores the possibilities of kinetic composition
afforded by Origami different techniques using squared module. Origami and
paper pleating techniques are one of the conceptual design approaches from
which Kinetics can be developed. The paper examines the possibilities of
different arrangements of folded modules to create environmental efficient kinetic
morphed skins. The paper aims to achieve different Kinetic origami-based
shading screens categorized by series of parameters to provide appropriate
daylighting. The main tested parameters are the form of Origami folds, the
module size and motion scenarios. Ten origami cases where explored first using
conceptual folded paper maquette modules, then parametrically modelled and
simulated at four times of the year, 21st of March, June, September and
December, taken every hour of the working day.

Keywords: Kinetic cellular façade, Origami, Parametric modelling, Parametric

simulations, Daylighting performance.

INTRODUCTION ingly, some origami and paper pleating techniques

Origami and paper pleating techniques are one of
the conceptual design approaches from which Kinet-
ics can be developed (Adrover 2015). Applications
of kinetic origami and folding concepts can be used
repetitively in modular large-scale facades. There-
fore, the folding technique concept is not originally
intelligent, but it represents the ability of controlling
the structure by moving part or all of it. Accord-
could be demonstrated into kinetic cellular facade.
Origami screen, which act as Kinetic solar screens,
present efficient solution that can be used externally
to minimize heat gain while providing appropriate
daylighting (Mahmoud and ElGhazi 2016) . Origami
transformation was analysed in terms of skin compo-
nents, defined as the transformation of arrangement
and geometry of individual kinetic parts and the mo-

SHAPE, FORM AND GEOMETRY | Applications - Volume 2 - eCAADe 34 | 399

 tion of control means between them. Origami of-
fers a finite set of paper-folding techniques that can
be cataloged and tested with parametric modeling
software. For this work, Rhinoceros and Grasshopper
have been chosen as a software platform to gener-
ate a parametric folding tool focusing on single sur-
face folding, particularly where surfaces can trans-
form from one configuration to another while retain-
ing their planarity (Dave and Brian 2011).
The paper focuses on optimizing the divisions
and number of folds and its movements. The study is
a way to achieve different deployable façade shading
systems categorized by a series of parameters that
describe the strengths and weaknesses of each tes-
sellation. Through the kinetic behaviour of Origami
geometries the research compares simple folding di-
agrams with the purpose to understand the poten-
tial of kinetic patterns' morphology for application
in adaptive facades. The possibilities of using a re-
sponsive folding technique to develop a kinetic sur-
face that can change its configuration are here ex-
amined through the variation of parameters that in-
fluence kinematics' form. Surface manipulation tool
that can transform the arrangement of folding pla-
nar surfaces, studied a family of Squared cell folding
geometries to provide daylighting using parametric
modelling. These early kinetic design explorations
introduce the fourth dimension of Time as a key el-
ement of the process of transformation.
BACKGROUND ON ORIGAMI
400 | eCAADe 34 - SHAPE, FORM AND GEOMETRY | Applications - Volume 2
Many origami explorations of building skins have
been tested in previous research, (Lee and Leou-
nis 2011) presented a surface manipulation tool that
can transform the arrangement of folding planar sur-
faces, while (Crawford 2010) studied a family of fold-
ing geometry to provide ventilation using parametric
modelling. The research team (Elghazi et al. 2014) (El-
ghazi et al. 2015), selected a 3D kaleidocycle origami
module to design an environmentally responsive ki-
netic screen. Origami Screens act as Kinetic solar
screens which have a great environmental perfor-
mance energy efficiency (Moloney 2011) (Mahmoud
and Elghazi 2016). Origami screens present efficient
solution that can be used externally to minimize heat
gain while providing appropriate daylighting (Elg-
hazi et al. 2015). Some projects apply the Origami
folded façade concept as static treatment to screen
the glazing, while others merged it with intelligent
kinetic systems. For example, a static folded alu-
minium façade shelters a biomedical research facility
at a hospital of Navarra, Spain.The building's key ar-
chitectural feature is the exterior louvers constructed
of 3mm-thick perforated aluminum panels arranged
in a tessellated origami pattern. Integrating kinetic
systems, the 25-story twin office towers in the United
Arab Emirates merges the origami façade concept
with the intelligent sensor-controlled screen that re-
sponds dynamically to the sun, folding to shade or
expose the building as shown in Figure2. Another
example was recently completed Q1 headquarters
building in Essen Germany is shaded by 3,150 kinetic
"feathers" that open and close based on user input
and sensor data as shown in Figure 3. Similarly, The Figure 1
Syddansk Universitet communications and design A selection of
building in Denmark features a climate-responsive ki- folded Miura
netic façade using similar configuration as shown in geometries.
Figure 4.
Figure 2
Al Bahar Towers,
The folding shading
system, opens and
closes according to
sun's position.
Figure 3
Q1, ThyssenKrupp
Quarter in Essen.
Figure 4
The Syddansk
Universitet
communications
and design building
in Denmark
features a
climate-responsive
kinetic facade.
PARAMETRIC MODELING
Parametric design refers to a practice of digitally
modeling a series of design variants whose relation-
ships to each other are defined through one or sev-
eral mathematical relationships (parameters). Para-
metric models can automatically adjust geometric
models in response to dynamic real-time data (Davis
et al. 2011), to produce a flexible forms that can be
animated (Leach 2009) . It is considered an essential
tool during this research due to the complexity of the
folding operations which was made to design a flexi-
ble form that when collapsed would be able to open
diaphragm enabling daylight to enter the space and
when expanded would form a planar flat surface. Ini-
tially, the parametric model began with a square. The
folds were composed by finding relations and con-
nections between the vertices, mid points of edges
of the square to its center point. The relations of lines
and points changes in order to open and close the
diaphragm. The main parameters of each Origami
unit form were the motion of the folds. Different
combinations of rotations and translations led to dif-
SHAPE, FORM AND GEOMETRY | Applications - Volume 2 - eCAADe 34 | 401
ferent characteristics of the diaphragm. The para-
metric model allowed for an evolving understand-
ing of the how each fold influenced the properties
of the diaphragm. When the geometry was digitally
unfolded and examined over the range of positions,
from being fully closed to fully open, it served as a
means for better understanding the geometry of the
diaphragm. It could also be used to unfold the di-
aphragm into a single form for the further fabrication
process.
DIVIDING A SQUARE
However, with origami, it is surprisingly easy to
achieve different divisions of a squared module. One
of the ways of using geometric shapes is by splitting
a shape up into simpler shapes or splitting it in a sys-
tematic way to make new shapes. At first, Simple divi-
sions or less complex pattern can be used to actually
divide the square horizontally or vertically, creating a
grid.
Every step adds as many lines, cease or tessella-
tion as possible which do not depend on each other
but previously existing lines. This means that you
can create different 3D configuration from the same
grid divisions. Pattern sequences are used to split the
square into number of triangular parts. It is accom-
plished by connecting lines between intersections,
corners of the square, or from the center of each side
to the two corners of the opposite side as shown in
Figure 5.
METHODOLOGY
To integrate all characteristics of architectural design,
a comprehensive methodology is required to con-
sider the aesthetic qualities of the façade and the
performance behavior of the skin. This paper ex-
plores origami-based Kinetic façades design to con-
trol daylight uniformity through three phases. The
first phase explores kinetic behaviour of origami ge-
ometries through paper maquettes to compare dif-
ferent simple folding configurations to understand
the movement at global scale. The second phase use
Parametric modelling tools to model different kinetic
 origami modules through the variation of parame-
ters that influence kinematics' form. The third phase
compare kinetic origami cellular façades' configura-
tions using parametric simulation in optimizing Day-
light Performance.
just in small conceptual models, but also in surpris-
ingly large-scale ones. Today, applications of deploy-
able origami and paper fold concepts range from
aerospace applications to maps, textiles, kinetic art
and sculpture.

The First Phase Figure 5

Paper modeling of transformable surfaces is helpful Square divisions.
to be able to visualize surface movement. Concepts
of folded modules are explored through paper ma-
quettes, although they may not be worked up into
fully resolved structures. Paper-folding is a technique
of alternating mountain and valley folds in an ar-
rangement that allows movement in a folded model
as shown in Figure 6. Figure 6
Folded paper
The Second Phase maquettes
The Second phase exploration of the folding geome- explorations.
tries was executed in the Rhinoceros CAD environ-
ment. Rhinoceros (Rhino) is a stand-alone, commer-
cial NURBS-based 3-d modelling tool. The paramet-
ric modelling script constructed the tested geome-
try by firstly; building the office room and dividing
the southern façade into modular units and secondly;
modelling the origami unit modules to be located on
it.
Origami derives from the Japanese words on
('folding') and kami ('paper'). It is originally the
ancient Japanese art of transforming a flat sheet
of paper into a sculpture or shape (Adrover 2015).
Two-dimensional object is transformed into a three-
dimensional one by only a series of folds. These folds
convert the paper object into a new entity with sur-
prising strength and kinetic properties. Paper glu-
ing or cutting, or assemblages of paper cuts, are
not defined as origami but are also explored as 'pa-
per pleats' (Adrover 2015). Starting with a piece of
paper and just by folding, ending up with diverse
The Office Space. In order to understand all the pos-
range of forms and patterns has attracted many peo-
sibilities and limitations of creating a performative
ple and resulted in hundreds of books and internet
building skin, a side-lit south-facing office space in
sites, origami courses as shown in Figure 1 (Sorguç
Cairo, Egypt has been selected as a case study. In
et al. 2009). In the 1960s, the geometrist, artist
office-type environments, where occupants typically
and computer scientist Ron Resch conducted rele-
cannot freely adjust their position and have rather
vant studies of deployable origami structures not

402 | eCAADe 34 - SHAPE, FORM AND GEOMETRY | Applications - Volume 2

 stringent visual comfort requirements, the building peated as many times as desired. In fact, the gener-
skin has to provide glare control using either clever ated module is used as input for the rectangular grid
façade orientations and design or automated shades component. This action allows a repetition onto the
(Reinhart 2014). The office could be considered as grid by module geometry, cells dimensions and num-
one of many identical rooms stacked together to ber of cells along the X and Y-axes inputs. Basically,
form a facade (Reinhart 2014). It was modeled using the pattern is originated by the main module, digi-
grasshopper, with spatial dimensions of 4m (width), tally defined by a series of points connected by lines.
8m (depth) and 3m (height) as shown in Figure 7. It's defined off the original geometry through a series
The measured space is 8 m deep to give the opportu- of commands: move, mirror, and rotate. The simula-
nity to clearly test the screen efficiency. The Origami tion of the digital folding of the model is decidedly
screen was allocated as a mediate interface in front of more complex to define since the kinetic movement
the glazing layer in the southern facade. of repeated folds must have their own axis and cen-
ter of gravity as well as be located on the grid with
Figure 7 arrangements to formulate the entire surface.
Spatial dimensions On the whole, Ten tested cases of folds are tested
of the office space. as in Figure 11. There are mainly two set of variables
on which the script is based upon. First of all, The mo-
tion variable of each folded module ranging from to-
tally closed to totally opened as shown in Figure 11
, only Six intermediate steps were tested as shown
in figure Figure 8. The second variable is the size of
the folded module. Three module sizes were tested
to divide a façade with dimensions 4m (width) x 3m
(height) which are; (2x1.5 m), (1x1m) and (0.5x0.5 m)
The Second phase exploration of the folding geome- as shown in Figure 9.
tries was executed in the Rhinoceros CAD environ-
Figure 8 ment. Parametric models connect points in a desired
The Six order and adjust the arrangement of folding planar
intermediate surfaces, to generate base lines of Origami pattern
motion tested using Grasshopper (McNeel 2010).
steps. Screen logic. Although the geometry of the skin is
important as it impacts the overall performance of
the building, due to the overall paper objective, a
simple squared module geometry was chosen as a
case study. The overall design of the skin is com-
posed of Origami modular folded in different direc-
tions, that acts as apertures. The horizontal and ver-
tical rotation and translation of the units creates the
depth of the skin; the simplest forms of folding be-
have similar to fins. Since the module is defined by
adjustable base lines, changing the numerical values
associated with each line can result in different shape
proportions. The squared folded module could be re-

SHAPE, FORM AND GEOMETRY | Applications - Volume 2 - eCAADe 34 | 403


The primary concept of the tested foldings of kinetic
origami skin, shown in Figure 10, is that when it
moves, either the corners of the square or the center
open to control different levels of daylight penetra-
tion using parametric modelling tool.
The Third Phase
The Daylighting Simulations will be integrated using
Grasshopper and Diva. The DIVA plugin was used to
perform the daylight analysis via integration with Ra-
diance and DAYSIM. Experimentation was conducted
using point in time simulations for year-round perfor-
mance, therefore, Iterative runs will be performed for
each configuration system, each run was planned to
be performed for four times of the year, March 21,
June21, September and December 21, taken every
hour of the working day from 8:00 am to 6:00 pm.
Daylighting evaluation is analyzed using two
codified metrics in LEED v4 to evaluate daylight au-
tonomy design. These two metrics are: Spatial
Daylight Autonomy (sDA) and Annual Sun Exposure
(ASE) metrics, which forms together a clear picture
of daylight performance. sDA describes how much
of a space receives sufficient daylight, which should
achieve (sDA 300 lux / 50% of the annual occupied
hours) for at least 55% of the floor area. sDA has
no upper limit on luminance levels, therefore, ASE
is used to describe how much of space receives too
much direct sunlight. It measures the percentage of
404 | eCAADe 34 - SHAPE, FORM AND GEOMETRY | Applications - Volume 2
floor area that receives at least 1000 lux for at least Figure 9
250 occupied hours per year , which must not exceed The size three
10% of floor area (USGBC 2013). tested variables
These two metrics were set to evaluate static (2x1.5 m), (1x1m)
buildings performance, which need to be modified and (0.5x0.5 m).
to adapt the dynamic buildings changeability. For
the dynamic system evaluation, some hourly perfor-
mance evaluations of the working hours must be es-
tablished to trace the daily, monthly and annually
façade response for climate changes, therefore HsDA
(Hourly spatial Daylight Autonomy) is used. HsDA
must exceeds 300 lux for at least 55% of the floor
area, where the Hourly Sun Exposure (HSE) is mea-
sured instead of the Annual Sun Exposure (ASE). HSE
measures the percentage of floor area that receives
at least 1000 lux per hour and shouldn't exceed 10% Figure 10
of floor area (Elghazi et al. 2015) The squared
Daylighting Performance Simulation. As previ- star-like cells are
ously mentioned, DIVA-Grasshopper is employed to scripted to act as
measure Daylight Autonomy (HsDA and HSE). The ge- apertures, with a
ometry of the skin was connected to the DIVA plug- variable hole
in. Each modeled surface was assigned a specific ma- opening diameter,
terial such as walls (50% reflectance), glazing (80% which will be
visual transmittance), ceiling (80% reflectance), floor optimized to
(20% reflectance) and exterior surface material (30% improve
reflectance). A set of sensors or nodes was located daylighting
above finish floor at (0.7m) high as shown in Figure performance.
12 to calculate (HsDA and ASE).
Figure 12
The set of sensors
located at 0.7 m
high.
Figure 11
Parametric
Variation of the
folding module:
The module was
reconstructed in
Grasshopper for a
parametric
alteration of
motion. The
rotation of folds
was parametrically
adjusted from 0 to
85 degrees (10
cases 10 steps).

SHAPE, FORM AND GEOMETRY | Applications - Volume 2 - eCAADe 34 | 405


The parametric model provides a constant feedback
on the explorations of the proposed scenarios of
Origami patterns to be animated for daylighting sim-
ulation. The main tested parameters are the form
of Origami folds, the module size and motion sce-
narios. The simulation includes 10 explored origami
screens with variations of 3 module sizes and 6 mo-
tion steps resulting in (180) cases tested on two sim-
ulation stages. First measuring illuminance at four
times of the year, March 21, June21, September and
December 21, taken every hour of the working day
from 8:00 am to 6:00 pm (7200 runs) to be able to
compare different configurations.
RESULTS
The process of parametric simulation for the Ten
tested cases took about seven days to be completed
and written in spreadsheets using SpeedSim-for-
DIVA plugin along with grasshopper (Wagdy 2015).
The daylighting performance results is represented in
406 | eCAADe 34 - SHAPE, FORM AND GEOMETRY | Applications - Volume 2
Figure 13
The figure shows
the 720 runs for one
of the cases where
the intermediate
sized modules (1m
x 1m) reached more
appropriate
Daylighting.
terms of SDA and ASE for different Origami modules.
The best possible solutions were found by reaching
the balance point between SDA and ASE. It aims to
maximize the sDA value while minimizing the effect
of ASE to get rid of direct sun exposure.
It was found that the large Origami module
(1.5x2m), resulted in a relatively high percent of HSE
and over-lit portion of space while the small modules
(0.5 x 0.5 m) results in a partially day-lit space. The in-
termediate sized modules (1m x 1m) could relatively
get rid of direct sun exposure while maintain appro-
priate Day-lit percentage as shown in Figure 13. In
addition, this size of apertures maintains a minimum
level of contact with outside.
All the kinetic folded tested cases resulted in
acceptable daylighting performance. Acceptable
daylight performance was mostly achieved at fully
opened cases and partially opened in winter time.
Unacceptable HsDA results always appears in De-
cember at 4:00 & 5:00 pm.
Figure 14
The best Four
configurations of 1
x1 1m size modules
which reach
appropriate
Daylighting.

SHAPE, FORM AND GEOMETRY | Applications - Volume 2 - eCAADe 34 | 407

 The simplest form of folding the square horizontally
creates an overhang that protects the interior space
from direct radiation, as well as creating a small scale
light shelf that re-directs daylight into the space be-
yond, resulting in a good HSDA percent for the deep
office space. Some kinetic complex Origami config-
urations, provided an appropriate daylighting rang-
ing between 75% and 90% daylit area. As shown in
Figure 14, Results of "A", the traditional horizontal ki-
netic screen, achieved HsDA (higher than 300 lux) for
75% of the floor area for 67.5% of the tested hours
and 90% of the floor area for 25% of the tested hours.
On the other hand, The HsDA results of "B" and "D" are
slightly lower than "A", they achieved HsDA for 75%
of the floor area for 60% of the tested hours while re-
sults of "C" achieved HsDA for 75% of the floor area for
only 50% of the tested hours. The HSE results of "D",
that receives more than 1000 lux, achieved 10% of
the floor area for only 7.5% the tested hours. Accord-
ingly, "D" is considered the lowest HSE and therefor
the most energy efficient. Finally, the horizontal ki-
netic screen "A" achieved very high HSDA while The
star-like kinetic screen "D" get rid of HSE high values.
On the whole, the graphs show how kinetic motion
variation of origami module could positively affect
daylighting performance moving from one position
to another through day and year as shown in Figure
14. This enriches the aesthetic qualities and façades
articulation of the façade while enhancing the day-
lighting performance behavior of the skin.
CONCLUSION
Kinetic Origami cellular façade can be used as ef-
ficient solution which responds dynamically to the
sun, folding to shade or expose the building; re-
sulting in minimized heat gain while providing ap-
propriate daylighting. Folding surfaces, particularly
complex crease configurations, can be modeled and
tested in variation using these algorithms. This
makes it possible to design and test folding pattern
configurations by creating a flat tessellation pattern.
It is recommended to test daylighting performance in
the different orientations to reach better understand-
408 | eCAADe 34 - SHAPE, FORM AND GEOMETRY | Applications - Volume 2
ing of origami screens efficiency. Besides, a full an-
nual simulation can trace the daily, monthly and an-
nually façade response for climate changes.
REFERENCES
Adrover, E. R. 2015, Deployable structure., Laurence King
Publishing
Crawford, S 2010 'A Breathing Building Skin.', Proceedings
of the Association of Computer Aided Design in Archi-
tecture ACADIA, p. 211–217
Davis, D, Salim, F and Burry, J 2011 'Designing Respon-
sive Architecture: Mediating analogue and digital
modelling in studio.', Proceedings of the 16th Inter-
national Conference on Computer-Aided Architectural
Design Research in Asia (CAADRIA),, Hong Kong, pp.
155-164
ElGhazi, Y, Wagdy, A and Abdelwahab, S 2015 'Para-
metric Driven Design for Kinetic System, Optimiz-
ing Kaleidocycle facade configuration for daylight-
ing Adequacy in hot arid climate.', IBPSA 2015, India
ElGhazi, Y, Wagdy, A, Abdelwahab, S and Gamal, A
2014 'Daylighting Driven Design: Optimizing Kalei-
docycle Facade For Hot Arid Climate.', Fifth German-
Austrian IBPSA Conference, RWTH Aachen University
Leach, N 2009 'Digital Morphogenesis', Architectural De-
sign, pp. 32-37
Lee, D and Leounis, B 2011 'Digital Origami: Model-
ing planar folding structures', ACADIA Regional 2011:
Parametricism: (SPC), pp. 25-29
Mahmoud, AH and ElGhazi, Y 2016, 'Parametric-Based
designs for Kinetic facades to optimize Daylight Per-
formance: Comparing Rotation and translation Ki-
netic motion for Hexagonal Façade Configuration',
Solar Energy, 126C, pp. 111-127.
McNeel, R 2010 'Grasshopper - Generative Modeling
with Rhino.', Seattle, USA: McNeel North America.
Reinhart, C and Stein, R (eds) 2014, Daylighting Hand-
book: Fundamentals, Designing with the Sun
Schenk, M 2011, Folded Shell Structures, Ph.D. Thesis, Uni-
versity of Cambridge.
Sorguç, A. G., Hagiwara, I and Selcuk, S 2009 'Origamics
in architecture: a medium of inquiry for design in ar-
chitecture', Metu Jfa 2
Wagdy, A 2015 'SpeedSim-for-DIVA version 1.0 Retrieved
4/18/2015', http://www.aymanwagdy.com/#!speedsim/cjg9.
Wagdy, A and Fathy, F 2015, 'A parametric approach
for achieving optimum daylighting performance
through solar Screens in desert climates', Journal of
Building Engineering, 3, pp. 155-170
Structural Architectural Elements Made of Curved Folded
Sheet Metal
Vlad Andrei Raducanu1 , Vasile Danut Cojocaru2 , Doina Raducanu3
1
National University of Arts Bucharest, FDADDesign Department 2,3 University
Politehnica of Bucharest, Faculty of Materials Science and Engineering
1
http://unarte.org/national-university-of-arts-bucharest.php
2,3
http://www.upb.ro/en/
1
zoster.ar@gmail.com 2 dan.cojocaru@mdef.pub.ro
3
ac_doina@yahoo.com

To deviate from conventional metallic structural elements is not an easy

achievement, especially if free forms with curved surfaces are wanted. One
approach that makes sinuous volumetric metallic shapes accessible is curved
folded sheet metal.The aim of the current application is to create a
reinterpretation of the classical column, an architectural element which is both
decorative and structural. This is achieved through curved crease folding of steel
sheet metal. To aid the form-finding process, a virtual simulation of the bending
process is done using computational means.

Keywords: curved crease folding, metallic material behaviour, computational

design, digital stress analysis

INTRODUCTION cessed by more accessible technology, such as 3-axis

In architecture, metallic structures are used increas-
ingly due to their low weight, slim profiles, easy
(dis)assemblage and facile recycling. The usual struc-
tural elements consist of extruded profiles such as
H or I beams which also usually lead to conven-
tional architectural displays. Making less standard-
ized, freeform structural elements, that could con-
tribute to the architectural aesthetics of the built en-
vironment, does involve at the present time major
costs if metallic three-dimensional printing is consid-
ered. A cost-effective alternative is to use sheet metal
that can be folded in order to achieve sinuous, aes-
thetically pleasing shapes, that through digital and
physical analysis can be tweaked for structural per-
formance. The sheet metal development can be pro-
CNC milling machines or CNC punch presses.
STATE OF THE ART
Curved crease folding (Huffman 1976) is an expand-
ing research domain that is still in development,
hence options to digitally simulate curved crease
folding designs are limited (Bhooshan et al. 2015).
Due to the cumbersome process of digitally simulat-
ing the deformations that arise when folding sheet
material along curved lines, working with physical
models is often the cornerstone in the development
of a curved crease design. The design alterations of-
ten go from the physical model to the digital one and
not the other way around (Tachi and Epps 2011), yet
the (folded) digital counterpart is necessary for en-

SHAPE, FORM AND GEOMETRY | Applications - Volume 2 - eCAADe 34 | 409

 semble visualization and especially if structural and
other types of digital analysis are desired in order to
inform the form-finding process.
The most basic way to achieve curved folded designs
is to generate the fold lines as intersections of a devel-
opable surface with a cutting plane, the result being
the reflected geometry of one of the two sides (Mi-
tani and Igarashi 2011) (see Figure 1), a method that
can be seen in David Huffman's famous designs (De-
maine et al. 2010). In this cases, a fully digital sim-
ulation, without feedback from a physical model, is
possible, but the design spectrum can be fairly lim-
ited.
For more complex shapes, the following method of
virtual simulation of curved folding process can be
used within Grasshopper (Tachi and Epps 2011):
1. A design solution is searched through the use
of physical models (usually paper models).
2. Once a plausible idea emerges, the develop-
ment is digitally reproduced, preferably in an
algorithmic manner, so that the folding pat-
tern can be easily altered in order to optimize
the solution.
3. The simulation of the folding process takes
410 | eCAADe 34 - SHAPE, FORM AND GEOMETRY | Applications - Volume 2
place using a rationalization of the original
development as a rigid origami pattern (see
Figure 2). Rigid origami uses polyhedral pla-
nar plates joined by rectilinear hinges (Miura
1970) rather than a continuous elastic sheet Figure 1
like paper or sheet metal. This conversion Curved crease
needs the ruling lines of the folded shape. folding through
These lines can be "discovered" on the phys- plane reflection.
ical model (one way of doing this is by us-
ing a ruler - Figure 3). A certain develop-
ment (in its original fluent form) doesn't de-
scribe a single folding outcome, it depends on
the way the different strips get bent (see Fig-
ure 4). The outcome is nevertheless a ruled
surface which therefore is developable and Figure 2
has ruling lines. Different configurations of The reinterpreted
the same development display differently ori- development as
ented ruling lines. This means that the rulings rigid origami
pattern determines the way the development pattern (quad
gets folded digitally. These ruling pattern can mesh).
be digitally generated, by defining an algo-
rithm based on the physical model examina-
tion (rather than manual reproduction from
scanned flattened models).
4. Analysis - the rigid plate interpretation is a
mesh that can be subject to different analyses,
be it structural or otherwise.
Figure 3
Discovering ruling
lines on the paper
model using a ruler.
Figure 4
Two resuting
surfaces (with
different rulings)
based on the same
development.
Figure 5
Paper model
studies for curved
folded column
design families.

CURVED FOLDED DESIGN DEVELOPMENT
The goal was to investigate diverse shapes that could
work as both structural and ornamental architectural
elements, such as the columns used in the classical
orders. We intended to use sheet metal that can be
very effective from a structural point of view but also
generates pleasing reflections when curved. Sev-
eral curved crease folding column designs were de-
veloped at model scale (see Figure 5), each with its
own advantages and disadvantages regarding differ-
ent aspects of aesthetical, practical and structural na-
ture. The chosen design (see Figure 7) references
the entasis and the fluting found in the classical ar-
chitectural orders (see Figure 6) in addition having a
hexagonal section better illustrates the idea of a cir-
cular section, which is specific to the concept of col-
umn. The design needed to include new character-
istics compared to the classical or even contempo-
rary columns, hence the idea of transparency arose.
This was achieved by splitting the column, and later
on also the option of using inner light that accentu-
ates the shape of the column in dark environments
and further enhances the ornamental aspect of this
structural element (see Figure 7).

SHAPE, FORM AND GEOMETRY | Applications - Volume 2 - eCAADe 34 | 411

 ALGORITHM Figure 6
The column is based on a number of 6 strips. The Entasis (bulging of
individual strips have a symmetrical arrangement of the profile) and
fold trajectories, which eases the determination of fluting (longitudinal
the ruling lines needed for the digital folding sim- grooves) present on
ulation (see Figure 8) . There are a few principles classical columns,
(Fuchs and Tabachnikov 1999) that relate the orienta- Paestum Basilika [1].
tions of the ruling lines to the way the developments
gets folded (see Figure 9). These principles were im-
plemented in the algorithm, and the digital simula- Figure 7
tion does return an acceptable approximation of the 1/3 scale prototype
physical model. In this way, the algorithm includes an - lightning effects.
automated adaptation of the ruling lines pattern that
generates the development and thus fine-tuning the
design requires fewer physical models. The physical
models are still necessary at times, since sheet ma-
terial doesn't conform exactly to its digital represen-
tation due to different aspects, like the fact that the
creases do have a tendency to resist folding after a
certain angle, whereas in the digital simulation the
folding lines are free-moving hinges.

Figure 8
Digital folding
simulation using
Kangaroo 2 - chain
reaction starting
Once the rationalized development is generated, a from one end.
planar quadrilateral mesh can be based on it and dig-
itally folded (Kergosien et al. 1994). The deforma-
tion can be performed using specific origami soft-
ware, such as Freeform Origami and Rigid Origami
Simulator (Tachi 1999) [2], but the Kangaroo plugin
for Grasshopper can also simulate the deformations,
which is more convenient since the same software
(Rhinoceros + Grasshopper) can be used for a larger
part of the project and tweaked to specific needs.

Figure 9
Illustrating curved
folding
mathematical
principles
(illustration from
Tachi and Epps
2011).

412 | eCAADe 34 - SHAPE, FORM AND GEOMETRY | Applications - Volume 2

Figure 10 Kangaroo 2 was used in the simulation with the fol-
The end parts of the lowing settings: the quads needed to retain their
strip do not remain planarity as much as possible, also their perimeter
collinear when and diagonals needed to remain the same length,
folding. in order to prevent deformation. No hinge restric-
tions were implemented at first. This made the dig-
ital model have a different final shape in the middle
section compared to it's physical counterpart, where
the two lateral strips are more open due to the ten-
dency of the sheet material to resist bending, there-
fore the next step was to implement this resistance in
Kangaroo. This however made the initial approach,
where one end got folded and the rest came into
place perfectly (see Figure 8), less than ideal, since
the fold didn't propagate to the other end (see Fig-
Figure 11 ure 11). In the end, virtual banks were used to fold
Bending resistance the digital module (see Figure 12). Another problem
of the hinges makes that arose is that the two ends of the module didn't
the folding process remain collinear (this can be seen in both the digital
fade towards the and the physical models - see Figure 10). When used
other end. as an ensemble in reality, they do remain collinear the
ends being fixed together (see Figure 13). Making
them collinear however generates a pinching effect
in the middle of the module, where it is at it's narrow-
est, which is a weak spot, as can be seen both from
physical tests as well as digital analysis (see Figure
13). To counteract this, local embossing of the sheet
metal will be explored (see section Further Develop-
ments)

Figure 12 Prototype
Virtual banks were A 1:3 prototype was build using 1mm pickled steel
used to fold the sheet metal that was welded after the folding process
digital (see Figure 14). A CNC punch press (Durma RP6) was
development. used to create different hole patterns for the creases
(see Figure 15). The folding process was made us-
ing a press brake. Preliminary tests showed that a
small dent in the middle portion of the crease con-
tour is needed (using the press brake locally) so that
the metallic sheet folds along the creases when the
extremities are folded all the way to their final angle.

SHAPE, FORM AND GEOMETRY | Applications - Volume 2 - eCAADe 34 | 413

 POST-FOLDING ANALYSIS Figure 13
Mechanical Structural Testing Digital structural
The prototype was mechanically tested for compres- analysis using
sion with an increment of 0.5 KN using an Instron Karamba - weak
3382 machine (figure 16). The force/displacement point in the middle
variation was linear for all tests. The tests were of the module.
stopped at a force of 5KN (500Kgf ) that exposed a
compression of only 0.39mm, which indicate a very
good mechanical behavior of the design. Further
tests will be made to test it's absolute limit.

FURTHER DEVELOPMENTS
The next stage involves implementing embossed
"dimples" using the same punch press to modify the
material properties. Strategic dimple patterns will be Figure 14
generated digitally depending on the distribution of Prototype - scale
loading forces resulted from the smooth surface pro- 1:3.
totype analysis.
Micro-scale testing of the results will be made
in order to obtain relevant material characteristics
around dimples; the local increase in strength is mon-
itored using mechanical tests coupled with hardness
tests. Some preliminary tests were made at a small
scale. Samples were prepared to be examined by a
scanning electron microscope (Tescan VEGA II - XMU)
Figure 15
which allows us to examine the local deformations
Prototype
(Figure 17). Further testing will be used to observe
development -
the influence of stress hardening through different
different hole
embossing patterns on the column as an ensemble.
patterns.
Metallic materials behavior will be studied from the
point of view of structural properties as well as their
influences upon fabrication. Further prototypes and
samples will undergo similar tests, in the end a 1:1
scale column is to be build.

414 | eCAADe 34 - SHAPE, FORM AND GEOMETRY | Applications - Volume 2

Figure 16 CONCLUSIONS
Compression Regarding the topic of the 34th eCAADe Conference -
testing of the "Complexity & Simplicity", the output of the research
prototype (using presented in this paper leans towards simplicity, from
Instron 3328 a geometrical point of view, but some of the most
machine). satisfying designs rely in fact on simple shapes, such
as Oscar Niemeyer's buildings (Figure 18). As sculp-
tor Constantin Brancusi said, ''Simplicity is complex-
ity resolved''. On the other hand, intensely complex
designs, such as Michael Hansmeyer's columns are
just as fascinating (see Figure 19) . Advanced com-
putational design methods are tools and should be
seen as such, they don't dictate a classification of the
outcoming results, but simply are means to explore
possibilities that were less tangible before.

ACKNOWLEDGEMENTS
The work was supported by a grant of the Romanian
National Authority for Scientific Research, CCCDI- UE-
Figure 17 FISCDI, project number 316/2014.
Dimple design Special thanks to Gregory Epps and the members
testing using of Robofold [17] for welcoming me at their studio and
scanning electron showing me the methods they use to manage curved
microscope (SEM - folding designs.
Tescan VEGA II -
XMU). REFERENCES
Bhooshan, S, Van Mele, T and Block, P 2015 'Discrete fu-
Figure 18 nicular structures with curve-crease-folded moulds',
Proceedings of the International Association for Shell
Oscar Niemeyer
and Spatial Structures (IASS) Symposium, Amsterdam
cathedral [3]. Demaine, ED, Demaine, ML and Koschitz, D 2010 'Recon-
structing David Huffman's Legacy in Curved-Crease
Folding', Origami5: Proceedings of the 5th Interna-
tional Conference on Origami in Science, Mathematics
and Education (OSME 2010), Singapore , pp. 39-52
Fuchs, D and Tabachnikov, S 1999, 'More on Paperfold-
ing', The American Mathematical Monthly, 106(1), pp.
27-35
Figure 19 Huffman, DA 1976, 'Curvature and Creases: A Primer on
Michael Paper', IEEE Transactions on Computers, 25(10), pp.
Hansmeyer's 1010-1019
Kergosien, Y, Gotoda, H and Kunii, T 1994, 'Bending and
colums [4].
Creasing Virtual Paper', IEEE Computer Graphics and
Applications, 14(1), pp. 40-48
Mitani, J and Igarashi, T 2011 'Interactive Design of Pla-
nar Curved Folding by Reflection', The 19th Pacific

SHAPE, FORM AND GEOMETRY | Applications - Volume 2 - eCAADe 34 | 415

 Conference on Computer Graphics and Applications,
Kaohsiung, Taiwan, pp. 77-81
Miura, K 1970, 'Proposition of pseudo-cylindrical con-
cave polyhedral shells', Institute of Space and Aero-
nautical Science, Report No. 442, pp. 141-163
Tachi, T 2009 'Generalization of Rigid Foldable Quadrilat-
eral Mesh Origami', Proceedings of the International
Association for Shell and Spatial Structures (IASS) Sym-
posium 2009, Valencia
Tachi, T and Epps, G 2011 'Designing One-DOF Mecha-
nisms for Architecture by Rationalizing Curved Fold-
ing', Proceedings of the International Symposium on
Algorithmic Design for Architecture and Urban Design
(ALGODE TOKYO 2011), Tokyo
[1] en.wikipedia.org/wiki/File:PaestumBasilika.jpg
[2] www.tsg.ne.jp/TT/software
[3] http://www.chocolate-fish.net/albums/Brazil/Bra
silia/Brasilia-Cathedral-outside.jpg
[4] http://www.michael-hansmeyer.com/images/columns
/columns1.jpg
[5] http://robofold.com/

416 | eCAADe 34 - SHAPE, FORM AND GEOMETRY | Applications - Volume 2

Collaborative Modeling with Symbolic Shape Grammars
Martin Ilčík1 , Michael Wimmer2
1,2
TU Wien
1,2
{ilcik|wimmer}@cg.tuwien.ac.at

Generative design based on symbolic grammars is oriented on individual artists.

Team work is not supported since single scripts produced by various artists have
to be linked and maintained manually with a lot of effort. The main motivation for
a collaborative modeling framework was to reduce the script management
required for large projects. We achieved even more by extending the design
paradigm to a cloud environment where everyone is part of a huge virtual team.
The main contribution of the presented work is a web-based modeling system with
a specialized variant of a symbolic shape grammar.

Keywords: collaboration, symbolic shape grammar, generative modeling

INTRODUCTION team members creates a lot of work overhead. Merg-

Procedural modeling is a popular approach in GIS,
urban planning and for CGI film effects. In such ap-
plications detailed architectural content has to be
generated on a very large scale, hence shape gram-
mars with symbolic matching (Stiny 1982) are prefer-
ably used to simplify the modeling process. Proce-
dural representation of objects allows to use a sim-
ple notation for complex symmetries and patterns. A
whole class of objects can be instantiated from a sin-
gle grammar. However, large scenes contain a vari-
ety of object classes, each exhibiting unique patterns.
The number and complexity of involved grammars
increases with each new class. Manual management
and efficient re-usage of the designs soon becomes
problematic.
Additional difficulties arise when the creative
process is not limited to a single user. Our experi-
ence with the standard industry tool CityEngine by
Esri [1] indicates that team work with a repository of
grammars is really difficult. Production rules have to
be manually imported and referenced across gram-
mars. Constant manual synchronization between
ing of updates requires a dedicated expert just for the
repository management. Efficient collaborative tools
are a big challenge for the field of generative design.
Contributions
We have created a collaborative design framework
based on the theoretic work of Müller et al. (2006)
which defined the successful CityEngine. Many as-
pects of the single-user oriented concept needed to
be revised. The most innovative features of the pro-
posed approach are:
1. A cloud repository of generative grammars.
2. A new language feature: procedural queries
that automatically apply rules from the cloud.
3. No management overhead for team projects.
4. No special syntax, grammars are simple C#
scripts.
Our experimental framework called Michelangelo [2]
implements the presented features. All modeling is
performed on the web, hence all grammars get im-
mediately stored in Michelangelo's cloud. The main

SHAPE, FORM AND GEOMETRY | Applications - Volume 2 - eCAADe 34 | 417

 theoretic novelty of the grammar definition is the
concept of procedural queries which replace symbols
as the main control element of the derivation pro-
cess. Queries are the key to collaborative modeling.
They automatically seek and link rules from the cloud
and determine the order of their execution.
COLLABORATION FOR GRAMMARS
None of the existing grammar-based modeling tools
was created with collaboration in mind. However,
it is mainly the participation of artists with hetero-
geneous skills and ideas (Edmonds et al. 1994) that
stimulates variety and drives the complexity of re-
sults to a new level. An optimal interface to a gram-
mar with 5 rules looks different than the interface to
100 rules (Gips 1999). A team of artists will easily
produce several such grammars. A cloud repository,
once properly filled, will host several orders of mag-
nitude higher amount of rules. With CityEngine, be-
ing the most popular tool implementing a symbolic
shape grammar, we have analyzed the state-of-the-
art pitfalls of collaboration in generative design.
CityEngine
CityEngine stores designs in a basic digital repository.
The best practice for team work is to fragment new
designs into short, well organized grammar snippets.
Their connections need to be declared directly in the
grammar files [3]. I.e., to link a symbol to a rule from
a different grammar it requires manual import and
mapping.
import B: “book.cga"
Book --> B.Book
In a collaborative environment, such static assign-
ment has several drawbacks:
1. The whole script becomes invalid if someone
renames the entry rule of the imported gram-
mar. Figure 1 illustrates the case on a simple
hierarchy of grammars. In the real world its
complexity would be much higher. Renaming
the Book rule to the more specific Hardcover-
Book requires to update a considerable num-
418 | eCAADe 34 - SHAPE, FORM AND GEOMETRY | Applications - Volume 2
ber of grammars (in Figure 1 marked dark or-
ange). Keeping permanent overview and pre-
cise synchronization is not feasible even for
small projects.
2. If someone designs an alternative grammar
for an item, other team members willing to
use it must manually import the new file or
rule into their grammars and include a condi-
tional to decide when to use which source.
import B1: “book.cga"
Book1 --> B1.Book
import B2: “book2.cga"
Book2 --> B2.Book
#decide on Book1 or Book2
Book --> case { ... }
3. The control is fully passed to the linked gram-
mar, which acts like a black-box. It is not pos-
sible to override its rules or imports. We con-
sider this limitation as the main obstacle for
emergence of alternative designs within a col-
laborative environment. It is not possible to
alter materials or model parts created by im-
ported grammars without changing the orig-
inal script. If it comes from a different au-
thor, it is necessary to read and understand it
first. Rewriting someone else's grammar file is
a bad practice which is mostly avoided by cre-
ating an edited copy leading back to problem
2.
During intense artistic work the administration of
a complex hierarchy of grammars can easily over-
whelm the authors. To support efficient team work
it is necessary to simplify the control over the gram-
mars.
Imperative vs. declarative design
Starting with the first shape grammars (Stiny 1982)
and L-Systems (Prusinkiewicz and Lindenmayer
1990), through GML (Havemann 2005) and CGA-
Shape (Müller et al. 2006), up to the most recent
PCGR (Silva et al. 2015) and CGA++ (Schwarz and
Müller 2015), all theoretic concepts are very close to
Figure 1
Example of a
hierarchy of
grammars.
Renaming the book
rule (yellow) or
adding a new one
would require to
manually introduce
changes in all
grammars linking
the book grammar
(orange). common imperative programming languages. They will be embedded automatically.
demand that the designer specifies a sequence of
modeling steps which creates the desired object. DECLARATIVE SHAPE GRAMMAR
Fixed linking of production rules is the main ob- The proposed declarative shape grammar is based on
stacle for collaborative modeling. Artists without a symbolic set grammar over an algebra of shapes
a computer science background are not skilled to (Wonka et al. 2003). String symbols provide only lim-
assemble and link complex scripts, however they ited meaning to shapes. Therefore, in the declarative
can easily describe their goals. Generative design grammar they are replaced by a more complex struc-
tools seem to overlook this fact, but for example in ture for shape description with semantics consisting
logic programming goal-driven computation is quite of:
popular. It utilizes the declarative programming
paradigm (Kowalski 1979). Applied to generative • G a set of string goals.
design it means that instead of specifying how, the • T a set of tags T∩G={}. Tags are goals which
designer should mainly describe what to generate. were fulfilled.
The imperative "then continue with the modeling op- • τ a Boolean flag marking terminal shapes.
eration X" attached to each modeling operation gets • A a set of named attributes (numeric or string).
replaced by the declaration "then try to transform the
output shape to S". Rules must be adapted to the semantics as well. In-
Replacing a pointer to a particular rule X by a stead of a symbol, a rule in the declarative grammar
shape description S of the desired result is a subtle contains a set of predicates over the shape semantics
syntactic difference with major impact on the collab- on its left side. The next section provides an exam-
orative modeling process. Production rules are still ple. There are predicates available for goals, tags and
listed in grammars, but there are no explicit connec- attribute names (with s being semantics of the eval-
tions between them. Such freedom allows to com- uated shape):
bine grammars automatically and arbitrarily in the
• IfGoal(s, g0 . . .gn ) = s.G ⊇ {g0 . . .gn }
background without exposing the complex manage-
• IfTag(s, t0 . . . tn ) = s.T ⊇ {t0 . . . tn }
ment tasks to the artist. At the same time, the spec-
• IfHas(s, a0 . . . an ) = s.Akeys ⊇ {a0 . . . an }
ification of shape descriptions introduces a query
mechanism to the grammar. Using a list of simple All of the predicate functions also have negated
goals, e.g. chair, wooden, the author can specify what counterparts with the prefix IfNot. For predicates
a non-terminal shape should become. If the reposi- based on attribute values there is a general predicate
tory contains a design that fulfills the given goals it If which directly evaluates the given function f de-

SHAPE, FORM AND GEOMETRY | Applications - Volume 2 - eCAADe 34 | 419

 fined over the semantics s.
• If(s, f) = f(s)
Derivation step
In imperative approaches the shape symbol directly
specifies the next rule to be applied.
Symbol1 --> commands Symbol2
In the declarative grammar the set of tags defines the
current state while the set of goals specifies the final
state to be reached. Hence, the equivalence of sym-
bols for rules matching has to be replaced by subset
relations over the shape semantics.
Predicates(Semantics1) --> commands
,→ Semantics2
Using the predicates described above one defines
conditions for the rule to be applicable to a shape. A
special command Fulfills checks for existence of goals
(using IfGoal) and converts them to tags.
A typical condition is that a rule can be applied
to a shape only if the set of its goals contains all goals
fulfilled by the rule. Further requirements can be
given for and shape attributes. E.g., a rule that ini-
tiates table decoration in a certain style may require
the shape to be tagged restaurant, table with goals
romantic, dinner and its size should be less than 0.8
meters in each dimension:
IfIs(s, "restaurant", "table") ∧
IfGoal(s, "romantic", "dinner") ∧
Fulfills(s, "setting") ∧
If (s.Size.X < 0.8 ∧ s.Size.Z < 0.8)
IMPLEMENTATION AND USAGE EXAMPLE
The system is implemented as a cloud service called
Michelangelo [2]. For the modeling basis we use a
variant of the CGA-Shape by (Müller et al. 2006).
Grammars are written as C# scripts providing a famil-
iar environment and powerful features at once. Let
us demonstrate the syntax using a very simple exam-
ple of a chair. The main difference to standard sym-
bolic grammars is that instead of a symbol the user
sets goals (a set of strings) for the axiom and shapes.
420 | eCAADe 34 - SHAPE, FORM AND GEOMETRY | Applications - Volume 2
They represent semantics of the desired modeling re-
sult.
new Axiom (" chair", "wooden ").Size
,→ (0.4 ,0.8 ,0.4);
Rules modify the model to match the given goals and
introduce more specific goals in a top-down manner.
The following rule divides the shape vertically into
three parts using relative sizing with respect to the
shape extent on the Y axis. It also adds a new goal to
each one of them. A randomized numeric attribute is
assigned to the support part using the Rnd(min, max)
function.
new Rules.Split(Axis.Y,
Relative (0.4).Goal (" support ").
,→ Attribute ("Leg.Size", Rnd
,→ (0.03 ,0.05)),
Relative (0.05).Goal (" seat "),
Relative (0.55).Goal (" lean "))
.Fulfills (" chair ");
Contrary to the standard symbolic matching, we in-
troduce a goal-driven declarative approach. Rules
are sought and applied to each shape so that its goals
get fulfilled. The previous example rule fulfills the
goal chair, thus it can be applied to the axiom. Ful-
filled goals become tags, hence all shapes produced
by the rule will be tagged chair. Tags and numeric
attributes accumulate in the shapes over the whole
derivation process.
Most of the rules fulfill at least one goal, but it
is not required. Matching of rules to shapes is fur-
ther controlled by predicates: IfGoal checks a goal
without fulfilling it, IfIs checks a tag (i.e. a fulfilled
goal), IfHas checks a named attribute. More com-
plex predicates are explained in the documentation
of Michelangelo.
A local goal (note the lowercase) can be fulfilled
only by the same grammar to avoid interference with
other goals in the cloud. The compiler takes care of
setting the corresponding Fulfills to be local as well.
Shapes marked Void() become terminals with no ge-
ometry.
new Rules.Split(Axis.X,
 Absolute(@Double ("Leg.Size ")). was queried instead and returned a rule assigning
,→ goal ("split -legs "), brown color to wooden objects. Our design was re-
Relative (1).Void (), fined automatically without any intervention neces-
Absolute(@Double ("Leg.Size ")). sary. This trivial example already shows the solution
,→ goal ("split -legs "))
to the first two pitfalls listed in section "Collaboration
.Fulfills (" support ");
for Grammars".
new Rules.Split(Axis.Z, Let us demonstrate the queries on a more explicit
Absolute(@Double ("Leg.Size ")). example. The following grammar adds an alternative
,→ Goal ("leg"), rule for the chair support. It only defines three rules
Relative (1).Void (), and relies that any other rules necessary for a wooden
Absolute(@Double ("Leg.Size ")). chair will be fetched from the cloud. That is equiva-
,→ Goal ("leg") lent to overriding a part of an existing design without
.Fulfills ("split -legs "); any direct knowledge about it. Hence, the third pitfall
The previous pair of rules produces the legs of the of CityEngine is solved as well. Figure 2 (right) shows
chair. To assure random, yet consistent sizing for all the resulting chair.
of them, the previously defined attribute Leg.Size is
Figure 2
accessed using the special @Double function. It im-
Brown material for
plicitly adds a corresponding IfHas predicate to the
the chair is
rule. Attribute getters can be composed to complex
automatically
expressions. To finish the basic chair model we add a
linked from the
simple rule for the lean construction.
cloud. Support of
the basic chair new Rules.Size(Axis.Z, 0.04)
model (left) can be .Fulfills (" lean ");
easily overriden At this point the grammar already produces a sim-
(right). ple chair (see figure 2 left) with variable leg thickness.
Note that using Void() is the only way of explicit ter-
minal assignment. Any other termination directive
would contradict the basic idea of connecting gram-
mars in the cloud. The derivation process shall con- new Axiom (" chair", "wooden ").Size
tinue until there are no more rules applicable. A pre- ,→ (0.4 ,0.8 ,0.4);
defined limit of derivation steps assures termination.
new Rules.Split(Axis.Y,
Cloud queries Absoulute (0.4).goal (" stand "),
The cloud-based derivation process is the most revo- Relative (1).goal ("leg"))
lutionary feature of the system. It allows efficient col- .Goal (" metalic ")
laborative modeling without any overhead. It might .FulFills (" support ");
seem that the chair grammar contained modeling
new Rules.Scale(Vec3 (0.5 ,1.0 ,0.5) ,
operations only. In fact, it already exposed procedu-
,→ Pivot.Middle)
ral queries. The concept of goals is the query mech- .Fulfills ("leg");
anism integrated within the grammar. There was no
rule for material assignment in the grammar, but the new Rules.Scale(Vec3 (0.25 ,1.0 ,0.25) ,
chair in figure 2 is brown. As the wooden goal of the ,→ Pivot.Middle)
axiom was not fulfilled by any local rule, the cloud .Fulfills (" stand ");

SHAPE, FORM AND GEOMETRY | Applications - Volume 2 - eCAADe 34 | 421

 Now there are at least two rules stored in the cloud
which fulfill the support goal. In such case, one of
them is selected in a non-deterministic way. Stochas-
tic selection is possible as well. Primarily, rules from
the current grammar get selected. If none fits, the
cloud of grammars is queried for rules matching the
current shape. The derivation process in the back-
ground seamlessly jumps between the local and the
cloud rules all the time.
The number of results returned from the cloud
can be immense. The more common an object is, the
more alternative designs are expected to be found.
Moreover, the cloud is a dynamic repository with new
designs coming in all the time. The amount and vari-
ability of results produced by each grammar will in-
crease over time as the cloud grows. Therefore, it
is advised to keep open goals for all shapes so that
grammars written in the future can automatically add
details to all related designs.
Cloud-based modeling
The cloud can also impose negative effects on the
design process. The more alternatives the less pre-
dictable the results. Wrong semantic annotations in
a single grammar can easily propagate to a large class
of related designs and produce unacceptable results.
Many interesting aspects emerging from the work
with a cloud need to be addressed in addition to the
basic concept of a declarative grammar.
Ambiguous rule semantics. Rule predicates should
be precise, but at the same time simple. Designing
a good matching function for each rule is not easy.
Keeping predicates too general may have negative
impact on the whole cloud. In the previous example
the chair terminated with a leg goal opened for each
of its legs. If someone later writes a grammar for the
human body with the following rule it will ruin many
chairs by attachment of human legs.
new Rules.Split(Axis.Y, /*foot ,ancle ,
,→ etc .*/).Fulfills“(leg");
It is very important to include predicates for the con-
text in which the rule should be applied. There are
two ways of limiting the rule context. It can be done
422 | eCAADe 34 - SHAPE, FORM AND GEOMETRY | Applications - Volume 2
manually by adding an IfIs predicate to each rule, or
it can be set for a block of rules with a using block.
using (Predicates.IfIs (" chair ")) {
/*All rules in this block will inherit
,→ the IfIs */}
Both chair grammar examples from the previous sec-
tion should be wrapped by such a Predicates block,
except for the first rule which fulfills the chair goal.
Synchronization of queries. Procedural queries are
executed independently and in random order. If
there is a number of shapes with identical goals a dif-
ferent sequence of production rules will be applied to
each of them. But often their appearance should be
synchronized, e.g, for windows on a façade. Synchro-
nization of shapes is possible using the Synchronize
object.
var windows = new Synchronize ();
Shapes to be synchronized get attached to the Syn-
chronize instance by calling the Sync function. Syn-
chronization then applies for their children as well.
For all synchronized shapes, as long as their goals al-
low, the same sequence of production rules gets ap-
plied. Synchronization of rule parameters is left for
future work.
new Rules.Repeat(Axis.X,
Absolute (1.5)
.Goal (" Window ")
.Sync(windows))
.Fulfills (" Floor ")
.IfIs (" Facade ");
Quality control. Michelangelo is still a research pro-
totype. The system is used by a closed group of
trusted users. Despite of several protection measures
a single malicious rule could destroy almost every
produced model. In the current development stage
it can be avoided by utilizing restrictions on linking
rules from the cloud. The default mode all can be re-
duced to employ only rules by friends, mine or from
this grammar.
For the future we plan to utilize gamification in
the style of Stack Overflow (Jin 2015). Users will col-
lect achievements and reputation points. Based on
 the reliability of the grammar's author new rules will
become accessible in the cloud only after approval by
respected users. At the same time, objective review-
ing of grammars will be encouraged.
COMPLEX RESULTS
Figure 3
Almost 100 chairs
combined
randomly from only
4 source designs.
Figure 3 shows a variety of chair models with parts
combined from four incomplete grammars. The im-
age was generated by a grammar which contained
only axioms for wooden chairs, but no rules were
given. All modeling decisions were left to the cloud.
Variance implied by the high number of possible
combinations is the first important use-case of the
presented method.
Instancing of complex model hierarchies is the
most powerful use-case of the presented approach.
Procedural models, if defined well, are able to fit
automatically into different space volumes. Figure
4 shows an office building with 2500 m2 on three
floors populated purely by procedurally generated
content. Shape grammars were in particular helpful
with furniture layout and design due to extensive in-
stancing of similar objects. The most frequent items
are 240 desks and office chairs. A part of them is visi-
ble on the floor plan of the top floor on Figure 5. De-
tailed variations of an office are presented in Figure 6
and Figure 7. It is mainly the arrangement of furniture
that changes in the examples.
Evaluation of the whole building axiom was per-
formed in less than two minutes. The generation pro-
cess took 47,284 derivation steps controlled by 329
rules which were combined from 27 grammars. The
SHAPE, FORM AND GEOMETRY | Applications - Volume 2 - eCAADe 34 | 423
interiors include items designed independently in a
time span of more than half a year. Nevertheless,
no management was required for the maintenance
or merging of the sources. The cloud automatically
derived the models using a combination of all the
stored grammars.
Limitations
During several informal sessions we collected feed-
back for Michelangelo. Designers appreciate the
web-based platform for providing easy scripting and
good overview out-of-the-box. On the other hand,
rendering performance of large detailed scenes in a
browser can not match stand-alone applications.
Our most ambitious experiment was the pre-
sented office building. For such architectural tasks
the main drawback is the lack of collision detec-
tion and missing support for floor planing. Floor
plans created for the office building would not com-
ply with a different one. A different boundary poly-
gon and different placement of vertical structures like
columns, pipes and elevators would be the main is-
sues. A dedicated rule would be necessary for a uni-
versal procedural approach to floor planing.
The current scope of Michelangelo are furniture
items. Our main goal is to improve the available de-
formation rules to allow a wider range of forms. The
current set is limited to simple taper, twist and shear
rules.
FUTURE WORK
We hope that more out-of-house artists will try out
Michelangelo. As the cloud will grow there will be
new challenges for keeping the system consistent
and focused on the goals given in the axioms. The
topic of cross-reviewing grammars is part of a larger
research package planned for the next months. We
see a large potential in turning the cloud intelligent.
By observing user behavior and collecting feedback
on the produced models the cloud should be able to
learn and provide a highly customized experience:
• Learn new concepts and select rules better. If
there are many rules without an explicit pred-
 Figure 4
An office building
with its full interior
generated from the
cloud.

Figure 5
Interiors of offices
in the top floor.

424 | eCAADe 34 - SHAPE, FORM AND GEOMETRY | Applications - Volume 2

Figure 6
Variations of a
medium office.

Figure 7
More office layouts.

SHAPE, FORM AND GEOMETRY | Applications - Volume 2 - eCAADe 34 | 425

 icate on the current shape semantics it is hard
to pick the best match. The cloud can use the
collected data to learn new predicates and im-
prove the decision process even in contexts
not handled by the original grammar.
• Adapt stylistic details based on learning pref-
erences of the user. By learning semantic pat-
terns preferred by the user it should be possi-
ble to predict which cloud rules he or she will
like more.
While the concept of an intelligent cloud has clear
priority, many more ideas for the development of the
system are waiting for implementation. The variety
of geometric operations should be increased to al-
low more attractive and complex forms. A visual in-
terface hiding the programming part would be for
sure appreciated by non-experts for simplicity and by
experts for efficiency. Interactive visual analysis fea-
tures for the produced models are becoming clearly
necessary as the complexity of the produced mod-
els grows. Last but not least, the complexity of the
generated models is limited by memory and band-
width of the display devices. Simplifications by level-
of-detail techniques and textures baking will be nec-
essary to reduce the data load.
Kowalski, R 1979, 'Algorithm = Logic + Control', Commun.
ACM, 22(7), pp. 424-436
Leblanc, L, Houle, J and Poulin, P 2011 'Component-
based Modeling of Complete Buildings', Graphics In-
terface 2011, pp. 87-94
Müller, P, Wonka, P, Haegler, S, Ulmer, A and van Gool, L
2006, 'Procedural Modeling of Buildings', ACM Trans.
Graph., 25(3), pp. 614-623
Prusinkiewicz, P and Lindenmayer, A 1990, The algorith-
mic beauty of plants, Springer-Verlag, New York
Schwarz, M and Müller, P 2015, 'Advanced Procedural
Modeling of Architecture', ACM Trans. Graph., 34(4),
pp. 107:1-107:12
Silva, P, Eisemann, E, Bidarra, R and Coelho, A 2015, 'Pro-
cedural content graphs for urban modeling', Interna-
tional Journal of Computer Games Technology, 2015,
p. N/A
Stiny, G 1982, 'Spatial relations and grammars', Environ-
ment and Planning B: Planning and Design, 9(1), pp.
113-114
Wonka, P, Wimmer, M, Sillion, F and Ribarsky, W 2003,
'Instant Architecture', ACM Trans. Graph., 22(3), pp.
669-677
[1] http://www.esri.com/software/cityengine
[2] http://michelangelo.graphics
[3] http://cehelp.esri.com

ACKNOWLEDGMENTS
This research was financed by the Austrian Science
Fund project Nr. FWF P24600-N23.

REFERENCES
Edmonds, E A, Candy, L, Jones, R and Soufi, B 1994,
'Support for collaborative design: agents and emer-
gence', Communications of the ACM, 37(7), pp. 41-47
Gips, J 1999 'Computer implementation of shape gram-
mars', NSF/MIT Workshop on Shape Computation, pp.
56-66
Havemann, S 2005, Generative Mesh Modeling, Ph.D. The-
sis, TU Braunschweig
Jin, Y, Yang, X, Gaikovina Kula, R, Choi, E, Inoue, K and
Iida, H 2015 'Quick Trigger on Stack Overflow: A
Study of Gamification-Influenced Member Tenden-
cies', IEEE/ACM 12th Working Conference on Mining
Software Repositories, Florence, pp. 434 - 437

426 | eCAADe 34 - SHAPE, FORM AND GEOMETRY | Applications - Volume 2

Revising Stereotomy through Digital Technology
Pedro de Azambuja Varela1 , José Pedro Sousa2
1,2
Faculty of Architecture, University of Porto + DFL/CEAU/FAUP
1,2
{pvarela|jsousa}@arq.up.pt

Stereotomy is usually regarded as a classic discipline in the field of architecture

which has lost its relevance to other construction methods. Recently we have
witnessed approaches to the stereotomic way of building using modern digital
technologies, bringing the discipline to the present. On the other hand, the
complex shaped buildings being built in the present seem to lack some kind of
structure visible in classic stone architectures. A shift in construction methods
must be attended to, with its metallic beams and concrete massifs. For an
understanding of what remains the same and what changes from the classic
literature, this paper tries to create analogies between contemporary architecture
surfaces and the discrete stone structures of the past. To achieve this, a concept of
stereotomy semantics is introduced, creating a different posture towards the
problem of division of surface in architecture.

Keywords: stereotomy, treatise, complex geometry, interpretation

ON STEREOTOMY contained in Philibert de l'Orme's "Le premier tome

Stereotomy's meaning (from the greek, "stereo" -
solid and "tomia" - section) gives us clues to its clas-
sical meaning. Cutting solids acknowledges the ex-
istence of a previous big mass of material which is
to be separated into smaller blocks by a stonemason
or a carpenter. This is a practice not so much con-
cerned with the actual physical means to cut through
stone, but with the intelectual means to find the
correct shapes to be produced. This construction
method has a strong tradition in architecture as it
spawned bold structures, a large number of theoreti-
cal works and, possibly most importantly, Descriptive
Geometry (1798), the synthesis of Gaspard Monge's
stereotomy classes.
One of the most important research works on
stereotomy in the architecture field came in 1567
with the first structured information on the subject
de l'Architecture". In this book, the author stresses the
importance of control in design and construction, so
that the architecture project may be undertaken ac-
cording to the architect's vision until its final materi-
alisation. This claim in favour of architects appeared
in the Renaissance in a clear opposition to Gothic tra-
dition. While in the late medieval cathedrals' ribbed
vaults and buttresses were controlled by the stone
masons, de l'Orme wants to affirm the architect as the
incontestable decision maker. This is only possible if
he possesses the technical means to project arches,
vaults and its components in its three dimensional
detail, and finally draft them accurately in paper, se-
curing the design with a document.
De l'Orme inaugurated a tradition of research
and publication on stereotomy, which encompasses
more than 50 treatises published until the beginning

SHAPE, FORM AND GEOMETRY | Applications - Volume 2 - eCAADe 34 | 427

 of the 20th century. Among the most relevant ones,
the works of Girard Desargues on projective geom-
etry (1640), de La Hire's important notions on stat-
ics (1678), Frézier's widely published and one the
most used and referenced treatises in Europe (1737),
and Monge's work on developing the language with
which most architects still project architecture today
- Descriptive Geometry (1798).
The industrial revolution introduced cast-iron
and later steel which replaced in many ways the use
of stone in architecture, a construction technique
strongly dependent in stereotomy knowledge. As
new construction systems evolved, a loss of interest
in individually cut stones gradually set in. To take ad-
vantage of the economical benefits of industrial mass
production, materials were designed and employed
following strict modular schemes, favouring designs
based on symmetry and repetition. This new logic of
modularity did not favor the exploration of unique
architectural designs requiring customized building
components.
THE INTEGRATION OF DIGITAL TECH-
NOLOGIES
Engineering, in the facets of CAD (design), CAE (engi-
neering) and CAM (manufacturing) has been able to
restore a much welcome enrichment in design possi-
bilities of complexly shaped architectures. The avail-
ability of these technologies allows architects to de-
sign unique buildings with unique construction sys-
tem and elements, as can be seen in the seminal
project by Frank Gehry, the Guggenheim Museum in
Bilbao. This project design process used parametric
software which could handle a consistent 3D model,
as well as report complete schedules of quantities
and dimensions. Although some architects used
mesh based modelling softwares in the turn of the
century, NURBS (non-ration b-splines) is a type of ge-
ometry representation which has gaining traction in
the latest years, together with visual programming
tools. Architecture software has seen engineering
tools embedded with FEM (finite element method)
analysis, creating a more holistic view of design. Dig-
428 | eCAADe 34 - SHAPE, FORM AND GEOMETRY | Applications - Volume 2
ital fabrication is becoming every time less bounded
to a small number of axis, large robotic arms as well
as more capable multi-material 3D printers can cre-
ate complex and accurate shapes. These tools create
a different point of departure when projecting archi-
tecture, expanding possibilities in all fields of archi-
tecture, including that of stereotomy.
Current research on stereotomy under the dig-
ital umbrella has produced very interesting results.
A retake on classical stereotomy with contempo-
rary technologies is the pioneering experiments of
Giuseppe Fallacara of Politecnico di Bari on vaults
made of digital fabricated stone voussoirs (Fallacara
2009), recently with the added component of post-
tension. The BLOCK group of ETHZ has an on-
going research in digital stereotomy, which "devel-
ops a method for tessellating free-form masonry-like
vaults" (Rippmann 2010). Justin Diles from Knowlton
School researched in stereotomic hollow fiber blocks
(Diles 2014). Other works like the ones from Matter
Design (Kaczynski 2011) or Supermanouvre (Larsen
2012) only add to the clear notion that stereotomy
as an architectonic theme is a continuing interesting
approach, but now highly powered by current digi-
tal technologies. The understanding of stereotomy
as a construction technique that depends on the
unity of a multitude of small blocks of a single fam-
ily finds a harmonious analogy in parametric strate-
gies, in which a shape is computed a great number
of times with subtle differences, accurately different
to be part of a greater whole understood as a unified
stereotomic project.
A brief analysis of the afore mentioned works
shows us that the relation to classical stereotomy
is being taken literally, on a direct relation to the
works produced under the treatises of de l'Orme and
Frézier. Although we can clearly detect differences
to the pre 20th century approach, be it in the mate-
rial employed, the much greater geometric freedom,
or auxiliary forces other than simple compression,
we propose a broadening of the field of stereotomy
beyond compressive volumetric structures, so that
a common line of thought can be traced between
structures built of smaller parts. We will define some
notions of an on-going research in how classical
stereotomy literature contains multiple hints and re-
lations to contemporary projects so that the design
of subdivided surfaces, as a natural descendent of
stereotomic design, is ought to have a clear notion of
the similarities and differences between the classical
cutting of stone and the contemporary homonym.
TOWARDS AN ENLARGED SEMANTICS
A stereotomic project may be described by the veri-
fication of three invariants: pre-figurative, technical/-
geometrical and compression statics (Fallacara 2003).
If we develop these concepts, we may assert that the
three fields present in a stereotomic architecture are
those of drawing, projective-constructive technique
and performance. The first enables us to preview the
final appearance of the architectonic object, the sec-
ond allows us to define it formally and performance
validates the solution under a determined interest.
This hypothesis allows us to observe stereotomy with
one of its components expanded: performance. This
performance may be tension statics instead of com-
pression, or maybe surface curvature or even fabri-
cation possibilities. By allowing us to broaden this
concept, stereotomy can be taken into new fields not
necessarily stone made. In classic stereotomy, statics
is perceived as its performance determinant; in this
context, we understand statics as a meaning within a
semantic stereotomy.
By understanding stereotomy as a multi-
semantic and methodical approach to subdivid-
ing surfaces, much of treatises' concepts can find
a healthy echo in their digital counterpart of today's
high end architectonic complex structures. The need
for the control of geometry is in great extent eased
by the computer, which does not dispense mathe-
matical and geometric principles for solving these
kind of problems. A revision of stereotomy treatises
can enlighten the design of subdivisions in various
kind of architectonic surfaces, augmented by digi-
tal means in a similar fashion as mathematics em-
powered stereotomy. For the illustration of the idea
SHAPE, FORM AND GEOMETRY | Applications - Volume 2 - eCAADe 34 | 429
of a semantic stereotomy, two different projects,
Chartreuse de Valbonne (1770-1780) and Olympias-
tadion Munchen (1968-1972) we will be evaluated
under these terms.
The vaults of Chartreuse de Valbonne show
a very particular interest in creating decoration
with the constructive stone blocks that support the
church roof. Decorative religious themes seem to
be a contrasting attitude towards military buildings'
austere efficiency; Frézier is worried with the impor-
tance of Theory over Practice as to optimise the de-
signs (Frézier 1738), an optimisation that should be
understood in the context of a treatise of stereotomy
written in the 18th century by a military man: struc-
tural stability and practical means of attaining the
right geometry and transferring it to construction
materials were the main drivers of his design. In
Valbonne, the brothers France d'Avignon show a
focused interest in the decorative curves, that lie
above cutting simplification. Going beyond austere
and economic constraints, decoration become this
project's stereotomic semantics.
On another hand, the tensile roof of Frei Otto's
Olympiastadion [1] is held by seemingly opposite
forces than those that support the stone vaults of
classical stereotomy; instead of compression, trac-
tion is what binds the whole structure together. The
idea of traction related to stereotomy is not new. Al-
though metal girders can be used to secure stones
together, although a more striking example is that
of the lighthouse known as Smeaton's Tower (Ron-
delet 1868) in which the strong horizontal forces of
the waves are held by a dovetail shaped interlock-
ing system that does not allow blocks to separate
from each other, creating a stereotomic traction sys-
tem; this idea was recently used by Jeanne Gang in
the Marble Curtain installation [4]. In Frei Otto's roof,
the structure relies in a carefully crafted network of
steel cables in tension, roughly following and gener-
ating the two main curvatures of the surface in an
intrinsic orthogonal layout. The longest and radial
cable direction becomes the guide for the continu-
ous rows of repetitive acrylic panels. The resulting
 grid of building elements is the stereotomy of this
structure, materialised in the acrylic panels that, al-
though not structural, are in a narrow relation with
the underlying steel grid. The direct relation between
force vectors and apparent division lines show its
stereotomic semantics, producing a consistent grid
consequential from the design intent and construc-
tion constraints down to its final appearance.
The Olympiastadion, together with other exam-
ples such as the Sydney Opera House, shows great
efforts towards the optimisation of construction el-
ements, producing a stereotomy that is bounded in
its time by the design and production possibilities
available. The advent of new technologies that en-
ables new design approaches is embodied by the
Valbonne vaults; the brothers Franque d'Avignon
turned their attentions to a more elaborate deco-
rative scheme in 1770 because structural stability
and fabrication drawings knowledge was mastered
enough and did not pose an obstacle to other design
intents. In a similar fashion, the introduction of the
computer in such processes means a gain in produc-
tion and accuracy, quantity and quality. This simple
fact, powered by algorithmic design and digital fab-
rication, broadens the gamut of possibilities as one
is not restrained to use repetitive modules in a com-
plex construction. A digital and contemporary inter-
pretation of classic stereotomy is considered, provid-
ing positive arguments in building a contemporary
stereotomic approach.
CASE STUDIES
The relation between digital technologies and an ex-
panded concept of stereotomy shall be explored in
three case studies. Curvature as a performance factor
is explored in the EMP Museum sheet metal façade;
the second example shows that decoration, or the ar-
chitectural language, drives the construction of the
concrete surface of Casa da Música; the last example
shows how history revision can play a decisive role in
a contemporary stereotomy that is driven by differ-
ent drivers only possible due to today's technologies.
430 | eCAADe 34 - SHAPE, FORM AND GEOMETRY | Applications - Volume 2
Casa da Música
OMA's project (1999-2005) for a concert hall in Porto
[2] is a concrete polyhedron which sits in the ground
with no immediate out a clear direction. This lack
or orientation is also found in the exterior enclosure
where there is hardly a vertical wall or horizontal di-
rection, except for the main windows. An important
role in the definition of the strange balance of this
building is played by stereotomy, helping create an
architecture expression by imprinting formwork lines
in its surface, simulating its own reality. This dialogue
between form and function in architecture can be
traced back to the old Egypt where capitals evoked
lotus, papyrus and palm trees, or Greeks who were
responsible for creating the classical vocabulary and
system. By Vitruvius' account, the logics of wood con-
struction were petrified; although some could con-
demn the misrepresentation of the constructive sys-
tem, classic language has been ever since one of the
main references in architecture design. The design
intent becomes as important than the constructive
system itself, lending to an expression system in it-
self.
Figure 1
1) Casa da Música's
natural appearance;
2) Construction
lines; 3) Imprinted
stereotomy.
Figure 2
Unfolded envelope
showing the
continuity of the
stereotomic
pattern.
Casa da Música's exterior walls are all built of concrete
cast in place [3]. The construction of the walls was
phased, because concrete has a limit on its height for
proper setting. Because concrete is a ductile mate-
rial while it is setting, horizontal lines appear natu-
rally in the top height of the concrete layers, regard-
less of formwork. The architects seem to have had a
different purpose for the appearance of the building,
denying its relationship to gravity and its horizontal
and vertical related vectors. Standard formwork pan-
els were chosen, creating a rectangular array. This
array of panels is aligned with a roof edge and from
there, it develops around the facets of the building
creating oblique directions which are aligned in an
intrinsic surface analysis but completely unaligned as
seen from the outside. A conflict is generated be-
tween the construction lines and the falsework (Fig-
ure 1), in a tension that clearly marks a stand from
the architect's point of view: the division lines appar-
ent in the surface ought to be oblique. The intention
of organising a prefabricated, mass produced false-
work, much like in the architecture of Tadao Ando,
is an expression of stereotomy for a subjacent idea
of organisation is implied by the apparent junctions.
These false work imprinted junctions are key to ex-
press the design intent in the architecture, gaining
prominence above the construction lines, for its ex-
SHAPE, FORM AND GEOMETRY | Applications - Volume 2 - eCAADe 34 | 431
pressive and intentional aesthetics, making the poly-
hedral nature of the project (Figure 2) change its own
stereotomy semantics.
EMP Museum
Gehry's EMP Museum (1995-2000) is covered in sheet
metal in a double curvature configuration. Sheet
metal does not admit double curvature, so the design
had to be rationalised into developable strips. Break-
ing down a curved surface to smaller parts which
bear a smaller curvature is an approach that can be
traced back to the earliest records on stereotomy
practices. The first forms of stereotomy relied in
shapes easily projectable to a plane, featuring vari-
ation in only one dimension, such as arches and bar-
rel vaults. The method for materialising this kind of
designs was mainly "par équarissement"; this process
relies in having two or more projections drawn in
a block of stone and then having the mason carve,
progressively from both sides until these meet. As
the discipline created more complex shapes such as
the skewed arch or arriére-voussure the transferring
"par équarissement" revealed its shortcomings as the
bounding box would be too big or the orthogonal
projections not sufficient to describe the complexity
of the angled shape.

Philibert de l'Orme introduced another method in his
"Le premier tome de l'Architecture" which allowed
for much more complex shapes to be worked out of
stone. The method, called "par panneaux", relies in a
different projection for each face of the voussoir by
rebating, which were complex geometric operations
for the 16th century. This achievement allows for true
size drawings in sheets to be put directly in the block
of stone as to carve progressively until these sheets
meet. The projection of each of the individual faces
(Figure 3) of a construction block - be it planar or not
- implies a target plane that should conform to a pos-
sibly curved surface. Depending on the ability of the
draftsman and stonemason, a high curved surface
might produce inaccurate results for its inaptness of
true size in direct projection into a plane.
The direct relation between division design and
materialisation possibilities creates a two way play, in
which the feasibility of imprinting a "panneau" is in-
fluenced by the design and the design is constrained
by the afore mentioned feasibility. This symbiosis is
also present in the EMP Museum sheet metal cover.
Seattle's music scene, embodied by Jimmy Hen-
drix, was perceived by Frank Gehry's design as an ex-
plosion of shapes and colours - these were embodied
in seven different volumes. The initial design foresaw
the facade built in shiny terrazzo as an analogy to the
lacquered finish of electric guitars. The possibility of
this volumetric stereotomy was discarded due to cost
issues, and it was replaced by a skin cover made of
sheet metal. Constraints in material dimensions and
curvature, as well as fixation technology were key in-
432 | eCAADe 34 - SHAPE, FORM AND GEOMETRY | Applications - Volume 2
fluences in the division design. The departing point Figure 3
was a grid of four-sided panels bounded by flow lines Projections of a
(curves of constant U or V), with maximum dimen- trompe's stone
sions of 10' x 15' and maximum square area of 100 voussoirs.
square feet (Shelden 2002). The first action was to
identify panel planarity through its mean gaussian
curvature: regions which fell under a pre-determined
tolerance were considered planar; others were subdi-
vided in smaller region so that fitness was acquired.
Once the surface was divided in regions whose cur-
vature was similar to planes - to a specified extend
-, a grammar was applied to further divide (Figure 4)
these according to maximum sheet dimensions (48"
x 96").
Figure 4
EMP Museum sheet
metal cover of one
of the volumes,
with analysis of
macro and micro
stereotomy.
The stereotomy design methodology of the surface
of the EMP Museum has two main moments: the divi-
sion into macro regions and the division within these
regions to create the final building blocks - micro
stereotomy. This final design could only be achieved
once the macro stereotomy could be unrolled so that
its design could be somewhat "imprinted", just as de
L'Orme's "panneaux". But instead of being bounded
by stonemasons capabilities and compression only
statics, Gehry was directed by the sheet metal prop-
erties and industry capabilities in deforming it to at-
tain curvature.
 Ponte Truchet
The Ponte Truchet (2007) by Giuseppe Fallacara has
in its design a clear idea: to transform the inter-
locking blocks flat vault seen in Frézier's work into a
ribbed vault by means of a topological transforma-
tion. The main source of inspiration for this vault it
is Jean Truchet, (1657-1729), a Carmelite man of sci-
ence, namely mathematics. One of his interests was
in tilings, and he registered a patent on a vault con-
struction system (Figure 5) which uses only one kind
of voussoir to attain doubly flat faces, perfecting the
previous Abeille design. A report published by the
French Royal Academy of Sciences praised the design
but argued that its construction would be too diffi-
cult for the varying curvature of the surfaces and the
need for a perfect fit.
Figure 5
Patent of Truchet's
original flat vault
design.
Figure 6
Fallacara's
reinterpretation of
the Truchet vault;
the intrados of two
voussoirs are
marked in red and
blue.
SHAPE, FORM AND GEOMETRY | Applications - Volume 2 - eCAADe 34 | 433
Fallacara saw this as an opportunity to put to test con-
temporary technologies. Computer driven multi-axis
machines make no distinction between straight lines
or highly varied curves. Not only a vault was fabri-
cated with this curved voussoir, but a deformation
was also applied to the initially flat vault, making it
into a barrel vault, a skewed barrel vault (Figure 6).
This is a subversive move, because the vault that was
composed of only one template that repeated itself
in all its extension was transformed into a structure
that is composed of unique voussoirs in half of it. Al-
though this could be understood as a drawback to-
ward the original Truchet design, it ought to be per-
ceived in its context: stone voussoirs are cheaper to
be produced by machines than by human labour. Be-
cause the design is parametric, and voussoirs were
not designed individually by hand, but automatically
through a set of rules; and stone working machines
do not work with templates or moulds but individu-
ally at each block, this design is utterly contemporary
although it links directly with the theoretical tradition
and finds in this historic design an opportunity to cre-
ate a vault that is decorative and structurally more ef-
fective than a simple barrel vault.
REVISING STEREOTOMY THROUGH DIGI-
TAL TECHNOLOGY
These examples provide an insight on how
stereotomy can have an alternative, yet in contin-
uum understanding to the classic theory. Despite the
elaborate and accurate nature of many stereotomic
designs produced in the 18th and 19th century, there
is a constant throttle in the complexity of these de-
signs, be it in the geometrical design, or the ability to
reproduce it physically with the necessary accuracy.
This throttle is deeply connected with the technol-
ogy available at the time which, although capable of
intricate designs, comes at a high human labour cost.
With the possibility of using the computer to
automate drafting tasks under specific rules, the
stereotomic design can achieve levels of perfection
not possible before; on the other hand, CNC ma-
chines allow us to fabricate the exact geometrical
 shape needed, instead of falling to La Gournerie's ar-
guments (Rabasa 2011).
Be it in structural elements, formwork, cladding
or even decoration purposes, a new kind of
stereotomy that depends on theory and reason as
much as it did three centuries ago is taking shape,
as the emergence of an expanded architectural mor-
phological freedom has not only taken place, but it
is already in its course. This new shapes have their
closest relative in stereotomy treatises, where clear
options are taken to its subdivision or regulation. An
analogy is not only possible but necessary, in order
to provide elements to evaluate and push forward
today's morphological possibilities.
Semantic factors such as the nature of materi-
als, the type of connections between parts, the forces
that connect them, planarity of surface, thickness of
the structure, density of subdivisions, all of these fac-
tors may somehow replace or augment classic needs
of stone stereotomy in a productive digital environ-
ment.
ACKNOWLEDGEMENTS
The work presented was co-financed by the
European Regional Development Fund (ERDF)
through the COMPETE 2020 - Operational Pro-
gramme Competitiveness and Internationalization
(POCI) and national funds by the FCT under the
POCI-01-0145-FEDER-007744 project, the previous
PTDC/ATP-AQI/5124/2012 research project, and the
SFRH/BD/93438/2013 PhD scholarship.
FIGURE CREDITS
Figure 1 - Photo from https://commons.wikimedia.org
/wiki/File:A_Casa_da_M%C3%BAsica.jpg and draw-
ings by the author.
Figure 2 - Drawing by the author.
Figure 3 - Delorme, P 1567, Le premier tome de
l'Architecture, F. Morel (Paris), p. 100.
Figure 4 - Original image changed by author.
https://commons.wikimedia.org/wiki/File:Razvan_-
Orendovici_-_EMP_SFM_Building_(4170362500).jpg
Figure 5 - Recueil des machines et inventions approu-
vées par l'Académie Royale des Sciences, 1699, p.
434 | eCAADe 34 - SHAPE, FORM AND GEOMETRY | Applications - Volume 2
163.
Figure 6 - Original image changed by author.
http://www.lithicdesign.it/portfolio/ponte-truchet/
REFERENCES
Diles, J 2014, 'Eigenforms', ACADIA 2014: Design Agency,
1, pp. 103-106
Fallacara, G 2003, Il paradigma stereotomico nell’arte del
costruire. Dalla natura sincretica della modellazione
digitale alla progettazione/costruzione di elementi ar-
chitettonici in pietra da taglio., Ph.D. Thesis, Politec-
nico di Bari
Fallacara, G 2009 'Toward a stereotomic design: Experi-
mental constructions and didactic experiences', Pro-
ceedings of the Third International Congress on Con-
struction History, p. 553
Frézier, AF 1738, La théorie de la pratique de la coupe
des pierres et des bois, pour la construction des voûtes
et autres parties de bâtiments civils et militaires ou
traité de stéréotomie à l'usage de l'architecture, J.D.
Doulsseker le fils
de l'Orme, P 1567, Le premier tome de l'Architecture, F.
Morel (Paris)
Otto, F, Trostel, R and Schleyer, FK 1973, Tensile struc-
tures; design, structure, and calculation of buildings
of cables, nets, and membranes, The MIT Press, Cam-
bridge, Mass
Rabasa, E 2011, 'La Gournerie versus Monge', Nexus Net-
work Journal, 13(3), pp. 715-735
Rippmann, M and Block, P 2010, 'Digital Stereotomy :
Voussoir geometry for freeform masonry-like vaults
informed by structural and fabrication constraints
Interdependent Geometrical Constraints', Proceed-
ings of the IABSE-IASS Symposium 2011, 9, p. 0
Rondelet, JB and Blouet, GA 1868, Traité théorique et pra-
tique de l'art de bâtir, Librairie de Firmin Didot, Frères
Shelden, DR(R 2002, Digital surface representation and
the constructibility of Gehry's architecture, Ph.D. The-
sis, Massachusetts Institute of Technology
[1] http://www.tensinet.com/database/viewProject/37
79.html
[2] http://oma.eu/projects/casa-da-musica
[3] http://www.afaconsult.com/portfolio/120921/127/
concert-hall-casa-da-musica
[4] http://studiogang.com/researchproject/marble-cu
rtain
SPATIAL ANALYSIS | Urban Context
Mapping Planned and Emerging Art Places in Singapore
through Social Media Feeds
Ludovica Tomarchio1 , Bige Tuncer2 , Linlin You3 , Bernhard Klein4
1,2
ETH Zurich, Future Cities Laboratory, Singapore-ETH Centre (SEC), Singapore
and Architecture and Sustainable Design, Singapore University of Technology and
Design (SUTD), Singapore 3 Architecture and Sustainable Design, Singapore Uni-
versity of Technology and Design (SUTD), Singapore 4 ETH Zurich, Future Cities
Laboratory, Singapore-ETH Centre (SEC), Singapore
1,4
{tomarchio|klein}@arch.ethz.ch
2,3
{bige_tuncer|linlin_you}@sutd.edu.sg

This paper presents a methodology to collect and visualize social media data
about art, in order to map art locations in cities using geo-localized data, and
comparing planning decisions with the actual use of spaces. As various social
networks have penetrated into the daily life of people, these become one
important and effective data source to understand how people perform 'arts'
around the city (Shah 2015). The case study for this methodology is Singapore, a
vibrant city where art and culture are being promoted in the light of an emerging
creative economy. The Singapore government promotes art and creates 'art
clusters', such as art districts, galleries, fairs and museums in the city.
Additionally, artists, creative entrepreneurs, consumers, and critics seek and
explore alternative spaces. Understanding where art and creativity are discussed,
broadcasted and consumed in Singapore is a key point to have better insights into
art space planning, and study its effects on the city.The paper will try to answer
the following research question:Is it possible to discover, through social network
data, spaces where art is produced, discussed, and broadcasted to an audience in
Singapore? How?

Keywords: social- media, art, creative city, creative places

INTRODUCTION have a specific cultural policy, promoting museums

Municipal planning policies are concerned with the
diffusion of culture in the city and its spatial localiza-
tion. The idea of the cultural city and cultural class,
promoted among others by Richard Florida (Florida
2005), has encouraged city-planning departments to
and cultural facilities. There is also the belief that
those facilities together with studio spaces and other
institutions should ideally be close one to the other;
the terms "arty neighbourhood", "cultural quarter"
(Roodhouse 2010) or "cultural cluster" (Cooke and

SPATIAL ANALYSIS | Urban Context - Volume 2 - eCAADe 34 | 437

 Lazzeretti 2008) indicate the area's commitment to
cultural development, the ability of actual artists to
live in the area, the access to inspirational cultural in-
stitutions or even the actual art economy that flows
through the area. The different facilities benefit from
the mutual proximity and the "art consumer" or artist
find a spatial continuity among the places she is inter-
ested in.
American sociologist Lyn Lofland defined
"parochial sphere" (Lofland 2006) as space "where we
meet like-minded people", and "who are involved in
interpersonal networks that are located within 'com-
munities' ". The idea behind "cultural clusters" is that
it is possible to recreate a parochial domain within a
neighbourhood revolving around art production and
consumption. This idea of "cultural cluster" is also
connected to the general goal of neighbourhoods'
regeneration and future gentrification (Roodhouse
2010, Cooke and Lazzeretti 2008).
While city planners make decisions based on
these theories, there is a novel way in which peo-
ple get to know and decide to go to art happenings:
social media. Using again the previous definition of
"parochial domain", through the use of social media,
like-minded people get in touch and then navigate
the space, regardless of spatial continuity, going di-
rectly to events that might interest them. Artistic pro-
duction and consumption do not need to have a spa-
tial continuity anymore. The idea of the "cultural clus-
ters" is not effective, together with the planning poli-
cies that, at different stages, in different cities, with
different strategies, work toward it.
The paper aims at exploring a methodology to
locate art venues and neighbourhoods where art is
produced, consumed and broadcasted, with crowd-
sourced data (social media data). The methodology is
applied to the case study of Singapore, a vibrant city
where art and culture are being promoted in the light
of an emerging creative economy. The final result is
a visualization platform, which can support decision-
making processes about cultural planning in the city.
In this paper, we intend to discuss not the results of
the case study application, rather the methodology
438 | eCAADe 34 - SPATIAL ANALYSIS | Urban Context - Volume 2
applied to collect, filter and analyse the data.
CASE-STUDY: SINGAPORE AND PLANNED
ART SPACES
In the last five years, Singapore's central government
actively nurtured and supported the art scene, pro-
viding grant schemes, art spaces, and art facilities.
Singapore government created data resources on the
website of National Art council (NAC) that includes
lists of recipients of grants and the location of 'art
spaces' such as facilities (studios, stages, spaces for
rehearsal, galleries) offered to artists. NAC defines
art as "Dance, Literary art, Music, Theatre, Visual Art
and Performance Art" [3]. For the sake of consistency,
only those forms of art listed by NAC will be consid-
ered in this research.
The government awards major grants to art in-
stitutions and universities, but also small galleries
to support their operating costs. Production/devel-
opment/marketing grants are awarded to individual
artists to support their personal research and career
in Singapore and abroad [3]. NAC has also an agenda
developed in concertation with the Urban Redevel-
opment Authority (URA) to provide artists with infras-
tructure in the city.
In particular, these 'art spaces' include:
• A "Framework for Art Spaces" (2010) to pro-
vide heavily subsidised housing and studio
spaces inside Singapore Land Authority prop-
erties
• Some shared facilities, owned or subsidised
by the government as spaces to perform, re-
hearse, and produce different forms of art
Together, NAC grants and dedicated URA spaces cre-
ate a very attractive and internationally competitive
package for emerging and established artists. The ra-
tionale behind this governmental measure becomes
even more evident as art precincts are developed
to regenerate certain areas [3]. The 'art belts' have
an urban agenda, with the purpose of creating spa-
tial continuities among art-related buildings and in-
crease real-estate values.
However, as pointed out in a local newspaper [4], boundaries to represent art belts, meaning the ad-
these policies do not always create the expected out- ministrative sub-zones. These boundaries are in fact
comes. Art districts strategically located in the city the contemporary planning tools used by the plan-
failed to attract both artists and costumers, forcing ning administration, thus representing the current
some international galleries to close. city planning practice.
Additionally, some art is by definition critical.
Other art deliberately avoids paternalistic support. EMERGING ART SPACES IN SINGAPORE
These artists continue to live and practice in Singa-
MAPPED BY SOCIAL MEDIA DATA
pore, probably relying more on informal and social
According to Evans and Foord (2008) there is very lit-
networks to stage and produce their work, rather
tle official or robust data on cultural activity and pro-
than institutional ones. This context creates a frame-
visions in cities. They draw a methodology of map-
work to justify a different methodology to map art in
ping art related facilities based on surveys, to be im-
Singapore.
plemented in a GIS platform. Among the different
In a first step, we examine and map these spaces
cultural mappings undertaken by planning authori-
that, according to Singapore central government and
ties very little has been shared and made public to
NAC [3] are considered, supported, and suggested as
a wider audience. An interesting reference is Cul-
'art places'. The mapping is done in two scales: build-
tureMap London (2004), an interactive resource that
ings and clusters. Art venues are buildings where
maps data on cultural engagement across the capital,
art is produced, consumed or broadcasted. These in-
originally funded by Arts Council England. It provides
clude studio spaces, rehearsal rooms, museum, and
some insight into the audience of cultural facilities in
curatorial offices. When art venues are close one to
the English capital. CultureMap uses Audiences Lon-
the other we talk about "cultural clusters'.
don's extensive data sets about cultural users, includ-
The mapped art venues consist of:
ing demographics, costs of tickets, and surveys [5].
The possibility to include crowd-sourced data to
• studio spaces allocated by the government to
map art venues has been explored by "Geography of
the artists;
buzz: Art, Culture and the Social Milieu in Los Ange-
• small- medium galleries economically sup-
les and New York" [2] a research by Elizabeth Currid
ported by the government, present in a list;
from the University of Southern California and Sarah
• small- medium exhibition spaces mapped by
Williams from Columbia University. The authors col-
the government;
lected 300.000 snapshots taken by photo agency
• museums supported or mapped by the gov-
Getty Images at more than 6,000 parties, openings
ernment;
and premieres in 2006 and 2007 and conducted GIS
• theatres supported and mapped by the gov-
and spatial statistics to provide a new spatial dimen-
ernment;
sion through which to understand cultural industries
• rehearsal spaces/recording spaces supported
and city geographic patterns.
by the government;
Fortunately, with the recent boom of social me-
• art university- education facilities mapped by
dia, users have reported a considerable number of
the government;
activities (i.e., check-ins, comments, pictures ) on-
• foundations or private collections mapped by
line. Using social media data, scholars have been
the government. [3]
able to map city users' behaviours and patterns: from
community neighborhood based on cultural affinity
Cultural clusters include the government 'art belts'.
(Cranshaw et al 2012; Yang, Zhang, and Qu 2016)
The size and extent of the art belt might vary, how-
to users' mobility patterns, from semantic interpre-
ever, in this paper we will refer to the planning

SPATIAL ANALYSIS | Urban Context - Volume 2 - eCAADe 34 | 439

 tations (Simeone et al. 2012) or city users' analysis
(Morandi et al 2013) to sentiment analysis (Ceron et
al. 2014), among others.
For the case study of Singapore, we can ex-
tract valuable patterns and narratives from the data
shared via social networks and digital devices, focus-
ing in particular on art venues. The data collected via
social media has been filtered with a word-filtering
process and visualized in a GIS platform (Figure 1).
The social media data are collected through open
APIs of social networks Instagram and Twitter. In-
stagram is a mobile application that enables online
photo sharing. It also allows text annotations. Users
take a picture, apply a digital filter to it, and share
it, usually with the location of where the sharing oc-
curred unless this function is specifically disabled.
Therefore, a majority of these feeds are geo-tagged.
Twitter is a micro-messaging application. The shared
content via Twitter is short messages shared with
followers. Twitter feeds are not automatically geo-
440 | eCAADe 34 - SPATIAL ANALYSIS | Urban Context - Volume 2
tagged, therefore the majority of feeds do not con-
tain a location. We only considered feeds that have
a geo-location and a text component. We omitted
feeds that do not contain a geo-location, or consist
only of images.
We selected Instagram and Twitter as our data
sources, because more than 2 million people are us-
ing Instagram and Twitter in Singapore [1], and these
two social media, together with Facebook, are the
most-used social media in the field of art [1].
In order to filter the social feeds, considering only Figure 1
the ones "talking" about art, we deployed a word fil- The project
tering process. This filtering process relies on 330 components, and
words (Table 1) that have been selected by analysing flow of the
more than 2000 art feeds in Singapore, posted by methodology.
public institutions as well as by private artists and col-
lectives. These art feeds have been processed by a
programme, which selected the most frequent words
that occurred in these.
The word list includes words in the four official
languages of Singapore: English, Malay, Mandarin,
and Tamil. The words are organised in three differ-
ent categories. The first list includes single words that
characterize the feed as an "art feed". The second list
includes word pairs: both words need to be present
in the feed. The final list also includes word pairs,
however, this time each pair includes words which
should not appear together in the text. We have im-
plemented a program that first scans each social me-
dia feed, and selects the ones that include any of
the words in the single word list. Then, the program
scans the remaining feeds for the word pairs in the
second list, and selects the feeds where both words in
a list item appear in a feed. Finally, the program filters
the collection of all selected feeds, and scans these
using the word pairs in the third list, eliminating the
feeds that contain both words in a list item. Addition-
ally, the program considers hashtags that are usually
part of the comments.
The data resulted in 8500 localised feeds about
art in the 7 days of observation (14-20 November
2015).
We performed accuracy studies to determine the
 reliability of this approach. The average accuracy is
75%. The presented method thus ensures a high
probability of detecting art related feeds in the con-
text of Singapore.
Having a one week data collection period is un-
doubtedly insufficient. However, in order to develop
the methodology and implement the analysis and vi-
sualization infrastructure for mapping the art scene
in Singapore, this approach forms a sufficient foun-
dation. Deriving conclusion regarding the planning
policy cannot be the purpose at this stage.
RESOLUTION OF THE ANALYSIS, SCALAR
AND TEMPORAL PARAMETERS
When analysing and interpreting social media data
it is important to consider the proposed interpreta-
tion as local and instantaneous; local because the
possible narratives extracted by the data depend on
the types of software people use (that changes from
place to place) and how they use it; instantaneous
because creating feeds relies on commercial soft-
ware, with many active users, and the methodology
changes together with the market following the way
people use their digital devices. Local culture could
deeply affect what a click means and how to read it.
In this study, we considered Instagram and Twit-
ter, two commercial platforms quite widespread
among the Singaporean society, and particularly
used by the artist community. In both of these plat-
forms, location is not the main purpose of the social
interaction provided by the software. As previously
explained, the users decide to share short messages
and pictures and they can choose to embed the feed
with latitude and longitude information.
Thus, users present a more implicit will to com-
municate a message about a place: the user's loca-
tion is not the main focus of her social interaction,
it is a somehow a side effect. This circumstance is

Table 1
List of English
Words for the art
related
word-filtering
process of the
feeds.

SPATIAL ANALYSIS | Urban Context - Volume 2 - eCAADe 34 | 441

 particularly meaningful, as these kinds of digital foot-
prints on the city can be considered as a more spon-
taneous pattern of the truly most attended places by
smartphone-equipped dwellers: these are the real
venues where people spend their urban lives and
from which they socially interact.
In the context of art, we interpreted the art
feeds venues as locations where art is produced, con-
sumed, and broadcasted. For example, many vi-
sual artists, while working in their studios, decide
to share the work in progress or the results of their
production: studio space, rehearsal spaces, record-
ing spaces, thus are included in our social media
mapping and are considered as the locations where
art is produced. On the other hand, museums, the-
atres, events, often offer the chance to art lovers to
make pictures or comments on the work of art they
are watching: those mapped contributions are the
places where art is consumed. But more and more
artists, art institutions, and events rely on public rela-
tion offices, probably working together with the cu-
ratorial team, often in charge of taking care of the
media communication. Thus, the mapping process
will include also all the locations where art is broad-
casted. With the purpose of social media mapping,
therefore, we include all the locations with sufficient
numbers of art feeds. These locations will be inter-
preted as art venues: places to produce, broadcast
and consume art.
Some separate observations regarding art
events include temporary occupancy of buildings
or open spaces for art-related happenings. The art
events are included in the mapping, as they are po-
tentially spotted by the social media word filtering
like any other art contribution. Their location, even
if temporary, is equally meaningful for the analysis,
as it is usually hosted in buildings or infrastructure
normally equipped for this kind of events, thus rep-
resenting a venue of interest.
The data collected have been visualized in a GIS
platform, creating an online map. Initially, we used a
commercial GIS to perform some analysis, but the fi-
nal representation has been done in Carto DB [6], an
442 | eCAADe 34 - SPATIAL ANALYSIS | Urban Context - Volume 2
open software for the representation of geographi-
cal data where it is possible to perform some analysis
through the Structure Query Language (SQL).
The layers in the map act as chapters of the con-
structed narrative. We try to establish the mutual re-
lationship among the places in Singapore supported
by the government as places of art, and the emerg-
ing ones. In order to understand the relationship be-
tween art venues and the contiguousness in space,
we worked at several spatial scales: individual build-
ings (such as galleries and museums) and districts.
We aggregated the data in buildings of Singa-
pore (Open Street Map)[7] and sub-administrative
boundaries [8]. In order to consider the different
types of utilization of the art facilities, we have con-
sidered different time-stamps: day, night, weekends
and weekdays. Roughly we could approximate stat-
ing that the venues where people tend to go during
the weekend could be considered more as locations
where art is consumed while the working days contri-
butions include also the spaces of productions. Sim-
ilarly, the locations active at night tend to be more
theatre or events locations, thus, places where art
is consumed, while the daily contributions present
a whole range of possible interpretations(museums,
rehearsal spaces, galleries, exhibitions spaces, studio
spaces).
The social media data is presented overlaying the
buildings and venues, considered by the NAC and the
URA officially as 'art spaces'. In our interpretation,
each official art building has an area of influence of
30m radius, in order to aggregate social media data.
This value has been selected considering the preci-
sion of the GPS and localization signal of social media
feeds.
VISUALIZATION PLATFORM
Trying to abstract and synthesize the enormous num-
ber of potential outcomes of these collected data, we
identified three main categories:
1. Discovery: The social media analysis reveals
the existence of certain art venues that are not
mapped by the government (Figure 2)
 2. Confirmation: The social media analysis spots
a series of venues actually corresponding to
those mapped by the government
3. Negation: The social media analysis reports
no or little utilization of the place referred to
as a certain art venue, supported by the gov-
ernment
Assessing confirmation, negation and discovery
depends on the statistical evaluation of the current
data. In particular, confirmation and negation is calcu-
lated considering the initial mapping of the officially
planned art venues, while in discovery the feeds are
aggregated on all the buildings and administrative
boundaries of Singapore. The medium value to de-
fine a confirmation or negation on the buildings de-
pends on their size. Smaller buildings score confir-
mation also with a smaller number of contributions,
scaling the threshold.
As an outcome of the map, it is possible to navi-
gate among confirmation, negation and discovery in
the scale of the districts and the buildings, and to
comment or understand the localization of art in Sin-
gapore. Each user can browse the data in the on-
line visualization in a graphic way, but also access-
ing the statistics and analytical representation of the
data. Comparing the two representations it is possi-
ble to extract not only observations on the actual art
life of buildings and neighbourhoods, but also infor-
mation about the planning strategy in the different
locations.
We created a simple metric to value to suc-
cess of the planned strategy for each district, de-
pending on the number of confirmation and nega-
tion, (Planned= (Conf irmation − N egation) ·
10
) and another metric
Conf irmation + N egation
to value how many buildings in the area deal with
art without consistent support from the government
(UnPlanned=(Discovery − Conf irmation)). The
first metric scores from +10 to -10; the maximum
score indicates that the art venues planned by the
NAC and URA, present in the observed district, show
a significant number of art contributions. This gen-
SPATIAL ANALYSIS | Urban Context - Volume 2 - eCAADe 34 | 443
erally means that the cultural planning strategy, re-
garding the buildings, is quite successful. Negative
values indicate a high number of Negations, thus the
planned art buildings don't have a consistent num-
ber of art feeds; this could be further investigated
looking into the reasons of the negative value, proba-
bly depending on accessibility or integration among
other factors. The Unplanned metric is positive with
a high score when the value of discovery is relevant,
meaning several buildings not supported by the gov-
ernment in any way, host an art related activity. The
two metrics together give some indications about
the district and how different art-related facilities in-
teract.
The visualization also helps in understanding the
different typologies of buildings on which the con-
tributions are aggregated. Combining the analytics
and the maps we aim at having a detailed represen-
tation of cultural life in Singapore.
POSSIBLE OUTCOMES
This paper does not aim to analyse the results, how-
ever, it is interesting to give an overview of the poten-
tials of this type of analysis. In particular, the obser-
vations presented here only showcase the possibili-
ties of this mapping, without providing any real in-
terpretations considering the limited amount of data
we are currently analysing.
Table 2 presents some metrics associated with
some administrative boundaries of Singapore as an
outcome of the visualization. It is already possible to
observe or at least to guess some patterns and trends
regarding the cultural life of Singapore.
For example, Gillman Barracks is a planned cul-
tural cluster of Singapore. With the support of gov-
ernmental subsidies the colonial architecture area
hosts several galleries, exhibition spaces, workshops
and spaces for rehearsal, and also the Centre for Con-
temporary Art of Singapore (CCA) research-based ex-
hibition space, connected to the Nanyang Technol-
ogy University of Singapore. We can see how the
general contribution is not too high with more art
feeds during the day rather than the night. This is
 Figure 2
Art in Places Map/
Showing Art Feeds,
Discovery Layer.

Table 2
List of some
Singapore
sub-administrative
boundaries with the
scores performed in
the analysis related
to their cultural
planning.

444 | eCAADe 34 - SPATIAL ANALYSIS | Urban Context - Volume 2

probably due to the fact that there are more stu-
dio spaces and exhibition spaces rather than perfor-
mance spaces. The value of the Planned metrics is -2,
negative, meaning that among the planned facilities
we count more negations than confirmations. Mean-
while, the value of Unplanned is zero, meaning that
there is no building not supported by NAC, hosting
art related activities. Somehow the introduced pro-
gramme by the government did not activate any sort
of spontaneous utilization in the nearby area.
The second area is Marina Area, which includes
the iconic theatre Esplanade that hosts national and
international performances, and Marina Bay Sands
Exhibition space. The general level of art feeds is
the highest among the different neighbourhoods
and there is the highest trend during the weekend.
The Planned metric scored 10, the maximum score,
meaning all the art facilities have a substantial con-
tribution. However, the level of Unplanned is again
0: no discoveries have been found in the area. Some-
how Marina Bay Sands and the Esplanade are iconic
buildings that work at the scale of single attractions
rather than art clusters.
Somerset is a commercial area with shopping
malls with some art installations in controlled envi-
ronments and some art facilities and theatres. It has a
high number of contributions and generally confirms
the planned buildings by NAC and URA. However,
there are also some discoveries: they are still aggre-
gated in the shopping malls and probably they are
art related events hosted in the commercial strips to
attract more visitors. Generally, in terms of consump-
tion of art, this mapping shows interest for "side-
effect" art locations: art located in buildings with
a different programme, such as shopping malls or
the airport, sometimes have more contributions that
dedicated spaces like museums or galleries.
Kampong Glam is the Arab quarter of Singa-
pore: it is an area, which is currently undergoing a
transformation, generally hosts parties and leisure
events and probably still enjoys low rental prices. The
Planned level is high confirming the success of all
the facilities located there, but also the Unplanned
SPATIAL ANALYSIS | Urban Context - Volume 2 - eCAADe 34 | 445
score is high showing several discoveries distributed
in the neighbourhood which aggregate mainly in
small shop-houses, probably studio spaces or inde-
pendent exhibition spaces.
Following this logic, it is possible to classify and
comment on the different areas. This paper's goal is
not to interpret the data but only to exemplify the
current methodology. For further interpretations, we
wish to combine real-time and cumulative social me-
dia data with other datasets such as rent prices and
demographic data.
CONCLUSION
The paper presents a new methodology in finding
creative quarters and creative hubs, which are the
subject of policy interventions and public-private in-
vestment. This kind of work has a specific value for
the city of Singapore, but generally is a valuable re-
source in order to value planning policies and strate-
gies affecting the art production and its spatial ef-
fects in cities.
The data released by the API of the two software
platforms is public, meaning the users did not decide
to limit the accessibility to the data. There are some
ethics concerns about the use of personal data for the
purpose of analysis, but the final visualization consid-
ers the data already aggregated, representing each
user only with their location. Under those circum-
stances, their privacy is, in all cases, maintained.
Currently, the data has been collected for one
week. In order to have a wider portrait of art spaces
in Singapore, we will extend this study and use real-
time data. However, we intend this study as a qualita-
tive research, with the purpose of defining a method-
ology for the use of data coming from social media
to extract information about cultural planning in the
city.
The current methodology will be improved also
regarding the detection of the art related feeds. Con-
sidering Instagram, further studies will go in the di-
rection of image sensing to understand the full range
of information provided by the feed: the comments
and the pictures.
 The methodology and the visualization pre-
sented here also offer an interesting case for further
considerations on cultural planning and how much
it is actually possible to plan and control the cultural
developments of our cities.
ACKNOWLEDGEMENTS
The research was conducted at the Future Cities Lab-
oratory at the Singapore-ETH Centre, which was es-
tablished collaboratively between ETH Zurich and
Singapore's National Research Excellence and Tech-
nological Enterprise (CREATE) programme.
REFERENCES
Ceron, A, Curini, L, Iacus, SM and Porro, G 2014, 'Every
tweet counts? How sentiment analysis of social me-
dia can improve our knowledge of citizens' political
preferences with an application to Italy and France',
New Media & Society, 16, pp. 340-358
Ciuccarelli, P, Lupi, G and Simeone, L 2014, Visualizing the
Data City: Social Media as a Source of Knowledge for
Urban Planning and Management, Springer, Cham
Cooke, P and Lazzeretti, L (eds) 2008, Creative Cities, Cul-
tural Clusters and Local Economic Development, Ed-
ward Elgar, Cheltenham, Glos, UK; Northampton,
MA
Cranshaw, J, Schwartz, R, Hong, JI and Sadeh, N 2012 'The
Livehoods Project: Utilizing Social Media to Under-
stand the Dynamics of a City', 6th International AAAI
Conference on Weblogs and Social Media, Dublin
Evans, G and Foord, J 2008, 'Cultural mapping and sus-
tainable communities: planning for the arts revis-
ited', Cultural Trends, 17, pp. 65-96
Firminio, JR and Graham, S 2001 'Typology for virtual
cities: the interplay between physical and virtual ur-
ban spaces.', EDA 2001 – European Digital Architect,
Prague
Florida, R 2005, Cities and the Creative Class, Routledge,
New York
Lloyd, R 2002, 'Neo-Bohemia: Art and Neighborhood Re-
development in Chicago.', Journal of Urban Affairs, 24
(5), p. 517–32
Lofland, LH nd, The Public Realm: Exploring the City's
Quintessential Social Territory, Transaction Publishers
Manovich, L 2012, 'Trending: The Promises and the Chal-
lenges of Big Social Data', in Gold, MK (eds) 2012, De-
bates in the Digital Humanities, The University of Min-
nesota Press
446 | eCAADe 34 - SPATIAL ANALYSIS | Urban Context - Volume 2
Morandi, C, Palmieri, R, Stojanovic, B and Tomarchio,
L 2013 'Digital mapping: the analysis of the social
realm of Urbino', NUL-New Urban Languages, Milan
Morandi, C, Palmieri, R and Tomarchio, L 2015, 'Digital
Social Media to Enhance the Public Realm in Historic
Cities', in Aiello, LM and McFarland, D (eds) 2015, vol.
8852, Lecture Notes in Computer Science, Springer In-
ternational Publishing
Roodhouse, S 2010, Cultural Quarters Principles and Prac-
tice., Intellect, Bristol, UK; Chicago, USA
Shah, RR, Shaikh, AD, Yu, Y, Zimmermann, W and Wu,
G 2015 'EventBuilder: Real-time Multimedia Event
Summarization by Visualizing Social Media.', 23rd An-
nual ACM Conference on Multimedia Conference, pp.
185-188
Simeone, L, Lupi, G and Iaconesi, S 2012 'Polyphonic im-
ages of the cities. Mapping new human landscapes
through User Generated Content', Northern World
Mandate, Cumulus Helsinki Conference., Helsinki
Yang, D, Zhang, D and Qu, B 2016, 'Participatory Cul-
tural Mapping Based on Collective Behavior Data in
Location-Based Social Networks', ACM Transactions
on Intelligent Systems and Technology, 7, pp. 1-23
[1] http://wearesocial.sg/blog/2015/01/digital-soci
al-mobile-2015/
[2] http://www.nytimes.com/2009/04/07/arts/design/0
7buzz.html?em&_r=0
[3] https://www.nac.gov.sg/
[4] http://theartnewspaper.com/market/five-gallerie
s-to-close-in-singapore-gillman-barracks/
[5] http://www.culturemaplondon.org/
[6] https://cartodb.com/
[7] https://www.openstreetmap.org/#map=5/51.495/
-7.844
[8] https://data.gov.sg/
Social Media as Complementary Tool to Evaluate Cities
Data Mining Innovation Districts in Boston

Nai Chun Chen1 , Takehiko Nagakura2 , Kent Larson3

1,2,3
Massachusetts Institute of Technology
1,2,3
{naichun|takehiko|kll}@mit.edu

High tech industries are playing an important role in the economic development
in the United States. While some cities are shrinking, the "innovation" cities are
growing. The attributes that cause some cities to successfully become innovative
is a very relevant 21st century topic and will be investigated here.Previous work
conduct city analysis through conventional government GIS or census data but
such analyses do not answer questions about the perception of citizens inhabiting
the city, and the activities they conduct. The novelty of this current project is to
make use of large-scale bottom-up data available from social media. Several
social media sources-CrunchBase, Twitter, Yelp, and Flickr- were data mined
pertaining to four innovation districts in Boston. We found that the success of
innovation districts in Boston were correlated with several important variables:
the most successful districts tended to occur near research institutions, in very
"mixed use" areas, and were unexpectedly not correlated with land and labor
prices, unlike technology districts in the past. Based on our study, we make
recommendations for the urban design that cities should put in place to increase
the potential for "innovation".

Keywords: Smart Cities, Social Media, Innovation District, Spatial Analysis,

Data Mining, Natural Language Processing

INTRODUCTION elements that work and ones that do not. A new

Why Innovation Districts? A remarkable shift is cur-
rently occurring in the spatial geography of new and
existing cities: the emergence of new high tech cities
or so called "innovation cities"(Simmie 2011). To cap-
italize on these untapped economic opportunities,
city planners and urban designers are beginning to
incorporate the unique demands by innovating cities
into city designs. However, such efforts have had
mixed success, and planners remain unsure of design
urban form is emerging, and gives rise to what we
are calling "innovation districts." Hence, "What are the
attributes of urban districts where innovation is oc-
curring?" becomes a critical question for urban plan-
ners. Some analyses have been done by previous re-
search, but most of them only focused on govern-
ment data or company surveys. We propose to in-
vestigate web-based data from social media in order
to generate broad and analytical perspectives of the

SPATIAL ANALYSIS | Urban Context - Volume 2 - eCAADe 34 | 447

 uses of urban spaces, for finding novel attributes in
innovation districts and evaluating the districts from
the perspective of residents, which helps urban plan-
ners design a better one.
Previous Work
Studying the attributes contributing to the success
of the Innovation districts is an important topic, and
some work has already been done to understand
these attributes. For example, "Innovating the City:
Challenges and Opportunities in establishing incu-
bators and Districts in Paris and Boston" is a Mas-
ter's Thesis in Urban Planning at MIT that is related to
this topic (Johnson 2014). This thesis offers insights
into the unique political, economic and cultural sys-
tems that shape innovation strategies in innovation
cities. But, in order to give translatable and readily
deployable advice for urban designers in Innovation
districts, one must delve further into understanding
the city attributes that facilitate innovation.
Particularly lacking in the study of the Innova-
tion district is an understanding of how residents per-
ceive the changes. Whether citizens feel benefited di-
rectly, and a deeper understanding of how they have
benefited, is unknown. This is due to the inherent
limitations of the types of data used so far. In the
Johnson study above, interviews, though in-depth,
are still limited in scope as they occur only to the
top companies participating in a project. Thus, they
project only one point of view. Other studies appear
to be restricted by the types of data that the govern-
ment has collected (Ciuccarelli et al. 2014). The cur-
rent project attempts to make important use of large
scale bottom-up data available from social media.
DATA AND METHODS
In this project, we utilized geo-located tweets from
Twitter, pictures from Flickr, and the density of
amenities from Yelp. We developed and used algo-
rithms to extract data from these social media (Kumar
et al. 2014).
448 | eCAADe 34 - SPATIAL ANALYSIS | Urban Context - Volume 2
CrunchBase
CrunchBase is a Web 2.0 platform that has become
increasingly popular with new businesses. It allows
startup companies to register, and update their com-
pany data, location, categories, launch date, em-
ployment data, etc. Using the CrunchBase Web API,
we compiled information regarding startups in the
Boston region, and applied visualization techniques
to examine the spatial patterns of high tech startups
in the Boston area.
Yelp
Yelp is a company that publishes crowd-sourced re-
views about local businesses, and was thus, an im-
portant source of information for us regarding the
spatial distribution of diverse types of businesses and
amenities.
Twitter and Flickr
Twitter and Flickr are both social networking services
that allow users to send and receive messages and
pictures, respectively. The key attribute is that all
of these public information are geo-located (Russell
2011). We also recorded timestamps of "tweets" on
Twitter, allowing us to understand the different lo-
cal temporal distributions of activities in the innova-
tion districts. Finally, we applied natural language
processing and sentiment analyses to the tweets.
Analyses of image tags from Flickr revealed that the
pictures uploaded in different innovation districts
tended to vary according to localized urban focus.
Application
These different social media sources were mined to
reveal urban issues such as the locations of improp-
erly used or underused land areas. The richness of
the data enables us to make inference about the rea-
sons behind such unpopular spaces. This social me-
dia data thus shows the image of the city from the
point of view of residents and allows urban designers
and government officials to solve these urban prob-
lems (Lynch 1960).
 ANALYSIS
Mining and gathering data from web-based API in
each district will indeed yield an enormous amount
of data. Thus, the next difficult technical challenge is
how to treat all of this data to make meaningful con-
clusions.
Crunch-Base as Innovation Trend Detector
Based on the data collected from CrunchBase, we
were able to apply visualization techniques to exam-
ine the spatial patterns of high tech startups in the
Boston area. Several findings can be seen from our
visualizations.
In general, it can be seen that traditional tech
companies in the 20th century were located in sub-
urban areas to take advantage of cheaper land and
wages. However, new high tech companies are more
often located in the city center in the Greater Boston
area, rather than the suburbs. We speculate that this
is in order to attract young talent graduating from
universities in the city. This observation also suggests
that the transformation of local industries through
the generations have also changed the preferences
of company locations from sub-urban to the city.
If we zoom into the Boston and Cambridge area,
we may see several of these patterns in greater detail.
Proximity to MBTA. Data visualization from Crunch-
Base reveals the relationship between startups, their
headquarters, and the MBTA. These startups tend to
be located close to MBTA stations, such as Kendall
Square, Central Square, Financial District, and the
South Innovation District. This patterns occurs pre-
sumably because the company is then able to attract
labor forces commuting by public transportation sys-
tem (see Figure 1).
Locations of the "Big" high Tech Companies. The
location of "big" tech companies, such as Facebook,
Google, and Microsoft, seem to have also become a
critical factor influencing smaller startups when they

Figure 1
Distribution pattern
of high tech
companies, based
on geolocated
company data from
CrunchBase.
(Startup companies
in cyan and
headquarters in
dark blue.)

SPATIAL ANALYSIS | Urban Context - Volume 2 - eCAADe 34 | 449

 Figure 2
Distribution of start
up companies, and
biotech companies
in green.

Figure 3
South Innovation
District and Kendall
Square have a
smaller proportion
of civic facilities
such as restaurant
(purple) and
grocery (green)
than Harvard
Square and
Financial District.

450 | eCAADe 34 - SPATIAL ANALYSIS | Urban Context - Volume 2

choose their locations. There are two possible rea-
sons: First, there could be a "flow of labor" from one
campus to the smaller ones. Many members of these
new startups might have been former employees of
the large tech companies. Second, there could be a
"flow of ventures." The big tech companies tend to
invest in startups that they better understand; local
startups thus have an advantage in these critical so-
cial relations.
Locations of the Institutions. In the Boston and
Cambridge area, research institutions also seem to
serve as anchors for startup clusters. We can see that
Harvard, MIT, BU, and many of the medical institu-
tions have startup clusters nearby (see Figure 2). For
instance, when talented graduate students and pro-
fessors from local institutions (such as MIT) invent
novel technologies, they often establish a startup
close to their base operations (their labs, which are
located locally).
Established Temporal and Spatial Distribution.
CrunchBase data also provide us insight about the
spatial distribution of different types of startup.
Within the Boston area, we can see, for instance, that
there are many biotechnology companies located in
Kendall Square. We may reasonably suspect that big
pharma companies like Novartis, and Pfizer provide a
hotbed for this type of startup to grow.
From the timestamp and geo-location data pro-
vided by startup companies in Crunchbase, we can
also trace the dynamics of startup developments in
the Boston and Cambridge area. There are two types
of districts popular among startup companies:
First, the innovation districts established by gov-
ernment, such as Kendall square and the South In-
novation District. The earliest tech companies es-
tablished here soon rapidly attracted new startups
into the area. Second, "self-emerging innovation dis-
tricts" that emerged spontaneously without govern-
ment intervention, such as Harvard Square, Central
Square, and the Financial District. Recently, these
have become popular districts for startups. Why they
spontaneously emerged is an interesting question.
The next section deals with this question.
SPATIAL ANALYSIS | Urban Context - Volume 2 - eCAADe 34 | 451
Yelp as an Amenity Indicator
One hypothesis of why these locations (Harvard
Square, Central Square, etc.) were preferred startup
locations is that they differ in some of the local ameni-
ties that are offered, compared to other areas of
Boston. We mined Yelp data to examine the diver-
sity and density of amenities in Boston. We com-
pared the "government established innovation dis-
tricts" with the "self-emerging innovation districts".
The first category of Innovation districts (including
Kendall Square, and South Innovation District) lack,
to this day, a diversity of local amenities, such as
restaurants, bars, and grocery stores. In contrast,
"new emerging innovation districts", such as Harvard
Square, Central Square, and Financial District, not
only have higher density of civic amenities, but also
have a greater variety (see Figure 3).
Land Use and Amenity Distribution. We examined
in more detail the amenity pie charts of four key ar-
eas: Kendall Square, South Innovation District, Har-
vard Square, and financial district. If we only focus
on "grocery stores" in the data visualization analy-
sis, it can be seen that the "self-emerging innova-
tion districts" also tend to have more groceries stores,
which plays critical role in providing daily supplies of
life. This information suggests to urban planners that
mixed land use is very important in the future for in-
jecting humanity to a new innovation district, and is
something deficient within the "government estab-
lished innovation districts." This is a consideration for
further improvement (see Figure 4).
Twitting-scape: Twitter as Activity and Land
Use Sensor
We suggest that Twitter can answer questions per-
taining to when and where activities are occurring. In
contrast to the civic data that can be provided from
Yelp, geo-location data from Twitter can provide in-
formation about whether a building or open space
is being used or not. We investigated the difference
between daytime and night-time usage of a piece of
land or building; more uniform use of the buildings
during the day and night means that the urban de-

sign goal for more efficient building use was achieved
(see Figure 5).
We can see, for example, that geo-located Twitter
data in Harvard square is active both during the day-
time as well as at night. In contrast, Kendall square
has a lot of data activity during the daytime, which
452 | eCAADe 34 - SPATIAL ANALYSIS | Urban Context - Volume 2
Figure 4
Kendall Square and
South Innovation
District have a low
proportion of civic
facilities.
decreases at night. This implies that human activi-
ties in Kendall square are not as popular as Harvard
square during the night. This informs the urban de-
signer that the mono-functional land use in Kendall Figure 5
square does not provide as many activities as the mix- Areas of high night
used land in Harvard Square, a consideration for fur- activity, analyzed
ther improvement (see Figure 6). from Twitter data.
Kendall Square and
Tweets as a Local Trend Speaker. Using Natural Lan-
South Innovation
guage Processing to mine tweets could help the ur-
District have less
ban designer understand important social trends and
activity at night.
priorities of local people. We use sentiment analysis
or opinion mining to identify and extract subjective
information from source materials. Generally speak-
ing, sentiment analysis aims to determine the atti-
tude of a speaker/writer with respect to some topic
or the overall contextual polarity of a document.
Figure 6
Red indicates mixed
use buildings, while
purple and blue
indicate mono
functional use. The
high prevalence of
mono function
buildings in
Cambridge and
South Innovation
District contribute
to the scarcity of
civic facilities and
night social
activities in these
areas.

Figure 7 of daily life in South Innovation District does not in

Analysis of tweets
with Natural
Language
Processing for
emotional content.
For instance, we can understand if people's emo-
tions are positive or negative (see Figure 7). One ob-
servation we made was that tweets related to the
South Innovation District innovation district contain
more negative words than tweets which is related to
Harvard square, implying that the perceived quality
many ways meet people's expectations.
Analyzing further the frequencies of words and
topics in tweets provides us with information about
what people care about and the image of the lo-
cal district within the citizen's daily life. For exam-
ple, some local amenities, such as CVS, parks, and
bars, frequently appear as the topics of Tweets in Har-
vard Square and Financial District. This implies that
these places seem to play important roles in local
life. Within Kendall Square, the cinema is the only
amenity topic that is significantly discussed; the rest
of the topics that show up in Tweets in this area are

SPATIAL ANALYSIS | Urban Context - Volume 2 - eCAADe 34 | 453

 related to "jobs" and "high tech". This problem is even
worse in the South Innovation District: no civic topics
show up at all. Only industrial-related words such as
"tech" and "startup" show up, suggesting an almost
monolithic focus within this area; not a sign of a vi-
brant district (see Figure 8).
Local Events and Activities Speaker.. Finally,
Tweets, in addition to analyzing spaces, can also be
used to analyze events. For example, "pi day" in Har-
vard square could be noticed from the Twitter data
of March 14, 2015. Thus, when and where major ac-
tivities happen can be observed from Twitter data
mined.
Flickr-scape: Flickr as Indicator of Urban
Life
Data mining images from Flickr can help understand
how people perceive the urban environment. Im-
age files collected from social media provide graphic
data, together with related information such as tag-
texts, geo-locations, and timestamps. In this way, we
possess not only the image content, but also data to
understand how people utilize urban space and how
people feel within the space (see Figure 9).
Ratio in Types of Image. For example, image data
from Kendall square innovation district most often
pertained to the urban space. Popular topics in-
cluded the local buildings, sculptures and other art,
landmarks, transportation, amenities, open space,
454 | eCAADe 34 - SPATIAL ANALYSIS | Urban Context - Volume 2
street, activities, and the Charles River. Thus, Those
targets are considered to be the popular and impor-
tant "image of the city" within the Kendall Square.
(Lynch 1960) Of particular note is that images of
the nearby Stata Building designed by Frank Gehry
on MIT campus account for a landmark within the
Kendall square area (see Figure 10).
In contrast, Harvard square Flickr data contain Figure 8
more images related to activities, street life, and open Analysis of tweets
space. This observation, combined with some of the with Natural
other observations above about Harvard square, sug- Language
gests a plausible reason why Harvard square spon- Processing for the
taneously developed into an Innovation district, de- popular topics
spite not being a government-designated one. More discussed in each
and more startups tend to come to this "mixed district.
used" area, where there are so many convenient civic
amenities, as well as a world class institution, Harvard
University.
APPLICATION TO URBAN DESIGN
Our social media analysis allowed the discovery of ur-
ban issues that were overlooked when the innova-
tion districts were first designed by the government
in a top-down fashion. For example, Kendall Square
revealed some land misuse, even though it is a popu-
lar location for startup companies (due to its proxim-
ity to MIT and big companies already present). From
Twitter data, it can be seen that there is very little
activity in this Square at night and on non-working
days. Also, many of its open spaces, encircled by
companies, are not being used. 
We have discovered a possible reason for this:
the prevalence of mono-functional land use and
enclosed open space in Kendall Square. Analysis
through Natural Language Processing supports this.
Tweets in Kendall Square primarily concern job and
employment-related topics, in contrast to Harvard
Square which features mixed-use land, and gener-
ates tweets including references to more events (like
Pi day) and leisurely locations (such as parks and
restaurants), at all times of the day.
Figure 9
Street-life related
and open-space
related images are
less frequent in
Kendall Square (left
of each pair) than in
Harvard square
(right of each pair).

Figure 10 streets, so that they do not have such an "enclosed"

Kendall Square: feeling to them (see Figure 12,13). Second, street
Image topic landscapes can be altered by urban designers to have
identification from more street furniture and amenities in order to facil-
Flickr. Images of itate a more mixed-use landscape for residents, and
Stata Center encourage more street activity. Third, government
designed by Frank officials can facilitate more social events such as festi-
Gehry on MIT vals in this area to stimulate activity in these unused
campus account for open areas.  In this way, appropriate urban modifica-
a significant tions are suggested from the social media data min-
portion. ing study that we conducted, to further improve the
innovation districts (see Figure 11).
Figure 11
Public events such CONCLUSIONS
as a festival can be All of these findings are in contrast to traditional tech
implemented by companies of the 20th century that were generally lo-
the Urban Planner cated in suburban areas to take advantage of cheaper
to stimulate land and lower wages, and may reflect new social
under-used open characteristics and needs of the 21st century. From
spaces in the our results, we made recommendations for the ur-
Kendall Square. ban plan and design that these cities should put in
Thus, based on this land use principle discovered place to increase the potential for "innovation". For
by social media, several suggestions for improving the city of Boston, for instance, it became very clear
this district can be made. First, government officials to us that Kendall Square is a district lacking in diver-
can change building and zoning codes to encour- sified amenities, and based on our findings, a plan to
age and reward buildings for putting civic facilities in increase local "mixed" land use will increase the ul-
the ground floors of buildings in the area, and even timate success of the Square. We suggest that our
encourage these civic facilities to extend onto the approach using social media analyses is a powerful

SPATIAL ANALYSIS | Urban Context - Volume 2 - eCAADe 34 | 455

 Figure 12
An example of
re-designing
existing buildings
(left) with open
retail stores on the
ground floors
(right).

Figure 13
Increasing mix use
zoning (left) as
opposed to
mono-function
zoning (right) can
increase night life
activity in
under-utilized
areas.

methodology to study the emerging "innovation dis-
tricts", an important topic of the 21st century.
ACKNOWLEDGMENTS
We thank C. Sun, M. Dennis, and S. Williams for dis-
cussions and paper preparation,C. Haung for help in
getting the user interface to run, J. Chen, C. Summit,
M. Lin and J. Nawyn for helpful discussions and com-
ments, members of Media Lab Changing Place group.
In memory of W. J. Mitchell.
REFERENCES
Ciuccarelli, P, Lupi, G and Simeone, L 2014, Visualizing the
Data City Social Media as a Source of Knowledge for
Urban Planning and Management, Springer
Johnson, K 2014, Innovating the City: Challenges and Op-
portunities in Establishing Incubators and Districts in
Paris and Boston, Master's Thesis, MIT
Kumar, S, Morstatter, F and Liu, H 2014, Twitter Data An-
alytics., Springer
Lynch, K 1960, The Image of the City, Cambridge, Mass.:
MIT
Lynch, K 1981, A Theory of Good City Form, Cambridge,
Mass.: MIT
Russell, M 2011, Mining the Social Web: Data Mining Face-
book, Twitter, LinkedIn, O
Simmie, J 2011, Innovative Cities, London: Spon
[1] http://www.brookings.edu/˜/media/Programs/metro
/Images/Innovation/InnovationDistricts1.pdf

456 | eCAADe 34 - SPATIAL ANALYSIS | Urban Context - Volume 2

Big Data for Urban Design
The impact of centrality measures on business success

Gideon Aschwanden1
1
The University of Melbourne
1
gideon.aschwanden@gmail.com

This paper investigates the role of spatial parameters in relation to the economic
dynamic embedded in the urban fabric. The key element explored in this study is
the role of the urban configuration and accessibility on the success of different
business sectors in Switzerland.The underlying hypothesis is that economic
markets are constant forces of change influencing the development of cities and
functions on all scales. Markets are institutions that reduce people's choices
based on a myriad of factors to a single number, the price. Accessibility is a
resource for each business that yields multiple values of benefits and transactions
in terms of economic properties. This project explores the interaction of multiple
measures of accessibility, calculated by Space Syntax analysis, with the success of
different markets represented by employment by business sector. 828548 business
locations and 44 spatial measures were used to derive associations between them.
The results show that the measure of 'Choice' correlates highly for smaller radii
and 'Integration' for larger radii with the total number of jobs. The result also
shows each sector has a specific set of accessibility measures that allows them to
thrive.

Keywords: Big Data, Centrality, Economy, Accessibility, Urban Design

INTRODUCTION different configurations of space and the success of

This project explores the relation between the built
form and the urban economy. The spatial configura-
tion and movement patterns in the urban fabric are
the result of human activity in the real world. Under-
pinning the urban fabric is a market structure that or-
ganises and influences behaviour and urban life. The
constant shaping of the built form is the result of the
same force that seeks to optimise the urban fabric
towards the needs of the people within the existing
constraints.This paper tries to reveal these forces in
individual business sectors. The focus is on the con-
figuration of the street network since it is the longest
lasting elements in the urban fabric. The street net-
work is also a resource that implicates costs of trans-
portation and accessibility to costumers for every
business location. Therefor, the relation between
the spatial configuration and economic success is ex-
plored to find the relationship between them. For de-
tails on economic models and spatial accessibility see
Narvaez et al. (2012).

SPATIAL ANALYSIS | Urban Context - Volume 2 - eCAADe 34 | 457

 In 'The social logic of space' Hillier and Hanson (1989)
argue that space itself carries patterns of behaviour.
In the consequent investigation of how urban sys-
tems behave Penn et al. (1993) showed that spatial
configurations govern the distribution of movement.
The configuration of the street network influences
the movement patterns and the movement patterns
evoke land use patterns and vice versa. This has been
shown in cases where an increase of street connec-
tivity has attracted a higher volume of people which
attracted different land uses (retail) which depend on
a high volume of costumers, and amplified the den-
sity of pedestrians (Hillier 1996). Therefor, if the street
network is guiding different movement patterns and
starting a multiplier effect, then it is argued that spa-
tial configuration can determine the success of busi-
nesses, which can be detected, in the number of em-
ployees.
METHOD
This paper explores employment by different busi-
ness sectors using space syntax analysis. A total of
828548 business locations with their business sector,
number of full-time and part-time employees was
available.
458 | eCAADe 34 - SPATIAL ANALYSIS | Urban Context - Volume 2
The spatial variables used were measures of In-
tegration, Choice and Total Depth at different met-
ric radii: 800m, 1600m, 2500m, 5000m, 10000m,
50000m. The basis for the graph is the street net-
work of Switzerland generated by TOMTOM, a com-
pany that provides mobile guiding services. By vary-
ing the distance radius upon which the integration is
calculated, the measure can be representative for dif-
ferent modes of transport. The cleaned network con-
sists of 1'356'105 links. Dangling links bellow 5m, par-
allel lines bellow 2m have been removed and nodes
and link end of 5m have been merged in order for the
algorithm to be able to calculate the measures in a
meaningful way. For details on the individual calcula-
tion methods and implications of Choice, Integration
and Total Depth see Aschwanden 2014.
Switzerland
Profile. Switzerland has grown historically in a de-
centralised way due to its historical federalism in or-
ganising land use and a high independence of towns
to plan and implement their land use and street de-
signs. Many town centres date back to medieval or
roman times and hold still memories of a fine meshed
street layout for pedestrians. In contrast to other Eu-
Figure 1
Swiss road network
indicating the 5
quintiles of Total
Depth with a radius
of 800m.
ropean cities this has not been destroyed in WWII
nor has it been altered by large-scale infrastructure
projects tracing highways through their centre. Nev-
ertheless, the dominating structures on the national
scale are the highways and the majority of the new
buildings within the last 50 years are constructed out-
side of traditional centres.
Street Network Analysis. This section shows a selec-
tion of variables tested using angular segment anal-
ysis. Figure 1 shows Switzerland with Total Depth
800m indicating small to large agglomeration cen-
tres, showing the location of more than 2700 munici-
palities. At a global level the Total Depth 10km high-
lights the large agglomeration centres like Zurich,
Basel.
Business Dataset
The business datasets consists of 828542 individual
data points including business sector, number of full-
time and part-time employees. This information is
collected 1995, 2001, 2005, 2008 resulting in a longi-
tudinal dataset over a period of 13 years. The choice
of business sectors was based on their homogenous
distribution across Switzerland and minimum of hav-
ing 100 locations leading to the following sectors:
wholesale trade, except of motor vehicles and motor-
cycles, Retail trade, except of motor vehicles and mo-
torcycles, Postal and courier activities, Hospitality Ac-
commodations, Food and beverage service activities,
Legal and accounting activities, Other professional,
scientific and technical activities, Rental and leas-
ing activities, Employment activities, Travel agency,
tour operator reservation service and related activi-
ties, Public administration and defence; compulsory
social security, Education, Human health activities,
Residential care activities, social work activities with-
out accommodation, creative, arts and entertain-
ment activities, libraries, archives, museums, botani-
cal and zoological gardens, gambling and betting ac-
tivities, Sports activities and amusement and recre-
ation activities, Activities of membership organisa-
tions, repair of computers and personal and house-
hold goods, other personal service activities.
SPATIAL ANALYSIS | Urban Context - Volume 2 - eCAADe 34 | 459
Feature Selection
With the combination of both business and spatial
accessibility measures linear models can be created
to find associations between them. To investigate
the relative importance of individual features hierar-
chical partitioning is applied. This method iteratively
reduces the features and calculates the resulting in-
crease of error in the predictive model. This allows
to remove features, spatial accessibility measures in
this case, that have a limited effect on the outcome. It
also removes features that can be substituted by oth-
ers and generates a model for prediction.
To test the model a bootstrapping algorithm is
applied that uses a random distribution of values for
prediction and returns the confidence intervals of the
prediction. The higher the confidence interval is the
better is the prediction capability of the model. This
is done with each model created by all feature combi-
nations. These models are then ranked according to
their confidence interval to identify the model with
the most accurate predictions. The features used in
these models are therefore the most relevant for the
individual business sector.
RESULTS
The location of jobs in Switzerland can be explained
best with the measures of 'Choice' radii 800m, 1600m,
2500m and 10000m as well as Total Depth n and In-
tegration 2500m and 5000m see Figure 3.
For individual business sectors the results are
more heterogeneous. Employment agencies are lo-
cated mainly in areas with high Integration radius
800 indicating that they need to be centrally located.
Hospitality (NOGA 55) on the other hand requires a
high value of choice with large radii and total depth
indicating their requirement for a good accessibil-
ity by individual transportation and centrality in re-
gard to the whole network. Retail (NOGA 47) thrives
when there is a combination of small radii integration
and large radii choice associated with pedestrians
and ease of access to highways simplifying logistics.
Some business locations are associated with high val-
ues of integration and do not relate to choice or to-
 tal depth like Creative, arts and entertainment activi- ties (NOGA 90). Libraries, archives, museums, botan- Figure 2
Swiss road network
indicating the
quintiles of Total
Depth with 10km
radius.

Figure 3
Confidence interval
for models
predicting the
number of
employment based
on different spatial
features.

460 | eCAADe 34 - SPATIAL ANALYSIS | Urban Context - Volume 2

Figure 4
Confidence interval
of models for
employment
prediction for
individual business
sector based on
NOGA codes
classifications
(national
classification of
economic
activities).

SPATIAL ANALYSIS | Urban Context - Volume 2 - eCAADe 34 | 461

 ical and zoological gardens (NOGA 91) on the other Hillier, B, Penn, A, Hanson, J, Grajewski, T and Xu, J 1993,
hand are more located where total depth is high. The 'Natural movement: or, configuration and attraction
same is true for Human health centres (Hospitals etc.) in urban pedestrian movement', Environment and
Planning B: planning and design, 20(1), pp. 29-66
whereas the residential care centres (NOGA 87) need
Narvaez, L, Penn, A and Griffiths, S 2012 'SPACE SYNTAX
an element of integration. ECONOMICS: decoding accessibility using property
value and housing price in Cardiff, Wales', Proceed-
DISCUSSION ings of the Eighth International Space Syntax Sympo-
sium, Santiago de Chile, pp. 1-19
The study shows that different spatial configurations
are congusive for different business sectors and that
locations with similar characteristics can be identi-
fied. The heterogeneity of models explaining the
labour market of a specific sector shows the complex
interaction between the built form and the labour
market. That the system is not producing a single
model that is magnitudes better than the next is to
be expected since the individual indicators are self-
similar and represents minute differences in the ur-
ban fabric.
The question of thresholds for the individual in-
dicators is not clear yet. Since only linear models are
evaluated this is not taken into account but might be
of interest when taking a decision.
The author is aware that this preliminary study
neglects several aspects like global market trends,
changes to the street network and longitudinal
trends that exist in the data. Additional geographic
factors like population density should also be in-
cluded since accessibility is useful if it is towards a
labour market or costumes. These aspects will be ad-
dressed in a later stage of the project.

REFERENCES
Aschwanden, GDPA 2014, Health and Place: an analysis
of the built environmentís impact on walking behav-
ior and health, Ph.D. Thesis, Diss., Eidgen\"ossische
Technische Hochschule ETH Z\"urich, Nr. 22014
Groemping, Ulrike 2006, 'Relative importance for linear
regression in R: the package relaimpo', Journal of sta-
tistical software, 17(1), pp. 1-27
Hillier, B 1996, 'Cities as movement economies', Urban
Design International, 1(1), pp. 41-60
Hillier, B 2007, Space is the machine: a configurational the-
ory of architecture, Space Syntax
Hillier, B and Hanson, J 1989, The social logic of space,
Cambridge university press

462 | eCAADe 34 - SPATIAL ANALYSIS | Urban Context - Volume 2

Development of The Method for Estimating Traffic Volume
of Pedestrians in An Underground Mall by Use of Watch
Cameras
Naoya Takagi1 , Atsushi Takizawa2
1
Osaka City University. 2 Osaka City University, CREST, Japan Science and Tech-
nology Agency.
1
t.naoya.oioi.etoomouyo@gmail.com
2
takizawa@arch.eng.osaka-cu.ac.jp

This paper describes a method for estimating pedestrian traffic volume by using
video cameras. In the Umeda underground mall in Osaka City, we estimated the
traffic volume without tracking technology and while protecting pedestrian's
privacy. We developed an original algorithm that roughly estimates the traffic
volume of pedestrians from sequential images of video cameras. We focused on a
line on each image cut out from video and made a new image which shows the
spatiotemporal distribution of pedestrians. We defined this image as 'time
historical image of pedestrian spots (THIPS)'. In a THIPS, a pedestrian is
regarded as a cluster of connected pixels with the same label. We captured the
spatiotemporal distribution of pedestrians by using these images. We found that
this algorithm requires a THIPS to estimate the number of pedestrians who
passed the spot for a few minutes and plural THIPSs to estimate their traveling
directions. Finally, we concluded that this algorithm is an efficient means of
estimating pedestrian traffic volume.

Keywords: Pedestrian Flow, Underground Mall, Spatiotemporal Distribution,

Watch Cameras, Background Subtraction, Integer Linear Problem

INTRODUCTION number of people to be monitored and the protec-

In recent years, the study of spatiotemporal distribu-
tion of pedestrians has been increasingly important
in the field such as urban planning, disaster preven-
tion planning, marketing; however, we have only re-
cently obtained detailed data. Although many re-
searchers are trying to estimate the distribution by
using the data of various sensors, some problems re-
main, such as the cost of introducing sensors, the
tion of their privacy.
The authors needed a spatiotemporal distribu-
tion of pedestrians to be applied to evacuation plan-
ning in a huge underground mall. It is difficult to get
data from cellular phones in underground malls be-
cause their signals are too weak. Conversely, because
many underground malls are already equipped with
video cameras, it is worth to consider the possibil-

SPATIAL ANALYSIS | Urban Context - Volume 2 - eCAADe 34 | 463

 ity to utilize them. However, underground malls are
densely populated spaces, and video cameras are in-
stalled with a shallow angle of depression. Therefore,
it was anticipated that the tracking technology was
not feasible.
The purpose of this study was to define a method
of using video cameras to estimate the spatiotempo-
ral distribution of pedestrians in a mall without dis-
turbing their privacy. By focusing on the passages
used by pedestrians, we estimated pedestrian traf-
fic volume to determine the congestion of a target
space.
DATA ACQUISITION METHODS FOR
PEDESTRIAN MOVEMENT
In recent years, in support of evacuation from natu-
ral disasters such as earthquakes and flooding, the
technology capturing the space-time distribution of
pedestrians and markets such as O2O services are be-
ing researched and developed to relax the conges-
tion in places where an unspecified number of peo-
ple congregate. The data source for this technology
is generally traffic census data, CCTV and small cen-
sors such as GPS and RFID. Indoor positioning tech-
nologies based on Bluetooth and Wi-Fi are also being
researched; however, these positioning technologies
are underdeveloped and are face high introduction
costs.
Kinoshita developed a method for estimating
real-time origin-destination flow of pedestrians in
a space in 2015, and Kataoka proposed a method
based on Markov random field theory to reconstruct
incomplete traffic data collected by various traffic
sensors in 2014.
TRACKING TECHNOLOGIES
Tracking technologies are necessary for extracting
the position of pedestrians in images and movies.
The core technology is machine learning with local
information such as facial attributes. However, those
technologies are influenced by the instability of im-
ages and movies as listed bellow, and this makes pre-
cise detection difficult.
464 | eCAADe 34 - SPATIAL ANALYSIS | Urban Context - Volume 2
As features can be used stably even in the conditions
listed below points, some indices based on slope,
color, movement and distance information are pro-
posed. Deep learning that automatically extracts fea-
tures from images and movies has also been attract-
ing attention in recent years.
• Individual difference of clothes and physical
size.
• Complex background makes human detec-
tion difficult.
• Direction and posture of a human also makes
human detection difficult.
• If human areas are overlapped, we may under
estimate them.
• View point of a camera especially depression
angle influences the performance of the de-
tection.
OUR PREVIOUS STUDY
In 2015, our previous study used the images from a
surveillance camera in an underground mall. How-
ever, at that location, because the ceiling height
and the depression angle of the cameras was low,
pedestrians were overlapped in the movie. Because
the background images were difficult to change
throughout the seasons and time, we used a simple
background subtraction method on the color value
of pixels in the images extracted from a movie. In
particular, we prepared images in time zones with no
pedestrians present and also images with pedestri-
ans present. We calculated the color information of
RGB, HSV and Lab of each pixel and obtained L1 and
L2 distances between pixels located in the same po-
sition of no pedestrian image and the target image.
We summed the distances between all pixels, used
them as explanatory variables of regression models
whose objective variable was the number of pedes-
trians in a target image and predicted the number of
pedestrians with a regression model.
Because this method requires extensive training
data that needs to be prepared manually and be-
cause the target area for the prediction is a little am-
biguous, we propose this study.
Figure 1 thermore, the Umeda underground mall is unified
Umeda with many buildings such as Osaka Station City. Thus,
underground mall because the Umeda underground mall has been un-
model for the intentionally enlarged many times, it is difficult for us
evacuation to go through other areas.
simulation.
PROFILE OF THE DATA
In this study, the target area was the Osaka City un-
derground mall area, as illustrated in Figure 2. We ac-
quired movies from a few dozen video cameras for
the first 10 min after 8:00, 10:00, 12:00, 15:00, and
18:00 on Sunday June 16, 2013 and on Monday June
17, 2013. The movie frame size was 320 × 240 pix-
els. The original movies were compressed into AVI
format. For the study, we cut out eight images per
second from a movie and blurred them with a Gaus-
sian filter so that the pedestrians' faces in the images
Figure 2 could not be distinguished. We developed our algo-
Umeda rithm with Microsoft Visual Studio 2010 OpenCV 2.3.1
Underground Mall (Open Source Computer Vision Library).
governed by Osaka
City. TIME HISTORICAL IMAGE OF PEDESTRIAN
SPOTS
We made an image by piling with a line in 2424 im-
ages that we captured over a span of 5 min to grasp
the behavior of pedestrians (see Figure 3). We de-
fined this image as 'time historical image of pedes-
trian spots (THIPS)', as shown in Figure 4. The width
of a THIPS expresses the change in time of the line,
and the height is the width of the original image. As
a consequence, this image shows the spatiotempo-
PROFILE OF THE UMEDA UNDERGROUND ral distribution of pedestrians. Below this, progress
MALL in the estimation for our algorithm is shown by this
As illustrated in Figure 1, the Umeda underground THIPS.
mall in the Umeda district has expanded to Chaya-
machi (most northern district), Dojima (most south- CLASSIFICATION OF THIPS BY WATCH
ern district), Doyamacho (most eastern district), and
CAMERAS
West Umeda (most western side), approximately 1.1
In making a THIPS, it is significant whether a line
sq. km. Moreover, the Umeda underground mall
is set although the behavior of pedestrians should
comprises Whity Umeda (northern and eastern side),
be appreciated as clear as possible. Thus, we intro-
Diamor Osaka (central side), Dojima Underground
duced a vertical section against the boundary be-
Shopping Centre (southern side), Osaka Garden City
tween spaces in order to understand the state that
(western side), a railway station and a terminal. Fur-

SPATIAL ANALYSIS | Urban Context - Volume 2 - eCAADe 34 | 465


pedestrians move to another space. We defined the
vertical section as the 'THIPS section', and the line on
movies as the 'THIPS line'.
On that account, as illustrated in Figure 5, we
classified THIPSs as follows: 'time historical image of
pedestrian horizontal spots (THIPS_H)', which is de-
noted by a horizontal line on movies; 'time historical
image of pedestrian vertical spots (THIPS_V)', which
is denoted by a vertical line on movies and 'time his-
torical image of pedestrian diagonal spots (THIPS_D)',
which is denoted by a diagonal line on movies.
ALGORITHM FOR THE ESTIMATION OF
SPATIOTEMPORAL DISTRIBUTION OF
PEDESTRIANS
We interpreted the information of pedestrians on
THIPSs to estimate the spatiotemporal distribution of
pedestrians. We propose an algorithm and describe
each method of calculation.
466 | eCAADe 34 - SPATIAL ANALYSIS | Urban Context - Volume 2
Figure 3
Form of a time
history image of
pedestrian spots.
Figure 4
A Time Historical
Image of Pedestrian
Spots (THIPS).
In a THIPS, a pedestrian is regarded as a cluster
of connected pixels with the same label. First, the
algorithm calculates the distance of every image by
subtraction and arranges the pixels. Second, it as-
sesses the value of each pixel as 0 or 1 with a thresh-
old; that is, whether a pedestrian exists or not, re-
spectively. Then, the algorithm defines the cluster
of a pedestrian in a THIPS based on a labelling al-
gorithm that examines neighbor pixels against a la-
belled pixel. Furthermore, a morphological opera-
tion and a fitting by least squares method are intro-
duced into the algorithm to decrease the misdetec-
tion of clusters. Figure 5
The Classification of
THIPS.
REGULATION OF BINARY THIPS BY MOR-
PHOLOGICAL OPERATION
It is inevitable that noise is detected and pedestrians
are divided because of the binary THIPS by subtrac-
tion. Therefore, the algorithm introduces a morpho-
logical operation to integrate divided clusters and to
remove noise.
The morphological operation works on dilating
or eroding clusters on images; it is usually used a cou-
ple of times. In addition, the 'opening operation' di-
lates after eroding clusters, and the 'closing opera-
tion' erodes after dilating. The 'closing operation' is
effective against combined clusters, and the 'open-
ing operation' is effective in limiting noise. In our al-
gorithm, to efficiently decrease noise, the dilating op-
eration was introduced not dilating clusters in case all
neighbor pixels are zero.
LABELLING PEDESTRIANS IN BINARY
THIPS
The labelling part of our algorithm checks a neigh-
bor's pixels around an objective pixel, labels the ob-
jective pixel under and moves the next objective
pixel using the below three rules. This labelling al-
gorithm repeats until a cluster is identified. The la-
belling algorithm is summarized as follows:
1. When the value of all checked pixels is 0, a
new label is given to the target pixel.
2. When only one value of all checked pixels is 1,
the label of the checked pixel is given to the
target pixel.
3. When the value of more than two pixels is
1 and the labels of checked pixels are differ-
ent, the labels of all checked pixels are merged
into the oldest label.
This labelling algorithm identifies some overlapped
pedestrians as a cluster and a pedestrian as some
clusters on a binary THIPS by misdetection of subtrac-
tion.
FITTING THE NUMBER OF PEDESTRIANS
BY LEAST SQUARES METHOD
The labelling algorithm shows the situation of pedes-
trians who are divided or overlapped. Thus, the num-
ber of pedestrians comprising a cluster on a THIPS
is important to determine an exact estimation of the
number of pedestrians. Therefore, in each cluster on
a THIPS, we calculated the centroid coordinate of a
cluster, the number of pixels occupied between the
time axis (the width of time), the number of pixels oc-
cupied between the space axis (the width of space)
and the number of pixels comprising a cluster (the
size of a cluster). We adjusted these variables to the
SPATIAL ANALYSIS | Urban Context - Volume 2 - eCAADe 34 | 467
function approximating the number of pedestrians
in a cluster with the solver tool of Excel by the least
squares method. The training data of the function are
the number of pedestrians in each cluster on a binary
THIPS, manually counted to the first decimal place.
The approximate function of the width-of-time
and the space-of-each-cluster variables attained a
0.835 accuracy rate. We regarded clusters with the
number of pedestrians under 0.1 as noise and elim-
inated those from our conclusion that considered
how many pedestrians comprised a THIPS. It fol-
lows that we obtained 207 pedestrians on the THIPS,
and the fitting operation cleaned 158 pedestrians as
noise from the THIPS. The fitting operation is effective
in not only removing noise but also in understanding
the situation of each overlapped or divided cluster.
ADJUSTMENT OF CLUSTERS ON THIPS
We see that the size and shape of clusters on a
THIPS varies widely. This is due to the projec-
tive transformation into two-dimensional from three-
dimensional. We think this distortion of clusters will
affect the estimated result. Moreover, on being mul-
tiplied by the correction coefficient, the distortion of
clusters will correspond to the variables listed in the
previous section along the classification of a THIPS
line.
To begin with, we explained 'time historical im-
age of pedestrian horizontal spots (THIPS_H)'. As illus-
trated in Figure 6, we assumed the situation where
pedestrians A and B who are of the same size and
line up horizontally moved in a straight line at the
same speed from the back to the front on the screen.
Those shapes of clusters became a trapezoid be-
cause pedestrians A and B became larger as they
approached the camera. The shape of pedestrian
A was different from that of pedestrian B in Figure
6, because pedestrian B walked nearer to the wall
than pedestrian A. However, the shapes of pedestri-
ans were not the same, but their sizes were the same
on account of those widths of space and time were
equivalent in each time. Accordingly, the width of
space of each cluster was better when averaged.

Subsequently, we explained 'time historical image of
pedestrian vertical spots (THIPS_V)'. As illustrated in
Figure 7, we examined a situation where pedestrians
A and B line up vertically on the screen. We under-
stood that the size of pedestrians A and B were rel-
ative to their standing position due to the operation
of projective transformation from a world coordinate
system to a screen coordinate system. Therefore, the
size of a pedestrian had to be relatively adjusted to
the coordinate of a cluster's bottom in the case of a
THIPS_V.
DATA STRUCTURE
We hold the data of a pedestrian as a list data struc-
ture. The list holds the information on their label,
coordinate of the centroid, coordinates of all pixels
and size of the cluster. Consequently, the size of the
list corresponds with the number of pedestrians in a
THIPS.
VERIFICATION OF THE ALGORITHM
We show the method of estimating the number of
pedestrians by a series of algorithms. Background
images and pedestrian images are blurred by a Gaus-
sian filter, whose kernel size is 5 and deviation is also
5. Moreover, the threshold is 110, which can be dis-
tinguished when the scanning lines disappear in bi-
narizing pixels. In morphological operation, we took
468 | eCAADe 34 - SPATIAL ANALYSIS | Urban Context - Volume 2
one closing operation but did not take an opening Figure 6
operation because it broke the shape of clusters. As The Shape of
a result, we output the data of pedestrians' list, fitting Cluster in
to linear approximate equation, which is introduced Horizontal Case.
the variables of spatiotemporal width and attained
accuracy rate 0.85.
Figure 8 shows the labelled pedestrian's image
which is made of the verification data. Additionally, Figure 7
Table 1 clarifies the comparison the number of esti- The Shape of
mated pedestrians with the actual number of pedes- Cluster in Vertical
trians. In the number of clusters, we obtained better Case.
result against that of just labelling operation. How-
ever, in the number of pedestrians by fitting opera-
tion, the result was not different from labelling one.
Therefore, we judged the adjustment of pedestrian's
information effective in enhancing the accuracy rate
of the approximate function.
Table 1
The Comparison
between Estimation
Number, Fittng
Number and Actual
ALGORITHM FOR ESTIMATION OF TRAF- Number.
FIC VOLUME OF PEDESTRIANS
Plural THIPSs are necessary to estimate pedestrian
traffic volume and travel direction. In addition, the
list data of pedestrians are arranged by their THIPSs
by considering the positional relationship between
each cluster. Moreover, our algorithm performs
the matching of clusters between THIPSs using inte-
ger linear programming (IPL) to estimate the cross-
section traffic volume.
PEDESTRIAN MOVING DIRECTION
We developed an algorithm to obtain each pedes-
trian's coordinates and travel direction from the cen-
troid of the cluster.
The information between different spots is nec-
essary to reveal a pedestrian's travel direction. Figure
9 shows the connection of labelled pedestrians be-
tween different spots. We define the travel direction
of pedestrian K as a vector Δu. It has two factors on
Figure 8
A THIPS with
Pedestrians
Labelled.

different microspaces, as shown in Eq. (1). As a result,
we understand the direction whether Δt is larger than
zero or not.
K|y= − K|y= = ∆u = (∆x, ∆t) (1)
Thus, we assess whether the moving direction, which
is estimated by the difference between centroids of
clusters, is correct or not and verify the accuracy of
our algorithm to estimate the moving direction of
pedestrians. In conclusion, we achieved compara-
tively high validity of the algorithm in Table 2.
Table 2
Correct Estimation
of the Direction of
Pedestrians.
CLEANING NOISE BY DISTANCE OF CLUS-
TERS BETWEEN THIPSS
To estimate the travel direction of a unique pedes-
trian, we have to be able to distinguish a particular
cluster when compared with clusters of other THIPSs;
those do not always exist in similar positions. Con-
sequently, we cleaned irrelevant clusters of other
THIPSs due to the large distance between them. We
focused on almost not changing the position of clus-
ters in case the intervals of THIPS lines were infinites-
imal. It would be the closest distance identified with
same pedestrian, calculating the distance of clusters
between THIPSs by the Ward method. Hence, it is
likely to be further than the distance between clus-
ters that are noises and those clusters which would
be same pedestrian on another THIPS. Here one of
the THIPSs is the source layer and the other is the
compared layer. The changes of minimum value of
the distance between those clusters are shown in Fig-

Figure 9
Moving Direction of
Pedestrians.

SPATIAL ANALYSIS | Urban Context - Volume 2 - eCAADe 34 | 469

 Figure 10
Transition Graph of
the Minimum
Distance between
Clusters and
Boxplot of Clusters.

Figure 11
The Matching
Conclusion
between
Clusters(upside:
Source Layer,
underside:
Compared Layer).

ure 10 (the interval between them is 5 pixels). The

n ∑
∑ m
minimize: {xi ,j ·distan cei ,j }
largest minimum distance between clusters was re- i=0j=0
garded as an outlier by outlier detection. subject to: if(the number of clusters are
∑
n ∑m
same){ xi ,j = i(= j)}
MATCHING CLUSTERS BETWEEN THIPSS
i=0j=0
BY ILP else (the number of clusters are different)
In matching clusters between THIPSs, we considered ∑
n ∑m

the combinatorial optimization that is the combina- { xi ,j ≥i(i > j)}

tion of clusters that is probably identified with same i=0j=0
xi ,j ∈{0, 1}
pedestrian between the source and compared lay-
ers which are cleaned of outliers. In other words, Figure 12
this combinatorial optimization is the problem which The Estimation
the distance of those clusters minimized. We solved Places of Traffic
this problem with ILP by IBM ILOG CPLEX, where Volume of
distan cei ,j is the distance between a clusteri Pedestrians.
on the source layer and a clusterj on the com-
pared layer, and xi ,j are binary variables describing
whether a combination with clusters exists or not.
The formulation of ILP is summarized as follows:
The minimizing problem of the sum of the dis-
tance between clusters on THIPSs

470 | eCAADe 34 - SPATIAL ANALYSIS | Urban Context - Volume 2

Figure 13
Algorithm for
Estimating Traffic
Volume of
Pedestrians.

Table 3
The Estimation
Result of Traffic
Volume of
Pedestrians in
Watch Cameras.

LABELLING CLUSTERS BETWEEN THIPSS
The source layer and the compared layer are inte-
grated on the basis of the matching results so as to
estimate the travel direction of pedestrians. Data
are structured as in Section "Data Structure" and also
held in the list structure.
VERIFICATION OF THE ALGORITHM
Although the influence that some noises are
matched to far cluster is ascertained, the perfor-
mance of labelling algorithm between THIPSs gener-
ally performs well. The result of matching clusters is
shown in Figure 11.
EXPERIMENT
I estimated the traffic volume of pedestrians over the
area illustrated in Figure 12 by this algorithm. The re-
sult is shown in Table 3. It describes no estimated
value is far from the correct number of pedestrians.
In addition, during times of low pedestrian flow, the
estimated value was almost equivalent to the actual
count as shown in the results of cameras 3 and 7.

SPATIAL ANALYSIS | Urban Context - Volume 2 - eCAADe 34 | 471

 However, the rate of error of some estimates exceeds REFERENCES
0.2. This is because the width-of-time of clusters was Kataoka, S, Yoshida, M, Furtlehner, C and Tanaka, K 2014,
smaller with passing time short in the cases of cam- 'Traffic data reconstruction based on Markov ran-
eras 1 and 2. In the case of camera 4 that tracks a com- dom field modeling', Inverse Problems, 30(2), p. 15
Kinoshita, Y, Sasazawa, M, Yoshida, K and Ishima, K 2015
plicated passage, the estimate was a different from
'Traffic Volume Estimation Method Using Data As-
the actual count because of the division of pedestri- similation –A Study of Traffic Volume Estimation in
ans. In addition, in the case of camera 5, it was dif- Railway Station, Part2-', Summaries of technical pa-
ficult to estimate the traffic volume because many pers of Annual Meeting Architectural Institute of Japan
pedestrians were overlapped. Architectural Institute of Japan, Kanto, p. 5346
Therefore, in low-volume situations where Takagi, N and Takizawa, A 2015, 'A Study on the Meth-
ods for Estimating the Distribution of Pedestrians
pedestrians were not overlapped or the shape of
in an Underground Mall by Use of Watch Cameras',
clusters were obviously independent of others. Our Computer Aided Drafting, Design and Manufacturing,
algorithm offers an efficient means of estimating 25(4), pp. 32-38
pedestrian traffic volume by video camera. Yamaguchi, Y, Yamashita, T and Fujiyoshi, H 2012, 'Hu-
man Detection Based on Statistical Learning', IPSJ
SIG Notes, 18, pp. 1-14
CONCLUDING REMARKS
This study demonstrates that it is not necessary to
install additional sensor devices to obtain data on
pedestrian traffic flow. However, the locations to es-
timate traffic volume are limited because cameras do
not change their position. Therefore, it is taken this
method and original means until now. This method
also has the advantage of obtaining traffic volume
from existing movies of cameras.
In this paper, we proposed a method to estimate
the traffic volume of pedestrians (see Figure 13). We
captured the spatiotemporal distribution of pedestri-
ans by focusing on a spot within camera images. Ac-
cordingly, we conclude that this algorithm is an effi-
cient means of estimating pedestrian traffic volume.
Moreover, we attained a high accuracy rate in clus-
ters that could be equivalent between selected spots.
Finally, there are some issues in the basic operation,
and we intend to solve these issues and enhance the
accuracy of our system of estimating pedestrian traf-
fic volume.

ACKNOWLEDGMENTS
The authors were offered data sets from Osaka City
Construction Bureau and Taniguchi laboratory in Os-
aka City University. This study was partially sup-
ported by Grant-in-Aid for Scientific Research (A) (No.
25240004).

472 | eCAADe 34 - SPATIAL ANALYSIS | Urban Context - Volume 2

 Synthetic Modelling of Pedestrian Movement
Tallinn case study report

Renee Puusepp1 , Damiano Cerrone2 , Martin Melioranski3

1,2,3
Estonian Academy of Arts
1,2,3
{renee.puusepp|damiano.cerrone|martin.melioranski}@artun.ee

This paper builds towards the argument that pedestrian traffic in the city can be
successfully simulated with agent-based computational models if pedestrians'
movement patterns are appropriately studied first. Furthermore, such simulation
models, when finely calibrated and supported by onsite observations, allow
planners to evaluate different urban design scenarios.We present a pilot study
carried out in the centre of Tallinn, and discuss a way of how pedestrian
movement simulations can be conceived. In the pilot study we recorded some 120
traces of pedestrians' movement and developed a prototype of an agent-based
computational model to simulate this movement. Additionally we investigated the
possibility of including solar analysis into the computational model. Already this
short exercise offered us some interesting insights into how certain spatial
qualities and features can drive pedestrian traffic making urban walkers to verge
off the shortest routes. The pilot study was carried out in the context of the High
Street project [1] for turning the centre of Tallinn into more pedestrian friendly
area by redesigning urban space, calming vehicular traffic and creating new
opportunities for businesses to flourish.

Keywords: agent-based model, urban analysis, pedestrian simulation, movement

patterns, solar analysis

Figure 1 BACKGROUND
Project area in Agent-based models have been used in urban and
Tallinn. spatial analysis before. Perhaps the best known is
the Space Syntax approach. For example, Penn and
Turner (2003) demonstrate that space layout in urban
settings affects search efficiency of agents. Space
Syntax tends to use pre-calculated visibility graphs
and their agents possess no knowledge of urban
space nor have they particular targets. They are sort

SPATIAL ANALYSIS | Urban Context - Volume 2 - eCAADe 34 | 473

 of random walkers (Batty 2003) purely reacting to the
degree of visibility in open space. However, this be-
haviour is probably not a standard case in any city. At
least in our short pilot study in the centre of Tallinn
we observed only a few occasions when people did
not seem to have a particular destination and were
simply guided by ad-hoc decisions. Following this
observation, our simulation model is opposed to the
agent-based model used in Space Syntax studies and
to the mere modelling origins and destinations. In-
stead, we assume that simulation agents "know" the
shortest path between their origin and their destina-
tion. Nevertheless, an agent may not always choose
the shortest route since its navigational decisions are
also affected by some local spatial features pertaining
to the configuration of space and urban morphology.
Such features are associated but not limited to wider
pavements, less noise, sunnier or greener space, and
the activities certain streets or spaces offer.
Some models, most notably in Space Syntax,
try to simulate pedestrian movement purely based
on physical configuration of space and urban layout
(Penn and Turner 2002, Penn and Turner 2003 ). The
speciality of our approach is to acknowledge that
pedestrians, due to socio-cultural background, move
differently from country to country and from city to
city. Thus, we first observe behaviour of pedestri-
ans on the site and then adjust our simulation model
based on these observations.
Our method is relatively new in the Nordic region
of Europe but it has developed upon well-known aca-
demic theories and approaches shared all over the
world. The analytics methods that we are going to
deploy for the Tallinn High Street larger scale study
are pretty solid and developed on the field as they
were recently used in commissioned project by SPIN
Unit together with the Estonian Academy of Arts for
the City of Turku (Cerrone et al. 2005) and for Urban
Design London with Transport For London. In the
first case we developed a study and a plan for revi-
talisation of the city centre and transportation plan-
ning. In the case of London, we have focused on the
area of King's Cross, using social media data to anal-
474 | eCAADe 34 - SPATIAL ANALYSIS | Urban Context - Volume 2
yse spatio-temporal patterns and assess the impact
of new development on retail clusters present in the
area. Working for Turku and London, gave us experi-
ence and confidence on our method and has consol-
idated our practice.
Cities are places of opportunity and innovation
and generate new emotions and spontaneous ac-
tivities. The complexity of human activities and ex-
change is what gives urban spaces particular dynam-
ics, making some streets more successful than oth-
ers. Attempts to model the key factors of pedestrian
movement can be found in the form of multiple sci-
entific approaches and analytic methods developed
in the fields of urban studies (e.g. Batty 2003), spatial
cognition (e.g. Tahrani et al. 2005) and travelling be-
haviour studies (e.g. Broach et al. 2009). So far, there
is not one theory or one model that can unify the
description of pedestrians' path choice into one as-
sumption due to the complexity of spatial dynamics
and behavioural patterns. In order to tackle the com-
plexity and dynamism of social and architectural ur-
ban features, it may be possible to assemble models
describing the complexity of the relations and colli-
sions between individuals and the city or simplify the
problem by reducing the amount of generalisations
and assumption by carrying out actual observations
in a limited and relatively small urban area. We chose
to focus on observations based modelling because
our interest as scholars is to bring people back to the
centre of our studies, where human behaviour is no
longer attempted to be generalised and described by
a model it characterise them.
We assume that good urban form, good acces-
sibility, social activities and commercial services lo-
cated along streets are all needed in order to create
a space for everyone. Streets generate movement
and attract people thanks to both their physical con-
figuration and the activities they afford and accom-
modate. Public space should promote integration
of people of different classes, ages and social hier-
archies. Higher levels of inclusion leads to a higher
number of people walking on the streets and using
the services offered.
 In order to provide new tangible spaces and propose
innovative improvement of the existing ones we start
by mapping and analysing pedestrians movement,
that is one of the main resources of small urban cores
or historic towns. We study the area near Viru Väljak
in Tallinn because it is the meeting point of the his-
toric town with the new urban core. It allows us to ad-
dress the need of fostering pedestrian movement in
the historic town to improve tourism and accessibil-
ity. The new urban core, which is mainly designed for
cars, demands to be redesigned to foster pedestrian
movement and permeability between the function-
alistic architecture that characterises it - better plan-
ning of pedestrian movement is necessary to the sur-
vival of both.
terest, during a 2-hour study. These tracks were later
mapped to a GIS application. Then a series of agent-
based simulation models were devised to mimic par-
ticular behavioural properties of pedestrian move-
ment that we observed. Once selected algorithms
were compiled into a single model, several iterations
of simulation model development and comparison
of simulation results with the recorded GPS tracks
was carried out. The simulation model was thereafter
modified and compared to the real world data until
the tracks left behind by agents bore strong enough
resemblance to the GPS tracks. If at first this appears
as an exercise of remodelling natural movement, our
approach can be used to simulate and study the im-
pact of proposed urban transformation in the scale of
10 city blocks.

Figure 2
THE PILOT STUDY
Our pilot study demonstrates how principles of syn-
The study area map
thetic modelling (Morse et al. 2008) can be used
shows two different
to simulate pedestrian traffic within an urban area.
types of gates - dots
Synthetic modelling method allows development
indicate bus stops;
and fine-tuning of computational simulation mod-
lines mark the area
els based on real-world data. In a nutshell, syn-
boundary crossable
thetic modelling is an iterative method of develop-
to pedestrians.
ing computational systems that simulate real-world
phenomena by the means of observation, modelling
and comparison. A synthetic modeller is not as much
concerned with the exactness of the model compo-
nents as with making sure that the model produces
similar patterns to those that are observed in real- Common methods deployed to study pedestrian
ity. In our case of simulating pedestrian movement movements use surveyors standing at certain gates
we modify and calibrate the model until walkers in and counting people passing the gates. Gates (see
the simulation model start choosing similar trails to Figure 2) are imaginary access and exit points, per-
those of real pedestrians. pendicular to a street and for this reasons they can
The first step of our study is mapping pedestri- only measure how many people move in or out a
ans' traces to quantify and characterise movement in certain street. Another approach is counting pedes-
the area. We organised an intensive workshop with trians arriving to origin points or destinations. In
masters students at the Architecture department of our study we combine both approaches to charac-
the Estonian Academy of Arts to trace and record terise pedestrian movement with surveyors follow-
pedestrian' paths in the area of Viru Väljak using GPS ing pedestrians from one gate to the next one. We
enabled smartphones. A group of 10-12 students divided surveyors into two groups: one group starts
recorded over one hundred pedestrians' tracks from recording tracks from a set of selected origins (trans-
their origin to their destination within the area of in- portation nodes) till their destinations (transporta-

SPATIAL ANALYSIS | Urban Context - Volume 2 - eCAADe 34 | 475

 tion nodes, retails, building etc.) or till they cross one
edge (gate) of the area of interest; the second group
starts recording tracks from a given edge (gate) of the
area of interest till another edge or till the path ends
into a destination.
What makes this method valuable to charac-
terise pedestrian movement is that once data is
loaded and mapped into GIS we can see what are
the major attractors within our area of interest - we
can map the path density to highlight the street seg-
ments where pedestrians are more likely to be pass-
ing by and we can observe behaviour patterns in re-
gards to path choice. Also, we can analyse individual
pedestrian journeys (see Figure 3) one by one. This
information will be later used to warp the computer
simulation in order to set attractors and flow density
based on these observations.
The field observations help us to identify features
476 | eCAADe 34 - SPATIAL ANALYSIS | Urban Context - Volume 2
that attract pedestrian traffic - magnets, as we call
them. These magnets make walking easier, more
pleasurable or more purposeful. Magnets can be spa-
tial (e.g. width of pavement), functional (cafes, shops)
or environmental e.g. noise, sunniness, general ther-
mal comfort (Mayer et al. 2008).
Despite the field work carried out in the pilot
study was extremely insightful for understanding
pedestrian movement, it was also very labour inten-
sive. This type of studies should be carried out regu-
larly and in much larger scale. Only then we can hope
of building and calibrating a truly validated simula-
tion model.
There were a few ways that we can consider ex-
tending the pedestrian traffic data collection meth-
ods.
1. One of the possible ways is to use active mo-
bile positioning method as described by Tiru
Figure 3
Pedestrian tracks
recorded by
students in outdoor
studies.
 (2011) - a method where "location request
can be initiated from the mobile device it-
self (by its user) or from outside." In the latter
case, the consent of the owner of the device is
needed. This means that for active position-
ing, we need to get people to install a mobile
application (e.g. OSMTracker) and send us the
data that they have collected, or we can de-
velop our own application that facilitates au-
tomatic collection of GPS data from devices.
2. We can use pedestrian detection cameras
and convert movement paths to GPS tracks.
A variety of methods have been developed
for tracking single or multiple pedestrians in
static or moving cameras by exploiting differ-
ent types of image information [2] Installing a
camera-based tracking system requires a net-
work of cameras to cover the entire study area.
3. Use a purpose made network of WiFi access
points to detect all WiFi enabled devices.
4. We can use a purpose made sensory network
in combination with low-cost tags or beacons
that are given to the pedestrians at the gates.
Potential options are a) Bluetooth Low Energy
(BLE) beacons and b) RFID tags.
INTEGRATING STUDIES OF NATURAL
LIGHT
The research already done in 1970's by William H.
Whyte (1980) has shown that solar access and ex-
posure are key factors in creating attractive settings
for pedestrian gathering in streets and plazas. Thus
one can assume that solar exposure level also affects
the pedestrian choice of routes. Perhaps with a few
exceptions of mid-summer's extreme weather con-
ditions in Tallinn and other cities in similar climatic
zone, exposure to sunlight usually makes streets
more attractive for walking. Therefore, direct expo-
sure to sun can be considered as a magnet in the pro-
posed simulation model. A study of how solar ex-
posure impacts spatial perception in urban context
has been connected to space acknowledgement in
SPATIAL ANALYSIS | Urban Context - Volume 2 - eCAADe 34 | 477
relation to analysis and effects within Virtual Reality
(VR) environments (Tahrani et al. 2005). In our case
the simulation of attraction to solar exposure is per-
formed in 2D environment with a focus on pedestrian
movement.
Unlike many other magnets pertaining the phys-
ical configuration of urban space such as the width of
the pavements, transparency of the facades, ground
surface materials etc., solar access is dynamically
changing throughout the day. When using it as
an input to the simulation model, one has to con-
sider the temporal parameters of the model. E.g. if
pedestrian movement is simulated during the morn-
ing rush hour then the solar analysis needs to be car-
ried out within the same hours of the day. However,
simulation can also be carried out within a specific
timeframe (e.g. one day, week, month).
Our tests of solar exposure calculations were car-
ried out by using the built-in analysis tools of Bent-
ley Microstation. This particular package was chosen
because of its broad handling of various geometry
formats from different sources and the possibility of
live links to WMS mapping data from Estonian Land
Board (ELB), and general support for geolocated in-
put data. Also according to our previous experience
with other popular analysis packages, the geometry
needed less repairing and re-meshing to become us-
able.
In order to carry out the study, a GIS database
with 3D building models (copyright: Tallinna Lin-
naplaneerimise Amet) was translated to DWG for-
mat from ArcGIS by the Tallinn City Planning Of-
fice's Geoinformatics and Cartography Department
(Tallinna Linnaplaneerimise Amet, Geoinformaatika
ja kartograafia osakond). To further refine the lo-
cation based solar exposure in terms of diffused
light calculations, a general weather file for Estonia
(so called Base Year data) was converted from Excel
spreadsheet to Energy + .EPW format, usable by sev-
eral climate analysis software packages in addition
to Microstation's built-in tools described above. To
make a streamlined workflow between different soft-
ware packages possible, i.e. easily machine-readable
 data for pedestrian simulation, the results were cus-
tomised by grey-scale values for hourly solar cover-
age, combined with an overlay of building data with-
out the solar analysis in RGB 255 0 0 - leaving only
the street level computations visible, and output as a
1:1000 scaled PNG raster image.
To make a streamlined workflow between differ-
ent softwares possible, i.e. easily machine-readable
data for pedestrian simulation, the results were cus-
tomised by gray-scale values for hourly solar cover-
age, combined with an overlay of building data with-
out the solar analysis in RGB 255 0 0 - leaving only
the street level computations visible, and output as a
1:1000 scaled PNG raster image.
The analysis results clearly demonstrated that
the well exposed street sides in the pilot study area
coincided with preferred routes of pedestrians, as
well as shadowed spaces with more neglected areas
in the urban environment in general. Although the
pilot study area is fairly small, there is little doubt
that we will be able to demonstrate the same corre-
lation between sunny areas and preferred pedestrian
routes with larger scale studies.
A further refinement considering the facade ma-
terials affecting street level glare and other factors af-
fecting pedestrians, although of significance (Hagita
and Mori 2013) has been decidedly left outside the
scope of this study.
THE SIMULATION MODEL
The software prototype for simulating the pedestrian
movement patterns is based on some earlier agent-
based models done by Puusepp (2011). For the pi-
lot study the model is developed in Netlogo that is
well suited for testing early concepts and for build-
ing rapid prototypes. Netlogo works reasonably well
for the size of the pilot study. For the larger scale
project we anticipate some scalability issues, but this
remains yet to be discovered.
While agent-based models have been heavily
used in crowd modelling (Procházka et al. 2015) and
evacuation (Helbing et al. 2002), there have been
only a few examples of using agent-based models
478 | eCAADe 34 - SPATIAL ANALYSIS | Urban Context - Volume 2
for studying pedestrian movement in urban spaces.
One of the most notable is perhaps the STREET model
(Haklay et al. 2001) that focuses on the simulation
of the behavioural aspects of pedestrian movement
in non-congested situations. The STREET model
demonstrates that agent-based simulations are pre-
ferred over traditional pedestrian simulation models
such as using Euclidean distances or shortest paths
through the network of gravity models due to their
flexible and distributed nature. Traditionally, move-
ment studies measure the outcome produced by the
current combination of street configuration and at-
tractor (offices, shops, cafes etc.) locations; such
models are better suited to modelling general pat-
terns of movement and not so well applicable to
model the movement of individuals. According to
the authors of the STREET model "some researchers
argue that the main generator of pedestrian move-
ment is the configuration of the street network it-
self, and that patterns of movement are largely deter-
mined by this configuration, rather than by the dis-
tribution of attractors within the network. This is an
extreme view, which is difficult to sustain without re-
course to ceteris paribus arguments."
Also, seldom are the cases where qualities of ur-
ban space is considered as part of the decision mak-
ing of pedestrian route selection mechanism. In con-
trast to traditional transportation planning models,
agents are directly embedded into the digital repre-
sentation of the studied urban environment and their
route selection mechanism operates on the locally
available data. Such models work better at small spa-
tial scales, where movement of individuals becomes
the focus of the computation. Arguably, local spatial
and environmental qualities have a major impact on
the route selection decisions at this scale.
Simulated pedestrians in our model are posi-
tioned on one particular location (origin) and given
a target (destination) in the beginning of the simu-
lation. They are simple reactive agents that move
closer to their assigned destination by hill climbing
pre-calculated proximity values towards their desti-
nation, but are simultaneously attracted to so-called
 magnets - urban qualities defined by the modeller of This setup reflects on the study that most people do
the simulation. Our agent has a notional body with not choose the shortest paths in the city (Zhu and
three sensors - one in front and two symmetrically on Levinson 2015). We assume that unlike public trans-
sides at a specific angle from the front sensor. These portation users pedestrians are more likely to base
sensors read both the proximity and the magnetic their choice of route on local spatial features. There-
values from the environment. fore, we propose a hypothesis that pedestrians are at-
tracted to some apparent features such as the width
Figure 4
of passages, attractive street frontages, existence of
Barrier map
greenery, exposure to sun, shelter from noise and ad-
showing barriers of
verse weather conditions, as well as to some other
pedestrian
less tractable magnets such as visual variety of urban
movement in the
morphology. Figure 5 shows how the magnetism at-
area- buildings,
tracts pedestrian traffic in the simulation model.
roads and areas not
There are two different kinds of input data to be
directly designated
loaded into the simulation model before the agents
for walking (e.g.
are set loose. The first kind of input is concerned
lawn).
with origins and destinations. In addition to the
above described computation of proximity values to
each designated destination, we also need a origin-
destination (OD) matrix that defines the frequency of
trips between all pairs of origins and destinations.

Figure 5
A test run of the
simulation model
showing how the Proximity values are calculated separately for each
magnetism of the destination and are inversely proportional to the
Viru Keskus topological distance from the target destination, tak-
shopping centre ing into account barriers such as buildings and roads
affects the (see Figure 4). The environment is made of a set of
pedestrian traffic. patches (pixels) with each patch containing a prox-
imity value to each destination point. This value
is propagated from patch to patch with the passed
value lessened in every iteration so that a gradient
field of proximity values is formed. The patches gain
the proximity value only from their immediate neigh-
bours, which means that passing on the values hap-
pens only locally. This type cellular automata mech-
anism has been modelled and described in detail by
Adamatzky (2001). A similar method has also been
used in Daedalus computer program for creating and While we can run the simulation in the qualitative
solving mazes [3]. mode without the OD matrix, it would only help us
In addition to proximity values, agents are also to understand the change in route selection of indi-
attracted to certain magnets in their environment. vidual pedestrians, but not give us the desired data

SPATIAL ANALYSIS | Urban Context - Volume 2 - eCAADe 34 | 479

 about the overall change of the pedestrian traffic Figure 6
numbers in the area. There are a number of ways Machine-readable
to compute the OD matrix. One of the ways to be solar exposure map
considered for the large scale study is based on the overlay on top of
gravity model (Torrens and Alberti 2000) In the grav- the barrier map.
ity model each origin and destination is given a mass
- a number of pedestrians that depart from and ar-
rive to this point. A simple OD matrix can be calcu-
lated taking into account the mass of each OD point
and the Euclidean distances between points. Getting
accurate data about such numbers alone is tremen-
dous amount of work. Ideally we would rely on exact
counting methods but in reality this would need to
be based on a combination of less accurate methods
such as functional mapping of city blocks, data col-
lected from transportation systems (in Tallinn's case
the "green card" data can be used) and available fi- Figure 7
nancial data from enterprises in the area. GPS tracks of
The second kind of data that can be fed into pedestrian
the simulation model represents magnetic values of movement.
spaces. These magnets or urban qualities are either
observed on the site or computed. Magnets can be
inserted into the simulation manually by "painting"
the streetscape with positive (attractive street sec-
tions) or negative (unpleasant areas for walking, e.g.
pedestrian tunnels) magnetic values. Alternatively,
the solar exposure maps (see Figure 6) can be loaded
automatically. In the latter case, brighter colours in
the solar exposure images are directly translated into
positive magnetic values.
Figure 8
FURTHER WORK Tracks of agents in
Initial calibration of the simulation model was car- the simulation
ried out by comparing GPS tracks recorded (see Fig- model, also
ure 7) during outdoor studies with the tracks gener- showing interior
ated within the simulation model (see Figure 8). The paths.
calibration of the simulation model was an iterative
procedure where the strength and positioning of ur-
ban magnets were finely tuned until the tracks left
behind by agents bore strong enough resemblance
to the GPS tracks. In addition, we have established a
mechanism of taking environmental factors such as
sun exposure into account.

480 | eCAADe 34 - SPATIAL ANALYSIS | Urban Context - Volume 2

With our early pilot study we have demonstrated
that it is possible to build an agent-based simulation
model combining outdoor observations and com-
puted qualities of urban space. We will take this ap-
proach further and use synthetic modelling method-
ology to build an agent-based simulation model that
can be turned into a mechanism for future-proofing
urban design decisions, predicting how different spa-
tial and functional scenarios affect pedestrian traffic
in the centre of Tallinn. In order to do so, we will
carry out an extended outdoor study observing how
people navigate the area in reality and simulate it via
computational means. We will carry out qualitative
and quantitative spatial analysis and test the hypoth-
esis of how different spatial features encourage or
discourage the route selection mechanism of pedes-
trians. Understanding spatial magnets allows us to
turn the simulation model into a stimulation model
- an active construction that help to solve issues in-
volved in designing new complex systems (Resnick
1994).
REFERENCES
Adamatzky, A 2001, Computing in Nonlinear Media and
Automata Collectives, Institute of Physics Publishing,
Bristol
Batty, M 2003, 'Agent-based Pedestrian Modelling', in
Longley, PA and Batty, M (eds) 2003, Advanced Spa-
tial Analysis: The CASA Book of GIS, ESRI, New York,
pp. 81-108
Broach, J, Gliebe, J and Dill, J 2009 'Development of a
Multi-class Bicyclist Route Choice Model Using Re-
vealed Preference Data', 12th International Confer-
ence on Travel Behavior Research, Jaipur
Cerrone, D, Lehtovuori, P and Pau, H 2005, 'A Sense of
Place. Exploring the potentials and possible uses of
Location Based Social Network Data for urban and
transportation planning in Turku City Centre.', Re-
port, 1, pp. 1-148
Hagita, K and Mori, K 2013 'Analysis of the influence of
sun glare on traffic accidents in Japan', Journal of the
Eastern Asia Society for Transportation Studies, Rio de
Janeiro
Haklay, M, O'Sullivan, D, Thurstain-Goodwin, M and
Schelhorn, T 2001, '"So go downtown": simulating
pedestrian movement in town centres', Environment
SPATIAL ANALYSIS | Urban Context - Volume 2 - eCAADe 34 | 481
and Planning B: Planning and Design, 28(3), pp. 343-
359
Helbing, D, Farkas, IJ, Molnar, P and Vicsek, T 2002, 'Sim-
ulation of Pedestrian Crowds in Normal and Evac-
uation Situations', in Schreckenberg, M and Sarma,
SD (eds) 2002, Pedestrian and evacuation dynamics,
Springer Verlag, Berlin, pp. 21-58
Mayer, H, Holst, J and Imbery, F 2009 'Human Thermal
Comfort Within Urban Structures in a Central Euro-
pean City', Urban Climate, Yokohama, pp. 2007-2010
Morse, AF and Ziemke, T 2008, 'On the role(s) of mod-
elling in cognitive science', Pragmatics \& Cognition,
16(1), pp. 37-56
Penn, A and A, Turner 2003 'Space Layout Affects Search
Efficiency for Agents with Vision', 4th International
Space Syntax Symposium, London, pp. 9.1-9.16
Penn, A and Turner, A 2002 'Space Syntax Based Agent
Simulation', International Conference on Pedestrian
and Evacuation Dynamics, Duisburg, pp. 99-114
Prochazka, J, Cimler, R and Olsvsevicvcova, K 2015,
'Pedestrian Modelling in NetLogo', in P, Sincak (eds)
2015, Emergent Trends in Robotics and Intelligent Sys-
tems, Springer, pp. 303-312
Puusepp, R 2011, Generating Circulation Diagrams for Ar-
chitecture and Urban Design Using Multi-Agent Sys-
tems, Ph.D. Thesis, University of East London
Resnick, M 1994, Turtles, Termites, and Traffic Jams: Explo-
rations in Massively Parallel Microworlds, MIT Press,
Cambridge, Massachusetts
Tahrani, S, Jallouli, J, Mareau, G and Woloszyn, P 2005
'Towards a Virtual Reality Tool for Lighting', Proceed-
ings of Computer Aided Architectural Design Futures,
Vienna, pp. 115-124
Tiru, M 2011, 'Overview of the Sources and Challenges of
Mobile Positioning Data for Statistics', Report, 1, pp.
1-5
Torrens, PM and Alberti, M 2000, 'Measuring Sprawl',
CASA Working Papers, Paper 27, pp. 2-43
Whyte, WH 1980, The Social Life of Small Urban Spaces,
Project for Public Spaces, New York
Zhu, S and Levinson, D 2015, 'Do people use the shortest
path? An empirical test of wardrop's first principle',
PLoS ONE, 10(8), pp. 1-18
[1] http://peatanav.ee
[2] http://www.bmva.org/apps:pedestrian
[3] http://www.astrolog.org/labyrnth/daedalus.htm
DIGITAL HERITAGE
Prototyping a Temporospatial Simulation Framework:
Case of an Ottoman Insane Asylum

Augustus Wendell1 , Burcak Ozludil Altin2 , Ulysee Thompson3

1,2,3
New Jersey Institute of Technology
1,2,3
http://www.njit.edu/
1,2,3
{wendell|bozludil|ust3}@njit.edu

Effectively communicating the history of a complex building site using only

conventional presentation techniques can be limiting. It is a challenge to present
temporal changes in the site simultaneously within a spatial framework. Whereas
2D drawings typically used in architectural historical scholarship are many times
cleaned abstractions, 3D models might create an illusion of "recreated reality."
Both representation types distance the reader from the primary sources rather
than being a mechanism to bring the reader closer to the sources. This paper
reports on an interactive simulation of a fin-de-siecle Ottoman insane asylum in
Istanbul. Our response to these challenges incorporates to the simulation both
the temporal aspect (spanning the life of the asylum and the daily life within) and
primary documents by using temporally aware 3D models and keyed descriptive
markers for building programming.

Keywords: Unity3D, Simulation, Digital Humanities, Game Engine, Temporal,

Architectural History

INTRODUCTION to accomplish what is predetermined, they open up

Digital technologies and digital design tools are
opening new territories in art/architecture histori-
cal scholarship, employing methods and techniques
used in managing complex design processes, design
research, architectural education as well as those in
digital humanities. These initiatives address issues of
digital heritage, digital archeology, digital art history,
digital spatial analysis alike. Digital models and digi-
tized data sets are becoming increasingly accessible
giving scholars methods and media that did not ex-
ist before and potentially changing the ways in which
their research questions are conceptualized and ap-
proached. In other words, rather than being "tools"
new ways to think and to produce knowledge as
well as create simplified representations of complex
knowledge.
Architectural history, by the nature of its subject
and interest, lends itself easily not only to the use
of digitized repositories and images, but also to dig-
ital approaches that include virtual modeling, sim-
ulation, geographic information systems, network
analysis and mapping, discourse analysis, and use of
metadata (Drucker 2013). Scholars are using various
methodologies and approaches to integrate innova-
tive technologies in architectural history, both in the
research phase and during the (re)presentation of

DIGITAL HERITAGE - Volume 2 - eCAADe 34 | 485

 findings. As many studies and projects employ new
tools and produce ever-evolving outcomes in digital
platforms, some theoretical questions and new chal-
lenges arise. The "problematized fidelity between
the model and its referent" has been long discussed
with the concern that the products are not "simple
reflections of reality" but contestable knowledge rep-
resentations (Bonde Maines et al. 2009).
Our proposal reports a work-in-progress tool
that presents and visualizes the evolution and work-
ings of a nineteenth century Ottoman asylum. The
information is gathered from a detailed research of
primary archival and printed documents involving
multiple disciplines: medicine, psychiatry and archi-
tecture. The specific case study is a building com-
pound in Istanbul, the capital of the Ottoman Em-
pire, that was used as the state mental hospital be-
tween 1873 and 1922. The building is originally a
sixteenth century complex by the famous Ottoman
court architect Sinan and is still standing albeit cur-
rently under heavy and damaging restoration. Re-
search shows that during the time of its use as an asy-
486 | eCAADe 34 - DIGITAL HERITAGE - Volume 2
lum, the building changed significantly to enable its
medical function. This project aims to visualize the
complex and disparate findings in an accessible and
interactive way by modeling the findings and adding
the dimension of time. In doing so, the objective is to
demonstrate the medicalization and modernization
of the theory and practice of psychiatry through the
analysis of reorganization of time and space.
VISUALIZING THE ASYLUM
In our project, we faced several challenges to demon-
strate the findings with methods traditionally used in
architectural history: the sources are predominantly
textual; there are very few visual resources describing
various states of the building; and the historical layers
added to the building render conventional visualiza-
tion methods ineffectual. As a response to these chal-
lenges and the resulting complicated presentation of
findings, our project employs an interactive 3D sim-
ulation toolset built within Unity3D, that allows visu-
alizing hypothetical spatial reconstructions and tra-
jectory tracking all continuously referring to tempo-
Figure 1
Unity3D Timeline UI
component.
ral data and incorporating primary documents. The
components of the project are:
Temporal Dimension
Instead of the traditional interest in finding about the
"original" building, this approach looks at the life of a
building and how it changed over time. The method
we propose is to implement a time slider component
into the user interface (UI) in Unity3D (Figure 1). This
component controls the visibility of changes at dis-
tinct time periods. The time slider is facilitated by in-
cluding time information within the model data.
Reconstructed Trajectories
Instead of treating space as an isolated entity, this
approach captures the life in the building. By trac-
ing the movement of the occupiers of the space (pa-
tients, doctors and the staff ), we know more about
the everyday life in the building, as well as the med-
ical treatment that took place through these trajec-
tories. It was assumed at the time that the life and
routines of patients in the asylum were crucial com-
ponents of the healing process. This component is
indicated through graphic spatial description mark-
ers.
Cross-Referencing Data
Instead of using the model as an abstracted represen-
tation of findings, this approach reconnects the pri-
mary sources to the spatiotemporal simulation. The
ability to combine important reference texts and im-
ages with spatial and temporal elements in simulta-
neous viewing directly associates the sources with
the visualization.
TEMPORAL DIMENSION
The inclusion of temporal data in spatial simulations
has been studied and implemented in several do-
mains of architectural simulation design and archi-
tectural history. Construction simulation and plan-
ning in both commercial BIM software and in aca-
demic writings is a well covered topic (Boton, Sylvain
et al. 2011; Clayton, Warden et al. 2002). In BIM soft-
wares like Autodesk Revit, one can design and selec-
tively view data from multiple project phases. The
limitation with these methods is the premeditated
nature of the timeline. For instance, construction
simulation depends on a linear sequence of steps and
BIM phasing assumes a predetermined number and
naming of steps to fit a project sequence into. Within
the topic of building lifecycle analysis, temporal data
has been visualized for more quantitative purposes
(Vaizidou 2007; Linnert 2000).
In scholarly works, 4D timelines have been in-
corporated in multiple asset based methodologies.
Projects developed to display temporal data on his-
toric buildings have previously been limited in their
subdivision of time and the building unit (Alkhoven
1991) or have been proprietary and unadaptable for
new scholarship (Bonde 2009). Architectural histori-
ans have developed projects incorporating temporal
data that resolve some of these limitations. The inclu-
sion of an external database of temporally coded sim-
ulation objects frees the simulations framework from
predetermining the timeline (Kalay et al. 2005; Favro,
Johanson 2010). However, the presence of a single
temporal coordination document limits the partici-
pation of external scholars.
Our application leverages the use of external
model assets with embedded temporal coding. Each
model asset contains metadata including a begin-
ning and an end date as well as a location. The benefit
of this approach is to keep the timeline open ended
and adaptable to different external asset inputs, in-
cluding shared resources shared resources from mul-
tiple scholars. The dynamic structure of the platform
opens the door for shared scholarship in spatial stud-
ies.
Using a system of importer methods within
Unity3D, elements that include pre-specified
"key=value" pairs have custom Unity3D components
added to the imported objects and the information
parsed into the correct fields. A centralized Manager
object is used to locate, connect, and manage all ob-
jects within the scene that have corresponding com-
ponents to build an abstract model for controlling
DIGITAL HERITAGE - Volume 2 - eCAADe 34 | 487
 element visibility at each demarcated period within
the automatically generated timeline slider (Figure
1).
RECONSTRUCTED TRAJECTORIES
Representing movement of people in a space has
been explored albeit less than temporal dimension
and embedding metadata in models. A study that we
have benefited from, "The Virtual Monastery," uses
interactive access analysis diagrams in representing
the individuals within the spaces of the abbey (Bonde
2009). In that case, the diagrams are not integral
parts of the 3D models but have a separate yet elabo-
rate system. In our project, we aimed to integrate the
trajectories directly within the simulation.
There are two major trajectories we can recon-
struct based on the available sources that would help
us understand the workings of the asylum: (a) daily
life: medical programming (doctor visits, administer-
ing medicine etc.) and daily routines (such as eat-
ing, cleaning, sleeping etc.) (b) admission process.
The difference between the two major types of tra-
jectories is whereas the latter is relatively linear in na-
ture, the former is heavily complex and nonlinear. In
other words, daily activities involved many patients
doing different things simultaneously. For instance,
patients in example wards would either duplicate ex-
isting programmatic/trajectorial locations/purposes,
or would not fit into a simple linear trajectory of a
day. Even a well-defined activity like dining is not
easy to diagram as patients from different wards (cat-
egorized according to the severity of mental illness)
meet in one of the two dining halls, as in the case
of the female section. It is not clear from the docu-
ments exactly how this was organized. The duplicate
origin of each daily routine trajectory for these pa-
tients became visually and experientially confusing.
The admission process, on the other hand, followed
a certain procedure that was--theoretically--followed
by all the admitted patients.
Reconstructing the trajectories of the occupants
is managed through the inclusion of combined
graphic and text objects located at key points in
488 | eCAADe 34 - DIGITAL HERITAGE - Volume 2
space. The text is often located at the entrance to a
space, includes the name of the space, information
pertaining to the use of the space and a graphic in-
dicator pointing to the space. These tags are non-
visible dummy objects added in the 3D modeling
stage imported into Unity3D with embedded loca-
tion data in XYZ space, applicable start and end
dates, and the visible text data.
CROSS-REFERENCING DATA: 3D DATASET
Our use of metadata at the 3D asset level builds on
and improves previous approaches keying metadata
to simulation variables (Boeykens et al. 2009; Kalay
et al. 2005). The inclusion of metadata within the
3D model resource in our project allows the sharing
of third party resources that are automatically inte-
grated in the simulation framework with little addi-
tional authoring or conversion.
In the first phase, the project 3D model dataset
is generated in 3D Studio Max. Historic drawings
are scanned into digital raster files, embedded and
traced in Autocad to create vector plans. These Au-
tocad files are imported as DWG files into 3D Studio
Max. 3D Studio Max polygonal geometry is created,
snapping to the imported splines to insure accuracy.
Building codification and dates are organized on in-
dividual layers. (Figure 2)
The 3D model with metadata is exported from
3D Studio Max into an FBX format file, which is a mod-
ern interchange file format with an open specifica-
tion for standardized parser design and coding. The
FBX file is then imported into the Unity3D project for
simulation where the metadata custom properties
added in the 3D modeling phase are parsed by an im-
porter method (Figure 3). These tags are converted
to Unity3D components used in managing timeline
based object display within the UI. The metadata tags
consists of:
• RestoModel A tag that should show up in
the user properties to alert the Unity3D plu-
gin that this model has additional data within
the user properties which the Unity3D plugin
will use to autogenerate a RestoModel com-
Figure 2
3D dataset building
codes and temporal
revisions organized
by layers in 3D
Studio Max.

Figure 3
3D Studio Max
custom properties
metadata.

DIGITAL HERITAGE - Volume 2 - eCAADe 34 | 489


ponent. (RestoModel is the identifier within
the Unity3D framework.)
• Anchor The three-dimensional vector (x,y,z)
that describes where 0,0,0 should be on this
model relative to the simulation.
• Tag A tag that allows a model to be marked
as belonging to some sort of special category,
such as title, block name and function. Multi-
ple "key = value" pairs can exist in any given
model.
• Description A string that describes the model
as it relates to the larger project
• Start The starting month, date and four-digit
year that corresponds to when the model
came into existence
• End The ending month, date and four-digit
year that corresponds to when the model
ceased to exist (If not provided, the model
is assumed to be destroyed when another
model in the RestoSet starts)
490 | eCAADe 34 - DIGITAL HERITAGE - Volume 2
Figure 4
Primary document
presentation tab.
Photograph source:
Sihhat Almanaki,
1933.
CROSS-REFERENCING DATA: PRIMARY
DOCUMENTS
By cross-referencing various types of information, we
aim to create a system that binds or connects pri-
mary documents and data sets in meaningful ways
to the simulation. This is done by a set of predeter-
mined metadata tags within a CSV file. We have used
Dublin Core as our starting point and added more
fields that are specific to the project and specific to
the tool. Adopting this widely accepted yet cus-
tomizable standard makes it possible to share infor-
mation among scholars and contribute into the sim-
ulation. Unity3D imports and parses this document
upon simulation launch, recognizes the tags, orga-
nizes and presents the related assets (photographs,
primary documents etc.) to the viewer through an
interactive interface. This use of metadata is similar
to the methods used with temporal and 3D datasets.
Primary documents (texts and images) are dis-
played in the UI combined with the active 3D view-
point. (Figure 4) The graphical UI elements allow
users access to information directly connected to
what they are examining within the 3D scene. By
automating the process of integration into one at
the phase of importing assets, it becomes possible
to reduce the workload of framework users down to
providing only assets and the locational context by
which they should be made available to other schol-
ars. The primary document metadata tags consists
of:
• Format
• Document identifier (for official documents)
• Date (of the document)
• Source
• Publisher
• Rights
• Language
CONCLUSION
By creating and representing connections within our
disparate research findings, we arrive at a simplicity
in understanding a complex analytical ecosystem/-
dataset. Through systematizing the preparation of
data, including metadata tagging and developing
matching importer/parser methods, we have created
a template for visualizing complex research datasets
and realizing potential interconnections beyond ex-
isting relationships/interconnections.
The ecosystem of existing Unity3D tools enables
the future incorporation of VR, aural and agent based
simulations among others. Original scripting de-
velopment within the platform leverages a num-
ber of accessible programming language options
(Javascript, C #) to provide options for further cus-
tom coded methods and tools. Building our simu-
lation does not represent the end of a process, but
rather a stage with opportunities for continued de-
velopment. By capitalizing on the ability to extend
the functionality and to create a platform for shar-
ing information, it becomes possible to use this initial
tool as a starting point in a wide variety of scenarios
where the visualization and manipulation of complex
data necessitates specific and unique solutions.
REFERENCES
Alkhoven, P 1991, 'The Reconstruction of the Past: The
Application of New Techniques for Visualization and
Research in Architectural History', Computer Aided
Architectural Design Futures: Education, Research, Ap-
plications, CAAD Futures ‘91 Conference Proceed-
ings, pp. 549-566
Boeykens, S and Neuckermans, H 2009 'Architec-
tural design analysis, historical reconstruction and
structured archival using 3D models: Techniques,
methodology and long term preservation of digi-
tal models', Joining Languages, Cultures and Visions:
CAADFutures 2009, PUM, pp. 119-132
Bonde, S, Maines, C, Mylonas, E and Flanders, J 2009,
'The Virtual Monastery: Re-Presenting Time, Hu-
man Movement, and Uncertainty at Saint-Jean-des-
Vignes, Soissons', Visual Resources: An International
Journal of Documentation, 25(4), p. 363–77
Boton, C, Sylvain, K and Gilles, H 2011 'Understand-
ing Pre-Construction Simulation Activities to Adapt
Visualization in 4D CAD Collaborative Tools', Com-
puter Aided Architectural Design Futures 2011 [Pro-
ceedings of the 14th International Conference on
Computer Aided Architectural Design Futures / ISBN
9782874561429], Liege (Belgium), pp. 477-492
Clayton, MJ, Warden, RB and Parker, T 2002, 'Virtual con-
struction of architecture using 3D CAD and simula-
tion', Automation in Construction, 11, pp. 227-235
Drucker, J 2013, 'Is There a ‘Digital’ Art History?', Visual Re-
sources: An International Journal of Documentation,
29(1-2), p. 5–13
Favro, D and Johanson, C 2010, 'Death in Motion: Funeral
Processions in the Roman Forum', Journal of the So-
ciety of Architectural Historians, 69(1), pp. 12-37
Kalay, Y, Kinayoglu, G and Kim, SW 2005 'Spatio-
Temporally Navigable Representation and Commu-
nication of Urban Cultural Heritage', Proceedings:
VSMM 2005 International Conference on Virtual Sys-
tems and Multimedia
Linnert, C, Encarnacao, M, Storck, A and Koch, V 2000 'Vir-
tual Building Lifecycle - Giving architects access to
the future of buildings by visualizing lifecycle data',
ICCCBE8, Stanford
Osman, M 1933, Sihhat Almanaki, Kader Matbaasi, Istan-
bul
Vaizidou, M 2007, 4D Building Model - A Conceptual Sys-
tem of the Time Dimension in the Building’s Life, Mas-
ter's Thesis, Vienna University of Technology; Build-
ing Science & Technology
DIGITAL HERITAGE - Volume 2 - eCAADe 34 | 491
Modelling Medieval Vaults: Comparing Digital Surveying
Techniques to Enhance our Understanding of Gothic
Architecture
Nicholas Webb1 , Alexandrina Buchanan2 , John Robert Peterson3
1,2,3
University of Liverpool
1,2,3
{njwebb|alexandrina.buchanan|j.r.peterson}@liverpool.ac.uk

Surveying tools such as laser scanning and photogrammetry are increasingly

accessible, providing opportunities as digital mediators to enhance our
understanding of architectural heritage. Here we discuss and compare the use of
both techniques as starting points to analyse medieval vaults at two sites in
England: Chester Cathedral and Exeter Cathedral. The project is inspired by the
work of Robert Willis, a Victorian scholar who hypothesised how medieval vaults
were designed and constructed; however, he did not have sufficient survey data to
fully prove his theories. We will discuss the accuracy of each digital survey
method in relation to our research that occurred at two distinct scales: the
overall geometry of vault rib arcs where vault bays were several metres in length
and width, as well as more detailed investigations of individual rib profiles where
millimetre accuracy is required. We will compare laser scanning with
photogrammetry in terms of their methodological and practical applications to
architectural heritage in the particular context of medieval vault design, in order
to assess the relative merits of each and aid decision-making as to which method
should be used in specific circumstances.

Keywords: Photogrammetry, laser scanning, point cloud modelling, medieval

vaults, digital heritage

INTRODUCTION we are investigating how these techniques compare

At the University of Liverpool a team of researchers
are studying medieval vault design and construction
in the British Isles (Webb and Buchanan 2016), and
consequently the most appropriate applications of
digital techniques to enhance our understanding of
this topic. Both laser scanning and photogramme-
try offer the ability to create point cloud models and
digital mesh models to a high level of accuracy in
defined environments (Corns 2015, Tallon 2014) and
based on our specific lines of enquiry.
This paper describes a pilot study to digitally
document the medieval vaults at Chester Cathedral,
and a later study at Exeter Cathedral, both in England
(Figure 1). The investigation forms part of a larger
project, where vaults at Wells Cathedral, Ely Cathe-
dral, Nantwich St Mary's Church and Ottery St Mary
Church have also been documented. The study is in-
spired by the research of the Victorian scholar Robert

DIGITAL HERITAGE - Volume 2 - eCAADe 34 | 493

 Willis who encouraged members of the Royal Insti- rib's curvature based on its association with a floor
tute of British Architects to survey medieval vaults in plan, its radius, its centre point(s) and the relation-
order to gain enough information to develop theo- ship of the centre point(s) to the rib's impost line. He
ries regarding their design and construction (Willis also hypothesises how the profile of vault ribs may
1842). However, due to the limitations of manual sur- be distorted on tas-de-charge stones, where ribs di-
veying techniques in comparison to the significant verge to form individual elements, which is unreliable
task of documenting the vaults, usually high up in the to establish by eye. The use of digital surveying tech-
cathedral interior, his challenge was not answered. niques makes this complexity apparent.
The introduction of digital techniques makes it
possible to survey existing buildings accurately and Figure 1
rapidly, and additionally offers the potential to pro- Vaults from the
vide a new analysis of significant works of architec- Chester Cathedral
ture through the exploitation and manipulation of Chapter House (left)
this data (Di Mascio 2015, Giannattasio et al 2013). In and Exeter
the context of medieval vaults, this has enabled us Cathedral Lady
to gather data for analysis in a matter of days, as op- Chapel (right) for
posed to the same task that would have taken many the large scale
weeks if completed manually, not to mention the investigation.
added risk of human error (Richens and Herdt 2009).
Nevertheless, the digital process still requires care-
ful planning and documentation, not only to deliver
transparency in relation to the methodology used,
but also to ensure that accuracy is provided and that
any errors or deviations are kept to an acceptable
LINES OF ENQUIRY FOR COMPARISON
minimum (Mitchell 1992). This is important, as in-
Although Willis made many suggestions for future re-
vestigations into vault geometry should be assessed
search on the design and construction of the vaults,
against the precision of the digital survey data, as,
we will focus on two lines of enquiry to demon-
like physical copies, there is a risk they may become
strate the potential advantages and disadvantages
a faded facsimile with each version, such as creating
of both photogrammetry and laser scanning at dif-
a digital mesh model from a point cloud model. Con-
ferent scales. In order to examine the curvature of
sequently, for the purposes of transparency and ac-
the vault ribs and relate these to the linear geome-
curacy it is important to be able to verify any data
try of the plan, the line of enquiry at a larger scale
against an equivalent dataset derived by alternative
documented the overall geometry of vault rib arcs,
means.
where vault bays were several metres in length and
Medieval vault design and construction is syn-
width, at the two case study sites; the Chapter House
onymous with the conference call of 'complexity and
at Chester Cathedral, generally dated to c.1250-1260
simplicity'. Whilst medieval vaults may appear to
(Hartwell et al 2011), and the Lady Chapel at Ex-
have simplicity and clarity in terms of their geometry
eter Cathedral, dated c.1280-1290 (Cherry and Pevs-
when analysed visually, the use of digital surveying
ner 2004). To do this we created orthophotos in
and modelling techniques has revealed an underly-
plan and section to assess any significant features
ing complexity demonstrating this is not necessarily
that were not apparent when studying the vaults
the case. For example, Willis discusses in his origi-
by eye. The line of enquiry at a smaller scale in-
nal article different methods of establishing a vault
vestigated Willis's hypothesis that the rib profile of

494 | eCAADe 34 - DIGITAL HERITAGE - Volume 2

 tas-de-charge stones were distorted for ease of con-
struction, where millimetre accuracy was required to
obtain satisfactory results. A respond supporting a
tas-de-charge comprised of three courses of worked
blocks in the Chapter House at Chester Cathedral and
a two-course tas-de-charge above one of the piers of
the south nave arcade at Exeter Cathedral, generally
dated as 1332-1334 (Erskine 1983), were used (Fig-
ure 2). For this investigation we used a combination
of orthophotos and point cloud models to ascertain
the profiles of the ribs as they met at the springing
point as accurately as possible. These studies relat-
ing to scale are vital as it may be apparent that the
survey data becomes more or less reliable depend-
ing on whether photogrammetry or laser scanning is
used, and reiterates the need for accuracy in such re-
search (Kersten 2007).
Figure 2
A pier (right
foreground) in the
south nave arcade
at Exeter Cathedral
was used for the
small scale study.
PHOTOGRAMMETRY PROCESS
The first method used to digitally document the me-
dieval vaults was photogrammetry. The only piece of
equipment required besides the camera was a tripod;
therefore they were logistically simple to transport to
the two sites. A 24MP Sony a77ii camera was used
to take photographs with approximately 200 images
captured at each site. Lenses with 35mm equivalent
lengths of 24mm, 45mm, 75mm and 150mm were
used for different elements of the surveys. Targets
were positioned in the cathedral interiors as com-
mon reference points between photographs (Figure
3), and several measurements were taken between
distinctive features as a method of scaling the images
once they were imported into the photogrammetry
software. A level ground plane was also desirable;
different methods such as plumb lines and spirit lev-
els were tested, however, as yet we have been unable
to find a definitive solution to accurately plot this.
Next the photographs were processed using
Adobe Lightroom, where RAW files were converted
into high quality JPEGs and unsuitable photographs
removed. Following this, the photographs were im-
ported into Agisoft Photoscan to begin the proce-
dure of creating three-dimensional digital models.
Photoscan automatically locates the targets posi-
tioned in the cathedral interiors during photography,
and the measurements taken are input as scale bars
and full three-dimensional co-ordinates for known
points on the ground plane. An accurate camera
alignment is then performed, and tensions between
the targets and scale bars are calculated. Finally, a
bounding box is drawn around the area to be inves-
tigated, which is then processed into and exported
as a high-density point cloud or mesh model. With
photogrammetry, getting the correct settings when
photographing the spaces was particularly impor-
tant; this was an issue at the Chester Cathedral pilot
study, which will be discussed later on. We also found
that capturing the geometry of the spaces carefully
could be problematic and had to be well planned
in advance; too many photographs resulted in the
Photoscan software crashing on a MacBook Pro with
DIGITAL HERITAGE - Volume 2 - eCAADe 34 | 495
 16GB of RAM, whereas using too few photographs
produced poor results.
LASER SCANNING PROCESS
The laser scanning was performed using a Faro Focus
3D X330, mounted on a tripod for most scans (Figure
3). The tripod and the two cases holding the scanner
and its proprietary spherical targets can be carried
by one person if necessary, or comfortably between
two people. Several scans were taken of the medieval
vaults; at the Exeter Lady Chapel and Chester Chapter
House two scans were taken in each vault bay with an
additional scan in the centre of each space for con-
text. Two scans per bay were taken to ensure that
details were captured along the entire width of the
rectangular bays. Two scans were taken of the col-
umn in the nave arcade at Exeter. Similarly to the
photogrammetry process, all laser scans were taken
from ground level. This did result in some small areas
of the scan being obscured or less detailed as they
could not be seen from below, however, these were
not significant enough to affect our research ques-
tions and justify using scaffolding or a basket crane.
Spherical targets were placed in the interior for pro-
cessing later on. At Chester we chose not to cap-
ture colour data, whereas at Exeter we decided on full
496 | eCAADe 34 - DIGITAL HERITAGE - Volume 2
colour scans, where the laser scanner captures colour
using the available light.
Faro's own scan software, Scene, was used to cre-
ate point cloud and mesh models. Once imported, Figure 3
the spherical targets were tagged in each scan, and Photogrammetry
through a registration process they were combined targets positioned
into a single point cloud model. We could then on the wall of the
create orthophotos; orthographically correct images Chapter House at
in plan and section, as a method of analysing the Chester Cathedral,
overall geometry of the two spaces studied at Ex- with the laser
eter and Chester. Following this, point cloud mod- scanner also
els of the combined scans were exported for further capturing data in
analysis, as well as mesh models. Faro Scene can di- the foreground.
rectly export mesh models of a single scan only, and
tests using other software such as Meshlab to con-
vert point clouds to mesh models proved unsatisfac-
tory, therefore mesh models of single scans were im-
ported into Rhino and combined when necessary for
the research. One of the main issues we had with the
laser scanning process was that resulting orthopho-
tos produced by combining several scans often de-
picted areas of shadow as well as changing light lev-
els that sometimes occurred across different scans; a
downside not considered by the medieval masons in
their quest for light, well-lit spaces.
LINES OF ENQUIRY: LARGE SCALE SUR-
VEYS
The first line of enquiry investigated the overall ge-
ometry of the vaults at the Chester Chapter House
and Exeter Lady Chapel in plan and section, focusing
on the position and curvature of the ribs. It is widely
accepted that masons first traced the plan of the
vault full scale on a tracing floor and then projected
arcs upwards to form the three dimensional geom-
etry, rather than making scale drawings or plotting
the full design beforehand (Porter 1997). Therefore,
having access to orthographically correct plans and
sections enabled us to interrogate the design and
construction processes. As mentioned previously, we
used orthophotos created from both photogramme-
try and laser scanning.
Figure 4 shows orthophotos in plan and section
Figure 4
Comparative
images of surveys
taken at Chester
Cathedral (left) and
Exeter Cathedral
(right), with the
photogrammetry
survey shown
above, the laser
scan survey in the
centre and the two
surveys overlaid to
demonstrate their
accuracy below.

DIGITAL HERITAGE - Volume 2 - eCAADe 34 | 497

 of the Exeter Lady Chapel and the Chester Chap-
ter House. Overlaying the two and looking at the
spaces as a whole demonstrated that photogramme-
try and laser scanning produce almost identical re-
sults, which could be used for accurate measurement
and investigation of the building features at a large
scale. The fact that both methods produced simi-
lar results provides mutual validation, especially as
this occurred at both case study sites. However, in
some scenarios the usability of the orthophotos be-
comes less successful at a detailed scale, as can be
seen from the highlighted areas in Figure 4. Graining
is apparent on the laser-scan produced images due
to the overlap of scans as mentioned previously, and
the sectional photogrammetry models have a lack
of accuracy in depicting the rib curvature in section,
which appears to be faceted rather than smooth.
The photogrammetry models were most successful
in plan as photographs were taken from the cathedral
floor capturing surfaces directly above. These images
leave us with a precise record of building elements
for analysis, such as sizes of individual rib stones, or
voussoirs, as well as sculptural details on rib bosses.
The plan images of Chester show us how the
apparently rectangular bays are actually parallelo-
grams, and the sectional images reveal how the
ridgeline rises considerably at each end to meet the
window and vestibule respectively; both factors that
are difficult to see by eye. The plan images at Exeter
demonstrate that each bay gets progressively longer
moving from east to west, suggesting poor setting
out at the start and/or a need to meet with an exist-
ing feature, and the sectional images show that the
vaults have a flat ridge line.
LINES OF ENQUIRY: SMALL SCALE SUR-
VEYS
The second line of enquiry utilised a Chester Chapter
House springer and Exeter South Nave springer to re-
view small-scale surveys. This follows up on Willis's
hypothesis that the templates (moulds from which
the ribs were cut) used for the tas-de-charge, where
the ribs are worked from a single block as they begin
498 | eCAADe 34 - DIGITAL HERITAGE - Volume 2
to diverge, did not use projection techniques (Willis
1842). By studying marks on a stone from a ruined
vault at Canterbury Cathedral, Willis suggested that
the masons might create a distorted springing. This is
because some of the individual tas-de-charge blocks
were cut horizontally, rather than radially from the rib
arc's centre as expected from typical arch design (Fig-
ure 5). Consequently, if the rib template is not ad-
justed when laid horizontally, it results in a distorted
rib profile along the tas-de-charge stone. Investigat-
ing this is of particular interest as, besides the initial
studies by Willis at Canterbury, tas-de-charge distor-
tions have also been documented by Pérez De Los
Ríos and Rabasa (2014) at Narbonne cloister vaults,
Valdecristo Charterhouse, La Trinidad Convent, Per-
pignan Chapter Hall and Lonja de Palma, as well as
documentation by Vidal (2011) at Santa María de
Huerta Monastery. However, it is yet it is unknown
whether this was universal practice.
To test whether or not this occurred at our case
study sites, we firstly had to study the orthophotos
created previously to ascertain whether or not cuts
were radial or horizontal. Working with the colour im-
ages for Exeter made mortar joints unclear, therefore
the colour data was discarded for this line of enquiry.
Examining the orthophotos of a selected springing in
the Chester Chapter House and another in the nave
aisle at Exeter clearly showed in both cases that their
tas-de-charge blocks were cut horizontally (Figure 5).
The next step used the digital data to take sec-
tions through the diagonal cuts, or mortar joints, of
the transverse ribs to establish the profile and then
test these against the tas-de-charge horizontal sec-
tion. If the two matched, this would support Willis'
theory that the ribs were distorted based on the ma-
sons not adjusting the rib templates. For this investi-
gation the point cloud models were most effective as
the mesh models had a faceted profile making them
less reliable. Once flattened, the points were traced in
red using MicroStation V8i to create the rib profiles,
and then compared by overlaying the different ex-
amples over each other. Figure 6 shows comparative
data captured at Chester using both laser scanning
Figure 5
At both Chester and
Exeter, horizontal
cuts are apparent in
tas-de-charge
stones when
studying the
orthophotos
created from the
survey data. In
addition to
tas-de-charge stone
cuts, the images
also demonstrate
the locations of
radial rib cuts.

and photogrammetry. The images show that results
for the photogrammetric study at Chester were un-
reliable, whereas the laser scanned version produces
an accurate rib profile for tracing. It is important to
state that the data captured at Chester was intended
for the large-scale investigations only, and whilst the
laser scan data also proved adequate for more de-
tailed investigations, we realised that for photogram-
metry to be effective a more strategic approach was
required.
At Exeter, based on our previous findings, we
selected a single pier and produced a photogram-
metric survey of it. As a strategy, the images of the
tas-de-charge and vault ribs springing from the aba-
cus were more focussed, and consequently the wider
context was ignored. The data was then processed
alongside the laser scanned version of the pier. Data
from Exeter in Figure 6 shows that the laser scanned
tracing of rib profiles was still more accurate than
the photogrammetric survey. This is particularly no-
ticeable for the upper rib cut, where the laser scan
data shows all the intricacies of the rib profile, while
the photogrammetry data shows the general shape
of the rib profile, but the details of the ribs are lost.
However, when we compare the profiles of the main
pier below the abacus line, we can see that the pho-

DIGITAL HERITAGE - Volume 2 - eCAADe 34 | 499

 Figure 6
Cuts taken through
mortar joints of
vault ribs and
tas-de-charge
stones are
compared. As the
most accurate
representation, the
laser-scanned
profile is then
traced in red and
overlaid with the
photogrammetry-
produced
images.

togrammetry version is comparable to the laser scan
version (Figure 6). This is because of two factors,
firstly the pier shafts are lower than the ribs and sec-
ondly the pier itself features in the majority of digi-
tal images used for photogrammetry, therefore it is
documented much more thoroughly than the ribs.
This tells us that photogrammetry could be a viable
method of investigating tas-de-charge distortions if
we could get close enough to the ribs using scaffold-
ing or a basket crane, however, as we have the use of
a laser scanner this method is still the most successful
in terms of both results and logistics.
Based on the results obtained primarily from
laser scanning we can state that the standard rib cuts
matched those of the tas-de-charge blocks in the se-
lected features at both Chester and Exeter. This sup-
ports Willis's theory that the overall rib profile is dis-
torted at the springing, probably, according to his in-
terpretation, for ease of constructive practice when
this construction shortcut would be invisible to the
viewer.
RECOMMENDATIONS AND REFLECTIONS
The surveys captured at Chester Cathedral were part
of a pilot study, whereas those captured at Exeter

500 | eCAADe 34 - DIGITAL HERITAGE - Volume 2

Cathedral occurred after further trials at several other
sites. Therefore we had the chance to refine our
methods and consequently, we should take into
account that the photogrammetry data at Chester
could be improved if we were to survey it again.
For example, Chester showed us that to study the
tas-de-charge using photogrammetry required an in
depth survey of the feature itself, rather than relying
on wider contextual data. Consequently, the pho-
togrammetry data at Chester was not representative
of that developed at additional sites such as Exeter.
On the other hand, the laser scan data proved to be
successful at all scales and incidentally, we decided
to lower the resolution of our scans for later sites
from approximately one point every one and a half
millimetres to one point every three millimetres, as
the datasets we were producing were too large. This
again demonstrates the accuracy that the laser scans
provide.
As a methodology, particularly when scans of
cathedral interiors are taken from the floor, we pro-
pose that both laser scanning and photogrammetric
survey methods are viable for producing orthopho-
tos to investigate the overall geometry of medieval
vaults, and that photogrammetry provides higher
quality orthophotos to study vault details such as rib
bosses. However, when using point clouds in three-
dimensional modelling software it is clear from the
research presented here, as well as data from other
sites such as Wells Cathedral, that photogrammetry
is not sufficient at capturing the intricacies of rib pro-
files when photographs are taken from the cathedral
floor. Therefore, the use of laser scanning is the su-
perior option, particularly as it gives very accurate re-
sults from several metres away.
We intend to continue documenting sites for the
project using both methods, since they can be car-
ried out in parallel and essentially, each technique
can be used as validation of the other in terms of
accuracy. Our next step is to investigate using the
photogrammetric models as wrap around images for
the laser scan models, therefore combining the su-
perior visual images of the photogrammetry with
the accuracy of the laser scans. Most importantly
to the project, the use of both laser scanning and
photogrammetry has enabled us to enhance under-
standing of medieval vaults, and support arguments
regarding of their design and construction initially
put forwards by Willis over 175 years ago.
ACKNOWLEDGEMENTS
We are grateful to the Paul Mellon Centre for Studies
in British Art for their research support grant to sur-
vey the vaults at Chester Cathedral, as well as funding
from the University of Liverpool's Interdisciplinary
Network Fund to survey the vaults at Exeter Cathe-
dral.
REFERENCES
Cherry, B and Pevsner, N 2004, Devon (Pevsner Architec-
tural Guides), Yale University Press, New Haven and
London
Erskine, A (Ed) 1983, 'The accounts of the fabric of Exeter
Cathedral, 1279-1353', Devon and Cornwall Record
Society, 26, p. xxxi
Giannattasio, C, Grillo, SM and Vacca, G 2014, 'The
medieval San Francesco convent in Cagliari:
From the architectural, Materical and Historical-
Stratigraphical analysis to the information system',
International Journal of Heritage in the Digital Era,
3(2), p. 413–429
Hartwell, C, Hyde, M, Hubbard, E and Pevsner, N 2011,
Cheshire (Pevsner Architectural Guides), Yale Univer-
sity Press, New Haven and London
Kersten, T 2007, 'Virtual Reality Model of the Northern
Sluice of the Ancient Dam in Marib/Yemen by Com-
bination of Digital Photogrammetry and Terrestrial
Laser Scanning for Archaeological Applications', In-
ternational Journal of Architectural Computing, 5(2),
pp. 340-354
Di Mascio, D 2015 'Analytical drawings of architectural
built heritage', Proceedings of EAEA12, pp. 199-208
Mitchell, W 1992, The Reconfigured Eye, Massachusetts
Institute of Technology
Porter, T 1997, The Architect’s Eye, Taylor & Francis, New
York
Richens, P and Herdt, G 2009 'Modelling the Ionic Capi-
tal', Proceedings of eCAADe 2013, pp. 809-816
Pérez De Los Ríos, C and Rabasa, E 2014 'Stretched tem-
plates in gothic tas- de-charge construction', Pro-
DIGITAL HERITAGE - Volume 2 - eCAADe 34 | 501
 ceedings of the First Construction History Society Con-
ference, pp. 333-342
Tallon, A 2014, 'Divining Proportions in the Information
Age', Architectural Histories, 2(1), pp. 1-14
Vidal, RM 2011 'Bóvedas sexpartitas: Traza, estereotomía
y construcción. Monasterio de Santa María de
Huerta', Actas del Séptimo Congreso Nacional de His-
toria de la Construcción, pp. 821-832
Willis, R 1842, 'On the Construction of the Vaults of the
Middle Ages', Transactions of the Royal Institute of
British Architects, 1(2), pp. 1-69
[1] Corns, A 2015, '3D Icons Ireland', http://www.3
dicons.ie/process
[2] Webb, N and Buchanan, A 2016, 'Tracing the Past
', http://www.tracingthepast.org.uk

502 | eCAADe 34 - DIGITAL HERITAGE - Volume 2

Behavioural Simulation for Built Heritage Use Planning
Davide Simeone1 , Ugo Maria Coraglia2 , Stefano Cursi3 ,
Antonio Fioravanti4
1,2,3,4
Sapienza University of Rome
1,2,3,4
{davide.simeone|ugomaria.coraglia|stefano.cursi|antonio.fioravanti}@uniroma1.it

This paper presents a system for simulating human behaviour in built heritage
artefacts aimed at supporting the decision-making processes for their possible
re-use.Its goal is to predict the mutual influence between the occupancy
phenomena and the architectural heritage environment, in order to optimise the
balance between efficiency requirements of spaces and preservation needs of the
heritage artefact. The proposed system is based on the integration of a BIM
environment with a game engine that allows the modelling of the built
environment and the simulation of its use phenomena at the same time. A central
role in the systems is played by the distribution of Artificial Intelligence among
Virtual Users, process entities (the activities) and the building components,
ensuring the coherent representation of the use processes and the direct
measurement of their impact on the artefact.

Keywords: Built Heritage, Human Behaviour Simulation, Agent-Based

Modelling, BIM

RATIONALE the intervention process aims at improving techni-

In the continuous evolution of cities, identification,
conservation, and adaptation of built heritage are rel-
evant aspects of their cultural and urban develop-
ment, both in the case of acknowledged historical
towns and more recent urban areas. Especially in Eu-
rope, a long debate arose about the possibility of re-
use to give new functions to heritage buildings. How-
ever, despite the necessity and the opportunity of
giving new life to these environments, several prob-
lems arise in terms of preservation and actual capabil-
ities of hosting new functions, users and activities. As
a matter of fact, the goals of preserving architectural
heritage and upgrading and re-using those spaces
seem to be often in conflict. This occurs whenever
cal performances, forcing new uses into an existing
building without taking into account the issues of
compatibility. In this context, specificity and unique-
ness of heritage artefacts have highlighted the lim-
its of current approaches, mainly based on the use
of too general norms and regulations. As a result, at
present, it is quite difficult for designers to conceive a
design that effectively balances preservation and re-
use instances.Today, digital models play a key role in
making heritage buildings part of the contemporary
urban life by encouraging the convergence of past
and present knowledge to facilitate their interpreta-
tion, allowing the virtual reconstruction of their con-
figurations over time as well as supporting the prefig-

DIGITAL HERITAGE - Volume 2 - eCAADe 34 | 503

 uration of their future (Kalay 2008). Despite the intro-
duction of several virtual simulation tools in built her-
itage, their main applications in this field are for com-
munication purposes (Serdar 2015; Boeykens 2011).
Simulations (being often just animations) have been
developed and introduced in museums and archae-
ological sites to visualise and communicate past uses
of the artefacts, sometimes with the possibility of
interaction between the users and the virtual non-
player characters, in order to enhance their learning
attitude (Affleck 2005).Nevertheless, the important
task of predicting and evaluating the capabilities of
a built heritage artefact hosting new users and uses
is still unsupported and completely left to designers'
expertise and imagination.To overcome this lack, the
objective of the research presented in this paper is
the development of a simulation model for the pre-
diction of occupancy phenomena related to the re-
use of built heritage, taking into account both the ef-
ficiency of the assigned new functions and their pos-
sible impacts on the artefact and its preservation.The
proposed model has been applied to different case
studies concerning architectural heritage planning,
each of them encompassing its own features, preser-
vation requirements and intended uses. As an ex-
ample, the model was tested in supporting decisions
about the different use scenarios of a renaissance
cloister in Rome (currently part of Sapienza University
Campus) in order to improve its efficiency and usabil-
ity as well as reduce mutual interference in activities
besides the risk of damages and decay.
STATE OF THE ART
Virtual prediction of built environment occupancy is
a quite popular topic in the CAAD world, even if it
is mainly focused on behavioural simulation in not-
yet-built environments. A big shift in this field was
the spreading of the Agent-Based Modelling (ABM)
paradigm in the early 90s. Since then, much re-
search has been carried on its application to architec-
tural design processes, in particular for the prediction
of people movement in special typologies of build-
ings including airports (Krijnen 2009), shopping malls
504 | eCAADe 34 - DIGITAL HERITAGE - Volume 2
(Zhu 2016 ), railway stations (Lee 2001), and urban
spaces (Yan 2004; Batty 2001). These applications rely
on ABM traditional principles, considering each per-
son as an autonomous entity provided with a lim-
ited set of behavioural rules which control and gen-
erate its behaviour, according to specific objectives
and to the surrounding context (environment and
other agents) (Macal 2007). While this approach has
already shown excellent results in simulating simple
autonomous behaviours - such as spatial mobility-
some limits emerged in their application to more
complex systems of actions, activities and coopera-
tion among people which are the cardinal elements
of a building use process. More recently, some re-
search has focused on the simulation of systems
of activities within buildings rather than just move-
ment prediction. Wurzer (2010) proposed an agent-
based system for early stage spatial planning while
the ISSUU system, introduced by Tabak (2004), relies
on an automatic, time-based organisation of activi-
ties and on their subsequent simulation. Integrating
ABM and process-driven approaches, Simeone (2012
and 2013) and Schaumann (2015) presented another
methodology - defined as Event-Based Modelling -
that relies on AI distribution not only for virtual users
but also for process entities - the Events - in order to
control and coordinate the performing of activities
in a coherent and structured process. The objective
of all these approaches was to support design deci-
sion about not-yet-built environments and to predict
and evaluate the impact of a proposed design on ac-
tivities and operations of the intended users of the
building. In addition to the operational dimensions,
these methodologies can also provide useful feed-
backs on other use-related phenomena, such as en-
ergy consumption and environmental control.While
in the field of new buildings there is an extended lit-
erature on this topic, only some research has focused
on ABM for heritage buildings, which was mainly lim-
ited to simulation of fire egress and similar safety is-
sues. At the date of this paper, a simulation model to
predict the mutual influence between the intended
re-use processes and the built heritage artefact it-
self, and to support the assessment of the impacts on of them made of its geometrical representation, its
both the users (and their operational needs) and on non-geometrical features, and its behaviour. There-
the artefact preservation, is still missing. fore, the simulation platform is a distributed intelli-
gence system, where AI and computational resources
SIMULATING USERS' BEHAVIOUR IN BUILT are assigned to (Figure1):
HERITAGE ENVIRONMENTS
• V-Users entities (as in current agent-based
Capitalising previous research on event-based mod-
systems);
elling (Simeone et al., 2012, 2013; Schaumann et
• process entities, events and scenarios (as in
al., 2015), the proposed model for behavioural sim-
activity-based modelling);
ulation for built heritage relies on simulating in a
• entities that are part of the built environment
virtual environment the use narratives constructed
(similarly to cellar automata systems).
through a combination of actors, activities and built
environment entities. Differently from pure Agent-
While the V-Users AI is oriented to the generation of
Based models, in which each Virtual User (V-User)
their autonomous behaviour, the built environment
performs an autonomous behaviour, in the event-
entities are provided with computational capabilities
based modelling this autonomy is reduced and par-
in order to monitor, measure, assess and update the
tially influenced by process AI entities, the events,
status of the built environment in accordance with
that control cooperative performing of activities and,
the impacts of the V-Users behaviour. The core of this
at a higher level, the coherence of the entire simu-
approach is the act of modelling the system building-
lated use process and of its adherence to real world
use-users as a set of agents, rather than having a "pas-
phenomena. This methodology is particularly effec-
sive" environment and V-Users acting in it. Relying on
tive in prediction of use phenomena in built her-
specific use scenarios developed by designers in col-
itage artefacts since it suits the articulation and com-
laboration with clients, the model is populated with
plexity of heritage buildings re-use, generated by
V-Users and "placed" activities indicating where an
users' behaviours, movements and activities. If com-
activity is going to be performed. Once activated, the
pared to event-based modelling and other simula-
"placed" activities can simulate their own performing
tion paradigms, an element of novelty, peculiar for
by controlling and coordinating V-Users' actions un-
the context of architectural heritage, is the focus not
til completion . In order to guarantee the coherence
only on the environment influence on users' activities
of the entire simulated scenario, the system is pro-
but also on the impact of re-use on the heritage arte-
vided with a Narrative manager which directly con-
fact preservation. Therefore, the built environment
trols and coordinates the V-Users behaviour accord-
has been modelled considering two different enti-
ing to the predefined temporal structure of activities.
ties typologies that refer respectively to the spaces
In this way, the autonomous and sometimes arbitrary
domain and the technological components articu-
actions of the V-Users are controlled and included in
lation. While previous work on event-based mod-
a more coherent simulation of the use of the built
elling mainly considered the environment domain in
heritage artefact. While, in the previous works, activ-
term of spaces (Simeone, 2012, Schaumann, 2015),
ities entities have been modelled as abstract entities
the inclusion of technological components allows to
for process formalisation, in this application to built
provide them with some behavioural rules to mea-
heritage, they have been provided with some physi-
sure different aspects of the interaction with the in-
cal features such as performing_position and area_-
tended use and users.This assumption relies on the
of_influence. The inclusion of these two aspects en-
idea of a more comprehensive model of the building
hances the phase of planning use scenarios. Archi-
and its use as a system of interacting entities, each
tects and managers can populate the virtual built en-

DIGITAL HERITAGE - Volume 2 - eCAADe 34 | 505

 vironment by placing different typologies of activi-
ties in accordance with the intended future use, while
the area of influence geometry shows the areas of
the environment that will be spatially influenced by
the performing activities. Both are controlled by pa-
rameters and can dynamically vary during the simu-
lation according to contextual occurrences and to the
status of the surrounding environment. Each activity
entity is provided with a programming script that 1)
verifies the conditions that are necessary for its per-
forming, 2) executes the set of related actions, and 3)
updates the status of the different entities involved.
Through this script, it is also possible to represent and
manage unplanned activities: a set of rules can con-
trol the activation of these activities only when and
where some conditions are satisfied during the sim-
ulation itself (Figure 2).
Smart building components for use impact
assessment
Another relevant aspect of the proposed model is the
choice of modelling the different components of the
artefact not as static, passive entities but as "smart"
elements provided with their own set of behavioural
rules (see also Lee 2007). This set of rules, imple-
mented by means of specific behavioural scripts in
the game engine, do not actually control behaviour
but they rather assess the variation of specific param-
506 | eCAADe 34 - DIGITAL HERITAGE - Volume 2
eters related to preservation of the element. While
the simulation is running, the built environment en-
tities, equipped with AI resources similarly to the V-
Users and the Activities ones, measure the effects of
users' behaviour on them (and on their preservation)
providing immediate feedbacks to the architect. For
instance, an antique floor can "feel" the loads related
to the users passing and stopping on it, or a fresco-
painted wall can signal that the presence of too peo-
ple nearby can cause damages. In this way, design-
ers and clients can intervene on the configuration of
the environment or on the planned use process (or
both), calibrating and optimising conservation and
use instances. The result is an ad hoc design solu-
tion for the artefact, that keeps into account its speci-
ficity and uniqueness. For this aspect, Building infor-
mation modelling is particularly effective in this con-
text since it allows to include in the artefact repre-
sentation variables and parameters related to non-
geometrical features that can be useful for the use
simulation. Although decay and preservation of an
artefact component are very complex aspects that
embed a large system of variables, conditions and
interactions with other elements and actors, simpli-
fied functions, and algorithms can be used to pro-
vide a qualitative feedback regarding the impact of
intended use and of users' behaviour. The choice of
considering the building components as active en-
Figure 1
The conceptual
framework of the
proposed system: a
BIM environment
integrated with a
Simulation platform
(a game engine) to
allow modelling of
the heritage
artefact and
simulation of its
intended use
processes.
Figure 2
The application of
the proposed
model to the use
simulation of Saint
Peter Cloister
(Rome, XV Century).
In this picture are
highlighted the
main elements of
the model: 1)
V-Users agents
(schematically
represented), 2)
placed activities (in
charge of
controlling the
performing of
collective and
coordinated
tities with their own ability to measure the impact (designed in the XV Century by Sangallo) originally
actions), 3) artefact
of use processes on their conservation, also ensure part of San Pietro Church and now embedded in the
components,
a good extensibility and flexibility of the model: by Engineering Faculty of Sapienza University of Rome.
provided with AI to
adding specific measurement rules, different aspects This environment currently hosts different activities
control
of artefact preservation can be assessed and consid- and users, with several variations during the day and
preservation
ered during the simulation. in case of special events: students and faculty mem-
variables.
bers use it to reach the classrooms located on two of
IMPLEMENTATION AND CASE STUDY its sides and as a resting place during breaks; tourists
From the implementation perspective, the platform visit it since it is well known in the history of the
has been developed using the game engine (Unity city of Rome; different configurations are created by
3D v.5) where agents' behavioural rules, activities- means of mobile partitions for exhibitions or for con-
performing rules, and narrative manager protocols ferences; during the summer, the cloister even hosts
are scripted in C. The artefact and its different el- some movie events and concerts. All these activities
ements can be modelled using a BIM software (in and use destinations rise two problems: the mutual
our experiments we used Autodesk Revit 2016) and interference and the risk regarding Architectural Her-
later exported to the game engine platform, in which itage artefact and its main components such as the
the entities are enriched with behavioural scripts in columns or the classical central font. By simulating
accordance with their typology and the preserva- these situations and, if relevant, their possible com-
tion aspects to consider. In our experiments, the binations, designers and facility managers are now
process of exporting the geometry of elements was able to test different spatial configurations and use
performed through the Unity for Autodesk Revit protocols and policies, assessing for each proposal
1.2 application, recently distributed by Autodesk.The pros and cons in terms of efficiency and built heritage
model has been applied to the simulation of different preservation (Figure 3, 4).
use scenarios of the 'Cloister of San Pietro in Vincoli'

DIGITAL HERITAGE - Volume 2 - eCAADe 34 | 507


CONCLUSIONS
This paper presents the development of a modelling
and simulation approach for prediction of possible
re-use phenomena of built heritage artefacts as well
as their impact on preservation. In the complex
decision-making process regarding interventions on
built heritage in order to give it new functions and
use, the proposed model allows balancing preser-
vation and use instances, providing a way to opti-
mise the design. In fact, it allows to virtually pre-
dict how the building will appear when finally oc-
cupied and in use and, at the same time, to verify
the impact of the intended use phenomena on its
different components. Through this prediction and
508 | eCAADe 34 - DIGITAL HERITAGE - Volume 2
Figure 3
The walkability map
computed through
the navigation
panel of Unity 3D.
assessment phase, architects and all the other ac-
tors involved in the process, including clients, will
be able to test different design solutions and to con-
ceive different use scenarios for the building, assess-
ing for each proposal the actual impacts on the her-
itage artefact. If compared with the complexity of hu-
man behaviour and interaction with a built environ-
ment, especially if considering architectural heritage
artefact, the model has necessarily to rely on some
simplified assumptions that make the large amount
of variables and variations of such phenomena more
manageable. Since the proposed model is currently
conceived to support early planning phases, where
qualitative predictions are more effective than quan-
Figure 4
The simulation of
an art exposition in
the Sapienza
Cloister during
normal functioning
of the faculty
(developed in the
Unity 3D
environment.

titative ones, the simplified assumptions still ensure
a good reliability of the simulation and useful feed-
backs to the different actors involved in the decision-
making process. Future work within this research
scope will focus on additional testing and calibration
of the model, in order to improve adherence to real
phenomena and more accurate estimation of users'
behaviour impact on the built heritage artefact.
REFERENCES
Affleck, J and Kvan, T 2005 'Reinterpreting Virtual Her-
itage', Proceedings of CAADRIA 2005, New Delhi , pp.
169-178
Aydin, S and Schnabel, MA 2015 'Fusing Conflicts Within
Digital Heritage Through the Ambivalence of Gam-
ing', Proceedings of CAADRIA 2015, Daegu, pp. 839-
848
Batty, M 2001, 'Agent-Based pedestrian modeling', Envi-
ronment and Planning B: Planning and Design, 28(3),
pp. 321-326
Boeykens, S 2011 'Using 3D Design Software, BIM and
Game Engines for Architectural Historical Recon-
struction', Proceedings of CAAD Futures 2011, Liege,
pp. 493-509
Kalay, YE 2008, 'Preserving Cultural Heritage through
Digital Media', in Kalay, YE, Kvan, T and Affleck, J
(eds) 2008, New Heritage: New Media and Cultural
Heritage, Routledge, New York, pp. 01-10

DIGITAL HERITAGE - Volume 2 - eCAADe 34 | 509

 Krijnen, T, Beetz, J and de Vries, B 2009 'Airport Schiphol:
Behavioral Simulation of a Design Concept', Proceed-
ings of eCAADe 2009, Istanbul, pp. 559-564
Lee, JYS, Lam, WHK and Wong, SC 2001 'Pedestrian sim-
ulation model for hong kong underground stations',
Proceedings of IEEE 2001, Hong Kong, p. 554–558
Lee, J, Sacks, R and Eastman, CM 2007, 'Specifying Para-
metric Building Object Behavior (BOB) for a building
information modeling system', Automation in Con-
struction, 15, p. 758–776
Macal, C and North, M 2007 'Agent-Based modelling and
simulation: desktop ABMS', Proceedings of WSC 2007,
Washington, pp. 95-106
Schaumann, D, Kalay, YE, Hong, SW and Simeone, D 2015
'Simulating Human Behavior in not-yet Built Envi-
ronments by means of Event-based Narratives', Pro-
ceedings of SimAUD 2015, Alexandria, pp. 7-14
Sehnaz, C and Gülen, C 2008 'Agent-Based System for
Modeling User Behavior in Shopping Malls', Proceed-
ings of eCAADe 2008, Antwerpen, pp. 635-642
Simeone, D and Kalay, YE 2012 'An Event-Based Model
to simulate human behaviour in built environments',
Proceedings of eCAADe 2012, Prague, pp. 525-532
Simeone, D, Kalay, YE, Schaumann, D and Hong, SW 2013
'Modelling and Simulating Use Processes in Build-
ings', Proceedings of eCAADe 2013, Delft, pp. 59-67
Tabak, V, de Vries, B and Dijkstra, J 2004, 'User Behaviour
Modelling', in Van Leeuwen, JP and Timmermans,
HJP (eds) 2004, Developments in Design & Decision
Support Systems in Architecture and Urban Planning,
Eindhoven University of Technology, Eindhoven, pp.
141-156
Wurzer, G 2010, 'Schematic Systems - Constraining Func-
tions Through Processes (and Vice Versa)', Interna-
tional Journal of Architectural Computing, 8(2), pp.
201-217
Yan, W and Kalay, YE 2004, 'Simulating the Behavior of
Users in Built Environments', Journal of Architectural
and Planning Research, 21(4), pp. 371-384
Zhu, W 2016, 'Agent-Based Simulation and Modeling of
Retail Center Systems', Journal of Urban Planning and
Development, 142(1), pp. 1-10

510 | eCAADe 34 - DIGITAL HERITAGE - Volume 2

3D Digital Reconstructions of Lost Buildings
A first critical framing

Danilo Di Mascio1 , Michele Chiuini2 , John Fillwalk3 ,

Pieter Pauwels4
1
Northumbria University 2,3 Ball State University 4 Ghent University
1
danilo.dimascio@northumbria.ac.uk
2,3
{mchiuini|jfillwalk}@bsu.edu 4 pipauwel.Pauwels@UGent.be

This paper presents a first critical framing of 3D digital reconstructions of lost

buildings, hence focusing on reconstructions using 3D models. This critical
framing is constituted by reflections on definitions, concepts, methodologies and
digital technologies. Much experimentation in the digital reconstruction of lost
buildings has produced valuable results during the last twenty years, but the
variety of methods, software types, and purposes should be reorganized into clear
operational protocols to allow sharing of information and collaboration across
disciplines. We have therefore investigated the diverse approaches towards
digital reconstructions of lost buildings and found that digital reconstructions are
not typically aimed at simple visualizations: they are often further developed in
virtual environments, information systems and tools for simulations and analysis.
In this sense they are not products but instruments for further research and/or
heritage conservation. Counter the large number of theoretical positions,
methods and tools specific to different disciplines, such as architecture and
archaeology, we can find the almost total absence of a comprehensive critical
assessment. This paper, which is part of a larger project, intends to open and
articulate a debate on how to define, analyze and critically systematize the
methodologies for the digital reconstruction of lost buildings.

Keywords: Digital reconstruction, lost buildings, 3d modelling, information

systems, virtual environments.

INTRODUCING 3D DIGITAL RECONSTRUC-
TIONS OF LOST BUILDINGS: WHAT, WHY
AND HOW
An analysis of several sources, supported by authors'
experiences in this field, have allowed to identify and
select key aspects of previous projects about digital
reconstructions of lost buildings. References (con-
ference papers, journal articles, books and websites)
have been checked from areas of intersections be-
tween digital technologies with architecture and ar-
chaeology. Moreover, websites of specific research

DIGITAL HERITAGE - Volume 2 - eCAADe 34 | 511

 projects have also been consulted. This preliminary
study has made evident the variety of characteristics
of this research field. Lost buildings and their forgot-
ten stories have always attracted the cultural inter-
ests and curiosity of scholars and lay people world-
wide. An early example in which 3D modeling soft-
ware is applied for lost buildings can be found in
the Old Minster in Winchester, UK [1]. A long list of
similar research initiatives have applied 3D model-
ing and virtual reconstructions to a wide variety of
no longer existing buildings and cities from the me-
dieval church of Cluny III, to Dudley Castle and the
ancient Rome [2]. A digital reconstruction process of
lost buildings is constituted by several aspects and
steps; hence the main question is related to how to
critically systematize this variety of heterogeneous
information. As far as we know, this work repre-
sents the first such critical framing that tries to con-
sider 3D digital reconstructions of lost buildings in a
more comprehensive way. Most of the previous pub-
lications are focused on the description of individ-
ual research projects, however there are some excep-
tions. One of the most relevant is the first chapter of
the publication by Novitski (Novitski, 1998), which is
entirely dedicated to lost buildings. This section of
the book presents descriptions of several case stud-
ies pertaining to various geographic locations, histor-
ical periods and building typologies, including clas-
sic Greek theaters, the Columbian village of Ceren,
the palace of Kublai Khan and the Larkin Building de-
signed by Frank Lloyd Wright. Each project was de-
veloped by scholars and experts with different back-
grounds: students, archaeologists, architects, archi-
tectural historians; every description includes also in-
formation about the main aims, sources and names
of the main software packages used. Nonetheless,
the main aim of Novitski's work is to describe a se-
lection of lost architectural treasures brought back
to light thanks to modern digital technologies, es-
pecially 3D modelling, rendering and animations.
Hence, there is not the explicit objective of sys-
tematizing digital reconstructions of lost buildings.
Other publications are focused on specific typolo-
512 | eCAADe 34 - DIGITAL HERITAGE - Volume 2
gies of buildings and historical periods, such as lost
palaces [3]. This research deals with virtual recon-
structions of historical, courtly architecture from me-
dieval and early modern periods. Other studies (An-
derson et al., 2009) focus on specific technologies
(in this case game technologies applied to serious
games) in a chosen field, cultural heritage. However,
the piece of writing describes serious games in cul-
tural heritage, whereas we aim to focus on technolo-
gies applied to lost building in general.We propose
here to organize the digital reconstruction process by
replying to three main questions: what, why and how
(Figure 1). These three questions define the main as-
pects of digital reconstructions of lost buildings and
at the same time can be used to build up a first crit-
ical frame for research. Next sections will elaborate
each of these three questions, respectively (defini-
tion and analysis), whereas the concluding sections
build up the targeted frame for the digital reconstruc-
tion of lost buildings (discussion and conclusion re-
spectively).
Figure 1
The three main
questions that
address the digital
reconstruction
process of lost
WHAT: LOST BUILDINGS buildings.
The expression "lost buildings" identifies buildings
that have lost their original form and that pertain to
one of the following three categories: partially lost,
totally lost and transformed. The word "building" is
here used in a wide multi-scalar way to include both
single buildings and artefacts, designed by profes-
sionals and vernacular, and settlements from small
villages to cities. However, we are aware of other
definitions that include, for example, unbuilt projects
(Glancey 2015), but these buildings deserve a fu-
ture specific publication. It is important to define a
classification, because for each category a different
methodology may apply. We are proposing three
general categories that represent what kind of lost
buildings can be modelled, namely: totally lost, par-
 tially lost and transformed (see Figure 2).
Figure 2
Diagram that shows
the first phase of
the digital
reconstructions of
lost buildings,
namely the choice
of what is to be
modelled. The
scheme highlights
also two other
aspects that
characterize those
lost buildings:
reason (how they
became lost) and
building/ì/site type.
WHY: OBJECTIVES AND THEORETICAL AP-
PROACHES
It is essential to highlight that the philosophies and
methodologies depend on the objectives of the re-
construction and on the disciplinary areas involved.
For instance a digital reconstruction of a building se-
riously damaged by a recent war may be finalized to
its physical reconstruction. In this case the model-
ing may be following architectural design processes,
which are different from the case of a historical or ar-
chaeological analysis of a building demolished cen-
turies ago. The methodological differences followed
in the digital modeling will thus depend on objec-
tives and disciplinary areas involved, mainly architec-
ture and archaeology, as well as on the extent of the
building remains and its documentation. In other
words, the two fields of reconstruction methodolo-
gies and digital modelling methodologies must be
seen as separate and clearly defined, but they end
up being intricately intertwined in a building infor-
mation system resulting from the modelling process.
The main conceptual, theoretical and methodologi-
cal approaches towards the digital reconstruction of
lost buildings lie within disciplines such as architec-
ture and archaeology. The methodologies of recon-
struction are conceptually related to the restoration
theories and to the methods of architectural history.
This requires any reconstruction to have a scientific
basis, using literary or visual documentation, as well
as physical evidence of the remains on site and other
surviving parts. The typical example is the recon-
struction of archaeological remains, starting from the
building foundations and ruins, which can be identi-
fied and dated thanks to literary documents and arte-
facts found in the excavations. This example leads
the discussion to four crucial questions that repre-
sent likewise lines of investigation:
1. How to represent conjectural parts (missing
parts) from the ones that are certain (existing
remains), how to indicate the level of uncer-
tainty, and how to allow testing different re-
construction hypotheses;
2. How to provide information about the re-
assembly of parts conserved in different
places, for instance when frescoes and sculp-
tures integral with the architecture, but also
objects of daily use, are stored in musea or
private collections. This situation typically
occurs in archaeological projects again (the
Parthenon, or Pompeii houses), but other
cases exist as well, for instance for the Chicago
Stock Exchange;
3. How to document the evolution of the site
over time: archaeology always uncovers
a number of layers belonging to different
phases of existence (while the classic graphic
reconstructions show buildings frozen at a
particular time in history);
4. How to link or refer the reconstruction to the
documentation that forms the scientific basis
and the socio-cultural context of the site. This
is an essential question for academic research
work as well as, in a different form, for cultural
DIGITAL HERITAGE - Volume 2 - eCAADe 34 | 513
 tourism or secondary education.
All these points cover important lines of investiga-
tion about lost buildings, especially the more tradi-
tional ones in archaeology and architectural history,
but represent only part of the studies that can be
undertaken. For example, in presence of totally lost
buildings, the first two questions should be formu-
lated in a different way, because their remaining ev-
idences are the sole archive records, which can be
more or less complete. The lines of enquiry can be fo-
cused on other aspects, such as 3D digital reconstruc-
tion of no longer existing buildings within their urban
context (Martens et al. 2010), or the study of com-
plex geometries of dismantled pavilions, such as the
one designed by Le Corbusier for the 1958 Expo Exhi-
bition held in Brussels (Sdegno, 2012). Moreover, in
disciplines such as architecture, studies on lost build-
ings are not limited to historical investigations. Digi-
tal reconstructions can also be more design oriented
such as in Di Mascio (2010). In this research the mass-
ing model of the fortified center of a medieval vil-
lage, destroyed by an earthquake, was digitally re-
constructed using historical plan views, historical pic-
tures and photos collected during onsite visits (Fig-
ure 3). The digital model has been used to identify
tangible and intangible memories, such as shape and
size of alleys and buildings, and the perceptual as-
pects on the alleys, that can inform the reconstruc-
tion of the fortified center (Di Mascio, 2010). We can
summarize the main objectives of digital reconstruc-
tions of lost buildings under four categories:
1. Historical/archaeological research;
2. Visualization and information systems for cul-
tural tourism;
3. Experiential virtual visit of totally lost sites;
4. Documentation and analysis for conservation
and reconstruction.
From the above, we are able to extend our framework
for digital reconstructions of lost buildings with an
additional decisive step, namely the why of the dig-
ital reconstruction, which consists of a specific over-
all purpose and a preliminary definition of a number
514 | eCAADe 34 - DIGITAL HERITAGE - Volume 2
of particular products that are aimed at. We have
outlined seven types of products, which then result
in the diagram listed below in Figure 4. However,
these products also represent further practical appli-
cations of the 3D digital reconstruction that use spe-
cific methodologies and software packages and as
such they will be further described in the next sec-
tion.
Figure 3
Diagram of the why,
namely the
definition of the
purpose of the 3D
digital
reconstruction. The
image mentions
also the “product”.
HOW: THE DIGITAL RECONSTRUCTION
PROCESS
As previously mentioned, the choice for a specific
digital technology ('why') is influenced by many
factors, from the conceptual and theoretical ap-
proaches, methods and tools within each discipline
such as architecture or archaeology. For this reason,
it is useful to highlight the main phases that charac-
terize the digital reconstruction process and study of
a lost building. We can then indicate to what extent
this digital reconstruction process is realized for par-
ticular types of buildings (what) and particular pur-
poses (why). The following three phases are consid-
ered (inspired by Pauwels and Di Mascio, 2014):
1. Data Collection and Analysis
2. 3D Digital Reconstruction
3. Multifaceted applications of the 3D model
(documentation, analysis, and enrichment)
4. Dissemination
Data Collection and Analysis
Every digital reconstruction starts with a data collec-
tion phase in which the main objective is to gather in-
formation to support the modeling process. Besides
traditional tools (on-site measurements and data col-
lection, both intrusive and non-intrusive), 3D laser
scanning technologies and photogrammetry tech-
niques have become mainstream in the last fifteen
years, and have greatly augmented the quantitative
knowledge of important buildings and sites (see for
example the Cyark archives [4]). Both techniques al-
low digital modeling to start with very accurate geo-
referenced data (e.g. archaeological evidence) of
morphologically complex sites. Every scan acquires
morphological and dimensional information, repre-
sented by point clouds. These documents are con-
verted through specific software packages into 3D
surfaces and meshes or used as reference points to
model 3D objects (surfaces and solids). The passage
from point cloud to mesh remains an important tech-
nical challenge that is considered by a large number
of computer scientists. The scans of remains located
in different locations, also very far from each other,
transformed in 3D objects, allow in a later stage to
digitally reassemble the fragments in a process that
can be called digital anastylosis (Thuswaldner, 2009;
Chiuini, 2011). Because of the large myriad of tech-
niques, devices, software and tools, however, the
granularity, level of detail and quality of point clouds
are often different and not appropriate for the par-
ticular object that is considered (too much or too lit-
tle detail). Furthermore, the fewer the building re-
mains are, the bigger the necessity is to search and
collect other documents. In the case of totally lost
buildings, the only references may be represented
by archive records. The digital reconstructions of the
Synagogues in Vienna, entirely destroyed in 1938, are
example cases of this situation (Martens et al., 2002).
In this latter case, the available drawings are often re-
drawn in vector format, in software such as Autodesk
Autocad, to be used as reference during the 3D mod-
eling.
3D Digital Reconstruction
The main objective of the 3D digital reconstruction
phase is to digitally recreate a 3D model of the con-
sidered object. The whole process is influenced by
three main concerns: the collected data (and the way
of collecting them) during the previous phase, the
features of the object that should be modelled and
further applications of the 3D digital reconstruction
(connected with the purpose of the digital recon-
struction, why). The digital reconstruction also has its
own value because it allows elaborating hypotheses
and evaluating them through 3D modelling; hence it
typically represents a moment of investigation. Lev-
els of detail and objects organization are strictly de-
pendent on the final aims of the project. A digi-
tal model created for the purpose of investigating
the constructive characteristics of a vernacular build-
ing (Di Mascio, 2013) or of an historical skyscraper
(Chiuini, 2011) has a different geometrical complex-
ity, level of detail and organization of elements com-
pared to a model aimed at simulate crowds' be-
haviour along the streets of an ancient city (Maïm et
al., 2007). 3D modeling software, such as 3ds Max
and Maya, are widespread solutions in digital recon-
structions processes. In recent years, also BIM soft-
ware is being applied to heritage projects. However,
the use of a specific software package for the digital
reconstruction process is strictly connected with the
successive uses. The modeling process is a design
problem: the design of an information system that
can make us understand, in its multifaceted complex-
ity, a lost building, site, or place.
Further applications and analysis of the 3D
model
The result of the digital reconstruction process can
often represent the starting point of further work that
DIGITAL HERITAGE - Volume 2 - eCAADe 34 | 515
 can have several purposes. This work can to some ex-
tent be articulated in seven main branches that iden-
tify further applications and analyses of the 3D digital
model: analytical representations, photorealistic vi-
sualizations, information management systems, doc-
umentation, simulations, augmented reality, virtual
reality.
Dissemination
The dissemination represents the last phase, but not
the least important; it can be the main aim of a re-
search project or the final step, where part of the find-
ings are packaged in ways that can be easily experi-
enced and understood by users. The results of the
applications and analysis of the 3D digital reconstruc-
tion constitute the main outcomes useful to spread
the information to a wider audience for different pur-
poses such as cultural tourism and education. All
these products can be displayed in museums, on-
line or included in products such as books or DVDs.
To conclude, the digital reconstruction process typi-
cally goes through four practical key phases, namely
a data collection and analysis phase, a digital recon-
struction phase, a further application of the 3D dig-
ital reconstruction and a dissemination phase. Very
diverse options exist in all these four phases, as is
schematically displayed in Figure 4. In order to make
appropriate choices in all four phases about the way
in which the digital reconstruction process should
take form, one should closely bear in mind the 'what'
and 'why' of the digital reconstruction process.
PRODUCTS OF THE 3D DIGITAL RECON-
STRUCTION
Of course, in many cases, the 3D digital reconstruc-
tions of lost buildings are directly aimed at produc-
ing specific product outputs. These are further ap-
plications of the 3D digital reconstruction that form
part of the reasons why a lost building is digitally re-
constructed. Of course, as can also clearly be seen
in the last section, these products are closely linked
to the actual purpose of the 3D digital reconstruc-
tion. The end products should always be devised and
516 | eCAADe 34 - DIGITAL HERITAGE - Volume 2
produced in service of a particular purpose, whether
this be data collection for conservation or restoration
projects, 3D graphic representations for touristic vis-
its, historic documentation or other. Diverse mul-
tifaceted uses can be identified for a 3D digital re-
construction. On this basis, we are proposing seven
categories that represent further applications and
analysis of the digitally reconstructed lost building.
These are very diverse products. Choosing to aim for
any number of these products should be done soon
enough, in combination with defining the purpose
of the digital reconstruction. Namely, these products
all use some form of 3D or data, but they are very di-
verse in nature and typically require a specific way of
modeling. In other words, in order to produce any
of these products ('why'), one will need a different
digital reconstruction process ('how'). Furthermore,
some products ('why') may be more appropriate for
lost, partially lost or transformed buildings ('what').
Furthermore, each of these applications can be com-
bined with the others.
Analytical Representations
These analytical representations allow to analyze,
document and communicate only a few aspects of
the building. Thanks to these graphic drawings it is
possible to split the characteristics of a building to
better study and represent them individually or in
sub-groups. Consequently, attention is paid to one
or more aspects, isolated from the whole building
and analyzed through one or more graphic works. In
fact, an overall view of the building can limit the anal-
ysis of a particular characteristic. Axonometric ex-
ploded views to study the aforementioned construc-
tive characteristics of buildings are example analyti-
cal drawings. In (Webb et al., 2011), digital represen-
tation techniques, inspired by forensic analysis, are
used to investigate, specific lines of enquiry related
to selected features of a building.
Photorealistic Visualizations
This is the traditional approach that aims at very accu-
rate architectural and realistic visualization of the lost
building, including the lost parts. The reconstruction
Figure 4
The How diagram
with the four
practical key
phases.
of Trajan's Forum in Rome includes an archaeologi-
cal investigation and a sophisticated modeling work
(Packer 2001). A similar purpose has led the digital
reconstruction of the Empire Exhibition held in Glas-
gow (Scotland) in 1938 and subsequently dismantled
[5].
Information Management Systems
Recent work in the field of archaeology, cultural her-
itage and architectural history tends to move be-
yond digital 3D modeling into more specific tech-
nologies for knowledge organization. This includes
data management, ontology engineering, multime-
dia systems, immersive environments, virtual and
augmented reality and semantic information mod-
eling (GIS and BIM). Thematic modeling integrated
with GIS is becoming common practice in archaeol-
ogy. Human interfaces, interaction and usability have
become an essential component of reconstructions.
The 3D model can also be enriched with information
pertaining to any knowledge domain using semantic
web technologies. The model can become an infor-
mation portal, designed to allow updating and alter-
native interpretations.
Documentation
The model allows producing architectural graphics
and other information for historical studies, conser-
vation, restoration or adaptive use. Rendered images
and animations of the digitally reconstructed portion
of the cornice of the Chicago Stock Exchange Build-
ing document the relationship of the terra cotta en-
velope and the steel frame that resulted from the 3D
digital reconstruction phase (Figure 5). The modeling
developed effectively as a heuristic process, which re-
vealed how an entire row of terra cotta blocks had
been damaged during construction because of lack
of coordination between the terra cotta production
and the steel design. When the digital terra cotta
blocks were positioned on the model of the frame, it
appeared that all blocks below the coping had been
chiselled on the back at the same level, which was
confirmed by visual inspection of the fragments. The
portion removed from each piece lined up perfectly
DIGITAL HERITAGE - Volume 2 - eCAADe 34 | 517
 with a horizontal steel angle mounted at the top of experience. The reconstruction of Hadrian's Villa [6]
the steel brackets. This work, perhaps hastily exe- focuses not so much on architectural accuracy, but
cuted while setting the terra cotta on site, cracked the on the analysis of astronomical alignments with spe-
material causing water infiltration over time (Chiuini, cific artifacts on site, using the NASA calendar for
2011). the year 125 CE. Also digital characters (avatars) pro-
vide realism to the experience by moving and talk-
Simulations ing. This type of model can be used at different levels:
Simulations can be interactive (e.g. for training pur- for the general public and for the historian. It is not
poses) or devised for specialist applications (bearing just a visual representation of a space, but an instru-
structures, lighting, acoustics, etc.) and can use dif- ment to analyze the architectural design intentions
ferent technologies such as virtual reality. and historical hypothesis in an interdisciplinary envi-
ronment.
Augmented Reality
Augmented reality applications allow overlapping A CRITICAL FRAMING FOR 3D DIGITAL RE-
the virtual model on a real site using a portable/wear-
able device. In (Madsen et al. 2015) an augmented
CONSTRUCTIONS OF LOST BUILDINGS
The overall diagram displayed in Figure 6 now pro-
reality installation, that includes two interactive de-
vides an excellent resource to compose a visual
vices (a tabled connected to a large TV screen and an
framework in which many existing and forthcoming
hand-held tablet), has been developed to allow peo-
digital reconstruction of lost buildings can be cat-
ple to experience the visual appearance of the lost in-
egorized.The diagram is composed by three main
teriors of the chapel of Koldinghus Castle in Denmark.
parts that represents likewise questions and impor-
The chapel was destroyed and reconstructed several
tant phases of a digital reconstruction process of lost
time and what people can see today is a ruin resulted
buildings: what, why and how. The first column is
of a basic restoration undertaken in the 1970s (Mad-
related to what. Hence, at the beginning it is neces-
sen et al. 2015).
sary to identify the lost building that is to be mod-
elled and its degree of loss. Reason and the identifi-
Virtual Reality
cation of the building/site type are useful to provide
Despite how it may seem, virtual tours have been
both a more comprehensive overview of what is to
around already for several years. In fact, it was pos-
be modelled and clues to search reference buildings
sible to experience, for the first time, a virtual tour al-
that are essential for filling the unavoidable informa-
ready in 1994, in the visitor center in the remains of
tion gaps.The second column refers to the purpose
Dudley castle in the West Midlands, England (Boland
of the 3D digital reconstruction, namely why we are
et al. 1996). This first virtual tour anticipates the re-
undertaking this kind of study. At the same time, it is
cent, and for several aspects, more advanced real-
possible to define the products that we are aiming at;
time virtual and interactive tours using game engines
they represent a concrete reply to the purpose. How-
technologies that have widely spread recently. Dig-
ever, this is still a planning phase that can be updated
ital modelling in general is today also conceived as
and modified afterwards according to the available
a way to embed information in a three-dimensional
documentation.The last part of the diagram is con-
representation of the physical world. The use of
stituted by the four key phases from Data Collection
game engines and VR device allow to navigate and
to Dissemination passing through the actual 3D Dig-
explore the reconstructed buildings in a virtual envi-
ital Reconstruction and its further applications. The
ronment; it can include interactive components and
how phase is oriented in the practical creation of a fi-
metadata. The integration of dynamic systems and
nal product that replies to a specific purpose and dis-
physical simulations such as gravity allow a realistic

518 | eCAADe 34 - DIGITAL HERITAGE - Volume 2

Figure 5
Rendered images
that document the
relationship
between the terra
cotta and the steel
frame of the cornice
of the Chicago
Stock Exchange
Building (Source: M.
Chiuini).

Figure 6
The final diagram
that represents
visual framework in
which many
existing and
forthcoming digital
reconstruction of
lost buildings can
be categorized. The
whole process is
constituted by 6
main steps and
influenced by
disciplinary areas.

seminate the result to a selected audience.
CONCLUSIONS AND FUTURE DEVELOP-
MENTS
This piece of writing presented a first critical framing
of 3D digital reconstruction of lost buildings based on
the result of an examination of example projects and
personal experiences in the field. The big amount
of research about 3D digital reconstruction of lost
buildings, developed in a span of over 30 years, pre-
sented a wide variety of theoretical positions, meth-
ods and tools influenced by disciplinary areas such as
architecture and archaeology. However, counter this
amount and variety of research, we found the almost

DIGITAL HERITAGE - Volume 2 - eCAADe 34 | 519

 total absence of a comprehensive critical assessment.
This piece of writing proposes a critical framing that
suggests a way of organizing these research initia-
tives by replying to three main questions that de-
fine every digital reconstruction process: what, why
and how. We recognized that the category of lost
buildings (what) is not restricted to heritage but con-
sists of every building, including the pavilions of re-
cent world fairs. For this reason the theoretical ap-
proaches, the aims (why), and the technologies and
methodologies (how) can be varied and not con-
fined to archaeology and architectural history, even
if these disciplines represent an important number of
investigation. Moreover, the piece of writing identi-
fies a set of products of the 3D digital reconstructions
that form part of the reasons why a lost building is
digitally reconstructed. These products are the result
of further applications and analysis of the 3D digital
reconstruction; they show, organize, communicate,
deepen and enrich the knowledge about a specific
lost building for several purposes. This first critical
framing constitutes the basis of further studies, such
as a state-of-the-art review.
REFERENCES
Anderson, EF, McLoughlin, L, Liarokapis, F, Peters, C,
Petridis, P and de Freitas, S 2009 'Serious games
in cultural heritage', 10th International Symposium
on Virtual Reality, Archaeology and Cultural Heritage
VAST State of the Art Reports
Boland, P and Johnson, C 1996 'Archaeology as com-
puter visualization: 'Virtual tours' of Dudley Castle c.
1550', Imaging the Past: Electronic Imaging and Com-
puter Graphics in Museums and Archaeology, Occa-
sional Papers 114, British Museum Press, p. 227–234
Chiuini, M 2011 'Digital Reconstruction of the Chicago
Stock Exchange', Digital Media and Its Applications in
Cultural Heritage. Proceedings of the DMACH Confer-
ence, Amman, Jordan, p. 10
Glancey, J 2015, Lost Buildings, Goodman Books, London,
United Kingdom
Madsen, JB and Madsen, CB 2015, 'Handheld Visual Rep-
resentation of a Castle Chapel Ruin', ACM Journal on
Computing and Cultural Heritage, Vol. 9(No. 1), p. 18
Martens, B and Peter, H 2010 'Displacing the frontiers of
reconstructed cultural heritage. Representation of
520 | eCAADe 34 - DIGITAL HERITAGE - Volume 2
the non-existing within an urban context', Proceed-
ings of the 15th CAADRIA Conference, Hong Kong,
China, pp. 63-72
Di Mascio, D 2010 'Preserving memories with digital me-
dia: a methodology for the reconstruction of Castel-
nuovo Village', Proceedings of the 15th CAADRIA Con-
ference, Hong Kong, China, pp. 83-92
Di Mascio, D 2013 'Understanding and Managing the
Constructive Characteristics of Vernacular Architec-
ture. Two raw earth houses', Proceedings of the 31st
eCAADe Conference - Volume 2, Delft, The Nether-
lands, pp. 435-444
Maïm, J, Haegler, S, Yersin, B, Mueller, P, Thalmann, D and
Van Gool, L 2007 'Populating ancient Pompeii with
crowds of virtual romans', Proceedings of the 8th In-
ternational conference on Virtual Reality, Archaeology
and Intelligent Cultural Heritage (VAST, Aire-la-Ville,
Switzerland, pp. 109-116
Novitski, BJ 1998, Rendering Real and Imagined Buildings,
Rockport Publisher, Gloucester, Massachusetts, USA
Packer, J 2001, The Forum of Trajan in Rome. A Study of
the Monument in Brief., University of California Press,
Berkeley, California, USA
Pauwels, P and Di Mascio, D 2014, 'Interpreting metadata
and ontologies of virtual heritage artefacts', Interna-
tional Journal of Heritage in the Digital Era, 3(3), pp.
531-555
Sdegno, A 2012 'Physical and Digital Models for Elec-
tronic Spaces: The 3D virtual re-building of the
Philips Pavilion by Le Corbusier', Proceedings of the
30th eCAADe Conference - Volume 1, Prague, Czech
Republic, pp. 747-754
Thuswaldner, B, Flory, S, Kalasek, R, Hofer, M, Huang, Qi-
Xing and Thur, H 2009, 'Digital Anastylosis of the Oc-
tagon in Ephesos', Journal on Computing and Cultural
Heritage, 2(1), pp. 1-30
Webb, N and Brown, A 2011 'Augmenting critique of lost
or unbuilt works of architecture using digitally me-
diated techniques', Proceedings of the 29th eCAADe
Conference, Ljubljana, Slovenia, pp. 942-950
[1] http://3dvisa.cch.kcl.ac.uk/project12.html
[2] http://www.digitale-kunstgeschichte.de/w/images
/0/0f/Messemer_Heike_The_Beginnings_of_Digita
l_Visualization_of_Historical_Architecture_IN
_Virtual_Palaces_Part_II_2016_p_21-54.pdf
[3] http://www.courtresidences.eu/index.php/publica
tions/e-Publications/#Volume%203
[4] http://www.cyark.org/
[5] http://www.empireexhibition.com/
[6] http://idialab.org/nsf-funded-virtual-simulatio
n-of-hadrians-villa/
Simplifying Architectural Heritage Visualization
AUGMENTEDparion

Derya Gulec Ozer1 , Takehiko Nagakura2

1
Istanbul Kemerburgaz University 2 Massachusetts Institute of Technology
1
www.deryagulecozer.com 2 http://cat2.mit.edu/takehiko/
1
deryagulecozer@gmail.com 2 takehiko@mit.edu

Among other historical artifacts, architectural heritage is the most difficult to

present in museums. There is a need for a high-tech visualization of cultural
heritage since it is important to visualize, share and analyze data for stakeholders
such as historians, archaeologists, architects and tourists. This study aims to
represent architectural heritage in terms of photogrammetry and AR methods for
the Parion Theater, Biga, Turkey, dates back to 1st-2nd century A.D. and has
been under excavation since 2005. The study uses MULTIRAMA, a method
previously developed by ARC Team (MIT) in 2013, which aims to represent the
"unseen" to such users by visualising and documenting via an app. The method
supports architectural heritage representation via the processes of, i)
documentation, ii) data process and modeling, and iii) presentation. This holistic
and low cost approach will focus on the problem of visualizing the digital
architectural heritage, and led light to future projects of a historical visualization
database throughout Turkey.

Keywords: Augmented Reality (AR), Cultural Heritage, Photogrammetry, Parion

INTRODUCTION media effectively. Among these technologies, Aug-

Documenting the inventories of archaeological cul-
tural heritage is very important regarding the knowl-
edge transfer of historical, national and cultural prop-
erties of nations over to the future generations (Benli
and Ozer 2013). In this respect, architectural rep-
resentation is adopting digital methods and digital
data at a fast pace, such as in the case of cultural her-
itage visualization which is often referred to as Digi-
tal Heritage. Within the last 20 years, the paradigm
shift to these new computer technologies makes it
necessary to use various communication tools and
mented Reality (AR) techniques are well known since
they contribute a lot to the representation process.
In this respect, use of AR tools and methods, besides
various sectoral use, is important to study and re-
search of their integration to historic visualization.
Certain artistic objects are difficult to present in
a different context, especially, removing architecture
from its original location to display in a different envi-
ronment would lose a good part of its artistic essence
since it is strongly tied to its context. Alternatively,
use of traditional scale models, photographs, videos,

DIGITAL HERITAGE - Volume 2 - eCAADe 34 | 521

 and drawings is a way of translating the original into
useful representations commonly deployed by archi-
tects. But these methods fragment the building into
isolated forms of different media, and an ordinary au-
dience is often left clueless about the original archi-
tecture. For instance, relating a technical drawing
and its scaled model to the cultural heritage is com-
plicated for people who do not get the architectural
language.
Architectural Cultural Heritage, appears as a
complex field to represent since there is a risk of los-
ing some of its physical characteristics while visualiz-
ing or documenting. In 2006, a symposium was or-
ganized to find solutions to Computer-Based Visual-
ization of Cultural Heritage, and later London Char-
ter was published with its first draft of the principles
and objectives of Digital Cultural Heritage by Hugh
Denard. This charter aims to establish broad princi-
ples for the research and communication of cultural
heritage based on computer-based visualization, in-
tellectual integrity of such methods and outcomes
(Beacham et al. 2006) [1].
Within the literature, early studies have started
with virtual archaeology (Reilly 1990), the credibility
of virtual representations (Ryan 1996), virtual restitu-
tions (Lucet 1997) and a search for alternative rep-
resentations for cultural heritage [3]. Later to the
end of the twentieth century, a new tool called Digi-
tarama has been developed by one of the co-authors,
which is a method placing 3D printed model of Ha-
gia Sophia, visualising the digital models and inner
space with two digital screens attached to two arms
(Nagakura 1997). It was followed by VR and AR stud-
ies as a training tool applications (Bound et al. 1999)
and virtual archaeology studies (Gillings 1999).
The technological approaches have grown
rapidly in the year 2000, with an important book
named Virtual Reality in Archaeology (Barcelo et al.
2000), valuable contributions were made to the field.
It was followed with computer simulations (Pasztor
et al. 2000), acoustical and illuminative archaeolog-
ical reconstructions (Pope and Chalmers 2000) and
immersive walk through experiences (Kadobayashi
522 | eCAADe 34 - DIGITAL HERITAGE - Volume 2
et al. 2000). Ray tracing techniques (De Nicola et al.
2000), virtual reality applications (Goodrick and Hard-
ing 2000; Frischer et al. 2000; Bonfigli et al. 2000) is
also developed in the same year. In 2002, an article
was published on the developing VR applications
in archaeology which shed light to cultural heritage
representations (Frischer et al. 2002). Additional to
these, similar to our work, a work named "Augment-
ing Kashgar" was published (Aydin and Schnabel
2014-2015). Their work is a virtual representation,
documenting a historical village via photogramme-
try. Apart from that, our study is aiming to visualise
an AR via an app which shows different representa-
tions of photogrammetric documentation.
Later, one of the co-author's team in MIT devel-
oped several methods called Deskrama (Nagakura
and Oishi 2006) and Ramalytique (Nagakura and
Sung 2014) (Nagakura et al. 2015) [2]. These meth-
ods were concentrating on virtual representations of
architecture and cultural heritage, and they were pro-
gressing in developing a new method called MULTI-
RAMA later in 2013.
MULTIRAMA (previously Ramalytique) was a
method developed by ARC (Architecture Represen-
tation Computation) Group at MIT in 2013, with the
collaboration of Takehiko Nagakura, Woong-ki Sung,
Daniel Tsai and Howard Burns. It was an interactive
AR interface to represent artifacts or buildings. Later,
in 2015, a new collaboration was made by authors, to
carry out research on Parion archaeological site using
MULTIRAMA.
In this study, a new collaboration was made be-
tween authors who aim to introduce and present
the architectural heritage found in the archaeologi-
cal sites using the "augmented reality" interfaces via
MULTIRAMA. Having been carried out under the con-
cept of "virtual archaeology", the projects have tack-
led the problem both technologically and in a pre-
sentational manner. In this study, an interface is sug-
gested for both the visualization and presentation of
the ancient Roman city, Parion (Biga/Turkey), which is
one of the most important cultural heritages of Ana-
tolia. Relating to the theme of the conference; com-
 plexity vs. simplicity, the visualization is available for
the use of architects, archaeologists, and restaura-
teurs; being user-friendly and an easy interface appli-
cable on tablets for tourists, students; nearly for ev-
eryone who are interested in archaeology. This holis-
tic and low-cost approach will focus on the problem
of accurate visualization and representation in the
cultural heritage of Parion (Ozer et al. 2016).
AIMS AND METHOD
This work has been carried out in three stages: i) doc-
umentation, ii) data process and modeling, and iii)
presentation (see Figure 1). During the documen-
tation stage, we enable the use of photogrammet-
ric methods using 123D Catch software. In the pro-
cessing and modeling stage, we provide tools for cor-
recting 3D photogrammetric images using AR ap-
plication with UNITY software. In the final stage,
the resulting accurate 3D reconstruction of the cul-
tural heritage site is presented to the audience with
a viewer AR application. In MULTIRAMA, the mod-
els are united with other drawings and 3D models in
the AR interface through the use of tablets to look at
physical architecture models (Ozer et al. 2016).
Figure 1
MULTIRAMA
method application
process (Ozer et al.
2016).
DOCUMENTATION
Parion is an Antique Roman city which dates back to
600 B.C. Among many architectural structures, Par-
ion Theater is an important discovery dates back to
1st and 2nd century A.D. In 2005, archaeological ex-
cavations started in the ancient city Parion, locating
on the northwest of the Troas Territory (Biga- Turkey).
Since then, excavations have been continuing in the
south necropolis within the seven areas of theatre,
roman baths, terraced structures, odeon and thermal
plant archeopark fields (Basaran 2013).
Situated on a sloped terrain of the city, Parion
Theater is a usual Anatolian theater in contrast to
Side Theater. Excavations started in 2006, and still
in 10 years, only the scene section is found and doc-
umented. There are surveys on the cavea part, but
none of it has been revealed yet. There are some
assumptions on the shape, form and height of the
cavea and scene part, and the final visualizations are
made depending this hypothesis. We should point
out that, none of this visualizations is certain, they
are only assumptions depending on historical find-
ings and debates with archaeologists in the site.
DIGITAL HERITAGE - Volume 2 - eCAADe 34 | 523
 Figure 2
Planimetric schema
of taking
photographs for
123D Catch.

Figure 3
A comparison of
the real/real (left)
and the virtual/real
(right) of the scene
of Parion Theater.

524 | eCAADe 34 - DIGITAL HERITAGE - Volume 2

Figure 4 stitched model is a mesh surface, and can be con-
Marker image for verted to a Rhino model to clean and work on it. The
MULTIRAMA mesh model was a very good representation of the
interface. virtual/real except for being distorted a few centime-
ters in 10 m. span (see Figure 3). After the file is
cleaned up for mesh holes in Rhino, it gets ready for
3Dprint, in .stl format.
In the final step, to make the AR, a target is
needed as a marker. The AR software will read the tag
(see Figure 4) for the visualization, and show the de-
sired image in the app. For visualization, parion_AR_-
v1.apk file is installed on the tablet or smartphone
which runs on Android. The .apk file was generated
by the team and is reorganized for this project using
UNITY software into an Android device (see Figure 5).
The file was copied onto the device and the file man-
ager app was used to launch the apk file.
In the exhibition, the 3d partial print of the the-
Figure 5 ater is placed on the table with the marker on it. The
Final visualization user who uses the smartphone or tablet can visual-
of the whole site. ize the AR via the app which can be downloaded to
the device (see Figure 6). The user may choose three
different visualizations mode; capture, sections, and
upper structure (see Figure 7).

RESULTS
As in the results; The AR is presented in an ex-
hibition setting in İstanbul Kemerburgaz University
and the web site (www.deryagulecozer.com). The
widespread effect will be increased with the use of
this project results to the academicians, students and
specialists in the related sectors of Digital Cultural
Heritage.
Project results will be executed in two ways:
Firstly, the results will directly contribute to the ar-
chaeological work (Parion, Biga) of visualization / rep-
The site was documented during summer excava-
resentation and restoration / restitution process. In
tions in 2015 with a regular camera. With the help
this related work, re(presentation) / preservation /
of a crane, aerial photos were taken which overlaps
restoration of Cultural Heritage Studies will gather
at each other at least 1/3 of the previous picture. It
speed. Secondly, deployment of VR / AR technol-
should be noted that 360-degree pictures should be
ogy use in digital heritage, with the introduce of this
taken for a good 3D Model (see Figure 2). After these
method to the academia, will ensure the continuity
pictures are taken, they were stitched in 123D Catch
of collaborative academic studies and collaboration.
software, which is free to use up to 70 pictures. The

DIGITAL HERITAGE - Volume 2 - eCAADe 34 | 525

 Figure 6
A user using smart
phone device via
MULTIRAMA and
visualizing the
unseen.

Figure 7
MULTIRAMA Parion
v1.0 Beta Interface.

526 | eCAADe 34 - DIGITAL HERITAGE - Volume 2

Figure 8
Future work of
using AR on site.
The outcome of the research and the benefit to the
related fields are summarized below:
• Use of a low-cost, holistic method utilizing AR
technologies to represent digital heritage,
• Reviving the context connection with the ar-
chitectural heritage since architects usually
documents the heritage via 2D drawings,
• An easy interface and visualization tool for ar-
chaeologists to analyze their works in a 3D set-
ting instead of 2D drawings,
• Providing tourists a different perspective
about the ancient architecture while visiting
archaeological sites,
• Suggesting an easily-usable, understandable
and informative medium with an user-friendly
interface running on a tablet / smartphone
for the children and students interested in ar-
chaeology,
• Contributing to the presentation and preser-
vation of Turkey's cultural heritage by docu-
mentation,
• Providing a means for the reconstruction of
the historical architecture.
FUTURE WORK
There are two future studies which will follow this
project. First is an MIT- Singapore University col-
laboration which will be focusing on structuring a
new database for historic sites around the world, also
some Turkish sites will be included such as Cappado-
cia. The second project is named as "Using Virtual Re-
ality (VR) and Augmented Reality (AR) Technologies
in Architectural Education", which will focus on in-
tegrating Cultural Heritage Visualization to architec-
ture education. This study aims to integrate "Cultural
Heritage Preservation and Documentation" and "Dig-
ital Design", through the development of a low-cost,
holistic method that utilizes VR / AR technologies to
represent Digital Heritage. This project also aims to
visualize the AR on site (see Figure 8).
ACKNOWLEDGEMENT
This study was made during post-doctoral research
in ARC Group (MIT), with a collaboration of authors
in 2015. We would like to thank Nikolaos Vlavianos,
Omer Ozturk, Mustafa Yildizli and Yichen An for
their contributions in the field work and modelling.
We also would like to thank the rest of ARC Group
(MIT) previously collaborated in Multirama project;
Woong-ki Sung, Daniel Tsai and Howard Burns.
The project presented in this paper was funded
by the postdoctoral research grant of TUBITAK 2219
fellowship and Istanbul Kemerburgaz University. We
would like to give sincere gratitudes to Parion Ar-
chaeological Expedition Team, Prof. Dr. Vedat Keles
and Asst. Prof. Dr. Ertug Erturer for providing the
study environment and for their invaluable support.
DIGITAL HERITAGE - Volume 2 - eCAADe 34 | 527
 An earlier version of this study has been published in
ITU AZ journal, DOI: 10.5505/itujfa.2016.66376. For
detailed information, please refer to the website,
www.deryagulecozer.com.
REFERENCES
Aydin, S. and Schnabel, M.A. 2014 'Augmenting Kash-
gar', Virtual Systems & Multimedia (VSMM) Confer-
ence, Hong Kong, pp. 12-15
Aydin, S. and Schnabel, M.A. 2015 'Fusing Conflicts
Within Digital Heritage Through the Ambivalence of
Gaming', 20th Proceedings CAADRIA, Hong Kong, pp.
839-848
Barcelo, J.A., Forte, M. and Sanders, D.H. 2000, 'The Diver-
sity of Archaeological Virtual Worlds', in Barcelo, J.A.,
Forte, M. and Sanders, D.H. (eds) 2000, Virtual Reality
in Archaeology, ArcheoPress, Oxford
Basaran, C. 2013, 'Araştırmalar', in Basaran, C. (eds) 2013,
Antik Troas’ın Parlayan Kenti: Parion, Ege Yayınları, Is-
tanbul
Beacham, R., Denard, H. and Niccolucci, F. 2006 'An In-
troduction to the London Charter', The Evolution of IC
Technology in Cultural Heritage, Papers from the Joint
Event CIPA/ VAST/ EG/ EuroMed Event
Benli, G. and Ozer, D.G. 2013, 'Use of Laser Scanning
for Cultural Heritage Documentation', International
Journal of Electronics, Mechanical and Mechatronics
Engineering, 3(1), pp. 447-454
Bonfigli, M.E., Calori, L. and Guidazzoli, A. 2000 'Nu.M.E.:
a WWW Virtual Historic Museum of the City of
Bologna', SAC Proceedings, Como, Italy
Bound, A.C., Haniff, D.J., Baber, C. and Steiner, S.J. 1999
'Virtual Reality and Augmented Reality as a Training
Tool for Assembly Tasks', Proceedings of 1999 IEEE In-
ternational Conference on Information Visualization
Frischer, B., Abernathy, D., Favro, D., Liverani, P. and De
Blaauw, S. 2000, 'Virtual Reality and Ancient Rome:
The UCLA Cultural VR Lab’s Santa Maria Maggiore
Project', in Barcelo, J.A., Forte, M. and Sanders, D.H.
(eds) 2000, Virtual Reality in Archaeology, Archeo-
Press, Oxford
Gillings, M. 1999 'Engaging Place: a Framework for the
Integration and Realisation of Virtual-Reality Ap-
proaches in Archaeology', CAA Conference, Birming-
ham
Goodrick, G. and Harding, J. 2000, 'Virtual Reality at
the Neolithic Monument Complex of Thornbor-
ough, North Yorkshire', in Barcelo, J.A., Forte, M. and
Sanders, D.H. (eds) 2000, Virtual Reality in Archaeol-
ogy, ArcheoPress, Oxford
528 | eCAADe 34 - DIGITAL HERITAGE - Volume 2
Kadobayashi, R., Nihimoto, K. and Mase, K. 2000, 'Immer-
sive Walk-Through Experience of Japanese Ancient
Villages with the Vista-Walk System', in Barcelo, J.A.,
Forte, M. and Sanders, D.H. (eds) 2000, Virtual Reality
in Archaeology, ArcheoPress, Oxford
Lucet, G. 1997, 'The virtual restitution of the mesoameri-
can site of Cacaxtla: a methodological point of view',
Archeologia e Calcolatori, 8, pp. 129-138
Nagakura, T. and Oishi, J. 2006 'Deskrama', Proceedings of
ACM SIGGRAPH 1998, New York
Nagakura, T. 1997, 'Digitarama', in Sakamura, K. and
Suzuki, H. (eds) 1997, The Virtual Architecture, Tokyo
University Digital Museum, Tokyo
Nagakura, T. and Sung, W. 2014 'Ramalytique: Aug-
mented Reality in Architectural Exhibitions', Confer-
ence on Cultural Heritage and New Technologies 19th
Prooceedings, Vienna
Nagakura, T., Tsai, D. and Choi, J. 2015 'Capturing History
Bit by Bit', eCAADe 33th Prooceedings, Vienna
De Nicola, R., Giordano, A. and Guarino, S. 2000, 'Ray-
Tracing Techniques In A Parallel Environment: The
Case Study Of A Horse Grave', in Barcelo, J.A., Forte,
M. and Sanders, D.H. (eds) 2000, Virtual Reality in Ar-
chaeology, ArcheoPress, Oxford
Ozer, D.G., Nagakura, T. and Vlavianos, N. 2016, 'Aug-
mented Reality (AR) of Historic Environments: Rep-
resentation of Parion Theater, Biga, Turkey', ITU A|Z,
13(2), pp. 0-20
Pasztor, E., Juhasz, A., Dombi, M. and Roslund, C. 2000,
'Computer Simulation in Stonehenge', in Barcelo,
J.A., Forte, M. and Sanders, D.H. (eds) 2000, Virtual
Reality in Archaeology, ArcheoPress, Oxford
Pope, J. and Chalmers, A. 2000, 'Pre-rendering acoustics
and illumination for archaeological reconstructions
Computer simulation of Stonehenge', in Barcelo,
J.A., Forte, M. and Sanders, D.H. (eds) 2000, Virtual
Reality in Archaeology, ArcheoPress, Oxford
Reilly, P. 1990, 'Towards a virtual archaeology', in Lock-
year, K. and Rahtz, S. (eds) 1990, Computer Applica-
tions in Archaeology, British Archaeological reports
(Int. Series 565), Oxford, pp. 133-139
Ryan, N.S. 1996, 'Computer based visualization of the
past: Technical realism and Historical Credibility', in
Higgins, P., Main, P. and Lang, J. (eds) 1996, Imaging
the past Occasional Paper 114, British Museum Press,
London, pp. 95-108
[1] http://www.londoncharter.org/fileadmin/template
s/main/docs/london_charter_2_1_en.pdf
[2] http://cat2.mit.edu/multirama/
[3] http://info.ee.surrey.ac.uk/Personal/R.Bowden/p
ublications/vrsig97old/postscript/014.ps.gz
An Intuitive Heritage Education System for Learning
Architectural Structures and Styles
Focusing on the historical Korean architectures

Jongwook Lee1 , Aram Min2 , Jihyun Lee3

1,2,3
Korea Advanced Institute of Science and Technology
1,2,3
{bellee21|ddmin88|jihyunl87}@kaist.ac.kr

Although national and international attention toward the cultural heritage is on a

rise, there is a lack of public attention toward cultural heritage sites. One of the
main problems is the lack of interaction and the visualization of cultural heritage
on the sites. Especially Korean historical architectures are structurally complex
and are not easily understood by common people. To improve the public
awareness of architectural heritages, we propose a model-based diagnosis system
to educate the visitors and tourists. The system is designed to guide the users to
model an appropriate architecture in accordance with the era, location, and the
usage. For the system, we built a robust set of cases based on the ontological
structure we designed especially for architectural heritage education. It basically
enables users to reconstruct buildings intuitively in six steps from bottom to top.
A system evaluation was conducted on the affective, cognitive, operative aspect of
the system at a heritage site. The results show well in terms of cognitive aspect
but was evaluated poorly in terms of the operative aspect.

Keywords: Historical Korean architecture, JESS rule engine, heritage education

system

INTRODUCTION and the visualization of traditional Korean architec-

As Korea shifts from a developing country to a de-
veloped country, national attention toward the cul-
tural heritage is on a rise. For example, Baekje (AD
892-936) Historic Sites have recently been listed as
one of UNESCO World Heritage and there are ef-
forts to gain public attention toward cultural heritage
sites through cultural festivals. While diplomatic and
public relations efforts advance, the cultural heritage
sites as tourist sites are still under poor conditions.
One of the main problems is the lack of interaction
tures. Today, the base stones are the only compo-
nent that remain along with vague description pan-
els. Just a field of architectural remains does not en-
gage the visitors and does not give much information
and context of how the place used to look like. Ko-
rean historical architectures are structurally complex
and are not easily understood by common, meaning
untrained, people. Even after when people visit the
cultural heritage sites with Korean historical architec-
tures, the understanding toward building structures

DIGITAL HERITAGE - Volume 2 - eCAADe 34 | 529

 is unclear because often, these cultural heritage sites
do not give attention to the architectural structure
themselves.
To overcome the problems of widening knowl-
edge gap between the experts and the common
people, and decreasing public awareness, we pro-
pose a model-based diagnosis system to educate
the visitors and tourists. To develop this system,
we reviewed the benefits of CBR for education,
summarized related systems previously suggested,
and explained the components of Korean architec-
ture. Next, we interviewed the expert to reflect
experts' heuristic knowledge for system implemen-
tation, gathered the historical Korean architecture
cases, and implemented the system. Finally, we ex-
ecuted the system evaluation through user tests for
system enhancement.
BACKGOUND
There have been many researches on the gamifica-
tion of architectural heritage and on ways to edu-
cate the public in virtual environments. For instance,
Champion (2008) made a list of categories of cultural
heritage games which include tourist game, puzzle
games, resource management games, historical bat-
tle games and others. As can be seen, educating the
architectural heritage to the public is being actively
researched to the point where a list of categories are
appearing. Of the categories, the 'puzzle games' is a
game where people solve quests by interpreting re-
mains and epitaphs to reconstruct architectural her-
itage indicating the public's interest toward architec-
tural heritage (Champion 2008). Meanwhile, Juvan-
cic (2008) researched for the optimal interface design
for educating the general public on architecture and
conducted the experiment for the interface design
of eco-spatial education system at a primary school
(Juvancic 2008). The result showed the effective-
ness of interactivity on learning outcomes. However,
there have not been education systems based on the
reasoning of architectural styles nor researches re-
garding the reasoning of architectural styles based
on the usage of the building (such as its purpose),
530 | eCAADe 34 - DIGITAL HERITAGE - Volume 2
geographical location, and built period. While the
architecture education systems are researched, re-
searches on digital reconstruction of architecture are
recently on a rise through the advancement of tech-
nology. For instance, Liu (2015) researched about de-
signing a historical architecture based on architec-
tural style rules of a certain period, and Apollonio
(2011) researched about estimating destructed archi-
tectures based on an architect's pattern book and ar-
chitecture cases (Apollonio 2011). However, there
have not been education system and gamification re-
searches for learning architectural heritage (style and
structure) through digital reconstruction. Ontology
is a useful tool to explain an object's concept and
relationship between the sub-elements of the ob-
ject. Ontology of an architecture can be represented
through components, spatial arrangements, mate-
rials, usage or function, and others. Stefano Cursi
(2015) defines the architectural heritage information,
and in order to explain the multiple context of these
information, the Cursi (2015) suggested an ontologi-
cal structure that are at large classified into the arte-
fact knowledge domain and the historical context
domain. Cursi (2015)'s research aims help the experts
in the field of architecture and preservation to share
and obtain information related to architectural her-
itage.
METHODOLOGY
In order to design a model-based diagnosis system to
educate the visitors and tourists. A model-based di-
agnosis tackles the problem of troubleshooting sys-
tems startingfrom a description of their structure
and function or behavior (Vittorio Brusoni, 1998). It
means tracking the causes (faults) is responsible for
a system failure. we first interviewed an expert to in-
vestigate what kind of historical cases to use. Then,
we built a robust set of cases based on the onto-
logical structure we designed especially for architec-
tural heritage education and we derived tasks, corre-
sponding rules and ways to control the rules.
Figure 1
System Overview.
Knowledge acquisition through expert in-
terviews
An expert interviewed was a researcher from the Na-
tional Research Institute of Cultural Heritage with ex-
perience of 19 years. From interviewing the expert,
we had two findings. First is, since we do not have
access to Chinese and Japanese architecture data in
a given amount of time, we should narrow down the
case library scope to the existing national treasure
architectures from 'after middle age architecture' in-
cluding early, middle, and late eras of Goryeo Dy-
nasty and early, middle, and late eras of Joseon Dy-
nasty. Second is, as the experts do, we should incor-
porate the location of the architecture into our sys-
tem as one of the constraint along with the era and
the usage of the architecture; therefore, the user in-
put information was revised to era, usage, and loca-
tion. For the era, the experts refer to the architectural
styles appeared during the same era and estimate the
architectural style of its golden age. As for the loca-
tion, the experts refer to the architectural styles ap-
peared near the site of excavation. To summarize,
we learned that we should narrow down the case li-
brary to existing national treasure architectures and
that we should include the location information as a
constraint as it plays an important role in identifying
architectural styles.
Ontological approach to well-indexed set of
cases
The ontological structure suggested in this paper
is designed for digital architecture education. As
shown in Figure 2, a 'wooden architectural heritage' is
divided into three classes which are context, compo-
nent, and style. The context class includes informa-
tion that affects the style of the building such as its
period when built, location of its existence, and us-
age or the function. The components class includes
four of the main parts of a building because Ko-
rean architectural heritages are mainly wooden and
have a precast structure. The four parts represen-
tative at each of their vertical placements are base,
bracket, frame, and roof. For example, on the bot-
tom of the building is the base component that in-
cludes elements such as basestones, steps, and sty-
lobates. Likewise, on the top of the building is the
roof component that includes elements such as dori,
rafter, and rooftile. Lastly, the style class includes all
possible components' types.
DIGITAL HERITAGE - Volume 2 - eCAADe 34 | 531

From the abovementioned classes, we only focus on
the style class since we are designing a system that
can illustrate the building styles. Of the components
under the style class, we use basestone type, sty-
lobate type, column type, bracket type, roof struc-
ture type, and roof type because these elements are
the most deterministic according to the built period,
building's usage, and geographical location which is
the context class in our ontological structure. With
these styles, architects estimate the construction pe-
riod of the historical building or reconstruct historical
remains.
Unfortunately, historical cases of the historical
Korean architectures are not well organized in liter-
atures. We had to obtain the information regarding
historical architecture cases from various sources on-
532 | eCAADe 34 - DIGITAL HERITAGE - Volume 2
line and old books. Korean architectural heritages Figure 2
are mainly wooden and have a precast structure. Ontology for
This can be divided into four parts according to their architectural
placements and functions: base, frame, bracket, and heritage education.
roof. Each part is composed of several sub-elements.
Of 14 sub-elements, stylobate, base stone, column,
bracket, and roof structure, especially, have different
styles according to the built period, building's usage,
and geographical location. With these styles, archi-
tects estimate the construction period of the histor-
ical building or reconstruct historical remains. As il-
lustrated in Figure 2, these six components enables
users to reconstruct buildings in six steps starting
with the stylobate (gidan) at the base of the archi-
tecture, base stones (choseok), columns (gidoong),
bracket (gongpo), roof structure (gagu), and ending
with the roof (jiboong).
Tasks and rules to control the rules
At large, there will be seven main tasks where each
task narrows down the list of architectural case ac-
cording to the input or selected information (see Fig-
ure 3). For instance, in Task 1, the cases are narrowed
down according to the initial data input by the user:
era, period, and location. From the remaining cases,
the cases are even more narrowed down according
to the stylobate type selected by the user in Task 2.
This process repeats until Task 7 where a final list of ar-
chitecture cases is reconstructed. While the compo-
nents are narrowed down systematically, to the users,
all components are visually listed.
Each task will be composed of three rules. First
rule will assert specification to the working memory.
Second rule will compare the asserted specification
with the list of cases to find those that match which
are the possible cases. Lastly, if the selected type is
not in the list of cases, then the third rule fires where
the system prints out a reason why the particular type
does not work. The tasks will be controlled using the
status slot of the facts. The status will first begin as
(status initial). However, once the narrowing down
using the user input on era, usage, and location is
completed, the cases that remain will have a (status
Figure 3
Six basic
components for
building case
library and case
example.
considerforstylobate). Once the cases are narrowed
down from the stylobate selection, then the cases will
be asserted as (status considerforbasestone).
IMPLEMENTATION
Using the built cases, tasks, rules, and control meth-
ods, we designed the system GUI and implemented
into a system using JESS for inference engine and
JAVA swing for interface design. Our system, Logi-
Rec, is divided into seven tasks (see Figure 4). The
'User Inputs' panel on top left is the panel that re-
trieves the initial information for the system to run
such as the building name, the era, the usage, and the
location. When the 'Submit' button is pressed, the
Figure 4
Tasks and Rules.
system will search for the cases that meet the require-
ments. For example, historical cases with the same
era, usage, and location are considered. However,
on the interface is the complete list of all the types
under the architectural component. The panel lo-
cated under the phrase "RECONSTRUCT YOUR BUILD-
ING HERE" is the 'Instruction' panel where the system
and the user can communicate especially to explain
why the selected type is wrong. Below that is the
'Workspace' panel which displays all the choices for
the building components. The 'Check' button veri-
fies whether if the selection is appropriate. On the
bottom is the 'Information' panel which provides ex-
planation for the building component with an image.
DIGITAL HERITAGE - Volume 2 - eCAADe 34 | 533
 Lastly, below the User Inputs panel is the 'Progress'
panel which shows the progress of the building be-
ing built for user experience and understanding.
The system is designed to guide the users to
model an appropriate architecture in accordance to
the era, location, and the usage. It basically en-
ables users to reconstruct buildings intuitively in six
steps from bottom to top starting with the stylobate
(gidan) at the base of the architecture, base stones
(choseok), columns (gidoong), bracket (gongpo),
roof structure (gagu), and ending with the roof (ji-
boong). The users' selections and the knowledge in
the system (which is the list of all the architectural
cases) are compared to diagnose the selection by
finding the discrepancy between the actual user se-
534 | eCAADe 34 - DIGITAL HERITAGE - Volume 2
lection and the ideal selection and informs the users
on why the selection is inappropriate.
SYSTEM EVALUATION
Using the implemented system, we executed the
system evaluation through user test for system en-
hancement. The quantitative evaluation method was
derived based on the evalution model of MOSAICA
System proposed by Barak (2009) with few adjust-
ments made adding questions about understanding
of architectural heritage and awareness of traditional
culture. The survey questions consist of 13 state-
ments soliciting level of agreement about affective,
cognitive, operative aspect of the system (see Table
1). Four questions are related to the 'Affective' cate-
Figure 5
Developed System.
 gory measuring whether or not operating the system
is interesting, enjoyable, and fun; six questions are re-
lated to the 'Cognitive' category measuring whether
or not the system is educational and provides new
knowledge at its use; finally, three questions are re-
lated to the 'Operative' category measuring whether
or not operating the system is easy to learn and do
not involve technical problems (Barak, 2009). Each
statements were measured on a 5 point likert scale. In
addition to the quantitative approach, we conducted
interviews to gain specific feedbacks for the improve-
ment of our system.
Table 1
System evaluation
questionnaire.
Figure 6
Results of
questionnaire.
We conducted the user test on the 28th of May in
2016 from 5 p.m. to 6 p.m. on site at Deoksugung
Palace. The participants were randomly gathered on
site as potential system users. A total of 10 tourists
who have no professional knowledge about architec-
tural heritage were asked to reconstruct the building
on site (Deoksugung Palace) by interacting with our
system on a laptop. After completing the task, the
participants answered the questionnaire. When the
evaluation was finished, interviews were conducted
to get feedbacks for system improvement.
RESULTS
From the survey results, we made a bar graph to visu-
alize which question had a positive rating and which
had negative rating. As shown in Figure 5, ques-
tions with the most positive rating were the ques-
tion 4 under affective aspect and question 8 under
cognitive aspect, which are consecutively, 'exploring
this system improved my understanding of architec-
tural heritage' and 'exploring this system enhances
open-mindedness towards traditional culture'. These
results indicate that tourists had positive response
about learning architectural heritage at the heritage
site and that the system actually helped to improve
their knowledge about traditional culture. On the
other hand, questions with the most negative rating
was the question 9 under operative aspect which is
'this system provides sufficient information of archi-
tectural heritage'. This result indicates that the partic-
ipants had a difficult time understanding the concept
of dividing the architecture into parts and classifying
the architectural styles based on the era, period and
location. The detailed results of the surveys are pre-
sented in Table 2 under positive and negative asser-
tions along with the results of the improvement feed-
backs from user interviews.
From Table 2, it is easy to see that our system was
evaluated well in terms of cognitive aspect but was
evaluated poorly in terms of the operative aspect. As
DIGITAL HERITAGE - Volume 2 - eCAADe 34 | 535
 our aim of this system was to provide an educative ex-
perience and new knowledge to the potential users,
we believe that the aim was achieved at this point of
our research.
CONCLUSION
The present system shows possible component cases
with input data and users can combine possible com-
ponents and construct possible Korean traditional ar-
chitectures. We hope that this 'model-based diag-
nosis system' will help to overcome the problem of
lack of interaction between the cultural heritage sites
and the visitors. Especially, we hope that his sys-
536 | eCAADe 34 - DIGITAL HERITAGE - Volume 2
tem to educate the young adult tourists interested
in Korean cultural heritage. By taking them step by
step to reconstruct buildings intuitively in six steps,
the tourists can understand the structures of archi-
tectures more in detail. Our system allows the users
to explore Korean traditional architectures and guide
them to reconstruct buildings intuitively piling up
the building bottom-up like building with building
blocks. Whether if our system evokes curiosity and
the desire to learn to the users was tested during
the system evaluation; the result showed that the
system performed well in terms of the cognitive as-
pect which measures whether or not the system is
Table 2
Evaluation matrix
and examples of
users’ remarks.
educational and provides new knowledge at its use. Computer-Aided Architectural Design Futures, 1, pp.
However, we found out that our system performed 383-401
poorly in terms of the operative aspect which mea- Gonzalez, A.J. and Dankel, D.D 1993, The engineering of
knowledge-based systems: theory and practice, Pren-
sures whether or not operating the system is easy to
tice Hall
learn and do not involve technical problem. As for Juvancic, M and Zupancic, T 2008 'Towards Effective In-
the limitation of our work, we defined the architec- terfaces for General Architectural Learning', Proceed-
tural style at a high-level division of era, usage and ings of the eCAADe Conference, pp. 569-576
location while each could be subdivided further. In Liu, Jin and Wu, Zhong-Ke 2015, 'Rule-Based Generation
addition, we assumed that buildings we included in of Ancient Chinese Architecture from the Song Dy-
nasty', Journal on Computing and Cultural Heritage,
our case library had maintained their original form
9.2, p. 7
instead of the transformed form created while un-
dergoing repair. Therefore, we had to exclude trans-
formed heritages from heritage cases. In response to
the system evaluation, for future works, we intend to
make the system comfortable to use and interesting
by adding the concept of gamification incorporating
the reward system. And to enhance understanding
regarding Korean historical architecture, we plan to
add a section that provides fundamental information
on Korean architectures explaining what each com-
ponents are and that they can be considered sepa-
rately like building blocks. Lastly, to improve the tran-
sition of the screen, we intend to incorporate intuitive
gestures by using pad with a touch screen.

ACKNOWLEDGEMENTS
This research is supported by BK21 Plus Postgraduate
Organization for Content Science.

REFERENCES
Apollonio, F. I, M, Gaiani and Z, Sun 2013 '3D model-
ing and data enrichment in digital reconstruction of
architectural heritage', International Society for Pho-
togrammetry and Remote Sensing
Barak, M, Herscoviz, O, Kaberman, Z and Dori, Y. J 2009,
'MOSAICA: A web-2.0 based system for the preserva-
tion and presentation of cultural heritage', Comput-
ers & Education, 53(3), pp. 841-852
Brusoni, V 1998, 'A spectrum of definitions for temporal
model-based diagnosis', Artificial Intelligence, 102, 1,
pp. 39-79
Champion, Erik 2009, Handbook of research on effective
electronic gaming in education, IGI Global
Cursi, S, Simeone, D and Toldo, I 2015, 'A Semantic
Web Approach for Built Heritage Representation',

DIGITAL HERITAGE - Volume 2 - eCAADe 34 | 537

VIRTUAL REALITY
Development of a Virtual Reality Solution for End User
Involvement in Interior Design
Kjeld Svidt1 , Jesper Bendix Sørensen2
1
Aalborg University 2 Evokon
1
ks@civil.aau.dk 2 jbs@evokon.dk

This paper describes development and test of a prototype Virtual Reality system
aimed at user involvement in hospital design. User needs and functional
requirements are captured by interviews and observations in three case studies of
ongoing projects in Denmark. Based on the identified requirements, a prototype is
developed based on a multitouch display for manipulating room layout in a floor
plan view and a set of Oculus Rift glasses for experiencing the design in Virtual
Reality. Together with users from the studied cases, test scenarios were performed
to identify possible benefits, challenges and further requirements to the system.

Keywords: User involvement, hospital buildings, interior design, virtual reality

INTRODUCTION et al. 2010). However, many resources are spent on

Within a period of 10 to 15 years, the Danish health
sector is investing approximately 6 billion Euros in a
number of new hospital buildings. During the de-
sign of these complex buildings, the design teams to
a large extent use physical full-scale mock-ups to sup-
port the dialogue with end users of different profes-
sions. This approach enables users to provide valu-
able input for the design of future working environ-
ment, logistical issues, discuss various solutions and
testing them through role playing. Involving users in
this way is important for the final solutions, but the
preparation of physical mock-ups can be costly and
it will be difficult to reach very many design alterna-
tives with physical mock-ups, both due to cost of the
mock-ups and the time it takes to dismantle and set
up new mock-ups. Additionally, it will be difficult to
physically create major sections, of the future build-
ing to study navigation and logistic aspects, since the
physical facilities that are available for user involve-
ment imposes restrictions on the size of these (Leicht
building and rebuilding physical mock-ups as well as
transporting users to the mock-ups, performing tests
of future working routines, evaluating, reporting at
redesigning the architects' design proposals.
The use of virtual mock-ups has the potential
to support a number of the issues identified with
physical mock-ups. Previously, technologies like Vir-
tual Reality have been associated with high costs for
both hardware and software, but recent year's de-
velopment has lowered these costs drastically. It is
now possible to buy well-functioning virtual reality
glasses at very affordable prices. This makes it inter-
esting to look for ways to support and expand the
existing working methods and processes with new
technological initiatives. For example, the use of vir-
tual mock-ups could provide new opportunities to
experience and try out more test scenarios, and also
to simulate and analyze situations than with physi-
cal mock-ups. (Johansson, 2012; Leicht et al. 2010;
Peavey et al. 2012).

VIRTUAL REALITY - Volume 2 - eCAADe 34 | 541

 The idea behind the current project is to investigate
the possibilities to partly replace physical mock-ups
with Virtual Reality mock-ups based on the design-
ers' 3D models. Based on a close collaboration with
a number of ongoing healthcare projects, this pa-
per examines how Virtual Reality can support user in-
volvement through the creation of virtual mock-ups,
using building models from the consultants existing
BIM-environment.
METHOD
Based on system development methods from Con-
textual Design (Beyer et al. 1998) a prototype of a Vir-
tual Reality system to facilitate the creation and use of
virtual mock-ups is developed and tested. The work
is based on three case studies, where observations of
current work practices, as well as interviews result in
a needs assessment in relation to the work associated
with mock-ups in 3 temporally dispersed stages. This
gave an understanding of the existing working pro-
cesses and needs of the involved actors prior to de-
veloping and testing the prototype system. In addi-
tion it gave valuable input to identify future user sce-
narios of actual work associated with mock-ups and
the use of mock-ups after trials are completed.
CASE STUDIES
Three ongoing projects in the healthcare sector were
chosen as case studies. The selected cases were cho-
sen to cover different aspects of user involvement
and different ways to support user involvement in
different stages of the design process.
Case 1 was Department for Trauma and Torture
Survivors located in Vejle, Denmark. This case fo-
cused on user involvement in the early design stage
before any kind of mock-ups were built. During brain
storm sessions, the users should think out and de-
scribe future working routines which should lead to
descriptions of requirements for the facilities includ-
ing e.g. requirements to room affinities. 15 employ-
ees in 3 groups should describe the expected 'patient
travel' through the new facilities mainly by means of
coloured post-its. No floor plans or other graphic il-
lustrations of the planned facilities were used at this
542 | eCAADe 34 - VIRTUAL REALITY - Volume 2
stage, so ideas regarding geometric aspects of the
new facility were only exchanged verbally.
Case 2 was part of a new hospital in Gødstrup,
Denmark. This case focused on user involvement in
the planning of layout of new medicine rooms. Phys-
ical mock-ups of design proposals were built in test
facilities near the construction site and users should
test and find the best solutions for future working
routines and collaboration in the full scale mock-ups.
Room boundaries were created by partition screens
as shown in figure 1 and equipment was represented
by more or less abstract artefacts made of different
materials. Figure 2 shows examples of artefacts e.g.
polystyrene sheets representing a Laminar Airflow
Bench and a group of users in the full scale mock-
up. The users furnished the room with the artefacts
to study possible future working routines. Facilitators
supported the discussions and brought additional
artefacts as requested by the users during the ses-
sion. During these exercises the users found it nec-
essary to start sketching floor plans of the room indi-
cating position of equipment to complement the full
scale mock-up and to keep track of discussed design
alternatives as well as comments explaining design
decisions.
Figure 1
Room boundaries
of full scale
mock-up were
created by partition
screens.
Case 3 was a new university hospital in Aarhus,
Denmark. In this case a role play preparing users to
work in the new facilities was studied. Scale mod-
els of the planned facilities were represented by large
prints of floor plans combined with Lego scale mod-
Figure 2
Users discussing
furnishing and
examples of
artefacts
representing
different types of
equipment.

els of equipment, patients and personnel. Figure 3 ronment.

shows an example of the set up. A facilitator con-
trolled the role play with participants from different
disciplines such as nurses, doctors, cleaning person-
nel and others. For different types of patients, the ac-
tors should play through scenarios from planning the
arrival of the patient, receiving the patient and fol-
lowing the route of patient and personnel through
different rooms during the treatment. Since only a
minor part of relevant equipment was represented
on the floor plans or as artefacts, many questions
and uncertainties arose during the role play regard-
ing which equipment would be available in specific
situations.
IDENTIFICATION OF FUNCTIONAL RE-
QUIREMENTS
The researchers' observations of existing processes
in the case studies and interviews with professional
process facilitators as well as end users from different
staff categories led to the identification of a number
of relevant functional requirements for a Virtual Real-
ity system which would have the potential to partly
replace or complement some of the work with phys-
ical mock-ups. The identified requirements are sum-
marized below.
The working processes often include closely re-
lated activities taking place in a number of rooms
with different functionality. Thus it should be possi-
ble to study aspects of affinity between rooms with
different function. The end users should have the
possibility to identify and comment on rooms' func-
tional requirements while they are in the virtual envi-
Ideas for alternative room layout often emerged
during the test sessions in the physical mock-ups.
In the virtual environment, it should be possible
to easily model room layout including positions of
walls and interior to quickly view and discuss con-
sequences of a design idea. In addition to this, it
would be important to be able to save comments, de-
scriptions of design ideas and decisions made in the
virtual environment to communicate this knowledge
back to designers and capture design rationale.
From a technical point of view, it would be an ad-
vantage to have a good connection to existing tools
used in the specification of requirements to rooms
and their equipment including CAD/BIM tools that
would be used by the designers. Similarly, it would
be relevant to have connectivity to existing docu-
ment management systems used by the design team
on the specific project.
In the case studies, it was found that missing
equipment or too simple or too abstract representa-
tions of equipment or building parts as e.g. windows
often led to confusion and misunderstandings dur-
ing the test sessions. Here, the virtual environment
has the potential to add value by better visual rep-
resentations of equipment and building parts. Thus,
the objects in the virtual environment should appear
visually realistic in size and shape. Figure 4 shows an
example of a toileting chair in sufficient level of detail
to the right and too simple to the left. The graphic
quality will especially help end users with little or no
experience in reading technical drawings who asked
for a realistic impression of the spatial aspects of the
room interior.

VIRTUAL REALITY - Volume 2 - eCAADe 34 | 543

 Figure 3
Users in role play
around tables with
floor plans and
scale models of
personnel and
equipment.

In all the case studies, the real time dialogue be- Figure 4
tween users proved to be valuable for the decision Toileting chair
processes. To have a similar functionality in the vir- shown in different
tual environment, several users should be able to in- level of detail.
teract with the model in real time, e.g. to enable 'role
playing' in the investigation of future working rou-
tines in the proposed building design.

PROTOTYPE DEVELOPMENT
Based on a review of available technologies, e.g. (Mi-
helj et al, 2014), (Johansson, 2012) and the detailed
studies of existing workflows and needs in the de-
sign process, a first prototype of a system to support
user involvement in the interior design was devel-
oped including a few of the identified needs identi- Figure 5
fied above. The prototype was based on the Unity First prototype
Game Engine, an Oculus Rift Head Mounted Display ready for test in lab.
(HMD) and a touch screen, where end users could in-
tuitively move equipment and room boundaries in
floor plans based on the architect's design proposal.
Figure 5 shows the prototype ready for use in the lab
and figure 6 shows a screenshot of the initial user in-
trerface.
During the development process, factors influ-
encing the user experience in the virtual environ-
ment where examined through a literature review,
where depth perception, complexity of the virtual
environment, presentation quality, motion tracking
and cyber sickness were some of the topics discussed
(LaViola 2000; Bowman et al. 2007). Two detailed
tests were carried out, where models from the build-
ing designers BIM-environment were transferred to
the virtual environment and tested with users to
identify potential obstacles.

544 | eCAADe 34 - VIRTUAL REALITY - Volume 2

Figure 6 strated that usability of the equipment plays a signifi-
Screenshot of touch cant role, since some of the users found it challenging
screen user to move objects on the touch screen and a number of
interface. users experienced varying degrees of cyber sickness.
Overall, the evaluations confirmed that the combina-
tion of head mounted displays and a touch screen
for manipulating the design can contribute to the
user involvement process in interior design of hospi-
tal buildings.

Figure 7
The clarification of needs and review of technical and
Users preparing test
human factors that affect the use of Virtual Reality
scenario.
formed the basis for a system design, where existing
workflows were redesigned into a vision for the fu-
ture work practice. A prototype allowing end users
and design facilitators to design and configure the
workspace, and thereafter test it in Virtual Reality was
then developed.

EVALUATION AND CONCLUSIONS

In order to validate the usefulness of the vision, a
number of evaluations of the developed prototype
were set up. Through "quick and dirty" feedback
sessions with users and experts of user involvement,
along with a series of field studies, the user experi-
ences of the interface and Virtual Reality were exam-
ined. Figure 7 shows users preparing a test scenario
and figure 8 shows a user navigating in the test envi-
ronment.
The evaluations showed that even high-end
computer systems can easily become overloaded re-
sulting in reduced performance of the VR system.
When a user involvement system should be used
Figure 8 for the construction of large hospital buildings, it is
User navigating in therefore essential to prevent such failures in perfor-
the virtual mance and to design the system so as to maximize
environment while the usefulness of the available hardware. The eval-
colleagues are uations of the developed prototype with end users
following her view confirmed that the virtual mock-ups have a potential
on the touch in user involvement. Users agreed that the system's
screen. basic features in several areas supported their needs,
especially on issues regarding spatial understand-
ing. However, the evaluation exercises also demon-

VIRTUAL REALITY - Volume 2 - eCAADe 34 | 545

 REFERENCES
Beyer, H and Holtzblatt, K 1998, Contextual Design: Defin-
ing Customer-Centered Systems, Academic Press
Bowman, DA and Mcmahan, RP 2007 'Virtual Reality :
How Much Immersion Is Enough?', IEEE Computer So-
ciety
Johansson, J 2012 'Patient Rooms of a California Based
Hospital: Benefits of Physical Mock-ups vs. Benefits
of Virtual Mock-ups', University of Minnesota, College
of Design, Working paper.
LaViola, JJ 2000, 'A Discussion of Cybersickness in Virtual
Environments', SIGCHI Bulletin, 32(1), pp. 47-56
Leicht, RM, Kumar, S, Abdelkarim, M and Messner, JI 2010
'Gaining End User Involvement through Virtual Re-
ality Mock-Ups: A Medical Facility Case Study', Pro-
ceedings of CIB w78 conference
Mihelj, M, Novak, D and Beguš, S 2014, Virtual Reality
Technology and Applications, Springer
Peavey, EK, Zoss, J and Watkins, N 2012, 'Simulation and
mock-up research methods to enhance design deci-
sion making', Herd, 5(3), pp. 133-144

546 | eCAADe 34 - VIRTUAL REALITY - Volume 2

High quality Virtual Reality for Architectural Exhibitions
Anette Kreutzberg1
1
The Royal Danish Academy of Fine Arts, School of Architecture
1
anette.kreutzberg@kadk.dk

This paper will summarise the findings from creating and implementing a visually
high quality Virtual Reality (VR) experiment as part of an international
architecture exhibition. It was the aim to represent the architectural spatial
qualities as well as the atmosphere created from combining natural and artificial
lighting in a prominent not yet built project. The outcome is twofold: Findings
concerning the integration of VR in an exhibition space and findings concerning
the experience of the virtual space itself. In the exhibition, an important aspect
was the unmanned exhibition space, requiring the VR experience to be
self-explanatory. Observations of different visitor reactions to the unmanned VR
experience compared with visitor reactions at guided tours with personal
instructions are evaluated. Data on perception of realism, spatial quality and
light in the VR model were collected with qualitative and quantitative methods at
two different occasions and setups after the exhibition, both showing a high
degree of immersion and experience of reality.

Keywords: Virtual Reality, Oculus Rift, GearVR, Exhibition display

INTRODUCTION AND BACKGROUND
As a part of the exhibition series World Architecture at
the Danish Architecture Centre [10] (DAC) in Copen-
hagen, the exhibition 'World Architecture - Snøhetta'
[13] opened to the public in June 2015. The exhibi-
tion, created by Snøhetta [12] and Danish Architec-
ture Centre, ran until September 27th 2015 and had
13.000 visitors. Experimental use of virtual reality was
implemented in the exhibition to demonstrate state
of the art communicative methods and tools as well
as expand the range of media used to present archi-
tecture in exhibitions.
The experimental VR model let the audience ex-
perience the vast space of the Central Hall in King Ab-
dulaziz Centre for World Culture [11] in a 1:1 scale as
if they were actually present in the building complex
situated in Saudi Arabia, well before the building was
inaugurated.
Architectural representation and experi-
encing architectural space
It is not enough to see architecture; you must expe-
rience it (Rasmussen 1959). There is a discrepancy in
the way we sense architecture and its representation.
For an architect it is important to be able to precisely
express oneself and make sure that visions and inten-
tions are fully understood. Architectural represen-
tations are primarily drawings or images, previously
handmade and nowadays mostly computer gener-
ated from 3D models, but it is still 2D representations
presented either on paper or on screen. Although of-
ten supplemented by scale models to touch, the 2D

VIRTUAL REALITY - Volume 2 - eCAADe 34 | 547

 representation remain the primary and purely visu-
ally focused method of representing architecture.
Adding interaction or motion to screen images
can help conveying space (Kreutzberg 2011) but
there is still a difference in envisioning the real space
within a frame of an image or animation and experi-
encing the space with your body in the real world. Ex-
periencing architectural space is a multi-sensory ex-
perience involving more than mere vision (Pallasmaa
2005).
VR has the potential of bridging the gap between
2D visualization of space and the bodily experience
of space in the real world by placing the viewer in the
virtual world, making the viewer part of the virtual
world and therefore experiencing the virtual world
with bodily senses.
Virtual space
Previous studies have found VR displayed on Ocu-
lus Rift DK2 [9] Head Mounted Display (HMD) to
be a promising media platform for visualizing and
demonstrating complex spatial 3D models, espe-
cially for non-experts untrained in reading technical
drawings (Kreutzberg, 2015).
VR is still an emerging medium, where hardware
and software are improved and refined constantly,
many of which are still only available in developer
versions. Strict limitations of the models concerning
poly count, texture size, lights and screen effects ap-
ply, to run a stable and comfortable VR application.
Some issues are acknowledged and important to
address especially concerning simulator sickness in
VR (Moss, 2011). It is of major importance to prevent
feelings of disorientation and nausea, and to protect
users from motor-visual functioning issues after use,
especially at a public exhibition. Creating a stable ex-
ecution with sufficient frame rate and limited draw
calls is the challenge.
The experience of realism in architectural VR
models relates to the use of real world model size,
detailed surface properties, shadows and reflected
light. They are all important for experiencing and un-
derstanding the architectural space but especially for
548 | eCAADe 34 - VIRTUAL REALITY - Volume 2
sensing the intended atmosphere, light is important.
Real-time rendering of reflected light and soft shad-
ows are very computational demanding and require
extensive optimization of the VR model for use with
contemporary available HMD's.
METHODS
Phenomenological studies were made several times
during the exhibition period, observing the be-
haviour of the visitors in the area with the VR HMD
at a distance. Differences in behaviour of groups, sin-
gles and groups on guided tours were observed.
Survey at cultural event
After the exhibition period a cross-sectional anony-
mous survey (Creswell 2014) was conducted at Kul-
turnatten [2] collecting data [6] from visitors (N=72)
at a cultural event displaying the KACWC VR model.
Questions relevant for this paper were focused on the
experience of realism, image quality, degree of vi-
sion and comfort as well as including a comparison
of viewing the VR model in Oculus DK2 and as stereo-
scopic 360 °rendered image in GearVR [8].
Qualitative empirical study
To get a more thorough understanding of the ex-
perience in the KACWC virtual space, data was col-
lected on sensory, bodily and atmospheric experi-
ences through video-recorded Walk-and-talk inter-
views with nine participants, inspired by Sensory
Ethnography methods (Pink 2009).
The interviews were arranged after the exhibi-
tion and the cultural event, and were conducted in a
meeting room at an architect's office. The interviewer
followed along on a computer screen and made con-
versation about the sensuous experience while it was
happening. The interviews lasted between 8:20 min-
utes and 13:35 minutes each. (Hermann 2016.)
After the interviews, the participants sum-
marised and discussed their individual experiences
in two focus group discussions.
 Exhibition design - practical setup
The exhibition Snøhetta: World Architecture was or-
ganised in three rooms each representing the exhibi-
tion subtitle: People Process Projects. The VR exper-
iment was placed in the first room representing the
people and their work environment. An Oculus Rift
DK2 HMD was mounted on an acrylic box discretely
placed on the wall, representing the very long office
working table. The connected laptop was hidden be-
hind the wall and not visible to the audience. A small
instruction label was placed in the box. See figure 1.
Figure 1
The HMD.
Figure 2
Central Hall in King
Abdulaziz Centre
for World Culture,
Screen View.
The HMD was modified to be used hand held without
having straps tied to the head for two reasons. Firstly,
a hand held experience restricts the movement of
the upper body avoiding vigorously turning and sec-
ondly, it places the hands "out of sight".
Some visitors had previous experience with the
Oculus DK2 or had seen it demonstrated online, and
they unfortunately wanted to use the heads traps,
and tried to tear the straps down that were secured
on the side of the HMD.
Proper visual instruction as to how to hold the
HMD with both hands instead should have been
more visible. The comfort of not having the HMD
strapped to the face was obvious though, and peo-
ple wearing glasses could actually use the HMD with
their own glasses.
VR model
The VR model was created by converting a Rhino [3]
solid model to several mesh models in 3dsMax [4]
and importing these mesh models into Unity [5] for
texturing, lighting and programming of interaction
and finally output the complete model to an exe-
cutable file.
Converting a huge conventional 3D Rhino archi-
tectural solid model representing 100.000m² to a low
poly mesh model suitable for VR is not an easy task.
All unwanted details were deleted prior to the mesh
conversion and still the final optimised mesh model
had more than 2 million triangular polygons. It was
not possible to optimise all the geometry automati-
cally without at the same time changing the appear-
ance of the original architecture. Several parts of the
model were therefor completely remodelled to opti-
mise and at the same time to lay out the topology for
clean UV mapping of texture maps and light maps.
All mesh optimizing and UV mapping were done in
3dsMax. See figure 2.
VIRTUAL REALITY - Volume 2 - eCAADe 34 | 549
 Lighting was created in Unity with Precomputed
Real-time Global Illumination with Forward Render-
ing in combination with Baked Global Illumination. It
allows for the use of hardware techniques like 'multi-
sample anti-aliasing' (MSAA) which can have a great
impact on image quality.
The different light sources were adjusted in in-
tensity, direction and colour in an attempt to best
represent the combination of daylight and artificial
light in the architectural model.The Unity Profiler was
used throughout the process as a tool to analyse and
ensure stability in performance of the final built.
RESULTS AND REFLECTION
Using VR in an exhibition space without personal in-
struction turned out to be a larger challenge than
first anticipated. First of all the VR headset was not
spotted by quite a lot of visitors - it blended in with
the wall displays too well! Secondly if spotted - quite
some bewilderment arose as to what was displayed
and what to do. Most visitors found out to look
around but not all tried to turn their body 180 de-
grees.
Initially it was intended to allow for movement
and exploration in the VR model, but concerns about
simulator sickness and the practical setup in the un-
manned exhibition turned the experience into an ex-
ploration from a fixed viewpoint.
Halfway through the 3-month exhibition period,
an update was made to the model creating a loop
with fade-to-black between two different positions
instead of displaying just one permanent viewpoint.
The loop was combined with plan views indicating
the active point of view. See figures 3, 4 and 5.
Looping between two positions did not work as
intuitive as expected - fade to black was set to 4 sec-
onds for a comfortable feel - but it was often inter-
preted as a fault - that the installation broke when the
screen turned black - and visitors took of the headset
before they could realize it was a transition into the
next position.
Many visitors completely lost track of the orien-
tation in real space in the exhibition, and were quite
disoriented when taking off the HMD again.
550 | eCAADe 34 - VIRTUAL REALITY - Volume 2
Figure 3
KACWC plan view A.
Generally, women seemed more focused on explor-
ing the architecture, sensing the light and general at-
mosphere accepting the immersion of being some-
where else while looking through the VR glasses,
whereas many men moved their heads and bodies
vigorously to test the latency and precision of the
technology behind the player.
There was a difference in the way people acted
if they were in a group, with a lot of audio to align to
the real world, and the single visitor without any com-
forting companions. The latter tended to be gentler
in the movements and were examining the headset
afterwards - and reading the instructions on the wall
carefully before they tried the experience.
Most people tried to take a step or two while be-
ing immersed in the VR environment - and it took
some tries to realize that locomotion was not possi-
ble.
On guided tours, the visitors were introduced to
the VR setup, and they could understand the loop-
feature and were generally very impressed. Explana-
tions as to what the model actually showed was also
explained by the guide.
It was a popular task in groups to film or photo-
graph each other wearing the HMD.
Cultural event
At the cultural event posters with renderings, draw-
ings and text gave visitors an opportunity to learn
more about the architectural project displayed in VR.
Visitors were guided in their VR experience and spent
approximately 5 minutes in total testing the differ-
ent HMD's. The survey results (N=72) were very pos-
Figure 4
KACWC position A.

Figure 5
KACWC position B.

itive with 31% of survey respondents experiencing
the realism to be "Fantastic" and another 43% respon-
dents experiencing the realism to be "Good". Degree
of vision and comfort were experienced as "Fantastic"
by 53% and 44% of survey respondents respectively.
See figure 6.
Visitors used Oculus DK2 to view the looping VR
model from the exhibition and had the opportunity
to compare it with a pre-rendered stereoscopic 360°
image of the same VR model (in this example ren-
dered from within Unity 3d game engine in a rather
blurry resolution!) displayed in GearVR on a Samsung
Galaxy S6 [7] mobile. The comparison was interesting
because the high quality VR model used static points
of views, and did not utilised the possibility of move-
ment. The same static point of view in a pre-rendered
stereoscopic 360° rendering can be rendered directly
from the render software without the need of time
consuming optimising for real-time rendering in a
game engine.

Figure 6
Experience of the
room with Oculus
Rift DK2.

VIRTUAL REALITY - Volume 2 - eCAADe 34 | 551


Difference in experience of realism and image quality
was noticed though including the smaller view frus-
tum in Gear VR (100°) compared to Oculus DK2 (120°).
See figure 7.
With the use of high quality renders, the GearVR
does have great potential - it is cheap, portable, and
you can store many 360-renderings on the mobile.
Walk-and-talk
The walk-and-talk interviews were conducted during
two days in November 2015.
Sixteen questions in three categories were in-
cluded in the interview guide, asking for comments
ranging from purely descriptive to emotional and in-
tuitive reflections.
Participants were architects (4) and users of ar-
chitecture (5) and the two groups had very different
approaches to VR. The architects reflected mostly on
the representation of architectural space whereas the
architect-users reflected on the virtual space and not
the representation of space.
Mass and volume. All participants described the dif-
ferent components of the space, especially the con-
trast between organic volumes emerging from regu-
lar straight ceilings and walls. Most participants also
commented on the large number of columns scat-
tered in the hall.
Materials. The participants perceived most of the
materials correct. They identified wood, metal, pol-
552 | eCAADe 34 - VIRTUAL REALITY - Volume 2
Figure 7
Experience of the
room with GearVR.
ished stone, glass and concrete, only the rammed
earth (walls) was generally confused with other ma-
terials like cork, wood or brick. One exception was a
participant confusing the polished stone floor with a
carpet.
Light. Many participants described in detail different
light settings, contrasts and emphasized the natural
or artificial appearance.
One architect commented on the different light
setups in the two positions in the VR model:
(Pos. B) "I can see they have been working with
artificial lighting in the ceiling. I am not particularly
mad about it, if I may say so. I think that the skylight
at the end of the room is impressive though, but it
feels like some daylight is missing, compared to the
other place I was before.
(Pos. A) Here there are the same ceiling lights
as in the other place, but here it is much lighter in
a pleasant way. It is because I can see that the light
comes from above. There is a place by the stairs
where some light is, it could be daylight."
In the final focus group discussion her conclusion
was:"I think what worked best in VR was the repre-
sentation of the light. [...] You got a great feel for
the character of light in VR, and you could experience
the various light settings, daylight, reflected light and
various types of artificial lighting. It is difficult to rep-
resent light in an image. Here I really think it [VR] has
potential." (AN, Interview C)
Figure 8 Haptic impulses. All participants expressed an urge
Videorecording of to explore the virtual space, especially "to look past
haptic impulses the edge and down". The visual experience made
causing unintended them forget the fact, that it was a stationary repre-
moving in real sentation, and they tried to walk towards the points
space. of interest. See figure 8.
Implicit sensory perception. "I have a feeling that if
I touched the pillar in front of me, I would feel stone.
Cold against the hand. I clearly recall the surface from
places I have been before." (SM, Interview B3)
The bodily dimension. "I'm surprised at how wild
this is. I can look up, and I can turn around. Which
is a full 360 degrees! I can look down." (JD, Interview
B5)
Eye height. The eye height in the VR model was fixed
and set to 165cm. Individual calibration of eye height
is possible with Oculus DK2, but since it was the ex-
hibition VR model that was being tested, the fixed
eye height was preserved. All participants were in the
height range of 165cm.-180cm.
"The simple and tangible, such as standing up,
means a lot. That you experience the space standing,
instead of sitting at a conference table and looking
down at a drawing or looking at a screen. [...] You
achieve a completely different physicality by stand-
ing up and turning your body. It gives a physical feel-
ing, which you cannot obtain in other ways. Other
than by being in a room in the real world." (JJ, Inter-
view C)
Atmosphere. For fully calling forth an atmosphere in
VR, light and materiality seem not to be sufficient; the
model must contain furniture, people and other arti-
facts to have a real feel of life to it.

CONCLUSION
Reconstructing and optimising the very large Rhino
3D model for use in VR was very time-consuming,
but resulted in a stable execution including real-
time rendered light. The model was on display in
the Snøhetta: World Architecture exhibition for 3
months.
The integration of VR in the exhibition was too

VIRTUAL REALITY - Volume 2 - eCAADe 34 | 553

 discreet in the unmanned area; many visitors did
not discover the HMD at all. For those that did,
they seemed to have an exiting experience, although
many did not understand what was on display - nei-
ther the content nor the technology.
The automatic loop between two fixed views in-
troduced halfway through the exhibition period was
not as intuitive as expected. Sufficient instructions
on what to expect and why, especially concerning
duration and (limited) interaction, proved to be very
important for a positive VR experience.
The perception of spatial and lighting qualities
in the VR model worked very well and was examined
thoroughly at walk-and-talk interviews after the exhi-
bition period. The comment from a user: "It is almost
too real," explains the urge to walk around and ex-
plore the VR environment although interaction was
restricted to fixed viewpoints. Since virtual reality is a
medium that attempts to replicate one's experience
in the physical world, users are likely to have an ex-
pectation that they will be able to interact with that
virtual world in the same ways they do outside of it.
User responds indicate that for perception of at-
mosphere in VR, light and materiality are not suffi-
cient alone; the model must also contain a certain
amount of furniture, people and other artifacts.
Further studies should extend the experiments
with representation of natural and artificial light in
VR and should include experiments with 360° stereo-
scopic renderings for faster iterative work.
ACKNOWLEDGEMENT
The Danish Arts Foundation [1] funded the develop-
ment of the VR model.
REFERENCES
Creswell, JW 2014, Research design: Qualitative, quanti-
tative, and mixed methods approaches, Sage Publica-
tions
Cutting, J 1997, 'How the eye measures reality and vir-
tual reality', Behavior Research Methods, Instruments,
& Computers, 29(1), pp. 27-36
Hermann, D 2016, Kroppen og sanserne i Virtual reality,
Master's Thesis, University of Copenhagen
554 | eCAADe 34 - VIRTUAL REALITY - Volume 2
Kreutzberg, A 2011 'Game Engines as Dynamic Tools
in the Design Phase', First International Congress of
Retevitruvio 2011, Bari, Italy, pp. 1585-1593
Kreutzberg, A 2015 'Conveying Architectural Form and
Space with Virtual Reality', Real Time - Proceedings
of the 33rd eCAADe Conference 2015, Vienna, Austria,
pp. 117-124
Moss, JD and Muth, ER 2011, 'Characteristics of Head-
Mounted Displays and Their Effects on Simulator
Sickness', Human Factors: The Journal of the Human
Factors and Ergonomics Society, 53(3), pp. 308-319
Pallasmaa, J 2005, 2008 edn., The Eyes of the Skin, Wiley-
Academy
Pink, S 2009, Doing sensory ethnography, Sage Publica-
tions
Rasmussen, SE 1959, 1980 edn., Experiencing Architec-
ture, The MIT Press, Cambridge
[1] http://www.kunst.dk/english/
[2] https://kadk.dk/kalender/kulturnatten-2015
[3] https://www.rhino3d.com/
[4] http://www.autodesk.dk/products/3ds-max/overvie
w
[5] http://unity3d.com/
[6] http://www.surveyxact.dk/
[7] http://www.samsung.com/dk/galaxys6edge-and-
gala
xys6/
[8] https://www.oculus.com/en-us/gear-vr/
[9] https://www.oculus.com/en-us/rift/
[10] http://www.dac.dk/da/dac-life/udstillinger/2015
/snoehetta---world-architecture/
[11] http://en.kingabdulazizcenter.com/
[12] http://snohetta.com/
[13] https://kadk.dk/en/news/ind-i-snohetta-forsknin
g-og-vr-briller-abner-op-arkitektur-det-er-by
gget
Design Tools and Complexity: Mobile Games and
Collective Imagination
Alexander Holland1 , Stanislav Roudavski2
1,2
University of Melbourne
2
https://unimelb.academia.edu/StanislavRoudavski
1
alexholland@me.com 2 stanislav.roudavski@cantab.net

This paper is based on a hypothesis that games can be used to support design
decisions in a variety of complex situations. To explore this proposition, the
research described below focuses on two aspects. Firstly, it experiments with the
potential of games to be socially provocative. And secondly, it applies the induced
provocations in support of collective imagination. This discussion is supported by
a practical case study: a working prototype of a smartphone game that simulates
urban cycling. The paper discusses utilisation of this game by diverse
stakeholders in a workshop that sought to advance decision-making in a
particularly vexatious stalemate.

Keywords: infrastructuring, virtual riding environments, collective imagination,

provocating games, design negotiation, mobile games, urban cycling

PROLOGUE: POCKETPEDAL Kelly is given a phone. "GET TO THE CITY," the

Kelly cannot remember the last time she rode her
bike. Maybe a couple of summers of ago, around the
bay. Nor does she play videogames - the closest she
gets is some Sudoku on her phone.
Most days Kelly drives. Even though she com-
mutes past handsome trees and grand buildings, it
is a stressful journey. Traffic is packed, and cyclists
swerve in front of her without warning. Kelly heard
they want to give more space to cyclists - as if finding
a park is not already a struggle.
A friend mentions a related workshop, open to
the public. Kelly is sceptical. She has filled feedback
forms before. When she arrives a group of people
is sitting down, eyes on the screens. "My best score
ever!" someone laughs. Looking at their screens, she
sees a world of blue cars and pink trees. "This is near
my work," she says.
screen instructs. A cyclist appears, riding in the
brightly coloured world - her city! Kelly taps the
screen and the rider moves forward. She taps again
and the avatar pedals faster. At the top left of the
screen is a score, increasing steadily. Kelly thinks she
can handle this. Some people lean over and start
cheering her on.
A double-parked taxi slides into view, blocking
her ride. Urgently, Kelly tap-tap-taps to avoid. A ma-
genta car closes in from behind. "GET BACK TO THE
BIKE LANE," her phone flashes. Kelly tries to return
into the lane, but the car clips her.
"GAME OVER!" the phone states. "YOU GOT A
SCORE OF 75, AND A BROKEN TOE!"
"Seventy-five?" someone notices. "Much better
than my first time!"

VIRTUAL REALITY - Volume 2 - eCAADe 34 | 555


INTRODUCTION: NEGOTIATING COM-
PLEXITY
The design value of experiences, such as those de-
scribed above, can be significant for understanding
constraints and opportunities. This paper argues
that such experiences become even more power-
ful when they are shared with others. For exam-
ple, the prologue's protagonist, Kelly, has an oppor-
tunity to experience a familiar environment via the
circumstances typical to other stakeholders [Figure
1]. Through this lens, she acquires a deeper un-
derstanding of others' motivations and behaviours.
Such experiences might prompt Kelly to rethink her
preconceptions, encouraging her to look at riders in a
new way. In provoking these deeper understandings,
PocketPedal asks: are cyclists irresponsibly erratic? Is
their behaviour avoidable? What can be done to sup-
port more equitable road conditions?
Increasingly, contemporary practice acknowl-
edges that design problems are necessarily complex,
dynamic and extended. This character of design chal-
lenges is particularly evident in urban settings. They
are not only objectively complex, but are also highly
contingent, dependent on stakeholders' value sys-
556 | eCAADe 34 - VIRTUAL REALITY - Volume 2
Figure 1
PocketPedal, a
mobile game of
urban negotiation.
The complex
interactions
between riders,
urban infrastructure
and traffic are recast
in game format.
The characteristics
of urban cycling are
condensed with the
aim to provoke and
challenge.
tems and political persuasions. In situations where
stakeholders disagree, progress to a particular so-
lution may be difficult, or even impossible, unless
the participants or potential beneficiaries can be per-
suaded to modify their attitudes, expectations and
future actions. Understandably, this can be difficult;
not only because of stakeholder reluctance, but also
due to the difficulties people encounter in appreciat-
ing experiences of others and imagining feasible al-
ternatives.
Urban cycling is utilised here as an illustrative
case-study with which to meet this significant chal-
lenge. Recent research demonstrates that designers
and planners need to broaden their knowledge of
specific modes of transport to recognise the diverse
and often conflicting needs of urban travellers. Par-
ticularly relevant to urban cycling, it has been sug-
gested that this recognition must be done before ad-
equate design solutions can be realised that achieve
greater participation in cycling (Bell & Ferretti 2015).
Compatible with this need, design can be un-
derstood as a process of imagining together. This
collective imagination is a public process even when
insular, authoritarian design methods are engaged.
In more inclusive scenarios, designing overlaps with
making of public cultures. Çınar and Bender (2007)
discuss collective imagination as a totality of daily
practices of urban dwellers. This paper seeks to em-
phasise how the particular attitudes and sophisti-
cated personal knowledge (what can be termed self-
expertise) of individual stakeholders can be imported
from the domain of the everyday into semi-formal
design environments, where they can contribute to
practical solutions. In such design settings, collec-
tive imagination emerges as a process of negotia-
tion. Here, the role of professional designers and
their design tools shifts from the production of final
outcomes, to supporting the essential components
of negotiation: promotion of empathy and trust, en-
couragement of conditions for safe experimentation,
erosion of stereotypes and cultivation of common
languages, amongst others.
Substantial and diverse work that engages with
similar ideas already exists, loosely clustered around
the notion of participatory design. Relevant themes
include: a shift away from the designer as the control-
ling expert (Jones 1979); a move beyond 'infrastruc-
ture' as a noun to 'infrastructuring' as a verb (describ-
ing a collaborative design framework) (Karasti 2014;
Clement et al. 2012); design for the inclusion of stake-
holders beyond "users" (Szebeko & Tan 2010; Bin-
nekamp et al. 2006); explorations of emergent and
interlinked ecologies (Thackara 2005; Fuller & Matos
2011); and the structured search for unexpected po-
tentials (Wood 2008). Seeking to extend this back-
ground through an innovative deployment of newly
available digital tools, the research discussed in this
paper explores opportunities provided by the intro-
duction of games into such participatory design pro-
cesses.
PROVOCATIVE GAMES
Traditional architectural representations are in-
tended to communicate already-established ideas.
Wishing to make the design process more inclusive
and open, the PocketPedal project seeks to deploy
media that reveals complex environmental inter-
actions. In this context, games emerge as useful
modes of engagement because they deploy genera-
tive mechanisms that can stage dynamic and immer-
sive processes. To use a definition proposed by Salen
& Zimmerman (2004, p.80), "[a] game is a system in
which players engage in an artificial conflict, defined
by rules, that results in a quantifiable outcome." An
influential school of thought promotes an argument
that simulations are the core and unique character-
istic of games (Frasca 2003). While the debate on
whether this is the case continues, game strategies
have been successfully expanding design methods
in a variety of contexts (Mayer et al. 2005; Kuboya
& Kimura 2005). When working through games, de-
signers do not shape stakeholders' experiences di-
rectly. They establish the rules of play; however, it is
the direct experience of games as open-ended pro-
cesses that encourages players to make meaningful
inferences.
Consequently, interactive technologies such as
games persuade in a variety of ways: through direct
experience, consequential actions, engagement with
other players and so on (Fogg 2003; Bogost 2007).
When aiming to persuade, games can make critical
arguments through the effects of gameplay events.
In such cases, the mechanism of in-game interactions
gains the capability to make political arguments and
suggest collective action.
Indeed, there exists a substantial body of work
on "serious" games, for example in education (Oak &
Bae 2014; Zielke et al. 2009; Walz & Deterding 2014).
Such games have been criticised for neglecting play-
ers' experience (Ferrara 2013) narrowing the game-
play options to the presentation of predetermined
outcomes, similar to other types of media (Pelletier
& Kneebone 2015).
The position of this paper is that games can play
serious roles in design processes without losing their
identity as games. Deployed strategically, games
have the potential to become useful design tools
that can lead to new discoveries where designers and
stakeholders create their own experiences in ways
un-envisaged, or at least undirected, by the original
VIRTUAL REALITY - Volume 2 - eCAADe 34 | 557
 game designers. Used in this context, games - and,
specifically, the PocketPedal prototype -- share the
conception and orientation of many strands of col-
laborative design. To evidence this point, the follow-
ing section discusses how a synthesis of urban cy-
cling and strategic game mechanics position Pock-
etPedal not as a cycling simulator but as a provocative
game that can be useful for participatory design.
POCKETPEDAL
PocketPedal is a game of real-time urban negotiation.
Played on the phone, the game offers a safe, accessi-
ble alternative to experiencing cycling along a busy
and dangerous inner-city road. In PocketPedal, play-
ers control a virtual cyclist and attempt to ride with-
out crashing. By navigating the virtual route, players
are exposed to the risks, challenges and high speed
decision-making associated with urban riding.
558 | eCAADe 34 - VIRTUAL REALITY - Volume 2
With an average penetration rate of 85% in the coun-
tries that are members of the Organisation for Eco-
nomic Co-operation and Development (OECD), and
over 100% in countries like Australia and the USA
(OECD 2016), people are at ease with phones. The ha-
bitual use of touch devices means many of the poten-
tial stakeholders are already somewhat familiar with
some aspects of the game's interaction model, re-
gardless of prior gaming experience. Deployed on
mobile devices, virtual worlds have the advantage
of not needing specific spaces that can host gam-
ing consoles, controllers or computer screens (Eriks-
son et al. 2005). They can instead be easily embed-
ded into many types of activities and benefit from the
richness of their surrounding contexts.
The virtual world of PocketPedal is a dangerous,
intense but very popular two-kilometre bicycle jour-
ney through inner-city Melbourne. An analysis of cy-
cling accidents revealed a segment of a major boule-
vard - St Kilda Road - where regular and serious Figure 2
events were particularly frequent (VicRoads 2016). A bike path crossing
PocketPedal implements road geometry, major inter- two lanes of traffic,
sections, car lanes and on-street parking spaces of representative of
the route. Combined with pavement markings re- the poor cycling
lated to bicycling and obstructions such as bollards infrastructure,
and lane changes, these elements define a course requires the player
that is demanding to navigate but rewarding to ex- to negotiate
plore [Figure 2]. multiple streams of
In contrast to previous virtual-cycling environ- fast traffic.
ments that often prioritise realism (Ranky et al. 2009;
Välkkynen et al. 2001; Badland et al. 2010; Herpers et
al. 2008; Kutz & Herpers 2008), PocketPedal focuses
on reconstructing complex and messy urban intensi-
ties. These infrastructural, logistical, cognitive, social
and political effects are simplified, exaggerated and
made accessible through play. Such safe accessibil-
ity - through the phone screen - is unconstrained by
previous cycling experience of the participants and is
accessible to a broad range of stakeholders.
Figure 3
A double parked
taxi slides into view,
blocking the path
forward. As in real
cycling,
PocketPedal’s
unprotected bike
lanes regularly force
a player into the
vehicular
road-space. Ahead,
a blue van opens its
front door; a
reminder that even
on the bike path
cyclists must always
be vigilant.
For instance, PocketPedal closely examines vulnera-
bilities of a cyclist in urban conditions. The game cre-
ates conditions where players operate in the state of
constant alertness typical for cycling, where risks of
bodily harm are much greater than that for motorists
(Garrard et al. 2010). PocketPedal's procedurally gen-
erated and automatically operated vehicles modify
road conditions in accordance with a weighting sys-
tem developed from site observations. Such condi-
tions change gradually, through multiple plays, but
also suddenly - in close proximity to the virtual rider.
Stationary vehicles become hazards by forcing play-
ers into traffic with their bulk or open doors [Figure
3]. Intruding into the experience of cycling, such sit-
uations closely resemble the dangers of parallel park-
ing as demonstrated, for example by Johnson et al.
(2013). An automated traffic system deploys individ-
ual vehicles with varying driving behaviours loosely
based on recorded driver-cyclist interactions (Walker
et al. 2014; Johnson et al. 2010). To give an exam-
ple, a vehicle can be reckless: less likely to slow down
near a player, and occasionally veering into the bicy-
cle lane. By contrast, a vehicle could be moving in
a way that matches the traffic flow and obeys road
rules.
The typical game mechanics of scoring and
health encourage players to examine the conse-
quences of poor cycling infrastructure and unpre-
dictable traffic. The game continuously assesses
a player's behaviour via a metric of 'road health'
that serves as a rough proxy for the fluctuating risk.
This rating, prominently displayed on screen at all
times, determines the amount of points the player is
awarded at ten-metre increments. Road health de-
cays when a player leaves the bicycle lane and in-
creases when a player follows its path. Good road
health is crucial for attaining a PocketPedal high score.
Through interaction with game mechanics such
as these, PocketPedal imposes cognitive and emo-
tional loads in players. On the road, a combina-
tion of infrastructural, behaviour and physical pres-
sures contribute to the delegitimision of cyclists (Al-
dred 2012). Most intense in urban conditions, the
need to resist this perceived message to 'get off the
road!' is psychologically taxing. To provoke such a
response in players, increasingly intense audio and
visual warnings are deployed when the virtual cy-
clist leaves the bicycle lane and enters vehicular road-
space. This sensorial feedback, coupled with the
threat of a collision and the consequential loss of
health and points are an example of an approach that
seeks to differentiate between the intensities of sim-
ulated urban environments. As cycling infrastructure
frequently changes and can be blocked by double-
parked cars and open vehicle doors, players, just as
real cyclists, have to rapidly evaluate and select ac-
tions in view of likely risks, while under time and scor-
ing pressures [Figure 4]. Unlike on the actual road,
these negotiations can be attempted in game multi-
ples times, on multiple cycling runs, without any risk
to life.
VIRTUAL REALITY - Volume 2 - eCAADe 34 | 559

Extending the ideas clustered around the work on
'persuasive games', the research discussed in this
paper also seeks to persuade, not directly, as pro-
paganda is meant to do, but circuitously, by self-
persuasion, in a mode that is more akin to the arts.
This approach deploys games as provocations that
seek to pose questions and urge participants to find
their own answers in the light of new experiences.
COLLECTIVE IMAGINATION
Based on the deployment of a fully functional pro-
totype in the mixed-stakeholder PocketPedal work-
shop, this paper argues that games can contribute to
design, especially when they are supported by other
design tools and design processes. Such games are
not intended as standalone products for individual
play. Instead, they are used in the presence of vari-
ous stakeholders including game developers as well
560 | eCAADe 34 - VIRTUAL REALITY - Volume 2
as expert architects, urban planners and engineers. Figure 4
Conventional gaming taxonomy would classify Local unavailability
PocketPedal as a single player game, yet in a de- of cycling
sign context it is strategically collaborative. Beyond infrastructure
acting as means of enjoyment or education, Pock- means a player is
etPedal seeks to provoke collective imagination. In suddenly
one example, the pixelated, colourful, non-real ren- surrounded by
dering system urges participants to focus on events moving vehicles.
rather than on appearances, encouraging a debate The player can
about identities of urban settings. In another exam- choose to continue
ple, performance and road-health points awarded by forward for extra
the game become invitations for communication be- points, or to turn
tween players of different backgrounds and experi- left to the relative
ences. safety of the
The interface between play and players is vital for kerbside lane. Road
successfully embedding electronic games into wider health, seen in the
participatory design activities. When a player 'loses' top right corner,
a game of PocketPedal by colliding with a vehicle, has dropped to
a physics engine calculates the force of impact and ‘poor’, indicating
presents the player with a crash type and a typical the dangerous
injury. A slow side swipe might result in a broken situation the player
toe; a high speed rear-end collision could lead to a has found herself in.
broken neck. One the other hand, if a player navi-
gates successfully, without crashing, a riding assess-
ment is given according to a player's compliance with
the cycling infrastructure. These crash and perfor-
mance messages, informative and comically ampli-
fied, become mechanisms for transferring the player
out of the virtual world and into an informed discus-
sion with surrounding people.
In such settings, participatory strategies such as
discussing, brainstorming, roleplaying or 'cardboard
computing' (Pelle & Kyng, 1991) can expand and en-
rich the interactions sustained within game environ-
ments. When games are embedded into a support-
ive setting, they do not need to be comprehensive,
technically complete or conclusive. As in the case of
PocketPedal, simulations that are embedded within
them can aim to be accurate or deliberately mislead-
ing; the burden of meaning-making falling not to the
games themselves, but to the participants who can
play them alongside with other tactics.
To test the usefulness of games as tools for
design negotiations, a fully functioning game pro-
totype was built and deployed in the PocketPedal
Workshop, a design workshop investigating the chal-
lenges of urban cycling along St Kilda Road. Work-
shop participants came from a diverse set of back-
grounds, with different, potentially conflicting, un-
derstandings of the site. Stakeholders included cy-
clists, motorists, transit users, a planner from the lo-
cal council and two health professionals. Data was
collected through recorded conversations, and two
questionnaires, one completed at the end of the
workshop and the other a week later.
An extensive report on workshop activities and
their impact on participants is beyond the scope of
this text. Instead, this paper will briefly discuss two
methods used to embed PocketPedal in participatory
activities, and the rich interactions evoked from this
collective imagining.
The first of these methods introduced stake-
holders to PocketPedal, gradually giving individuals
greater control. In this method, a volunteer plays
PocketPedal on a large screen, riding according to
the instructions given by other participants. Each of
these playing partners is given control over a simple
task associated with cycling: increasing or decreas-
ing speed, opening or reacting to car doors and so on.
Together, these participants form a collective cycling
'brain' that needs to interpret the conflicting com-
mands of the group and navigate the game world.
Through this approach, specific aspects of cy-
cling can be brought into focus. For example, the
player's movement and speed can be controlled in-
dependently, by two different people. Others can
modulate the actions, for example by adjusting the
collective's attitude to risk. Feedback indicated that
subdividing tasks in this way was useful:
• "I felt more confident analysing the game
when we had one task assigned to us."
• "It made me realize there is a reason that cy-
clists sometimes have to ride in what may
seem an "erratic" manner"
The analysis needed for such distributed cycling
changes individual games into experiences of group
negotiation. The decision of one participant has con-
sequences for all. For example, if the participant con-
trolling speed decides to go faster, the task of the per-
son reacting to doors becomes more difficult. The ef-
fect is the collective moderation of behaviour. The
group has a vested interest in 'winning' and collec-
tively overrides less successful instructions, 'bleed-
ing' commands across discrete tasks:
• [Speed Controller] "Go faster!"
• [Group] "No!!"
Thus, participants' actions are negotiated in the
midst of experience and have immediate conse-
quences: unsuccessful instructions leading to crash-
ing. During the workshop, participants reported
strong emotions, motivated by the immediate expe-
rience of the game.
• "When actually riding you have a protective
bubble of hope - that the cars and trucks will
avoid you. The game removes this comforting
assumption and brings home the fact that the
riders are so vulnerable."
• "By playing the game I have experienced vi-
ciously the feeling of cycling on St Kilda Road."
In contrast with shared play on a single screen as de-
scribed above, the second method for provoking col-
lective imagination involved individuals, each with
their own phone unit, playing PocketPedal at the
same time and in the same room. When participants
can take complete control of the virtual cyclist they
can adjust the experience for their needs. For exam-
ple, players can rerun situations they want to experi-
ence again, with new tactics.
Compared to group play, individual play re-
quires simultaneous personal management of mul-
tiple tasks: navigation, hazard detection, obstacle
avoidance, risk assessment, prediction and so on. The
flexibility of individual play makes it effective for ex-
ploring and testing cycling tactics in a safe environ-
ment. Yet PocketPedal played through this method
is still experienced collectively. Players share scores,
VIRTUAL REALITY - Volume 2 - eCAADe 34 | 561
 complain together about the game's inaccuracies,
cheer each other on and commiserate with each
other. Interactions like these create a safer space for
collaboratively examining cycling. For example, cy-
clists and motorists could triangulate a potentially in-
flammatory discussion about road interaction on a
shared language of high scores:
• A: "Two-twenty-eight. Two hundred and
twenty-eight. Just saying."
• B: [Turns around in shock] "You got a score of
228??"
Follow-up interviews a week later indicated that play-
ing PocketPedal in the workshop setting had a lasting
effect on how some saw the route:
• "Especially for the motorists who have not rid-
den a bike since they were kids and see rid-
ers as bloody nuisances rather than legitimate
road users. It changes their consciousness."
• "When I drive up, my focus is getting to the
destination quickly without really taking in my
surroundings. Playing the game has influ-
enced the way I think in St Kilda Road now. I'm
almost hyperaware of anything around me."
Through PocketPedal, a near-automatic daily com-
mute becomes enriched with experiences of others.
A cyclists' behaviours that might have been abstract,
unpredictable and frustrating become understand-
able as consequences of decisions made by individ-
uals who attempt to do the best when navigating a
complex and difficult urban environment.
Methods that can help to disrupt habits and pre-
conceptions can be productive in design many sit-
uations. Mundane experiences can obstruct com-
plex processes from being understood, precluding
productive negotiation and the invention of broadly
suitable alternatives. Participant feedback indicates
that imagining together through games can facili-
tate exchanges between diverse individuals with dis-
tinct worldviews and experiences. Negotiation be-
tween players, trial and error testing in a virtual en-
vironment, advice giving, friendly rivalry and score
562 | eCAADe 34 - VIRTUAL REALITY - Volume 2
comparison create a reciprocal exchange of mean-
ing. This engaging interaction can lead to individ-
ual and collective discoveries that would be impos-
sible without such facilitation. Provocative games
have the potential to reinforce such collaborative ex-
changes.
CONCLUSION: LUDIC DESIGN TOOLKIT
This paper describes PocketPedal, a smartphone
game that generates constructive design provoca-
tions. The discussion of PocketPedal as an effec-
tive aid for collective imagination proposes that such
games can be useful in design. This usefulness
emerges from the deployment of games alongside
other design tools, an assembly conceptualised here
as a Ludic Design Toolkit. A detailed discussion of
such toolkits cannot be accommodated within the
confines of this paper. Furthermore, the application
of such toolkits to large-scale practical problems is
yet to be attempted and will have to be reported in
subsequent publications.
Instead, this paper contributes by proposing ex-
perimentally motivated conceptualisations that can
enrich and extend the spectrum of design methods.
This spectrum already includes approaches that can
be conceptualised as toolkits. For example, multiple,
mature and widespread toolkits have significantly re-
duced time and expertise needed to create graphical
user interfaces. Toolkits also exist in architectural and
urban design (e.g., see Wart et al. 2010) where they
can be known as methods, standards, regulations or
survey and analysis techniques.
To expand this concept with additional options,
the research described in this paper combines smart-
phone gaming with participatory-design techniques
that include assessing video footage, playing card-
based games, enacting role-playing scenarios and
encouraging discussions between diverse sets of
stakeholders.
This work's interest in games is derived from their
capabilities as simulations. Simulations are com-
monly understood as representations. As representa-
tions, they aim to establish a close relationship with
some source environment, even in the cases where
this relationship cannot be predictive, or even repeat-
able. By extending such ambitions and sometimes
deliberately contradicting them, games can establish
alternative goals. Rather than trying to accurately re-
produce already-existing situations, they can stage
various types of speculative systems that seek to in-
terrogate the present or enact and critique possible
futures. These systems can aim to explore imagin-
able interactions, stimulate learning and interroga-
tion, provoke and even deliberately mislead to solicit
engaged participation and meaningful response.
Noteworthy effects of direct participation made
possible through the introduction of games and ob-
served during this research project allowed: 1) in-
action questioning of the typical and common sim-
plifications of conventional urban roles such as
drivers, pedestrians or cyclists; 2) exposure of impli-
cations of the perceived and practised habitual be-
haviours emanating from such stereotypical roles in-
cluding commuting, parking or joyriding; 3) trans-
positions permitting one person to experience the
world through the eyes of another: an elderly woman
inhabiting a fast cyclist or a non-cyclist struggling
to negotiate the traffic on a bike; 4) cultivation of
sustained reflection about values, for example, ex-
pressed as binary tensions of game dynamics: con-
venience vs diversity, transit vs destination, commut-
ing vs experiencing, convenience vs safety or self-
assuredness vs collaboration.
In conclusion, the preliminary evidence suggests
that the use of games can help to achieve and sus-
tain productive participation. This use of provocative
games in support of collective imagination does not
replace the capabilities of already-establish tools but
extends them in unusual and exciting ways.
REFERENCES
Aldred, R 2012, 'Incompetent or Too Competent? Nego-
tiating Everyday Cycling Identities in a Motor Domi-
nated Society', Mobilities, 8(2), pp. 252-271
Badland, HM, Opit, S, Witten, K, Kearns, RA and Mavoa,
S 2010, 'Can Virtual Streetscape Audits Reliably Re-
place Physical Streetscape Audits?', Journal of Urban
Health, 87(6), pp. 1007-116
Bell, W and Ferretti, D 2015, 'What Should Planners Know
about Cycling?', in Bonhom, J and Johnson, M (eds)
2015, Cycling Futures, University of Adelaide Press,
Adelaide, AU, pp. 321-356
Binnekamp, R, van Gunsteren, LA, van Loon, PP and
Barendse, P 2006, Open Design: A Stakeholder-
Oriented Approach in Architecture, Urban Planning,
and Project Management, Ios Press, Amsterdam
Bogost, I 2007, Persuasive Games: The Expressive Power of
Videogames, MIT Press, Cambridge, MA
Clement, A 2012 'Probing, Mocking and Prototyping:
Participatory Approaches to Identity Infrastructur-
ing', Proceedings of the 12th Participatory Design Con-
ference: Research Papers - Volume 1, New York, pp.
21-30
Cınar, A and Bender, T 2007, 'The City: Experience, Imag-
ination and Place', in Cınar, A and Bender, T (eds)
2007, Urban Imaginaries: Locating the Modern City,
University of Minnesota Press, Minneapolis, MN, pp.
xi-xxvi
Eriksson, D, Peitz, J and Bj\"ork, S 2005 'Socially Adapt-
able Games', Proceedings of the 2005 Digital Games
Research Association (DIGRA) Conference: Changing
Views: Worlds in Play, Vancouver, pp. 1-8
Ferrara, J 2013, 'Games for Persuasion: Argumentation,
Procedurality, and the Lie of Gamification', Games
and Culture, 8(4), pp. 289-304
Fogg, BJ 2003, Persuasive Technology: Using Computers to
Change What We Think and Do, Morgan-Kaufmann,
Amsterdam; Boston
Frasca, G 2003, 'Simulation versus Narrative: Introduc-
tion to Ludology', in Wolf, MJP and Perron, B (eds)
2003, The Video Game Theory Reader, Routledge,
New York, pp. 221-235
Fuller, M and Matos, S 2011, 'Feral Computing: From
Ubiquitous Calculation to Wild Interactions', The Fi-
breculture Journal(135), pp. 144-163
Garrard, J, Greaves, S and Ellison, A 2010, 'Cycling Injuries
in Australia: Road Safety', Journal of the Australasian
College of Road Safety, 21(3), pp. 37-43
Herpers, R, Heiden, W, Kutz, M, Scherfgen, D, Hartmann,
U, Bongartz, J and Schulzyk, O 2008 'FIVIS Bicycle
Simulator', Proceedings of the 2008 Conference on Fu-
ture Play: Research, Play, Share, New York, pp. 244-
247
Johnson, M, Charlton, J, Oxley, J and Newstead, S 2010,
'Naturalistic Cycling Study: Identifying Risk Factors
for On-Road Commuter Cyclists', Annals of Advances
in Automotive Medicine, 54(Jan), pp. 275-283
Johnson, m, Newstead, S, Oxley, J and Charlton, J 2013,
VIRTUAL REALITY - Volume 2 - eCAADe 34 | 563
 'Cyclists and Open Vehicle Doors: Crash Characteris- 229-235
tics and Risk Factors', Safety Science, 59, pp. 135-140 Walker, I, Garrard, I and Jowitt, F 2014, 'The Influence of a
Jones, JC 1979, 'Designing Designing', Design Studies, Bicycle Commuter', Accident Analysis and Prevention,
1(1), pp. 31-35 64, pp. 69-77
Karasti, H 2014 'Infrastructuring in Participatory Design', Walz, SP and Deterding, S (eds) 2014, The Gameful World:
Proceedings of the 13th Participatory Design Confer- Approaches, Issues, Applications, MIT Press, Cam-
ence: Research Papers - Volume 1, New York, pp. 141- bridge, MA
150 Wood, J 2008 'Changing the Change: A Fractal Frame-
Kuboya, M and Kimura, S 2005, 'The Game of “Forming work for Metadesign', Changing the Change: Design,
a Coalition Government”', in Shiratori, R, Arai, K and Visions, Proposals and Tools, Turin, pp. 1-8
Kato, F (eds) 2005, Gaming, Simulations and Society: Zielke, Ma, Evans, MJ, Dufour, F, Christopher, TV, Don-
Research Scope and Perspective, Springer, Tokyo, pp. ahue, JK, Johnson, P, Jennings, EB, Friedman, BS,
11-20 Ounekeo, PL and Flores, R 2009, 'Serious Games
Kutz, M and Herpers, R 2008 'Urban Traffic Simulation for for Immersive Cultural Training: Creating a Liv-
Games', Proceedings of the 2008 Conference on Future ing World', IEEE Computer Graphics and Applications,
Play: Research, Play, Share, New York, pp. 181-184 29(2), pp. 49-60
Mayer, I, Leijten, M, Carton, L, de Jong, M, Scalzo, R, [1] https://www.data.vic.gov.au/data/dataset/crash-
Dammers, E and Verwest, F 2005, 'The Urban Net- stats-data-extract
work Game: A Simulation of the Future of Joint [2] http://www.oecd.org/internet/broadband-statisti
City Interests', in Shiratori, R, Arai, K and Kato, F cs-update.htm
(eds) 2005, Gaming, Simulations and Society: Re-
search Scope and Perspective, Springer, Tokyo, pp.
39-48
Oak, JW and Bae, JH 2014, 'Development of Smart Multi-
platform Game App Using Unity 3D Engine for CPR
Education', International Journal of Multimedia and
Ubiquitous Engineering, 9(7), pp. 263-268
Pelle, E and Kyng, M 1991, 'Cardboard Computers:
Mocking-it-up or Hands-on the Future', in Green-
baum, J and Kyng, M (eds) 1991, Design at work,
Lawrence Erlbaum Associates, pp. 169-195
Pelletier, C and Kneebone, R 2015, 'Playful Simulations
Rather than Serious Games: Medical Simulation as a
Cultural Practice', Games and Culture, 11, pp. 1-25
Ranky, R, Sivak, M, Lewis, J, Gade, V, Deutsch, JE and
Mavroidis, C 2009 'VRACK — Virtual Reality Aug-
mented Cycling Kit: Design and Validation', 2010
IEEE Virtual Reality Conference, Waltham, MA, pp.
135-138
Salen, K and Zimmerman, E 2004, Rules of play : game de-
sign fundamentals, Cambridge, Mass. ; London : MIT
Press, c2004.
Szebeko, D and Tan, L 2010, 'Co-designing for Society',
Australasian Medical Journal, 3(9), pp. 580-590
Thackara, J 2005, In the Bubble: Designing in a Complex
World, MIT Press, Cambridge, MA
Valkkynen, P, Heinila, J, Lainio, S, Lakaniemi, S and Vaata-
nen, A 2001 'Using Exercise Cycle as a Haptic Input
Device in a Virtual Environment', Immersive Projec-
tion Technology and Virtual Environments 2001: Pro-
ceedings of the Eurographics Workshop, Stuttgart, pp.

564 | eCAADe 34 - VIRTUAL REALITY - Volume 2

BIM Collaboration in Virtual Environments
Supporting collaboration in co-located and distributed settings

Uwe Woessner1 , Joachim B. Kieferle2

1
High Performance Computing Center Stuttgart 2 Hochschule RheinMain
1
woessner@hlrs.de 2 joachim.kieferle@hs-rm.de

Combining BIM and Virtual Reality improves one of the core concepts of BIM: to
foster the collaboration between different disciplines. In this paper we present
two new approaches. One is the collaboration with BIM/VR in co-located as well
as distributed Virtual Environments. Different aspects of collaboration are
discussed and shown. The other new approach is the documentation of BIM/VR
meetings. The documentation can be generated semi-automatically from within
the Virtual Reality environment first for traditional office documents and second
for the documentation directly in the BIM model. Initial tests with projects from
practice as well as numerous interviews with practitioners have proven the
relevance and benefit for practice.

Keywords: BIM, Virtual Reality, Remote and co-located collaboration, CSCW

INTRODUCTION 1. Collaboration: co-located meetings and

The integration of multiple disciplines is one of the
main advantages of BIM. Whereas the technical is-
sues have been explored quite intensively and solved
to a certain degree, there is still a substantial need
for tools and research in how to support the hu-
man issues of interaction and meetings between ar-
chitects, engineers, owners and others. This paper
builds upon previous research on the integration of
BIM and Virtual Reality (VR) (Kieferle and Woessner
2015). There, we presented a Revit Plugin which
provides a real time bi-directional link between the
BIM system and a virtual environment. This system
not only provides an immersive 3D visualization of
the BIM model but also allows modifications of the
model in Revit and from within the Virtual Environ-
ment (VE). The focus in this paper is now on two new
approaches for BIM/VR based meetings:
meetings in distributed VEs
2. Documentation: in form of meeting minutes
in the BIM model and in traditional office ap-
plications.
In general, VR is proven to be an efficient communica-
tion platform. Within this paper we present how this
can be applied to regular formal meetings and how
it can be extended to spatially distributed VR meet-
ings with a focus on BIM. The main research ques-
tions, that we discuss and show solutions for in this
paper are:
• how to enhance collaboration in BIM/VR at lo-
cal and remote locations
• how to provide enough nonverbal clues to re-
mote partners for efficient communication
• how to record BIM/VR sessions (local and re-
mote) for protocol purposes

VIRTUAL REALITY - Volume 2 - eCAADe 34 | 565

 Meetings mostly bring together a group of people
to jointly reconcile conflicts, reach a group judg-
ment or decision and solve a problem (Romano and
Nunamaker 2001). They further researched that most
meetings last between 1/2 to 2 hours, a usual time
frame, well suited for BIM meetings in VEs.
To make a meeting successful, visualization is
one of the key aspects for clarification and to avoid
misunderstandings (Seifert, 2015). These aspects are
directly supported in AEC meetings in VEs, since all
participants don't need to abstract from plans or
other scaled representations but can directly interact
with the virtual building model at scale 1:1. This new
setup however raises questions about how to docu-
ment a joint session and how to interact in various
settings from co-located VEs over distributed VEs up
to mixed types of VEs and combinations of VEs and
desktop workspaces.
COLLABORATION: DESIGN MEETINGS IN
REAL AND VIRTUAL ENVIRONMENTS
Throughout a building life cycle, there are many sit-
uations where regular team discussions are needed,
from idea finding meetings, architectural design re-
views, financial planning meetings over construction
planning up to regular construction meetings and
beyond.
These meetings can be held in different ways: as
traditional meetings, group meetings in one virtual
environment or remote collaborative in spatially dis-
tributed virtual environments.
Traditional scenario: All participants meet
in a real world environment
In a real world construction site environment, team
discussions with all participants are very intuitive.
Informal communication is mostly done verbal in
face to face discussions. Nonverbal clues such as fa-
cial expressions or body gestures are also important
for a successful communication. First and foremost
the pointing gesture is important to provide context
for the discussion. Without being able to point to
an object on a plan, a model or the real world, iden-
566 | eCAADe 34 - VIRTUAL REALITY - Volume 2
tifying objects by verbal description only becomes
time consuming and error prone. Modifications to
a model are the most direct form of communication
which leads to least misunderstandings. Paper plans
can hardly be modified but they can efficiently be ex-
tended by sketches on tracing paper for example.
The formal communication is done through ex-
change of plans and a protocol of the meeting.
Figure 1
Typical setup of a
team meeting. a
monitor showing
Word minutes on
the left ,the CAVE in
the center and a
person adding
comments to
minutes on the
right.
Co-located VE scenario: Local, computer
supported, all participants are physically in
the same virtual environment
In a virtual environment like a CAVE (Cave Automatic
Virtual Environment; figure 1) or powerwall, all par-
ticipants physically share the same space, time and
virtual world. Interactions like in the real world envi-
ronment are used, with some limitations or different
requirements:
• verbal: no difference to real world collabo-
ration; participants can talk freely with each
other
• non verbal: body gestures - people can see
each other, thus there is not much difference
to traditional meetings in nonverbal commu-
nication with one exception: Due to the dif-
ferent perspectives, pointing at virtual objects
with your fingers does not work. You have to
use a virtual pointer or a laser pointer to do so.
• modifications: the BIM model can be changed
directly from within the VE by moving objects
 or changing parameters to improve the de-
sign. Sketches are not possible yet but might
be implemented in the future.
So the informal communication is done through a
joint experience of the 3D world.
Formal communication can be realized through
"Revit design options". Meetings can be documented
on the fly. Annotations can be placed anywhere in
the model and snapshots can be taken any time.
Both, annotations and snapshots, are automatically
added to meeting minutes in an office document
which can be freely extended.
Distributed VE scenario: Remote collabora-
tion, participants are located in distributed
virtual environments
The long distance collaboration becomes an issue,
as direct interactions like speaking or direct body
gestures have to be supported or replaced by other
methods:
• verbal (voice): direct communication is re-
placed by audio conferencing with room mi-
crophones for group discussions. This re-
quires good echo cancellation but once that
is in place, there is not much difference in ver-
bal communication.
• nonverbal clues: The position of other par-
ticipants is represented by avatars. They do
not show the whole person but only a very
abstract representation consisting of glasses
(showing gaze), pointer hand (showing point-
ing direction), a foot pad (showing assumed
position of feet) as well as an ID label in order
to identify the other participants.
• modifications: as in the Co-located scenario,
bi directional communication with the BIM
model is possible from all participating sites
in a distributed VE.
Informal communication is still possible as in the co-
located scenarios but in general remote collaborative
sessions seem to work better if they are a bit more
formalized. We achieve this through three different
collaboration modes:
• Master/Slave mode: All views in the collabo-
rative are tightly synchronized and only one
of the participants, the "Master" can interact
with the application. This is good for situa-
tions, where one of the participants presents
the current state of the planning or construc-
tion to the rest of the team. In this situation, it
is not desired that the other participants look
at other areas or even interact with the scene.
As needed, the master role can be passed on
to any other participant.
• Tight coupling mode: All participants share
the same space, their views are synchronized
but they can freely interact with the applica-
tion. In more creative situations, you might
want every participant to be able to modify
the object in discussion. For example if you
want to jointly work out a new entry door -
some of the participants might modify the ge-
ometry, others might work on the materials or
document and annotate the object.
• Loose coupling mode: In this mode, the views
are not synchronized which means that all
participants can walk through the building in-
dependent of each other and everyone can in-
teract with the application.
Figure 2 shows a collaborative session between a user
in a five sided cave on the left side and a user in
front of a 3x3 tiled display. The session is currently in
"Loose coupling" mode. Both users share the same
data-set, a hydro power station, but look into differ-
ent directions. In both images, you can see glasses,
hand and feet of the avatar of the remote user.
DOCUMENTATION: SEMI-AUTOMATED
MINUTES OF MEETING
Documenting the results of team sessions in virtual
environments is crucial for a successful team collab-
oration. Up to now this has been done only manu-
ally in our VR environment, mostly with hand notes
during the meeting or even after the meeting. Thus
VIRTUAL REALITY - Volume 2 - eCAADe 34 | 567
 Figure 2
Collaborative
session between
users in a CAVE and
in front of a 3D tiled
display.

the documentation, especially when numerous top-
ics were discussed, lacked precision, structure and a
good degree of self explanation / visual reference.
This is, why we implemented two methods to ease
the documentation: Semi-automated documenta-
tion in an office document and semi-automated doc-
umentation in the BIM model. Both methods overlap
to a certain degree.
BIM / VR meeting documentation in office
documents
Any time during a team discussion in VR, the partic-
ipants can take 2D snapshots of the current 3D view
(see figure 3). They can also make annotations in
form of colored arrows with attached labels. To add
a new annotation, you have to point to an abject and
click the 3D mouse button. This will place an arrow
at that location and add a label with a consecutive
number. The number is specific enough to refer to
in an office document. However if required, one can
change the label to any text through the tablet user
interface. This user interface also allows to change
the color and relative size of the arrow. The actual
size of the arrow is adjusted so that it looks similar in
size independent of model scale and distance to the
viewer. Annotations which are far away are rendered
larger so that they are still readable from the users
position. In collaborative sessions, it depends on the
current collaboration mode whether any participant
can modify and annotate. In Master-Slave mode, one
can only modify her or his own annotations but not
annotations created by a remote user. In tight cou-
pling mode however, you can modify all annotations.
The absolute scale of the annotations is determined
locally so that labels are readable from any location
and on any type of device.
Figure 3
Office document
with automatically
taken snapshot
from the CAVE and
manually added
comments.
A comprehensive video recording plugin allows to
record live 2D or 3D videos from a VR session.
Through a tablet user interface, the user can chose
between different frame rates, encoders and resolu-
tion and other encoder specific parameters. In a col-
laborative session, videos are only recorded in the lo-

568 | eCAADe 34 - VIRTUAL REALITY - Volume 2

 cal VE. Especially for conducting research on the be- tation, documentation directly in the BIM model (fig-
havior of users in virtual buildings, the motion and ure 4) provides a much higher degree of information.
gaze direction of the user can be recorded. Voice With the bi-directional link between the VR software
recording is not yet possible but we plan to integrate COVISE [2] and the BIM software Revit, interactions
that. can be made in both environments. However infor-
In many current projects not all participating plan- mation is strictly stored in the BIM database so that it
ners are working with BIM software. Thus, there can later be retrieved and worked on.
is a need of providing traditional documentation
Figure 4
in form of an office document alongside the BIM
Annotations in
database (figure 3). These documents typically con-
Revit model.
tain a header, providing information on participants,
date and venue of the meeting and the document
versions which are discussed during the meeting.
This general meeting information is typically struc-
tured in templates and then filled in manually. Au-
thoring of individual action items however can be ef-
ficiently supported by the VR system. Therefore we
implemented a direct link from the VE to Microsoft
office applications Word, Excel and PowerPoint. At
When walking through the virtual building in VEs, lo-
the current development state, whenever 2D snap-
cal and remote participants stop at a certain posi-
shots are taken in the VR environment, they are auto-
tion and add annotations directly in the VR environ-
matically inserted into the Word document that the
ment to mark critical issues. Annotations carry at-
minutes keeper holds on her or his computer. She
tributes such as position, orientation and the creator
or he adds text to the screenshots describing the dis-
(owner) of the annotation. This information is auto-
cussions that took place, on the fly. The snapshots
matically stored in the BIM model as TextNotes. For
are extended by information about their position and
example discussed modifications are labeled with a
orientation in the 3D model.
due date, responsible person and one or multiple ap-
In a follow up meeting, the snapshots can there-
proval check marks. Those are processed after the
fore be revisited in the VE. Snapshots can be double
meeting in a formal way, item by item and rather than
clicked in the office document and the VE jumps to
trying to keep track of those on paper, this informa-
the location of that snapshot. This allows a quick way
tion is added to the BIM model as attributes to 3D
to navigate to the location in question without the
objects or labels. Whereas the documentation in real
need of manually walking through the building and
world environments is kept in a written protocol, thus
searching for the previous position.
not directly linked to the building elements, with our
new approach the discussed information is directly
BIM / VR meeting documentation in Revit
linked to the building elements in the 3D BIM model,
document
even at the exact 3D location. Thus the BIM model
As explained beforehand, working in virtual environ-
can later be revisited either in the desktop applica-
ments differs from sessions in a real world environ-
tion or in a VR environment to check whether every-
ment. This requires and enables other methods of
thing has been taken care of. Users of the BIM sys-
documentation, which can even be of higher value
tem can filter for newly created annotations, work on
than the traditional ones. Whereas the office docu-
them and mark them as resolved, approved and so
mentation is well known from traditional documen-
on.

VIRTUAL REALITY - Volume 2 - eCAADe 34 | 569

 TECHNICAL IMPLEMENTATION
The implementation is based on Autodesk Revit as
a BIM software running on a Windows workstation.
Revit is extended by a C# Plugin which realizes the
communication with the VE. The VE is based on
COVISE/OpenCOVER, an open source VR Framework
which has been extended by a plugin that communi-
cates with Revit. Microsoft Word, PowerPoint and Ex-
cel are extended by Add-Ins which implement com-
munication between the respective office applica-
tion and the virtual environment OpenCOVER. All
of the components discussed in this paper are li-
censed under the LGPL v2.1 and thus can be used for
free in non-commercial and commercial applications.
Source code of COVISE is published as a Git repository
on github [2]. All plugins communicate with each
other through TCP sockets. Therefore the virtual en-
vironment can run on a different workstation or even
a computer cluster.
The communication protocol is based on the CO-
VISE Messaging infrastructure. The COVISE Kernel
provides a C++ communication library (coNet) which
provides classes for raw sockets, connections and
Messages. Messages have a simple header consisting
of a type, size, sender and payload and can be sent
over connections to other applications. The message
payload can be a binary buffer or a TokenBuffer which
provides serialization of a number of standard data
types, strings, integers floats, doubles and so on.
A C# implementation of this library has been de-
veloped so that both the Revit plugin as well as the
office plugins which are written in C# .NET could be
implemented. A message thread listens for data on a
socket, stores received messages in a message queue
and informs the application of the new messages. In
the main thread of the Add-In, messages are taken
from that queue and passed on to a message handler.
570 | eCAADe 34 - VIRTUAL REALITY - Volume 2
In case of Revit, we differentiate between two
types of messages, read only messages and mes-
sages which require write access to the BIM database.
For the latter, a transaction is generated and on
successful completion of the message handler, this
transaction is committed and thus all modifications
are stored in the database. In case of a failure, if for ex-
ample not all constraints could be satisfied, the trans-
action is cancelled and all of the database modifica-
tions are undone.
Distant collaboration typically poses high re-
quirements on network performance. Network per-
formance is mainly described through three parame-
ters: Bandwidth, latency and jitter. In our case, band-
width is typically not of importance, as the amount
of data transferred during a collaborative session is
an avatar position and orientation which is a couple
of bytes only at a maximum frequency of 20 mes-
sages per second. Everything else is sporadic trans-
fer of small data updates such as parameter modifi-
cations or positions of moved objects. Latency how- Figure 5
ever is of special importance if concurrent manipula- Ribbon user
tion of objects is allowed because then, each inter- interface of the
action has to be locked before it can be safely ap- Revit plugin.
plied, which requires sending a request for modifica-
tion and waiting for a reply. Latencies in the range of
seconds are still common during trans-atlantic com-
munications. To overcome these limitations, we im-
plemented a lazy locking mechanism which allows
to start an interaction immediately and only cancels
that interaction if a rare conflict with a remote inter-
action is encountered. This allows for fluent interac-
tions like in non-collaborative environments even in
high latency settings. Jitter negatively impacts band-
width and latency critical cases but as we successfully
hide latency in the application, jitter also does not re-
ally impact our application performance. Bandwidth
is only an important aspect during startup of a collab-
orative session when the whole BIM dataset has to be
transferred from the BIM workstation to all participat-
ing VEs.
Bandwidth measurement during an end user
study show a maximum bandwidth requirement dur-
ing initial transmission of the geometry from Re-
vit to the VR environment of only 20 MBit/s. This
does not even saturate a wireless connection. Dur-
ing normal operation, only view point information
and other synchronization data is transferred at not
more than100 KBit/s. More complicated interactions
which require communication to all partners in a col-
laborative session and to Revit such as moving anno-
tations can reach a bandwidth of up to 600 KBit/s.
These numbers clearly show that collaborative ses-
sions even over wide area networks with multiple
participants are definitely feasible.
EVALUATION
The developed methods have been tested with prac-
titioners on real life projects from small to large scale.
These first tests and the feedback from the practi-
tioners show, that the new approach of team collab-
orations in distributed VEs is promising and might
change the way how team sessions in BIM/VR envi-
ronments can be carried out and documented in the
future. Remote collaboration on the BIM model as
well as the collaboration in distributed VEs already
works very well. The evaluation sessions generated a
lot of further user requests, the most important ones
are:
• additional information for snapshots and an-
notations such as time stamps, room numbers
or axis information.
• linking viewpoints of snapshots with Revit
cameras and linking annotations to bullet lists
in the word document and text notes in Revit
at the same time.
• extending avatar pointing rays.
• Providing better visual and / or acoustic feed-
back when taking snapshots.
• improvement in office document - separate
insert position for snapshots and annotations
to prevent conflicts when editing minutes
while taking new snapshots.
• Multiple physical input devices for multiple
users so that everyone in a session can point
at objects and add annotations.
• Annotations in form of 2D sketches should be
supported.
• In early versions of BIM databases, room in-
formations are often not available yet. It was
thus recommended to include axis labels in
the room information and location of snap-
shots and annotations.
• The pointer of the avatar is too short and
should be extended to see more clearly where
they are pointing at.
• Two types of markers should be supported:
Temporary markers, just for discussions which
are not stored in the database and permanent
markers.
In some companies, PowerPoint is used to write
meeting minutes the request was therefore to also
implement automatic snapshots in PowerPoint. This
would also allow to better store video captures in the
minutes. A tighter link between annotations in the
Revit database and the word document was also sug-
gested so that BIM attributes on text notes such as ac-
ceptance marks can easily be synchronized between
the office document and the BIM database.
CONCLUSION AND FUTURE WORK
The presented methods show the potential of fos-
tering collaboration with BIM in VEs as well as the
new methods for documenting meetings in the vir-
tual model. One of the most interesting benefits is
the spatial linking of meeting information to virtual
building elements. This increases precision in the
communication. By linking VEs and jointly discussing
projects, users from all locations can interact with
the virtual model and thus attach information to the
BIM database for later retrieval. However as the pre-
sented methods tackle VR, the BIM database as well
as Human Computer Interaction, it needs a lot of fur-
ther research, development and testing.
Next steps will be the implementation of the user
requests and a formal evaluation of the performance
within such a collaborative working environment.
Additional future work beyond this paper will be fo-
cussing on improving navigation in VEs by automat-
VIRTUAL REALITY - Volume 2 - eCAADe 34 | 571
 ically extracting floor plans from the BIM database
and providing this as a navigation aid on a tablet or
multi touch table as well as the integration of Aug-
mented Reality and touch tables / planning tables to
further link the virtual and the real world.

ACKNOWLEDGEMTNS
We would like to thank the team members in Züblin
construction company (team led by Alexander Kuhn)
and OPTIMA Pharma (team led by Ralph Eisen-
schmidt) for their testing and most valuable feed-
back.

REFERENCES
Bassanino, M, Fernando, T and Wu, KC 2014, 'Can vir-
tual workspaces enhance team communication and
collaboration in design review meetings?', Architec-
tural Engineering and Design Management, 10(3-4),
pp. 200-217
Bassanino, M, Wu, KC, Yao, J, Khosrowshahi, F, Fernando,
T and Skjaerbaek, J 2010 'The impact of immersive
virtual reality on visualisation for a design review
in construction', Information Visualisation (IV), 2010
14th International Conference, pp. 585-589
Romano Jr, N and Nunamaker Jr, J 2001 'Meeting Anal-
ysis: Findings from Research and Practice', Proceed-
ings of the 34th Annual Hawaii International Confer-
ence on System Sciences ( HICSS-34)-Volume 1 - Vol-
ume 1, Washington, DC, USA, pp. 1072-
Kieferle, J and Woessner, U 01 'BIM Interactive - About
combining BIM and Virtual Reality - A Bidirectional
Interaction Method for BIM Models in Different En-
vironments', Proceedings of the 33rd eCAADe Confer-
ence, Vienna University of Technology, Vienna, Aus-
tria, pp. 69-75
Seifert, JW 2015, Besprechungen erfolgreich moderieren:
Kommunikationstechniken für Leiter und Teilnehmer,
GABAL Verlag GmbH
Wössner, U, Schulze, JP, Walz, SP and Lang, U 2002 'Eval-
uation of a collaborative volume rendering applica-
tion in a distributed virtual environment', Proceed-
ings of the 8th Eurographics Workshop on Virtual En-
vironments (EGVE), pp. 113-122
[1] http://usa.autodesk.com/adsk/servlet/index?id=2
484975&siteID=123112
[2] https://github.com/hlrs-vis/covise

572 | eCAADe 34 - VIRTUAL REALITY - Volume 2

VR or Not VR - No Longer a Question?
Wolfgang Dokonal1 , Michael Knight2 , Ernst Dengg3
1,3
TU Graz 2 University of Liverpool
1,3
{dokonal|dengg}@tugraz.at 2 mknight@liverpool.ac.uk

Virtual Reality (VR) software has developed to the point where, for the architect
who is averagely technically adept, it can be incorporated into the design process
with reasonable effort and costs. For VR to be an effective design tool, it must
add value to the design process and should give insights and opportunities not
available by other methods.Previous research by the authors reported on the
results of an international student workshop which focused both on the workflow
(to prepare the architectural models for the new VR systems) and the spatial
perception that users experienced. In this paper, we continue to explore the
question: "Can low cost VR be an effective addition to the architects' design
toolbox, or does it still remain a "far-fetched, high-tech expensive folly?"To do
this we are working with a larger group of students, a more developed workflow
and we are also expanding this to architects in practice. We will be assessing
both the practicality of integrating VR into the design workflow and the spatial
perception of the designer when interacting with the model. We are experimenting
with additional interface tools.

Keywords: Virtual Reality, Google Cardboard, Low Cost Interfaces

INTRODUCTION still unclear - at least outside of academia and the

2016 has the potential to become a turning point
for architects in the use of VR and/or Augmented
Reality (AR) in the design process. VR has been on
the agenda for a very long time but has never really
managed to have a significant impact on the design
workflow of a "normal" sized architectural practice.
The reasons are quite clear: up to now, a substan-
tial amount investment, both financial for the equip-
ment, complex workflows that are necessary to bring
the geometry into the systems together with com-
plex technical settings with a lot of technical issues
(e.g. synchronisation of different PC's, projectors etc.)
which needs a lot of preparation to get it working re-
liably. And after all this work, the main benefits were
Global companies. There has been a considerable
amount of research done which is concerned with
high cost and large VR installations such as Power-
walls and CAVEs but, for the 'average' small architec-
tural practice these have been of little practical use.
So, until now VR has never been a viable option
for 99.9% (or even less) of architectural firms. Most of
them would not have considered using VR for even
for final project presentations and even less for using
it during the design process - mainly because of the
reasons mentioned above. Recent advances and de-
velopments in low cost VR hardware and software, in
particular those that use mobile phones as a compu-
tational device, have given all architects and design-

VIRTUAL REALITY - Volume 2 - eCAADe 34 | 573

 ers the opportunity to use VR as an effective part of
the design process. The new changes in the whole
system are driven by the interests of the gaming in-
dustry and today, this a powerful and economically
flourishing industry with a great deal of available re-
sources.
With the new devices that are already available
or will be available very soon the "game" is chang-
ing and we think that architects should finally make
use of this opportunity. With the Oculus Rift Con-
sumer Version, Sony's Morpheus, HTC's Vive, Sam-
sung's Gear and the improved Google Cardboard
among others, there is a wide range of extremely low
to very low cost VR devices available for the general
public. Microsoft Hololens is now available for devel-
opers in the US and Canada and also seems to have a
lot of potential. In our opinion it is now the right time
for architects to rekindle an interest in these systems
because they have the potential to add a new level
to the design as well as the presentation process. To-
gether with free or low cost software and an easy to
use workflow there is the opportunity for everyone
who is interested. As in many areas, the car industry
leads the way. Car manufactures have been using ex-
pensive and complex VR systems as an integral part in
the design process for decades and now start to use
the low cost VR versions for giving potential buyers
an immersive experience in configuring their future
car. But architecture is starting to catch up : The Pre-
fabrication - Housing industry is starting to show an
interest in these new devices and there are some ex-
amples how they make use of the low cost VR and AR
systems. The APP Roomle from an Austrian producer
is a good example. (www.variohaus.at). Also the new
Zaha Hadid project for Graz already uses VR and AR
technologies for smartphones - see the "Argos Graz"
App to visit the unbuilt project
THE 2016 SYSTEMS
The current research focuses on the use of smart-
phones with Google Cardboard and the Oculus Rift
systems. We used two approaches to bring the 3D
geometry models into the VR systems. The first was
574 | eCAADe 34 - VIRTUAL REALITY - Volume 2
to use the game engine software 'Unity3D' to prepare
the 3D geometry for the Oculus Rift systems. We also
used Unity3D to create Google Cardboard versions
for Android phones. Unfortunately, we could not cre-
ate iOS versions due to Apple requiring (paid) devel-
oper registration.
The second approach was to use the 'Ku-
bity'(www.kubity.com). This provides a cloud based
service where 3D models are uploaded to a server
and prepared automatically for the use in combina-
tion of Android smartphones in 2D panorama form
and Google Cardboard devices for full 3D. This is a
straightforward process where the geometry is up-
loaded; a QR code is displayed on the PC screen as a
link to your own model in VR. The QR code is scanned
from the smartphone app and the model is down-
loaded. That sounds perfect and in some ways it is,
but it comes with a number of disadvantages. There
is no collision detection implemented yet and there
are some problems with the movement inside the
models - we will discuss this problem in more details
later.
However, development in this field is very fast
and we can assume that there will be updates to 'Ku-
bity' that will resolve some of the known issues - e.g.
a recent update adds the possibility to use the open
wavefront obj file format. Currently the service is
free, but it might be expected to be monetised at
some point. There are probably also some issues of
data security when architects plan to use this app in
the design process that are more important. We as-
sume that there will be a certain reluctance to upload
preliminary designs to an unknown server when you
work as an architect on a project or a competition.
This time it was no issue for the models we used.
THE TASK FOR THE STUDENTS
As last year, we wanted the students to model their
own flat using either SketchUp or other modelling
software; the assumption being they would have a
good knowledge of their own living environment.
Because of that they are very good candidates for ex-
periencing these "real world" environments in the VR
 systems and give feedback about their experience us-
ing them.
To give an extra 'virtual' dimension to the re-
search process we added two small architectural
built projects that we had received from an architec-
tural practice so the students could also experience
projects unknown to them. These projects were two
houses in rural locations in the UK (see Figure 1). To
keep the scale similar to the student's flats, only part
of the houses were modelled. Unfortunately the de-
signers themselves could not be part of the work-
shop but provided feedback on the spatial experi-
ence that it gave them.
Figure 1
Modelled interior of
unfamiliar house
(Design by Darren
Ward).
THE WORKSHOP THIS YEAR
The workshop was structured similar to last year and
was again taken by Masters and undergraduate stu-
dents from TU Graz (Austria) and the University of
Liverpool (UK). During the first phase in Graz stu-
dents modelled their own flats using SketchUp. Dur-
ing the second phase in Liverpool, students devel-
oped their models further and then tested them us-
ing the VR equipment. They completed a question-
naire which recorded their experience and accuracy
of spatial perception. One difference this year was
that the VR software/Google Cardboard integration
had developed rapidly since last year. This resolved
some, but not all of the problems encountered.
PROBLEMS, POSSIBLE SOLUTIONS AND
THE QUESTIONNAIRE
We experienced some of the same problems as last
year (see "New Interfaces - Old Models" Dokonal,
Knight, Dengg, 2015) together with some new ones.
One of the main problems we encountered last
year was the import and translation of the SketchUp
models into Unity3D. We were looking at the use of
Okinos Nugraf as a means of translating/preparing
the files, but this has been superseded by the in-
clusion of a direct SketchUp importer in Unity itself.
This meant we had far fewer problems in bringing
the geometry into the systems, both with Unity and
'Kubity' which also directly translated for Cardboard.
The need for a well disciplined and constructed 3D
model remains which some of the students found to
be a greater challenge than VR itself. Itself, but all
students quite quickly were able to experience their
models virtually.
The problems this year were not in the model
import, but in navigating around the models in VR.
This was partly hardware caused by the first gener-
ation Google Cardboard's unreliable magnet switch
and partly software when using Kubity. To move for-
ward inside "Kubity" you have to use a tab on the
screen of the smartphone - rather complicated with
the first versions of Google cardboard we had avail-
able for the workshop. To use the apps produced by
Unity3D we had to use the magnetic switch to move
inside the VR model. The magnetic switch turned out
to be very unreliable - with different Smartphones we
had different results. The original plan was that ev-
eryone should use her or his own phone. (Kubity is
also available for iOS)
It is worth restating that one of the main goals of
the exercise is that students can use the workflow we
developed during these workshops to use the VR app
as a design tool for their own projects in the future.
Having a lot of problems to move inside a VR model
is not very helpful in terms of immersion and leaves
the user with an experience of frustration rather than
immersion.
VIRTUAL REALITY - Volume 2 - eCAADe 34 | 575
 We had a great deal of discussions how to overcome
these kind of problems. One of the possible solu-
tions we called the mechanical finger because we dis-
cussed that the simplest way would be to have a me-
chanical finger which does the tapping on the screen
when you have the VR device on your head - ideally
connected to the feet. It was not possible during the
short workshop to get all the necessary parts for our
ideas to build it. The workaround for the limited time
available during the workshop was to use only the
smartphones that were working more reliably with
the magnetic switch.
The result of the questionnaire is not signifi-
cantly different to previous results and might also
have been influenced by the fact that we severely
struggled with navigation problems. Therefore we
could directly compare results with last year's work-
shop, the same questions were asked again; the re-
sults were almost identical (2015 figures in brackets):
• Viewing in a VR system helped to understand
the design better 95% (94%)
• Spatial perception was much clearer and
closer to reality than a 3D CAD model 96%
(94%)
• They could easily orientate themselves in the
flats 92% (90%)
Where the survey differed was in two areas. Firstly,
the question about the 'atmosphere' created was
much more positive. This is due to being able to cre-
ate much more realistically rendered models than the
'white card' materials that we used last year and the
need for more detail/materials was again reduced as
a requested improvement. However, the lack of re-
alism in the movement methods in Kubity (i.e. lack
of collision detection and gravity) meant that some
respondents felt that this affected the degree of im-
mersion in the experience.
576 | eCAADe 34 - VIRTUAL REALITY - Volume 2
Figure 2
Eezzee click
concept (with
apologies to Heath
Robinson).
EEZEE CLICK - THE "MECHANICAL FINGER"
The discussions during the workshop inspired the
thought that the Cardboard VR device would gain a
lot in terms of immersion (besides all those other as-
pects like level of detail etc) if a naturalistic navigation
methods could be devised - i.e. moving freely inside
the virtual world just by walking.
As we wanted to have it as a low cost / DIY ver-
sion within our 'workflow for everyone' all the sophis-
ticated systems using cameras and other tracking de-
vices were not an option. We wanted a solution avail-
able on a student's budget and within their capabil-
ity to construct. Moreover, we didn't want to limit
our walking range like the Oculus Rift (Tracking area
1.75m x 4m) and the HTC Vive (5m x 5m) and the
physical cage of CAVES, respectively. This is one of
the biggest advantages of our system.
Thankfully also Google seemed to have realised
that the magnetic switch was not reliable enough
and the second version of the Cardboard viewer re-
placed the magnetic switch with a simple kind of me-
chanical finger extension to make it possible to tap
on the screen with the device on.
Our mechanical finger only had to synchronise
feet movement to a tap on Cardboard and not on the
 smartphone itself which reduced the risk of the in-
famous Spiderweb App (broken screen) significantly.
The solution was to equip Cardboard with a solenoid
and connect it to switches on both feet. To keep it
simple we started with a wired version.
Figure 3 At the moment the solenoid is mounted to the
Eezzee click test set Google Cardboard with cardboard and Gaffa tape
on Google and the switches are pinned to the soles of both
Cardboard second shoes of the person using the device. A 12 Volt bat-
edition. tery pack produces the energy for the solenoid. The
circuit is wired in such a way that the solenoid is
pushing the cardboard tab whenever the person puts
both feet on the ground. That was the easiest way to
manage the movement without any additional elec-
tronics - there are other options in rewiring that we
will discuss later. One of the problems with that cir-
cuit is that if the person stand with both feet on the
ground the tab is pressed constantly. This is no prob-
lem for the Virtual world - you simply stop and stand
like in the real world. But it is a slight problem in the
real world because the solenoid is working constantly
and therefore overheating.
FIRST TESTS WITH THE "EEZEE CLICK"
The result of our mechanical finger is quite stunning
- it simply works. It makes a big difference to be able
to move in the virtual world by just walking in the real
world. But you should make sure that the real world
is big enough to move inside without banging your
head.
Still there are some problems in movement us-
ing the Kubity app. First of all there is the problem
of a lack of real gravity or physics. This means that
you cannot fix the eye level to a constant height that
equates to eye level. So if you look up or down dur-
ing walking you end up either flying up in the direc-
tion that you are looking or digging in the ground.
Whereas flying is sometimes quite useful it is not so
nice to end up underneath your model. Another is-
sue is that there is no calibration for the width of any
step inside the VR model when you use Kubity at the
moment. It worked quite well with all the models we
tested but at the moment we would have to rescale
the model and re-upload it to Kubity if we would
want to do any changes. This should be no problem
when we use Unity3D to produce cardboard apps out
of the geometry but this has yet not been tested fully.
OUTLOOK FOR "EEZEE CLICK"
We discussed several ideas what new possibilities our
enhanced device could have with different settings -
some of them are not really within architecture.
At the moment with our serial circuit we can only
walk not run. Whilst this is no problem to experience
a house or a flat in VR for normal purposes - normally
you do not run in a flat - it might be interesting for
special tasks inside houses. For example, if you would
want to test the design of an escape route in the Vir-
tual world you should be able to run. Nobody walks
away from a fire. Being a software for the gaming in-
dustry Unity3D offers many opportunities for fire and
explosions that you can script as events into the ge-
ometry. So this could be an opportunity to test the
ability of different groups to escape a certain geom-
etry in a certain time (But make sure that you have
enough space in the real world...)
Another situation for the use of running in the
Virtual Model would be the experience of a city
model like a jogger which also gives a different ex-
perience
For all these cases we would need to be able
to switch the circuit to parallel mode which should
VIRTUAL REALITY - Volume 2 - eCAADe 34 | 577
 also be possible within our low cost approach. This
switching of navigation modes is possible by creating
on onscreen interface that would allow the choice of
walk/run/Superman modes. This is planned for the
next version.
A DIFFERENT INTERFACE APPROACH -
XBOX CONTROLLER
The second interface method used a standard Mi-
crosoft Xbox controller. Whilst this removes the com-
mon problem of the user being able to move the
head though 360 degrees whilst using a static key-
board, it required modification to allow a natural
walking movement.
The controller is supported by the Unity Inter-
face controller and by Android smartphones, so it
was a simple matter to map two buttons to a script
which moved the camera one step per activation.
Two push buttons were installed on the heels of the
users' shoes, and the wires extended to the Xbox con-
troller. The Xbox controller was attached to the An-
droid smartphone with a USB OTG cable. If the user
wants to move by 'walking on the spot', this 'walking'
movement is achieved by alternatively raising and
lowing each heel in turn. Turning is a matter of swiv-
elling on the balls of the feet. It is also possible to
walk naturally, but a 'second' is required to prevent
accidents such as walking into physical objects.
COMPARISON - 'EEZZEE CLICK'(TM) AND
XBOX CONTROLLER
There is no doubt that the simplicity and almost low-
tech approach of 'eezzee click' enables anyone to
view their designs in a VR environment with a min-
imum of both time and financial investment. There
are currently limitations to Kubity, but this is to be
expected in a new application. Of more concern is
the data security issue of sending (and possible stor-
age) of potentially sensitive design data on a remote
server. However, as 'eezzee click' can also be used
with Unity (by creating a native Cardboard app) there
is also
The Xbox controller method allows for more flex-
578 | eCAADe 34 - VIRTUAL REALITY - Volume 2
ibility and adaptability that 'eezzee click'(TM), but is
more complex to initially setup. The advantage of us-
ing Unity is that it has a full physics engine that allows
more natural movement and interaction with the vir-
tual objects, but again, these require setting up
BACK TO THE OLD DAYS - THE EXERCISE
BIKE WITH HMD DEVICES
In working on this project it was also decided to re-
visit the early VR work, particularly the exercise bike.
Some of the problems encountered on this can now
be resolved because of advances both in the flexi-
bility and capability of the software and the lower
(or free) cost of professional quality game authoring.
One of the main problems with original bike was the
handlebar movement which was achieved by adapt-
ing an old Microsoft mouse using gear wheels de-
signed for radio controlled models. The problems
arose not in the physical side, but in a lack of flexibility
in configuring the software. Small movements were
fine, but anything more than 45degrees meant rotat-
ing the handlebars a full 360 degrees - not a natural
movement. With the flexibility of Unity3D, it is now
possible to correct these and, when coupled with the
use of a HMD, produce a much more natural and im-
mersive experience.
To use the exercise bike as an interface would
be especially be great for experiencing city models
by cycling through them in the Virtual environment
- and you don't have to be afraid to crash into a wall
in the real world. We are still working on that and will
be able to give you an update in August.
CONCLUSIONS
Although we still experienced several problems dur-
ing our workshop and the following experimental
sessions the potential of the use of especially the low-
est cost smartphone based HMD is very obvious to us.
Smartphones are ubiquitous today and the costs for
the HMD devices very low - sometimes they are given
away with the phone. Companies already see the
potential for clever presentations - see introduction.
Students took to the creation of VR and use of the VR
models with comparative ease. All produced models
and were able to see how it could be easily integrated
into the design workflow. Some tried this by revising
and re-visiting their models in the classic circular de-
sign>test>revise scenario. This was largely achieved
on their own laptops with their smartphones and it
was only when the Oculus Rift was used that a higher
end computer was required. Most of the participants
indicated that they would at least try this on their
next design project and a smaller number were very
enthusiastic.
Bringing together free or low cost software,
some carefully selected cheap components, some
ingenuity and an easy to follow workflow VR bring
added value in the design process. In our opinion VR
in the design process will be no longer a question, it
will be a reality soon - for students and for the aver-
age architectural firm, VR will be a valid tool in the
designers toolbox.
With thanks to: Darren Ward at RedRaven Design
(www.redraven.net) for permission to use the Long
House in Cockermouth for testing and Ben Devereau for
the 'eezzee click' concept sketch.
REFERENCES
Abdelhameed, W A 2014 'Creativity and VR Use, Rethink-
ing Comprehensive Design: Speculative Counter-
culture', Proceedings of the 19th International Con-
ference on Computer-Aided Architectural Design Re-
search in Asia (CAADRIA 2014), Kyoto, pp. 407-416
Dokonal, W, Knight, M W and Dengg, E A 2015 'New In-
terfaces - Old Models', Real Time - Proceedings of the
33rd eCAADe Conference - Volume 1, Vienna Univer-
sity of Technology, Vienna, Austria, pp. 101-106
Af Klercker, J and HenrichsÈn, J 2001 'Can simulations
in VE support architects in solving complex design
problems?', ACCOLADE - Architecture, Collaboration,
Design. Delft University Press (DUP Science), delft, pp.
77-82
Knight, M W and Brown, A G P 1999 'Working in
Virtual Environments through appropriate Physi-
cal Interfaces', Architectural Computing from Turing
to 2000 [eCAADe Conference Proceedings], Liverpool,
pp. 431-436
Knight, M W and Brown, A G P 2001 'Towards a natu-
ral and appropriate Architectural Virtual Reality: the
nAVRgate project', Past, present, future. Proceedings
of the Ninth International Conference on Computer
Aided Architectural Design Futures, Eindhoven, pp.
139-149
Krakhofer, S and Martin Kaftan, M 2015 'Augmented Re-
ality Design Decision Support Engine for the Early
Building Design Stage', Emerging Experience in Past,
Present and Future of Digital Architecture, Proceedings
of the 20th International Conference of the Association
for Computer-Aided Architectural Design Research in
Asia (CAADRIA 2015), Daegu, pp. 231-240
Kreutzberg, A 2014 'New Virtual Reality for Architec-
tural Investigations', Fusion - Proceedings of the 32nd
eCAADe Conference - Volume 2, Newcastle
Kreutzberg, A 2015 'Conveying Architectural Form and
Space with Virtual Reality,', Real Time - Proceedings of
the 33rd eCAADe Conference - Volume 1, Vienna Uni-
versity of Technology, Vienna, Austria, pp. 117-124
Miltiadis, C 2015 'Virtual Architecture in a Real-time, In-
teractive, Augmented Reality Environment - project
Anywhere and the potential of Architecture in the
age of the Virtual', .), Real Time - Proceedings of the
33rd eCAADe Conference - Volume 1,, Vienna Univer-
sity of Technology, Vienna, Austria, pp. 61-68
Mourel, A, Eloy, S, Dias, M S and Pedro, T 2013 'How
Space Experimentation Can Inform Design: Immer-
sive Virtual Reality as a Design Tool,', Proceedings of
the 17th Conference of the Iberoamerican Society of
Digital Graphics, Valparaíso, pp. 182-187
Shemesh, A, Bar, M and Yasha, J G 2015 'Space and Hu-
man Perception', Emerging Experience in Past, Present
and Future of Digital Architecture, Proceedings of the
20th International Conference of the Association for
Computer-Aided Architectural Design Research in Asia
(CAADRIA 2015), Daegu, pp. 541-550
VIRTUAL REALITY - Volume 2 - eCAADe 34 | 579
BIM | Concepts
Automated Generation of BIM Models
Patrick Janssen1 , Kian Wee Chen2 , Akshata Mohanty3
1,3
National University of Singapore 2 CENSAM, Singapore-MIT Alliance for Re-
search and Technology, Singapore
1
patrick@janssen.name 2 kianwee@smart.mit.edu
3
akshatamohanty@gmail.com

In early stages of architectural design, highly simplified minimal models are often
preferred while in the later stages maximal Building Information Models (BIM)
are required that include the relevant information for detailed design
documentation. This research focuses on the transition from minimal to maximal
models and proposes a semi-automated workflow that consist of two main steps:
analysis and templating. The analysis step starts with the minimal geometric
model and decorates this model with a set of semantic and topological attributes.
The templating step starts the decorated model and generates a transitional BIM
model which can then be readily altered and populated with high resolution
building information. A demonstration of two test cases shows the feasibility of
the approach.

Keywords: BIM, parametric modelling, interoperability

INTRODUCTION switch from conceptual modelling tools to BIM tools.

Architectural design typically relies on different tools
at different stages of the design process. In the
early stages, minimal models that are highly simpli-
fied are often preferred, for two reasons. First, min-
imal models are easier to build, thereby minimis-
ing the 'sunk costs' for options that get discarded.
Second, minimal models are much faster to analyse
and simulate, thereby allowing designers to receive
timely feedback. In the later stages, maximal models
are required that include all relevant information for
detailed design documentation. Typically, Building
Information Models (BIM) are created that progres-
sively increase the amount of information, including
information from other design consultants and spe-
cialists (Eastman, 2008).
In the transition from minimal to maximal mod-
els, there is a key point when designers need to
Before this point, planar entities such as walls and
floor slabs may be modelled as single polygons and
linear elements such as columns and beams may be
defined as polylines. Once the switch is made to BIM
tools, entities need to be more precisely defined as
building elements, requiring them to be thickened
and materials and other details to be defined. We re-
fer to this as the materialisation process.
Materialisation Process
In this research, we focus on materialisation pro-
cesses where the tools used for conceptual mod-
elling differ from those used for BIM modelling. There
are cases where the conceptual modelling is per-
formed within the BIM tool, which then simplifies to
some extent the materialisation process. However,
designers often prefer to use different tools. Com-

BIM | Concepts - Volume 2 - eCAADe 34 | 583

 monly used tools include Trimble SketchUp and Mc-
Neel Rhino for conceptual design, and Graphisoft
ArchiCAD and Autodesk Revit for BIM modelling.
This research aims to explore the feasibility of
semi-automatically converting minimal geometric
models into transitional BIM models consisting of ex-
plicitly defined building elements, but with only lim-
ited building information. Although these models
may not yet be maximal in their information content,
their usefulness lies in the fact that they can be di-
rectly imported into existing BIM tools, where they
can be easily modified and further developed.
This materialisation process is currently difficult
to automate. For buildings designs with high levels of
repetition, this may not be an issue as the time taken
to manually remodel the design in the BIM tool is low.
However, if the design includes any more complex
non-repetitive configurations, then the remodelling
becomes time-consuming and error prone.
The aim of semi-automating this materialisation
process is to enable more fluid and flexible work-
flows that do not require BIM models to be manu-
ally recreated. However, this does not require the
whole modelling process to be fully automated. In-
stead, workflows need to be developed that min-
imise the effort required to remodel the design. As
a test case, the research will focus on a materialisa-
tion process in which the geometric models are im-
ported as DXF models, and the transitional BIM mod-
els are exported as Industry Foundation Classes (IFC)
models. The DXF file format is a commonly used data-
format for representing 3D geometry. The IFC file for-
mat (ISO 16739:2013) is an information-rich object-
based file format for representing building informa-
tion. The use of an open standardised file format en-
sures that the approach remains workflow agnostic,
allowing users to link together diverse tools and sys-
tems to support various forms of collaboration and
exchange.
One of the key requirement relates to the rep-
resentations used to model IFC elements. When
boundary representations are used, it becomes very
difficult to further modify and developed the model
584 | eCAADe 34 - BIM | Concepts - Volume 2
using BIM tools. In contrast, when parametric rep-
resentations are used, the process of importing IFC
models into BIM tools generates native BIM elements
that retain their intelligent behaviour are therefore
easy to modify. The aim is therefore to use parametric
representations wherever possible, and to only revert
to boundary representations where it is absolutely
necessary.
Existing Approaches
Within the existing ecosystem of modelling tools and
plugins, there are a number of approaches that may
address some of challenges associated with the ma-
terialisation process. Three main approaches will be
briefly reviewed: geometric modelling tools that sup-
port IFC export, BIM modelling tools that support
geometric import, and graph-based parametric BIM
modelling systems.
Perhaps the most straightforward approach is to
use a geometric modeller that is capable of export-
ing IFC. One well know example is SketchUp, the Pro
version of which now supports IFC export. Using this
approach, the user simply models the building ele-
ments as objects in SketchUp, and then assigns an IFC
type to each of these objects. The downside with this
type of approach is twofold. First, the full 3D geome-
try must be modelled, and as result there is not really
any advantage over modelling it directly in the BIM
tool. Second, the resulting IFC elements are gener-
ated using boundary representations, which means
that they can no longer be easily modified once im-
ported into BIM tools.
Another approach is to import the geometric
model directly into a BIM tool, and then convert the
geometric entities to BIM elements. An example of
this is Revit, which allows geometric solids to be im-
ported as massing models into Revit. The faces of
the conceptual mass can then be selected and con-
verted to selected BIM elements such as walls and
slabs. The downside is that most BIM elements can-
not be created in this way. For example, openings
in the faces of solids can be imported, but cannot
be subsequently converted to Revit elements such as
doors and windows.
A third approach is to consider using existing
parametric modelling tools as a platform for defin-
ing and customising the rules for the materialisation
process. Graph-based parametric modelling systems
such as McNeel Grasshopper [1], Bentley Generative-
Components [2], Autodesk Dynamo [3], and Sidefx
Houdini [4] could be used. A number of workflows
and plugins have been developed for generating BIM
models using such parametric modelling tools. In
general, there are two approaches: tightly coupled
and loosely coupled.
With the tightly coupled approach, systems are
coupled through the Application Programming Inter-
face (API) provided by the BIM system. In this case,
graph-based systems communicate via the API of the
BIM system, directly instantiating geometry in the
BIM model each time the graph-based model is ex-
ecuted. There are numerous examples of this ap-
proach. Grasshopper has plugins for connecting to
both ArchiCAD (ArchiCAD Connection [5]) and Revit
(Hummingbird/Whitefeet [6]). Dynamo has a built-in
connection with Revit, and GenerativeComponents
has a built-in connection with AecoSIM [7]. The
downside of all these solutions is that they only work
with specific BIM tools and are therefore not work-
flow agnostic.
With the loosely coupled approach, systems are
coupled through model exchange. The graph-based
system typically generates an IFC model that can be
directly imported into the BIM system. An example
of this approach is the GeometryGym [8] plugin for
Grasshopper. In theory, it is possible for designers
to create their own customised materialisation pro-
cedures using a plugin like GeometryGym. However,
the downside would be the high complexity involved
in doing so. This is primarily due to the fact that
graph-based parametric modelling tools are not well
suited to creating such rule-based procedures. In
particular, the control flow of these types of proce-
dures is typically complex, with a heavy reliance on
conditional and looping constructs, for which graph-
based systems are not well suited.
Paper Overview
This research proposes a generalised system for semi-
automating the materialisation process. The paper
consists of three main parts, with the first part de-
scribing a proposed system, the second part focusing
on a set of experiments, and the third part discussing
future challenge.
PROPOSED SYSTEM
A system for the semi-automated generation of tran-
sitional IFC models from geometric DXF models is
proposed. The proposed system focuses on typical
building elements such as slabs, roofs, walls, doors
and windows. These types of elements have the ad-
vantage that they can be defined using the paramet-
ric representations within IFC. In the proposed sys-
tem, the materialisation process is divided into two
parts: the analysis step and the templating step.
The analysis step starts with the geometric DXF
model and generates a new model that is enriched
with a set of semantic and topological attributes, re-
ferred to as the decorated model. The analysis rules
infer these attributes from the size, orientation, and
relationships between geometric entities in the DXF
model. Other than adding attributes, no changes are
made to the geometry by the analysis rules.
The templating step starts with the decorated
model and generates the transitional IFC model.
Each of the geometric entities in the decorated
model are mapped to parametrically defined objects
in the transitional IFC model. The template rules are
matched against geometric entities in the DXF model
based on attributes created in the analysis step. If a
rule matches, the IFC element specified in the rule is
added to the IFC model.
In general, the system is envisaged as a web-
based application that can be used to materialise
models. Figure 1 shows a materialisation process
consisting of four main steps: 1) upload the geomet-
ric DXF model; 2) execute the analysis rules; 3) cus-
tomise execute the templating rules; and 4) down-
load the transitional IFC model.
Step 3 is important as it allows the materialisa-
BIM | Concepts - Volume 2 - eCAADe 34 | 585

tion process to be customised to the needs of the de-
signer. The customised rules can take into account a
variety of factors, including the scale of the project,
the architectural design language, and the type of
construction. Through an iterative process of tweak-
ing rules, designers will over time be able to develop
libraries of personalised rules.
In step 4, the transitional IFC model is down-
loaded and imported into the BIM application, where
the IFC elements can be replaced with more complex
native BIM objects. There is therefore no expectation
that this materialisation process will create the whole
model. It is more likely that the process creates a BIM
model that can be readily populated with higher res-
olution information of the building project.
Geometric DXF Model to Decorated Model
The DXF model is assumed to contain highly simpli-
fied representations of a building following a set of
modelling conventions, using only planar polygons
and lines. Building elements such as walls, slabs,
doors and windows are represented using polygons,
while building elements such as columns and beams
are represented using lines. The analysis step first
assigns unique IDs to the entities in the model, and
then analyses the individual entities and their topo-
logical relationships.
For lines, attributes are created to store the
length and the direction vector. For polygons, at-
586 | eCAADe 34 - BIM | Concepts - Volume 2
Figure 1
Proposed system
for
semi-automating
the materialisation
process.
tributes are created to store the number of ver-
tices, perimeter, area, and normal vector. For poly-
gons, specific shapes are recognised, and a series
of Boolean attributes are created. These include is-
convex, is-rectangular and is-stable. A stable polygon
is defined as one whose lowest edge is horizontal.
Lines and polygons are also categorised according to
their angle of inclination, into one of four categories:
is-horizontal (0°), is-sloping (1 to 45°), is-leaning (46 to
89°), and is-vertical (90°).
The relationships between lines and polygons
are analysed for a given tolerance. For each type
of relationship, attributes are created with the at-
tribute value being a list of IDs of the entities in-
volved in the relationship. Contact relationships are
any lines or polygons that touch each other in some
way. Attributes are created for various types of con-
tact between points, edges, and faces. For poly-
gons, two additional types of relationships are co-
planarity and containment. Co-planarity refers to all
polygons that are co-planar to a given polygon, with
a single co-planar attribute being created for each
polygon. Containment refers to a relationship be-
tween co-planar polygons, where the interior of one
polygon completely contains another polygon. At-
tributes are created for both contained-by and con-
tains.
Decorated Model to Transitional IFC Model
The templating step generates IFC elements from the
entities in the decorated model. This step starts by
creating the IFC spatial structure. The first three lev-
els of this spatial structure are predefined, and consist
of the IFCProject, IFCSite and IFCBuilding. Under the
building level, a number of IFCBuildingStories need
to be created. These building stories will contain all
the IFC elements that belong to that storey, including
slabs, walls, windows, columns and beams.
The templating step automatically groups the
geometric entities in the decorated model accord-
ing to building storeys they belong to. First, floors
are detected by identifying all horizontal polygons
with upward pointing normal. These floors are then
grouped into co-planar sets, and sorted according
to their z-heights in ascending order. These groups
then represent the building stories. The remaining
geometric entities in the model are then added to
these groups according to their topological relation-
ship to the floor polygons.
The geometric entities in each building storey
group are then processed, starting with the lowest
storey. For each group, an IFCBuildingStorey is cre-
ated and the geometric entities in the group are then
mapped to IFC elements using templating rules. The
templating rules have two parts: a pattern and a tem-
plate. The pattern part of the rule specifies the type
of geometric entity (either line or polygon) and a set
of attributes that the entity must have. The template
part of the rule specifies one or more IFC elements to
be instantiated.
The geometric representation of the IFC element
is generated based on the shape, position, and orien-
tation of the polygon or line. For example, an IFC wall
[9] can be represented using a number of parametric
representations. The Swept Solid representation of a
wall defines the solid wall object by sweeping a pla-
nar profile using either linear extrusion or revolution
techniques. The clipping representation defines the
solid wall object as a result of series of Boolean oper-
ations in a Constructive Solid Geometry tree. Finally,
the IFCWallStandardCase wall type is a parametric ob-
ject that defines a wall by the centre line and param-
eters that specify the width and height.
In addition, a number of settings are applied to
the IFC element. For example, for walls, the settings
include the thickness and materials of the wall, as
well as the position of the wall relative to the poly-
gon. (For example, the polygon may represent either
the wall centreline, the inner face or the outer face.)
As an illustrative example, a simplified version
of the templating rules for generating IFC walls are
shown below. In this case, the rules are designed to
identify polygons that are walls. Such polygons must
be stable, but the shape may vary, and it may be ver-
tical or leaning. These attributes will affect the rep-
resentation that is used. In addition, the 'contained-
by' attribute is used to distinguish between walls and
openings (such as doors and windows). For walls, the
'contained-by' attribute should be empty, while for
openings, it should contain an ID of the polygon in
which the opening is to be created.
• IF polygon is-stable, is-vertical, is-rectangular
and contained-by is empty THEN IFCWallStan-
dardCase, Representation=Curve2D.
• IF polygon is-stable, is-vertical and contained-
by is empty THEN IFCWallStandardCase, Rep-
resentation=Clipped.
• IF polygon is-stable, is-leaning, is-rectangular
and contained-by is empty THEN IFCWall, Rep-
resentation=SweptSolid.
• IF polygon is-stable, is-leaning and contained-
by is empty THEN IFCWall, Representa-
tion=Clipped.
Similar rules exist or generate other types of el-
ements. For example, rules for doors and win-
dows have patterns that match polygons where the
contained-by attribute is not empty. If a polygon is
contained by a wall and its lowest horizontal edge is
touching the lowest horizontal edge of the wall, it is
categorised as a door. Otherwise, if it is not touch-
ing the lowest horizontal edge of the wall, then it is
categorised as a window.
Default templating rules may initially be used,
BIM | Concepts - Volume 2 - eCAADe 34 | 587
 but designers need to be able to customise these Figure 2
rules to suit their needs. Designers can make either (a) House
globalised or localised customisations. Globalised conceptual model
customisations include general modifications to the in Rhinoceros3D (b)
default setting. Localised customisations only apply House IFC model in
to certain subsets of elements, defined through the ArchiCAD (c)
use of layers in the DXF file. For example, the designer Section generated
may want to have a variety of different types of walls, from the IFC model
with different settings. In such a case, the wall poly- in ArchiCAD.
gons can be placed on different layers, and templat-
ing rules can then be created that include the layer
name in the pattern. This then ensures that these
rules will only apply to polygons on those layers.

EXPERIMENTS
The feasibility of the proposed system is demon-
strated through a number of experiments in which
the creation of a set of analysis and template rules
were tested. These experiments used Rhino for the
conceptual modelling, Houdini for defining and exe-
cuting the rules, and ArchiCAD for importing the IFC
BIM model.
In these demonstrations, Houdini allowed for a
fast way of prototyping and testing rules. For gener-
ating IFC models, an exporter was implemented for
Houdini using the IfcOpenShell Python library. This
allows the exported IFC model from Houdini to be im-
ported into ArchiCAD, where plans and sections can
be generated.

House Example
In order to illustrate how these different rules may op-
erate, a simple example may be considered, as shown
in Figure 2. The example is based on the IfcOpen-
House [10], where an extra level is added to increase
the complexity. The example consists of two floors,
six windows; three on each level, one door on the first
level and a roof at the top level.
In the analysis step, the DXF model imported
from Rhino is analysed and a decorated model is gen-
erated with additional attributes. For this example,
there are no columns and beams, and as a result the
model only contains polygons.
In the templating step, the decorated model is con-
verted into a transitional IFC model. This step starts
by identifying all the floors as horizontal polygons
with upward pointing normal, of which there are two.
Since these floors are not co-planar, two separate
IFCBuildingStories are generated: a ground storey and
a first storey. The remaining entities in the geometric
model are then categories into the different building

588 | eCAADe 34 - BIM | Concepts - Volume 2

storeys, depending on the height of the polygons rel-
ative to the floors. This results in 8 polygons on the
ground floor (4 walls, 3 windows, and a door) and 9
polygons on the first floor (4 walls, 3 windows, and 2
roofs).
The templating rules are then applied to each
storey in turn, thereby generating the various IFC el-
ements for each floor. In most cases, the automated
mapping of the geometric entities to IFC elements
is straightforward. Walls are mapped to IFCWallStan-
dardCase, with a thickness of 200mm. Floors are
mapped to IFCSlab of type Floor, and roofs to IFC-
Slab of type Roof , both with a thickness of 300mm.
Windows and doors are mapped to IFCWindows and
IFCDoors respectively, with a thickness of 50mm. For
the windows and doors, an IFCOpeningElement also
needs to be added to the containing wall to define
an opening, with a thickness that is equal to that of
the wall.
The walls are slightly more complex due to the
fact that different representation are required for the
ground floor and first floor. On the ground floor, the
walls are constant height, so an IFCWallStandardCase
with a Curve2D representation can be used. In this
case, the wall height is determined by the distance
between the two floors. On the first floor, the walls
need to be connected to the sloping roof, and as a
result the walls vary in height. An IFCWallStandard-
Case with Clipped representation is therefore used.
Figure 2b shows the transitional IFC model imported
into ArchiCAD and Figure 2c shows a resolved section
generated from the IFC model.
Complex Tower Example
A similar procedure as the one used for the House ex-
ample is applied on a complex tower as shown in Fig-
ure 3. First, a minimal geometric model of a multi-
storey building was created using Rhinoceros3D.
In order to create additional complexity, a twist-
ing tower is generated by turning the floor plates
on each level (Figure 3a). In addition, columns and
beams are also added, modelled as vertical and hor-
izontal lines respectively. The default profile for the
columns is a 500mm diameter circle, while for the
beam it is a 500mmx500mm square.
Figure 3 shows how the minimal model can be
quickly translated into a transitional IFC model. Once
the IFC model is imported into a BIM application, it
can be readily altered and populated with higher res-
olution data, and used to generate a variety of plans
and sections for further development.
FUTURE WORK
The two test cases show the feasibility of a semi-
automated workflow for generating BIM models from
minimal conceptual models. However, there are
clearly many limitations that require further research
and exploration. The aim of this approach is to be
able to automate the generation of BIM models for
standard types of building elements and configu-
rations, thereby allowing architects and designers
to focus on the more complex parts of the mod-
elling. However, even the standard building ele-
ments present numerous challenges.
As an example, the issue of variable wall thick-
nesses may be considered. Figure 4 shows a plan
view of a wall intersection, with different wall thick-
nesses. In (a), the walls in the geometric model are
aligned, the designer may want the same alignment
to be maintained in the BIM model, so that the walls
remain flush. However, if the polygons in the geo-
metric model are assumed to represent the wall cen-
tre plane, then this will result in a misalignment, as
shown in (b) and (c). In order for the walls to end up
flush, the designer would have to misalign them in
the geometric model, as shown in (d). Such an ap-
proach would clearly be highly unintuitive.
An alternative approach would be to develop
more intelligent rules that could infer such condi-
tions. For example, the rule that generates the thin-
ner wall may be able to detect that by shifting the
wall slightly to the right, this misalignment can be
avoided. However, any such 'intelligent' behaviours
will also need to have overrides, just in case there is a
situation where a designer actually intends the walls
not to be flush.
BIM | Concepts - Volume 2 - eCAADe 34 | 589
 Figure 3
Complex tower
example. (a) The
minimal geometric
model in Rhino. (b
& c) The IFC model
imported into
ArchiCAD model.
(d) A section
generated from the
ArchiCAD model.

Figure 4
Misalignment of
walls connections
and intersections
due to varying wall
thickness. (a, b & c)
Polygon alignment
in the geometric
model result in a
misalignment of
the walls in the BIM
The challenge is therefore how to keep the process ledge, Oxon model. (d, e, &f )
as a whole simple and intuitive for the designer. This [1] http://www.grasshopper3d.com/ Polygon
requires a balance between, on the one hand, more [2] https://www.bentley.com/en/products/product-lin misalignment in the
e/modeling-and-visualization-software/generat
advanced rules with intelligent inference, and on the geometric model
ivecomponents
other hand, intuitive and simple ways for the de- [3] http://www.autodesk.com/products/dynamo- results in a
signer to override this intelligence. Future research studio/ alignment of the
will explore how such a balance can be achieved. overview walls in the BIM
[4] https://www.sidefx.com/ model.
[5] http://www.graphisoft.com/archicad/rhino-grassh
REFERENCES opper/
Eastman, C 2008, BIM Handbook: A Guide to Building In- [6] http://www.grasshopper3d.com/group/hummingbird
formation Modeling for Owners, Managers, Designers, [7] https://www.bentley.com/en/products/brands/aeco
Engineers and Contractors, Wiley sim
Janssen, P and Stouffs, R 2015 'Types of Parametric Mod- [8] https://geometrygym.wordpress.com/
elling', Proceedings of the 20th International Confer- [9] http://www.buildingsmart-tech.org/ifc/IFC4/fina
ence of the Association of Computer-Aided Architec- l/html/
tural Design Research in Asia CAADRIA, Hong Kong, [10] http://blog.ifcopenshell.org/2012/11/say-hi-to-
pp. 157-166 ifcopenhouse.html
Patrick, J 2014 'Visual Dataflow Modelling: Some
thoughts on complexity', Proceedings of the 32nd
eCAADe Conference, Newcastle, UK, pp. 547-556
Woodbury, R 2010, Elements of Parametric Design, Rout-

590 | eCAADe 34 - BIM | Concepts - Volume 2

Information and Construction: Advanced Applications of
Digital Prototyping in the Housing Industry
Basem Eid Mohamed1 , Frederic Gemme2 , Aaron Sprecher3
1
Abu Dhabi University 2 BONE Structure 3 McGill University
1
basem.mohamed@adu.ac.ae 2 fgemme@bonestructure.ca
3
aaron.sprecher@mcgill.ca

This study elaborates on recent efforts in applying Digital Prototyping strategies

to realize a prefabricated construction system that allows for significant
flexibility and adaptability in housing design. The rational of the described model
is based on combining advanced BIM modeling with structural analysis, towards
achieving high accuracy in the design phase, leading to subsequent precision in
fabrication and assembly of a specific building system; the BONE Structure. Such
an application aims at delivering significant levels of detailing in design and
production of the system's components, thus supporting the intention of
pre-defined assembly on jobsites, leveraging quality, and reducing waste. The
paper represents a phase from a continuous research endeavor that aims at
exploring technological enablers for mass customization in the housing realm,
based on advanced levels of digitization of the design and production processes.

Keywords: Housing, Prefabrication, Digital Prototyping

INTRODUCTION design flexibility; one that is considered vital in the

Contemporary advancements in digital design tools,
coupled with new protocols of digitized manufac-
turing processes have offered the Architecture, Engi-
neering, and Construction (AEC) industry the means
to allow for higher control over the design, fabrica-
tion, and construction processes. Through the use of
advanced design and modeling software, designers
and engineers are continuously developing abilities
to establish parametric relationship between parts,
analyze structural performance, and simulate com-
ponents production and assembly, towards a more
efficient and precise model for construction. While
these technologies have allowed for significant time
and cost control throughout design and construc-
tion processes, another crucial factor is allowing for
housing realm.
Schneider and Till (2005) defined flexible hous-
ing as buildings that can adapt to the changing needs
of occupants, comprising different possibilities of
pre-construction layout selection, as well as the abil-
ity to change one's housing over time. Addition-
ally, it includes the capability of integrating technol-
ogy throughout occupation in response to change
in demographics, or adaptive re-use. As a result of
wide modes of applicability, there are numerous ap-
proaches for achieving flexibility.
Within the context of this research, we ex-
plore how managing and manipulating information
throughout the design, fabrication, and assembly
of the BONE Structure system can lead to design

BIM | Concepts - Volume 2 - eCAADe 34 | 591

 flexibility on multiple levels. The first level exam-
ines flexibility at initial design stages, with focus on
enabling participatory design in the form of pre-
construction choices to homebuyers. The second
level is denoted with post-occupancy adaptability,
aiming at responding to prospective changes in oc-
cupants' needs over time. The framework for both
levels stems from the structural system and its de-
sign model, supported by advanced applications of
Digital Prototyping strategies. On the one hand, the
physical characteristics of the system's components
and their configuration establish the basis to analyze
the capacity of housing designs to accommodate
variations requested by homebuyers. On the other
hand, Digital Prototyping empowers high precision
in visualization, fabrication, and assembly of build-
ing components, thus enables implementation of de-
sign variations efficiently. The aim of such a frame-
work is to respond more effectively to wider sector of
customers and market demands, given the variation
in socio- demographic patterns, thus expand market
shares.
THE STRUCTURAL SYSTEM: BONE STRUC-
TURE
The BONE Structure is a proprietary construction sys-
tem that combines cold-formed steel components
and insulating materials in an integrated manner.
The steel components are all designed to be assem-
bled using solely screws and bolts, eliminating the
need for cutting, piercing and welding on jobsite.
As a consequence, the amount of residual matter
resulting from the construction process is reduced
significantly. The system differentiates itself from
traditional steel construction system by combining
the advantages of a post-and-beam structure; larger
spans, size of openings, the integration of the struc-
tural system with the thermal envelope, and the pre-
cision of assembly provided by pre-manufactured
components.
Pertaining to form, the BONE Structure is an or-
thogonal system that has been designed using a
5 feet incremental module in the X-Y plane and 6
592 | eCAADe 34 - BIM | Concepts - Volume 2
inches module in the Z direction, leading high level
of standardization of components. Given that the
system is always undergoing research and develop-
ment, the latest version of the system (10.0) includes
the half-grid feature, providing more flexibility to the
architectural design.
All components have an overall length of max-
imum ten feet in order to facilitate procurement.
Structural columns have a composite square profile,
4 inches x 4 inches, assembled in plant employing
self-tapping screws. An additional H-shaped profile
can be added on the exterior side of columns to en-
hance their resistance to lateral loads. Stronger mem-
bers are designed using a thicker material whenever
required. Figure 1 represents views of the structural
system components.
Figure 1
An overview of the
structural system.
Floor and roof structure main components are vari-
able lengths of a seventeen inches deep profile man-
ufactured using eleven gauge galvanized steel, the
maximum standard span being 25 feet. These pro-
files are characterized by large openings that are per-
formed to remove some weight to the components
and to allow for better efficiency for trades such as
electricity, HVAC and plumbing. Stronger members
are designed using a thickness of 0.1875 inch when
required. Secondary joists are put in the transverse
direction between the joists with a typical spacing of
20 inches.
Once land preparation is completed, and the
foundation is poured and cured, steel components
and fasteners are delivered. The prefabricated struc-
tural components are anchored to the foundation by
using cast-in anchors available in three standardized
dimensions. The anchors are positioned using spac-
ing templates, eliminating the use of a measuring
tape. Assembly instructions are obtained from digital
models, thus capitalizing on the efficiency of building
processes.
Floor structure is completed by adding a 0.75
in thick thirteen plies plywood panel over the steel
components. These panels are pre-pierced and pre-
cut using CNC (Computer Numerical Control) ma-
chining centres. The assembly is performed using
self-tapping screws as holes in plywood panels are
aligned with holes in the steel structure underneath.
The plywood panels are supported on all edges and
define a structural diaphragm. On the roof, the di-
aphragm is achieved by a Structural Insulated Panel
(SIP) fastened to the structure by screws installed
from the interior side. These SIPs are composed of
two layers of Oriented Strand Board (OSB) with an Ex-
pensed Polystyrene (EPS) layer in between. The typ-
ical thickness of EPS is 10.5 inches, giving an insulat-
ing rating of R-40. In some cases, additional structural
components such as structural lintels above sliding
doors and garage doors, overhangs, balconies, ter-
race structures and canopies are then added. Open-
ings for doors and windows are pre-framed using
pre-cut plywood parts with machined holes and
grooves.
The subsequent steps are denoted with cover-
ing the complete wall area with 3 inches thick EPS
panels. These panels are cut in three configurations
to fit exactly in between the structural components.
They are held in place by z shaped profiles. The whole
wall is finally covered with a 2.5 inches thick layer of
polyurethane sprayed foam. The depth of the z pro-
files is calibrated to exceed the polyurethane and re-
ceive the exterior finish. A nylon thermal break posi-
tioned between the z shaped profile and the struc-
tural columns prevents metallic continuity through
the thermal shell. On the interior, wall and ceilings
are covered with galvanized steel furring profiles that
are clipped to the structure, making it ready for dry-
wall installation. Exposed components such as archi-
tectural stair structure, exposed cross bracings, ter-
race structure and sunshades are typically installed
when spray foam application is completed. The shell
is then completed by installing doors and windows
and sealing the perimeters of these components.
The described system model combines innova-
tive structural design, with digital practices in design
and fabrication towards exploring new territories in
the housing industry. The process proposes a new
logic for modularity and standardization of compo-
nents to allow for various configurations. Accroding
to Salingaros and Tejada ( 2001), modularization of
design empowers product configuration by simulta-
neously taking advantage of the economies of scale
and scope. With the advent of digital design and fab-
rication technologies, modularity in design has re-
cently gained new ground by offering high degrees
of flexibility in design, especially through the possi-
bility of numerous assemblies of prefabricated com-
ponents, as well as by allowing for material use op-
timization. The presented building system extends
its flexibility based on a robust modular approach, in
addition to specific physical characteristics of its com-
ponents. The aim is to allow for different configura-
tion scenarios in response to users' demand for inter-
nal and external design variations.
DIGITAL PROTOTYPING IN DESIGN, FABRI-
CATION, AND ASSEMBLY
One of the emerging methods of computer mod-
elling is Digital Prototyping; defined as the mean by
which engineers, and designers can explore products
virtually before being built. It is a process that al-
lows for designers to visualize, validate, simulate, and
optimize, products data throughout the product de-
velopment process within an advanced a digital en-
vironment. Such a process is structured on differ-
ent levels of automation, depending on the nature of
the product, and operations involved in the process
(Bullinger et al. 2000).
Digital Prototyping combines Computer Aided
Design (CAD) technologies with Virtual Reality (VR)
to allow producing prototypes more efficiently, thus
optimizing the product development process. One
of the strategic advantages of Digital Prototyping is
real-time decision making at early stages of product
development. This enhances the process and makes
BIM | Concepts - Volume 2 - eCAADe 34 | 593
 it more efficient, leading to allowing for earlier modi-
fications, and optimization of the prototype. Further-
more, virtual prototypes enable qualitative evalua-
tion of product qualities, thus eliminating errors that
might occur within the fabrication process (Kanai et
al. 2009).
Bullinger et al. (2000) defined a crucial compo-
nent of Digital Prototyping that is the Digital Mock-
Up (DMU); a purely digital test model of a techni-
cal product. Such a model enables for a current and
consistent availability of multiple view of the prod-
uct shape, function, and technological coherences.
This constitutes the basis on which modeling and
simulation performed and communicated for an im-
proved configuration of the product. Commonly, vir-
tual products are employed as a reference for test-
ing the design regarding its feasibility, functionality,
and efficiency prior to production of physical pro-
totypes. This is done by building a comprehensive
model that brings 3D data into a single 3D model.
This is done either for exclusive products, or ones
integrated within a system. In such a virtual envi-
ronment, possible defects can be detected and cor-
rected in design before building physical products.
The model also enables unprecedented precision in
establishing the link between design and manufac-
turing, thus improving productivity.
Applications of Digital Prototyping has been suc-
cessful in various domains, including aircraft con-
struction, shipbuilding, and the automotive indus-
try. Also, various software platforms have been em-
ployed, including for instance, Autodesk Inventor,
Solidworks, and CATIA by Dassault. Nevertheless,
the selection criteria of the platform relates to the
capability of devising a workflow that is compatible
with the nature of the product, level of technology
involved, and users participating in the product de-
velopment process (Jelena, and Maver 2003). We
present in this research a digital workflow model that
complements the nature of the developed building
system with focus on realizing a set of goals mainly
denoted with design flexibility, precision in fabrica-
tion and assembly, towards improving quality and
594 | eCAADe 34 - BIM | Concepts - Volume 2
performance. These goals are realized through an in-
tegrated approach to Digital prototyping that stems
from implementing a comprehensive environment
of various software tools and platforms. the follow-
ing sections will elaborate on the workflow structure.
The BIM Protocol
Tools, platforms, and environments constitute the
three levels of Building Information Modeling (BIM)
applications. While BIM tools are task-specific appli-
cations that produce a specific outcome, including
model generation, drawing production, scheduling,
and cost estimation, BIM platforms are design appli-
cations that offer data to multiple users. They pro-
vide primary data-models that manage the informa-
tion on the platform. Finally, BIM environments al-
low for the management of one or more data streams
that integrate BIM tools and platforms within an or-
ganization. They support interoperability within a
system, or the development of an efficient workflow
through the exchange of data between applications,
thus facilitating its automation. It eliminates the du-
plication of structural, electrical, and mechanical data
previously generated by other applications (Eastman
et al. 2008).
The design process of BONE Structure system is
focused on implementing an effective data sharing
and management environment to allow for collabo-
ration between various team members. The process
is structured as a series of operations that functions
either simultaneously, or consecutively depending
on the nature of data input/output, thus bringing
performance to the prefabricated housing industry
by depending on advanced digital modeling. The ef-
fective implementation of an effective BIM environ-
ment tends to increase productivity in all phases from
conceptual design to on site assembly of the build-
ing system components. The digital process follows
a robust structure starting with modeling, analysis,
validation, and leading to a comprehensive digital
model.
Commonly, the design is modeled on Autodesk
Revit for visualization, and validation of data at the
early conceptual stage, based on a set of system
guidelines. This also allows for quality management
to transfer precise information throughout the work-
flow. Such a process employs a set of pre-designed
elements stored within a database in the form of
component families, thus becomes a series of virtual
assembly tasks. Once various tasks are completed ,
the model is then used for referencing purposes in
relation to contextual data. This allows for envision-
ing each part of the puzzle towards maximizing the
efficiency of material orders and reduce waste. Ad-
ditionally, REVIT worksets facilitate internal collabo-
ration with other team members and centralize data
management.
The organization of the modeling environment
is informed by the standardization and modularity
of BONE Structure's technology, where the BIM fami-
lies database comprises detailed models of structural
columns and joists, in addition to walls, floor and roof
panels, openings, stairs, exterior canopies and bal-
conies. These scomponents have been parametri-
cally designed in order to support design flexibility
while keeping configurations within a feasible range.
For instance, window and door families can be easily
configured as per detail and profiles provided by dif-
ferent suppliers, whereas implementing a set of au-
tomated functions would allow for new window or
door models to be developed by simply changing the
profile sweep. Accordingly, this level of BIM applica-
tion aims at achieving the following:
• Develop, and validate the architectural de-
sign;
• Export the structural components for ad-
vanced structural analysis;
• Provide referential information for digital pro-
totyping of individual components.
Further to the functions mentioned above, a crucial
operation of the BIM strategy is denoted with permit
application, whereas the issuing of plans is optimized
through the use of a set of Revit's automated func-
tions. These functions generate code notes based on
individual families that assigned to a specific project
depending on its location, thus reduce the amount
of writing in plans. Additional texts and notes are
uploaded to the project and can be added as a tag.
In a similar fashion, predefined tables enables au-
tomating the calculations of unprotected openings,
lot coverage, gross floor area, setbacks and other in-
formation needed for a the permit application.
The intention of the described BIM environment
goes beyond the norms of visualization to commu-
nicate with clients, and issuance of drawings, to be-
coming an important fragment of the integrated Dig-
ital Prototyping strategy. It represents the first level
of creating a digital model that is completely free
of any technical conflicts. The structure of such an
environment allows technical components to be up-
dated in real-time in accordance with design deci-
sions, based on developing system specific paramet-
ric families that are informed by BONE Structure sys-
tem design guidelines. Furthermore, the automating
part of the the modeling operations aims at improv-
ing quality, and speeding up the modeling process.
Automation is achieved using a set of algorithms that
contributes to the course of placing structural com-
ponents such as columns and beams, openings, con-
nections, roof/floor panels, in addition to code notes.
Establishing an integrated workflow
The application of BIM in the construction industry
has gained considerable momentum over the past
few years. Companies have been developing suc-
cessful implementation strategies in order to remain
competitive, while improve production quality, and
methods of communication. However, there are al-
ways two key issues that require greater focus and ex-
ploration: workflow and interoperability. These top-
ics become more complex at the point where collab-
oration between different software tools is required
for the sake of work sharing and data transfer, es-
pecially disciplines overlap such as architecture, me-
chanical and structural design (Smith 2014).
The distinctive approach of BONE Structure sys-
tem's implementation model to Digital Prototyping
stems from the notion of developing a collaborative
BIM | Concepts - Volume 2 - eCAADe 34 | 595
 workflow that allows for effective data transfer be-
tween various operations of design, analysis, simu-
lation, fabrication, and assembly. As mentioned in
the previous section, the design process is initiated
through a conceptual framework. Once further elab-
orated, the creative effort is promptly developed into
a comprehensive 3-dimentional object representing
the building. Such a step is subsequently comple-
mented with structural design, where the BIM model
is exported to ROBOT for performance and integrity
analysis. At this stage, the effectiveness of data trans-
fer is insured through assigning performance prop-
erties to components built in Revit, to be then auto-
matically recognized by Robot. This allows the struc-
tural engineer to identify various elements and vali-
date the stability of the structure, leading to virtually
simulating members to environmental loads based
on building codes and/or contextual regulations. Re-
actions on members and connections are also com-
puted using numerical methods, enabling for pre-
cise definition of appropriate profiles to withstand
the calculated loads.
We have critically examined various alternatives
with regard to interoerability within the utilized BIM
environment, specifically in the case of structural
analysis. The goal was to develop a model of data
sharing between Revit and the structural analysis tool
without the need to reinserting or modifying struc-
tural families. Following various attempts, we ended
up with Robot after developing a series of functions
that allows structural components to be automated
according to architectural modifications, thus result-
ing in significant time savings. Once validated, the
outcome is then synchronized back to Revit through
another automated Robot function for design val-
idation. The finalized BIM file is then exported to
the Digital Prototyping platform; Autodesk Inventor,
where building foundation, exterior doors, windows
and roof in a non-editable format, are used to provide
reference levels and size of interconnected elements.
Autodesk Inventor is employed to generate pre-
cise allocation and assembly of structural members
within the design. The intricacy of the process relies
596 | eCAADe 34 - BIM | Concepts - Volume 2
on computational design capabilities of the design
team, where a series of special algorithms were de-
signed and coded to automate a set of operations.
For instance, one of the first algorithms in the loop
acts as placement agent that assigns structural com-
ponents following a specific pattern extracted from
Robot data, then generates required connections.
The process at this stage is optimized with regard
to time, and precision. furthermore, another algo-
rithm performs, and simulates a number of the as-
sembly tasks in real-time, thus overcoming possible
complexities associated with the building process.
The implementation of this algorithm takes the 3D
BIM model into a level of 4D and 5D BIM operations to
track assemblies, and extract quantities for cost anal-
ysis. Schedules are also pre-organized, and thus cost
can be automated according to a specific model, ei-
ther per sq.ft. or linear ft. Figure 2 demonstrates a
completed model on Inventor and the level of detail
obtained.
Realizing flexibility in design
Design for flexibility involves a certain logic of tech-
nological applications with regard to the provision of
services that allow for various configuration of com-
ponents. Such a logic is based on a clear differentia-
tion between elements that are fixed, and others that
are open to change and variation. Commonly, build-
ings are aimed to change in response to social, eco-
nomic, or environmental aspects. However, one of
the main limitation of implementing flexibility is high
cost, and complexity associated with linking design
to construction components ( Friedman 2011).
Perhaps one of the leading constructional prin-
ciples to facilitate flexibility in housing is "supports",
developed by Habraken in the 1960s, for the Dutch
housing sector. This theory distinguished between
two fundamental components: "supports" and 'in-
fills". While "supports" are regarded as the physical
entity, or the rigid part of the building, "infills" repre-
sent the flexible part that could be adjusted on dif-
ferent levels: social, industrial, economic and organi-
zational. The system was designed to facilitate varia-
 tions of floor layouts over time, while also accommo-
dating the design of dwellings to meet the diverse
standards of normally accepted housing in any par-
ticular society (Habraken 1999).
Along with integration of information technolo-
gies, digitization of design and production processes,
followed by sharp increase in demand for personal-
ized goods and products, new production paradigm
emerged; mass customization, a process that was for-
mally systematized by Joseph Pine in 1993. With re-
gard to architecture, Kolarevic (2003) explained how
mass customization holds great potential in architec-
ture, as buildings can become superbly unique and
highly customized products. Utilizing design and
fabrication tools in the form of CAD/CAM technolo-
gies, modern digitally integrated production pro-
cesses have produced a paradigmatic shift in pro-
duction ideology. This method permits optimal vari-
ance in response to differing local conditions, such as
uniquely shaped and sized structural components or
variable openings.
We believe that flexibility in design can be re-
garded as a core component in achieving customiza-
tion, whereas the level of customization relies primar-
ily on the degree of flexibility and its relevance to the
building system's technological capabilities. Accord-
ingly, we propose two significant flexibility models
being an integral part of the BONE Structure system:
micro and macro. While micro flexibility is denoted
with enabling participatory design in the form of pre-
construction choices to homebuyers, macro level ex-
plores the notion of variation in volume, in addition
to post-occupancy adaptability. Both flexibility lev-
els are enabled through the modular approach to the
Figure 2
A screen-shot from
Autodesk Inventor
of a 3D model
demonstrating
levels of details in
components and
connections in
project employing
fully standardized
parts.
design of the structural system to accommodate de-
sign variations, and supported by integrated appli-
cation of Digital Prototyping. The implementation
strategy of the digital strategy relies primarily on data
management, and product documentation.
Pertaining to micro flexibility, the integrated
workflow enables editing and changing parts and
components through manipulating parametric data.
It allows for real-time modification in building spa-
tial layout, while preserving the use of standard struc-
tural components. Additionally, employed to surpass
conventional capabilities of the software platform,
the placement algorithm, in addition to supplemen-
tary plugins, operate to enhance the level of flexibil-
ity to accommodate tremendous design variation. In
other words, flexibility in design is enabled through
flexibility in the data workflow strategy. Micro flex-
ibility is specifically intended for projects customiza-
tion, a growing trend in the prefabricated housing in-
dustry. While the common application of customiza-
tion covers the hierarchy of layout selection, to in-
ternal and external appearance of the housing unit,
what we are working on targets higher levels of cus-
tomization that allows for variations in housing lay-
out without affecting the structural integrity. Sup-
ports are columns, and infills are non-bearing walls
that can placed and modified according to users'
needs.
On the other hand, macro flexibility explores
variations in building volume and envelope, either
at early design stages, or even in the form of po-
tential post-occupancy adaptability in response to
future changes in socio-demographic characteris-
tics of occupants. In addition to being made pos-
BIM | Concepts - Volume 2 - eCAADe 34 | 597
 sible by the structural system framework, digital
project documentation would facilitate adaptabil-
ity, while overcoming structural and technical chal-
lenges. Given that parts and components of the
structure are assembled using bolts and fasteners,
and all of these connections are documented within
the comprehensive 3d model, variation in building
envelope can be achieved effectually. while in the
case of pre-construction design, the placement al-
gorithm tolerates configuration of special parts and
components, by replacing standard ones, yet post-
occupancy adaptability can be achieved by disas-
sembling components, then reassembled following
other configuration in response occupants needs. To
elaborate on customized design specifically, figure
3 represents a project that dictated the design and
fabrication of special parts to suite the design, while
keeping the principles; materials, process, and stan-
dardized connection details of the BONE Structure
system. The flexibility in design and optimization in
production are embedded in the process based on
the digital strategy included particularly in the coded
algorithms.
Digitally supported fabrication, and assem-
bly
The focus of the applied integrated digital strategy
is directed towards a multi-level modeling environ-
ment that benefits abundantly from Computer Aided
Design/Computer Aided Manufacturing (CAD/CAM)
598 | eCAADe 34 - BIM | Concepts - Volume 2
processes. It is structured based on a set of mod-
ules devised to aid in the process of modeling, simu-
lation, leading to precise fabrication, and assembly of
components. Commonly, a BONE Structure residen-
tial home project would include between 200 to 400
components in variable quantities, for a total in the
range of 8000 to 15000 individual steel parts. These
components are divided into twelve families, based
on the type of material and manufacturing process.
In addition to just-in-time virtual tracking of
components during design, the application of Dig-
ital Prototyping informs the fabrication and assem-
bly of parts. Autodesk Inventor generates fabrication
data to feed different Computer Numerical Control
(CNC) machinery involved in the process with precise
instruction. In addition to standardized parts, many
of the projects involve the design and production
of special parts for the sake of customized housing.
These parts are designed, tested, optimized, simu-
lated, and fabricated by simply manipulating param-
eters of the standardized parts. This also leads to sim-
plifying the workflow from design to manufacturing,
as collaboration occurs dynamically, thus maintain-
ing information accuracy.
The fabrication of parts relies primarily on a Figure 3
comprehensive file-to-factory process, where vari- Screen-shot from
ous fabrication instructions are extracted from 3- Autodesk Inventor
dementional models then sent in the form of digi- of a 3D model for a
tal files, including bills of materials. These instruction special design
are used to operate CNC machines, such as punch project. The image
presses, laser cutting, bending and stamping, for demonstrate level
the fabrication of various components. Additionally, of details attained
robotized welding is used for complex assemblies. while using special
Once fabrication is completed, the assembly pro- components.
cess is initiated two phases: pre-assembly in-factory
and on-site assembly. Simultaneously with fabrica-
tion, the engineering team works on pre-assembly
instructions. These instructions document quantities
of families and hardware components, steel parts,
fasteners and welding instructions, if applicable, as
well as precise sequential assembly procedures as
shown in figure 4. Accordingly, the process becomes
a predefined systematic practice that leverages qual-
 ity, saves time, and minimize waste remarkably, to- REFLECTIONS AND FUTURE WORK
wards a sustainable model. The interest in BIM applications in practice and
academia has been growing significantly in the last
Figure 4
decade. Many efforts have focused specifically on
A screen-shot from
methodologies to manage data and information
Autodesk Inventor
throughout the design and production of buildings.
of a 3D model of a
The advent in digital tools has opened new terri-
components and its
tories that can be considered as a leading edge of
assembly sequence
a chain of innovation in the whole building indus-
prior to site
try. The housing industry, and specifically the pre-
delivery.
fabricated housing sector, has witnessed great de-
velopment whereas digital design and manufactur-
ing strategies have inspired the most diverse research
and pragmatic solutions to contemporary industry
challenges. Nevertheless, the influence of these tech-
Based on the same virtual model, on-site assem- nologies only becomes effective when core systems
bly instructions are also generated to provide job- are embedded within a collaborative platform of in-
site team with precise instructions. These instruc- teroperability. In such a case, different participants
tions are provided in two formats. First, color- ultimately exchange data, and contribute towards an
coded three-dimensional drawings, comprising sym- integrated process; design, analysis, simulation, fab-
bols, part number, notes, and sections as partially rication, and assembly (Harry 2008).
demonstrated in figure 5. On the other hand, a cloud- We present in this paper represents a compre-
based 3D visualization interface is utilized to commu- hensive technological approach that we believe is
nicate with assembly team throughout the building considered a mile stone towards rethinking the way
process. The process becomes a catalog based as- we design and build housing, through combining a
sembly that required a trained team of profession- state-of-the-art structural system, and a digital mod-
als. Future intentions are focused on a more simpli- eling strategy; Digital Prototyping. Whereas Digital
fied process to reach a Do It Yourself (DIY) model, thus Prototyping has been applied in many engineering
making the system more affordable. and design domains, yet in architecture, such an ap-
proach is still in an exploratory phase, primarily re-
Figure 5 served to public and costly buildings. The described
A screen-shot from integrated platform in this paper tends to comple-
Autodesk Inventor ment the tactility of the structural system, towards ef-
of a 3D model of ficient building practices. This relates directly to the
components and concept of data management and collaborative qual-
their assembly ities, leading to effective communication between
instructions for design team members, as well as fabrication and as-
jobsite. sembly team , through establishing a coherent work-
flow to handle design flexibility. Furthermore, to
surpass conventional capabilities of the platform, a
computational design has been coded , and imple-
mented specifically to enhance various digital pro-
cesses. These computational tactics capitalize on the

BIM | Concepts - Volume 2 - eCAADe 34 | 599

 structural integrity of the design system, through au- Smith, P 2014, 'BIM Implementation – Global Strategies',
tomating a set of instruction within the modeling Procedia Engineering, 85, pp. 482-492
process, thus advancing the flexibility approach.
This research project is part of an ambitious en-
deavor towards exploring potential application of
mass customization in architecture, and specifically
in the housing industry. The proposed flexibility ap-
proach, in the form of combining a structural system
with a technological strategy, is seen as a crucial ele-
ment to realize the paradigm of mass customization.
Future work will explore further applications of com-
putational BIM, advanced automation towards more
effective interoperability, and alternative digital fab-
rication and assembly models.

REFERENCES
Bullinger, H.-J, Warschat, J and Fischer, D 2000, 'Rapid
product development — an overview.', Computers in
Industry, 42, pp. 99-108
Eastman, C, Teicholz, P, Sacks, R and Liston, K 2011, BIM
Handbook: A Guide to Building Information Modeling
for Owners, Managers, Designers, Engineers and Con-
tractors, Wiley
Friedman, A 2011, Decision Making for Flexibility in Hous-
ing, Urban International Press, United Kingdom
Habraken, N. J 1999, Supports: An Alternative to Mass
Housing, Urban International Press; 2nd edition
Harry, G 2008, 'Prefabricated Construction using Dig-
itally Integrated Industrial Manufacturing', ARCC
Journal, 5, pp. 48-65
Jelena, P and Maver, T 2003 'Digital prototyping in the
architectural design studio', Proceedings of SIGRADI
2003, Rosario
Kanai, S, Higuchi, T and Kikuta, Y 2009, '3D digital pro-
totyping and usability enhancement of information
appliances based on UsiXML', International Journal
on Interactive Design and Manufacturing (IJIDeM), 3,
pp. 201-222
Kolarevic, B (eds) 2003, Architecture in the digital age : de-
sign and manufacturing, NY Spon Press, New York
Pine, B.J 1993, Mass customization : the new frontier in
business competition, Harvard Business School Press,
Boston, Massachusetts
Salingaros, NA and Tejada, DM 2001, 'Modularity and the
Number of Design Choices', NEXUS NETWORK JOUR-
NAL, 3(1), pp. 99-109
Schneider, T and Till, J 2005, 'Flexible Housing: Opportu-
nities And Limits', Arq, 9, pp. 157 - 166

600 | eCAADe 34 - BIM | Concepts - Volume 2

Algorithm-Aided BIM
Harri Humppi1 , Toni Österlund2
1
Tampere University of Technology 2 University of Oulu
1
harri.humppi@gmail.com 2 toni.osterlund@oulu.fi

This paper investigates the relationship of Building Information Modeling (BIM),

Algorithm-Aided Design (AAD) and general Computer-Aided Design (CAD). As
a result of the developments in CAD, new tools have recently emerged that enable
designers to utilize some of the main aspects of BIM and AAD. With this new
modeling approach, called Algorithm-Aided Building Information Modeling
(AAB), designers can use algorithms to generate parametric object models. The
geometric objects contain embedded metadata that can be further utilized in the
design and construction processes. This paper investigates how the new
approach to modeling positions itself in the field CAD. The main result of the
investigation is that the new modeling approach can be seen as a part of larger
assembly that connects two design approaches of Object-Oriented Design and
Algorithm-Aided Design.

Keywords: Algorithm-Aided Design, Building Information Modeling, Digital

Design methodologies

INTRODUCTION defined scripts composed in either textual or graph-

New design methods have emerged through the
transition from analog to digital design. This tran-
sition has started only some decades ago and un-
doubtedly there are still many significant reforms to
come. At the moment, Building Information Model-
ing (BIM) is one of the most used digital design ap-
proaches within the building industry. In BIM ap-
proach, design commences by modeling geomet-
ric building components with their embedded meta-
data. The data contained in the objects can be
used during the design process but also through the
whole life cycle of a building. On the other hand,
the use of Algorithm-Aided Design (AAD) methods
are also becoming more common and integrated
into the workflows of design studios. In algorithmic
processes the form generation is controlled by user-
ical syntax. Scripts allow designers to generate and
control outcomes that would otherwise be too com-
plex to handle by manual modeling.
These two design approaches have been devel-
oping rather separately until recent years. Currently,
new tools have been developed that combine as-
pects of them both. They enable designers to gener-
ate and control object models through scripting and
exchange them with BIM tools (e.g. Mirtschin 2011).
These new tools affirm that advantages of BIM and
AAD can be used in the same design and modeling
processes.
At the moment, there are only few investigations
in to the integration of BIM and AAD on the theo-
retical level. Most of the research done in this area
focuses only on a certain narrow field related to the

BIM | Concepts - Volume 2 - eCAADe 34 | 601

 topic. A large portion of the current research fo-
cuses only on developing new software and toolsets.
There is a need to investigate the related theoreti-
cal aspects and ponder upon their implications, be-
yond the current tool development. On this basis,
this paper attempts to clarify the relationship be-
tween Algorithm-Aided Design and Building Infor-
mation Modeling and the views that they open to-
wards new methods of modeling and design.
This paper builds on the work of Humppi (2015),
which investigates the topic also from practical level.
The research methodology consists of literature re-
view, investigation to the state-of-art in design tools
and software, and case studies which help in building
the theoretical understanding on the subject.
CONCEPTS OF DESIGN PROCESS
Regarding the topic of the research, the meaning of
some basic terms are not yet settled (see Figure 1).
Here design is seen as an abstract concept con-
taining all aspects that are related to the design
event. Different design approaches are often com-
bined in the same design process. Design methods (or
methodologies) are different ways to design; they de-
scribe the interaction that happens between the de-
signer and the tool. The most essential digital design
methods regarding the research topic are modeling,
analysis, simulation and optimization.
Design tools are used to generate design solu-
tions. In the field of digital design, tools are usually
software, add-ons, plugins or custom scripts. All in
all, the design process is a series of methodological
steps that the designer takes in order to find a suffi-
cient design solution. In practice, the design process
is nonlinear, so repetition is needed to find the final
solution.
Three eras of CAD
Historical eras of Computer-Aided Design (CAD) are
shortly discussed through Figure 2. According to
Aish (2013) the history of practical CAD can be split
into three eras: the 2D Drafting era, the Building
Information Modeling era and the Design Computa-
602 | eCAADe 34 - BIM | Concepts - Volume 2
tion era. The diagram reveals an integrative path
of BIM and Algorithm-Aided Design, called here as
Algorithm-Aided Building Information Modeling (AAB).
All of these design approaches are shown as sub-
sets of Computer-Aided Design. The main data-flows
from design to construction are shown as separate
entities, but in reality the borders between eras and
data-flows are blurry and overlapping.
The 2D CAD era started in the 1980s, when com-
puters were adopted to present building designs as
multiple 2D drawings. The only essential change was
that computers partially replaced traditional draw-
ing on paper. Thus, the CAD era failed to utilize
the full creative potential of computers. (Aish 2013)
The development of BIM started also in the 1980s
(Aish 2013) and nowadays BIM is a commonly used
method for controlling building design processes. In
BIM designer can manually adjust parameters that
define the properties of each object. Thus, a BIM
model behaves like an assembly of rather indepen-
dent objects (Boeykens 2012). Parametric data of
these objects can be exchanged with other object-
oriented design software, where the structural com-
ponents are expressed as geometric objects with
properties. BIM provides the ability to digitally coor-
dinate often complex building processes from design
to actual construction (Garber 2014, 14). Nowadays,
BIM has reached a stable position in the field of build-
ing design and it has also overcome many limitations
of 2D CAD drawing.
Lastly, the Design Computing era introduced
the distinction between generative description of
the building and the generated model (Aish 2013).
Scripts and graphs are used to generate buildings,
forms and structures. The repetitive execution of a
script enables the exploration of many design alter-
natives. The main focus in the era has been the use
of algorithms to generate geometrical models that
do not contain any data except the geometry itself.
This approach can be called Algorithm-Aided Model-
ing (AAM). Thus, the parametric data and the geome-
try need to be exported and exchanged separately to
other design tools and eventually to fabrication and
Figure 1
Concepts of design
process. The grey
area refers to
design tools as also
in the following
circle diagrams.
construction.
BIM and AAD are digital design approaches
which have been used separately until recent years. A
comparison of BIM and AAD shows that both design
approaches have many advantages but also draw-
backs (e.g. Denis 2014, 14). For example, definition
of relations between objects is more challenging in
BIM than in AAM. On the other hand, BIM relies on
objects that can contain data facilitating the design
process.
Ambiguous terminology
The concept related to the design utilizing algorith-
mic logic and generative processes can be called
Algorithm-Aided Design (AAD) (e.g. Österlund 2013;
Tedeschi 2014) and is also often called Parametric De-
sign (e.g. Boeykens 2012; Harding 2014) or Genera-
tive Design (e.g. Ferreira and Leitao 2015; Feist et al.
2016).
There is no settled term for describing the new
modeling method that both utilizes algorithms for
generative modeling and generates objects with em-
bedded metadata. In research, it has been called
Algorithmic-based BIM (e.g. Feist et al. 2016), Gen-
erative BIM (e.g. Mirtschin 2011) and Parametric BIM
(e.g. Aish 2013). Combining aspects from BIM and
AAD, here the term Algorithm-Aided Building Infor-
mation Modeling (AAB) is proposed and used.
The terms Algorithm-Aided Design and
Algorithm-Aided Building Information Modeling de-
scribe well the main characters of these two design
approaches. These terms can be derived from the
abbreviation CAD, all of them containing the word
'aided', which refers to the aiding role of computers
and algorithms in design processes.
In addition, there is a difference between an ob-
ject and geometry. Geometry refers to an item that
does not contain any data other than the geometrical
representation itself. The term object is used to de-
scribe geometry that contains embedded metadata.
Data refers to knowledge that is presented or coded
in some form suitable for better usage or processing
BIM | Concepts - Volume 2 - eCAADe 34 | 603
 with computer.
Another important term related to research topic
is 'parametric modeling'. It can refer not only
to design through scripting but also to modeling
in BIM. For example, parametric modeling can be
divided into 'object modelling', 'associative mod-
elling', 'dataflow modelling', and 'procedural mod-
elling' (Janssen and Stouffs 2015; Janssen 2015).
Thus, parametric modeling can be used to describe
all modeling methods that use parameters to control
geometrical and object models, either through algo-
rithmic processes or object properties. AAB then de-
fines methodology, which combines the aspects of
parametric, algorithmic and object-oriented model-
ing.
The terms 'parameter' and 'attribute' are often
used in literature and in software interchangeably to
describe the data that is used to control and define
geometry or objects. No clear distinction exist be-
tween their uses, nor is there a settled definition that
604 | eCAADe 34 - BIM | Concepts - Volume 2
would make a difference between them (Weygant
2011, 25). Also, such terms as 'property' and 'feature'
can be used to describe the features of geometry.
Relation of BIM, AAD and AAB
The main starting point for the research has been
the awareness that Algorithm-Aided Building Infor-
mation Modeling (AAB) tools connect BIM and AAD.
This paper can be seen as a formal concept analysis
where the relationships between these concepts are
investigated.
One main notion has been that AAB is not an in-
dependent field of modeling. Instead, it is a combi-
nation of BIM and AAD. The second notion has been
that BIM refers to modeling and AAD refers to design
approach, so these concepts are not equal. On these
basis, there exists a need to form a more comprehen-
sive theoretical framework of concepts that are re-
lated to this new modeling method.
Figure 2
Three eras of CAD.
Adapted from (Aish
2013). Sizes of the
grey arrows refer to
current usage of
each modeling
method.
AAB TOOLS AND INTEROPERABILITY
One of the main challenges in digital design pro-
cesses is to ensure that the data flows fluently
through the whole design and construction pro-
cesses. Practically, how to ensure that all needed data
can be exchanged repeatedly through the process.
Interoperability refers to the ability to exchange data
between software, thus eliminating the need to repli-
cate data (Eastman et al. 2008, 66). A single com-
puter software cannot support all of the tasks that are
needed in the design of a project, so interoperabil-
ity is needed to enable fluent and automated design
processes.
In algorithmic modeling workflows the interop-
erability between tools is often lacking because pa-
rameters and geometry are exchanged separately.
Thus modeling methods combining algorithmic pa-
rameter control and object creation can potentially
foster these processes. They offer a possibility for
linking parameters to geometry. As Figure 3 shows,
here AAB tools are seen as a connection between
Object-Oriented Design tools and Algorithm-Aided
Design tools. On both sides of the figure, these tools
can be divided to parametric modeling tools and per-
formance evaluation tools.
Data-exchange between designs tools are
needed in the design process, so the main data-
exchange related to research topic are also presented
in the figure. Data-exchange between tools can be
split to direct and indirect file translations (Janssen et
al. 2015). AAB tools produce objects that can also be
exchanged directly or indirectly with object-oriented
design tools.
Figure 3 shows that AAB enables designers to
take advantage of both algorithmic and object-
oriented modeling tools. For example, designer
can use a Graphical Algorithm Editor (GAE) (e.g.
Grasshopper, Dynamo) to define the reference ge-
ometry for objects. Then AAB tools (e.g. Geome-
try Gym, Grevit, Grasshopper-Archicad Live connec-
tion) can be used for object creation in GAE. Per-
formance evaluation plugins can be used simultane-
ously with these modeling tools. Alternatively, ob-
jects can be exchanged with external object-oriented
performance evaluation software.
AAB and performance evaluation
As the previous section focused on design tools, this
section investigates design methods. Figure 4 shows
that AAB tools enable designer to combine two para-
metric modeling methods in the same modeling
workflow. Therefore, designer can proceed from al-
gorithmic modeling tool to object-oriented model-
ing tool or do the opposite. On this basis, there are
two perceived main directions for modeling work-
flows.
Maybe more obvious is the modeling workflow
where the designer generates new objects in GAE
and exchanges them to object-oriented tools. In the
opposite workflow the designer defines objects man-
ually in object-oriented modeling tools and then ex-
changes them to GAE. Then parameters of objects
can be modified algorithmically. Both of the pre-
sented workflows can be combined in the same mod-
eling workflow. Shortly, one of the main advantages
of AAB is that the designer can simultaneously cre-
ate or modify large amounts of objects through al-
gorithms defined in the script.
Algorithms can be used to generate object mod-
els on three levels; on model level, object level and ob-
ject joint level. On the model level algorithms can be
used to define reference geometry for objects. Then
the script can be used to control the shape and the
relations of the model. On the object level parame-
ters of objects can be controlled algorithmically. On
the object joint level algorithms can be used to de-
fine joints between objects. All in all, algorithms can
make modeling faster and more open for repetitive
parametric changes.
In Figure 4 are also presented performance eval-
uation methods that can be categorized to analysis,
simulation and optimization. The main aim of the
performance evaluation workflow is to find an opti-
mal result for a design problem. The most optimal re-
sults can be found by using iterative design process,
where the evaluation of a current design proposal, a
BIM | Concepts - Volume 2 - eCAADe 34 | 605
 proposition for alternatives, solving problems in col-
laboration and determination of the basis for next the
loop (Mirtschin 2011). This process can be manual
or handled automatically through optimization pro-
cesses.
AAB also facilitates double-iterative perfor-
mance evaluation processes. This means that these
processes can be managed in two stages. The pro-
cess consists of two loops, which are utilized sequen-
tially. In practice, this means that preliminary eval-
uations can be done in GAEs. These evaluations are
relatively fast, so the process can contain multiple
iterations. Designer can manage these simulations
606 | eCAADe 34 - BIM | Concepts - Volume 2
independently. When the sufficient precision level
of performance is achieved, the process can be con-
tinued in external performance evaluation software.
AAB tools generate objects, so secondary simulations
can be handled in external object-oriented software.
All in all, AAB can make external simulation processes
faster and facilitate iterative workflows because all
parameters related to simulation and objects can be
defined in GAE.
AAB as a part of digital design
Previous sections have focused on design tools and
methods. Here the main focus is in digital design ap-
Figure 3
Parametric tools
and AAB
data-exchange.
Figure 4
The relation of
Building
Information
Modeling (BIM) and
Algorithm-Aided
Modeling (AAM).
proaches. Figure 5 presents the main diagram, which
shows the discussed concepts together. The figure
affirms that AAB is not only connecting design tools
and design methods but also connecting two sepa-
rate digital design approaches called Object-Oriented
Design (OOD) and Algorithm-Aided Design (AAD). This
connection opens many new possibilities to CAD by
enabling many design processes that have not been
possible before. AAB tools enable designers to use
object-oriented and algorithm-aided design meth-
ods in the same design process. In other words, script
based design methods, such as modeling, analysis,
simulation and optimization, can be combined with
manual object-oriented design methods. This opens
new possibilities for collaborative design.
These two digital design approaches have been
rather separate until recent tool developments. Both
of these sides focus on different aspects of digital de-
sign and construction. OOD focuses on standardiza-
tion of design, fabrication and assembly, as AAD en-
ables non-standard procedures in them.
As a consequence of industrial revolution, in-
dustrial processes are commonly based on mass-
production, on which the OOD approach relies on.
Logics of form-making have been derived from in-
dustrial construction standards (Denari 2012). Even
current BIM software are based on these standards.
AAD relies on mass-customization, where the effi-
ciency can emerge in structures and construction
even if all building objects are singular (ibid). Cus-
tomization was challenging before because there
were no tools to fabricate custom objects. However,
technological enhancements in building component
production have enabled many fabrication processes
that were not possible before.
BIM | Concepts - Volume 2 - eCAADe 34 | 607

HYBRID DESIGN PROCESS
The use of AAB in design do not exclude other
computer-aided modeling methods or craft-based
methods such as drawing and physical model mak-
ing. For example, both AAM and BIM have some
advantages compared to each other, so designers
should be able to combine them in the same model-
ing workflow. One design tool can rarely respond to
all the needs of designers. Consequently AAB tools
can not correspond to all the needs either, so AAB
should be seen as a part of Hybrid Process.
608 | eCAADe 34 - BIM | Concepts - Volume 2
Figure 5
The relation of
Object-Oriented
Design (OOD) and
Algorithm-Aided
Design (AAD).
Hybrid Process refers to design processes that
combine a wide range of design methods, strategies
and tools (Sevaldson 2005, 318). This approach is
commonly used in architectural design. Then each
design method, including AAB, can be used cre-
atively and effectively.
AAB facilitates Hybrid Process by connecting two
digital design approaches that have been separate
before. The development of AAB could possibly lead
to situation where the relation between OOD and
AAD is even more integrated than currently. Maybe
then BIM, AAM and AAB could be used simulta-
neously to control the same model and workflow.
For example, the same object model that is defined
through scripting could be simultaneously manually
modified in BIM software.
Developing AAB
This paper introduced AAB by proceeding from digi-
tal design tools to digital design approaches. In prac-
tice, a design process of a building has the opposite
order. Firstly, there is a need to know why we are de-
signing before we can know how to design and what
we are designing. In order to develop AAB methods
and tools we need to know why we are developing
them. Design methods and design tools can only be
developed if the objectives of the design are known.
On this basis, here is suggested that the research and
development of AAB should also proceed from the-
ory to practice. As the Figure 5 showed, this paper in-
troduced a theoretical framework that positions AAB
in the field of digital design and CAD. This result can
be seen as a starting point for further theoretical re-
search.
REFERENCES
Aish, R 2013, 'First build your tools', in Peters, BP and Pe-
ters, TP (eds) 2013, Inside Smartgeometry: Expanding
the Architectural Possibilities of Computational De-
sign, John Wiley & Sons, Ltd, Printer Trento Srl, pp.
36-49
Boeykens, S 2012 'Bridging building information mod-
eling and parametric design', Proceedings of the Eu-
ropean conference on product and process modelling,
Reykjavik, pp. 453-458
Denari, N 2012, 'Precise Form For An Imprecise World', in
Marble, SM (eds) 2012, Digital Workflows in Architec-
ture, Birkhäuser, pp. 28-32
Denis, F 2014, Tool for Augmented Parametric Building
Information Modelling for Transformable Buildings,
Master's Thesis, Brussels Faculty of Engineering BR-
UFACE – ULB/VUB
Eastman, C, Teicholz, P, Sacks, R and Liston, K 2008, BIM
Handbook: A Guide to Building Information Modeling
for Owners, Managers, Designers, Engineers and Con-
tractors, John Wiley & Sons, Ltd, New Jersey
Feist, S, Barreto, G, Ferreira, B and Leitão, A 2016 'Portable
generative design for building information mod-
elling', Proceedings of the 21st International Confer-
ence of the Association for Computer-Aided Architec-
tural Design Research in Asia CAADRIA 2016, Mel-
bourne, pp. 147-156
Ferreira, B and Leitão, A 2015 'Generative Design
for Building Information Modeling', Proceedings of
eCAADe 2015, Vienna, pp. 635-644
Garber, R (eds) 2014, BIM Design: Realising the Creative
Potential of Building Information Modelling, John Wi-
ley & Sons, Ltd, Printer Trento Srl
Harding, J 2015, Meta-Parametric Design: Developing a
Computational Approach for Early Stage Collabora-
tive Practice, Ph.D. Thesis, University of Bath
Humppi, H 2015, Algorithm-Aided Building Information
Modeling: Connecting Algorithm-Aided Design and
Object-oriented Design, Master's Thesis, Tampere
University of Technology TUT
Janssen, P 2015 'Parametric BIM Workflows', Proceedings
of the 20th International Conference of the Association
for Computer-Aided Architectural Design Research in
Asia CAADRIA 2015, Daegu, pp. 437-446
Janssen, P and Stouffs, R 2015 'Types of parametric mod-
eling', Proceedings of the 20th International Confer-
ence of the Association for Computer-Aided Architec-
tural Design Research in Asia CAADRIA 2015, Daegu,
pp. 157-166
Janssen, P, Stouffs, R, Chaszar, A, Boeykens, S and Toth, B
2015, 'Custom Digital Workflows with User-Defined
Data Transformations Via Property Graphs', in Gero,
JSG and Hanna, SH (eds) 2015, Design Computing
and Cognition, Springer, pp. 511-528
Mirtschin, J 2011 'Engaging Generative BIM Workflows',
Collaborative Design of Lightweight Structures - LSAA
2011, Sydney
Sevaldson, B 2005, Developing Digital Design Techniques.
Investigations on Creative Design, Ph.D. Thesis, Oslo
School of Architecture and Design AHO
Tedeschi, A 2014, AAD_Algorithms-Aided Design: Para-
metric Strategies Using Grasshopper, Edizioni Le
Penseur
Weygant, RS 2011, BIM Content Development: Standards,
Strategies, and Best Practices, John Wiley & Sons, Ltd
Österlund, T 2013, 'Design Possibilities of Emergent
Algorithms for Adaptive Lighting System', Nordisk
arkitekturforskning: Nordic Journal of Architectural
Research, 1, pp. 159-185
BIM | Concepts - Volume 2 - eCAADe 34 | 609
Divide and Conquer, Mix and Match
A top-down and bottom-up approach to building information visualization

Helga Tauscher1 , Raimar J. Scherer2

1
TU Kaiserslautern, Department of Civil Engineering, Facilities Management and
Facilities Engineering 2 TU Dresden, Faculty of Civil Engineering, Institute of
Construction Informatics
1
helga.tauscher@bauing.uni-kl.de 2 raimar.scherer@tu-dresden.de

This paper seeks to explore the process of generating visual representations from
building information models in a formal way. Based on the reference model of the
visualization pipeline a more advanced model is developed which allows for the
consideration of different levels of detail in all stages of the visualization pipeline.
The construction of complex visualizations from simpler building blocks is
complemented by the successive subdivision of building information to generate
the partial visualizations from. The contributions of this paper are threefold.
First, it provides a formal model for information models, for visualization models
and for the connection between the two sides expressed with mathematical logic.
Second, it generalizes the idea of construction multimodels and utilizes it in a new
way in the context of visualizations. Third, it connects the multimodel concept as
a model of complex information models to the visualization pipeline as a model of
the visualization process.

Keywords: Building Information Modelling, Multimodel Visualization, Formal

Model

INTRODUCTION sive integrated information models to allow for a con-

The visualization of building information models be-
comes important with the growing complexity of
these models. In this section, we will first introduce
our approach to the visualization of these models,
our motivation and related work.
Background and Motivation
Building information modelling (BIM) has evolved
into a widespread method during the last decade.
Due to the development of increasingly comprehen-
tinuous application of information technology, less
and less information is established in the form of tra-
ditional documents and drawings. Instead, these vi-
sual representations are only derived on the fly by
software applications leaving few opportunities for
architects and engineers to shape the visual repre-
sentations which they are supposed to work with.
However, visual representations are essential to
the understanding of the information accumulated
during the design and planning phase of architec-

BIM | Concepts - Volume 2 - eCAADe 34 | 611

 tural construction projects. Visual analytics is an
emerging research area exploiting the superiority
of human vision over algorithmic methods when it
comes to analyse huge amounts of data (e.g. Keim
et al. 2010). Its possibilities for architecture and con-
struction are still to be explored.
In previous work the authors have worked to-
wards a method to enable domain specialists, i.e. ar-
chitects and engineers, to be actively involved in the
visualization process and to create their own visual-
izations. A prototypical framework was developed
that can generate visualizations from building mod-
els using a specification, and various construction-
specific visualizations have been studied and classi-
fied using the framework.
Approach: A formal model of the visualiza-
tion process
This paper presents the underlying model of the vi-
sualization process. Our model is based on the ref-
erence model of the visualization pipeline as intro-
duced by Haber and McNabb (1979), shown in Fig-
ure 1. Their model divides the visualization process
into three steps: filter, map, and render. These steps
do successively convert a given bulk of information
into a final visual representation. This model pro-
vides a rough, high-level view of the visualization
process, highly simplifying the real conditions found
at all stages of the pipeline: the high diversity on the
data or information model side just as the complexity
of informative visualizations.
According to the studies of Miller (1955) the abil-
ity of the human mind to handle complex issues is
limited and the preferred way of handling such is-
sues is to successively break them down into smaller
grouped issues that are easier to handle. The sug-
gested model of the visualization process is geared
to support the human mind in breaking down com-
plexity and to facilitate the specification of complex
visualizations.
Towards this goal, the visualization pipeline is ex-
tended from a simple linear process to cover more
complex scenarios in appropriate detail. The basic
612 | eCAADe 34 - BIM | Concepts - Volume 2
elements of the model are visualization techniques.
These are mappings from selected data to a specified
visualization structure. Simple visualization tech-
niques can combined to build a more complex one.
Conversely, a given complex visualization can be de-
constructed into its simpler building blocks.
Figure 2 shows the combination of two simple
visualization techniques into a complex visualization
technique. The operation of combining two tech-
niques consists of a top-down part and a bottom-up
part. In the top-down part the given input data is dis-
assembled by selecting subsets for each of the sim-
ple visualization techniques and for the coordination
of the subsequent assembly of the visualization tech-
niques. The bottom-up part combines the visualiza-
tion parts resulting from the basic visualization tech-
niques using a coordination model.
The process outlined above is described in the
paper as a formal model using set theory and pred-
icate logic (e.g. Pinter2014). This allows for a precise
formulation of the model yielding a threefold benefit.
First, it fosters a better understanding of visualization
processes in the context of construction specific en-
vironments and developments. Second, it serves as
a base for the implementation in a software applica-
tion. Third, it allows for further studies with a quanti-
tative approach, where methods to benchmark visu-
alization techniques could be based upon.
Related Work
There are other approaches to configurable visu-
alization generation. Brunetti et al. (2013) use
a component approach, where the transformation
stages of the visualization pipeline are encapsulated
and can be connected freely. They formally de-
scribe the connection between the components by
input signatures and then match compatible compo-
nents automatically. Chi (1999) extends the visualiza-
tion pipeline paradigm by complementing the three
transformation steps with four in-stage transforma-
tions. He proposes to combine operators of seven
types (three stage-changing and four in-stage oper-
ator types) to establish a more fine-grained view on
Figure 1
The visualization
pipeline - a
reference model of
the visualization
process. the visualization process and its configuration.
The idea of splitting and combining building in-
formation models into smaller or larger parts has
been studied for some time and from different an-
gles. Willenbacher (2002) looks at the subject from
the top and shows how partial models can be ex-
tracted from and kept synchronized with larger mod-
els. Fuchs (2015) in contrast tackles the problem from
the bottom and shows how elementary models can
be connected with link models to represent more
complex information structures. These approaches
constitute a basis for the model in this work.
THE MODEL
The following subsections describe the model we de-
veloped in details, starting with the top-down part,
followed by the buttom-up part, and concluding with
the connection of these two parts.
Divide and conquer
The top-down part of a complex visualization tech-
nique extracts submodels from a given building
information model. On the basis of the entity--
relationship model as proposed by Chen (1976), we
define an information model in this context to be any
tuple of the following three components:
• a set O of objects,
• a set A of attribute functions over subsets of
these objects,
• and a family R of relationship sets concerning
subsets of these objects.
We call the subsets of O that share their participa-
tion in common attribute functions and relationships
classes and denote them with C.
Relationship sets describe associations between
objects that are relevant in the context of the model.
A relationship set R ∈ R is defined as a set of rela-
tionships, each describing a number associated ob-
jects as an n-tuple. A relationship set R is thus a rela-
tion in the mathematical sense:
{(o1 , . . . , on )|o1 ∈ C1 ∧ · · · ∧ on ∈ Cn } (1)
or R ⊆ C1 × C2 × · · · × Cn . (2)
Attribute functions describe properties of objects
that are relevant in the context of the model. An at-
tribute function A ∈ A is defined as a function from
a class C ∈ O of objects to a value set V. The value set
V is the set of possible characteristic property values
for the property in question.
A : C → V with C ∈ O. (3)
In accordance with set theory (e.g. Pinter 2014) the
function A and its graph, the set of all ordered pairs
(o, A(o)) ∈ C × V , are considered to be essentially
the same thing. Thus, an attribute function A can be
expressed similar to a relationship set as
{(o, v)|o ∈ C ∧ v ∈ V } (4)
or A ⊆ O × V. (5)
This definition of an information model is deliber-
ately broad to not narrow the potential input data
to certain domain areas or implementation specifics,
and to be applicable to all sorts of potential building
information models.
In the disassembly process, submodels of the
given information model are selected from the infor-
mation model. A given information model is said to
be a submodel of another information model if the
following conditions are satisfied:
• the set of objects in the submodel is a subset
of the set of objects in the original model
• every graph of an attribute function in the
submodel is subset of the graph of any at-
tribute function in the original model
BIM | Concepts - Volume 2 - eCAADe 34 | 613

• every relationship set in the submodel is sub-
sets of any relationship set in the original
model
Thus, a model M ? = (O? , A? , R? ) is a submodel
of another model M = (O, A, R) if and only if the
following conditions are satisfied:
O? ⊆ O (6)
∀A? ∈ A? ∃A ∈ A : A? ⊆ A (7)
∀R? ∈ R? ∃R ∈ R : R? ⊆ R (8)
By selecting a subset of the objects in an information
model O? ∈ O it is possible to also select the sub-
sets of attribute functions and relationship sets con-
cerning only the selected objects, such that the se-
lection itself is a valid information model according to
our definition. More generally, based on a partition of
its object set, a given information model can be split
614 | eCAADe 34 - BIM | Concepts - Volume 2
Figure 2
Combination of two
simple visualization
techniques to form
a complex
visualization
technique.
into multiple elementary submodels and a remaining
link model, consisting only of relationship sets.
In order to split a model M = (O, A, R) into
n submodels M1 , M2 , . . . , Mn , the following steps
must be conducted:
1. Partition O according to an equivalence relation
Rsplit ∈ O × O into n partitions O1 , O2 , . . . , On
2. Partition each A ∈ A into n partitions
A1 , A2 , . . . , An , such that each partition contains
the object-value-tuples with the objects beeing from
the respective object set partition:
∀i ∈ {1, . . . , n}∀(o, v) ∈ Ai : o ∈ Oi . (9)
Each attribute function Ai is then the restriction of A
to the respective object set partition Oi : Ai = A|Oi .
3. Partition each relationship set R ∈ R into n
partitions R1 , R2 , . . . , Rn , and a reminder R0 , with
each object related through a relationship set Ri (ex-
cept the reminder R0 ) being in the object set Oi of
the corresponding submodel Mi :
∀i ∈ {1, . . . , n}∀(o1 , o2 , . . . , ok ) ∈ Ri
∀j ∈ {1, . . . , k} : oj ∈ Oi .
Each relationship set Ri is then the restriction of R to
the respective object set partition Ri : Ri = R|Ol
and the reminder R0 is
R0 = R\(R1 ∪ R2 ∪ · · · ∪ Rn )
or R0 = R\R1 \R2 \ · · · \Rn .
4. For every partition Oi of O collect the correspond-
ing set of non-empty attribute function partitions
and the corresponding set of non-empty relationship
set partitions to form an elementary model Mi as a
tuple (Oi , Ai , Ri ) with
{ }
Ai = Ai |Ai = A|Oi ∧ Ai 6= ∅ and
A∈A
{ }
Ri = Ri |Ri = R|Oi ∧ Ri 6= ∅ . (15)
R∈R
5. If partitions of the relationship sets remain as re-
minders, then every such partition R0 forms a link
model connecting the elementary models
LM = {R0 |R0 6= ∅}R∈R .
This procedure yields multiple submodels Mi that
each suffice the formal definition of an information
model given in equations (1) to (5). The resulting set
of submodels together with the link models corre-
spond to the concept of a multimodel as suggested
by Fuchs (Fuchs et al. 2011, Fuchs 2015). Fuchs
proposes to connect independent self-contained el-
ementary models using link models spanning multi-
ple of the original models. Such a multimodel con-
sists of a set EM of elementary models and a set LM of
link models. A multimodel with n elementary mod-
els and m link models can be described as a tuple
(EM, LM ) of these sets:
({M1 , . . . , Mn }, {L1 , . . . , Lm }) or
( )
{Mi }i∈{1,...,n} , {Lj }j∈{1,...,m}
Our approach inverts the view at the relation be-
tween part and whole as opposed to the multi-
(10)
model. It also disconnects the idea of the multimodel
from the existence of physically separate information
(11)
models in different files and instead transforms it into
a virtual structuring principle. Thus we are introduc-
ing a generalized notion of the original multimodel
approach.
(12)
Mix and Match
(13) Supplementing the top-down approach on the in-
formation model side, the bottom-up approach as-
sembles elementary visualizations on the respective
other end of the visualization pipeline, the visualiza-
tion models generated by a visualization mapping.
A visualization model is defined to be any tuple
of two components:
• a set O of visualization objects
(14) • a set A of attribute functions over subsets of
these objects
Let T be the set of possible points in time during
the presentation of a visualization and R2 be the Eu-
clidean coordinate space of the 2D area available for
the visualization. Similarly, for 3D visualizations, R3
would be the Euclidean coordinate space of the avail-
(16) able 3D volume. In the following we will only con-
sider 2D visualizations. Then a visualization model
can be transformed to a subset of space and time us-
ing an expansion function
( )
Exp : O → P R2 × T . (19)
Again, the definition is able to host a wide range of
potential visualization models. It is agnostic about
the semantic meaning of the attribute functions em-
ployed. It assumes that, in a valid visualization model,
for every class of visualization objects, the values
of its geometric attribute functions consistently de-
scribe the space it occupies in the visualization, with-
out underspecification or overspecification. Thus, for
every visualization object, its indexed set of attribute
(17) values can be correlated to a unique subset of R2 ×T .
Given this assumption, all spatial properties - like
(18) size, orientation, and the spatial relations between
BIM | Concepts - Volume 2 - eCAADe 34 | 615
 objects, such as neighbourhood, connection, and
overlap - can be derived from the attribute functions.
This is why, in contrast to the definition of an infor-
mation model defined in the previous section, there
is no family of relationship sets for the visualization
model. Adding relations to the model would intro-
duce overspecification and thus potential inconsis-
tencies. However, topological relations can serve as
input for layout algorithms, such as for Circular Car-
tograms (Dorling 1996). In such cases, relations re-
place attribute functions; the values of the attributes,
and thus the final geometric definition, are then de-
termined by the layout algorithm.
The reminder of this section will describe a
model for the combination of these expanded vi-
sualization parts to form an overall visualization us-
ing the visualization combination methods: Overlay,
Layout, Animation, and Interaction. These combina-
tion methods use different means of arranging and
connecting the single visualizations, either in layout
space or in visualization time.
Two visualization models M1 = (O1 , A1 ) and
M2 = (O2 , A2 ) can be combined by forming the
union of their objects sets and merging the attribute
functions into a common attribute function set A.
The simplest way to merge A1 and A2 is to form their
union. The combined model M is then
M = (O1 ∪ O2 , A1 ∪ A2 ). (20)
However, if two attribute functions A1 ∈ A1 and
A2 ∈ A2 are know to denote the same attribute on
the semantic level, their graphs can be merged into
a single attribute function, instead of including both
attribute functions separately into the merged visu-
alization model.
Since every attribute function in A = A1 ∪ A2
describes exactly one corresponding property, there
is a function as between the unified attribute func-
tion set A and the set of property semantics S - as :
A → S. The merged set of attribute functions is then
{ }
∪{A|as(A) = s}A∈A s∈S . (21)
We call a visualization model unambiguous, when
it does not contain any two objects occupying the
616 | eCAADe 34 - BIM | Concepts - Volume 2
same space at the same time, that is
∀o1 ∈ O∀o2 ∈ O : (22)
o1 6= o2 → Exp(o1 ) ∩ Exp(o2 ) = ∅. (23)
Even though the original models M1 and M2 might
have been unambiguous, their combination may in-
troduce ambiguities. A method to ensure unam-
biguous combinations is to use bounding boxes. A
bounding box is defined by the minimum and maxi-
mum values of the dimensions in the expansion of a
visualization object. For a two-dimensional visualiza-
tion model, the bounding box is derived as follows.
The expansion of an object set O is the union
of the result of applying the expansion functions
from Equation 19 to each single object: E =
∪o∈O Exp(o). The single dimensions can be ex-
tracted as
{ }
X = x|(x, y, t) ∈ R2 × T , (24)
{ }
Y = y|(x, y, t) ∈ R2 × T and (25)
{ }
T = t|(x, y, t) ∈ R2 × T . (26)
Let the functions min and max assign the minimum
and maximum value to a totally ordered set. For a
given set X these functions are defined such that
∀x ∈ X : min (X) ≤ x ≤ max (X). (27)
Using these functions, the bounding box BB in space
and time is then defined as the tuple of
min (X), max (X), (28)
min (Y ), max (Y ), (29)
min (T ), max (T ). (30)
The combination method based on bounding boxes
can be applied to any dimension of the visualization
space and is shown here for the dimension X in a two-
dimensional visualization model.
An offset function with an offset δx maps space-
-time onto itself such that the image is shifted by δx
in x direction:

Of f set : E → E with (31)
Of f set(x, y, t) = (x + δx, y, t). (32)
Setting δx to either max (X2 ) − min (X1 ) or δx =
max (X1 ) − min (X2 ), it results that min (X1 ) ≥
max (X2 ) or max (X1 ) ≤ min (X2 ) respectively,
such that there is no overlap.
Although the bounding-box is but one method
to ensure unambiguity whilst combining simpler vi-
sualizations into more complex ones, it is not suitable
for every unambiguous visualization. Figure 3 shows
a case which works and one that requires other lay-
out methods.
Figure 3
Arranging
compound objects
in space--time: While the combination method overlay does not in-
Overlapping case volve an offset function, layout is based on an offset
(left) and
function in the x- or y-direction, and animation and
nonoverlapping interaction are based on t-direction offset. Animation
case (right).
and interaction differ only by the way the time off-
set is specified: while animation has specific points in
time and time offsets defined, for interaction the off-
sets are specified dynamically through user triggered
or other events. All offset functions are chainable.
Connecting the Parts
Finally a visualization technique is constructed by
connecting the information model and the visualiza-
tion model side with the goal of creating a visualiza-
tion model from the information model. For a simple
visualization technique the visualization elements
are generated directly from information model ele-
ments. Complex visualization techniques integrate
a top-down part, one ore more visualization tech-
niques (simple or complex), and a bottom-up part.
Simple visualization techniques. Simple visualiza-
tion techniques operate directly on a given informa-
tion model and a visualization model to be created.
They are defined by three elements:
• a set of value mappings vm ∈ V M ,
• a set of value choices vc ∈ V C, and
• an object mapping om.
In the following, to distinguish between objects, at-
tribute functions and relationship sets in the informa-
tion model (on the data side) and in the visualization
model, we will add the suffixes D and V respectively
to the variable names already introduced earlier.
Value mappings establish relations between a
value set Vd in the model data to be presented and a
value set Vv in the visualization model. The relation
must be bijective, hence an invertible function in or-
der to be able to present all values unambiguously
and in a unique way. Let Vd be a value set in the
model data (the range of an attribute function) and
Vv a value set for a visualization attribute. A value
mapping vm is a function from Vd to Vv :
vm : Vd → Vv . (33)
Value choices are functions which specify how to se-
lect values for given data objects. A value choice vc
assigning values from a value set Vd to a subset Cd
of objects in Od can be expressed as
vc : Cd → Vd . (34)
The simplest case of a value choice is an attribute
function Ad ∈ Ad . More complex cases navigate
the object tree by combining specified relational at-
tributes before applying an attribute function. Finally
they might involve statistical functions to be applied
to selected values to retrieve the final value.
An object mapping om is a logical function from
a subset of Od to a set of objects Ov in the visualiza-
tion model.
om : Cd → Ov with Cd ⊆ Od . (35)
Together these three elements form mapping rules.
These rules are built by successively defining the
values of visualization object attributes depending
on values selected from specified data objects. The
value mapping vm must be preceded by the selection
of values from the information model (value choice
vc)and the correlation of the respective data and vi-
sualization objects (object mapping om).
BIM | Concepts - Volume 2 - eCAADe 34 | 617

Figure 4 shows how an object mapping, a value
choice and a value mapping work together to form
an attribute function in the visualization model. The
attribute function Av ∈ Av results as a concatena-
tion of these functions.
A v : O v → Vv (36)
Av = vm ◦ vc ◦ om−1 (37)
Complex visualization techniques. The construc-
tion of more complex visualization techniques from
simpler ones is based on the approach to complex
structures on the data side (top-down) and on the
model of complex visualizations (bottom-up). From
the given information model, multiple submodels
are extracted. Each of these is used to generate a part
of the visualization. These parts are then joined to
form the final visualization. The combination of the
single visualization parts may be guided by a ded-
icated coordination model, also selected as a sub-
model from the input model.
An integration method is characterized by the
following properties:
On the data side:
• whether submodels are needed or whether
the visualization part is data-independent
• whether the selection of submodels is inde-
pendent or guided by a coordination model
On the visualization side:
618 | eCAADe 34 - BIM | Concepts - Volume 2
• the dimension in which to integrate (combi- Figure 4
nation method to be used) Resulting attribute
• whether the specification of each subvisual- functions in the
ization is full or incremental visualization model.
Regarding the mapping:
• whether the position in the chosen combina-
tion dimension is independent or guided by a
coordination model
• whether the coordination model contains its
own set of mappings
• whether the visualization techniques (simple
or complex) are reused across subvisualiza-
tions
Since visualization techniques encompass the
whole visualization pipeline, the integration of vi-
sualization techniques can be seen as two parallel
pipelines that join at a certain point during the pro-
cess to produce a single visualization. Depending on
the stage in the pipeline at which the join occurs, dif-
ferent patterns emerge, as shown in Figure 5. These
patterns represent three different integration meth-
ods.
First, it is possible to represent each elementary
model in one independent "view". The visualizations
are interlinked on the view level. A second more in-
tegrative approach employs independent visualiza-
tion models that are joined in one view. The most in-
tegrative approach is to blend two elementary mod-
els by mapping onto a single visualization model. If
the integration is carried out in the data selection and
restructuring stage, the mapping consists of a single
simple visualization technique. These three types of
integration methods have been studied in more de-
tail in Tauscher and Scherer (2011).
SUMMARY AND OUTLOOK
A model of the visualizattion process was presented
that extends the reference model of the visualiza-
tion pipeline to accommodate more complex visu-
alizations. The model was described formally with
predicate logic. There is still work to do regarding
the connection of the top--down and bottom--up
Figure 5
Integrating two
visualization
techniques at
different levels in
the pipeline.
parts, since the complex visualization techniques are
to date only described in an informal way. The formal
description of the visualization process allows for a
precise formulation of the model yielding a threefold
benefit.
First, it fosters a better understanding of visual-
ization processes in the context of construction spe-
cific environments and developments. Other ap-
proaches focus ether mainly on the data or on the vi-
sualization side, while the approach presented here
seeks to connect both sides with an integrative ap-
proach.
One recent development in the construction
area is the multimodel, which attempts to connect
growing amounts of information while it maintains a
separation of concerns by keeping the diverse mod-
els separated. A generalized notion of the multi-
model may serve as a general structuring principle
for large information stocks and as such may be used
in other areas than visualization.
Second, the formal model may serve as a base
for the implementation in a software application.
The current prototypical implementation of a config-
urable visualization framework lacks functionality in
the area of complex visualization and might be ex-
tended based on the formal model described here
in the future. Third, it allows for further studies with
a quantitative approach, where methods to bench-
mark visualization techniques could be based upon.
We acknowledge the support of the German
Academic Exchange Service in the form of a Travel
Grant which enabled the attendance of this confer-
ence.
REFERENCES
Brunetti, JM, Auer, S, Garćia, R, Klímek, J and Nevčaský,
M 2013 'Formal Linked Data Visualization Model',
15th International Conference on Information Integra-
tionand Web-based Applications & Services (iiWAS), Vi-
enna
Chen, PPS 1976, 'The Entity-Relationship Model: To-
ward a Unified View ofData', ACM Transactions on
Database Systems, 1(1), pp. 9-36
Chi, EH 1999, A Framework for Information Visualization
Spreadsheets, Ph.D. Thesis, University of Minnesota
BIM | Concepts - Volume 2 - eCAADe 34 | 619
 Dorling, D 1996, Area Cartograms: Their Use and Creation,
Quantitative Methods Study Group of the Institute
of British Geographers
Fuchs, S 2015, Erschließung domäneübergreifender Infor-
mationsräume mit Multimodellen, Ph.D. Thesis, TU
Dresden
Fuchs, S, Kadolsky, M and Scherer, R 2011 'Formal De-
scription of a Generic Multi-Model', WETICE - 20th In-
ternational Conference on CollaborationTechnologies
and Infrastructures, Paris, France
Haber, RB and McNabb, DA 1979, 'Visualization Idioms:
A Conceptual Model for ScientificVisualization Sys-
tems', in Nielson, GM, Shriver, B and Rosenblum, LJ
(eds) 1979, Visualization in Scientific Computing, IEE
Computer Science Press, Los Alamitos, CA
Keim, D, Kohlhammer, J, Ellis, G and Mansmann, F (eds)
2010, Mastering the Information Age: Solving Prob-
lems with Visual Analytics, Eurographics Association,
Goslar, Germany
Miller, GA 1955, 'The Magical Number Seven, Plus or Mi-
nus Two: Some Limitson Our Capacity for Processing
Information', Psychological Review, 101(2), pp. 343-
352
Pinter, CC 2014, A Book of Set Theory, Dover Publications
Tauscher, H, Voigt, M and Scherer, RJ 2011 'Integrat-
ing visual presentations of construction multimod-
els: Visualization design space exploration', ConVR
2011: Proceedings of the 11th InternationalConfer-
ence on Construction Applications of Virtual Real-
ity2011, Weimar, Germany
Willenbacher, H 2002, Interaktive verknüpfungsbasierte
Bauwerksmodellierungals Integrationsplattform für
den Bauwerkslebenszyklus, Ph.D. Thesis, Bauhaus-
Universität Weimar

620 | eCAADe 34 - BIM | Concepts - Volume 2

Developing a Tangible User Interface for Parametric and
BIM Applications Using Physical Computing Systems.
Emad Al-Qattan1 , Philip Galanter2 , Wei Yan3
1,2,3
Texas A&M University
1,2,3
{emadkkqattan|galanter|wyan}@tamu.edu

This paper discusses the development of an interactive and a responsive Tangible

User-Interface (TUI) for parametric and Building Information Modeling (BIM)
applications. The prototypes presented in this paper utilizes physical computing
systems to establish a flexible and intuitive method to engage digital design
processes.The prototypes are hybrid UIs that consist of a digital modeling tool
and an artifact. The artifact consists of a control system (sensors, actuators, and
microcontrollers) and physical objects (architectural elements). The link between
both environments associates physical objects with their digital design
information to assist users in the digital design process. The integration of
physical computing systems will enable the objects to physically respond to
analog input and provide real-time feedback to users. The research aims to foster
tangible computing methods to extend the capabilities of digital design tools. The
prototypes demonstrate a method that allows architects to simultaneously interact
with complex architectural systems digitally and physically.

Keywords: Physical Computing, Parametric Design, BIM, Tangible UI

INTRODUCTION riencing the physical qualities of digitally conceived

Today, architects and designers use different meth-
ods to physically interact with digital models, which
are mostly based on GUIs. Some of the tools in use
may present more elaborate means of interaction
than others, such as in the case of immersive environ-
ments, e.g. virtual reality. Yet they remain as graphi-
cal representations of physical objects, in return they
lack to convey many of the physical qualities asso-
ciated with such forms. Additionally, graphic-based
methods do not offer the user the same means of in-
teraction that naturally occurs in the physical envi-
ronment. GUIs in the current practice of design limits
- to some extent - the user from exploring and expe-
designs
As commonly known, GUIs were developed in
the 1970s as new means of interaction that surpassed
CUIs (Ishii 2008). However, current practices in digital
design utilizing advanced modeling methods such as
geometric relationships and design information may
suggest the need for an alternative form of interac-
tion to manage and comprehend the complexity as-
sociated with digital models.
Physical design objects are an extraordinary
medium for interaction and feedback, and are of
great value to the iterative design process as earlier
research suggests (Globa et al. 2010). Nevertheless

BIM | Concepts - Volume 2 - eCAADe 34 | 621

 they remain as static productions of dynamic pro-
cedures. The absence of a live link between physi-
cal and digital design environments will most com-
monly result in the production of static forms where
the digital and parametric framework is cutoff from
the physical outcome. In other words, digitally fab-
ricated parametric models provide limited value to
the iterative design process, as they are incapable of
physically demonstrating or updating their paramet-
ric relations, geometry, and information.
Consequently, the research takes initiative to
propose a method that extends the functionality of
current tools and design methods to arrive at a novel
approach for engaging parametric and BIM-based
applications. The method will enable interaction to
take place in the physical environment using design
objects (elements), and provide real-time response
and feedback to the user through an object's kinetic
behavior. The prototypes are developed to demon-
strate the method through integrating digital mod-
els and artifacts. The artifact is the system's physi-
cal component, which consists of a control system
- with sensors and actuators - and physical objects
representing architectural elements. In this research
the artifact consists of a physical computing control
system and a full scale section of a paneled surface
with four panels. The artifact is linked to a digital
model of the panels that is generated using Rhino
and Grasshopper (Figure 1).
The research focuses on tangible UIs and interaction,
and the prototypes presented in this paper are at-
tempts to explore these notions. Physical computing
systems provide the essential components to trans-
late physical interaction into numeric data that could
622 | eCAADe 34 - BIM | Concepts - Volume 2
be used by the digital model. Physical computing
also assists in the process of creating real-time data
exchange between the digital and physical environ-
ments, which will allow users to interact with both
models simultaneously. The benefits of the proposed
method closely relate to the principles of tangible in-
teraction presented by Hornecker and Buur (2006),
which can be summarized in the following points:
first, physical and material qualities of design ob-
jects provide essential feedback to the iterative de-
sign process; second, the existence of objects in ac-
tual conditions (physical context) allows for direct
bodily interaction; and third, the link between both
environment will assist users to develop a complete
understanding of an architectural system by associ-
ating physical objects with their digital information.
Tangible Computing
Tangibility, interaction, and response are key terms
that define the current research, thus providing ex-
planation for each will create the proper context for
the work presented in this paper, and clearly state the
different types of hybrid systems used in design to-
day.
Tangible User-Interfaces associate digital infor-
mation with physical forms, thus proving users with
the ability to manipulate digital information through
their direct interaction with objects (Ishii 2008). Re-
search in tangible UIs and interaction started in 1995
by Fitzmaurice, Ishii, and Buxton prior to the intro-
duction of the term TUI by Ishii in 1997. Their work Figure 1
explored the link between "graspable" objects and The hybrid system;
graphical representations to establish a form of phys- left, the artifact
ical interaction. The graspable objects were specific consisting of a
in their method of interaction and data input, unlike control system
GUI components, which are mostly generic in nature, using physical
i.e. could be used for various applications (Fitzmau- computing
rice et al. 1995). components and
Tangible UIs link digital and physical environ- architectural
ments together to create a hybrid medium for physi- objects (panels).
cal interaction, which is most commonly established Right, a digital
by utilizing physical computing systems. Physical version of the
computing, in a broader sense is defined as the pro- paneled surface
modeled using
Rhino and
Grasshopper.
 cess of translating analog data into a virtual envi-
ronment (O'Sullivan and Igoe 2004), and its system
mainly consist of: microcontrollers, sensors, and ac-
tuators to create the interactive interface. Physical
computing systems vary in their degree of interac-
tion and response, and could be found in a wide
range of applications in the arts and sciences.
Hybrid interfaces vary in their degree of inter-
action and application. For that reason, it is best to
describe their relationship to Milgram's spectrum of
mixed reality (MR), which has two extremes, one be-
ing the real environment and the other being the vir-
tual environment, and in between are augmented re-
ality and augmented virtuality respectively (Milgram
and Kishino 1994) as show in Figure 2.
Figure 2
Milgram's spectrum
of mixed reality
(Milgram and
Kishino 1994). Augmented virtuality, describes the interaction with
digital environments; and augmented reality, de-
scribes the interaction with objects situated in their
physical environment (Hughes et al. 2005). The de-
gree of user interaction varies from one extreme to
the other along the spectrum. An example is vir-
tual reality (VR), which is a technology based on in-
formation processing that generates and simulates
graphical representations of actual - or hypothetical
- objects and environments (Tegmark 2007). Gener-
ally, VR environments provide users with an elabo-
rate style of interaction (involving a wider range of
senses) compared to conventional GUI-based tools,
yet objects in VR remain reserved in their native en-
vironment.
Conversely, physical computing systems process
analog information to communicate with a digital en-
vironment (O'Sullivan and Igoe 2004). In this research
physical computing is used to create a TUI for inter-
action by linking physical objects with their digital
information, a process referred to as embodied inter-
action. The process is defined by an alternative MR
than the previously stated where the continuum is in
a closed loop rather than defined by two extremes
(Salim et al. 2011; Dourish 2001). Comparing physi-
cal computing and other mixed reality models for de-
sign helps to establish the context for this research
and highlight the key differences between the differ-
ent types of hybrid systems.
Related Work and Significance
Previous research has shown the possibility of estab-
lishing a form of communication between the digital
and physical environments (Kensek 2014), which uti-
lizes environmental sensors to detect humidity, co2,
and daylight levels to inform BIM geometry and phys-
ical models. Another example uses a different ap-
proach in utilizing physical computing through the
use of Wii remote controllers to interact with para-
metric models (Salem et al. 2011). Both examples
provide insight into the use of different types of ana-
log inputs and physical interaction to engage digital
parametric and BIM-based models. Many more ex-
amples in architecture and design exist today that
demonstrate variations of physical computing se-
tups. In addition, software tools, including Firefly (a
plugin for Grasshopper), have also streamlined the
communication process between the two mediums,
which in return has provided users with the opportu-
nity to easily establish communication between dig-
ital and physical environments.
In this paper similar communication methods
and tools are adopted to manage data exchange be-
tween analog sensors and digital models. Tools uti-
lized for this research are updated versions of the
ones used in previous works, such as Dynamo, Fire-
fly, etc. However, the significance of the work is
found in the manner in which physical computing
systems are utilized. The proposed method and pro-
totypes mainly investigate user interaction and sys-
tem response to inform parametric models. The in-
terface utilizes different analog sensors that promote
different styles of user interaction rather than sensing
the physical environment. The prototype focuses on
providing tangible objects that represent architec-
tural elements rather than remote controllers, which
users could easily identify and are familiar with, thus
BIM | Concepts - Volume 2 - eCAADe 34 | 623
 making the physical interaction with digital designs
more intuitive for them. Last the prototype utilizes
digital fabrication methods and machinery, and cus-
tomized parts to enable complex physical responses.
The research focuses on establishing a sophisticated
medium for design interaction and response that en-
courages the users' personal involvement as design-
ers in a digitally based process.
Tools and Techniques
In the experiments different variations were done to
the physical computing setup. Each attempt investi-
gates different types of sensors, actuators, and meth-
ods to link both environments.
For clarification purposes, hardware and soft-
ware tools that are utilized for this research are orga-
nized in three groups:
Modeling tools: are used for geometric genera-
tion and transformation, and design data manipula-
tion. The tools used for the experiments include: Re-
vit (a BIM authoring tool) and Dynamo (a visual pro-
gramming environment and add-on for Revit); and
Rhino (a 3D modeler) and Grasshopper (a visual pro-
gramming environment and a plug-in for Rhino).
Data management tools: are used for data
(analog and digital) exchange between the two en-
vironments. Tools include: Processing (an open
source programming language and Integrated De-
velopment Environment, i.e. IDE), PLX-DAQ (plugin
for MS Excel and Arduino, developed by Parallax Inc.,
[1]), Firefly (developed by Andrew Payne, [2]), and Mi-
crosoft Excel.
Physical computing components: such as sen-
sors, actuators, and microcontrollers are used to cre-
ate the prototypes' control system and their kinetic
capabilities. The analog sensors chosen for the ex-
periments are operated through user input, and each
provides a different form of interaction. The sensors
include: rotary potentiometers and ultrasonic dis-
tance finders (referred to as proximity sensors). The
actuators enable objects to demonstrate physical re-
sponses, which provides users with physical feed-
back regarding the object's mechanics and kinetic
624 | eCAADe 34 - BIM | Concepts - Volume 2
performance. The actuators used for this experiment
are mainly servo motors (standard and high torque).
The microcontroller type used for the experiments is
an Arduino (MEGA 2560, [3]) that includes a single
CPU chip with limited RAM and ROM to manage the
data exchange process between the different com-
ponents of the system. Arduino is also a reconfig-
urable devices making it useful for a wide range of
real-time applications, thus making it a proper choice
for this research.
Framework and Prototyping
The hybrid system consists of two parts that are
linked together using a physical computing system.
The system is operated by user analog input, once
the system receives data, the models in both environ-
ments respond accordingly in a synchronized man-
ner providing visual and physical feedback (Figure 3).
Figure 3
prototypes’
workflow.
Each experiment investigates a different method of
physical interaction and TUI. The variations made to
the physical computing system (using different sen-
sors and actuators) and linkage methods were for the
purpose of improving the interactive experience of
the UI. The experiments are listed below according to
their type of linkage:
• Physical to digital: Dynamo, Arduino, prox-
imity sensor, potentiometer, MS Excel, and
PLX-DAQ.
• Digital to physical: Dynamo, Arduino, actua-
tors, MS Excel, and Processing.
• Hybrid system: full scale physical objects,
Grasshopper, Firefly, Arduino, proximity sen-
sors, and actuators.
 Prototype 1: Physical to Digital
The first experiment tested a simple form of physical
interaction with a unidirectional link that allows sen-
sor data to flow from the control system to the digital
model. The link was established using MS Excel and
the plugin PLX-DAQ (Figure 4).
Figure 4
workflow for
Prototype 1 (circuit
documented using
Fritzing, [4])
Figure 5
Left, the control
system using an
Arduino circuit and
two sensors
(proximity and
potentiometer);
right, the two
masses modeled in
The prototype includes two types of analog sensors:
Dynamo.
a rotary potentiometer and a proximity sensor (ultra-
sonic distance finder). In this example the artifact in-
cludes no physical objects, and the prototype is op-
erated by directly interacting with the sensors. The
digital model generated in Dynamo consists of sim-
ple geometric masses and two transforming param-
eters. Each parameter is linked to one of the masses,
and to one of the sensors on the control system (Fig-
ure 5). As a result, if a sensor in the control system is
operated the digital mass linked to it will display one
Figure 6 type of transformation.
Serial port Each mass in Dynamo is assigned to display one
communication type of the following transformations (motions): ro-
using the Arduino tation or horizontal movement. The masses' para-
IDE, MS excel, and metric behavior is constructed to reflect the manner
the PLX-DAQ in which the user will interact with the control sys-
plugin. tem; e.g. if the potentiometer is operated by rotating
its handle, the mass linked to it will rotate.
Data communication between the control sys-
tem and Dynamo was established through PLX-DAQ
and the computer's serial port. The setup using PLX-
DAQ allowed for a live data stream from the sensors
to the digital environment. Sensor data is sent by
Arduino in a CSV format (Comma Separated Values)
to PLX-DAQ, and afterwards was arranged in an MS
Excel spreadsheet. MS Excel arranges the values of
each sensor in a separate column, and the latest value
from each in a new raw (Figure 6). Data arrange-
ment and formatting was initially set in the Arduino
code. In Figure 6, sensor values in the MS Excel table
were rounded up. The column on the left, states "V"
for value to notify the user of the type of data being
received; the middle column includes the distance
measurements in inches provided by the proximity
sensor; and the column on the right includes rotation
angles obtained from the potentiometer.
The MS Excel file with sensor values is saved on
the computer's local drive. Dynamo afterwards lo-
cates the file, extracts the values, and sends them to
their assigned parameters in the model. The data ex-
change process in this setup is automated and in real-
time. This was achieved by using a MACRO for MS Ex-
cel, which automates the save function, and as a re-
sult, the CSV file will always include the updated sen-
sor values for Dynamo to use.
Interaction with the sensors matches the type of
parameters created in Dynamo; e.g. potentiometer
is a rotary sensor, which provides angles of rotation;
and the proximity sensor measures distance, which
provides movement values. Therefore, each param-
eter in Dynamo is setup to receive its values from its
corresponding sensor.
BIM | Concepts - Volume 2 - eCAADe 34 | 625

The first mass created is located at the origin of the
world coordinate system, and displays rotation. Its
angles of rotation values are provide by the poten-
tiometer. Once the mass receives sensor data, it will
respond accordingly by rotating in place (around the
origin and the z-axis).
The second mass created is linked to the proxim-
ity sensor and displays horizontal movement, which
corresponds to the manner in which the user will
interact with the sensor. Horizontal movement
demonstrates a simple parametric relationship that
could be constructed and operated physically. To fur-
ther explain, the proximity sensor detects nearby ob-
jects, and then it measure its distance from them. The
sensors and the orange cardboard (shown in Figure
5) demonstrate a similar relationship to the masses
in Dynamo; e.g. if the cardboard is moved closer to
or away from the sensor the masses will demonstrate
similar behavior. The sensor measures the distance
to the cardboard and sends the data to Dynamo, and
then used as parameter input values. The transla-
tion parameter moves the second mass horizontally
closer or farther away from the first mass along the
X-axis (Figure 7).
The experiment tested a simple form of TUI
where the users interact with the digital model
through operating the sensors. Sensors' setup re-
lates to geometry transformation in Dynamo (rota-
tion and movement), thus making the interaction
process straightforward for the user. The link in this
example has made the process of monitoring the
data flow and navigating the UI inconvenient, as a
result of running the different program applications
together for data exchange between the control sys-
tem and the digital model.
Prototype 2: Digital to Physical
In this experiment the data exchange process is re-
versed making the flow of data from Dynamo to the
626 | eCAADe 34 - BIM | Concepts - Volume 2
artifact to test the prototype's kinetic responses. The Figure 7
artifact in this experiment consists of two servo mo- Masses receiving
tors with a cardboard piece attached to each of their analog data from
rotating arms. The cardboards help in creating a sim- the sensors. Mass 1
ple physical representation of the digital model that (left) rotates around
would provide visual and physical feedback regard- the Z-axis and
ing the systems performance. Both environments origin, and is
are linked together through utilizing the tools imple- transformed by the
mented in Prototype 1, however the plugin PLX-DAQ potentiometer; and
in this example is replaced with Processing (Figure 8). Mass 2 (right)
Processing improved the link between the arti- moves horizontally
fact and the digital model by reducing the number of along the X-axis,
running program applications down to two, i.e. Pro- and its translation
cessing and Dynamo. The UI in this setup made the values are obtained
process of monitoring dataflow and navigating the UI from the proximity
more convenient in comparison to the previous ex- sensor.
periment.
Figure 8
Workflow for
Prototype 2.
The digital model from Prototype 1 is used for this
experiment with some modifications done to the
masses; both masses in this example demonstrate
rotational motion. The reason is to correspond the
digital models' type of transformation with the ser-
vos' rotational movement. The servos and digital
masses are operated through a single parameter in-
put, which provides angles of rotation values. In ad-
dition, the Dynamo model is imported into Revit to
expand the functionality of the system by including
a BIM authoring tool. This approach helps to estab-
lish a framework to associates digital design informa-
tion with physical objects for future work. The Re-
vit model and the artifact showed synchronized re-
sponses when parameter values were updated us-
ing a number slider (Figure 9). The prototype helped
demonstrate the physical limitations when attempt-
ing to create a kinetic and responsive object. This in-
 cludes constructing digital and physical models with lights, and two microcontrollers. The physical object
matching dynamic capabilities, e.g. digital objects represents a full scale section of an architectural ele-
must use similar types of motion that correspond to ment, i.e. a paneled surface with four Panels (Figure
the actuators. 10). The control system and the physical objects in
this example create an interactive and responsive ar-
Figure 9
tifact capable of demonstrating complex kinetic be-
Left, the artifact;
havior in physical space.
and right, the Revit
The artifact is linked to a digital model through
masses imported
the plugin Firefly to enable real-time and bidirec-
from Dynamo. The
tional data exchange. The digital model of the pan-
example shows
eled surface is created using Rhino and Grasshopper
both masses
(Figure 11). The objective of using an alternative dig-
rotating once the
ital modeling tool is to test the artifact's interactive
values of the
capabilities with a different modeling environment.
rotation parameter Creating a setup that allows digital and physical ob- The main consideration when selecting a modeling
in Dynamo are jects to demonstrate matching motions will help tool for the research was its parametric capabilities
updated. users to relate responses from both environments to and/or association with BIM information.
each other, thus making the interaction meaning- Each physical panel on the surface is capable of
ful and controllable for them. The model in Revit, rotating in two axis - referred to as Pan and Tilt mo-
showed a synchronized response using the linkage tions - to demonstrate rotations at different angles.
method described. Further attempts to utilize BIM The panels' pan and tilt motions were achieved by
databases are not explored in this example. How- attaching each panel to two servos using a U and
ever, the aim was to establish a form of communica- C shaped aluminum brackets (provided by Lynxmo-
tion between the physical object and the BIM author- tion, [5]). The assembly will allow each panel to ro-
ing tool for further investigations. tate in two different axis (Figure 12). The rotation an-
Figure 10 gles for each type of motion is obtained from a sin-
Artifact, showing gle proximity sensor in the control system. The mi-
the physical crocontroller receives the sensor values and remaps
architectural them to match the servos' range of motion, after-
objects (panels). wards the new angle values are sent to both the pan-
els and the digital model.

Figure 11
Workflow for
Prototype 3.

Prototype 3: Hybrid System

Prototype 3 integrates the methods previously stated
to establish an interactive and responsive hybrid sys-
tem. The control system includes: eight high torque
servo motors, two proximity sensors, sixteen LED

BIM | Concepts - Volume 2 - eCAADe 34 | 627


The aim of the prototype is to provide users with full
scale architectural objects with actual building mate-
rials. This will enable the prototype to convey physi-
cal qualities of the design in addition to digital infor-
mation. Therefore, to construct and fabricate the dif-
ferent architectural elements, having actual materi-
als, digital fabrication machinery and methods were
used. The two type fabrication methods that were
used for this experiment are: additive manufacturing,
using 3D printing (stereolithography); and subtrac-
tive manufacturing, using laser cutting. The material
used for the panels is Plexiglas with various thickness
to manage the panels' weight. In this context, the use
of digital fabrication methods logically tied the de-
sign and manufacturing processes together to create
a complete workflow for the project (Figure 13).
Physical limitations associated with design ob-
jects and the mechanics of their dynamic behavior
commonly pose a challenge to the process of devel-
oping tangible, interactive, and responsive UIs. The
specifications of the electromechanical components,
and the physical characteristics of the architectural
objects - including material and geometry - deter-
mine many aspects of the system's performance. The
cost value of the prototype at this stage including
building materials, customized parts, and electrome-
628 | eCAADe 34 - BIM | Concepts - Volume 2
chanical and electrical components may affect its im- Figure 12
plementation in certain design settings, such as in Testing the
a design studio. However, implementing interac- response of the
tive systems in an architectural project may provide artifact (left) and
the opportunity for further savings during a project's digital model
design and construction phases. AEC professionals (right).
have used similar methods with full scale physical
mockups for simulations (e.g. wind and solar stud-
ies). With this present approach, the system provides
the benefit of integrating both the physical and digi-
tal models together for establishing information and Figure 13
geometry relations, and the possibility of iterating Left, the proximity
the design, because of the objects' kinetic capabili- sensors’ 3D printed
ties. stand; middle, the U
In addition, sophisticated setups, such as the and C aluminum
prototype presented, are not energy demanding. brackets (by
The artifact in this experiment is running on 10 volts, Lynxmotion) ; right,
separated into two circuits with 5v for each. How- the Plexiglas frame
ever, the challenge is to determine the adequate lev- and panel support.
els of electricity and type of circuit required to prop-
erly run and operate the physical computing compo-
nents.
Figure 14
The artifact being
presented at an
exhibition (without
the digital model)
for visitors to
interact with.
In summary, Prototype 3 tested a bidirectional and
live link to interact with digital and physical mod-
els simultaneously with a set of sensors that were
straightforward and intuitive for the users to oper-
ate. The proximity sensors allows users to freely op-
erate the hybrid system by simply maneuvering their
hands in front of them, which is less restrictive than
 other types of sensors (Figure 14). LEDs are used
as an additional feature to the artifact, they provide
users with basic types of information, e.g. the system
is active or a certain panel, or object, is being oper-
ated. Therefore, the capabilities of the artifact using
a sophisticated physical computing setup and actual
building materials allowed for a complete design ex-
perience that offers feedback in physical and digital
form.
Discussion
The interaction with physical world objects makes us
Figure 15 experience their true qualities through our senses,
Artifact (left) and thus providing us with more information than what
BIM model linked to could be perceived on a screen. Consequently, hy-
evaluate a kinetic brid systems aim to make these tangible qualities
panel’s design for available for architects to properly engage in digital
shading. design processes.
Physical computing systems helps users to phys-
ically interact with parametric models in their na-
tive environment. Prototypes 1 and 2, used a simple
setup for the artifact that allows users to establish a
relationship between digital and physical objects' re-
sponses, and sensor analog input. This connection
was further explored in the development of Proto-
type 3. The process of creating Prototype 3 is closely
related to a conventional design process with a feed-
back loop where its objects' kinetic behavior, geome-
try, and relationships were constantly informing each
other during the progress of the work.
Nevertheless, the control system in the proto-
types presented only provided analog data to the
system where the parametric relationships are pre-
defined and constructed by the user. As commonly
known, the prevalent feature of a parametric mod-
els is in the explicit creation and use of relationships.
Currently, the process of constructing parametric re-
lationships to manage geometric and non-geometric
information pose a challenge for architects, espe-
cially when working on complex architectural sys-
tems. Tangible UIs and interaction could be fostered
to establish these relationships in the digital model.
To further explain, the interaction may take place in
the physical context, where the architect thinks of re-
lationships directly between the design objects, and
the hybrid system translates the architect's physical
relationships into the digital environment. Therefore,
utilizing physical computing systems to detect physi-
cal object relationships and reconstructs them in the
digital environment provides a benefit to the para-
metric design process and practice that extends the
functionality of modeling tools. Future work for this
research will investigate this feature for the develop-
ment of the hybrid system.
An example of work in progress is included in this sec-
tion that demonstrates the use of physical comput-
ing systems in the context of kinetic shading design.
The work focuses on the aspect of extending the ca-
pabilities of design tools by associating design ob-
jects' physical properties, BIM's geometric and non-
geometric information, and simulation tools such as
DIVA (Figure 15) to provide AEC professionals with a
hybrid system that allows them to properly explore
and evaluate an architectural system. Non-geometric
information refers to digital design data that is asso-
ciated with an architectural element in a BIM model;
such as: an element's parameters, material proper-
ties, etc. The artifact includes a single panel with a
reflective surface that is capable of rotating in three
axis; two axis are used for sun tracking, and one axis
is used to provide users with an additional level of
control. Users operate and control the hybrid system
through digital parameters and smart devices (e.g.
smart phones, tablets, etc.) that are linked to both
the artifact and the digital model. The Revit model
in this experiment is adapted from the work done by
Jeremy Roh (School of Architecture - UNC Charlotte,
[6]), which provides a good example for testing the
BIM | Concepts - Volume 2 - eCAADe 34 | 629
 system. However, further adjustments were made to
the Revit model and to the Dynamo graph to ensure
their compatibility with the artifact.
Conclusion
The research focuses on developing a Tangible, inter-
active, and responsive UI for parametric and BIM ap-
plications through utilizing physical computing sys-
tems. The research aims to investigate the possibil-
ities of extending the functionality of design tools
by linking digital and physical design models. The
experiments focused on developing methods that
would allow AEC and non-design oriented users to
expressively engage parametric and BIM-based de-
signs. The kinetic capabilities of the hybrid system
provides users with means to physically experience
the qualities of a design while providing them with
essential feedback in physical and digital form, which
is related to the design's geometry, parametric rela-
tionships, and information. In summary, the value of
the system to users which mixed reality applications
may not offer are: first, having the ability to interact
with digital designs through physical objects, which
directly relates to the manner in which we naturally
engage objects in our physical environment; second,
the physical existence of objects in an actual con-
text provides users with the ability to experience first-
hand the qualities of space, geometry, and materials;
and third, the system's kinetic ability provides feed-
back in a physical form that would assist in further
understanding the relationships governing an archi-
tectural system, and as a result, it would better inform
our design process.
REFERENCES
Dourish, P 2001, Where the action is: the foundations of
embodied interaction, MIT Press, Cambridge
Fitzmaurice, GW, Ishii, H and Buxton, B 1995 'Bricks: lay-
ing the foundations for graspable user interfaces',
Proceedings of the ACM CHI 1995, Denver, Colorado,
pp. 442-449
Globa, A, Donn, M and Twose, S 2010, 'Digital to physi-
cal: comparative evaluation of three main CNC fab-
ricating technologies adapted for physical modeling
in architecture', Journal of Architectural Computing,
630 | eCAADe 34 - BIM | Concepts - Volume 2
10(10), pp. 461-480
Hornecker, E and Buur, J 2006 'Getting a grip on tangible
interaction: a freamework on physical space and so-
cial interaction', Proceedings of ACM CHI 2006, Mon-
treal , pp. 437-446
Hughes, CE, Stapleton, CB, Hughes, DE and Smith, E
2005, 'Mixed reality in education, entertainment and
training: an interdisciplinary approach', IEEE Com-
puter Graphics and Applications, 26(6), pp. 24-30
Ishii, H 2008 'Tangible bits: beyond pixels', Proceedings
of the Second International Conference on TEI 2008,
Bonn, Germany , pp. 15-25
Ishii, H and Ullmer, B 1997 'Tangible Bits: Towards Seam-
less Interfaces between People, Bits and Atoms', Pro-
ceedings of the ACM CHI 1997, Atlanta, Georgia, pp.
234-241
Kensek, K 2014, 'Integration of environmental sensors
with BIM: case studies using Arduino, Dynamo, and
the Revit API', Informes de la Construccion, 66(536), p.
e044
Milgram, P and Kishino, F 1994, 'Taxonomy of mixed real-
ity visual displays', IEICE Transactions on Information
Systems, E77-D(12), pp. 1321-1329
O'Sullivan, D and Igoe, T 2004, Physical computing: sens-
ing and controlling the physical world with computers,
Thomson
Salim, FD, Mulder, HM and Burry, JR 2011, 'Form foster-
ing: a novel approach for interacting with paramet-
ric models in the embodied vitality', Journal of Infor-
mation Technology in Construction, 16, p. 133–148
Tegmark, M 2007, 'The mathematical universe', in Chiao,
RY, Cohen, ML, Leggett, AJ, Phillips, WD and Harper
Jr., CL (eds) 2007, In visions of discovery: shedding
new light on physics and cosmology, Cambridge uni-
versity press
[1] http://www.parallax.com
[2] http://www.fireflyexperiments.com
[3] http://www.Arduino.cc
[4] http://www.fritzing.org
[5] http://www.lynxmotion.com
[6] http://www.jrohdesign.com
BIM | Applications
Captivity or Flexibility: Complexities in a Dimensional
Customization System
Salman Khalili-Araghi1 , Branko Kolarevic2
1,2
Faculty of Environmental Design, University of Calgary
1,2
{skhalil|branko.kolarevic}@ucalgary.ca

Houses are essentially one-of-a-kind products that should reflect individualized

differences of inhabitants who live in them. Homebuyers and homebuilders alike
are thus captivated by the difficulties of housing customization. Achieving
customer satisfaction depends on the flexibility of customized solutions, though
the challenge of flexibility lies in the complexity of design validation. Constraints
may be seen as design limitations, but they could provide for the efficiency of
design validation. This paper addresses the complexities in the adoption of mass
customization in the housing industry, and presents a dimensional customization
system which would effectively use building information modeling (BIM)
software, parametric design, and automatic verification of dimensional
constraints to merge customization and validation.

Keywords: Mass Customization, Housing Industry, Building Information

Modeling, Parametric Modeling, Automatic Constraint Satisfaction

INTRODUCTION try focusing on affordability and low-cost production.

Social and cultural norms affect the way people
house themselves. Although houses (or apartment
buildings) can be characterized by sameness, it is the
socio-cultural heterogeneity that differentiates our
needs. Noting this lack of variety in housing, Kolare-
vic (2015) points out that customer demands tend
to be heterogeneous, while physical buildings are
largely homogenous.
The sheer necessity and the high demand for
housing after the Second World War made architects
and homebuilders rely on the economies of produc-
tion from other industries. Economic reality was the
real reason for applying industrial methods based
on efficiency and economy of production in quan-
tity. Mass production, as a Fordian paradigm of the
20th century, was embraced by the housing indus-
Repetitively-produced housing, however, was recog-
nized as monotonic and was noted for its failure to
provide for variety.
From a technological perspective, the recogni-
tion of disparity between physical homogeneity and
cultural heterogeneity compels one to search for a
customization system. As part of ongoing research,
this paper proposes a framework for developing a
flexible customization system that could allow for
both individuality and profitability through customer
participation in design, while fulfilling the viability
of individualized housing solutions. Specifically, this
paper focuses on the challenge of customer partic-
ipation at an early stage of the design process. It
argues that dimensional customization is a suitable
paradigm for the housing industry; houses essen-

BIM | Applications - Volume 2 - eCAADe 34 | 633

 tially do reflect the cultural and social heterogene-
ity of customers and could be tailored as one-of-a-
kind products (Kolarevic 2015). The following sec-
tions first discuss mass customization in the hous-
ing industry, second describe the necessity, potential
benefits and challenges of customer participation in
the design process, and then provide a description
of a proposed dimensional customization system, in-
cluding its characteristics. Following that, the im-
plementation of the system is presented, concluding
with a discussion of the limitations of present work
and possibilities for future studies.
MASS CUSTOMIZATION
Mass customization, as a post-Fordian paradigm, is
a way of delivering highly customized products or
services while still maintaining mass production effi-
ciency. The term "mass customization" was originally
coined by Stan Davis in (1987) as a business strat-
egy in which "the same large number of customers
can be reached as in mass markets of the industrial
economy, and simultaneously [...] be treated individ-
ually as in the customized markets of pre-industrial
economies" (Davis 1987, 169). Similarly, Joseph Pine
(1993) defines the goal of mass customization as pro-
viding enough variety in products and services at rea-
sonable prices.
Over the past decade and a half, there were sev-
eral attempts to embrace mass customization in ar-
chitecture (e.g. Kieran and Timberlake 2003, Larson,
Tapia and Duarte 2001, van den Thillart 2004, Duarte
2001). Due to potential design complexity, different
and lengthy review and permitting processes, and in-
herent fragmentation in the building industry, home-
builders hesitate to allow greater customer involve-
ment in the design process. In the building industry,
homebuilders want to maintain the efficiency of pro-
duction process by using standard models, though
homebuyers demand houses that are individually
configured and reflect their personal needs (Nah-
mens and Bindroo 2011, Hofman et al. 2006). Such
an approach, however, does not provide for flexibil-
ity and variety and may not result in customers find-
634 | eCAADe 34 - BIM | Applications - Volume 2
ing the solution they are looking for. Thus, the chal-
lenge remains how to satisfy customer demands and
insure profitability without sacrificing the efficiency
of mass production. The following section discusses
customer participation as a fundamental of mass cus-
tomization and its challenges in the building indus-
try.
CUSTOMER PARTICIPATION
Mass customization is not merely product variation,
as customers play a key role in housing customiza-
tion. Alvin Toffler (1990) identifies the term "pro-
sumer" to describe the producing and consuming
role of the customer in which the customer becomes
the co-designer. Mass customization should allow
customer participation in the design process to de-
fine and modify individualized solutions (Piller et al.
2004, Salvador et al. 2009, Franke and Piller 2004).
Mass Customization Challenges
The basic application of mass customization in the
housing industry, which has been growing in pop-
ularity, is to offer customers a limited number of
choices in the forms of either printed or electronic
catalogues, using a multiple-choice approach. Sears,
Roebuck and Co., from 1908 to 1940, can be seen as
the most established company that offered a cata-
log of house plans with added advantage of mod-
ification possibility (Thornton 2004). In web-based
customization systems, homebuyers are allowed to
navigate housing solutions at different levels, includ-
ing layout, elevations, and interior elements. For ex-
ample, Resolution: 4 Architecture's website [1] of-
fers 'modern modular' homes by developing a se-
ries of predefined typologies based on the possibil-
ity of variation within a standardized system. Living
Homes' website [2] offers several stylistically different
house designs with various customization options re-
garding exterior/interior elements including finish-
ing, appliances and HVAC systems. BluHomes' web-
site [3] offers standard modular prefabricated homes,
and allows customers to make decisions within a
"conceptual design" process, and manual 'code and
zoning research' at the final step.
Multiple-choice approach offers customers lim-
ited number of solutions to choose from. Due to the
complexity of the design process, homebuilders pre-
fer to limit the solution space, as the entire range of
alternatives would have to be designed up front. Be-
cause the designs are created without customer in-
volvement, there is no promise of customer satisfac-
tion - which is highly contingent upon the flexibility
of solution space. Increasing the number of options
could potentially increase the chance of finding sat-
isfactory solutions; however, it could also be argued
that having to choose among a large number of so-
lutions could substantially increase the complexity of
the customization process. The potential complexity
inherent in the flexibility of solution space could lead
to information overload, which is seen as a drawback
(Huffman and Kahn 1998, Zipkin 2001). One conse-
quence of information overload is 'mass confusion',
i.e. customer's inability to cope with making a de-
cision among a large number of solution, making it
more likely that the customer will become confused
(Pine 1993, Huffman and Kahn 1998). If customers are
dealing with a great variety (making it appear com-
plex), they may not be able to explore all possibilities
and therefore may not find the ideal solutions. In ei-
ther scenario, customers might become dissatisfied
and uncertain about their ability to continue partici-
pating in the co-creation process and thus turn back
to standard options (Huffman and Kahn 1998, Franke
and Piller 2004) .
Another drawback of customer participation is
the customer's inability to match needs with cus-
tomizing products. The success of customer partic-
ipation is highly dependent on the stability of cus-
tomer preferences, customer awareness of their pref-
erences, and customer ability to transfer them into
product specification (Simonson 2005, Kramer 2007).
Customer need-related information is attributed as
sticky, where the stickiness is defined as "the in-
cremental expenditure required to transfer that unit
from one place to another in a form that can be
accessed by a given information" (Franke and Piller
2004, 404). Simonson asserts that many customers
cannot precisely determine their needs, and their de-
cision could be changed based on situational cues
(Simonson 2005). In the conventional design pro-
cess homebuyers and the designer become involved
in a back-and-forth process for several design iter-
ations to find satisfactory solutions. Every transac-
tion implies information exchange, which depends
on understanding customer need-related informa-
tion. Customer preferences, however, are often ill-
defined, and as noted by Zipkin (2001), "customers
often have trouble deciding what they want and then
communicating or acting on them [...] There are situ-
ations in which customers clearly articulate their re-
quirements. More commonly, however, customers
are unsure" (Zipkin 2001, 82). Thus, translating pref-
erences into product specification determines cus-
tomer satisfaction and, in fact, is bounded by the ef-
ficiency of communication between customers and
the designer. Therefore, this becomes the source
of additional cost for the company to transfer cus-
tomer's needs into the product specification. Deal-
ing with the challenge of need elicitation and prod-
uct specification, mass customization is subject to
shift design tasks to customers by allowing them to
express their needs and participate in the design of
their own product (Thomke and von Hippel 2002, von
Hippel 2001).
From the company's perspective, customer par-
ticipation in the design of individualized houses
presents the challenge of design validation - ensur-
ing that individualized solutions are viable. Due to
the complexity of design validation, homebuilders
offer filtered information about housing options and
prefer to limit customer participation at the level of
choice selection. None of the industrialized hous-
ing companies supports customization at the level
of housing geometry. As noted by Lampel and
Mintzberg (1996), the level of customization depends
on customer involvement. Tailored customization is
identified as the customer involvement in customiz-
ing a standard product at an early stage of the design
process (Lampel and Mintzberg 1996). Tailored cus-
BIM | Applications - Volume 2 - eCAADe 34 | 635
 tomization at the level of determining various dimen-
sions of the house's exterior and interior offers the
highest level of customization. The deeper the level
of customization, however, the higher the complex-
ity of design validation. Customizing the overall spa-
tial layout of building geometry is extremely contin-
gent upon substantial design verification process to
ensure that variations comply with design, engineer-
ing and manufacturing constraints. The process of
manually evaluating and adjusting changes against
constraints is error-prone, costly and time consum-
ing, and could be a major driver of complexity. The
most prominent approach is to offer customers a flex-
ible solution space from which they could explore in-
finite number of housing alternatives, while ensuring
the validity of customized solutions.
PROPOSED SYSTEM FRAMEWORK
Dimensional customization in the building industry
requires customer involvement at an early stage of
the design process so that a house could have a
unique shape and form; building geometry can be
tailored individually using dimensional parameters.
The flexibility offered by such a large solution space
can be enhanced by technologies that could pro-
vide for customer participation in the design of mass-
produced but still highly customized houses.
Implementation of a dimensional customization
toolkit is based on using building information mod-
eling (BIM) software for a parametric definition of
the geometry, a simple but layered interaction with
the parametric model, and automatic verification
of design variations. BIM technology could en-
hance the communication between the customer,
designer, and manufacturer from the schematic de-
sign of customized products (houses) to construction
documentation (Garber 2014). Every single modifi-
cation could be automatically replicated within the
entire model, and geometrical (and other) informa-
tion of customized solutions could be extracted and
transferred directly for manufacturing purposes, us-
ing file-to-factory processes.
In dimensional customization, maintaining rela-
636 | eCAADe 34 - BIM | Applications - Volume 2
tions between elements of the model and paramet-
ric variation of essential dimensions are essential to
derive geometry numerically. Parametric modeling
enables the customer to explore design variations by
applying different dimensional parameters to a sin-
gle model. This process can produce an infinite num-
ber of individualized solutions that differ from each
other, but remain within the boundaries defined by
the designer. Instead of designing multiple houses
or offering limited number of options, the designer
becomes involved in designing a 'meta' house, pro-
viding for the flexibility of solution space within an
infinite number of dimensional variations. The cus-
tomer could then interact with the parametric model
via an interactive toolkit to design their individual-
ized houses.
Dimensional customization system introduces
a deeper level of customization by providing
parametrized geometry that complies with design
constraints. Design and performance limitations,
rules and regulations imply constraints. In a tradi-
tional process, if customized houses do not comply
with rules and regulations, the designer may need to
consult other team members (i.e. engineer, manu-
facturer), and inform the customer how to appropri-
ately modify the design so that it complies with the
constraints. If the proposed solutions do not satisfy
the customer (as if often the case), the entire cus-
tomization process needs to start over. Automated
constraint checking could significantly benefit the
process of design validation as the entire model is
automatically updated to accommodate changes.
Companies could benefit from constraint-based
parametric modeling, in that customization is per-
formed over a single model that allows the customer
to explore a vast number of solutions, while simulta-
neously complying with design rules. To realize this,
all constraints should be quantitatively embedded
into the parametric design model and represented
at an early stage of the design process. Doing so
helps ensure that essential features are maintained
as design changes to avoid evaluation reworks.
SYSTEM IMPLEMENTATION
The prototype discussed in this paper is imple-
mented within Autodesk Revit®, the most commonly
used three-dimensional building information mod-
eling software in the industry. The plug-in is imple-
mented through API based on three main functions:
dimensional constraint definition, spatial zone defi-
nition and the satisfaction of constraints. The follow-
ing sections present the technical characteristics of
the system and the implemented interface.
Product Architecture and Structural Hierar-
chy
Mass customization in housing reflects the dynamic
configuration of housing elements within the static
context of the building. In this way, the form is
interpreted on the basis of the possibility of dif-
ferent combination of elements made by the cus-
tomer, rather than conceptual definition of what con-
stitutes a building. As a fundamental element in
mass customization, product platform refers to a set
of components, modules, or parts that are physically
connected and allow the efficient development of
derivative products (Meyer and Lehnerd 1997). Mod-
ularization refers to organizing complex products by
decomposing them into simpler parts, so they can
be managed independently (Ulrich 1995). Supple-
mentary to product platform is product family, which
refers to a group of related products where product
derivatives can be obtained by customizing the com-
ponents of product family while maintaining the ba-
sic product platform.
The concept of a product platform and a prod-
uct family in the building industry can be approached
through industrialized building systems. By dividing
a building into two distinct parts, support (chassis)
and infill, John Habraken (1972) proposes a new con-
cept to support flexibility in building. While chassis
refers to the platform that constitutes permanent and
shared parts of the building, infill refers to the inter-
changeable elements of the building. The support
should be flexible enough to accommodate changes
in infill that could derive from social, cultural, or tech-
nical issues, and are intended to "accommodate and
outlast infill changes, to persist largely independent
of the individual occupants' choices, while accom-
modating changing life circumstances" (Kendall and
Teicher 2000, 33).
The flexibility of housing configuration is at-
tributed to its hierarchical structure, which in turn is
determined by how a house could be decomposed.
The definition of hierarchy enables the designer to
determine the design context and the relations be-
tween configurable elements, and in turn governs
the ways in which elements could be configured.
In dimensional customization, a building is de-
fined as an assembly of spaces. A building, P , is
defined by a set of spatial zones (modules), Z1 , Z2
,...,Zm each of which could also be comprised of sev-
eral subzones, Zi,1 ,Zi,2 ,. . .,Zi,j . A building is also
defined as a set of elements, E1 ,E2 ,. . .,Ep referred
to as customizable components, which are consid-
ered as members of spatial zones. A configuration
can be explicitly represented over the components'
attributes reflecting in the form of dimensions, which
are associated with building elements. The same set
of elements can yield different configurations when
dimensions of elements are modified. This hierarchy
of a building is used as a fundamental basis for clas-
sifying variations in all sorts of dimensional configu-
rations.
The definition of hierarchical structure is imple-
mented through the "Zone Definition" function, and
an overall hierarchical structure of zones is repre-
sented through the "Configuration List", so they can
be selected, providing access to the elements of the
selected zones for the configuration process.
Geometric and Dimensional Relation
A dimensional relation is an explicit representation of
a geometric object's placement relative to others. In
the dimensional customization system, the definition
of a dimension is a bilateral relation that determines a
distance between two elements placed axially along
X and Y directions.
Identifying a dimensional relation is highly con-
BIM | Applications - Volume 2 - eCAADe 34 | 637
 tingent upon logical properties of a geometric re-
lation supported by internal functionality of Re-
vit®. Thus, the interchanging of elements is sup-
ported through three types of geometric relations
between elements: parallel, perpendicular and axial
(i.e. aligned). Figure 1 illustrates six scenarios sup-
ported by Revit®, where the first two show uncon-
nected elements, and others show geometrical rela-
tions among connected elements. The transforma-
tion of an unconnected element (image i and ii) does
not make the other elements move. Maintaining the
topological relation of connected elements is held by
joining geometry. When elements intersect, Revit®
creates a join by default, in that one of the joined
elements cut the other and a union between two
shapes is created; elements connected at the time of
transformation. If two elements are connected ax-
ially at their edges (image iii), Revit® automatically
joins them, eliminates any edges, and makes them a
single element. In L-shape connection, where two el-
ements are connected on the basis of two edges, the
transformation of either element involves changes
in the length of the other. In T-shape connection,
where two elements are connected based on one
edge, the transformation of horizontal element, A, in-
volves changes in the length of vertical element, B,
(image iv). If three elements (or more) are connected
along their edges, only two elements can get joined
by Revit®. If two axial elements are joined, they be-
come a single element and act as T-shape connec-
tion between two perpendicular elements. Joining
between two perpendicular elements, however, acts
as L-shape connection, but without any relation to
the third element (image v). In that case, not only
the third element is not joined, but also it is totally
detached from others, and therefore the transforma-
tion of elements results in topological violation.
Constraints in Design
There have been several researches exploring the
essence of constraints in architectural design. A con-
straint specifies relations that must be maintained.
According to Kolarevic (1993, 61), constraints are de-
638 | eCAADe 34 - BIM | Applications - Volume 2
fined "as user-defined dependencies or relationships
among one or more entities that are maintained by
some kind of constraint satisfaction mechanism."
Applications of constraints can be distinguished
between constraint solving and constraint checking
approaches. The constraint solving method uses
constraints as a design driver to explore validated so-
lutions. This method is typically used when a gener-
ative approach to design is applied, where architects
can encode design rules that can be then explicitly
recorded as constraints. In this approach, the gener-
ated solutions fully meet the design constraints, al-
though the customer has no influence in the design
process. Constraint checking, on the other hand, al-
lows for customer participation. What makes con-
straint checking compelling in the customization of
housing is that it offers a promise of automatic design
verification to see if individualized solutions meet
all the constraints, and adjusts the solutions if nec-
essary. For example, Niemeijer (2011) explores the
application of automatic constraint checking in the
building industry for simplifying communication be-
tween the customer and the design. The system pro-
totype deals with entering and checking constraints
on an imported IFC model created by commercial
CAD packages. Embedding design codes and laws,
the system checks the model to determine if it meets
all the constraints, and adjusts it if necessary. Con-
straint checking is an appropriate approach in mass
customization as it allows the customer to interact
with the customizable house, while compliance with
constraints prevents design errors and conflicts at an
early stage of the design process.
Dimensional Constraints. Autodesk Revit® supports
dimensional constraints in two discrete environ-
ments, the conceptual mass and the main model-
ing. The former, conceptual mass environment, al-
lows a user to create 3D complex geometries and
define dimensional constraints in the form of para-
metrical equations, and the latter supports geomet-
rical modeling in the BIM environment. The concep-
tual mass environment, however, is not supported
by BIM technology. Similar to traditional CAD tools,
 this environment does not govern non-geometrical
information used to create a real building such as
price, fabrication and schedules. The created geome-
try could then be exported to the main modeling en-
vironment, but that results in the loss of dimensional
constraints. The Revit® BIM environment supports a
limited range of dimensional constraints; first, there
is a lock constraint, which is applicable over aligned,
linear and angular dimensions; second, the equality
constraint determines equal length over several di-
mensions; and third, the level constraint references
the top and the bottom of an element to a horizon-
tal plane whose height can be controlled by the user.
The challenge is yet to define a set of dimensional
constraints in the form of parametrical representa-
tion.
Based on the research in the prefabricated hous-
ing industry, the most commonly used dimen-
sional constraints have been identified and imple-
mented in the system: CF ix , CGreaterT han ,
CLessT han , CBetween , and CM tp . First, the
fixed constraint, CF ix , ensures that the dimensional
value v, as a distance between two elements, re-
mains always fixed, as represented in the equa-
tion {v = vf ix }. What makes this constraint dif-
ferent from Revit's lock constraint is that it also lim-
its the movement of the selected element. With
the lock constraint, if an element is moved, the
other element moves too, maintaining a fixed dis-
tance. With the fix constraint, neither of elements
can be moved, and if they do, the system would
return them back to the position where they were.
The other two constraints, CGreaterT han and
Figure 1
Six types of
building element
relations.
CLessT han , enable customers to explore an infi-
nite number of possibilities, with dimensions that
are either equal/higher or equal/lower than a value,
{v ≥ vmin } or {v ≤ vmax }, respectively. The fourth
constraint, CBetween , ensures that the dimensional
value v is always within a particular range of mini-
mum and maximum values, or {vmin ≤ v ≤ vmax }.
The last constraint, CM tp , ensures that a dimen-
sional value, v, is always a multiple of an increment
value, as shown in {v⁄vmtp ∈ N>0 }. The applicabil-
ity of such a constraint is determined by the limita-
tions of manufacturing systems in the prefabricated
housing industry. Each element can have multiple di-
mensions assigned to it, each of which can only be
associated with a single constraint. Each constraint
C involves some subset of the variables and specifies
the values these variables can simultaneously take.
The definition of dimensional constraints is im-
plemented on the basis of two steps: first is the
"Dimension Labeling" function, where the designer
needs to apply desired labels to dimensions, and sec-
ond is the "Constraint Definition," which enables the
designer to apply dimensional constraints. Dimen-
sions are not interrelated and each of those can only
get associated with a single constraint. Figure 2 illus-
trates an example of a dimension labeling panel and
constraints definition dialog box, in that the first col-
umn (parameter) shows dimension names associated
with the selected element, the second column (con-
straints) enables the designer to select among five
constraints, and the last column (value) is where the
parametrical values are entered.
BIM | Applications - Volume 2 - eCAADe 34 | 639

Constraint Satisfaction
The implementation of automatic design validation
is based on the constraint satisfaction problem (CSP)
paradigm. According to a general definition of CSP,
a model is composed of a set of dimensional param-
eters, D1 , D2 , D3 . . . Dm , each of which has a do-
main of possible positive values (Di ) and sets of con-
straints (Ck ) which specify the allowable combina-
tion of values (Russell and Norvig 1995). In that fash-
ion, given a set of dimensions and constraints, the
problem is finding the dimensional values associated
with building elements (walls, doors, and windows)
that could satisfy all dimensional constraints.
Constraint satisfaction is implemented based on
linear equations using numerical algorithms fulfill-
ing the reconfiguration process. The propagation of
changes is performed through recursive evaluation
of constraints and the dynamic update algorithm to
arrive at validated solutions.
Reconfiguration
The reconfiguration of building elements can be sup-
ported in two ways: numerical manipulation of di-
mensional values and interactive manipulation of
building elements. Both types of reconfiguration re-
quire pre-selection of a single element over which
the changes are applied. In the first method, once
an element is selected, the system activates the as-
sociated dimensions of the selected element. A user
can then easily change dimensional values numeri-
cally. The second method allows interactive reconfig-
uration within the model by two types of Euclidean
transformations of translation and stretch. Once an
640 | eCAADe 34 - BIM | Applications - Volume 2
element has been selected, the former allows a user Figure 2
to simply relocate a building element while the latter Two steps of
enables the user to change the length of an element dimensional
at the points of free edge. constraints
The process of manipulation is implemented definition: a)
based on changes to a single element at the time. Dimension Labeling
The geometrical reconfiguration, however, can be Panel and b)
limited by several dimensional constraints, due to the Constraints
fact that an element can be associated with multiple Definition Dialog
dimensions. Thus, the overall geometrical reconfigu- Box.
ration is the result of constraint satisfaction involving
all dimensions associated with the selected element.
In the reconfiguration process, constraint expla-
nation and solution recommendation are essential to
derive at valid configurations. If there is a constraint
violation, the system simply refuses to carry out a
transformation that results in conflicts. Dealing with
conflict resolution, the user could be overwhelmed
by the complexity of reconfiguration process. Build-
ing on that line of thinking, the dimensional cus-
tomization system provides for the explanation of
constraint violation through a notification message
in the natural language, and then recommends a sat-
isfactory solution complying with the constraints.
TEST CASE
For the test case, we used a typical two-story house.
The model was created in Autodesk Revit® and then
integrated with the customization toolkit. The build-
ing is divided into six zones, with a total of 15 sub-
zones. The first floor includes three zones. The first
zone consists of a garage; the second zone consists
of a den, powder room, stairs and hallway; and the
third zone includes a living room, dining room and
a kitchen. The second floor includes three zones.
The first zone consists of a bedroom, laundry room
and a bathroom; the second zone consists of a bed-
room, stairs, hallway and a storage space; the third
zone consists of a master bedroom, en-suite bath-
room and a walk-in closet. The perimeter bound-
ary of the building is limited by the site conditions
to 12x20 m2 . Figure 3 illustrates the 3D representa-
tion of the model and the 2D representation of the
 floor plans, consisting of 125 dimensional parame-
ters and constraints defined in both the first and sec-
ond floors. Figure 4 illustrates five customized de-
sign solutions explored through the system. Interior
appliances are placed in the model during the post-
processing configuration.
Figure 3
3D Representation
of the case study
model and
associated zones.
Figure 4
Five geometrically
customized
housing solutions
complying with
dimensional
constraints.
DISCUSSION AND CONCLUSION
This paper presents an ongoing research project that
explores the complexities associated with the appli-
cation of mass customization in the housing indus-
try, with a particular focus on dimensional customiza-
tion. The apparent disparity between heterogeneous
customer demands and a rather homogenous hous-
ing fabric provides an impetus to search for a dimen-
sional customization system that could allow for the
design of one-of-a-kind houses. Homebuyers and
homebuilders alike are "captivated" by the difficulties
associated with housing customization, such as bur-
den of choice selection and design validity of individ-
ualized solutions.
The proposed constraint-based dimensional cus-
tomization system should provide greater customer
satisfaction by allowing customers to have a much
greater influence on the overall design of their
houses, including the variation of key dimensions as-
sociated with the exterior and interior of the house.
Constraints may be seen as a limitation, but over the
course of design-configuration-validation, it offers a
promise of ensuring design viability, i.e. that a de-
signed house is affordable, producible, and safe.
The potential benefits of BIM technology have
been proven in the mass customization, but none
of the architectural design tools have been explic-
itly designed for this purpose. As a consequence,
there are several technical impediments in imple-
menting customization features within existing tools.
Two major issues have been identified in Autodesk
Revit®. The first is topological violation; Revit® does
not maintain connectivity (the "join" constraint) be-
tween more than two elements connected at a sin-
gle edge. The second flaw is editing the footprint of
a roof; specifically, a pitched roof is supported differ-
ently from the walls. Revit® does not provide an API
to edit the roof boundary, and its changes can only be
performed by mapping model lines in an edit mode.
Thus, dimensional customization is tightly limited by
the inherent characteristics of Revit® and the result-
ing issues cannot be addressed using the existing
API. Dealing with such Revit® difficulties implies the
necessity of either offering customers limited flexibil-
ity for design customization or having the designer
perform post-processing purification in the design.
Future work will focus on understanding social
and cultural challenges that influence customer de-
cision making. Allowing the customer to be more in-
volved in the design of individualized houses, while
dealing with various technical difficulties that may
emerge, raises questions about the acceptance of
such a system by the population at large. In co-
creation activity, the professional designer is re-
placed by the customer who may not have the suffi-
cient domain knowledge and the necessary skills and
confidence to perform the (re)configuration tasks.
There seems to be a trade-off relation between a
perceived value that a customization system could
provide and its perceived complexity. The customer
might be faced with either the risk of uncertainty
or dissatisfaction; in both scenarios, the role of the
BIM | Applications - Volume 2 - eCAADe 34 | 641
 designer is crucial to ensure that not only the cus-
tomized solutions are valid, but also that the process
of housing customization brings value to the cus-
tomer. As argued at the beginning, affective factors
associated with the design experience of customiza-
tion could potentially bring a higher valuation to the
entire process from the perspective of the customer.
Additional research is needed to better understand
customer behavioral intentions when using the di-
mensional customization system and how customer
attitudes are shaped to close the gap between cus-
tomization, validation, and efficiency.
REFERENCES
Davis, S 1987, Future Perfect, Basic Books
Duarte, J 2001, Customizing Mass Housing: A Discursive
Grammar for Siza, Ph.D. Thesis, Massachusetts Insti-
tute of Technology
Franke, N and Piller, F 2004, 'Value Creation by Toolkits for
User Innovation and Design: the Case of the Watch
Market', Journal of Product Innovation Management,
21(6), pp. 401-415
Garber, R 2014, BIM Design, Realizing the Creative Poten-
tial of Building Information Modeling, Wiley
Habraken, J 1972, Supports: An Alternative to Mass Hous-
ing, Architectural Press, London, UK
von Hippel, E 2001, 'User Toolkits for Innovation', Jouranl
of Product Innovation Management, 18(4), pp. 247-
257
Hofman, E, Halman, J and Ion, R 2006, 'Variation in
Housing Design: Identifying Customer Preferences',
Housing Studies, 21(6), pp. 929-943
Huffman, C and Kahn, B E 1998, 'Variety for Sale: Mass
Customization or Mass Confusion?', Jouranl of Retail-
ing, 74(4), pp. 491-513
Kendal, S H and Teicher, J 2000, Residential Open Building,
Routledge
Kieran, S and Timberlake, J 2003, Refabricating Architec-
ture, McGrow-Hill
Kolarevic, B 1993, Geometric Relations as a Framework for
Design Conceptualization, Ph.D. Thesis, Harvard Uni-
versity
Kolarevic, B 2015, 'From Mass Customisation to De-
sign ‘Democratisation', Architectural Design, 85(6),
pp. 48-53
Kramer, T 2007, 'The Effect of Measurement Task Trans-
parency on Preference Construction and Evalua-
tions of Personalized Recommendations', Journal of
642 | eCAADe 34 - BIM | Applications - Volume 2
Marketing Research, 44(2), pp. 224-233
Lampel, J and Mintzberg, H 1996, 'Customizing Cus-
tomization', Sloan Management Review, 38(1), pp.
21-30
Larson, K, Tapia, MA and Duarte, JP 2001, 'A New Epoch-
Automated Design Tools for the Mass Customization
of Housing', A + U Arachitecture and Urbanism, 366,
pp. 116-123
Meyer, M H and Lehnard, A P 1997, The Power of Product
Platform, The Free Pass, New York
Nahmens, I and Bindroo, V 2011, 'Is Customization Fruit-
ful in Industrialized Homebuilding Industry?', Jour-
nal of Construction Engineering and Management,
137(12), pp. 1027-1035
Niemeijer, R A 2011, Constraint specification in architec-
ture : a user-oriented approach for mass customiza-
tion, Ph.D. Thesis, Technische Universiteit Eindhoven
Piller, F, Moeslein, K and Stotko, CM 2004, 'Does Mass
Customization Pay? An Economic Approach to Eval-
uate Customer Integration', Production Planning and
Control, 15(4), pp. 435-444
Pine, JB 1993, Mass Customization: the New Frontier in
Business Competition, Harvard Business Press
Russell, S and Norvig, P 1995, Artificial Intelligence: A
Modern Approach, Prentice-Hall, Egnlewood Cliffs
Salvador, F, de Holan, PM and Piller, F 2009, 'Cracking
the Code of Mass Customization', MIT Sloan Manage-
ment Review, 50(3), pp. 71-79
Simonson, I 2005, 'Determinants of Customers' Re-
sponses to Customized Offers: Conceptual Frame-
work and Research Propositions', Journal of Market-
ing, 69(1), pp. 32-45
van den Thillart, CCAM 2004, Customized Industrializa-
tion in the Residential Sector, Mass Customization
Modelling as a Tool for Benchmarking, Variation and
Selection, SUN, Amsterdam, Netherlands
Thomke, S and von Hippel, E 2002, 'Customers as Inno-
vators', Harvard Business Review, 80(4), pp. 74-81
Thornton, R 2004, The Houses that Sears Built, Gentle
Beam Publications, Alton, IL
Toffler, A 1990, Future Shock, Random House, New York
Ulrich, K 1995, 'The Role of Product Architecture in Man-
ufacturing Firm', Research Policy, 24(3), pp. 419-440
Zipkin, P 2001, 'The Limits of Mass Customization', MIT
Sloan Management, 42(3), pp. 81-87
[1] http://www.re4a.com
[2] http://www.livinghomes.net
[3] http://www.bluhomes.com
Tracking Changes in Buildings over Time
Fully Automated Reconstruction and Difference Detection of 3d Scan and
BIM files

Martin Tamke1 , Mateusz Zwierzycki2 , Henrik Leander Evers3 ,

Sebastian Ochmann4 , Richard Vock5 , Raoul Wessel6
1,2,3
The Royal Danish Academy of Fine Arts, Schools of Architecture, Design and
Conservation 4,5,6 University of Bonn - Department of Computer Science
1,2,3
{martin.tamke|mzwi|heve}@kadk.dk
4,5,6
{ochmann|vock|wesselr}@cs.uni-bonn.de

Architectural and Engineering Communities are interested in the detection of

differences between different representations of the same building. These can be
the differences between the design and the as-built-state of a building, or the
detection of changes that occur over time and that are documented by consecutive
3D scans. Current approaches for the detection of differences between 3D scans
and 3D building models are however laborious and work only on the level of a
building element. We demonstrate a novel highly automated workflow to detect
differences between representations of the same building. We discuss the
underlying tools and methods and the ways to communicate deviations and
differences in an appropriate manner and evaluate our approach with a rich set
of real world datasets.

Keywords: 3d scan, BIM, Machine learning, Point Clouds, Big Data

INTRODUCTION: WHY ARE CHANGES IN
BUILDINGS OBSERVED?
The monitoring of the progress on building sites, the
control of geometric deviations between planned
and as-built geometry and the check of planarity of
building elements during construction planning are
essential tasks in AEC. Further applications are lo-
cated in the area of Facility Management and plan-
ning of renovation tasks, where the detection of dif-
ferences between a documented state and the cur-
rent state are a prerequisite for decision making and
design, and where the quality of a 3D model that rep-
resents the current state of the building has to be as-
sessed. The currently fastest and most precise way
to document the geometry of an existing building is
by means of Point Clouds, which are generated by
3d laser scanners or photogrammetric devices.Point
clouds consist of millions of unrelated coordinates.
This is diametrically opposed to BIM, the current state
of the art paradigm for architectural planning and fa-
cility management, which represents relational ob-
ject oriented data in a hierarchical structure.
Hence, finding differences between these two
types of datasets first of all necessitates to represent

BIM | Applications - Volume 2 - eCAADe 34 | 643

 them in the same digital model and the same 3D co-
ordinate system, i.e. to overlay point cloud data and
BIM objects or to create a way to semantically link
them.
Secondly it has to be understood, what types of
differences shall be detected among a set of repre-
sentations of the same building. Differences of inter-
est can theoretically be observed on any level of gran-
ularity in a building, from the extension of a build-
ing to small changes in surface ornament. A huge
search space is emerging, due to the complexity and
size, that is inherent to buildings and therefore as well
its representation. In order to filter and express rela-
tions between the many different elements the build-
ing profession has hence developed the concept of
scales. This concept is fundamental to the structuring
of Building Information Models, where a high level
description of a building shape is in depended scales
broken down into the smallest constituting parts and
relations between the scales can be expressed in mul-
tiple ways.
This scalar and relational understanding of parts
and whole provides a powerful approach to start and
structure the search for differences. It allows for a dis-
crimination between high level building related el-
ements such as walls, floors, ceilings, depended el-
ements, such as doors, windows and even smaller
parts and elements, which are not related, such as
furniture or other temporary elements. And it allows
to search for differences on one level of detail and
express found differences on a higher one, granting
overview of e.g. changes in the whole building, and
impact on related elements on a smaller scale in BIM.
This goes along the interest of professionals, who will
be interested in questions: which walls or windows
have been removed, were added or changed in di-
mension. Here it is important to acknowledge, that
the 3d scans are only capturing a certain moment in
time and that most building elements are static in
size and position. The "move" of wall by one meter, or
the change in the dimension of a window is in reality
a removal of one and an addition of another element.
An algorithm might hence observe, that a window is
644 | eCAADe 34 - BIM | Applications - Volume 2
at the same position but in a different size, or that a
similar sized wall is in close vicinity to a previous ex-
isting one. The essence of a difference detection is
concerned about missing or added elements.
Stakeholders will as well be interested in an
overview, where in a building these changes take
place and are eventually amassed, and where the
most severe ones take place (Figure 1). Other differ-
ences of interest, as on the level of building services
(Ducts, Pipes etc.), can as well be expressed on the
level of objects. And while a tracking of differences
of these is possible when they are exposed, image
based methods do not work, when they are hidden.
However, a professional might deduct that hidden el-
ements changed, when the geometries of related el-
ements, such as the wall they are enclosed in, change.
Figure 1
An an analysis
between a 3d scan
and a modelled BIM
reveals missing
elements (red) and
elements with
difference in size
(blue) and type
(window instead of
door or vice versus -
pink). Source: LE34.
We argue for a difference detection between build-
ing data, which takes point of departure in the most
common method of how this type of data is struc-
tured - as building elements. This allows for more ad-
vanced types of query across scales and provides an
effective filter for clutter (furniture, scanned persons
etc.), as all of these can not be associated to building
elements. For the difference detection we are look-
ing first of all, whether elements have been removed
or added between two or more representations of
a building. Related questions, whether an element
has been eventually moved or resized, whether it has
only be partially removed or how much a element de-
viates from a planned or previous state are steps for
subsequent detailed analysis.
Figure 2 prototypes (Tuttas 2014) use octree space partition-
The detection of ing on correlated Point Cloud and BIM models in or-
deviations to a der to create a search space, where differences can be
plane is a typical found.
use case, that takes Another principal approach (Tamke 2015), is to
place on the level of take a point of departure in the BIM model and search
a single building STATE OF THE ART iteratively in the correlated point cloud for points in
element. Here the Current approaches for a difference detection be-
vicinity to each surface of e.g. a wall (Figure 3). The
deviation of an tween Point Cloud data from photogrammetry (Tut-
visual representation of all remaining points provide
outdoor wall (left) tas 2014) or 3d Laserscan (Bosché 2010) and 3d ge-
a good overview and indicate areas that require fur-
to the vertical ometry in form of 3D CAD or BIM data concentrate
ther inspection. Elements, which are not present in
plane. The result is usually on the level of a single building element.
the BIM file can be detected (in the figure e.g. not
visualised in a color Related commercial implementations are avail-
modelled suspended ceilings in corridors).
gradient (right) and able in software such as CloudCompare (Cloud-to-
Both approaches are able to provide a general
the deviations are cloud distance), Bentley Pointools (Clash Detection
overview of differences and indicate elements, which
indicated as and Differencing), PointSense (Clash Detection, De-
are modelled in BIM, but miss or deviate substantially
numbers on an formation Analysis), Geomatic (Automated Accuracy
in the point cloud. They are however not able to dis-
overlaid coarse grid. Analyzer), ClearEdge3D EdgeWise (Quality Assurance
criminate sufficiently clutter and detect added ele-
Tool). Research indicates an extension of this ap-
ments, such as walls, nor do they take into account,
proach in order to identify e.g. the correct position-
that due to the use of vision based technologies
ing of a door (Anil 2011). Common to the workflows
building elements can be overshadowed by other el-
in all these implementations is, that users are tasked
ements.
to more or less manually correlate a Point Cloud and
BIM model of a building in a common 3D space, iden-
tify a single building element and start an algorithm,
Figure 3
which searches e.g. for the distance of points in
Difference
the Point Cloud to the elements. Upon completion
detection through
users receive information of differences and devia-
iteratively search of
tions (Figure 2) for this particular element. This work-
points in a point
flows is overall very laborious.
cloud, that are
Less attention was given to point-cloud- to -
outside a certain
point-cloud difference -detection. This area is cur-
distance from each
rently mainly investigated for the field of terraform-
BIM elements (red).
ing (Girardeau-Montaut 2005) or as general com-
WORKFLOW FOR DIFFERENCE DETECTION
puter science related problem in the case of real-time
3d scanning, where an example of an implementa- BETWEEN POINT CLOUDS AND BIM
tion is available in the PCL library. However these We propose a workflow to detect differences and de-
approaches do per se not answer the questions on viations between interior representations of build-
building element level, which stakeholders from the ings in BIM and point cloud formats which operates
building profession have. None of the commercial on the level of building elements. It's point of ori-
implementations is able to provide an overview of all gin lies in technologies developed in the EU funded
the differences, that have been detected and to ex- DURAARK project (Beetz et al. 2016), where tools
press the quality of these in numerical terms, as for were developed, which are able to reconstruct a BIM
instance in a measure of overall accuracy. Research model from point clouds (Ochmann 2016). These

BIM | Applications - Volume 2 - eCAADe 34 | 645

 tools reconstruct ca. 90% of the interior building ele-
ments found in BIM on LOD2 (Tamke et al. 2016) faith-
fully. We consider this amount sufficient for our re-
search prototype. Being able to reconstruct BIM from
point clouds provides an efficient means to compare
Point Clouds of the same building at different times
through the comparison on the base of a comparison
of the two derived BIM models.
Our approach towards the Difference Detection
of Point Clouds and BIM consists of four distinct steps:
1. Registration: Orienting the representations
(point clouds and BIM models)
2. Association: Subsets of points are associated
to either subsets of points or building el-
ements depending on whether the tool is
used for point-cloud-to-point-cloud or point-
cloud-to-BIM association. During this step,
unnecessary information such as furniture or
scanned people, is excluded from the analy-
sis. The focus is set on the building itself and
it has to be checked particularly for false neg-
ative detections.
3. Data Analysis: The difference between the
datasets is calculated
4. Refinement and user feedback: The results of
step 3 are verified and refined and qualitative
and quantitative feedback is provided to the
users.
In our approach we try to achieve a great extend of
automation and hence efficiency in each step, even-
tually trading in precision for speed.
Step 1: Automated registration of point-
cloud and BIM models
The registration process is roughly based on work by
Makadia et al. (Makadia et al. 2006) who achieve a
coarse pre-alignment using correlations between Ex-
tended Gaussian Images in the Fourier domain, and
subsequently perform a finer registration using the
Iterative Closest Point algorithm (ICP).
Taking the assumption that the "up" direction
is the z-axis and that most wall and/or structures
646 | eCAADe 34 - BIM | Applications - Volume 2
are respectively vertical or horizontal, we can re-
strict ourselves to a successive alignment in the two-
dimensional domain.
Vertical planes in both representations are ex-
tracted and their dominant normal directions are
determined by clustering the directions of projec-
tions of the extracted planes into the xy-plane. The
largest cluster of directions is then considered to be
the most dominant direction, and the second domi-
nant direction is constructed orthogonally to this di-
rection. In case of BIM models, the planes are ex-
tracted directly from the surfaces of wall entities; in
case of point clouds, a Random Sample Consensus
(RANSAC) plane detection is performed, see (Schn-
abel et al. 2007) for further details. The dominant
directions in both datasets provide hints for possible
rotational alignments of the datasets around the z-
axis, i.e. we assume that aligning the dominant direc-
tions to each other yields a valid rotational alignment
of the datasets, which sufficiently restricts the search
space in the rotational domain.
For each possible rotation around the z-axis, the
projections of vertical planes along the z-axis are ren-
dered as lines; one per building representation. Then
the alignment maximizing the overlap between the
two resulting line images is found using a Fourier
transform based approach. The computed image
alignment transformation is then transferred back
into world coordinates and applied to the building
representations. This results in an alignment of the
datasets in the xy-plane. For determining the transla-
tional alignment along the z axis, a similar approach is
used: Horizontal planes that potentially correspond
to floor surfaces and vertical planes are rendered into
an image, viewing the building representations from
the side. Aligning the images to each other then re-
sults in a translational offset which is applied to the
building representations (Figure 4).
The output of this first step is an RDF file (RDF
transformation file), which describes the transforma-
tion that is necessary to achieve the registration of
the two models.
Figure 4
Step 1: Automatic
alignment.
Different
representations
(here BIM shown in
cyan and
point-cloud shown
in blue) are
automatically
aligned in the three
dimensional space. Step 2: Association of point subsets to IFC posed as a graph labeling problem, which is solved
building elements using an energy minimization approach. Openings
Since our proposal for a difference analysis is based are detected and classified as doors and windows by
on building elements, we use a technique developed means of supervised learning.
in the DURAARK project (IFC reconstruction) in order An intermediate data structure of the process is
to detect building elements, such as rooms, floors, the representation of a building story as a halfedge
ceilings, walls, doors, and windows along with their graph in which faces are interiors of different rooms
relations in a Point Cloud (Figure 5). In the following and edges are walls between adjacent rooms. The
we will briefly outline this method, for further details neighborhood relation of rooms is also directly con-
we refer to (Ochmann 2016). tained in this structure, i.e. two rooms are neighbor-
ing if and only if they share a common edge. This rep-
Figure 5 resentation allows for fast queries to which high-level
The second step building element (room, wall, floor, ceiling) a given
associates Points to 3D point belongs. This allows us to associate all mea-
building elements sured points of a point cloud to the aforementioned
(room, wall, floor, building elements.
ceiling). The resulting BIM model is stored in IFC format.
Furthermore a RDF file is created. This RDF file holds
the information on relationships between a point
cloud and a BIM file. The association is expressed as
pairing of point cloud subsets with respective BIM ge-
ometry, and the UUID number of the particular ge-
ometry indicates the association. Points in the point
cloud, which are not associated to building elements,
can be determined by a simple boolean operation.
In a first step a coarse segmentation of the point
cloud separates rooms and outlier points, filters out Step 3: Data Analysis
clutter outside the building. Infinitely long vertical The aim of the data analysis is to detect and visu-
wall candidate planes are generated by means of alize the differences between two or more building
a RANSAC algorithm (Schnabel et al. 2007), which datasets. Our approach uses first of all the UUIDs
yields candidates for wall entities that are shared be- of building elements, which were either provided by
tween rooms. The determination of a globally plausi- the users as BIM file or were generated in the previ-
ble configuration of connected wall elements is then ous steps from point clouds. Specifically we are using

BIM | Applications - Volume 2 - eCAADe 34 | 647

 the generated RDF files. The associated point cloud Figure 6
data is in a subsequent step used for verification and Visualisation of our
refinement of the initial analysis. difference
To perform the Difference Detection, we have detection operating
developed a custom Grasshopper components li- on a point cloud
brary, which is based on the Volvox plugin for Rhi- and BIM model.
no/Grasshopper (Zwierzycki et. al 2016). Thanks to Clutter and outliers
Grasshopper's visual programing interface, the feed- are coloured pink,
back is instantaneous and detection of changes can identical elements
be visualized. All above takes place in a custom are grey, and
Grasshopper components library utilizing partially elements which
linked C++ APIs. exist only in one
Before the actual analysis, data from the RDF files model are colored
has to be interpreted. This takes place in the follow- We tackle the problem of analysing multiple files at red. Note that
ing process: once, again using set theory. The simplicity of this kinetic building
approach enables the user to create his/her own de- elements (door
1. Parsing of the RDF transformation file - this file
tection rules for the filtering of building elements on leafs) are not yet
contains transformation matrices assigned to
the fly. detected in this
UUID identifiers which reference either partic-
version of the
ular points in the point cloud or building ele-
Step 4: Refinement of detection and feed- association
ments in the BIM model.
back to user algorithm. Source:
2. Parsing of the RDF association file - the asso-
The output of the analysis in the previous step in- Rislökka.
ciation file contains a list of indices of points
from the point cloud, tagged with an UUID dicates whether an element is present or not in the
identifier which connects them with particu- other representation (Figure 6).
lar part of BIM geometry.
Figure 7
3. Transformation and splitting of the point
Point Cloud of a
clouds - based on the former files, the point
room which was
clouds collection can be oriented in space to
extended by taking
match the BIM model, after which it is split
down two walls.
into parts based on the RDF association file.
The red points
The interpretation of data is based on UUID identi- indicate the areas,
fiers of each element. The collection of the UUID where these were
identifiers is analyzed concerning building elements located and show
unique to each data set, building elements common as well, that the
for all datasets, building elements appearing in sev- removal was only
eral datasets etc. The analysis is performed on sets partial. An overview
of UUID identifiers using simple set theory logic. Not We find however that sometimes elements are only of the spatial
matching UUIDs for the same element in different partially scanned or that parts of a building element situation is shown
BIM files are identified through a control of geometric are removed. Figure 7. shows for instance the case in Figure 8.
position and dimension. of walls, which have been removed except a part in
Given the association data of multiple point the top, which remains as ceiling joist. In this case
clouds, it becomes necessary to compare all of them points can still be associated to the building element
at once to get a better overview of the final situation. in the BIM representation and the algorithm would

648 | eCAADe 34 - BIM | Applications - Volume 2

indicate, that the element didn't change. We use ad-
ditional data, which is delivered with the RDF file in
order to refine the difference detection. The RDF as-
sociation file contains information on the amount of
points which theoretically should be associated with
each of the BIM elements - the number of potential
points scanned in perfect conditions. Comparing this
measure with the actual number of points referred to
at building element, allows us to deduct parts of the
building which are missing from the point cloud. This
also reveals the improperly scanned (obstructed) el-
ements of the building. In these cases it can be in-
dicated to the user, that a further visual inspection is
required.
It is important to note that some changes can
be improperly assigned to the point cloud, as caused
by the noise in the data. Noise refers here to the
instrument-inducted noise (whose value is given by
the manufacturer) as well as points reflected from
furniture or other non-related objects. It is neces-
sary to perform a qualitative analysis of the associa-
tion to identify its errors. By randomly sampling the
point cloud and measuring the closest point distance
to the associated geometry, we obtain the values of
standard and average deviation. The data is com-
piled as a set of histograms (one for each associated
pair), which serves as an assistance for an eventual in-
spection.
EVALUATION OF THE DIFFERENCE DETEC-
TION
The functioning of the approach was positively eval-
uated on 3 real world data sets, of which one is de-
scribed in the following paragraph. This set of data
has one characteristic change to detect - the removal
of two walls. A special difficulty is here, that only a
part of the wall is removed (Figure 8), while a ceiling
joist remains. It contains the ground truth (IFC file), a
1st scan (Time 1), which shows the same state of the
building as the ground truth and two scans (Time 2
and 3) acquired over an extended period of time after
the removal of the walls and finally an IFC file which
includes additions and removals of elements. The al-
gorithm detected the change from Time 1 to 2 truth-
fully and indicated no changes between Time 2 and
3.
The previous evaluation demonstrates the valid-
ity of the approach. We extended the evaluation and
tested, whether the approach is able to detect mul-
tiple changes (additions and removal of elements)
over multiple files. For this we prepared a set of syn-
thetic IFC files, which have the described characteris-
tics in terms of changes. We execute our difference
detection (Figure 9) and all changes are correctly de-
tected.
DISCUSSION AND CONCLUSION
The rapid advance in 3D scanning and point cloud
generating technologies as well as the proliferation
of BIM in all areas of the building profession creates
the foundation for a new information driven prac-
tice. It is based on data from sensors which provide a
representation of the real world in almost real-time,
and from linked databases, which hold the many
sources of information that describe a building's per-
formance, design intent, and specifications on all lev-
els.
But with the vast amount of data that is accessi-
ble for institutions, architects, and engineers, comes
an issue of data management and analysis. Various
software tools developed in the DURAARK project try
to simplify and automate currently tedious tasks. In
this paper we presented the IFC reconstruction and
Difference Detection algorithms and their implemen-
tation, in our case Grasshopper. The evaluation has
proven the principal usability of the approach. An
evaluation with further real world datasets should
take place. A difference detection between different
states of a building represented in Point Clouds on
the base of reconstructed BIM files might sound tau-
tologic in the first place. The use of the tool in prac-
tice examples demonstrated the advantages of the
approach especially under four aspects:
1. A point cloud can be easily compared with ex-
isting BIM files.
2. Clutter in 3d scans, such as furniture and
BIM | Applications - Volume 2 - eCAADe 34 | 649
 Figure 8
A ground truth IFC
model (left) is
compared with 3
point clouds
acquired over an
extended period of
time. The removal
of two walls
(highlighted) is
properly detected
in the second and
third scan.

Figure 9
Automatic analysis
of changes
between 3 IFC
model of the same
building. The
ground truth (Time
0) is shown on the
left. Removed
elements are drawn
in red, while
additional elements
are drawn in green.
scanned persons, is effectively filtered. Vice 4. Indications for further search are provided:
versa it provides a process of elimination of all Being able to overlook a whole model allows
points that are with a high probability identi- users to identify critical areas, that might need
cal between datasets and hence a smaller set further inspections.
to search for false negative etc.
3. Most importantly a hierarchy for the search is Some issues need to be addressed in the future:
provided, which follows the common archi-
tectural understanding of building elements. • The Difference Detection tool utilizes the data
The approach provides a mean to get an from the Association and Registration pro-
overview of differences on the level of the cess to highlight the differences between data
whole building, while a detailed investigation sets. The quality of the difference detection is
on the level of the element (for instance by therefore directly linked to the quality of the
means of a histogram of the distance between input data. The errors, which were exposed in
points points in certain distance to the build- the Difference Detection algorithm can be di-
ing element) can provide numerical values to rectly traced back to the input from the previ-
assess overall accuracy. ous algorithms.
• The association of points to building elements

650 | eCAADe 34 - BIM | Applications - Volume 2

 is the result of the preceding IFC reconstruc-
tion. The underlying algorithm detects cur-
rently only a narrow set of building elements.
This should be extended to more building el-
ements, like columns, heaters or any other in-
frastructural part in a building. The recog-
nition or filtering of these would ease differ-
ence detection as noise is reduced. Addition-
ally, the algorithm is currently limited to pro-
cessing each story individually without taking
their linkage via staircases or elevator shafts
into account. The ability to differentiate really
complex and data heavy point clouds of com-
plete buildings remains a future task. A fur-
ther possible extension for IFC reconstruction
would be to recognize non-planar elements,
like curved walls or arches.
• The faithful import of IFC files into the BIM
modelling environment (Rhino with Visual
Arq plugin) proved to be a challenge. This
might be caused by the plugin, but points def-
initely at ambiguities inherent to the IFC for-
mat.
The approach presented in this paper allows to man-
age and process big data efficiently, further research
is however required on fast and efficient noise re-
moval, object recognition and the general precision
in the point cloud processing.
ACKNOWLEDGMENTS
The research was funded by the European Commis-
sion within the FP7 framework Programme (Grant
Agreement no 600908). We thanks the collaborating
companies, that provided DURAARK with a rich set
of point clouds and BIM files. For this paper espe-
cially LE34 in Copenhagen and Plan3D in Berlin. All
datasets are available on data.duraark.eu [1].
REFERENCES
Anil, E, Tang, P, Akinci, B and Huber, D 2011 'Assess-
ment of Quality of As-is Building Information Models
Generated from Point Clouds Using Deviation Anal-
ysis', Proceedings of the SPIE Vol. 7864A, Electronics
Imaging Science and Technology Conference (IS\&T),
3D Imaging Metrology
Besl, P and McKay, H 1992, 'A Method for Registration of
3-D Shapes.', IEEE Transactions on Pattern Analysis and
Machine Intelligence, 14(2), pp. 239-256
Bosch\'e, F 2010, 'Automated Recognition of 3D CAD
Model Objects in LAser Scans and Calculation of As-
built Dimenstions for Dimensional Compliance Con-
trol in Construction', Advanced Engineering Informat-
ics, 24(1), pp. 107-118
J, Beetz, I, Blümel, B, Fetahui, U, Gadiraju, M, Hecher,
T, Krijnen, M, Lindlar, M, Tamke, R, Wessel and R,
Yu 2016, 'DURAARK: Enrichment and Preservation
of Architectural Knowledge', in S, Münster, M, Pfarr-
Harfst, M, Ioannides, P, Kuroczyński and E, Quak
(eds) 2016, How to manage data and knowledge re-
lated to interpretative digital 3D reconstructions of
Cultural Heritage?, Springer LNCS, Vienna
Makadia, A, Patterson, AI and Daniilidis, K 2006, 'Fully Au-
tomatic Registration of 3D Point Clouds', Cvpr, 1, pp.
1297-1304
Ochmann, S, Vock, R, Wessel, R and Klein, R 2016, 'Auto-
matic reconstruction of parametric building models
from indoor point clouds', Computers \& Graphics, 54,
pp. 94-103
Schnabel, R, Wahl, R and Klein, R 2007, 'Efficient RANSAC
for point-cloud shape detection', Computer Graphics
Forum, 26(2), pp. 214-226
Tamke, M, Edvardsen, DF, Beetz, J, Evers, HL, Krijnen, T,
Hecher, M, Zwierzycki, M, Panitz, M, Wessel, R, Vock,
R, Ochmann, S and Gadiraju, U 2016, DURAARK Eval-
uation Report D7.4, DURAARK
Tamke, M, Evers, HL and Ochmann, S 2015, DURAARK SME
Use Case - Design and Retrofitting D7.2, DURAARK
Tuttas, S, Braun, A, Borrmann, A and Stilla, U 2014 'Com-
parision of photogrammetric point clouds with BIM
building elements for construction progress moni-
toring', International Archives of the Photogrammetry,
Remote Sensing and Spatial Information Sciences - IS-
PRS Archives, pp. 341-345
Zwierzycki, M, Evers, HL and Tamke, M 2016 'Parametric
Architectural Design with Point Clouds - Volvox', Pro-
ceedings of the 34th eCAADe Conference, Oulo, Fin-
land
[1] http://data.duraark.eu
BIM | Applications - Volume 2 - eCAADe 34 | 651
Toward Space Oriented BIM Practices
Aida Siala1 , Najla Allani2 , Gilles Halin3 , Mohamed Bouattour4
1
ENAU/ MAP-CRAI 2 ENAU 3 MAP-CRAI 4 ENAU
1,3
{siala|halin}@crai.archi.fr 2 najla.allani@gmail.com
4
med.bouattour@hotmail.fr

When performing design tasks, architects think in terms of space, and act on
physical elements. They often use various representation means to shape and to
communicate the complex aspects of space. Architectural representation is often
driven by visual perception whilst current BIM practices seem to be based on
semantics associated with scheduling building items (element, position, quantity,
etc.). The reduction of architectural sensitive approaches to merely technical
ones, reveals only quantitative and restrictive information that does not reflect the
architect's multi-sensorial experience. This paper examines some recent model
proposals which include descriptions of architectural space concept, and tries to
suggest a possible synthesis of this work. It focuses on cooperative practices
necessary to unveil the sensitive dimension of the architectural design, and
presents a state of existing BIM tools based on relevant tasks used in these
practices in order to acquire more knowledge about the concepts which ensure a
cooperative work taking into account the sensitive spatial aspects.

Keywords: Cooperative design, Architectural space, BIM, Qualitative property,

Topology

INTRODUCTION to qualitatively describe a space. It plays a leading

The overall sensation of a space cannot be under-
stood by an analysis of its measurable properties. The
overall ambiance of a space is given by all the inter-
connected computational factors of which it is com-
posed (Drozd 2009). Shadow, light, transparency,
depth, sound, smell, heat, airflow, materials, tex-
tures, etc., are all elements that simultaneously par-
ticipate in discovering and perceiving a space (Bon-
naud 2012). Referring to physical experience, it
is observed that architecture engages and assem-
bles a large number of sensory dimensions: light-
ing and thermal ambiance, openness and sunshine
of a space, etc. Topology is also an important aspect
role in evoking feelings in space (ex. communicat-
ing spaces, space with exterior relation, etc.). The
perception of ambiance induces a questioning about
the representation of sensitive spatial aspects. Ac-
cording to Mark Crunelle, it is complex to translate
through visual features all perceptions provided by
our various senses (Crunelle 1996). That is why, ar-
chitects always used representations evoking the ob-
servers' sensations to reveal the intended ambiance
of designed spaces.
BIM models are defined as a consolidated base
of parametric objects organized into multiple layers
of information describing the physical and functional

BIM | Applications - Volume 2 - eCAADe 34 | 653

 characteristics of a building (BIM handbook 2014).
BIM modelling introduces a technical approach into
digital cooperative practices. Such an approach is
based on the use of measured standards and codi-
fications, transforming all building information into
quantitative data. Only constructive information is
taking into account during the design phases. How-
ever, a sensitive space approach is not yet integrated.
Visual representations used in BIM tools do not al-
low the user to recognize all elements properties ex-
isting in BIM models. BIM models provide informa-
tion about spaces that not include the relationship
between them (Brodeschi 2015). Current BIM prac-
tices cover limited points of view that cannot appre-
hend all the architectural knowledge (form and func-
tion) expressed in the philosophical discourse (Bhatt
and Kutz 2012).
In this study, we discuss the sensitive space ap-
proach during architectural design phases. The first
chapter focuses on recent research having addressed
data modelling of architectural space concept. The
second chapter looks at relevant tasks enabling col-
laborating on spaces, taking into account their sensi-
tive aspect. Then, the third chapter explores the ca-
pability of current BIM tools to realize such tasks. Fi-
nally, we try in the last chapter to formalize in a first
data model all core concepts allowing taking into ac-
count qualitative and topological information about
spaces. Our long-term objective is to propose an al-
ternative approach based on these types of sensitive
information to assist cooperative work during the de-
sign phases.
BACKGROUND
Architectural space is often considered as a construc-
tion result. An empty element bounded physically
or virtually, in which we can experience a feeling of
enclosedness. The generally accepted perception of
space within science today is as a relation between
things (Ekholm 2000). A building is constructed to
provide a set of interior and exterior spaces. It is de-
signed on the basis of a specific program designat-
ing the properties of those spaces (shape, location,
654 | eCAADe 34 - BIM | Applications - Volume 2
ambiance, openness, sunshine, relation and commu-
nication between, etc.). Most of existing space mod-
els treat only the intrinsic properties of a space (ex.
geometric data, location, quantity, etc.) and abstract
properties related to human experience. The IFC
model supports, in part, this type of information. It
contains data describing the quality of space includ-
ing:
• Thermal requirements (required temperature,
minimum and maximum values depending
on the season, etc.).
• Lighting requirements (lighting values,
whether the space requires artificial lighting
(true or false), etc.).
• Occupancy requirements (activity type, head-
room required for the activity assigned, num-
ber of users, areas rate by users, daily duration
of the activity, whether an external view is de-
sirable, etc.)
• Safety requirements (fire risk factor, whether
the space is intended to serve as a storage of
inflammable material, or as an exit, etc.)
• Space covering requirements (nature and
thickness of floors, walls, ceilings, etc.).
However, this model does not take into account the
rest of qualitative spatial requirements, like the open-
ness of a space, its sunshine, its acoustic ambiance,
etc.
Proposing a method for automatic updating of
spatial requirements during the design phases, Kim
specifies two types of space requirements: activity-
based requirements and space-based requirements
(Kim 2015). These requirements cover only spatial in-
formation supported by the IFC model and do not in-
clude requirements related to the topology of spaces.
Topological requirements are usually requested by
the future user of the building, focusing mainly on
information about the boundary and the neighbour-
hood of spaces. E.g. The openness of a space to the
outside, its sunshine, its orientation and its relation-
ship with the rest of surrounding spaces (proximity,
communication, container / content, etc.).
Based on an analysis of the conceptual model of
spaces, space boundaries and enclosing structures
advanced by Bjork (Bjork 1992), Ekholm and Fridqvist
present a modelling for spaces including descriptions
of their topology in a building (Ekholm and Fridqvist
2000). Using a spatial view, they introduce the con-
cept of user organization to conceptually separate
between the intrinsic properties of the built envi-
ronment and the properties of human activities that
are to be housed within. But there is no reflection
in the two previous models that addresses the de-
sign requirements for the topology of spaces or for
their qualitative properties related to the user activ-
ity. Among existing recent models, it is only the work
of Bhatt who takes into account, implicitly, topolog-
ical requirements (Bhatt 2011). It proposes a spa-
tial modelling based on modular ontology, includ-
ing both qualitative and topological requirements, to
verify compliance of the spatial properties proposed
during the design phases with regulatory require-
ments. E.g. the range space of a sensory device, the
distance between emergency exits in public spaces,
etc.
We have just shown that spaces are taken into
account, in existing data models, particularly in the
IFC model as a set of quantitative and topological
properties. Quantitative properties can be guided
by qualitative requirements. Currently addressed
qualitative requirements cited above are related to
designers' field of intervention. E.g. MEP engi-
neer determines the thermal requirements, electri-
cal engineer determines the lighting requirements,
etc. However, the user requirements are given in
terms of further qualitative properties (e.g. an open
and sunny living room communicating with a large
kitchen). Hence, information about qualitative and
topological spatial requirements must be present in
a useful model for design, checking and validation
of spaces during the teamwork evolution. In our
work, we specifically focus on qualitative and topo-
logical requirements expressed by the future user of
the building (the owner). In continuity with the work
of KIM, this study aims to integrate these types of re-
quirements in current BIM practices and to suggest a
decision support method allowing designers to con-
trol them.
RELEVANCE OF SPACE CONCEPT IN COOP-
ERATIVE PRACTICES
According to Ireland, spatial problems are complex
(Ireland 2015). He argues that to organize spaces, ar-
chitectural practices tend to flatten spatial problems
into something quantifiable so that they can be man-
aged and planned. He states that "Approaching the
configuration of space in the standard way raises the
question whether any richness is lost? There is often a
qualitative disconnect between the articulation of spa-
tiality in the built environment and the spatiality of be-
ing". To overcome this issue, we propose to integrate
spatial logic in BIM practices. Indeed, the architec-
tural space concept is a shared knowledge between
all AEC actors, unlike physical structural or MEP ele-
ments, which define it in BIM models. This integra-
tion will enhance designers' cooperative work, and
enables them to act on the container (building ele-
ments) taking into account the quality of the content
(spaces) through sensitive dimension expression.
The architectural space; a design entity
loaded with sense
Evoking the sensitive dimension of space raises the
issue of its representation as a negative empty and
immaterial entity: did the designer need BIM tools
to translate the feeling of space and communicate
it? How to represent what cannot be seen, but
can be felt? How can we represent both physical
and abstract information necessary to describe sen-
sitive knowledge about a space? All this informa-
tion seems very useful, especially during the design
phases, when user requirements are given in terms
of spatial qualities (e.g. the first intentions required
by a house owner might be: a large kitchen, an open
and sunny living room, etc.) and spatial topology (liv-
ing room near the kitchen, bedroom upstairs, etc.).
Current BIM practices must therefore translate this di-
versity of properties by moving towards new repre-
BIM | Applications - Volume 2 - eCAADe 34 | 655
 sentative and cooperative methods (associating rep-
resentation, description, relations, etc.). In addition
to adding and visualizing spaces, BIM tools func-
tionalities must allow designers to enrich, to navi-
gate on and to check sensitive spatial data to ex-
press all owners' requirements. They must also al-
low them to receive, instantly, all information about
changed spaces, responsible actors and resulting re-
quired tasks, according to each actor needs.
To provide a useful conceptual framework in
order to lead a collaborative situation functional
analysis, Ellis and Wainer established a functional
clover, which defines three cooperative areas: co-
production, coordination, and communication (Ellis
and Wainer 1994). Based on this functional clover, we
highlight in the next sections the relevance of space
concept in cooperative activity.
Space as a co-production aid
Spatial logic in BIM practices facilitates co-production
of spaces. Actors manipulate space as a generic en-
tity including a set of building elements (walls, ceil-
ings, floors, doors, windows, etc.). This practice is
helpful, especially, when the project program is gov-
erned by regulatory requirements on space (quan-
titative, qualitative and topological requirements).
Some specific public projects (museums, hospitals,
courthouses, airports, etc.) or projects having graph-
ical charters (banks, stores, etc.) have this kind of re-
quirements. For example, French hospital operating
theater design is governed by regulatory spatial re-
quirements, e.g. 1. "A relaxation room can be internal
or external to the operating room. It must have a natu-
ral lighting." ; 2. "The pre-anesthetic room must be next
the operating theater. It is desirable to provide mood
lighting and / or specialized lighting." ; 3. "A doctors'
cloakroom is compulsory, it ensures changing theater
dress... It should connect common circulation areas to
the operating theater." ; 4. "The recovery room is a part
of an operating theater. It is considered a buffer zone
between the theater and the common area of the hos-
pital." (Translated by the authors from: French circular
DH / FH / 2000/264 on 19 May 2000 on the establish-
656 | eCAADe 34 - BIM | Applications - Volume 2
ment of operating theater boards in public health in-
stitutions).
Such spatial requirements are only known im-
plicitly by architects and designers, who are con-
tinually trying to satisfy during the advancement
of the design. Thus, a checking process is indis-
pensable. BIM practices should enable designers to
enrich qualitative and topological properties of
space in order to take into account requirements dur-
ing the design process (Siala et al. 2016). These
qualitative and topological aspects of a space should
be generalizable to all spaces of the same type. For
example, in requirement 1) above: "natural lighting"
must be generalized on all "relaxation rooms" of the
hospital. Also, in requirement 3) each "operating the-
atre" in the hospital must be separated from the com-
mon circulation by a "cloakroom".
During the conceptual design phases, traditional
communication means, like minutes of meetings,
show certain limits, especially when coordination in-
volves certain particular structural or MEP elements
(a dropout beam, a pipe reservation in a slab, etc.).
Indeed, the localization of observations in the de-
signed building is difficult, particularly when actor
has not attended the coordination meeting. That is
why, often the architect must annotate the observa-
tions on building plans in order to localize and then
communicate them to the interested actors. Thus,
allow actors enriching spaces with meeting obser-
vations facilitates the reuse of information by all ac-
tors and the identification of actors' actions. To struc-
ture all the information about spaces, it will also be of
interest to enable designers to enrich spatial data
associating notes (to report an issue, point out a
remark, etc.), documents and links (mood lighting
data sheet, specific lighting website, etc.).
At the operation and maintenance phase, the
integration of the building user in BIM practices
will provide the possibility to take into account the
changes he brought to his private spaces. Indeed, the
decomposition of BIM models into individual "space-
objects" (a detached and shared part of the whole
model) allows the building user to enrich spatial in-
 formation adding his changes and thus give asset
managers a clear vision of the real state of the entire
building at a time t.
Space as a coordination aid
During architectural design process, coordination
activities cover, in most cases, changing the state
of some building elements (add a wall, enlarge a
window, move a column, etc.). Existing BIM prac-
tices enable the structuring of information concern-
ing various building elements, classifying them by
type. Thus, to coordinate on a specific element im-
plies consulting all lists of elements (walls, columns,
beams, etc.). In this context, a spatial logic navi-
gation will provide not only a space-oriented coor-
dination (by zone, floor, space type, etc.), but also
the identification of the actors who are concerned
by a specific activity. For example, to coordinate
about a 'kitchen design' needs all element layers
forming this space: enclosing walls, ceilings, open-
Table 1
Relevant tasks for
collaborating
taking into account
sensitive spatial
dimensions.
ings, equipment, etc. Based on selected layers, it
could be easy to determine the necessary actors to
coordinate this activity (e.g. in this case: architect,
designer, electrical engineer, MEP engineer, kitchen
designer, etc.). To ensure a spatial logic navigation
through the project, a definition of relating enclosing
elements, openings and even eventual sub-spaces, is
necessary. So that selecting a space involves the se-
lection of all building objects belonging to it. With
this feature spatial logic navigation becomes a reality.
The integration of qualitative and topological spatial
information in current BIM practices would make it
possible to check whether designed models sat-
isfy required constraints, during updating models.
This will aid the co-production of spaces by orient-
ing designer's vision towards the desired spatial qual-
ity. This checking will highlight problematic changes
which do not respect the requirements previously
enriched by the architect. For example, in require-
ment 3) of the previous section: "A doctors' cloak-
BIM | Applications - Volume 2 - eCAADe 34 | 657
 room is compulsory, it ensures changing theatre dress...
It should connect common circulation areas to the op-
erating theatre." A method must enable BIM tools to
verify if proposed design solutions satisfy this topo-
logical requirements (for example by focusing on the
location of those three spaces: operating theatre,
cloakroom, common circulation areas).
During the evolution of design, models are ex-
changed between actors, so that each one intro-
duces adjustments according to his related com-
petence and field of intervention. In this context,
current BIM practices should allow checking auto-
matically changes that have been carried out on
spaces during updating models. This will help to
determine the impact of changes on the quality of
spaces, and then to communicate relevant updat-
ing changes to interested actors. For example, once
an architect has expanded a space, the MEP engi-
neer should have the notification allowing him to re-
vise heating and cooling requirements. Checking up-
dating changes also enables BIM tools to associate
changes with the responsible actor. This will facili-
tate the detection of all actors' actions and ensure
a more detailed tracking of the design progress.
Space as a communication aid
Respectively to the previously intended two check-
ing process, current BIM practices must ensure the
distribution of the resulting information flow, accord-
ing to each actor's needs. So that each actor re-
ceives only relevant information relating to his
filed of intervention, and then acts quickly. E.g.
1. An architect has changed the function of a space
from "staff office" to "conference room". The acous-
tic engineer should have the notification to act on
the acoustic treatments of enclosing walls, accord-
ing to the new function; 2. During the operation of
a residential building, a user has changed the spa-
tial arrangement of a bathroom. The plumbing en-
terprise should have the notification to verify the pip-
ing connection. Likewise, the asset manager should
have the notification allowing him to actualize the
BIM model taking into account the latest changes.
658 | eCAADe 34 - BIM | Applications - Volume 2
Based on the conveyed information, the BIM man-
ager can emit various suggested tasks on spaces
for each concerned actor, in order to structure the co-
ordination activity.
In this chapter we explained how BIM practices
can offer new opportunities to collaborate on sensi-
tive spatial information. Table 1 summarizes required
tasks of this approach. In the following chapter we
will highlight limitations of current BIM tools to pro-
vide such tasks.
STATE OF CURRENT BIM PRACTICES
In order to establish the need for a new vision of
spaces in current BIM practices, existing BIM tools are
first investigated.
Methodology
According to Table 1, we have first defined a set of
simple modelling tasks. These tasks have been estab-
lished to highlight fundamental limits of the various
BIM tools with regard to tasks described in the pre-
vious chapter (allowing the cooperation on sensitive
spatial data). Then, sets of qualitative and topologi-
cal information were associated with a set of spaces
modelled by a CAD tool, then registered under IFC
instance to be tested. Regarding topological infor-
mation, several kinds of link relations have been used
in this model, in order to make connection between
spaces (e.g. adjacency, intersection, content of sub-
spaces, etc.).Qualitative data concern spatial require-
ments supported by the IFC model.
Finally, handling these types of data by differ-
ent BIM tools enabled the detection of the limits of
each tested tool to represent and manage initially en-
riched semantics, to continue enriching them and fi-
nally to cooperate throughout. The observation work
was carried out on a set of CAD tools (ArchiCAD, Revit,
Allplan, and Tekla) and groupware (BIMsync , A360 ,
BimTRACK , Tekla BIMsight and dRofus ). The results
of this study are illustrated in Figure 1.
Figure 1
Ability of tested BIM
tools to support
defined tasks.
CAD tools
Creating space involves the automatic setting of its
quantitative properties. Tested tools are limited to
spatial requirements supported by the IFC model and
do not consider the rest of the qualitative spatial
requirements such as those concerning the open-
ness of a space, its sunshine, etc. Likewise, they
do not offer the opportunity to add the topologi-
cal constraints explained in the previous chapter. In
most of the investigated tools, a control approach al-
lowing verifying the compliance of design solutions
with related specified requirements is missing. CAD
tools serve, among others, to organize spaces during
the architectural design. However, they do not un-
derstand the connection between spaces and even
the relation linking them to the enclosing elements.
Tested CAD tools allow designers adding notes to a
space (name, function, area, headroom, etc.), but do
not permit further enriching it with associated doc-
uments, links and other personalized remarks like
notes that report issues, meetings' observations, etc.
Reporting issues is made possible by several inte-
grated modules into CAD tools like A360 and Bim
Track. Such modules provide the possibility to illus-
trate and share screen shots or recoverable views of a
design situation, but they do not link remarks to cor-
responding spaces. With tested tools, it was possible
to navigate on designed spaces within 2D or 3D vi-
sualization. However, given that all of these tools do
not understand the topology of architectural objects,
they do not provide navigation with a spatial logic.
Most of tested tools allow team-working on the
same model. They enable designers to select and
borrow desired building objects from the BIM model,
working on and then synchronizing data. Using the
same approach, we can imagine integration of the
building user in the collaborative environment, in or-
der to facilitate communication during asset man-
agement phase. Feature which is not provided at
present by existing BIM tools and which requires
that "space-objects" should be defined and loaned
by asset managers to the building users. Although
CAD tools offer advanced collaboration features, they
show some limits when it comes to architectural
space and specifically to spatial quality of the de-
signed building. To facilitate collaboration, most of
these tools can detect changes made to the building
objects whilst updating models. A simple compari-
son between two versions serves to identify all build-
ing elements that have changed. Information about
changes brought to spaces remains, however, unad-
dressed by all tested tools.
BIM | Applications - Volume 2 - eCAADe 34 | 659
 Thus, considering spaces in such cooperative prac-
tices represents a relevant alternative guiding collec-
tive activity to manage and validate the spatial qual-
ity of the building throughout the design phases.
Groupware
Created with CAD tools, building objects can only
be visualized and managed by groupware. Most
of tested groupware interpret space semantics au-
tomatically assigned by the CAD tool (quantitative
intrinsic properties), and ignore properties enriched
by the designer (Qualitative and topological require-
ments). DRofus has the most advanced features al-
lowing the management of spatial information. It
enables the introduction of spatial requirements in-
cluding equipment, some personalized qualitative
requirements and even proximity requirements be-
tween spaces. A possibility for checking models
pursuant to specified requirements is made possible
by integrating dRofus module to a CAD tool (Revit).
DRofus and Tekla BIMsight offer the ability to enrich
information about a space associating notes and doc-
uments. In A360, the information about spaces was
not taken into account neither in the 3D visualization,
nor in the list of hierarchical objects. This last tool
interprets BIM models only as a set of building ele-
ments (walls, columns, beams, doors, windows, etc.),
spaces are thus completely ignored. Furthermore,
no tested groupware permits navigation within the
designed building with spatial logic, nor allows, like-
wise, the assembly of meeting observations by space.
Groupware seem to be more appropriate than CAD
tools for manipulation by building users. Therefore,
providing a possible decomposition of models on
groupware presents a pertinent approach, making
possible for users to annotate spaces with changes
made during the operation of the building.
Most of the investigated groupware allow the
identification and management of compliance issues
during updating models, by sharing views, annota-
tion, and conflict detection (this last feature is sup-
ported only by Telka BIMsight and BimTRACK). How-
ever, they do not cover in any case incompatibility
660 | eCAADe 34 - BIM | Applications - Volume 2
related to the spatial quality of the designed build-
ing, and even related to the form of spaces due to
changes in enclosing elements. For example, shift-
ing a wall position may generate the expansion of a
space and the reduction of another. Consequently,
those spaces needs in terms of lighting, cooling,
heating, etc. also change.
So, foresee the checking of updating changes in
space, in current groupware, helps to orient the de-
signer's interest to improving the spatial quality of
the designed building.
TAKING INTO ACCOUNT QUALITATIVE
AND TOPOLOGICAL SPATIAL INFORMA-
TION
Our purpose in this chapter is to define a first data
model describing the sensitive aspects of spaces.
Based on the observations illustrated in our analysis
work on the state of the architectural space concept
as addressed in the recent data models and in the cur-
rent BIM practices.We define below the core concepts
of our space model.Brodeschi states that the built en-
vironment may assume different functions depend-
ing on the physical configuration or the activities to
be exercised in (Brodeschi 2015). In our model, archi-
tectural <Space> has two types of properties classi-
fied as follows:
• <Quantitative Property> which refers to the
intrinsic space data. E.g. geometric data, lo-
cation, quantity, etc.), and
• <Qualitative Property> which refers to extrin-
sic space data and depends on the interpreta-
tion of a human observer (e.g. the cooling or
heating mood of a space, its openness to the
outside, etc.).
A <Space> may allow the unfolding of one or more
activities. E.g. a bedroom must allow the following
activities: rest, play, read, sleep, etc. Each <Activity>
belongs to a given <Activity Type> that requires a
specific ambiance. E.g. <Day Activity>, <Night Ac-
tivity>, etc. Thus, designing a space involves first to
define the user activities that will be housed in and
Figure 2
Space model
including
qualitative and
topological
information.
then to determine ambiance fostering the unfolding
of each <Activity Type>. Each <Ambiance> has spe-
cific requirements (see Figure 2).
A space is governed by a set of design require-
ments depending on the <Space Type> in which it
belongs, following a topological preference or ac-
cording to ambiance required by the <Activity> that
will be housed in. Qualitative requirements may
also be associated with a <Space Type>. Thus each
<Space> must obey the common requirements for
its <Space Type>. Requirements related to the space
<Topology> define its relations with the remaining
building spaces. E.g. Communication relation or
closeness relation between two spaces, openness re-
lation or extension relation with the outside, etc. All
these requirements are variable, unlike regulatory re-
quirements. They may vary from a project to another
according to the climatic and geographical condi-
tions of the project, etc. and especially according to
the future user preferences.Our first space model is
illustrated in Figure 2 using Express-G diagram sym-
bols. In this model, we tried to take into account sen-
sitive spatial data highlighted in this paper. Based
on defines core concepts, a control protocol is pos-
sible to detect the spatial quality issues over qualita-
tive and topological requirements previously set by
the designer based on the requested program.
CONCLUSION AND FUTURE WORK
This study presents a review of architectural space
as a key concept to reveal a sensitive dimension of
architecture in current BIM practices during design
phases. Observing recent research having dealt with
space data modelling has allowed identifying the lim-
its of these models to integrate sensitive spatial data.
Exploring existing BIM tools has enabled this study
to determine missing tasks allowing designers to de-
sign and co-produce spaces, to coordinate and also
to communicate about spaces. In particular, about
spaces sensitive aspects. As shown in Figure 1, only a
small number of tasks are supported, even partially,
by tested tools. We noticed that tasks 6, 7, 9, 10, 11
and 12 are not addressed by all of existing BIM tools.
Visibly, dRofus has the greatest ability to enable re-
maining tasks.
A future work will include improving this first
model. The final global model will be used to imple-
ment a decision support method allowing designers
to control the spatial quality of the designed build-
ing, during cooperative work. DRofus seems to con-
BIM | Applications - Volume 2 - eCAADe 34 | 661
 stitute an interesting support for this future work, in-
cluding the integration of all identified missing tasks
discussed in the third chapter. The overall goal is to
achieve an efficient space-oriented prototype allow-
ing the unveiling of the sensitive dimension of archi-
tecture, hidden currently in BIM practices.

REFERENCES
Bhatt, M and Kutz, O 2012, 'Ontological modelling of
form and function in architectural design', Applied
ontology, 7, pp. 1-32
Bjork, Bo-Christer 1992, 'A conceptual model of spaces,
space boundaries and enclosing structures', Au-
tomation In Construction, 1(3), pp. 193-214
Bonnaud, Xavier and Chris, Younès 2012, 'L’actualité sen-
sorielle de l’architecture contemporaine', in Galli-
mard, Coll (eds) 2012, Architecture et perception, La
découverte
Branz, Building and construction (eds) 2014, New
Zealand BIM Handbook, BIM Acceleration Commit-
tee, New Zealand
Brodeschi, Michal, Putievsky Pilosof, Nirit and E. Kalay,
Yehuda 2015 'The definition of semantic of spaces
in virtual built environments oriented to BIM imple-
mentation', 16th International Conference CAAD Fu-
tures, Sao Paulo, Brazil., pp. 331-346
Crunelle, Marc (eds) 1996, L’architecture et nos sens, ULB,
Bruxelles
Drozd, Celine, Meunier, Virginie, Simonnot, Nathalie and
Hegron, Gerard 2009 'What Tools and Modes of
Representation to Reflect an Architectural Atmo-
sphere?', In, 9th European Architectural Endoscopy As-
sociation Conference, Cottbus, Germany, pp. 77-88
Ekholm, Anders and Fridqvist, Sverker 2000, 'A concept
of space for building classification, product mod-
elling and design', Automation In Construction, 9, pp.
315-328
Ellis, Clarence and Wainer, Jacques 1994 'A Conceptual
Model of Groupware', CSCW 94, ACM Conference on
Computer Supported Cooperative Work, North Car-
olina, USA, pp. 79-88
Ireland, Tim 2015 'An artificial life approach to configur-
ing architectural space', 33rd eCAADe Conference, Vi-
enna, Austria, pp. 581-590
Siala, Aida, Allani, Najla, Halin, Gilles and Bouattour, Mo-
hamed 2016 'Donner du sens à l’espace architec-
tural. Vers une intégration de la dimension sensible
de l’architecture dans les pratiques BIM', Séminaire
de Conception Architecturale Numérique SCAN’16

662 | eCAADe 34 - BIM | Applications - Volume 2

TopoBIM: Web-based Spatial Topology for Early Design
Participation
Mark Meagher1 , Phillip Langley2
1
University of Sheffield 2 University of Sheffield, Bryden Wood Limited
1,2
{m.meagher|arp12ppl}@sheffield.ac.uk

TopoBIM is a 3D web-based viewer for BIM data that facilitates the capture of
stakeholder knowledge related to project requirements and constraints in early
design. The software provides an interface for viewing 3D models and data for
selected room types and adding topological annotations. The use of topological
representation is proposed as a method for facilitating knowledge capture,
allowing decisions about the details of plan layout to be deferred and widening
the potential for participation in the early stage design process. Topological
representation is widely employed in the engineering disciplines, but is not
commonly used as a means of capturing schematic information in early design.
TopoBIM is proposed as a methodology and workflow for the introduction of
topology in early design, and as an example of lightweight, bespoke software that
informs design by allowing stakeholders to perform specific tasks using BIM data,
without the constraints and limiting complexity of BIM authoring environments.

Keywords: Early design, Topological representation, Participation, BIM,

Knowledge capture

INTRODUCTION licenses (Kassem et al. 2012). The current project

The early stages of architectural design are charac-
terized by limited information, rapid cycles of design
development, and input from multiple client, consul-
tant and stakeholder groups. The need to capture
detailed, structured building information in early de-
sign is particularly important in building types such
as hospitals that involve complex dependencies be-
tween space organization, equipment, and building
environmental services. Obstacles to stakeholder
participation in the early stages of building design in-
clude the need to master complex design software;
the challenges for stakeholders of working with ar-
chitectural representation; and the cost of software
addresses challenges to stakeholder participation in
early design through the development of bespoke
software for web-based creation and editing of build-
ing information.
This paper describes an application of bespoke,
web-based software as a means of providing access
to building information in early design and capturing
input from stakeholders regarding building require-
ments and constraints. The project proposes the use
of topological representation as an effective and ac-
cessible method for a wide range of stakeholders to
view building information and describe programme
requirements and constraints. While topological rep-

BIM | Applications - Volume 2 - eCAADe 34 | 663

 resentation is widely used in engineering disciplines
and for specialized architectural analysis, its use by
stakeholders in early design is still limited, and there
is a benefit in transferring this visualization type and
method for abstract thinking from engineering to the
design disciplines and the early design stages.
BIM IN EARLY DESIGN
Building Information Modelling (BIM) is a design pro-
cess involving the production of detailed 3D mod-
els with associated project data. The use of BIM is
rapidly becoming industry standard practice in the
UK, with all companies tendering for public sector
contracts required to implement level 2 BIM, which
mandates the use of open data formats and recom-
mends a series of data drops documenting project
development at key stages in the design process.
Data Drop 1 occurs concurrently with the preparation
of the project brief and documents the requirements
and constraints gathered during a period of consul-
tation with stakeholders.
A significant limiting factor in the effective imple-
mentation of BIM is its use in the early stages of de-
sign. There are multiple obstacles to the use of cur-
rently available BIM authoring software in early de-
sign; two of the most pressing issues can be summa-
rized as the complexity of the interface and the limita-
tions of the underlying data model. The BIM user in-
terface is designed to enable the production of highly
detailed 'virtual buildings' by expert users, and cur-
rently available software does not offer task-specific
interfaces to limit this complexity for stakeholders or
managers who need access to a limited subset of the
tools or data available in the application. In early
design a limited range of tools are needed for mod-
elling, data management and analysis, and providing
the full range of functionality to every user can invite
error through misuse of an unfamiliar interface.
In early design there is an opportunity to inform
decisions by providing simplified access to BIM data.
A promising model for providing access to BIM data
at this stage of design is the use of bespoke software
applications such as those proposed by Derix (2010)
664 | eCAADe 34 - BIM | Applications - Volume 2
who writes "Where lightweight computational sim-
ulations exist that can be assembled into integrated
hybrid workflows, design intentions for implicit archi-
tectural properties can gradually emerge through the
empathetic coupling of heuristics. The lightweight
applications must be limited in functionality (parsi-
monious) and visualize in real-time the intentions of
the simulation as it searches, to render its heuris-
tics transparent." We have built on this proposal for
a collection of simple, lightweight applications that
together provide the designer with diverse views of
complex phenomena. Also important is the modular-
ity of these applications, which must allow easy data
exchange between each other as well as with BIM au-
thoring software.
In addition to simplicity and modularity, we pro-
pose that transparency should be an essential re-
quirement for software development and workflow
in early design. Resnick et al. (2000) discuss the im-
portance of making software less opaque (i.e. un-
derstandable) to its users in response to the ubiq-
uity of what they call 'black boxes': tools whose
form and function offer few clues about the rules in-
forming their design and operation. Most architec-
tural software is a black box, due both to proprietary
protection of source code and to an overwhelming
complexity that limits understanding to experts who
themselves understand only a subset of the software
and its functionality. In contrast to this, transparency
is the principle that software and the workflows it en-
ables should be understandable to non-experts in its
inputs, outputs and basic rules of operation. The in-
put and output file formats for our application are hu-
man readable, allowing (in principle) for the creativ-
ity and intuition of the designer to play a role in their
analysis and use (Raymond 2003).
The design brief is a document produced in
the earliest stage of building design as a record of
client requirements, project objectives, building pro-
gramme, sustainability goals, and other project con-
straints and parameters. In highly complex building
types such as hospitals the preparation of the de-
sign brief involves input from a range of stakeholders
Figure 1
Diagrammatic
representation of
the Diagnostic
Imaging
department in a
hospital, showing
the level of detail
used in RIBA stages
0 and 1. The
diagram indicates
connection and
adjacency without
specifying precise
plan layout. Image
source: Bryden
Wood Limited.

Figure 2
The topoBIM user
interface. On the
left is a browsable
IFC tree structure,
and on the right are
the editing controls
and visualization
options. Image
source: Authors.

BIM | Applications - Volume 2 - eCAADe 34 | 665

 with expertise in a particular aspect of the building's
construction or operation, information which is com-
piled into a document that can then be used as the
basis for cost, feasibility, carbon, and other types of
analysis. This stage of design typically corresponds
with the development of diagrammatic plan layouts
indicating required circulation, connection and adja-
cencies between spaces (see Figure 1); and massing
models displaying building geometry and site con-
figuration. In the United Kingdom this stage of de-
sign is referred to as Royal Institute of British Archi-
tects (RIBA) stages 0 and 1, and corresponds to a sug-
gested data drop that includes a list of spaces in the
building and a summary of requirements and con-
straints including finishes, mechanical and electrical
services, equipment and furniture for each room or
room type.
The design brief is developed in consultation
with the client organization and panels of building
users, operators, community groups and other stake-
holders. The document outlines a set of require-
ments that become the basis for subsequent stages
of design, and which can be used to check that the
design specifications are consistent with client and
stakeholder requirements. This consultation typi-
cally takes place in meetings with representatives
of the various stakeholders, and can then be orga-
nized as a written summary and Data Drop using
the COBie (Construction Operations Building infor-
mation exchange) format. It is also common to pre-
pare Room Data Sheets (RDS) that summarize the re-
quirements of each room in the facility.
The limitations imposed by software often arise
through the interface, but can also result from an un-
derlying data model that fails to allow for the storage
of necessary information. By providing a structured
format for building data, the Revit data model or data
exchange formats like the IFC (Industry Foundation
Classes) eliminate ambiguity by limiting the possibil-
ities for unstructured extension. This is a benefit un-
less the data that needs to be stored has no place in
the structure. For example, the information needed
for automated checking of compliance with green
666 | eCAADe 34 - BIM | Applications - Volume 2
building standards such as LEED or BREEAM is im-
perfectly captured by proprietary BIM authoring en-
vironments such as Revit, and also by exchange for-
mats such as IFC or COBie, requiring the use of exter-
nal databases to enable the storage of data required
for this particular application (Biswas et al. 2013).
Similarly, topological relationships between objects
in the BIM model are not captured by the IFC data
format and must be automatically inferred (Krishna-
murti et al. 2014) or manually added as an external
data source.
TOPOLOGICAL REPRESENTATION
The term 'topology' is used to refer to graph theory,
a method for describing relationships between ob-
jects. Graphs consist of a collection of objects, re-
ferred to as nodes or vertices, and pairwise connec-
tions or edges. In architecture graphs have been
used to represent relationships of adjacency and con-
nectedness between rooms within a building plan
(Langenhan et al. 2013) and spaces or streets within
the city (Hillier et al. 1983, Vasku 2013). Topology
has been used extensively in artificial intelligence re-
search to support the automated production of op-
timal space layouts using expert systems, evolution-
ary algorithms and other approaches to finding op-
timal configurations with a given set of constraints
(Liggett and Mitchell 1981, Eastman 1973, Jo and
Gero 1998). In these applications parametric relation-
ships have been used as a formalization of design
rules and topology as an abstract representation of
relationships (for example of adjacency and connect-
edness) in the architectural plan. This application of
topology for automated space layout is intended as
an extension of human design capabilities, replacing
the most repetitive work of a technical draftsman in
order to allow designers to focus on more complex
tasks (Eastman 1973).
Topology has also been explored as a link be-
tween qualitative, human-readable design represen-
tations of design information and their correspond-
ing quantitative, machine-readable data formats.
Topological representation has been proposed as a
simplified form of design communication analogous
to sketching in its capacity to focus on a limited set
of relationships within the building (Medjdoub and
Yannou 2000). Topology can equally be used as a
link between human-readable design representation
and a machine-readable data format through the au-
tomated or semi-automated translation of architec-
tural drawings into topological data structures (Lan-
genhan et al. 2011).
In this project our intention has been to use a
topological representation as an intuitive abstraction
of spatial requirements in the hospital that can be un-
derstood and used by stakeholders in the design pro-
cess to capture knowledge related to the design and
operation of the hospital. It is convenient in terms of
data storage and visualization that this information
is also machine-readable, but we have deliberately
avoided automated data analysis in favor of man-
ual evaluation and processing of the information col-
lected using our interface, assisted by visualization.
TOPOBIM USE CASE
TopoBIM is a 3D web-based viewer for BIM model
data (geometry and semantic information) that facil-
itates the capture of stakeholder knowledge related
to project requirements and constraints for the de-
sign brief in RIBA stages 0 and 1. The software pro-
vides an interface for loading 3D models and data
for selected room types, and the ability to define (di-
rected or undirected) connections between rooms.
Each group of connections occupies one layer, and
using multiple layers it is possible to define different
types of connection between the same rooms (see
Figure 2).
The interface deliberately avoids the complex-
ity of BIM authoring environments such as Revit: al-
though a virtual building model with a high level
of detail is required in later stages of design, this
level of detail can prove distracting and unnecessar-
ily time-consuming in early design when the knowl-
edge needed to produce a detailed model is not nec-
essary or indeed available. We have chosen graphs as
a type of representation that allows design require-
ments and constraints to be defined without specify-
ing the kind of precise spatial information required to
produce a plan layout. For example, the software al-
lows one to select the rooms required in a given hos-
pital department and then determine the sequence
in which these rooms would be encountered by a
particular user of the hospital. The captured infor-
mation can then be exported as an adjacency list to
software such as Gephi, an open source tool that pro-
vides access to multiple algorithms for graph visual-
ization and analysis (Bastian et al. 2009). Most im-
portantly, the information captured from stakehold-
ers can inform the development of the design brief.
Although the graph information is both human read-
able and machine readable, we have focused in this
project on the manual process of capturing and eval-
uating the data required for producing the design
brief.
The use case considered in this paper is a hypo-
thetical consultation with stakeholders in the design
of a private hospital. The first application of topo-
BIM with a client is currently underway and will be
reported in future work; in this publication we re-
port a hypothetical use case describing the tests con-
ducted within Bryden Wood Limited during the pro-
cess of developing the software. In this initial use
case we have envisioned the use of the software to
support the preparation of design briefs for hospi-
tals and other building types involving complex de-
pendencies between space organization and equip-
ment. Directed graphs have been used as a simple
and direct means of representing multiple, overlap-
ping requirements while making use of data from
past BIM projects as a starting point in the design
process (Langenhan et al. 2013). Nodes represent
rooms, which can be selected from a library of rel-
evant room types exported in IFC format from BIM
models for past projects (see Figure 4). Because
room types in hospitals and other institutional build-
ing types are often standardized, and because Room
Data Sheets are a standard format for communicat-
ing building data to clients and stakeholders in early
design, there is a benefit at the design briefing stage
BIM | Applications - Volume 2 - eCAADe 34 | 667

in using the room as the principle unit of design and
analysis. Each room model contains semantic infor-
mation and geometry for the room itself and asso-
ciated furniture and equipment, allowing users of
the software to inform design proposals with plau-
sible assumptions about the requirements for stan-
dard room types. The following is an outline of the
proposed topoBIM workflow.
Each participant in the consultation process re-
668 | eCAADe 34 - BIM | Applications - Volume 2
Figure 3
The topoBIM
workflow. Source:
Authors.
ceives a link to an instance of topoBIM pre-loaded
with a set of room models specific to a given de-
partment in the hospital (see Figure 3). A collection
of rooms can then be selected from this list by the
user and loaded into the 3D viewer. Each room ap-
pears as a simplified volume with associated equip-
ment and furniture (see Figure 4). Multiple 3D lay-
outs are available in the interface. Directed connec-
tions can be added by clicking on two rooms in the
 desired sequence. Connections occupy user-defined
layers that can be used to represent any type of infor-
mation about the relationship between rooms in the
department. Each user will be able to create layers
to capture information, assigning their own name to
identify each category of information. For example,
a nurse or other medical professional could choose
to add information to layers representing the desired
adjacency of rooms and the flow of patients and staff,
with separate layers for inpatients and outpatients.
Rooms such as operating theatres are associated with
a complex constellation of rooms and circulation pat-
terns for staff and patients, and one of the hypothe-
ses that will be tested in the implementation of this
project is the effectiveness of topological represen-
tation as a means of visualizing and capturing these
relationships, with a separate graph for each type of
information.
Figure 4
A single room in the
topoBIM interface.
Associated furniture
and equipment are
included as 3D
models with
corresponding data
from the IFC.
Source: authors.
Once all desired layers have been defined, the graphs
can be saved as individual adjacency list text files and
viewed in a force-directed graph visualization that al-
lows users to see an abstract layout based on each
of the graphs they have defined. At the end of the
consultation process we expect to have several hun-
dred individual graphs, each representing a particu-
lar aspect of the department in question. Among the
graphs we anticipate collecting from users in the con-
sultation is information about movement of people;
adjacency and connection of rooms; and connection
of rooms to air handling units (HVAC system). An im-
portant feature of this software is that it also enables
users to create layers representing information they
have not explicitly been asked to provide: these user-
defined layers offer the possibility of a bottom-up
definition of what information is most important in
defining the final layout of the facility. Once consulta-
tion is complete the data will be communicated back
to clients stakeholders to confirm their accuracy.
DISCUSSION
In the TopoBIM graphical interface the abstraction of
topological representation is used deliberately as a
way to defer decisions about detailed aspects of de-
sign, analogous to the use of a hand sketching in
early design (Medjdoub and Yannou 2000). The use
of a 3D interface for representing both rooms and
connections was chosen as a means of preserving
an understanding of the relative volume and spatial
qualities of rooms. The 3D interface also facilitates
analytic views that distribute rooms and equipment
on the z-axis, for example according to their level in
the IFC hierarchy.
The simplicity of TopoBIM in terms of interface
and representation is a conscious response to the
complexity of BIM software and its representation of
complex building types. The project makes use of
three types of simplicity to support access to and
augmentation of spatial information by clients and
other stakeholders in the design process. First, the
simplicity of the software itself means that it provides
a very specific function, the creation and viewing of
topological information, a function not available in
industry standard BIM applications which describe
geometry and semantic information but not topol-
ogy. Second, the interface presents a deliberately ab-
stract representation of the building as a topologi-
cal network which ignores details of spatial layout in
order to focus on dependencies and constraints that
must be articulated in the design brief prior to spec-
ifying a specific design solution (Coates et al. 2005).
Third, the software further simplifies the experience
of each user by allowing them to focus on only those
layers of topological information relevant to their ex-
BIM | Applications - Volume 2 - eCAADe 34 | 669
 Figure 5
Three types of
connections as
viewed in the
topoBIM interface
(left) and in Gephi
(right). The Yifan Hu
Proportional layout
algorithm is used in
Gephi for
visualization of
graphs created in
topoBIM. Node
color visualizes the
number of
connections.
Source: Authors.

pertise: for example a surgeon can create and review
only connections related to the circulation route for
surgical staff in a particular section of the hospital,
without the distraction of considering overlapping
and adjacent activities within the building.
Currently there is a gap in the BIM-based design
process between the BIM authoring environment on
one hand, and on the other standard data formats
such as COBie and IFC. Both the proprietary author-
ing software and the open data standards are restric-
tive in their own way to stakeholder participation,
and can benefit from the addition of modular soft-
ware applications which add specific functionality to
the BIM design process. We propose that stakeholder
participation can be enhanced when these applica-
tions are designed in a way that promotes the overall
transparency of the design process, a goal that can
be accomplished in part through the use of human-
readable (text-based) file formats, parsimonious user
interfaces and modularity of functionality and de-
sign. The intention is to enable additional oppor-
tunities for designers to apply their creativity and
intuition in problem-solving, rather than relying on
the computer to automate the process of evaluating
building data in early design.
The purpose of topological representation in this
project is to isolate a specific element of the BIM data
model for use by a particular stakeholder. This ap-

670 | eCAADe 34 - BIM | Applications - Volume 2

proach could allow stakeholders who don't need ac-
cess to the full BIM model to access only that part
which is relevant to their work, and to annotate it with
their knowledge of building requirements and con-
straints.
CONCLUSIONS
This project has focused on the collection of stake-
holder knowledge in early design rather than anal-
ysis of this information. As discussed earlier there
is a long and well-established research trajectory in
building performance analysis based on topological
representation, and we have deliberately not incor-
porated analysis in topoBIM. Following the principles
of simplicity and modularity described above, our
aim in this project has been to provide only the func-
tionality needed to collect information that might
not otherwise be recorded and store this information
within the BIM model. We anticipate that the analysis
and use of this information will become the subject of
a future project.
In future work we will investigate the application
of topoBIM at multiple scales. Previous research has
demonstrated the validity of topological representa-
tion for a range of scales including design for manu-
facture and assembly (DfMA), the whole building and
the city. Although our current functional prototype
focuses on the scale of a single building oor or de-
partment, it can be made to work with these addi-
tional scales with minor modification. We will also in-
vestigate in future work the modification of existing
BIM data formats to include the possibility of captur-
ing topological information such as that produced by
our application. One exciting possibility to explore in
future work is the integration of topological user an-
notations in a future version of the IFC.
TopoBIM has been developed as a collaboration
between researchers at the University of Sheffield
School of architecture and London architecture firm
Bryden Wood Limited. The use of topological rep-
resentation for knowledge capture was developed
in a series of projects at Bryden Wood Limited, the
most recent of which is described in this paper. Fund-
ing for this project was provided by the University of
Sheffield through an Impact, Innovation and Knowl-
edge Exchange (IIKE) scheme funded by the Engi-
neering and Physical Sciences Research Council (EP-
SRC); and by Bryden Wood Limited.
REFERENCES
Bastian, M, Heymann, S and Jacomy, M 2009 'Gephi: an
open source software for exploring and manipulat-
ing networks', Proceedings of the International Con-
ference on Web and Social Media (ICWSM), pp. 361-
362
Biswas, T, Wang, T-H and Krishnamurti, R 2013, 'From de-
sign to pre-certication using building information
modeling', Journal of Green Building, 8(1), pp. 151-
176
Coates, P, Derix, C, Lau, T, Parvin, T and Puusepp, R 2005
'Topological approximations for spatial representa-
tions', Proceedings of the Generative Arts conference,
Milan
Derix, C 2010 'Mediating spatial phenomena through
computational heuristics', Proceedings of ACADIA
2010, New York, pp. 61-66
Eastman, CM 1973, 'Automated space planning', Articial
Intelligence, 4(1), pp. 41-64
Hillier, B, Hanson, J, Peponis, J, Hudson, J and Burdett, R
1983, 'Space Syntax', Architects Journal, 178(48), pp.
67-75
Jo, JH and Gero, JS 1998, 'Space layout planning using
an evolutionary approach', Articial Intelligence in En-
gineering, 12(3), pp. 149-162
Kassem, M, Brogden, T and Dawood, N 2012, 'BIM and 4D
Planning: A holistic study of the barriers and drivers
to widespread adoption', Journal of Construction En-
gineering and Project Management, 2(4), pp. 1-10
Krishnamurti, R, Toulkeridou, V and Biswas, T 2014, 'Com-
municating semantics through model restructuring
and representation', in Kensek, K and Noble, D (eds)
2014, Building Information Modeling: BIM in Current
and Future Practice, Wiley
Langenhan, C, Weber, M, Petzold, F, Liwicki, M and Den-
gel, A 2011 'Sketch-based methods for researching
building layouts through the semantic fingerprint
of architecture', Proceedings of the 14th International
Conference on Computer Aided Architectural Design
Futures, Liege
Liggett, RS and Mitchell, WJ 1981, 'Optimal space plan-
ning in practice', Computer-Aided Design, 13(5), pp.
277-288
BIM | Applications - Volume 2 - eCAADe 34 | 671
 Medjdoub, B and Yannou, B 2000, 'Separating topology
and geometry in space planning', Computer-Aided
Design, 32(1), pp. 39-61
Raymond, E 2003, The Art of UNIX Programming, Addison
Wesley, Boston MA
Resnick, M, Berg, R and Eisenberg, M 2000, 'Beyond black
boxes: Bringing transparency and aesthetics back to
scientific investigation', Journal of the Learning Sci-
ences, 9(1), pp. 7-30
Vasku, M 2013 'Generative improvement of street net-
works based on space syntax', Proceedings of eCAADe
2013, Delft, pp. 367-374

672 | eCAADe 34 - BIM | Applications - Volume 2

Parametric Architectural Design with Point-clouds
Volvox

Mateusz Zwierzycki1 , Henrik Leander Evers2 , Martin Tamke3

1,3
The Royal Danish Academy of Fine Arts, Schools of Architecture, Design and
Conservation 2 CITA
1,2,3
{mzwi|heve|martin.tamke}@kadk.dk

This paper investigates the efforts and benefits of the implementation of point
clouds into architectural design processes and tools. Based on a study on the
principal work processes of designers with point clouds the prototypical
plugin/library - Volvox - was developed for the parametric modelling environment
Grasshopper in Rhino. The prototype allows us to discuss the necessary technical
layer for the task, benchmark the tool, and finally to evaluate the benefits, that
this approach has for architectural practice, through a series of use cases.

Keywords: point-clouds, Architectural Design Tools, 3d Scan, multithreading,

.net

INTRODUCTION: POINT CLOUDS IN ARCHI-
TECTURAL DESIGN
The last years has seen a massive proliferation of 3D
scanners in all building related disciplines[1][2]. Af-
fordable scanners are today available for all sorts of
applications from airborne scans with drones, over
high-resolution indoor scans to handheld scanners,
which provide real-time data. 3D capturing technol-
ogy is becoming increasingly faster and mobile (see
Figure 1) and provides a new link between the build
environment and its representation in architectural
design.
Point clouds as a geometrical class
The output of all 3d capturing devices are point
clouds. Their sheer size poses a challenge to the cur-
rent computational infrastructure and processes of
design.Point clouds are computationally heavy and
the processing of them is demanding in terms of
hard- and software. Their proliferation is hence di-
rectly linked to the advancements in computational
power. Point Clouds are a relatively new geomet-
rical class, which was only introduced in the 2000s
(Rusinkiewicz and Levoy 2000). Since then only spe-
cialist software was able to handle point clouds with
reasonable speed. The current market of point cloud
software is oriented towards professionals from en-
gineering and land surveying industries. The leading
software as those from 3D laser manufacturers, such
as FARO scene [3] or Leica Cyclone [4], from CAD pro-
ducers, such as Bentley Pointtools [5] or Autodesk Re-
Cap [6]. Design and engineering processes can cur-
rently not take place in Point Clouds, which neces-
sitates a transformation of these into CAD formats
by means of software, such as FARO Point Sense [7],
Clearedge Edge Wise[8], Trimble Real Works[9], Sca-
lypso [10], PointCap [11]. The mentioned tools are
expensive: a Faro Scene licence costs e.g. around

BIM | Applications - Volume 2 - eCAADe 34 | 673


11.000€ [12]. The functionality of these tools is ma-
ture and they have e.g. their own advanced graphic
pipelines. All of this software is however conceived as
processor of point cloud datasets before their inges-
tion into stakeholders design software. For this task
a set of generic workflows is offered, which allows
specialists to clean, subsample and transform and to
some extend analyse point clouds and their quality
(Tamke et al. 2016). The outcome of these processes
is either a digital model, in a format that is directly
editable in a CAD package (e.g. polygon, Nurbs or
BIM formats) or a processed point cloud, which can
674 | eCAADe 34 - BIM | Applications - Volume 2
Figure 1
Historical
development of 3D
Capturing
Hardware and Point
Cloud generating
Devices. (Image
sources: Faro, Leica,
Matterport,
Microsoft, Google,
Geoslam, RIEGL).
be read by some architectural modelling programs,
such as Recap for Autodesk Revit or E57 and XYZ
point cloud formats for Archicad. These design en-
vironments do not foresee a further interaction with
the point clouds. Users have currently hardly any
means to manipulate the point clouds during the de-
sign process. This results for instance in retrofitting
projects in a situation, where the core of the design,
the representation of the building to be designed, is
in fact not accessible by the architectural designer.
This results in an overly complex workflow, where
even simple tasks, such as the transformation of point
clouds to fit to other models or the focus on areas of
interest, through selection and cropping, requires a
manipulation in an external software and a consec-
utive re-import. Recent advances in hard- and soft-
ware provide means to overcome the unusual dis-
tance between the designer and the material to de-
sign with and to enable designers to engage and rep-
resent the design with point clouds in an individual
and project specific approach.
ARCHITECTURAL WORKFLOWS WITH
POINT CLOUDS
An investigation and collaboration with stakehold-
ers, who are experienced in the use of point clouds,
and the review of nearly 500GB of point cloud and de-
rived BIM data in the European DURAARK[13] project,
revealed a general pattern of activities, in which data
from a raw point cloud is transformed into a simpli-
fied dataset that is considered useful for a design task
(Tamke et al. 2015). We could identify 5 steps, see fig-
ure 2:
1. Retrieve data - the collection and import of
different datasets, which provide insights into
the existing structure, such as point clouds
and BIM files, but as well photos, 3D legacy
plans, 3D models, geoinformation.
2. Allocation - Subsampling of datasets which
exceeds the resolution capable by the user's
hard- and software. Locating of different
datasets to each other and transformation of
these, so that they fit to each other geometri-
cally.
3. Analysis and Verification - Generation of a
qualitative understanding of the dataset. As-
sessment of the precision of the dataset, Iden-
tification of the difference between an ide-
alised design model and the scanned reality,
such as non planar walls or floors or devia-
tions.
4. Focus & Abstraction - Interaction with the
dataset through filtering of noise, focus on
parts and elements in the dataset, which are
relevant for the task, addition of new ele-
ments and transformation of the existing. Of-
ten transfer of the point cloud into the data
format of the design environment (Polygons,
Nurbs).
5. Design - Refinement and visualisation of the
data and integration into the further planning
process.
As previously mentioned the investigations into
the state of the art of point cloud related processes
revealed that current point cloud software restricts
users to a very limited set of workflows, analysis and
representations, and is restrictive in terms of input
geometry. This implies for instance, that design en-
tities cannot be linked to point clouds and that the
analysis of point clouds cannot be adapted to the de-
sign cases or language of the architects.
DESIGN INTEGRATED PARAMETRIC EDIT-
ING TOOL FOR POINT CLOUDS - VOLVOX
The five recurring steps: Retrieve data, Allocation,
Analysis & Verification, Focus & Abstraction and De-
sign, form the base functionality of Volvox, a plugin
developed by the authors for the highly popular Rhi-
no/Grasshopper visual programming environment.
It provides basic and advanced point cloud manip-
ulation and analysis tools such as: reading/writ-
ing various file formats (including industry standard
E57[14]), transformation, merging of multiple point
clouds, various subsampling techniques, cropping
based on geometry or scalar field, comparison of
point cloud with other point clouds or meshes and
various manipulation tools based on scalar fields. It
took one month to develop and release the Volvox
(version 0.1.0.0), with a basic point cloud manipula-
tion toolset. After four more months of development,
a grand update was released (version 0.2.0.0), which
brought multiple changes and new tools to Volvox.
One of the biggest problems encountered during the
development was a lack of .NET libraries for reading
and writing E57 file format. The widely used library
(libE57) is written in C++ which is not easy to inte-
BIM | Applications - Volume 2 - eCAADe 34 | 675

grate with Grasshopper (which itself is written in .NET
as well). A new library was developed - E57Lib, which
is the first .NET E57 read/write library. The library is
now published on the DURAARK GitHub.[15]
Implementing point clouds in a parametric
modelling workflow - Techniques and Tech-
nologies
Grasshopper is a node-based, algorithmic modeling
application, utilizing the RhinoCommon library ge-
ometry classes (e.g. Meshes, Nurbs geometry, vari-
ous types of curves). Any type of geometry sourced
from RhinoCommon has to be wrapped in a sepa-
rate class to be used in Grasshopper's visual work-
flow (tasks include here: describing how to draw
the geometry, apply transformation etc.). This wrap-
ping class enables the developers to seamlessly add
new types of geometry, which can be modified with
the already existing Grasshopper nodes. The de-
velopment of such a class for the RhinoCommon
PointCloud class was the first step in the creation of
676 | eCAADe 34 - BIM | Applications - Volume 2
Figure 2
Illustration of a
general pattern of
activities, in which
data from a raw
point cloud is
transformed into a
simplified dataset
that is useful for a
design task.
Dataset:
Plan3D/Berlin.
Volvox.It was quickly realized, that the node-based
nature of Grasshopper is not optimal for handling
memory expensive data types such as point clouds
- each node creates most of the times a local copy of
the data to work on. An algorithm, which loads for ex-
ample the point cloud from a file, subsamples it and
then moves it, creates 3 separate copies of the data.
Our solution is to develop a programmable node
(called "Engine") which can copy the point cloud only
once and which executes a set of parametric instruc- Figure 3
tions on it, see figure 3.The parametric nature of the A simple
instruction set for the Engine enables new ways of Grasshopper setup
working with point clouds. The user can work on for loading and
the Grasshopper solution with a highly subsampled editing a E57 Point
cloud. And once it's ready the same stack of opera- CLoud file: Load,
tions can be executed on a full resolution cloud - sim- VoxelSubsample,
ply by changing the subsampling parameter. Transform and
Volvox consists of the Engine, Instruction com- BoxCrop. This setup
ponents and further components, which operate in is using the
linear Grasshopper fashion. These are components multithreaded
which don't require a local copy of the point cloud, CloudEngineX.
and are mostly meant for analysis (average point,
closest point etc.)The plugin utilizes multithreading
on multiple levels. The Engine itself works in the
background thread. Some of the Instructions, as well
as regular components, are also multithreaded. In
case of the Engine, the thread separation has the ad-
vantage, that users can use the main Grasshopper
thread simultaneously. The use of multithreading in
other areas is meant to speed up the overall work-
flow, giving a better, more responsive user experi-
 ence. hagen, Denmark and display it in scale in plans and
sections.
Table 1
Benchmark of Multiple crops from a point cloud. Relevant details
standard for design can be extracted by means of multiple crops
operations in based on euclidean and freeform shapes.
Volvox and Cloud The utility of the filtering of point cloud con-
Compare. We tent for architectural design was explored in a stu-
measure the time dent workshop at CITA, KADK [17]. The Volvox Plu-
required for similar gin enabled architectural design students to manipu-
tasks and late point clouds in Gigabyte size and extract from full
acknowledge that building scans within minutes the information, which
the functionality of was useful to their design tasks, see Figure 5. Crop-
Evaluation of technological approach ping spaces from point clouds enables architects to
the algorithms of The competitive speed of Volvox is revealed in a
the two embed their design in a realistic looking context or to
benchmark of Volvox against one of the most pop-
applications can extract as-built information in an easy and aesthetic
ular and versatile point cloud editors - CloudCom-
not be 1:1 way, see Figure 6.
pare[16] - a freeware tool for Windows, Mac and Linux
translated. developed by (Girardeau-Montau et al. 2005), see Ta-
Figure 4 ble 1.
Dynamically
created Plan and Using Point Clouds in Architectural Work-
Section of a point flows
cloud of the Nikolaj The evaluation of the tool took place in a series of
Kunsthalle, use cases with stakeholders. These use cases were
Copenhagen, explored in common projects and workshops be-
Denmark. Left tween researchers and practitioners. They investi- Interactive comparison of models. Analysis of the
displays the full gated both existing and novel ways to address point deviation of point cloud in relation to points, planes,
point cloud. Right clouds. The question was to which extend the design meshes and other point clouds.
displays a Plan and processes with point clouds can be improved or even The monitoring of the progress on building
a Section of this charter into new fields, when a customizable compu- sites, the control of geometric deviations between
point cloud. tational environment for point clouds is used. planned and as-built geometry and the check of pla-
Dynamic creation of Plans and Sections. Sections narity of building elements during construction plan-
can be dynamically created within point clouds. ning are essential tasks in AEC (Tamke et al. 2015).
Architects work traditionally with plans and sec- The Volvox plugin gives the user the ability to calcu-
tions. This approach can be easily facilitated with the late and store e.g. deviations between point cloud
Volvox Plugin. We find, that within renovation tasks, and a mesh of the same object. The deviation from
plans and sections of the existing building stock are each point to the mesh is measured and stored in the
either missing or deviate from the as-built. The sec- point cloud as UserData for each point. This UserData
tion component in Volvox creates dynamic sections can then be retrieved and used to colorize points
on the fly. These can thereafter either be used as the in the point cloud according to a user defined color
base drawing, to adjust existing drawings or as un- range. Figure 7 shows the comparison, which the
derlay for new drawings. Figure 4 shows how Volvox Danish land surveying company LE34[18] did using
is used to analyse the Nikolaj Kunsthal in Copen- Volvox. They found deviations between an as-built
point cloud scan and the modelled representation

BIM | Applications - Volume 2 - eCAADe 34 | 677

 of a tunnel entrance to Nørreport Subway in Copen- Figure 5
hagen, Denmark. The company used Volvox, as their Context model
standard software tool could not handle the mesh- (point cloud) with
based representation of the tunnel, which the com- actual site is
pany received from their collaborators. The versatility cropped out.
of Rhino/Grasshopper allowed here a fast conversion
and user defined display of deviations.
Another common need for a deviation analysis
is the deviation from a plane. In energy renovation
tasks of existing building stock it is important to know
about the deviation of an exterior wall to the optimal
plane. With Volvox the best-fit plane of a facade can Figure 6
be found and the distance from each point to that Focus on the
plane can be stored as UserData, see Figure 8. infrastructure of the
Automated creation of architectural geometries. building in a large
External point cloud engines can be directly accessed point cloud
from Volvox and enable e.g. automatically generated through overlay of
architectural geometries from indoor 3d scans (e.g. DU- multiple cropping
RAARK IFC reconstruction DURAARK, CloudCompare operations.
[16]).
The creation of architectural as-built geometry
can be a time consuming task. The open struc-
ture of the Volvox plugin enables the user to ex-
tend the core components of Volvox with other li-
braries and/or point cloud engines. IFC reconstruc-
tion and association algorithms from the EU funded
DURAARK project were implemented in Volvox. This
enables the Volvox Plugin to overlay and associate
point�clouds and BIM models parametrically (see
Figure 9), and/or to automatically reconstruct a BIM Figure 7
model from indoor point cloud scans. This process Comparison and
takes place in a fraction of the time it would take display of deviation
to manually model the reconstruction (Ochmann et between a
al. 2016). One example from the DURAARK project designed mesh
shows a decrease in time spend of approx. 70%, see representation and
Figure 10 (Tamke et al. 2015). a point cloud scan
of a tunnel entrance
to Nørreport
Subway in
Interfacing on Point Level. Volvox grants access to Copenhagen,
the point level within a point cloud. Denmark done by
Users can program expressions to attach addi- LE34. Red: >20mm,
tional data as Rhino UserData to point clouds, e.g. green: 0mm, blue:
< - 20mm.

678 | eCAADe 34 - BIM | Applications - Volume 2

Figure 8
Deviation from
plane of a facade of
the Nyborg Strand
Hotel and
Conference Center.
Red < -100mm,
green: 0mm, Blue >
100mm.

Figure 9
Parametrically
overlay and
association of point
cloud subsets to
BIM elements
through a link
between external
tools and Volvox.

Figure 10
Automatic
reconstruction of
BIM geometry
using Volvox
extended with the
IFC reconstruction
algorithm from
DURAARK
(Ochmann et al.
2016). The time
needed for
algorithm to
reconstruct model
was approx. 10 min
per building/floor.
(Point cloud from
Symetri[19]). BIM | Applications - Volume 2 - eCAADe 34 | 679
 the distance per point to target or the height of Figure 11
points above a plane. These values can be visualised Left: Original point
through interactive coloration of the clouds. Fig- cloud. Middle:
ure 11 shows from the left the original colored point Point cloud with the
cloud, the colorization of the height of the points z-height stored as
with a custom color range and the point cloud culled UserData and
by Height UserData above a certain threshold. The colored. Right:
point cloud used here is a scan of an installation de- Point cloud culled
signed and built at the AA_AArhus Visiting School at a certain height
2015[20]. using the stored
The expression can as well operate using sim- UserData.
ple mathematical formulas or functions available in
Figure 12
.NET (trigonometry, conditionals, operators), see Fig-
Cloud Expression
ure 12.
i.e. culling and
3D print of point cloud. Voxelisataion allows to 3d coloring points
print any point cloud. from a Point Cloud
With the Volvox Plugin it is possible to mesh a Grid by expression
point cloud using the voxelization component. The based scalar field.
approach was first implemented by (Shepherd and
Treddinick 2015). Our component creates a colored
watertight mesh from the point cloud. The output
mesh can be directly 3D printed. Figure 13 shows the Figure 13
voxelization of the installation from the AA_AArhus Voxelization of a
Visiting School 2015. From left to right it shows the point cloud using
original point cloud, a coarse voxelization and a fine Volvox. Left:
voxelization. Original point
cloud. Middle:
Evaluation of use in design cases Coarse voxelization.
The evaluation showed the benefits of the integra- Right: Fine
tion of point cloud functionality into design environ- voxelization.
ments. Stakeholders fed back, that already the abil-
ity to focus and refine in a context specific way pro- Figure 14
vides new insights for architects, as they can at any The Volvox plugin
time overlay a model with the physical reality with- allows to work in
out disturbing information clutter from the rest of simultaneous
the point cloud. The simultaneous representation of representation of
CAD, parametric environment and large point cloud CAD, parametric
data was as well seen as beneficial, see Figure 14. and large point
Users with a modest knowledge of visual script- cloud data.
ing are furthermore able to quickly create customised
analysis and design evaluation tools. This allows
them to engage in their design in novel ways with
point clouds and create novel design driven tools and
workflows in this field.

680 | eCAADe 34 - BIM | Applications - Volume 2

CONCLUSION
The integration of a point cloud editing functionality
into architectural design environments has proved
to be beneficial for designers, as this frees the users
of point clouds to be authors and create the data,
rather than sole consumers of data - the role that cur-
rent point cloud software reserves for them. The in-
tegration of point cloud functionality into a paramet-
ric environment was evaluated positively, because
users could adapt existing workflows and create new
ones in a project specific and personal way. To use
a parametric modeling environment as front end for
the point cloud integration proved to be technically
challenging. Our approach turned however out to
be quite performative - being as good or even bet-
ter in a comparison with the state of the art Open
Source tools for point cloud editing. The invention
of a single node point cloud engine enables effi-
cient workflows, which allow for preliminary experi-
mentation and speculation with point clouds and a
later high quality output of professional results. The
Volvox plugin has been downloaded more than 800
times within the first four months after it's release
from its repository on food4rhino [21]. The source
code for the plugin and the custom developed .NET
E57 read/write library are furthermore published as
open-source on GitHub. As the Volvox plugin is split
into separate libraries, it's core can be used in custom
scripts and other components - taking advantage of a
seamless integration of point clouds in Grasshopper.
ACKNOWLEDGMENTS
The research, that led to Volvox took part in the
frame of the EU funded DURAARK project (FP7 - ICT -
Digital Preservation, Grant agreement No.: 600908).
The project wouldn't have been possible without
the strong support of industrial stakeholders, such
as LE34 Denmark, Symetri, Sweden & Denmark and
Plan3D Germany.
REFERENCES
Girardeau-Montaut, D., Roux, M., Marc, R. and Thibault,
G. 2005 'Change detection on point cloud data ac-
quired with a ground laser scanner.', ISPRS Workshop
- Laser scanning 2005, Enschede, The Netherlands
Hullo, J.F., Thibault, G. and Boucheny, C. 2015, 'Advances
in Multi-Sensor Scanning and Visualization of Com-
plex Plants: the Utmost Case of a Reactor Building.',
ISPRS - International Archives of the Photogrammetry,
Remote Sensing and Spatial Information Sciences, XL-
5/W4, 2015, pp. 163-169
Ochmann, S., Vock, R., Wessel, R. and Klein, R. 2016 'Auto-
matic Reconstruction of Parametric Building Models
from Indoor Point Clouds.', Computers & Graphics, 54,
pp. 94-103
Rusinkiewicz, S. and Levoy, M. 2000 'QSplat: a multireso-
lution point rendering system for large meshes.', Sig-
graph 2000. ACM, New York
Shepherd, P. and Treddinick, J. 2015 'Rapid Processing of
3D Colour Point Clouds for 3D Printing.', Proceedings
of the International Association for Shell and Spatial
Structures (IASS) Symposium 2015, Amsterdam Future
Visions 17, Amsterdam, The Netherlands
Tamke, M., Edvardsen, D.F., Beetz, J., Evers, H.L.,
Krijnen, T., Hecher, M., Zwierzycki, M., Panitz,
M., Wessel, R., Vock, R., Ochmann, S. and
Gadiraju, U. 2016 'DURAARK Evaluation Re-
port D7.4', FP7 – ICT – Digital Preservation, Grant
agreement No.: 600908, http://duraark.eu/wp-
content/uploads/2016/02/DURAARK_D7.4.pdf
Tamke, M., Evers, H.L. and Ochmann, S. 2015
'DURAARK SME Use Case - Design and
Retrofitting D7.2', DURAARK - EU research po-
ject FP7 – ICT – Digital Preservation Grant
agreement No.: 600908, http://duraark.eu/wp-
content/uploads/2015/04/DURAARK_D7_2.pdf
[1] http://www.smartgeometrics.com/blog/surveyors/3
d-laser-scanning-modeling-market-expanding/
[2] http://www.bizjournals.com/orlando/news/2015/11
/19/faro-technologies-ceo-talks-layoffs-chall
enges-and.html?ana=yahoo
[3] http://www.faro.com/en-us/products/faro-softwar
e/scene/overview
[4] http://hds.leica-geosystems.com/en/Leica-Cyclon
e_6515.htm
[5] http://www.bentley.com/en/products/product-lin
e/reality-modeling-software/bentley-pointools
[6] http://www.autodesk.com/products/recap-
360/over
view
[7] http://faro-3d-software.com/CAD/Products/PointS
ense/index.php
[8] http://www.clearedge3d.com/
BIM | Applications - Volume 2 - eCAADe 34 | 681
 [9] http://www.trimble.com/3d-laser-scanning/realwo
rks.aspx
[10] http://www.scalypso.com/
[11] http://www.pointcab-software.com/en/
[12] http://www.3dscannerstore.co.uk/software/
[13] http://duraark.eu/
[14] http://www.libe57.org/
[15] https://github.com/DURAARK/E57Lib.NET
[16] http://www.danielgm.net/cc/
[17] https://kadk.dk/en/case/hybrid-registrations-wo
rkshop-2014
[18] http://www.le34.dk/
[19] http://www.symetri.com/
[20] http://aarch.dk/info/studies/professional-studi
es/summer-school/rethinking-patterns/rethinki
ng-patterns-gallery/
[21] http://www.food4rhino.com/project/volvox

682 | eCAADe 34 - BIM | Applications - Volume 2

Index of Authors
A Cannaerts, Corneel 1-397
Abdelmohsen, Sherif 1-529 Carl, Timo 1-591
Abramovic, Vasilija 1-661 Cerovsek, Tomo 1-485
Achten, Henri 1-63, 1-613, 1-661 Cerrone, Damiano 2-473
Agkathidis, Asterios 1-291 Chelkoff, Grégoire 2-275
Ahmadi, Shahab 1-213 Chen, Kian Wee 2-583
Akleman, Ergun 1-69, 1-335 Chen, Nai Chun 2-447
Al-Qattan, Emad 2-621 Cheng, Chi-Li 1-133
Alacam, Sema 1-253 Chiuini, Michele 2-511
Allani, Najla 2-653 Cojocaru, Vasile Danut 2-409
Almeida, Ana de 2-349 Coraglia, Ugo Maria 2-503
Altin, Burcak Ozludil 2-485 Cursi, Stefano 2-503
Androutsopoulou, Eirini 2-71
Asanowicz, Aleksander 1-195 D
Aschwanden, Gideon 2-457 Das, Subhajit 2-117, 2-217
Aydemir, Zeynep 1-387 Day, Colin 2-217
Ayres, Phil 1-413 Decker, Martina 1-603
Aşut, Serdar 1-321 Dengg, Ernst 2-573
Dewberry, Michael 2-217
B Dobiesz, Sebastian 1-691
Baharlou, Ehsan 1-549 Dokonal, Wolfgang 2-573
Bajsanski, Ivana 1-185 Doumpioti, Christina 1-521
Balaban, Ozgun 2-89 Dounas, Theodoros 1-229, 1-241
Baquero, Pablo 1-73 Duarte, José Pinto 1-501, 2-189, 2-365
Baranovskaya, Yuliya 1-571 Dörstelmann, Moritz 1-571
Barczik, Günter 2-109
Bardyn, Jean-Luc 2-275 E
Beirão, José 2-305 Eisenbach, Philipp 1-93
Beirão, José Nuno 2-31 ElGhazi, Yomna Saad 2-399
Białkowski, Sebastian 1-283, 2-255 Eloy, Sara 2-349
Bingöl, Cemal Koray 2-41 Elsen, Catherine 2-159
Blonder, Arielle 1-581 Elshafei, Ahmed 1-529
Borhani, Alireza 1-69, 1-335, 1-639 Erdine, Elif 1-163
Bouattour, Mohamed 2-653 Erioli, Alessio 1-219, 1-435
Buchanan, Alexandrina 2-493 Evers, Henrik Leander 2-643, 2-673
Budig, Michael 1-83
F
C Ferreira,
Caetano, Inês 1-405, 1-465 Maria da Piedade 1-501

Authors - eCAADe 34 | 683

 Ferreira, Victor 2-365 Holland, Alexander 2-555
Fillwalk, John 2-511 Holzheu, Stefanie 2-209
Fioravanti, Antonio 2-315, 2-503 Hou, June-Hao 1-133
Flores, Hector 2-19 Humppi, Harri 2-601
Hysa, Desantila 1-357
G
Galanter, Philip 2-621 I
Gamal, Noha 2-275 Ilčík, Martin 2-417
Garcia, Manuel Jimenez 2-265
Gemme, Frederic 2-591 J
Gerber, David 1-455 Jabi, Wassim 2-147
Giannopoulou, Effimia 1-73 Jancart, Sylvie 2-159
Gomaa, Mohamed 2-147 Janssen, Patrick 2-59, 2-583
Goodship, Paul 2-79 Jovanovic, Marko 1-185
Grajper, Anna 1-691 Juntunen, Eveliina 1-669
Grobman, Yasha Jacob 1-581 Juntunen, Johan 1-669
Grohmann, Manfred 1-93 Jupp, Julie 2-95
Gruber, Anna 1-179
Guerritore, Camilla 2-189 K
Gül, Leman Figen 1-493 Kalantar, Negar 1-69, 1-335, 1-639
Kallegias, Alexandros 1-163, 1-475
H Katodrytis, George 1-329
Hahn, George 1-603 Kepczynska-Walczak,
Halici, Suheyla Muge 1-253 Anetta 1-59, 1-299
Halin, Gilles 2-653 Khabazi, Zubin 1-83
Halıcı, Süheyla Müge 1-493 Khalili-Araghi, Salman 2-633
Ham, Jeremy 1-377 Khallaf, Mohamed 2-95
Hansen, Ellen Kathrine 1-275 Kieferle, Joachim 1-229
Hanzl, Malgorzata 2-49 Kieferle, Joachim B. 1-329, 2-565
Harding, John 1-423 Kim, Jaehwan 1-53
Haring, Mario 1-179 Kim, Simon 1-631
Harris, Libertad M. 1-603 Klein, Bernhard 2-437
Harvey, Lawrence 1-377 Klerk, Rui de 2-305
Hauck, Anthony 2-217 Knight, Michael 2-573
Haymaker, John 2-117 Koenig, Reinhard 2-249
Heinrich, Mary Katherine 1-413 Kolarevic, Branko 1-25, 2-633
Henriques, Konatzii, Panagiota 1-169
Goncalo Castro 1-511 Kontovourkis, Odysseas 1-169
Henriques, Kos, Jose 2-31
Gonçalo Castro 1-537 Kotnik, Toni 1-43, 1-561, 2-391
Herneoja, Aulikki 1-63, 2-179 Kouchaki, Mohammad 1-213

684 | eCAADe 34 - Authors

Kovács, Ádám Tamás 1-269 Menges, Achim 1-549, 1-571
Kretzer, Andreas 1-501 Michalatos, Panagiotis 1-445
Kreutzberg, Anette 2-547 Min, Aram 2-529
Kruse, Rolf 2-109 Mitov, Dejan 1-185
Kulcke, Matthias 2-359 Mohamed, Basem Eid 2-591
Kwiecinski, Krystian 2-349 Mohanty, Akshata 2-59, 2-583
Kępczyńska-Walczak, Mohite, Ashish 1-561
Anetta 1-283 Moleta, Tane 2-137
Monesi, Roberto 1-435
L Montas, Nelson 1-73
Langenhan, Christoph 2-199 Montenegro, Nuno 2-31
Langley, Phillip 2-663 Mullins, Michael Finbarr 1-275
Larson, Kent 2-447
Lee, Ji-Hyun 2-379 N
Lee, Jihyun 2-529 Nagakura, Takehiko 2-447, 2-521
Lee, Jongwook 2-529 Nan, Cristina 1-345
Lee, Sang 2-209, 2-297 Narangerel, Amartuvshin 2-379
Leitão, António 1-405, 1-465 Noel, Vernelle A. A. 1-311
Li, Ruize 2-59 Nourbakhsh, Mehdi 2-117
Lima, Fernando 2-31 Nováková, Kateřina 1-613
Lo, Tian Tian 2-137
Lorenz, Wolfgang E. 1-205, 2-229, 2-359 O
Luusua, Anna 1-669 Ochmann, Sebastian 2-643
Luís, Romão 2-339 Okuda, Shinya 1-113
Olascoaga,
M Carlos Sandoval 2-19
Mahdavinejad, Oliveira, Rui 1-123
Mohammadjavad 1-213 Ozer, Derya Gulec 2-521
Mahmoud,
Ayman Hassaan Ahmed 2-399 P
Marchal, Théo 2-275 Pak, Burak 1-387
Markkanen, Piia 2-179 Panagoulia, Eleanna 1-621
Martens, Bob 1-485 Pantazis, Evangelos 1-455
Martins, Pedro Filipe 1-153 Papadopoulou, Athina 1-367
Mascio, Danilo Di 1-59, 2-511 Paraizo, Rodrigo Cury 1-537
Massoud, Passaint 1-529 Passaro, Andrés Martin 1-537
Matcha, Heike 1-237 Pasternak, Agata 2-371
Mateus, Luis 2-365 Patterson, Charlie 1-345
Meagher, Mark 2-663 Pattichi, Eleni 1-475
Meijer, Winfried 1-321 Pauwels, Pieter 2-511
Melioranski, Martin 2-473 Payne, Andrew 1-445

Authors - eCAADe 34 | 685

 Pedreschi, Remo 1-345 Stouffs, Rudi 2-59, 2-329, 2-379
Peterson, John Robert 2-493 Stricker, Didier 1-501
Petzold, Frank 2-127, 2-199 Svidt, Kjeld 2-541
Pihlajaniemi, Henrika 1-669 Szoboszlai, Mihály 1-269
Pirhosseinloo, Hengameh 2-275 Sørensen, Jesper Bendix 2-541
Plotnikov, Boris 2-127
Poinet, Paul 1-549 T
Prado, Marshall 1-571 Taborda, Bruno 2-349
Prohasky, Daniel 1-377 Takagi, Naoya 2-463
Puusepp, Renee 2-473 Takizawa, Atsushi 2-169, 2-463
Tamke, Martin 2-643, 2-673
R Tan, Elvira 2-59
Raducanu, Doina 2-409 Tarkka-Salin, Mirva 1-669
Raducanu, Vlad Andrei 2-409 Tauscher, Helga 2-611
Rejepava, Alina 1-179 Tepavcevic, Bojan 1-185
Remy, Nicolas 2-275 Tessmann, Oliver 1-93
Retsin, Gilles 1-143 Thompson, Emine Mine 1-651
Roman, Miro 2-287 Thompson, Ulysee 2-485
Roudavski, Stanislav 2-555 Thomsen, Mette Ramsgaard 1-33
Rumpf, Moritz 1-93 Thurow, Torsten 2-199
Tian, Lo Tian 1-77
S Tomarchio, Ludovica 2-437
Santos, Filipe 2-349 Toulkeridou, Varvara 2-217
Sara, Garcia 2-339 Toyama, Takumi 1-501
Schenkenberger, Benjamin 1-501 Trento, Armando 2-315
Scherer, Raimar J. 2-611 Treyer, Lukas 1-261
Schleicher, Simon 1-621 Triantafyllidis,
Schnabel, Marc Aurel 1-77, 1-377, 2-137 Georgios 1-275
Schubert, Gerhard 2-127 Tuncer, Bige 2-89, 2-437
Schwartz, Mathew 1-53 Turunen, Heidi 1-103
Schwinn, Tobias 1-549
Siala, Aida 2-653 V
Simeone, Davide 2-503 Varela,
Sousa, Jose Pedro 1-123 Pedro de Azambuja 2-427
Sousa, José Pedro 1-153, 2-427 Varinlioglu, Guzden 1-253
Spaeth, A. Benjamin 1-229 Varoudis, Tasos 2-249
Spaeth, Benjamin 1-241 Vella, Irina Miodragovic 2-391
Sprecher, Aaron 2-591 Verbeke, Johan 1-63
Stals, Adeline 2-159 Verniz, Debora 2-365
Stepper, Frank 1-591 Vock, Richard 2-643
Stojakovic, Vesna 1-185

686 | eCAADe 34 - Authors

W
Webb, Nicholas 1-59, 2-493
Weber, Markus 1-501
Wendell, Augustus 2-485
Wessel, Raoul 2-643
Wimmer, Michael 2-417
Wit, Andrew 1-631
Woessner, Uwe 2-565
Wortmann, Thomas 2-239
Wurzer, Gabriel 1-205, 2-229

X
Xu, Wenfei 2-19

Y
Yan, Wei 2-621
You, Linlin 2-437

Z
Zali, Parastoo 1-213
Zarzycki, Andrzej 1-53, 1-681
Zhenyi, Lin 1-113
Zolfagharian, Samaneh 2-117
Zupancic, Tadeja 1-63
Zwierzycki, Mateusz 2-643, 2-673
Zünd, Daniel 1-261

Ä
Äikäs, Topi 1-93

Ç
Çolakoğlu, Birgül 2-41

Ö
Österlund, Toni 2-601
Özkar, Mine 1-357

Authors - eCAADe 34 | 687

688 | eCAADe 34 - Authors