Académique Documents
Professionnel Documents
Culture Documents
Volume 2 - eCAADe 34 | 1
Editors
Aulikki Herneoja
Toni Österlund
Piia Markkanen
Oulu School of Architecture
University of Oulu
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.
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
Edited by
Aulikki Herneoja
Toni Österlund
Piia Markkanen
Volume 2 - eCAADe 34 | 3
Sponsors of the eCAADe 2016 Conference
4 | eCAADe 34 - Volume 2
Theme
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.
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
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
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
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
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
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
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.
Figure 2
Unsupervised
Clustering of
Neighborhoods
(Zoom).
Figure 3
Tweets
Classification.
Table 1
Reference values for
proximity (and
diversity). This
indicators also
apply for SPC.
Table 2
Meaning of the MXI
(Hoek 2008).
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.
only the nearest amenities, diversity indexes consid- cators play an important role in this approach, as long
ered all surrounding amenities, seeking a greater bal- as they enable the visualization and evaluation of dif-
ance (see Figure 7). This trade-off context was a key ferent scenarios for vacant areas occupation, guiding
for providing greater proximity, diversity and vari- interventions and giving hints from building poten-
ety indexes, meaning that services are nearer and in tial and mix of uses distribution. The Mixed-use in-
larger quantities along the district, which suggests dex and Spacematrix indicators proved to be useful
more walkability. These three components of WIC in the algorithmic implementation of this approach,
can play an important role in TOD planning tasks, as given that they consider objective features for mea-
long as they identify which services are more or less suring diversity and density, respectively.
available for a particular plot, street, block or even the
entire district. On the other hand, it also allows one Limitations, further developments and final
to see which areas of the district are better or worse remarks
supported in relation to different categories. Despite the usefulness of a computational tool for
Multi-objective optimization provides a set of so- supporting TOD oriented urban planning processes
lutions that are intended to be considered equally through evaluation and optimization of specific ur-
good (pareto-optimal solutions). It is an important ban features, we identify some limitations in the pre-
possibility for urban planning processes, as long as sented approach. First, we recognize that Walk In-
it strengthens the planners' role in considering "non- dex does not fully incorporate the diverse features
programmable" aspects, for stipulating subjective that can influence the walkability of an urban area.
criteria and priorities for decision making. Therefore, other variables that measure connectiv-
Despite of not being directly optimized, density indi- ity, density of streets network, and pedestrian friend-
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
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.
the information for the location of resource cell they sources and the new populations of the model will
find. Land cells, besides the information, have states; be spawned at the locations with the highest po-
it can be a land with a potential value, a resource cell tentials, where the potential is computed with the
or an agent can be attended to the cell as its home amount of interactions of agents at certain locations,
location. In Figure 2, it is shown how new entities are or cells, and with an additional amount of random
adding up to the simulation at every iteration. At ev- potentials. The random potential value added to the
ery iteration, one new resource and three agents are land cells are the simplification of natural advantages
being added to the simulation. cells have. These behaviors also will help the simula-
tion grow gradually according to the smallest inter-
Figure 2 actions in the model.
Population and
Resource Cell Model's Parameters
mechanism. 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-
The growth of the model depends on the interac- ity is defined in the smallest scale of the model with
tions of individuals and agents, in time. The new re- the interaction of agents in local neighborhoods,
Malgorzata Hanzl1
1
Lodz University of Technology
1
mhanzl@p.lodz.pl
142). The correlation between individual prefer- the usage of a hierarchy of typologies. Three ba-
ences, values and intentions, and actual behaviour sic levels were used as a starting point for mod-
and actions, is subject to Portugali's theory of self- elling - that of region (Figure 2), that of town, and
organisation (2000). Compared to Gidden's struc- that of neighbourhood. The analysis led to the pin-
turalism, which focuses on society and groups, the pointing of certain cases which although considered
point of departure for Portugali (2000) are individu- Jewish because of the high population share, at the
als and their personal choices. The above considera- same time represented different classes and varied
tions are in line with the empirical studies of the re- significantly, including the way the structure could
lations between Jews and Poles, especially in large be mapped. The basic level was a single neighbour-
cities, where more complex socio-cultural processes hood, with such features analysed as culture related
could occur. This is one of the possible paths for the notions of rhythms and sociometric layouts. The
further development of this research. heuristic method of highlighting those elements of
The current paper proposes a thesis that Jewish urban environments which were key for a given case
communities in pre-war Poland represented an ex- have been applied. The theory, based on the an-
ample of a self-organising society, one which could thropological concept of meaning and the quest for
be considered a prototype of contemporary post- its reflection in spatial order as a feature attributed
modern cultural complexity. The mapping of this to physical forms of urban structures, supported the
complexity is a challenge, a method for which is ad- classification and understanding.
dressed in the current study.
Database
METHOD While the making of the complete dataset needs
The understanding of the relations between the pat- a bottom-up approach with individual case studies,
terns of everyday life and the social habits of the com- the general framework may already be defined, with
munities needs contemporary analytical tools. The an assumption however that the classification may
complexity of patterns has been approached with be further extended. In order to achieve comprehen-
sive results which could fully render the reality, the
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.
Figure 1
The Möbius
interface.
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-
the centre of the building footprint to the closest wide variety of urban modelling scenarios.
transport node. Population densities are also visu- The case study demonstrates how Möbius can
alised by colouring the polygons. Figure 9 shows the support complex parametric GIS workflow. In the
resulting map in QGIS. case study, a workflow is developed that switches be-
tween the Geographic Information System and the
Figure 8 parametric modelling system multiple times, using
The attribute table each tool for what it is best at, mainly analysis and
for the building generation, respectively.
footprints in Through this research, some potential benefits
Möbius. 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
CONCLUSIONS AND FUTURE RESEARCH express rules, e.g. for determining building height
A parametric modelling system called Möbius has and plot coverage from the proximity attributes,
been described that supports smooth exchange of within Möbius or the Geographic Information Sys-
data with Geographic Information Systems. In par- tem, as procedural expressions, higher-order func-
ticular, Möbius provides a novel approach for visual tions, or grammar rules. Ideally, workflow flexibility
programming that integrates associative and imper- should allow users to select and develop their own
ative programming styles and uses a rich topologi- preferred workflow in support of existing design pro-
cal data structure allowing custom data attributes to cesses and preferential working methods.
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
Eirini Androutsopoulou1
1
National Technical University of Athens
1
iandroutsopoulou@gmail.com
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.
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
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.
MEDELLIN AND THE URBAN CABLE-CAR simultaneously physical transformations, social pro-
Medellin, the second largest city in Colombia with grams and participation. To do this the term 'Social
a population of around 2.5 million, has recently be- Urbanism' was used, which was 'a metaphor for an
come one of the most talked about cities in the world, integrated approach to transport and urban develop-
due to its rapid transformation from a violent, drug ment, and for the power of the strategic potential of
cartel controlled city to a city that aims to be more this integrated approach to address urban inequal-
socially inclusive. This transformation has been en- ity' (Levy 2013). Yet, whilst it was only a metaphor,
capsulated by new forms of urbanism that aim to en- it helped to enable a large-scale urban upgrading
gage the poor and a municipality that inspires to in- project for the whole city, with a focus on the urban
tegrate all of its citizens into one city. When Sergio poor.
Fajardo was elected Mayor in 2004, one of his main The most important urban project to develop
goals was to create an 'equal city for all and were all from this was 'Proyecto Urbano Integral' (Integral Ur-
citizens can construct relations stimulated by a city ban Project) or PUI. It was, and still is, an urban up-
rich in services, culture and public space' (Brand and grading project created and administrated by the
Davila 2013). This ambition developed into a new ap- municipality with the aim of increasing the quality of
proach to implementing urban projects combining life of inhabitants, focusing on areas where poverty
Figure 4
Normalised
Integration (NAIN)
weighted 'without'
(left) and 'with'
(right) speed.
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'.
mary mode of the data comes from fitness applica- are certain parameters related to that workout that
tions, the other data sources complement fitness ap- is stored, these are workout ID, type of the sport,
plications, the date and time of the workout, distance covered,
There are many different personal fitness appli- duration of the workout, calories spent during the
cations on the market. While their interfaces dif- workout, average speed, minimum and maximum el-
fer, what they do is basically same: that is to record evations, and ascent and descent during the work-
users' location within a timeframe and report it back. out. These are the basic parameters that are available
The most important personal fitness applications are when users' have basic subscription. If a user has pre-
Google Fit, Strava, Endomondo, Nike+, and Runk- mium subscription, other features such as heart rate
eeper. For our study, initially we choose Endomondo data are also available.
as data source. The reason for this selection is that, An Endomondo activity webpage is cre-
in Endomondo every workout is presented in a web- ated by the ID of the workout such as: en-
page which is listed sequentially which makes data domondo.com/workouts/23007256/. This ID is se-
collection easier. quential. As of June 2016, there are over 741 mil-
lion workouts however some of these workouts are
Table 1
deleted or private so it is not possible to access those
An example of a
workouts. Endomondo presents workouts in its in-
workout data.
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
Endomondo is one of the most used personal fitness user if the user has not selected to be a private user.
applications. It has over 20 million users worldwide Some users also associate their social media accounts
[4]. In Endomondo every user has a unique ID and with their identity. Profiles of users are also presented
every workout such as a run is stored with a workout in a webpage and an example of a URL of a profile
ID as presented in Table 1. In every workout there page is endomondo.com/profile/4356/. In the profile
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-
ban scales may ultimately result in poor design per- nition of the urban microclimate focuses on the rela-
formance in terms of how the physical features of tionship between a city block, or a small cluster of city
each scale impact on the wind flow profiles intended blocks (neighbourhoods) and the wider city topology
to be realised by the building design and or urban and or the natural topography
plan. The different scales of the physical features of
Considering the different spatial scales within these research studies reflect not only differences in
the urban microclimate, wind conditions can be the units of analysis but also how different architec-
modelled and measured relative to four levels of tural and urban features impact on wind flow. Figure
physical features relating to a building or a city's: (i) 1 illustrates this scale, which defines an architectural-
geographical location, (ii) land topography, (iii) (ur- urban spectrum that accounts for 2D and 3D features
ban) spatial morphology and topology and (iv) build- that define the building façade, building envelop,
ing geometry. Two main types of wind conditions city block, a cluster of city blocks, neighbourhoods,
can adversely impact on both the architectural and precincts, and the city as a whole.
urban scales: (a) stagnant-to-low wind flow and (b) The gap in understanding wind flow profiles rela-
high-to-extreme wind flow. In the case of stagnant- tive to the dependencies between the scales of archi-
to-low wind flow, wind velocity and permeability in- tectural and urban physical features reflects the dis-
creases the risk of airborne diseases and pollution. connect between the architectural and urban plan-
Studies of these conditions have aimed at improving ning disciplines. This 'disconnect' is to the detriment
wind flow and developing urban planning guidelines of meaningful design for urban microclimates and for
to promote ventilation. These studies generally focus achieving the positive effects of wind flow within and
on the interactions between building forms relative around buildings and cities. The approach of this re-
to a defined 'grid' of buildings. The unit of analysis search therefore acknowledges the need to investi-
and definition of the urban microclimate focuses on gate the mixed dependencies between the architec-
the relationship between building- and urban mor- tural and urban scales so as to identify the relation-
phology. In the case of high-to-extreme wind flow, ships between beneficial wind flow profiles, the phys-
wind velocity and permeability increase the risk of ical features that can support them across scales and
building damage. Research studies in this regard the resulting design qualities that define the 'urban
have focused on the interface between urban mor- microclimate'
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-
Framework of a Generative MDO Simula- the design analysis. It compares the outcome re-
tion Sequence for Wind Flow sults from analysis module with design constraints to
In order to deal with a complex design problem filters out design solutions. Evaluation module dis-
and several design objectives, this PhD research cards all solutions that do not meet building and city
project proposes an integrated approach to cou- compliance constraints. It ranks the remains of de-
pling parametric modelling techniques with MDO sign solutions according to their performance based
techniques. The framework combines architectural- on wind flow criteria defined at different levels within
urban-topography parameters in one platform for the urban canopy layer, e.g., at the pedestrian level,
performance-driven optimization for wind flow con- at ≤100, ≤200, etc.
dition. The structure of the proposed framework is Sub-Routine - External Constraints module:
based on five stages, with a series of steps across Consists of other architectural and urban design con-
them; these stages are carried out in sequential a straints such as building regulations and codes, and
manner as shown in Figure 3. zoning ordinance. These constraints granted from
Synthesis module: Using a generative paramet- outside sources such as city councils and other re-
ric approach, generate all possible design solutions lated authorities.
using architectural and urban design variables within Optimization module: This module works as a
a single geometric model so as to manipulate the val- space search mechanism, searching for the optimum
ues of geometric, morphologic and topologic design design alternatives within the domain of feasible and
parameters, and the relationship between the differ- performance solutions. The aim of the optimization
ent parameters. module is to evaluate and choose the fittest of the
Analysis module: Direct translation of build- available and feasible alternative designs based on its
ing geometry, spatial morphology and topology to- performance. However, if the optimized design solu-
gether with related wind flow parameter settings into tion does not fit the performance criteria, a designer
the wind flow simulation engine that utilises CFD to can implement changes in the initial design parame-
test solutions. As a result, analysable wind flow pro- ters using the synthesis module based on the simula-
files can be obtained directly from the model without tion results.
additional modification of geometry before analysis
results are then transferred to the Evaluation module
Evaluation module: Refers to overall results of
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 a large pool of architectural and urban design solu-
overall building or urban design both play a vital tions, i.e. two layers within the solution space of al-
role on the final design performance. Design deci- ternatives. To automatically generate such a solution
sions are not typically aiming to satisfy a single ob- space, it is necessary to first formally define the de-
jective rather, it aims at searching for best design sign problem into a series of design objective func-
solution that compromises between competing ob- tions, variables, and constraints. These definitions
jectives. It requires finding alternative design solu- are then used to generate an associative parametric
tions and analysing their performance impacts up- design model, which implicitly describes a bounded
front. However, designers in the early stage of the and a topologically fixed solution space. The defini-
design process deal with different domains that they tion of multiple objective functions provide the ba-
may not have experience in. Limitations surrounding sis for specifying design parameters and constraints
experience levels and the fidelity of information may at different levels of the urban canopy layer using
therefore affect the design decisions. In response, a corresponding wind flow measures. The specifica-
variety of design disciplines have adopted a paramet- tion of these internal layers depends on the city and
ric design approach so as to work within a process building profiles relative to existing heights. Conse-
that includes a performance analysis feedback loop quently, there will be a trade-off between the differ-
that supports early design decision-making. Utilising ent objective functions and wind flow profile opti-
parametric design in the early stage of design process mization relative to maximization of air ventilation
supports the exploration of a larger solution space versus minimisation of hazardous wind conditions.
due to the number of alternative solutions generated Design variables defined as the parameters that the
via the manipulation of the values of design parame- designer controls influence the design constraints
ters. It enables the exploration of both architectural and objective function and are evaluated in the anal-
and urban design performance, providing an analysis ysis phase; where design constraints are the func-
feedback that contributes to the designer's decisions tions that must be satisfied during the optimization
about a complex problem. In addition, it strengthens process. In order to create a flexible yet defined de-
the flexibility of the design process. sign workflow, there are a total of five categories of
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.
Background and Introduction sign project, not all criteria necessary to recognize
An important part of designing architecture consists a possible solution for the best one are known and
of producing and testing alternative possible solu- have to be found out as part of the design process.
tions for the given architectural design problems, and The process has two large interconnected problems:
then evaluating the test results to produce a next firstly, production of alternatives takes up a lot of
generation of alternatives. Then repeating the pro- time: they have to be developed mentally and rep-
cess until, with luck (and time running out before a resented as drawings and, ideally, physical models.
submission deadline) the process converges and one Alas, the more time used for production, the less
alternative is chosen as an appropriate solution. Such available for evaluation. Together with the human
a production- and test-intensive process has proven tendency for laziness, this secondly often seduces de-
to be feasible because usually at the start of a de- signers to quickly narrow the production of alterna-
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).
Figure 4
Screenshot of Tool
Prototype showing
100 automatically
generated massing
options.
Figure 6
Automatically
generated massing
studies manually
developed as
smooth surfaces
enveloping the
voxel arrangments
directly output from
the tool prototype.
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
plex datasets, to calculate the location of depart- sign stage to explore more design alternatives before
ments and wards. To solve the structural and spa- choosing optimal solutions. ISSO provides an inter-
tial optimization problem in non-healthcare projects, face and process to bring architects, engineers, and
Nimtawat and Nanakorn (2009) employed a grid- stakeholders together early on in the project cycle,
based topology optimization method that automati- imparting the opportunity to define shared objec-
cally generates beam-slab layouts. Mora et al. (2006) tives and generate spatially and structurally promis-
developed a prototype called StAr which enables ing design solutions, understand the tradeoffs, and
engineers to find the best solutions from architec- make and communicate better-informed decisions.
tural and structural designs. Delgado and Hofmeyer
(2013) proposed a virtual toolbox that generates op- METHODOLOGY
timized structural design solutions from spatial de- ISSO has four main phases (Figure 1). First, users de-
signs through simulating the iterative interaction be- fine their spatial-structural requirements such as list
tween spatial and structural designs. However, their of program spaces, adjacencies, qualities, load cases,
current implemented model only generates struc- and section types. Next, the spatial optimization ker-
tural layouts for spatial designs without optimizing nel optimizes the space layout problem, based on
its structural layout. the spatial requirements. Third, these spatial lay-
A major limitation of these studies is the lack of outs are optimized using structural optimization ker-
an integrated environment for establishing paramet- nel. Finally, users construct pareto-optimal solutions
ric relations and dependency among architectural and then select optimum solutions based on decision
and structural elements for design generation, analy- theory and social network techniques. The following
sis, and decision making. This study aims to develop sections describe the details of the process.
ISSO, an integrated spatial-structural optimization
toolkit to help designers during the conceptual de-
Figure 3
Design process of
parametric
structural design
optimization.
User Requirements room is likely to be more relevant for the design team
First, project design teams (including architects and than staff toilet. After completing the spatial require-
engineers) input contextual data into the system, in- ments, users input structural analysis data such as
cluding template layouts as shown in Figure 4(a). building materials, a list of sections, and loads and
They represent layouts from industry best practices, supports.
constraining spatial allocation of program elements
to industry tested layouts. Next, architects input fa- Spatial Optimization Kernel
cility program data, including program names, quan- The first stage of ISSO seeks to quickly generate and
tity, area, dimension or spatial aspect ratio expecta- analyze spatial design alternatives (Figure 2). To be-
tion, as well as priority values for each program ele- gin, the architect develops a building program, con-
ment, on a scale of one to 10. For example, an exam sisting of a list of spaces, their required dimen-
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
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.
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.
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- In order to accommodate for these requirements, to-
Rhino. dates to the physical model translate in real gether with the motivation to keep the configuration
time to Grasshopper and in turn to any fur- as fast and easy to operate as possible, it was decided
ther logic which is defined. On top of cre- to implement a single Grasshopper component. A
ating a physical input stream, the compo- TCP (for reliably establishing connections) and UDP
nent extends upon the usual functionality of (for streaming data once the connection has been
Grasshopper by allowing for dynamically de- established) client which was described in the previ-
fined input geometry which frees the user ous section has been adjusted and maintains the live
from configuring input parameters and allows stream between the CDP platform and Grasshopper.
them to focus on designing while providing An additional program was developed in order to cre-
constant input from any running simulations, ate a bridge to a Grasshopper component (Figure 4)
maintaining a feedback loop which informs whose input parameters consist of the connection
the design. details, custom configuration settings and a Boolean
2. Accurately translate existing physical geome- flag allowing to turn live streaming of data on and
try to the digital model. Rather than a position off. The output parameters consist of the footprint
of a generic or previously defined geometry, surfaces and heights for extrusions and mesh geom-
the actual 3D representation of the model is etry for free form shapes. A distinction is made be-
transmitted, processed and visualized by the tween the physical and the digital geometry passed
component. The system implementation sup- from the CDP platform. The physical would often cor-
ports a wide array of physical models, ranging relate to the actual design while the digital to the sur-
from extruded blocks (represented as planar roundings. That separation is particularly important
when running simulations where a clear distinction The process begins as a digital model is loaded to
between the context which is fixed and the design the CDP platform and is extended with physical ob-
which is subject to change. jects. Once the Grasshopper plugin has been config-
Once data is received and processed by the ured to the IP address and port of the CDP platform
Grasshopper component, it is available for the vast a live stream first constructs the combined model
array of manipulations and simulations which exist in Grasshopper and then responds in real time to
either natively in Grasshopper or in custom plugins. changes made to the model. From this point data is
Though most of the development focused on pro- available for further manipulations in Grasshopper.
viding a generic solution to serve as a base for other The solar rights plugin has several inputs- the ge-
Grasshopper definitions, several workflows, mostly ometry on which solar access is tested on, the geom-
in the field of climate simulations, have been devel- etry which potentially blocks solar access, and con-
oped further as test cases. These range from rather figuration settings such as grid size and sun positions
straightforward simulations such as shading studies tested upon. The output is a polysurface represent-
up to radiation, energy use and outdoor comfort sim- ing the maximum heights of the new buildings which
ulations. To better illustrate the proposed workflow, would still preserve solar access to the surroundings.
an implementation of a solar rights envelope simu- The polygons of the surrounding buildings are
lation based on the method proposed by Capeluto chosen and their facades are extracted and passed to
and Shaviv (2001) which determines the maximum the solar rights plugin as is the geometry of the physi-
height of new development in a site while preserv- cal objects. The other settings are preconfigured and
ing direct sun access to surrounding buildings is pre- can be optionally overwritten for advanced usage.
sented (Figure 5). Once the simulation is set to run, a polysurface ap-
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.
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.
Table 1
Characteristics that
various parties take
into consideration.
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.
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 reflection ratio of 91.0 %. The scale model envelope
FABRICATION PROCESS was fabricated by a laser cutter machine using ply-
Test cases were conducted using a simple rectangu- wood sheets. Using other materials to produce the
lar room with dimensions of (4*4*3) m. This room was envelope would not affect the testing process as the
designed to be similar to the one used in other vali- most important factor is the final reflectivity of the
dation studies such as CIE 171:2006 technical report. surface material. The inner surfaces of the envelope
This room has an unglazed opening on the southern were painted with 5 layers of white spray with 30 min
façade with dimension of (3*2) m. The opening is time interval between each layer in order to achieve
filled with a different parametric pattern in each test. consistent and homogeneous reflectivity. The model
All inner surfaces were white painted with an average scale is 1:10 in order to fit the overall scale of the test-
ing area and to fit the photometric sensors as well.
PREDICTION OF ERRORS who studied 3ds Max and set the error band between
Interpreting the results was based on assessing the 10% and 20% of under-estimation. Thus, an average
amount of discrepancies. The eventual result will be of 15% under-estimation is used in this study. As a
evaluated after checking the available error bands. A consequence, there is an overall predicted error band
number of researchers have outlined different causes for this study of 55.0 % (40.0 % + 15.0 %). This is pre-
of errors when using scale models in validation dicted as an underestimation of the lux levels in soft-
(Schiler 1987; Love and Navvab 1991; Thanachare- ware analysis when compared to the physical scaled
onkit et al. 2005; Aghemo et al. 2008). These ex- model.
pected causes are as follow: For further detailed assessment, the discrepan-
cies were categorized into two stages of error bands:
• Parasitic light penetration into scale models.
the lower error band (20.0 %) and the higher error
• Sensors levelling and placement in the scale
band (55.0 %). The lower error band was chosen
models.
to reflect a high accuracy analysis and to represent
• Surfaces reflectance.
the maximum error by Reinhart and Breton's study
• Lux meter calibration.
(2009). The higher error band is the maximum limit
• photocells size (sensing aperture).
for error. Discrepancies which exceed the 55.0% re-
Different literature sources, including the ones men- flect a lack of accuracy by the software.
tioned previously, were checked to gather the final
error margins for this study. These sources have used DISCUSSION
different pieces of software besides using scale mod- Based on the findings from the four tests, it is gener-
els or real buildings for validation. The next table (Ta- ally noticed that both Revit and 3ds Max are under-
ble 1) shows the reported error bands by each litera- estimating the illuminance values as compared to
ture source: the physical scaled models. This under-estimation by
From Table 1, it is concluded that scale mod- software was expected as discussed previously. In
els mostly overestimate the results when compared test 1, where a simple opening model is used, Revit
to real tests. This over-estimation is agreed to be has shown an overall better performance than 3ds
up to 50% and in extreme cases could reach 60%. Max. However, the differences in discrepancies be-
However, the average reported error of 40% is used tween Revit and 3ds Max were not high, as the ma-
for this study. On the other hand, the reported er- jority of discrepancies in both pieces of software lo-
ror bands from comparing software to real cases or cated within the lower error band (20 %) while no dis-
benchmarks vary between 10% to 25%. Since this crepancies exceeded the higher error band (55 %).
study focuses on 3ds Max and Revit, the used error In addition, the discrepancies in both Revit and 3ds
band is based on the recommendations from Rein- Max were at their lowest values nearest to the open-
hart and Breton (2009) and Tsountani and Jabi (2014)
To summarise, the average of discrepancies from Fig- The last three columns in Table 2 show the percent-
ure 6 show satisfactory results, as Revit exceeds the age of points on the sensor grid that have discrepan-
higher error band only in the winter case scenario in cies within and above error bands per each test case.
test 4, with a small difference of 2.0%. On the other For example, the first row of values in the table shows
hand, the maximum of discrepancies from Figure 7 that in Revit winter case, 94.0% of points are located
shows that both Revit and 3ds Max exceed the higher within the lower Error band. This means that 15 out
error limit in tests 3 and 4. However, 3ds Max has of 16 sensor points have discrepancies values that are
shown that the majority of its discrepancies are lo- lower than 20.0%. The following graphs in Figure 8
cated below the higher error band with percentage outline the percentages of these points in each test
between 69 and up to 100 % of sensor points (see Ta- case.
ble 2).
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.
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.
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,
Figure 7
Classification
precision with
optimized dataset
for each mp.
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
ly that, using the proposed method, the Opt dataset only for CAEP and not for other classifiers, we verify
is more clearly divided into two classes of sub the dataset with other classifiers. Here we use logistic
datasets. regression and a support vector machine (SVM) with
Let us return to the Opt result in Table 3 and ex- a linear kernel implemented in Weka 3.6.13, which
amine the itemsets in detail. In the case of class P, is often used for classification problems (Hall et al.
we can often observe the combination of the long ax- 2009). Considering the imbalance in the number of
ial line, wide visible area, and closeness to the obsta- records in each class, the misclassification cost of mi-
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.
two words: concentration and communication. We ambient factors in the environment (lighting, level
also discussed briefly of well-being in knowledge of privacy, noise) to their individual preferences and
work environment, creativity and knowledge shar- be able to work in distraction free environment when
ing. In our research project we approach de- needed.
sign of knowledge work environments in a holistic Activity-based offices with desk-sharing and
manned while acknowledging the impact of individ- multispace offices provide opportunities to switch
ual users' personal preferences regarding knowledge workstation to meet the requirements of the work
work supporting factors. One view of network of task. Surprisingly, not all knowledge workers use
key parameters that affect well-being and creativity the opportunity to switch the workstation and thus
is presented in Figure 3. their satisfaction to their environment is decreased.
In order to use contemporary knowledge work However, the individuals who switch several times
environment most efficiently and in a manner that a day have significantly higher satisfaction to their
supports both well-being and creativity, their users work environment. (Hoendervanger et al. 2016). Fur-
should demonstrate more choice-making, control thermore, regardless the desk-sharing policy, people
and active workstation switching according to task- tend to claim workstation with personal items while
requirements. Users should also be able to adjust not using those (Appel-Meulenbroek et al. 2015). It
the grammar rules comprise: (1) contextual features; Level 3_Volumetric Additions: it regulates the spa-
(2) building functions; (3) constructive features. This tial configurations deriving from the allocation of in-
level will be extended with rules for making demoli- cremental additions. Together with level 2, it per-
tions. In fact, before design can proceed it is neces- mits one to explore the range of possible solutions
sary to remove from the floor plan all the elements within the framework defined by the grammar rules,
that are not required for conversion into housing, thereby permitting to find a solution that suits the
such as internal partitions, specific office appliances, purpose of the refurbishment, both in terms of func-
and additional stairs and elevators. tional and performative behaviour.
Level 1_Design Matrix Definition: it consists in the Level 4_Internal Organization: it will be used
definition of a system of axes (structural grid, central only to verify if the division into dwellings deriving
core and circulation loop) and a system of bound- from the previous phases is able to generate inte-
aries (admissible perimeters for the allocation of ad- rior spaces in an adequate manner. The minimum
ditions) related to the initial building features and to dimension required for each room has been estab-
the range of transformation possibilities, constrain- lished according to existing regulations and three
ing the alignment of walls and the assignment of main internal zones have been identified: day area,
functions on the design matrix. This derives from the night area and service area. Once allocated the main
superimposition of axes and boundaries and it pro- zones, the subsequent subdivision into rooms will
vides the metrics for the composition of new inte- generate detailed organization of the dwelling.
rior and exterior spaces, defining modules and sub- Level 5_Facade Grammar: from the combina-
modules to use in the organization of space. tion of the previous levels, it is possible to gener-
Level 2_Division into Dwellings: it is meant to ex- ate the corresponding elevations and a 3D model,
plore the ability of the floor plan to accommodate thereby allowing one to understand the formal con-
a specific or variable functional programme and it sequences of the different refurbishment strategies.
will be described in detail in the following sections. Level 6_Allocating Technical Installations this level
It has to be considered together with level 3, which will be added in future work.
deals with the assignment of incremental additions,
because they are strongly interdependent.
Table 1
Stages and steps of
Level 2_Division
into Dwellings,
which have to be
followed in the
adaptation of a
floorplan.
the building under consideration (Step 1.1) and ap- dwellings in the functional programme following the
plied the design strategies defined in the previous chosen compositional principles (Step 3.2). In paral-
phases (Step 1.2). In Stage 2, it is selected the cir- lel with the execution of the Step 3, takes place the al-
culation strategy that is more adequate, considering location of the incremental additions of the dwellings
the functional programme and the features of the being allocated (Level 3_Volumetric Additions). In
specific building (Steps 2.1 and 2.2). The circulation Step 3.3 it is verified the compliance of the dwellings
schema is the starting schemata for the subsequent generated in Step 3.2, after allocating incremental
application of rules. Stage 3 includes the allocation additions, with area requirements for each dwelling
of dwelling types and it starts with the allocation of type, readapting their dimensions if necessary to sat-
the bigger dwellings in the corner position (Step 3.1) isfy them. In Stage 4, the access to dwellings is guar-
and proceeds with the positioning of the remaining anteed (Steps 4.1 and 4.2) by connecting them to
the dwellings in the programme to allocate, to pre- ing specific buildings in accordance with such poli-
vent further allocation of the same dwelling in subse- cies. For this purpose, it is rather important to com-
quent steps of generation, and it is added to the set plement software implementing the grammar with
D' of the dwelling types in the programme already al- software for evaluating solutions from various view-
located in the floorplan. points, which could provide performance indicators
and inform the generation of solutions matching de-
A TOOL FOR FACILITATING THE DECISION sired criteria. The final objective of the proposed tool
MAKING PROCESS is not to generate optimized solutions but to provide
The main goal of the ongoing research is the cre- a universe of adequate solutions with known perfor-
ation of a "scenario-making" tool that encompasses mances thereby making it possible to understand the
all the principles behind the adaptation of obso- trade-offs among different design alternatives and
lete office buildings into housing. This tool could the relation between geometry and performance or,
have a significant role in implementing transforma- in other words, between form and function. In this
tion policies for existing building stocks as it could be way, designers and developers could gain an ade-
used in finding the most suitable strategy for adapt- quate understanding regarding the effects of their
of the main issues is the mathematical definition of The chapter, "Comparison with the database",
similarity. Humans filter out a lot of detail differences presents a possible implementation of a correspond-
and thus are able to correlate geometric objects that ing similarity comparison. It is based on the specific
are roughly similar. Following this approach, a com- allocation of the geometrical topography of a sketch
parison algorithm is presented, which can recognize entered by the user (query) to rooms in a database.
similar geometries even in the case of locally insignif- Geometrical dependencies that are directly defined
icant, topological differences. Humans not only dis- by the user are tested against the database. Other
tinguish between the same and different, but also approaches are also possible, and a free approach
have a "feel" for greater or lesser degrees of correla- is presented in the outlook. Regardless of the im-
tion. This approach has also been transmitted in the plementation, this fingerprint considers the similar-
form of a similarity measure. ity of the three-dimensional shape as a criterion of
Thus one problem is the recognition of correla- the overall similarity comparison. For example, the
tions when geometries are not oriented identically, fingerprints can be combined for the throughways,
i.e., they are skewed relative to one another. Addi- room type and the similarity of the room shape.
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.
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-
Figure 7
Excemple of finger
print.
P1x = P6x = x1
P1y = P2y = x2
P2x = P3x = x3
(2)
P3y = P4y = x4
P4x = P5x = x5
P5y = P6y = x6 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
Variant test -> P5 , PC -> P6 , PD -> P1 , PE -> P2 , PF -> P3 .
After the reduction, only those candidates with a cor-
PAx = PBx = xa
responding topological description are considered.
PAy = PFy = xb
This refers to the number of points in the loops
PBy = PCy = xc
(Samet 2006) that form the geometry and the re- (3)
PCx = PDx = xd
duction of the point elements by means of com-
PDy = PEy = xe
mon variables. The more symmetrical a geometry is
PEx = PFx = xf
As the cities become denser, competition for visibil- the computer technologies in architecture and de-
ity between buildings and businesses increase. With sign becoming increasingly sophisticated, the disci-
densification, we also find more and more spaces that pline has transitioned from simply emphasizing pro-
fall out of the desirable visual perception. Another ductivity toward generative design experimentation
compelling problem stems from the leftover or in- and simulation. Now we use digital tools and tech-
between urban places where no one cares to go. On niques to simulate, explore, organize, and solve prob-
paper, these places may not be seen problematic. lems with enhanced visual or structural complexity.
But from the pedestrians' point of view, they may With the ongoing parametric design research project
stand out as dark spots that appear undesirable and ColorTracker we aim to explore one specific element
thus avoided. This research is motivated by the de- of environmental variables in architectural and urban
sire to use new encoding technology to translate the design: Color.
contextual color combinations for architectural and With the preceding points in mind, this on-
urban design. ColorTracker (hereafter CT) envisions going research project consists of two continuous
a unique tool to augment the simulation and devel- phases: the theory-based research and the project-
opment of chromatic formal composition for archi- based one. While the theory-based research aims
tectural design. to establish the conceptual framework, the project-
Digital technologies have greatly augmented ar- based research, based on the scientific yet theoreti-
chitectural work. They help increase quality and pre- cal premises of the first phase, directly engages the
cision, as well as efficiency and performance. With advanced inquiries in software development and de-
Figure 4
coordinate system;
Aleatorix.
source that directly affects the color rendition. The devices. The app is aimed at an architectural and ur-
zone 0 indicates pure white and the zone 10 pure ban use for color-to-geometry conversion with a mo-
black when printed on white paper. The zone gra- bile device. The CT Mobile App utilize hand-held de-
dients are measured using the neutral photo grey vices such as smartphones or tablets equipped with
(80% grey). The color temperatures refer to three a camera for site analysis and design. Here the app
main reference points: 2700K (incandescent light), is intended to work as location-based mobile appli-
3200K (tungsten photo light) and 5500-6500K (sun- cation that translates color disposition of architec-
light). The white balance on the video camera in- tural and urban places immediately at the scene into
put is calibrated according to the three color tem- three-dimensional data-geometry. Such hand-held
perature references. The exception to the color tem- mobile devices may also be networked; exchange the
perature references include fluorescent lights (green- visualized color data; and superimpose the informa-
ish tint) and sodium vapor street lights (orange). For tion on different locations.
these two kinds of lights, we will rely on the auto-
matic white balance function of the camera. Application & Use
CT can be used to assess and evaluate colorscapes
CT Mobile App (Work in Progress) and identify the perceptual black holes as well as to
The CT Mobile App will take advantage of the photo- evaluate how they can be improved in order to re-
and video-capturing capabilities of today's mobile store the chromatic balance and composition of the
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
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
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-
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.
being comprehensible for an architectural office as it cency relationships in Excel (Figure 4a) and the ana-
stands. What truly lies at the heart of the problem lytical capabilities of the NetLogo Network Extension.
is that both are not accessible, in architectural terms. On a technical level, we use a self-written extension
For this paper, we tried to make that crossing-over called NetLogoExcelBridge which allows NetLogo to
between both worlds more easily viable, using our be run from within Excel. Upon being started, NetL-
own implementation which we coined "SpaceBook". ogo establishes a channel through which it can read
and write data from Excel, which we then use to get
SPACEBOOK adjacencies given as half-matrix (Figure 4a) and ana-
SpaceBook is a mash-up between Excel and NetLogo, lyze/visualize that on a graph level (Figure 4b).
utilizing the ease in which it is possible to input adja-
Customizing graph analysis algorithms for • For closeness and betweenness, the re-
adjacency relationships lations (negative=-1, desirable=0.5, manda-
Being open-source, NetLogo makes it easy to under- tory=1) need to be mapped to positive
stand and extend the graph measures as well as the weights since the underlying algorithm ex-
graph visualizations which are the key to the discus- pects distances between nodes as input. If
sion we want to raise with this paper. As an example, the relation is negative, we map this to a high
we could easily re-implement closeness and eigen- number that simulates "infinity" (1000000 in
vector centrality so as to take negative relationships our case). Desirable relations are assigned
into account. In that process, we noticed that rela- a positive weight (1), mandatory ones are
tionships need to be mapped to edge weights, since mapped to 0 (i.e. there is "no cost" when cross-
centrality algorithms have different semantics and ing between the linked spaces or functions).
thus expect different input (also see Table 1): • 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.
Thomas Wortmann1
1
Singapore University of Technology and Design
1
thomas_wortmann@mymail.sutd.edu.sg
A NEED FOR DESIGN SPACE CARTOGRA- Architectural design space exploration and architec-
PHY tural design optimization rely on the notion of design
The challenge of representing high-dimensional data spaces to organize design variants and processes
in the two or three dimensions that are visually (Woodbury and Burrow 2006). Since these spaces of-
understandable by humans appears in many disci- ten have more than two or three dimensions, they
plines, and -with the advent of Big Data in the last are difficult to visualize. In parametric design, the
decade- has emerged as a field of study in its own number of dimensions of the design space equals the
right. This paper presents a novel method to visualize number of design parameters. The difficulty of rep-
high dimensional parametric design spaces and their resenting such abstract spaces motives the need for
performance - such as efficiency in energy and ma- multivariate visualizations, which, for example, helps
terial consumption- with applications in computa- designers understand the relationships between pa-
tional design space exploration. Specifically, the visu- rameters and the similarities characterizing group-
alization method presented here supports the under- ings of design variants.
standing of design problems in architectural design ADO combines the notion of a parametrically de-
optimization (ADO) by allowing designers to move fined design space with one or more numerically ex-
back and forth between a high dimensional design pressed performance criteria. In ADO, performance
space and a low dimensional "Performance Map". criteria typically pertain to reducing a building de-
representation of data as closed polylines. Although tion, prevents associating the plots with useful linear
such visualizations are easy to construct and under- mappings, and impedes estimating original data at-
stand, they become hard to read when representing tributes accurately."
many data, since the polylines tend to overlap. Natu- An important advantage of parallel and radial vi-
rally, visualization methods become harder to under- sualizations is that their coordinate axes allow the es-
stand as number of parameters, i.e. coordinate axes, timation of numerical parameters. This advantage
increases. contrasts with other multivariate visualization meth-
The visualization method applied in this paper - ods such as Self-Organizing Maps (Kohonen 1982),
Star Coordinates- also introduces one coordinate axis which cannot directly represent numerical relation-
for each parameter, with the axes typically arranged ships. The ordering of coordinate axes can influence
radially (see Figure 2). In contrast to Parallel Coor- the insightfulness of parallel and radial representa-
dinates, however, Star Coordinates displays a datum tions. The figures presented here order the coordi-
not as a closed polyline but as a single point. As nate axes according to the order of the parameters in
a point-based representation, Star Coordinates can the underlying optimization problem.
represent many more data than Parallel Coordinates
without overlaps, and even display a continuous field Visualizing Optimization Results
to represent the space of the data, although at the In the optimization field, fitness landscapes are usu-
price of making individual parameters less readable. ally represented with a Matrix of Contour Plots. Such
Note that, although in theory Star Coordinates might a matrix consists of n2 − n two-dimensional plots,
represent several data with an identical point, the with each plot representing the relationship between
Performance Map circumvents this limitation by pre- a pair of parameters. The plots on the diagonal
ferring better performing design variants. of the matrix are left blank or show the relation-
RadViz - a method closely related to Star ship of the performance objective to a single vari-
Coordinates- places data with a spring system, with able (see Figure 4). In other words, this visualiza-
each parameter represented by a spring. However, tion consists of two-dimensional "sections" or "con-
Rubio-Sanchez et al. (2016) conclude in a compar- tours" through the higher dimensional space. This
ison of the two methods that RadViz "introduces architecturally-inspired metaphor clarifies an impor-
non-linear distortions, can encumber outlier detec- tant limitation: Although the number of contour
the projected points in the two-dimensional map (2) value of each parameter with the vector represent-
and approximate the performance of unexplored de- ing the corresponding coordinate axis, and by then
sign variants (i.e. design variants with unknown per- adding the resulting scaled vectors. In other words,
formance values) by interpolating their performance the two-dimensional embedding p ∈ R2 of the n-
values based on the corner points of the triangula- dimensional design xn ∈ Rn is a linear combination
tion's triangles (3) or by estimating their performance of n coordinate vectors v:
values via a surrogate model (4). Note that, since the p = x1 v1 + x2 v2 + . . . + xn vn
Performance Map is surjective and based on interpo- A potential complication for Star Coordinates is
lating between evaluated designs, it does not repre- that two designs with different parameters can map
sent the whole design space, but rather a collection onto an identical point. For example, all designs
of design variants that is relatively similar to already which have the same value for all parameters (all
evaluated designs. This similarity, however, increases zero, all one, etc.), map onto the origin of the co-
the likelihood of performance estimates to be accu- ordinate axes. Since the two-dimensional mapping
rate. In other words, it is of little use to represent por- consists of linear combinations of more than two,
tions of the design space that have not been previ- pairwise linearly independent vectors in the plane,
ously explored at all. a two-dimensional point maps to a theoretically in-
finite number of parameter sets in the higher dimen-
Star Coordinates sional space. This mapping to more than one pa-
The method projects evaluated design variants us- rameter set implies that Star Coordinates is not bijec-
ing Star Coordinates. Although other arrangements tive, but surjective: Every parameter set maps onto
are possible, the coordinate axes are typically spaced a unique point, but this uniqueness is not true in re-
equally around a circle, which is also true for the ex- verse. In terms of creating the Performance Map, this
ample presented here. Note that, to avoid skew- complication is resolved by preferring the better per-
ing the visualization due to differences in scale be- forming design variant when several variants map to
tween parameter values, it is advisable to normalize the same point. (In practice variants rarely overlap).
the parameters to the same range, i.e. [0, 1]. Star As a result, the Performance Map is slightly biased to-
Coordinates determines the two-dimensional posi- wards better performing designs.
tion p of a n-dimensional design by multiplying the
the optimization results and surrogate model derived -which also includes penalties for exceeding stress
from a ten-minute run with RBFOpt, a model-based and deflection constraints- the colors are applied on
optimization algorithm. During ten minutes, 362 de- a logarithmic scale. In this way, more detail is visi-
signs were evaluated through a structural simulation, ble for the better, lighter designs. The colors follow a
with the lightest design weighing 240 kilograms. (For palette proposed by Niccoli and Lynch (2012) that is
details on the benchmark problem and optimization more perceptually adequate than conventional rain-
method see (Wortmann and Nannicini 2016).) bow palettes due to its linearly increasing luminance.
The color palette and scale are identical for all all fig-
Performance Map ures.
In Figures 2 and 3, every evaluated design is repre- Figure 2 indicates only the colored points of the
sented as a colored circle, with the position of each evaluated designs on the left, with the Delauney tri-
circle corresponding to the design's parameters, and angulation used to interpolated between them on
the circle's fill color corresponding to its weight. Due the right. Figure 3 shows the approximated Perfor-
to the large weight difference between the best and mance Map derived from the surrogate model. Note
the worst evaluated designs (240 kg and 1940 kg) how in Figure 3, the visualization indicates two sepa-
rate areas of good, i.e. light designs: One closer to the timization result can be indicated because the con-
center, and one in the upper left corner. These group- tours plots do not contain the remaining evaluated
ings indicate two different types of solutions, with points. Also note that the two different areas of good
each type displaying similar parameters. The visual- designs visible in Figure 3 are not identifiable in Fig-
ization also shows that the central group has been ex- ure 4. This comparison between Star Coordinates
plored extensively, while the peripheral group con- and a Matrix of Contour Plots confirms the disadvan-
sists only of a small number of explored points. The tages of contour plots identified above, such as the
visualized surrogate model indicates that this area fragmentation of the design space into a quadratic
might contain very high-performing design variants number of contours and the inability to identify non-
and thus is promising for future exploration. linearities between more than two parameters. The
next section discusses potential applications for the
Comparison with Matrix of Contour Plots Performance Map, with an emphasis on interactive
The contour plots in Figure 4 represent the identical design space exploration and optimization.
optimization results and surrogate model as Figure
3. Note that, in this visualization, only the best op-
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.
Figure 17
Detailed data for
PreAssemble (top)
and MatrixFree
Solver (bottom).
Figure 18
Memory usage
depend of solver
type and FE mesh
density.
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
Figure 2
Softmodelling -
Graphical map of
selection modes
and
mesh-manipulation
types.
topological transformation. The isolation of those multi-tactile interface. This allowed participants to
areas allows for an increase in the system's effi- create their own structures. The exhibition also in-
ciency, thus avoiding the use of computationally ex- cluded a series of 3D printed models designed by
pensive processes in any other area of the object. MadMDesign, which aimed to test the efficiency of
Once the local data matching process is completed, the software as a design tool. The models were a re-
particles and springs get reconnected to the new as- sponse to the challenge set by Actiu London Show-
sociated vertices and edges. This enables the object's room to create 24 different objects in only eight
physical properties to be updated with a new topo- hours (e.g. Figure 3). The feedback collected dur-
logical structure. ing the event was fundamental for the progression
towards the next stages of the software.
TOWARDS A DESIGN AND FABRICATION Since then, Softmodelling has been tested in sev-
eral workshops and installations, aiming to resolve
WORKFLOW
medium scale structures through the use of flexible
Data to matter and matter to data.. Softmodelling
materials. Alongside this process, the contribution
was first tested during Resonate 2014 in Belgrade,
of workshop participants has resulted in the creation
and was officially launched during London Clerken-
of a variety of pavilions and installations, aiming to
well Design Week (May 2014) at the Actiu London
measure both material behaviour and the efficiency
Showroom, where the software included its first
of the software as a design/fabrication tool. weaving these flexible elements together, in order to
The material experiments and physical studies create localised stiffness in areas of the structure with
that took place during these several events became higher stress levels, results in increasing the overall
extremely instrumental to the development of the stability of the structure.
software. This paper will focus on two installations, A series of pavilions are developed as an out-
which correspond to the two stages of the software's put of Softmodelling, through an assemblage of two
development: membrane structures manipulation, types of flexible materials: PVC pipes and stretch-
and flexible tubular elements which perform in a able textiles. While the PVC pipes bundle into struc-
structural symbiosis with the membrane itself. tural objects, the textile pieces get stitched together
to generate continuous surfaces. Thus the multiple
HYBRID FLEXIBLE STRUCTURES: membranes generate compression forces to encour-
Softmodelling 1.0. The prototypes developed dur- age the PVC pipes to stay connected, while the linear
ing the first stage, make use of the force density structure simultaneously pushes the membranes to
methods (Schek 1974) integrated in the software to maintain their state of tension. This results in struc-
create inhabitable structures. These structures, are tural balance producing stiffness within the architec-
driven by basic principles of rapid reoccupation sys- tural element.
tems, which offer a variety of flexible spaces. The pro- Computational membranes. Prior to the official
totypes were conceived from the idea of utilising low launch of the software, a primitive version of Soft-
cost materials with a high degree of flexibility. Inter- modelling was used for the design of the first large
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.
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.
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.
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
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.
There is a certain objectivity - a theme. This theme is character is algebraic. Although it is ambiguous what
being repeated, always in a slightly different way. It a computational drama actually is, its stage script is
is not mimetically tied to its representation, it is it's about engendering potentials and possibilities of ar-
offspring. A fugue inhabits the algebraic symbolic ticulating abstract objects in the world of data. It
space. In a similar manner, it would be challenging looks promising. In this play, books are going to be
to see how a figure of a book in the world of data abstract objects, finite objects in terms of letters and
could be different than a systemic Image of the City. words, but infinite in any other way: interpretation,
As we have already mentioned, Lynch has abstracted reading, meaning, translating, discussing... To be pre-
from the physical city. He is representing cities on cise, books are going to be our Actors. Let us play
the level of text, grammars and structures. If we take with text on the level of information.
his exercise seriously, as a theme for a fugue, what do
we get? What happens if we want to play with text Orthography - Actors - Pre-Specific Objects
on the level of information, data, lists and indexes? Book, actors, libraries. Books are our actors. They are
What are figures and fugues of books? What is their getting ready for the stage. They are never alone,
character, mood, how many faces do they have? they are always a part of some collections, libraries,
datasets, ensembles; they are moving in the move-
DRAMA ment of the others. Actors are not specific - they are
This is going to be a play. A fugue is going to be of a symbolic kind. One can say that they have an
staged almost like an ancient Greek comedy with its algebraic character which makes them opportunistic
actors, a stage and a stage play. This drama does to a meaning. Their mood is dependent on the library
not have a moral. Its actors are computational - their they inhabit and their face changes when looked at.
the same time a reflection of the whole library and a we read them. Knowledge and information become
measure of a specific book: relative to the way we look at them, just like in quan-
tum physics. Contradicting pictures do not exclude
• Word frequency vector: building 10 24 0 45 82 each other anymore. On the contrary, they develop a
39 0 609 18 15 21 50 29 60 1 0 85 11 249 40 33 different picture all together (Barad 2012). We are be-
49 16 233 72 53 31 2 0 26 1 75 179 121 1... yond objective and subjective, practical and poetic.
• Letter frequency vector: c 64758 86137 23964 As Eco would say, we are lost in the infinity of lists,
10744 33271 26169 41685 32659 32708 27677 but we are ready for our stage play.
49217 18427 57030 21832 6937 6964...
• Bigram frequency vector: new_york 0 84 11 4 Ichnography - Play - Articulation
0 46 0 5 1 2 21 0 62 6 0 1 3 13 5 2 7 120 1 30 34 Act 1 - Abstraction 1a - Concepts. Actors are on the
218 0 13 0 0 0 176 36 36 0 5 81 0 0 51 0 97... stage. A play begins. What is happening on the stage
is not straightforward and intuitive. As Serres would
Kinds of metrics one can apply to a text depend on put it: "Objects, in the distance, change their skins,
the richness of the stage. There are many of these they send one another kisses." (Serres 2000). There is
and they do not have to belong to the same kind. one process that appears consistent even from a dis-
Since informational faces are algebraic, they can be tance: similar indexes attract each other; similar in-
transformed from words and letters, to synonyms dexes group together. The whole library is in motion;
and antonyms, to Kindle and Google ratings, to taste concepts are negotiating and being negotiated. Self-
and moods of any intensity. Our probability space is Organizing Map is an algorithm that articulates this
relative to that. It changes according to the libraries interplay of indexes. This stage play is a double ar-
we would like to read from, and according to the way
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
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.
We may at first view the simplicity vs. complex- to dominate human affairs. In short, the questions
ity contrast as a matter of degree and progression: regarding simplicity and complexity hinge on how
something starts as a simple, singular entity and pro- rapidly and pervasively the reticulate configuration
gresses into a complex one, gaining multitude of operates.
qualifications and variations. We can also take for ex- Against the backdrop of reticulation, interdepen-
ample the pace and scope of such progression and dence includes the performance-critical, contextual
how fast and pervasive it may eventually become. As elements, the prerequisites, that are necessary for an
often, we break down a complex object or situation object or an event to take place. Certain prerequi-
into simple constituents so that we can understand sites bring about radical changes that propel the de-
it clearly. In a reticulate formation, complexity in velopment of human culture at an astonishing rate.
essence indicates a context of relationship and con- In the context of today's Internet-driven cultural mi-
nectedness. Especially since the emergence of the In- lieu, Java, the W3 standards, fiber optic infrastructure
ternet, not only the expanse but also the frequency at and the smart, networked mobile devices for exam-
which the reticulate configuration operates has come ple stand out as such prerequisites. Each of them
The Virtual World Builder Toolkit interface, in the ment upon the structure of ontologies representing
Automated Semantically Responsive System, is inte- building or urban types and typology. The works
grated in their own Shape Grammar Interpreter (SGI), presented here already pave the way for (universal)
using 3D transformation engine jMonkey Engine [5] - building or urban knowledge descriptions.
a free, open source game engine.
Most systems rely heavily on knowledge descrip- CONCLUSION
tions, which are then used as a control mechanism This paper presented an overview of approaches to
for shape grammar derivation. Knowledge in the Semantic Design Systems in the fields of architec-
ontologies is also used to restrict parameters in the ture and urbanism (for both real and virtual environ-
shape grammars. ments), more specifically regarding the combination
Regarding computer implementation, only the of computational ontologies and shape grammars.
Generative with Semantic Control System uses an in- A review of control mechanisms to regulate
terface which is familiar in architectural design: the shape grammar derivation was presented, suggest-
Grasshopper 3D [6] plugin for Rhinoceros [7]. ing the need for more flexible and adaptive knowl-
One important topic for future developments edge representations than just descriptions. Knowl-
of semantic design systems would be the agree-
WORKSPACES TO ENHANCE WELLBEING thetic and economical aspects but also wellbeing
AND PRODUCTIVITY and productivity of employees.
In order to improve the quality of an architectural Nowadays companies not only recognize the
artefact, a central task for multidisciplinary design connection between employee productivity and
teams is to test the tentative design solutions and see mindfulness, but also they are looking at good data
how well they work 'in practice' before, during and af- to support it.
ter the construction: from digital model to real world. The fact is, there is a major shift underway in
The building quality does not involve only aes- the workspace from employee wellness to wellbe-
SHAPE, FORM AND GEOMETRY | Grammars and Concepts - Volume 2 - eCAADe 34 | 329
rule may apply to a given shape. Beirão (2012, 228- to be made dependent on the total number of rooms,
236) offers a survey of (implementations of ) shape which is only known at the end of the production pro-
grammar interpreters so far, and must conclude that cess. Instead, Li (1999) suggests adopting (at least)
they have common limitations. Many of them com- two descriptions, the first one a diagram with the
pute only on two-dimensional shapes; and most do number of rooms as independent parameter(s), and
not apply subshape detection and therefore do not the second one the plan with the room sizes as de-
support emergence. Also, "very few shape grammar pendent parameters (of the number of rooms). By
interpreters allow for the implementation of rules op- staging the production of the diagram before the
erating with symbols" (Beirão 2012, 235), never mind production of the plan, the assignment of values to
other attributes, or a description function. the dependent parameters can be postponed until
In this paper, we will address the problem of after the assignment of values to the independent
developing an implementation of a shape grammar parameters.
interpreter supporting varying shape grammar for- Duarte (2001) defines a parallel grammar as sep-
malisms, by focusing on the implementation of par- arating different representations or aspects of a de-
allel and compound shape grammars. We will re- sign into different computations that interact with
view the literature on parallel and compound shape each other. Specifically, Duarte (2001; 2005) con-
grammars and propose an algebraic treatment facili- siders a discursive (parallel) grammar incorporat-
tating a modular approach, based on a similarity be- ing a shape grammar and a (textual) description
tween algebraic abstraction and procedural abstrac- grammar−as well as a set of heuristics, where the lat-
tion (Frank 1999). ter is intended to constrain the rules that are appli-
cable at each step of the design generation. While
PARALLEL AND COMPOUND SHAPE the shape grammar operates on shapes and the de-
scription grammar on textual (including numeric) de-
GRAMMARS
scriptions, their rules are commonly coupled, with
Originally, a parallel shape grammar was defined by
the description rule part constraining the shape rule
Stiny (1975, 37) and Gips (1975, 7) as a shape gram-
part. This combination of a shape grammar and a
mar intended to be used in the parallel generation
description grammar follows Stiny's (1981) definition
of shapes, that is, "whenever a shape rule is used, it
of a description function to augment a shape gram-
is applied simultaneously to every part of the shape
mar in order to construct design descriptions. Where
to which it is applicable" (Gips 1975, 7; Stiny 1975,
Stiny (1981) considers a description function as made
37). This is in contrast to the more common serial
up of functions, with each function assigned to a
application of shape rules, where at each step of the
shape rule and computing in parallel to the shape
generation a shape rule is applied to only one part of
rule, Duarte instead denotes the functions as descrip-
a shape. However, more recently, the term parallel
tion rules and the description function as a descrip-
(shape) grammar has been adopted in the context of
tion grammar. Otherwise, these operate in exactly
parallel computations on multiple descriptions.
the same way and with the same intention. In fact, al-
Li (1999) defines a parallel grammar as a gram-
though Stiny nowhere adopts the term parallel gram-
mar operating on different descriptions with the ob-
mar in this sense, Knight (1999; 2003a) explicitly at-
jective to resolve parametric dependencies. For ex-
tributes the concept to Stiny (1981). Knight (2003a)
ample, consider a shape grammar generating a plan
also offers a definition of a parallel grammar as "a net-
where the number of rooms is depended on the se-
work of two or more grammars that operate simulta-
lection of rules applied. In this process, it is almost
neously."
impossible to constrain the boundaries of the plan at
Admittedly, Li (1999) and Knight's (2003a; also,
the same time, as this would require the room sizes
330 | eCAADe 34 - SHAPE, FORM AND GEOMETRY | Grammars and Concepts - Volume 2
Duarte 2001) interpretation are not unequivocal. of grammars that do not operate in parallel. For ex-
Following Stiny's (1990) definition of a design as ample, Beirão (2012) uses the term compound gram-
"an element in an n-ary relation among drawings, mar to denote a composition of several discursive
other kinds of descriptions, and correlative devices grammars, where each discursive grammar formal-
as needed," Li (2001; 2004) considers seven drawings izes a so-called Urban Induction Pattern (UIP), en-
(from plan diagram to plan, section and elevation) coding a typical urban design operation or design
and nine descriptions (specifying measures of width, move. Knight (2004), instead, suggests a distinction
depth, and height, among others), in his specifica- between synchronized and a-synchronized parallel
tion of a shape grammar for (teaching) the architec- grammars, where the latter allows for sequential pro-
tural style of the Yingzao fashi. However, he consid- duction stages as proposed by Li (1999; 2001).
ers only four grammar components to define his par-
allel grammar, corresponding to four stages in the SHAPE ALGEBRAS
production (Li 1999). However, in Knight's (2003a) While Stiny avoids the term parallel grammar when
parallel grammar interpretation of Stiny's (1981) ex- referring to parallel computations, he does empha-
ample, the two grammar components of the parallel size parallel computations in multiple algebras. For
grammar−the shape grammar and the description example, when a rule applies to a shape composed
function−apply hand in hand: "the rules of a paral- of points and line segments, though the rule may
lel grammar may be linked so that the application require both one or more points and one or more
of a rule in one grammar triggers the application of line segments to be present within a prescribed spa-
one or more rules in other grammars" (Knight 2003a). tial relationship, the rule computes with points and
In fact, Knight (2003a) suggests the same interpreta- line segments in parallel. The overall shape rule com-
tion for Li's (2001; 2004) grammar, with each drawing putation actually combines two computations−one
and description specifying a component grammar in with shapes of line segments and one with shapes of
the parallel grammar, for each stage. Duarte (2001) points−"that are carried out in parallel and influence
takes a similar position when describing his discur- one another mutually" (Stiny 1990).
sive grammar applied to the houses designed by the Points and line segments, and by extension other
architect Alvaro Siza at Malagueira. Even though, spatial elements, can be considered to adhere to an
strictly speaking, he only refers to two grammars−a algebra (specifically, a generalized Boolean algebra
shape grammar and a description grammar−, he (Krstic 1999)), that is ordered by a part relation ('≤')
specifically acknowledges that both grammars in- and closed under the operations of sum ('+'), prod-
clude several sub-grammars. "These sub-grammars uct ('·'), and difference ('−'), as well as relevant trans-
correspond to viewpoints in the shape grammar (e.g. formations. For example, points may belong to the
first floor plan), and to features in the description algebra U02 and line segments to the algebra U12
grammar (e.g. morphology)" (Duarte 2001). View- (in two dimensions) (Stiny 1992); Uij denotes the al-
points define separate drawings (sketches, plans, el- gebra of spatial elements of dimension i, e.g., 0 for
evation, envelope, etc.) and features individual de- points, 1 for line segments, 2 for plane segments, em-
scriptions. bedded in a space with dimension j, e.g., 1-D, 2-D,
When the rules of a parallel grammar are linked, 3-D. Stiny (1992) extends the notion of algebras to
the linked rules can be expressed as one compound shapes with attributes: labeled points belong to the
rule (Knight 2003a). Therefore, some authors con- algebra V02 , while weighted line segments belong
sider compound grammars as an alternative term to the algebra W12 . Then, shapes of line segments
to parallel grammars, though others adopt the term and labeled points can be said to belong to an alge-
compound grammars also to denote compositions bra that is the direct product of the algebras U12 and
SHAPE, FORM AND GEOMETRY | Grammars and Concepts - Volume 2 - eCAADe 34 | 331
V02 , U12 ×V02 . Consequently, a shape rule applying the algebra's signature contributes to the module's
to shapes of line segments and labeled points can be interface (the class methods in object-oriented pro-
considered to combine two shape rules, one in U12 gramming), but the interface can be extended to in-
applying to line segments and one in V02 applying clude, among others, the part relation. Nevertheless,
to labeled points. having a (even partially) uniform interface, i.e., shar-
Any selection of shape algebras, including la- ing the same class methods, greatly eases the imple-
beled and weighted shape algebras, can be com- mentation of a general shape grammar interpreter.
bined using the direct product into a compound al- Unfortunately, as Knight (2003b) acknowledges,
gebra of compound shapes that are made up of a "the algebras that designers use, informally or for-
mix of various spatial elements, and optionally aug- mally, are many." Beyond labels and weights, shapes
mented with labels or weights. Chase (1999) notes can be augmented with attributes of any kind: "aes-
that "this is common in maps, as map features may thetic, formal, functional, structural, and so on. For
be distinguished by different element types (for ex- example, points can have diameters, lines can have
ample, lines representing roads, points representing thicknesses, planes can have colors, solids can have
cities), or labels used to distinguish elements with materials" (Knight 2003b). "The only condition is that
the same basic geometry but different semantics (for the operators of any shape algebra are defined on
example, lines can represent roads and rivers)." But, all its elementary objects, are recursively applicable,
examples in architectural representation abound as and are closed" (Yue and Krishnamurti 2013). Fortu-
well. Compound shapes may be expressed in unions nately, the notion of an algebra as derived from exist-
of the sets that form shapes from different algebras, ing algebras can be extended to augmented shapes,
with the understanding that basic and augmented e.g., of labeled points (Stouffs 2008). Defining an al-
spatial elements only interact if they are of the same gebra for labels is straightforward. Similar to points,
kind (same dimension and attribute kind, if any), and a label can be said to be part of another label only
are independent otherwise (Stiny 1992). if these are identical. Then, the operations of sum,
This notion of compound shapes, as resulting product, and difference correspond to the set oper-
from compound algebras defined by the direct prod- ations of union, intersection and difference. Labels
uct of basic shape algebras, is conducive to a modu- do not exhibit any allowable transformations, unless
lar (or procedural) implementation of a shape gram- we consider case transformations. However, the op-
mar interpreter (e.g., Frank 1999). Each basic shape eration of direct product on algebras will not support
algebra can define a single module (or a class in an an attribute behavior, and it is not straightforward to
object-oriented programming paradigm), and mod- consider an alternative operation on algebras. Krstic
ules can be combined to define compound algebras, (1999) offers an unintentional hint.
facilitating a variety of shape grammar formalisms.
The algebra's signature specifies the operations ALGEBRAIC ABSTRACTIONS
of the algebra, at a minimum, the operations of sum Krstic (1999) notes the difference between an alge-
('+'), product ('·'), and difference ('−'). The allowable bra of (maximal) spatial elements and an algebra of
transformations can be considered external to the al- shapes. The former is a partial algebra as the opera-
gebra, as part of the signature or as part of the al- tions of sum, product, and difference are not closed.
gebra's carrier, i.e., the set of elements of the alge- The sum of two spatial elements is a spatial element
bra (Krstic 1999; 2012). The part relation ('≤') is not only if the two elements overlap or are both part of
an operation of the algebra but can be expressed in another spatial element and have boundaries that
terms of the operation of product: t(a) ≤ s ⇔ t(a) · s = overlap. In general, the operations of sum, product,
t(a). From a modular implementation point of view, and difference on spatial elements are only defined
332 | eCAADe 34 - SHAPE, FORM AND GEOMETRY | Grammars and Concepts - Volume 2
if the spatial elements exist in the same subspace, cluding sum ('+'), product ('·'), difference ('−'), and re-
where the dimension of the subspace equals the di- duce ('r') can be defined for a discrete behavior as fol-
mension of the spatial elements. For example, two lows:
line segments must be part of the same infinite line, ∀ X, Y ∈ P(A) ⇒
two plane segments of the same infinite plane, and X +Y =X ∪Y
two volumes of the same 3D hyperplane. This sub- X ·Y =X ∩Y (1)
space is denoted the carrier of the spatial element, X −Y = X \Y
but in order to avoid any confusion with the carrier of r(X) = X
an algebra, we will instead refer to the co-descriptor of
the subspace. Then, the operations of sum, product, In a discrete behavior, the operations of sum,
and difference on spatial elements are defined only product, and difference correspond to the normal set
if the spatial elements have the same co-descriptor, operations of union, intersection and difference. The
that is, they are co-equal. Note from above that, in reduce operation reduces any set to a set of maxi-
and of itself, this is not a sufficient condition for the mal elements. Under a discrete behavior, any set is
operations to be closed. maximal because any duplicates are automatically re-
Having established the difference, Krstic (1999) moved. The algebras U0 (shapes of points; we omit
goes on to focus solely on shape algebras. How- the dimension of the embedding space for simplicity)
ever, it is possible to derive shape algebras from par- and L, sets of labels, can be defined in this way. Note
tial algebras of spatial or other elements in a general that descriptions, from an algebraic point of view, be-
way. First, let us assume a two-sorted partial alge- have exactly as labels and, thus, the algebra D of sets
bra with carrier {A, P(A)} and signature including the of descriptions can be defined in this way as well.
operations of combine, common and complement on Before we address other spatial elements, let us
members of A. We consider a two-sorted algebra be- first consider a behavior for weights (e.g., line thick-
cause we want the operations of combine, common, nesses or surface tones), as is apparent from drawings
and complement to extend upon the respective op- on paper−a single line drawn multiple times, each
erations of sum, product, and difference, and at the time with a different thickness, appears as if it were
same time be closed for co-equal (spatial) elements. drawn once with the largest thickness, even though
For example, the combine of two co-equal spatial el- it assumes the same line with other thicknesses (Stiny
ements that do not have overlapping boundaries will 1992).
be the set of the two spatial elements. Thus, each An algebra with carrier P1 (A), the set of all sin-
of the operations of combine, common, and comple- gleton subsets of A, and signature including sum ('+'),
ment takes as argument two elements of A and return product ('·'), difference ('−'), and reduce ('r') can be
an element of P(A), the set of all subsets of A. defined for an ordinal behavior, applying to weights,
in terms of the two-sorted partial algebra with car-
Deriving Algebras from Two-sorted Partial rier {A, P(A)} and signature including the operations
Algebras of combine, common and complement, as follows:
To derive a shape algebra from this two-sorted partial ∀ {x}, {y} ∈ P1 (A) ⇒
algebra, we need to distinguish the desired behavior. {x} + {y} = combine(x, y)
Each behavior will result in a different derivation. For- {x} · {y} = common(x, y) (2)
tunately, we can reuse behaviors for different kinds of {x} − {y} = complement(x, y)
spatial or other elements. The simplest behavior is a r({x}) = {x}
discrete behavior, applying to both points and labels.
An algebra with carrier P(A) and signature in- For weights, we know that the result of the op-
erations of combine, common, and complement on
SHAPE, FORM AND GEOMETRY | Grammars and Concepts - Volume 2 - eCAADe 34 | 333
two singleton weights is always a singleton weight. the collection of boundaries of co-shape Y that lie
We use this knowledge to define the operations of outside of co-shape X. 'same-side(b(X), b(Y))' denotes
sum, product, and difference in terms of the opera- the collection of boundaries of both co-shapes X and
tions of combine, common, and complement. Again, Y where the interiors of the respective co-shapes lie
the reduce operation results in the argument (single- on the same side of the boundary, and 'opposite-
ton) set itself. The algebra N of singletons of numeric side(b(X), b(Y))' the collection of boundaries of both
weights can be defined in this way. co-shapes X and Y where the interiors of the respec-
Deriving a shape algebra for spatial elements tive co-shapes lie on opposite sides of the bound-
other than points from a two-sorted partial algebra ary. Then, the boundary of the co-shape result-
is a little bit more complicated because of the need ing from the co-combine operation is formed by
to consider co-equal shapes. We take a two-step ap- the 'outside(b(X), Y)', 'outside(b(Y), X)', and 'same-
proach. First, we derive a sub-algebra for co-equal side(b(X), b(Y))' collections, and the co-shape can be
shapes of spatial elements, next we define a shape constructed from the union of these collections. The
algebra for a single type of spatial elements from this co-common and co-complement operations are sim-
sub-algebra. ilarly defined. In the case of the co-reduce oper-
A sub-algebra with carrier P(A) and signature in- ation, each spatial element in the co-shape is co-
cluding co-combine, co-common, co-complement, combined with the co-reduced remainder of the co-
and co-reduce can be defined for an areal behavior shape. From an implementation point of view, this
as follows: recursive definition of co-reduce may not be the most
∀ X, Y ∈ P(A) : efficient; actually, the same can be said about the
∀ x ∈ X, ∀ y ∈ Y , co(x) = co(y) ⇒ other operations−the classification of boundary seg-
co-combine(X,
Y) = ments with respect to another co-shape can be com-
∪ outside(b(X), Y )
construct outside(b(Y ), X) puted once for all of the classes inside, outside, same-
same-side(b(X), b(Y )) side and opposite-side. Obviously, these definitions
co-common(X, Y) = only serve as abstractions of the actual procedures,
∪ outside(b(X), Y )
construct inside(b(Y ), X) (3) we refer to Stouffs and Krishnamurti (2006) for actual,
same-side(b(X), b(Y )) and efficient, algorithms.
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 )) ('−'), and reduce ('r') can be defined for an areal be-
co-reduce(X)
{ =
co-combine({x}, co-reduce(X \ x)) ∃ x ∈ X
havior, applying to line segments, plane segments
∅ otherwise and volumes, in terms of the sub-algebra with car-
rier P(A) and signature co-combine, co-common, co-
The operations of co-combine, co-common, co- complement, and co-reduce, as follows:
complement, and co-reduce only apply to co-equal ∀X, Y ∈ P(A) ⇒
shapes. Instead of comparing the shapes for co- X + Y = r(X ∪ Y ) ( )
∪ {x ∈ X : co(x) = c},
equality, it is checked that all spatial elements have X · Y = c co-common
{y ∈ Y : co(y) = c}
the same co-descriptor ('co'). Then, the operations X − Y ∪= (4)
can be expressed in terms of the boundaries ('b') ∪ c {x ∈ X : co(x) =(c ∧ ¬∃y ∈ Y : co(y) =)c}
∪ {x ∈ X : co(x) = c},
of each co-equal shape (Krishnamurti and Stouffs c co-complement {y ∈ Y : co(y) = c}
∪
2004; Stouffs and Krishnamurti 2006), here termed r(X) = c co-reduce({x ∈ X : co(x) = c})
co-shape. Specifically, 'outside(b(X), Y)' returns the
collection of boundaries of co-shape X that lie out- The operations of product, difference and reduce
side of co-shape Y. Similarly, 'inside(b(Y), X)' returns are expressed directly in terms of the operations of
334 | eCAADe 34 - SHAPE, FORM AND GEOMETRY | Grammars and Concepts - Volume 2
co-common, co-complement and co-reduce on the Deriving Augmented Shape Algebras
respective co-shapes. In the case of complement, co- Next, we can tackle the issue of an algebra for aug-
shapes from X for which there exists no (co-equal) co- mented shapes. Rather than referring to a shape al-
shape in Y also form part of the result. A similar ap- gebra and an attribute algebra (e.g., L or N ), we will
proach can be taken for the operation of sum, how- instead refer to the partial algebra for the spatial ele-
ever, for simplicity, we prefer to express the opera- ments, just as we define the shape algebra from the
tion of sum in terms of the operation of reduce on partial algebra of its spatial elements. The behav-
the combined sets of spatial elements. Note that the ior of an augmented shape algebra, after all, mimics
algebras U1 (shapes of line segments), U2 (shapes of the behavior of the underlying shape algebra, with a
plane segments) and U3 (shapes of volumes) can all few differences. For example, consider two overlap-
be defined in this way. ping line segments. Without attributes, these com-
We should note that while an areal behavior ap- bine. With attributes, they combine only if they share
plies to shapes of line segments as well, from an im- the same attributes or, otherwise, of the segments
plementation point of view, it would be more effi- are identical. Otherwise, different segments (or parts
cient to define an interval behavior for shapes of line thereof ) will necessarily have different attributes and
segments. Additionally, while these behaviors cover will need to be represented separately. The behav-
shapes of different kinds of spatial elements, and sets ior of the attribute shapes, for each of the different
of labels and singletons of weights, other behaviors line segments (or parts thereof ), however remains
can be identified to apply to other kinds of attributes, the same.
for example, for material rankings (Knight 1993). In- An algebra with carrier P(A × P(B)) and signa-
stead, we will now continue to derive compositions ture including sum ('+'), product ('·'), difference ('−'),
of algebras under the direct product. and reduce ('r') can be defined for a discrete behavior,
in terms of the (attribute) algebra P(B) with signature
Deriving Algebras with the Direct Product including sum ('+'), product ('·'), difference ('−'), and
The direct product applies to all algebras that share reduce ('r'), as follows:
the same signature, specifically, the algebras U0 , U1 , ∀ X, Y ∈ P(A × P(B)) ⇒
U2 , U3 , L, D, and N , we have previously defined. X + Y =
We will address some implications of this in the dis- ∪ {(x, Bx + By ) : (x, Bx ) ∈ X ∧ (x, By ) ∈ Y }
{(x, Bx ) : (x, Bx ) ∈ X ∧ ¬∃(x, By ) ∈ Y }
cussion below. Here, we define a shape algebra with
{(x, By ) : (x, By ) ∈ Y ∧ ¬∃ (x, Bx ) ∈ X}
carrier P(A) × P(B) and signature including sum ('+'), X ·Y =
{(x, Bx · By ) : (x, Bx ) ∈ X ∧ (x, By ) ∈ Y } (6)
product ('·'), difference ('−'), and reduce ('r') in terms X −{ Y =
of the shape algebras with carriers P(A) and P(B) and ∪ {(x, Bx − By ) : (x, Bx ) ∈ X ∧ (x, By ) ∈ Y }
identical signatures, as follows: {(x, Bx ) : (x, Bx ) ∈ X ∧ ¬∃ (x, By ) ∈ Y }
r(X)
{ =
∀ (A1 , B1 ), (A2 , B2 ) ∈ P(A) × P(B) ⇒ {(x, Bx )} + r(X \ (x, Bx )) ∃ (x, Bx ) ∈ X
(A1 , B1 ) + (A2 , B2 ) = (A1 + A2 , B1 + B2 ) ∅ otherwise
(A1 , B1 ) · (A2 , B2 ) = (A1 · A2 , B1 · B2 ) (5)
(A1 , B1 ) − (A2 , B2 ) = (A1 − A2 , B1 − B2 ) Comparing this to the discrete behavior for a
r(A1 , B1 ) = (r(A1 ), r(B1 )) non-augmented shape algebra, we observe that for
the operations of sum, product and difference, if a
An algebra of shapes of points and line seg- spatial element is shared between both (augmented)
ments, U0 × U1 , and an algebra of shapes of line shapes, we combine both attributes of the spatial
segments and sets of descriptions, U1 × D, among element under the same operation. In the case of
others, can be defined in this way. the operations of sum and difference, we may need
to add spatial elements, with their original attribute,
SHAPE, FORM AND GEOMETRY | Grammars and Concepts - Volume 2 - eCAADe 34 | 335
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 of a drawing of line segments and labeled points,
We have adopted a constructive, algebraic approach the segments and points operate in the same (2D or
to defining shape algebras and augmented shape al- 3D) space and transformations of shapes of line seg-
gebras. This allows for a variety of algebras to be de- ments and shapes of labeled points necessarily need
fined from an array of basic (partial) algebras of spa- to go hand in hand. Instead, when shape algebras
tial and other elements. For example, we can de- operate in a different space (or drawing), transfor-
fine an algebra of shapes of line segments and la- mations may differ. The sum of algebras applies in
beled points, U1 × V0 , from basic partial algebras the first case, the (direct) product in the second case.
of points, line segments and labels. Other opportu- However, as we left transformations out of the pic-
nities arise. Rather than restricting ourselves to la- ture, the direct product of algebras as we defined it
bels and weights as attributes, we can consider other is applicable to both cases. We refer to Krstic (2012)
attributes, such as color, in the same way. Consider for a treatment of transformations in the context of
a basic algebra of sets of colors, C, and assume we shape algebras.
express the combination of a shape algebra with an
attribute algebra with the operator '∧', termed attri- CONCLUSION
bution, then we can define an algebra of shapes of We presented an algebraic approach to describing
colored plane segments as U2 ∧ C. Similarly, we compound shapes that includes the definition of
can then write V0 = U0 ∧ L and W1 = U1 ∧ non-spatial algebras and the combination of shape
N . Additionally, we can define an attribute algebra algebras and non-spatial algebras under an opera-
as a direct product of two basic attribute algebras, tion of attribution. This algebraic abstraction serves
e.g., U0 ∧ (L × N ) defines an algebra of shapes of as a procedural abstraction for the modular imple-
points with both labels and weights as attributes. We mentation of a general shape grammar interpreter.
may consider the operation of direct product to dis-
tribute over the operation of attribution, resulting in
(U0 ∧ L) × (U0 ∧ N ) = V0 × W0 .
REFERENCES
Beirão, JN 2012, CItyMaker: designing grammars for urban
While we commonly consider algebras of shapes
design, Ph.D. Thesis, Delft University of Technology
augmented with algebras of (sets of ) non-spatial Chase, SC 1999, 'Supporting emergence in geographic
elements, and the direct product of (augmented) information systems', Environment and Planning B:
shape algebras, other combinations are possible as Planning and Design, 26(1), p. 33–44
well. Consider the algebra D of sets of descriptions. Duarte, JP 2001, Customizing mass housing: a discursive
Though it is a non-spatial algebra, most authors con- grammar for Siza's Malagueira houses, Ph.D. Thesis,
MIT
sider descriptions parallel to drawings, not as at-
Duarte, JP 2005, 'Towards the mass customization
tributes thereof. Nevertheless, Beirão (2012) consid- of housing: the grammar of Siza's houses at
ers descriptions as attributes to spatial objects, allow- Malagueira', Environment and Planning B: Planning
ing for objects to refer to alternate, though similar, and Design, 32(3), p. 347–380
descriptions. Thus, U1 × D is as valid as U1 ∧ D. Of Frank, AU 1999, 'One step up the abstraction ladder:
course, some limitations in the way we combine alge- combining algebras', in Freksa, C and Mark, DM (eds)
1999, COSIT’99, LNCS 1661, Springer, Berlin, pp. 95-
bras may still be preferable, reflecting on the kinds of
108
design descriptions (whether drawings or other de- Gips, J 1975, Shape Grammars and their Uses: Artificial Per-
scriptions) we tend to use. ception, Shape Generation and Computer Aesthetics,
Knight (2003b; see also, Krstic 2012) also con- Birkhäuser, Basel
siders a sum of algebras that requires algebras to be Knight, TW 1989, 'Color grammars: designing with lines
specified in the same space. For example, in the case and colors', Environment and Planning B: Planning
and Design, 16(4), p. 417–449
SHAPE, FORM AND GEOMETRY | Grammars and Concepts - Volume 2 - eCAADe 34 | 337
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
SHAPE, FORM AND GEOMETRY | Grammars and Concepts - Volume 2 - eCAADe 34 | 339
(Stiny 1980). Shape grammars can analyze existing SPECIFIC SHAPE GRAMMARS
languages (analytic grammars, e.g. Stiny and Mitchell Shape grammars can analyse and describe (1) one
1978), generate new coherent ones (synthetic gram- singular language in specific grammars or (2) a mul-
mars, e.g. Stiny 1980), or transform existing de- tiplicity of languages in generic grammars.
signs (transformation grammars, e.g. Eloy and Duarte Firstly, a large number of shape grammars are
2012). On another hand, shape grammars can con- from the analytic (specific) type, intending to cap-
template one language (specific grammars) or more ture one historical style and generate new designs
than one language of designs (generic grammars). within the same style. A style is identified by com-
Style in analytic grammars is seen as an explanatory mon features that appear repeatedly in a set of ob-
tool, used as post hoc classification, while in synthetic jects. Those features may refer to the structure,
grammars style is seen as a generative tool, driver for shape, size, colour, texture, ornaments, materials, etc.
the exploration of new designs. We considered three key issues about the style defi-
Shape grammars largely explore the concepts of nition in specific shape grammars: the style identifi-
(a) style and (b) type. Style and type both charac- cation, inference and evaluation. (a) For the identifi-
terize a group of objects that share similar features, cation of the features that define a style there are two
but the first one is context-dependent and the sec- main sources: written descriptions or rules, given by
ond one is context-independent. (a) According to art or architecture historians in manuals or treatises
Chan (2000), a style can be related to one person (e.g. Yingzao Fashi in Li 2001) or guidelines given
(individual style - e.g.: Hepplewhite style in Knight by a brand identity (e.g. Harley-Davidson brand in
1980), group of persons (group style - e.g. Thonet Pugliese and Cagan 2002), and a corpus of designs
style in Barros et al. 2011), place (regional style) or pe- that are instances of the style. Notice that the first
riod of time (period style - e.g.: Queen Anne style in source may not exist, if style is not explicitly prede-
Flemming 1987). One may also include an art move- fined. (b) The inference of the style is made by encod-
ment as a combination of all features (e.g. De Stijl ing the written features and the patterns in designs
in Knight 1989). (b) According to Ahmad and Chase (given by stylistic analysis) into formal rules and pa-
(2012), types may characterize functions (e.g.: office rameters, into what is considered to be the core of the
chairs) or morphologies (e.g.: cantilever chair). Flem- style. (c) The style evaluation is made by three tests
ming (1987) noticed that style disregards differences defined by Stiny and Mitchell (1978): the grammar
of type (e.g. Thonet style is found in chairs and ta- has to (T1) recreate the designs of the corpus, (T2)
bles) and type disregards differences of style (e.g. a recreate other existing designs of the same style and
four-legged chair can be a Thonet or Hepplewhite). (T3) generate new designs within the style. The new
Moreover, the same style may be found across differ- generated designs must be plausible in appearance,
ent design domains (e.g. De Stijl is an artistic style, ar- function and symbolism (Stiny and Mitchell 1980).
chitectural style, and design style). Styles and types The application of the three tests is exemplified in
can be subdivided (e.g. the De Stijl style contains dif- Orsborn et al. (2006). The formal definition of the
ferent individual styles, each one with different stylis- style given by specific rules and parametric ranges
tic periods, which were compared in Knight 1989). may or not confirm the preliminary understanding.
Among style, other specific languages are character- Several issues arise with this definition of style
ized by shape grammars. It is the case of the brand by shape grammars. Ahmad and Chase (2012) argue
identity (e.g. Pugliese and Cagan 2002) and prod- that a language may contain designs that not belong
ucts families or collections (Castro e Costa and Duarte to a style, may not contain all designs from a style,
2014). and omit style aesthetic qualities. The style definition
is frequently incomplete, either the definition of his-
340 | eCAADe 34 - SHAPE, FORM AND GEOMETRY | Grammars and Concepts - Volume 2
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 Table 1 summarizes the application of these rules
inside it. and parametric ranges in the different types of gram-
mars. (1) Grammar A uses R1 and/or R3 rules with
Figure 3
parametric ranges restricted to A [a1, a2]. (2) Gram-
Venn diagram of
mar B uses R2 and/or R3 rules, with parametric ranges
two overlapping
restricted to B [b1, b2]. (3) The Common grammar
sets.
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
Consider these two sets as being two specific gram-
rules, new parameters, and/or the relaxation of para-
mars, grammar A and grammar B. If grammar A and
metric ranges (for e.g. into an infinite interval).
grammar B have some overlapping features, their
The application in this process will be detailed in
combination will derivate into more three grammars:
the next section, using two specific grammars as an
the Common Grammar - which characterizes an am-
example. The clarification of the differentiation be-
biguous language that can either be A or B, the Hy-
tween specific languages can help in the develop-
brid Grammar - which characterizes a language that
ment of new designs. This reasoning for the gener-
mixtures features of A and B, and the Generic Gram-
alization of two specific grammars can be extrapo-
mar - which characterizes a more generic grammar
lated to the generalization of n specific grammars to
that contemplates more than both A and B. Notice
a generic grammar (introducing more sets) and also
that we not considered the Distinctive Grammars, as
of n designs to one specific grammar (when each set
they cannot generate complete designs.
represents a design).
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 MULTIPURPOSE CHAIR GRAMMAR
language A but not in the language B. (R2) Specific A parametric generic grammar for multipurpose
rules of B are used to generate designs in the lan- chair design was developed in four stages, listed in
guage B but not in the language A. (R3) Common or Table 2. The stage 1.1 comprised the development
unrestricted rules are applicable to all the specific lan- of an analytic grammar for multipurpose chairs of
guages, and thus can be used to generate designs in one designer (Daciano da Costa). The stage 1.2 em-
both A and B languages. (R4) New rules are not used braced the extension of the previous grammar in or-
to recreate any specific language. der to incorporate the generation of multipurpose
Table 1
Restrictions in rules
and parameters of a
generic grammar.
SHAPE, FORM AND GEOMETRY | Grammars and Concepts - Volume 2 - eCAADe 34 | 343
Table 2
Evolutionary stages
of the multipurpose
chair grammar.
chairs of a second designer (Jasper Morrison), orig- into simpler rules) and deleted (by generalization of
inating a more generic grammar that characterizes equivalent rules). For instance, notice that the Da-
multipurpose chairs with wooden frame. The stage ciano Grammar on stage 1.2 has a bigger number of
2.1 involved the development of a generic gram- rules than in the stage 1.1. The grammar in stage 1.1
mar from the analysis of twenty-six chairs of different comprises a corpus of one designer and includes 23
types. This stage emphasized types to the detriment rules, while in the stage 1.2 comprises a corpus of
of styles, and thus is able to generate a significantly two designers and contains 44 rules. The develop-
larger number of solutions. The stage 2.2 consid- ment of the ontology followed the development of
ered the incorporation of an existing specific gram- the grammar, where parts were added and the se-
mar (Barros et al. 2011) to the grammar developed mantics readjusted.
in the last stage. The first two stages characterize
two individual styles, the third different types of de- Figure 4
signs, and the fourth a product family (the variation Domain
among the designs occurs only in the inner frame of decomposition.
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 Corpus
2000). The grammar was developed through the The criteria for the corpus selection followed a hierar-
process of transformation (Knight 1989) of the previ- chy of sets (Figure 4), where each set contains all the
ous analytic grammar, by rule addition, rule change, sets below it. Each level corresponds to a category,
parameter addition and parameter change (relax- which may contain more than one set. The product
ation of parametric ranges). As mentioned in Cas- design domain is the more generic level, and was de-
tro e Costa and Duarte (2014), the transformation composed into more specific levels until the bottom
also involves the reformulation of the original rules, level of individual chairs. Some levels could be or-
where rules were added (by decomposition of rules dered differently, other levels could be added, and
344 | eCAADe 34 - SHAPE, FORM AND GEOMETRY | Grammars and Concepts - Volume 2
some sets can be inferred from existing ones. Fig- Rules
ure 4 distinguishes between categories of type (light The initial shape of the grammar represents the seat
grey) and categories of style (dark grey). Notice that and back planes and the bounding box underneath
types refer to nouns with an initial lower letter (com- the seat where the legs are placed (see the left-hand
mon noun) and styles refer to nouns in an initial cap- side of the rules in Figure 5). The initial shape is pa-
ital letter (proper noun), therefore one may consider rameterized accordingly to the anthropometric stan-
types as being more generic than styles. dards for chairs. The rules are successively applied
The corpus of the grammar is composed by ten to the initial shape, by adding parameterized parts
chairs, five from the Portuguese designer Daciano da of the chair, until a final solution is reached. Those
Costa (1930-2005) and five from the English designer parts are arranged in six groups, which correspond
Jasper Morrison (1959-). The chairs of Daciano (Mar- to different functional regions of the chair: Legs, Seat,
tins 2001) date from 1966-1977 and were designed Back, Stretchers, Base and Arms. Rules only draw the
for the interiors of four Portuguese hotels, being all left side of the designs, as they are bilaterally sym-
produced by a Portuguese company (Móveis Sousa metrical. Figure 5 shows the two rules of the Legs,
Braga). The chairs of Morrison (Morrison 2015) date respectively the Front Leg and the Back Leg. Each
from 1988-2011 and were produced by three com- leg has its endpoints parameterized; four parame-
panies - Cappelini (Italy), Murani (Japan) and Vitra ters correspond to the width and depth position in
(Switzerland). The chairs are all from the multipur- the seat plane and the splay and rake angles. Table
pose type and have wooden frame. The chairs have 4 shows the application of those rules and parame-
different typologies, such as the number of legs, the ters to the grammars (Daciano, Morrison, Common,
presence or lack of arms, the number of stretchers, Hybrid and Generic), following the process described
the seat shape, and so forth. Table 3 lists the chairs before for the overlapping of two specific grammars.
of the corpus, chronologically ordered, plus two ex- The generic parametric ranges were defined in the
tra chairs from the control group. For the purpose stage 2.1. Legs include another rule that adds cur-
of the grammar, the corpus was simplified into lines vature (used in three Morrison chairs), but this was
(that represent the frame) and planes (that represent deleted in the stage 2.1, as only curvatures in corners
the seat and back pieces). The solid shape gener- were considered. In this case, the more generic is the
ated from the structure was introduced latter. Figure grammar, the less detail their rules tend to have.
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
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- statement is illustrated in Figure 6 by the chair D4 and
long to the specific language. Figure 6 depicts two the chair D8, which belong to the same style (Daciano
new designs for each style (D7 and D8, M7 and M8). style) but represent two different types of chairs (side
Test 4: the grammar generates new designs from chair and armchair). Thus, one can identify products
the Ambiguous style (A1 and A2). This is a very re- of the same family from existing designs, for instance,
stricted style, with the smallest solution space from the chairs D2 and D5 from the corpus. On another
the ones analyzed in this paper. hand, the second statement is illustrated in Figure 6
Test 5: the grammar generates new designs from by the chairs D4 and M2, which belong to different
the Hybrid style (H1 and H2). The chair H1 is the clos- authors but have similar configurations, thus may in-
est to the chair Thonet No. 14 (Barros et al. 2011) that dicate related influences.
this grammar can generate. Each individual style is defined by the restric-
Test 6: the grammar generates new designs in tion of rules and parametric ranges. The compari-
the Generic style. As show in Figure 6, this includes son of two individual styles was made by comparison
new types of chairs as monolithic base chairs (G1) and of rules and parametric ranges, defining specific fea-
cantilever chairs (G2). G3 represents the chair Thonet tures used by each one, common features, and hybrid
No. 14 (Barros et al. 2011) in its generic form. features. However, these characteristics do not con-
A qualitative characterization of the two de- clusively describe one designer's style, only by the
signer styles was made by comparison of the spe- analysis of few products and few design aspects. The
cific rules and the specific parameter ranges, for each test 2 failed (the grammar could not generate other
group of parts. (1) Daciano style (D) uses four and designs of the same style), which means that either
three-legged chairs, while Morrison style (M) always (a) five designs are not enough to define a designer's
uses four legs; none of them uses one or two-legged style or (b) the designer does not have a consistent
chair. D legs are vertical while M legs have slight an- style. Moreover, the identification of style in chairs
gles and curvatures. (2) The seat front radius in D is with a high degree of abstraction may be over sim-
0 while in M varies between [0,60]; and the seat rear plifying. Future work should include the evaluation
radius in D is {0,100} while in M varies between [0,80]. of the style in more detailed chairs by other persons
Both D and M uses all types of seat rails except the X- beyond the grammar developer, for instance by the
shaped ones. (3) Both vary between an open back authors of the corpus that are still alive (in the present
and a solid back, using back outer frame elements case only Jasper Morrison) and by their collaborators
but not inner frame elements. (4) Only one chair in in the design studio, so that one can conclude if the
M style uses stretchers. (5) Only one chair in D style new generated designs are actually in the style and
uses base rails (a side rail). (5) In D style, arms are sup- what is their degree of style (Chan 2000).
ported by the front or/and back leg extension; while Comparisons can also be made between two
in the only example of an M armchair, arms are sup- methods for designing generic grammars that cor-
ported by the back leg extension. As in seat rails, the respond to the stage 1.2 and the stage 2.1 (Table 2).
radius of the armrest arc is in D half of its length, while The first method is the one presented in this paper
in M this proportion varies. One can conclude that and comprises an extension of a specific grammar to
the Daciano style uses more straight rigid lines and a generic one. The second method is the one pre-
proportions than the Morrison style. sented in Garcia and Romão (2015) and comprises
design of a generic grammar from the analysis of a
DISCUSSION corpus of designs with different types. Although the
The same style may describe different types, and the second method covers a greater variety of types of
same type may describe different styles. The first designs (as one can observe by the chairs H1 and G3
SHAPE, FORM AND GEOMETRY | Grammars and Concepts - Volume 2 - eCAADe 34 | 347
in Figure 6), and thus has a wider solution space, is and Franco, J (eds) 2015, The next city - New technolo-
does not ensure that it can predict all the possible gies and the future of the built environment [16th In-
types. In fact, there are two rules in the stage 1.2 that ternational Conference CAAD Futures 2015. Sao Paulo,
July 8-10, 2015], Springer, pp. 600-619
are not applicable in the stage 2.1. The combination
Knight, T 1980, 'The generation of Hepplewhite-style
of existing specific grammars of the same class would chair-back designs', Environment and Planning B:
be an interesting field to explore in the future. 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-
REFERENCES vironment and Planning B: Planning and Design, 16,
Agarwal, M and Cagan, J 1998, 'A blend of different pp. 51-98
tastes: the language of coffeemakers', Environment Lee, HC and Tang, MX 2004 'Evolutionary Shape Gram-
and Planning B: Planning and Design, 25, pp. 205-226 mars for Product Design', Seventh International Con-
Ahmad, S and Chase, S 2012, 'Style representation in de- ference on Generative Art, Politecnico di Milano Uni-
sign grammars', Environment and Planning B: Plan- versity, pp. 309-322
ning and Design, 39, pp. 486-500 Li, A 2001, A shape grammar for teaching the architectural
Barros, M, Duarte, J and Chaparro, B 2011 'Thonet Chair style of the Yingzao fashi, Ph.D. Thesis, Massachusetts
Design Grammar: A Step towards the Mass Cus- Institute of Technology
tomization of Furniture', CAAD Futures 2011: Design- Martins, JP (eds) 2001, Daciano da Costa: designer, Fun-
ing Together, pp. 181-200 dação Calouste Gulbenkian, Lisboa
Benrós, D, Hanna, S and Duarte, JP 2012 'A generic shape Morrison, J 2015, A book of things, Lars Müller, Zürich
grammar for the Palladian Villa, Malagueira House Orsborn, S, Cagan, J, Pawlicki, R and Smith, RC 2006, 'Cre-
and Prairie House', Design Computing and Cognition, ating cross-over vehicles: Defining and combining
Texas, USA, pp. 321-340 vehicle classes using shape grammars', Artificial In-
Chan, CS 2000, 'Can style be measured?', Design Studies, telligence for Engineering Design, Analysis and Manu-
21, pp. 277-291 facturing, 20, pp. 217-246
Castro e Costa, E and Duarte, JP 2014 'Generic Shape Pugliese, M and Cagan, J 2002, 'Capturing a rebel: mod-
Grammars for Mass Customization of Ceramic Table- eling the Harley-Davidson brand through a motor-
ware', Design Computation and Cognition DCC, pp. cycle shape grammar', Research in Engineering De-
437-454 sign, 13, pp. 139-156
Duarte, JP 2001, Customizing Mass Housing: A Discursive Stiny, G 1980, 'Kindergarten grammars: designing with
Grammar for Siza's Malagueira Houses, Ph.D. Thesis, Frobel's building gifts', Environment and Planning B:
Massachusetts Institute of Technology Planning and Design, 7, pp. 409-462
Duarte, JP and Beirão, JN 2011, 'Towards a Method- Stiny, G and Mitchell, W 1978, 'The Palladian gramar', En-
ology for Flexible Urban Design: Designing with vironment and Planning B: Planning and Design, 5,
Urban Patterns and Shape Grammars', Environment pp. 5-18
and Planning B: Planning and Design, 38(5), pp. 879- Stiny, G and Mitchell, W 1980, 'The grammar of paradise:
902 on the generation of Mughul gardens', Environment
Eloy, S and Duarte, JP 2012 'A transformation grammar- and Planning B, 7, pp. 209-226
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
Wood Mass-Customized Housing
A dual computer implementation design strategy
SHAPE, FORM AND GEOMETRY | Grammars and Concepts - Volume 2 - eCAADe 34 | 349
and several shape grammars have addressed the WOOD MASS-CUSTOMIZED HOUSING
housing problem and the need to make available de- Choice of the house construction system has been
sign systems that respond to the inhabitants needs determined by its popularity in some european coun-
(Duarte 2001; Colakoglu 2005; Benros et al. 2011; tries as well as the possibility of its prefabrication off
Eloy 2012; Coimbra and Romão 2013). The main dif- site and partial automation of manufacturing pro-
ficulty to fully use these systems in design practice cess. In order to facilitate mass-customization cho-
is the delay of the computer implementation of ar- sen construction system have to enable construction
chitectural design processes, namely by the use of flexibility and permit just-in-time production. The
shape grammar logics. Computerized design tools use of industrialized methods of production allows to
that offer design alternatives supporting the design save time and permit cheaper costs and quality con-
ways of thinking and working are still rare. trol. For these reasons a light wood frame construc-
Current systems allow some aspects of genera- tion was chosen as the main structure of the building.
tion of shape and are based on a built-in drawing ed- In consequence developed shape grammar reflects
itor, e.g. Rabo-de-Bacalhau grammar tool (Strobbe the capabilities and constraints of the chosen build-
2015) or in links to a CAD software, e.g. Grappa (Grasl ing system.
2012) for Revit (McKay et al. 2012). These systems Chosen building technique uses vertical and hor-
are essentially academic based and not yet aiming at izontal structural members, studs, which provide a
being used in the real design scenario. Design ap- stable frame to which interior and exterior wall cov-
plications like Flemming (1987) Queen Anne Houses erings are attached. The structures are strengthened
tool, Li (2002) Yingzao fashi grammar, Grasl (2012) with rigid panels of oriented strand board (OSB) used
Grappa tool and Ligler and Economou (2015) Ent- to form wall sections. Sections of the walls can be
elechy Grammar are systems that allow to develop prefabricated off-site therefore the whole building
designs according to languages of past styles. Those could be erected in just a few weeks.
tools are used specially in an historical and design un- Wall framing in light frame construction system
derstanding perspective and not as tools to design includes the vertical and horizontal members of ex-
new solutions for architecture problems. Other im- terior walls and interior partitions (see Figure 1). The
plementations like Strobbe (2014) Rabo-de-Bacalhau walls are build with the use of standardized studs
grammar tool allow to generate new design solutions which are placed in the axial spacing of 60 cm. This
that can then be used by the house owners to do their structural guideline is encoded in the shape gram-
refurbishment. The goal of the presented research is mar rules. Therefore the shape rules are operating
to deliver such a tool that can deliver designs able to on rooms which dimensions are a multiplication of
be immediately used as a design project to build a 60x60 cm module. Moreover the maximum permis-
house. sible span of the ceiling, up to 6 m, affects the choice
Catalogues of house designs are very popular in of the house typology, therefore also the localization
several countries. Future owners are provided a cat- of load bearing walls as well as the width of the build-
alogue of standardized solutions for them to choose ing in general and rooms in particular.
but those solutions are restricted to a small number The possible design solutions include detached
of changes and the act of choosing is based on look- houses that can have either one or two floors, a com-
ing to all solutions and not to the ones that may re- pact rectangular floor plan, where only the garage
spond to the clients need. In this research we aim sticks out of the rectangular perimeter. The houses
at looking for systems that allow users to play with have a fixed width, a center axis with a right and left
a house design tool allowing them to find the best fit side of the same width, and the house length may
house to their own personal wishes. vary. The position of the houses may be parallel or
350 | eCAADe 34 - SHAPE, FORM AND GEOMETRY | Grammars and Concepts - Volume 2
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 of shapes positioned in the same coordinate system.
design standards. In order to preserve characteristics Processes use the usual procedural notions for cap-
as privacy, conditions for placing rooms are defined turing sequences, alternatives and property tests for
which suggests e.g. placing the social areas closer the application of shape rules and conditional shape
to the entrance and the private ones deeper in the rules during the design process.
house. To give an idea of the computational model used
The developed grammar uses the algebra U12 in this shape grammar implementation, for produc-
with bi-dimensional, linear base, floor plan of the ing houses layout design solutions, the following ba-
building. Automatically generated house layout so- sic shapes (see Figure 3) are considered to represent
lutions based on client's inputs and a set of design front and back markers and 0.6x0.6 m cells with dif-
rules enable clients to participate in the design pro- ferent room occupation:
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 Each identifier represents the picture positioned in
brief and generating new solutions fulfilling modified the respective coordinate system xOy.
user requirements. Positioning a shape defined in this way within a
different coordinate system x'O'y' is represented by a
IMPLEMENTATION pair (shape, transformation) specifying the transfor-
Two different implementations for the design sys- mation (translation, rotation, scale, etc.) required for
tem are presently under development. One based positioning the coordinate system xOy associated to
on shape grammars supplemented with processes the defined shape into the new coordinate system
detaches shape knowledge from procedural knowl- x'O'y'.
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 For instance, in Figure 4, cell is positioned into the co-
a simplified form of sub-shape detection. Shapes ordinate system x'O'y' by a scale S(c,d) and rotation
are represented by identifiers associated with pic- R(q) followed by a translation T(a,b) and thus is rep-
tures and positioned in a coordinate system refer- resented by (cell, T(a,b)R(q)S(c,d))).
ring additionally to the applied Euclidian transforma- Using these basic shape representations, the fol-
tion. Shape compositions are represented by sets lowing shape compositions (Figure 5)
SHAPE, FORM AND GEOMETRY | Grammars and Concepts - Volume 2 - eCAADe 34 | 353
,→ placeSingleBedRoom2 or ( Figure 5
,→ placeSingleBedRoom1 and An exemplary
,→ placeSingleBedRoom2 ). shape composition.
are represented respectively by: homeRestrictions = (next(vestibule ,
,→ garage) && next(toilet ,garage) &&
basement3x3 = {(cell ,Id), (cell ,T ,→ next(technicalRoom ,garage) && (
,→ (0.6 ,0)), (cell ,T(1.2 ,0)), (cell ,T ,→ next(fron ,homeOfice) || next(fron ,
,→ (0 ,0.6)), (cell ,T(0.6 ,0.6)), (cell ,→ homeOfice) || next(fron ,homeOfice)
,→ ,T(1.2 ,0.6)), (cell ,T(0 ,1.2)), ( ,→ ) && (next(diningRoom ,fron) ||
,→ cell ,T(0.6 ,1.2)), (cell ,T(1.2 ,1.2) ,→ next(dinningRoom ,Kitchen)) && (
,→ )}, vestibule3x3 = {(cves ,Id), ( ,→ next(homeOfice ,fron) || next(
,→ cves ,T(0.6 ,0)), (cves ,T(1.2 ,0)), ( ,→ homeOfice ,livingRoom)) && (next(
,→ cves ,T(0 ,0.6)), (cves ,T(0.6 ,0.6)), ,→ bathRoom ,doubleBedroom) || next(
,→ (cves ,T(1.2 ,0.6)), (cves ,T(0 ,1.2) ,→ bathRoom ,singleBedroom)) && some(
,→ ), (cves ,T(0.6 ,1.2)), (cves ,T ,→ doubleBedroom ,next(doubleBedroom ,
,→ (1.2 ,1.2))}. ,→ back)) && some(singleBedroom ,next(
,→ singleBedroom ,back)) && (next(
Conditional Shape Grammar Rules are pairs of shape
,→ kitchen ,fron) || next(kitchen ,
compositions plus a condition allowing it applica- ,→ dinningRoom) || next(kitchen ,
tion. The following shape grammar rule (see Fig- ,→ homeOfice)) && (next(livingRoom ,
ure 6) is represented by placeVestible with pre- ,→ fron) || next(livingRoom ,kitchen)
condition (next(matched(basement3x3),fron), base- ,→ || next(livingRoom ,dinningRoom)).
ment3x3, vestibule3x3).
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
Using identifiers for shape grammar rules allow its is followed in the design process. Operationally, the rule.
reference in processes describing the design devel- design process applies repeatedly shape rules to the
opment. Processes may also have names allowing it intermediate shapes obtained at each step accord-
reuse in other processes. The following processes de- ing to the sequence order established in the process.
scribe the steps for producing houses layout design Due to non-determinism, each application of the pro-
solutions: cess to an initial shape composition may produce dif-
home = placeBasement; placeEntrance; ferent designs by forward chaining using some op-
,→ placeSemipublic; placePrivate ; erational preference in the choice of the alternatives.
,→ verify(homeRestrictions). Each time a test process fails or a shape grammar rule
placeEntrance = placeVestibule; fails to apply, the system backwards trying to build a
,→ placeGarage; different solution.
,→ placeTechRoomandToilet . The tree of Figure 7 exemplifies the generation
placeSemipublic = placeHomeOffice and process of houses layout design included in circles
,→ placeKitchen and placeDiningRoom
fulfilling the design rules present so far for a 17x14
,→ and placeLiving Room.
placePrivate = placeDoubleBedroom and
module house with the following rooms and dimen-
,→ placeSingleBedrooms and sions: 3x3 vestibule, 3x3 technical room, 9x6 garage,
,→ placeBathroom. 3x2 toilet, 3x3 storage, 2x7 kitchen, 4x7 dining room,
placeSingleBedRooms = 5x7 living room, 4x7 double bedroom, 4x7 single
,→ placeSingleBedRoom1 or bedroom, 4x7 single bedroom with hall and 3x5
354 | eCAADe 34 - SHAPE, FORM AND GEOMETRY | Grammars and Concepts - Volume 2
Figure 7
An example of the
generation process
of house layout
design.
SHAPE, FORM AND GEOMETRY | Grammars and Concepts - Volume 2 - eCAADe 34 | 355
Figure 8
Alternative Shaper
Workbench.
356 | eCAADe 34 - SHAPE, FORM AND GEOMETRY | Grammars and Concepts - Volume 2
Figure 9
Genetic Algorithm
Implementation.
thus measuring the aptitude of a generated layout in fixed orientations. The experimentation results are
terms of rules' fulfillments. The areas and rooms rel- very favorable, with Shaper-GA evolving to feasible
ative positioning rules are incorporated into the fit- floor plans (two examples on the left-hand side of
ness function and the violation of any of them results Figure 9) within reasonable computational time. The
in a fitness penalization. The algorithm also needs visual interface will display all of the feasible layouts
the specific user information on the floor plan to be (left and right arrows at the bottom of Figure 9). Also
generated, namely, which rooms and side dimen- any of the remaining (lesser) fitted solutions found
sions are to be inserted to begin the generative pro- in the final evolutionary generation can be visual-
cess and design evolution towards feasible layouts. ized (example on the right-hand side of Figure 9).
This approach, Shaper-GA, begins by randomly The interface is prepared to show product layouts
generating individuals (layouts) where each com- where some of the relative rooms' positioning rules
prises the predefined needed rooms. However, be- have been relaxed thus allowing for the emergence
ing random, most of these layouts will not be fea- of other designs where the user might see fit.
sible because of overlaps or rules' failure of compli-
ance. Through the referred incorporation of penal- RESULT AND DISCUSSION
ties in the fitness function and usage of this knowl- Both mentioned implementations are actually sup-
edge in parental selection and for reproduction, the porting test experiments for generating house lay-
best fitted individuals (houses' layouts that best fol- outs that comply with a language of design and re-
low the design rules) are selected and their genes spond to user requirements. Up to now, for small
propagated to the offspring. From the overall set of houses' layouts, the process shape grammar imple-
parents and their offspring, the better fitted houses mentation produces solutions in just a few seconds,
are chosen for the next generation, Thus, the re- faster than the genetic algorithm shape grammar im-
peated process evolves the better layouts until being plementation. However, the number of potential so-
able to generate completely feasible floor plans. lutions is combinatorial on the number of divisions
Shaper-GA has undertaken the necessary trial ex- and their side dimensions. Thus, when the number of
periments with fixed specifications much alike the ex- divisions increase and their sizes change, we are ex-
ample described in the previous section: all the re- pecting that the process shape grammar implemen-
quired rooms have fixed dimensions, and thus also tation will take a considerable more amount of time
SHAPE, FORM AND GEOMETRY | Grammars and Concepts - Volume 2 - eCAADe 34 | 357
just to produce a solution. That's where the genetic Duarte, JP 2001, Customizing mass housing: a discursive
algorithm counterpart plays an important role. grammar for Siza, Ph.D. Thesis, Massachusetts Insti-
Shaper-GA implementation is being upgraded tute of Technology
Eloy, S 2012, A transformation grammar-based methodol-
to allow for evolutionary readjustment of the rooms
ogy for housing rehabilitation, Ph.D. Thesis, Universi-
dimensions, according to predefined interval range dade Técnica de Lisboa
values. Meanwhile, the best way to integrate the Eloy, S, Vermaas, P and Andrade, M (s.d.), 'The Quality of
connections between rooms into the evolutionary Designs by Shape Grammar Systems and Architects:
generation is also being studied. More research will A Comparative Test on Refurbishing Lisbon', Journal
be done in order to develop fully efficient prototype of Architecture and Planning Research, (forthcoming)
Flemming, U 1987, 'More than the sum of parts: the
software for both approaches. For that we'll be using
grammar of Queen Anne houses', Environment and
heuristics to find paths of promising generating de- Planning B: Planning and Design, 14(3), pp. 323-350
signs and speeding up the algorithm so that several Grasl, T 2012, 'Transformational Palladians', Environment
optima are quickly found. One additional approach and Planning B: Planning and Design, 39(1), pp. 83-95
will also be explored for the cases where there are Knecht, K and Konig, R 2010, 'Generating Floor Plan Lay-
non-optimal solutions, i.e. there are no solutions sat- outs with K-d Trees and Evolutionary Algorithms',
Generative Art Conference, , pp. 239-253
isfying all the shape conditions. For that situation the
Kwiecinski, K and Slyk, J 2014 'System for customer par-
genetic algorithm approach will be used to integrate ticipation in the design process of mass-customized
a multi-criteria representation of customer's prefer- houses', Fusion: Data Integration at its Best, Vol 2, pp.
ences. 207-215
The results found so far, make us believe that I-kang Li, A 2002 'A prototype interactive simulated
these generative tools are promising ones that satisfy shape grammar', Design education: Connecting the
Real and the Virtual, pp. 314-317
the users' needs and can be used by non designers in
Ligler, H and Economou, A 2015 'Entelechy I: Towards
order to explore housing design solutions. a Formal Specification of John Portman's Domestic
Architecture', Real Time - Extending the Reach of Com-
ACKNOWLEDGEMENTS putation, Volume 1, pp. 445-452
McKay, A, Chase, S, Shea, K and Chau, HH 2012, 'Spa-
This work was partially supported by Center For
tial grammar implementation: From theory to use-
Advanced Studies Warsaw University of Technology able software', Artificial Intelligence for Engineering
through Outgoing Scholarship No. CAS/37/POKL. Design, Analysis and Manufacturing, 26(02), pp. 143-
159
Pasternak, A and Kwiecinski, K 2015 'High-rise Building
REFERENCES Optimization A Design Studio Curriculum', Real Time
Benros, D, Granadero, V, Duarte, JP and Knight, T 2011, - Extending the Reach of Computation, Volume 1, pp.
'Automated Design and Delivery of Relief Housing: 305-314
The Case of post-Earthquake Haiti', CAAD Futures: Santos, F and Esmerado, J 2015, 'A different shape gram-
Designing Together, pp. 247-263 mar approach for automatic design generation', In-
Caldas, LG and Norford, LK 2002, 'A design optimization ternational Journal of Advances in Computer Science
tool based on a genetic algorithm', Automation in & Its Applications, 1, pp. 90-97
Construction, 11(2), pp. 173-184 Steadman, P 2008, The Evolution of Designs: Biological
Coimbra, E and Romao, L 2013 'The Rehabilitation De- Analogy in Architecture and the Applied Arts, Rout-
sign Process of the Bourgeois House of Oporto: ledge
Shape Grammar Simplification', eCAADe 2013: Com- Stiny, G 1980, 'Introduction to shape and shape gram-
putation and Performance mars', Environment and Planning B: Planning and De-
Colakoglu, B 2005, 'Design by grammar: an interpreta- sign, 7, pp. 343-351
tion and generation of vernacular hayat houses in Strobbe, T 2015, Computer-Aided Exploration of Architec-
contemporary context', Environment and Planning B: tural Design Spaces: a Digital Sketchbook, Ph.D. The-
Planning and Design, 32(1), pp. 141-149 sis, University of Gent
358 | eCAADe 34 - SHAPE, FORM AND GEOMETRY | Grammars and Concepts - Volume 2
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.
THE ALGORITHM simple facade elements with fixed main outlines and
The complexity of a given architectural object or ob- two openings with varying corner-points. The co-
ject arrangement can be measured on the level of ordinates of the corner-points are translated into bi-
gradient relations, the gradients itself being the rep- nary DNA (1st step: encoding). After using crossover
resentations of the proportion of each rectangle that (2nd step; with or without fitness check beforehand)
is defined by each pair of geometric points. Using and mutation (3rd step) on a population of n-parents
the gradient analysis means regarding the frequency (chromosomes), angle quotient and length quotient
of repetition of gradients, comparing every signifi- are used as fitness-values (4th step) to determine the
cant point with all the other points in a 2D CAD rep- segment-size on Goldbergs weighted roulette wheel
resentation of the object one by one and succes- (see Coates 2010, König 2010). By digitally spin-
sively listing pairs of points and their gradients (Kul- ning the roulette wheel, members for the succeed-
cke and Lorenz 2015). In addition the distance be- ing breeding population are picked to be proposed
tween points is taken into account; the gradient anal- (element of chance); out of these the user choses
ysis returns an angle quotient and a length quotient, 1 or 2 favourites. Taking this choice into account
each of which can be used as a fitness-value within the next breeding population goes through the algo-
a genetic algorithm. This is tested by the authors on 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).
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 algorithm is rewritten in PHP and using the <canvas>
returns a set of designs starting with the initial in- element in HTML5 (Figure 6).
dividuals in the left column, with fitness decreasing Left uninfluenced in the individual evolutionary
from top to bottom row. From left to right differ- steps, the genetic algorithm aiming for the lowest
ent time steps are visualised with a finishing result angle-quotient possible applying the gradient anal-
on the right. Differences between elements of the ysis as a fitness value shows certain tendencies that
last column are minimised as a (local) optimum is might not be wished for in a real building environ-
reached (Figure 4).The initial aim to reach a design ment, e.g. window/door-openings tend to go for ex-
that produces a minimal angle-redundancy quotient treme i.e. slim proportions resp. decrease in the size
and a minimal length-redundancy quotient is ques- of their area. In a fully automated context this would
tioned by the results. Left unchecked by rules and call for corrective rules and constraints, in the case
constraints this aim produces openings decreasing in of an interactive genetic algorithm avoiding extreme
size, which is the logical consequence if the only goal proportions may be left to the user, if he/she wishes
is to escape complexity in proportion as much as pos- to avoid them. In addition the degree of complexity
sible (Figure 5). reduction could also be chosen by the user adjusting
the angle-quotient; an ideal angle-quotient could be
Figure 5 set by the user, either beforehand or in the course of
Openings searching for the facade wished for. Functions could
decreasing in be added accordingly to the proposed IGA.
size/tending toward
extreme
proportions.
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.
THE FITNESS FUNCTION AND USER
Currently the determination of the best position of
CHOICE INTEGRATION user intervention within the GA, thus making it inter-
The integration of gradient analysis in an interactive active, is still ongoing. An advantage of using inter-
genetic algorithm aims rather at didactic and pro- active genetic algorithms instead of fully automated
fessional application as opposed to customer-use, ones is that fewer iterations are needed to come to
therefore it should be possible to optionally add the a conclusion, but more importantly a dialogue be-
fitness-values of each individual of the population tween the algorithm and the user is introduced. With
the user can choose from in text. This is done in a the integration of Gestalt related methods like the
two step process; first, the favourites are chosen by gradient analysis, interactive genetic algorithms be-
visual preference, than the fitness-value representing come didactic instruments on the aesthetic and cog-
the degree of complexity is made to appear and the nitive level, combining theory and design practise.
user can alter his decision, which is now influenced To further develop an asynchrone dialogue be-
by intuition, perception and analysis-output. To al- tween designers and users in a form that conse-
low for comfortable interactivity and easy access the
SHAPE, FORM AND GEOMETRY | Grammars and Concepts - Volume 2 - eCAADe 34 | 363
Figure 6
The algorithm in
PHP and HTML5.
quently upholds the initial idea of user-involvement of Complexity in the Design of Architecture and
through an IGA that still allows for professionally re- Designobjects', Martens, Bob; Wurzer, Gabriel; Grasl,
sponsible solutions, a next step will be to list and for- Thomas; Lorenz, Wolfgang; Schaffranek, Richard. Real
Time – Proceedings of the 33rd International Confer-
mulate necessary rules and constraints that can be
ence on Education and Research in Computer Aided
used for an accompanying or closing analysis of the Architectural Design in Europe, Wien, pp. 415-424
(final) instances that are created by the genetic algo- König, R 2010, Simulation und Visualisierung der Dy-
rithm in collaboration with the users' choices. namik räumlicher Prozesse, VS Verlag für Sozialwis-
senschaften, Wiesbaden
Poirson, E, Petiot, J, Aliouat, E, Boivin, L and Blumenthal,
REFERENCES D 2010 'Study of the convergence of Interactive Ge-
Coates, P 2010, The Programming of Architecture, Rout- netic Algorithm in iterative user’s tests: application
ledge, New York to car dashboard design', Proceedings of IDMME - Vir-
Kulcke, M and Lorenz, W 2015 'Gradient-Analysis: tual Concept 2010, Bordeaux
Method and Software to Compare Different Degrees
364 | eCAADe 34 - SHAPE, FORM AND GEOMETRY | Grammars and Concepts - Volume 2
3D Reconstruction Survey of Complex Informal
Settlements
Towards an understanding of the genesis of form
SHAPE, FORM AND GEOMETRY | Grammars and Concepts - Volume 2 - eCAADe 34 | 365
growing process, and ultimately, develop guidelines most flat. The latter case addressed Rocinha, one of
for the design of planned settlements in similar to- the largest favelas in Rio that shares the steep site
pographic conditions. The research uses the favela features of favela Dona Marta. However, due to its
Dona Marta as a case study and encompasses the large size, Rocinha does not have a homogeneous ur-
methodological steps described in the next section. ban fabric and still poses security challenges, which
There were several reasons for choosing favela makes it more difficult to study. This meant that the
Dona Marta. It is located in Rio de Janeiro and it has study could not rely on an accurate survey of the
had a big media exposure since the 1990s. It ap- site. To avoid this limitation, the current study uses
peared in two video clips (They don't care about us, cutting-edge 3D reconstruction technology to survey
from Michael Jackson and Put in a love song, from Be- the pacified favela, which is the focus of this paper.
yoncé and Alicia Keys) and two movies (the Brazilian The automated image-based 3D reconstruction of
Elite Squad, the enemy within and the North Ameri- sites and buildings has its roots in the work by Noha
can Fast Five.) This exposure to the public makes of Snavely, which resulted in the open source Bundler
favela Dona Marta a paradigm of the image that peo- software (Snavely 2006). This software implements
ple have formed regarding Brazilian favelas. More- the structure-from-motion approach, in which large
over, it is located in the south part of the city, near collections of images can be oriented in a single co-
a prime area and close to Christ Redeemer, an im- ordinate frame to produce a sparse 3d point cloud
portant Brazilian monument. Not least important, model. Then the bundle of images can be used to
this favela has been pacified, which enabled us to es- produce a dense point cloud model. One of the first
tablish links with the Residents Association and, ulti- tools for this final procedure was the patched-based
mately, have access to the favela. multi-view stereo (PMVS) and the clustering views for
multi view stereo (CMVS) software packages devel-
STATE OF ART oped by Furukawa et al. (Furukawa 2014). After this
The ultimate goal of the research is to build a com- work, other tools have been developed that take ad-
putational model of favela Dona Marta. The idea is vantage of the possibility of using the graphics pro-
to capture the rules underlying the genesis of the cessing unit (GPU) [1] and multi-core processing (Wu
form into a shape grammar, while understanding the 2011).
relations between such form and physical site fea-
tures, socio-economic conditions, and technological METHODOLOGY
aspects, including the availability of materials. The The first methodological step is concerned with gath-
idea of using grammars to capture the generation ering and creating accurate representations of the
of urban form has its roots on previous work on the settlement, including 2D drawings and 3D models,
development of a shape grammar for the Marrakech which will permit to study the relationships between
Medina (Duarte et al. 2007). Like favelas, the Mar- site features, particularly topography, and the com-
rakech Medina has a complex shape that grew in- plex shape of the built structures. The second step
crementally over time without a clear plan; it was is to develop a shape grammar aimed at explaining
the result of local rules expressing social conventions the genesis and structure of the form. This paper is
and physical conditions, rather than the outcome of a focused on the 3D reconstruction of the case study,
top-down approach. However, the steep topography that is, on the development of a 3D digital model
associated with favelas, makes it a unique case. More using various computer-based techniques. This step
recent work has used grammars to describe Mozam- consists of two sub-steps, one that uses Google Earth
bican slums (Barros et al. 2013) and even Brazilian images, described below, and another that uses im-
favelas (Dias 2014). The former case has a signifi- ages gathered directly in the real environment, the
cantly different structure, partly because sites are al- latter to be addressed in a future article.
366 | eCAADe 34 - SHAPE, FORM AND GEOMETRY | Grammars and Concepts - Volume 2
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 the Dona Marta favela seems to be densely built, a
2D site plan of more accurate analysis permitted by the developed
favela Dona Marta 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
Finally, the 3D model and the site plan obtained were Program administered by the Brazilian National
represented in the same coordinate system to guar- Council for Scientific and Technological Develop-
antee that spatial information was aggregated, en- ment (CNPq).
abling cross-referencing. The process of taking im-
ages and creating the 3D model strictly followed REFERENCES
the steps recommended by reference authors (Kout- Barros, P, Beirão, J and Duarte, JP 2013 'The Language of
soudis 2013). The 3D model obtained as a result en- Mozambican Slums', Proceedings of the 31st eCAADe
ables one to see the volumes of the built structures Conference 2013, Delft, pp. 715-724
Bartos, P, Sobavá, W, Pukanská, K and Weiss, G 2014,
and how they go over the hill, being barred only by 'Analysis of low cost photogrammetric procedures
stones or higher slant, as initially intended. However, in the process of historical objects survey', Interna-
because data was not taken directly from the real en- tional journal of science commerce and humanities,
vironment, the level of definition is low and limited 2(3), pp. 203-210
by the resolution offered by Google Earth. 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
CONCLUSION Davis, M. 2006, Planet of Slums, Verso, New York
The creation of the slum's map was important to un- Dias, MA 2014, 'Informal Settlements: a shape Grammar
derstand the process by which public spaces were approach', Journal of Civil Engineering and Architec-
formed. Preliminary results suggest that in these ture, 8(11), pp. 1389-1395
Duarte, JP, Rocha, JM and Soares, GD 2007, 'Unveiling
kinds of unplanned, informal settlements, pathways the structure of the Marrakesh Medina: a shape
are defined by building aggregation and, in a smaller grammar and an interpreter for generating urban
degree, by the need for guaranteeing the continu- form', Artificial intelligence for Engineering Design,
ity of movement through the urban space. Although Anallysis and Manufacturing, 21, pp. 317-349
SHAPE, FORM AND GEOMETRY | Grammars and Concepts - Volume 2 - eCAADe 34 | 369
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.
SHAPE, FORM AND GEOMETRY | Grammars and Concepts - Volume 2 - eCAADe 34 | 371
the plot development in dense city centers is the pro- tal sundial designated for the moment of equinox. It
vision on the insolation of apartments. The planned illustrates the lengths and angles of the shadow cast
investment cannot significantly reduce the quality of on the full hours of local solar time by a bar with a
life of other users of the city. The analysis of the im- nominal height and its multiples. This tool, printed
pact of the planned investment on the environment on tracing paper, as it was inserted in the book "Sun
is an important step of its planning process. in Architecture", is used by superimposing it on the
analyzed building plan with the surrounding build-
Local Building Regulations: Insolation ings drown to scale.
According to the Minister of Infrastructure Regula- The sundial can be determined for certain loca-
tion on the technical conditions to be met by build- tion, but usually it was drawn for each city only once
ings and their location guidelines on lighting, insola- and it was not taking into account differences in lo-
tion and occlusion of windows are specified. In this cations within the whole area of the city (Figure 1).
article only aspects related to insolation of premises The resulted in the need for some amendments for
are discussed and the process of their analysis is de- each analyzed location or taking large safety margins
veloped in detail. (Lose 2015). In densely built-up locations it is not al-
In the above-mentioned provisions, the issue of ways possible to accept ranges of safety beyond the
insolation of premises is expressed in the number of minimum required insolation duration, specified in
hours for which each window should be illuminated. regulations. For this reason it is necessary to apply
Those provisions concern only apartment buildings, tools that enable to conduct precise insolation anal-
schools, kindergartens and nurseries. The insolation ysis through the use of computer models.
time shall not be less then 3 hours, measured on the
spring and autumn equinoxes, between 8 am and 4 Figure 1
pm for buildings of educational and care functions Sundial for Warsaw
and between 7 am and 5 pm for apartment buildings. (Twarowski 1996).
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
Digital Insolation Analysis Methods
the number of windows requiring analysis can reach
On the market there is a number of software for au-
hundreds.
tomated analysis of lighting conditions. These pro-
grams include Google SketchUp with a Shadow Anal-
Sundial Method
ysis plugin, Autodesk Revit, Rhinoceros 3d with its
Currently, the method most commonly used in
graphical algorithm editor Grasshopper and addi-
Poland for insolation analysis of windows was pro-
tional plugins for weather and energy analysis, such
posed and disseminated by Mieczyslaw Twarowski
as Ladybug and Honeybee that connects it with Ener-
(1996). This method involves using a kind of horizon-
372 | eCAADe 34 - SHAPE, FORM AND GEOMETRY | Grammars and Concepts - Volume 2
gyPlus, Radiance, Daysim and OpenStudio software. ing other software (Kwiecinski and Pasternak 2016).
Listed computer programs enable precise determi- Repeated changing of software during the de-
nation of sunlight vectors on a given date and time, sign process is cumbersome (Pasternak and Kwiecin-
but does not give the possibility to check whether ski 2015), therefore a new analysis procedure was cre-
a particular window or apartment meets the con- ated in the software that was used for the initial anal-
ditions described in the building regulations, men- ysis and determination of the MBV in the PRC Anal-
tioned above. In order to verify whether a window ysis. It allows to reduce the need to move from one
is provided with the right amount of light, a method program to another and at the same time overcome
for counting the time in which this window is illumi- its the limitations in terms of possible amount of an-
nated needs to be established. alyzed windows.
The creation of this procedure in a parametric
Figure 2
environment allows also to create a wide variety of
Maximum Building
detailed optimization procedures depending on the
Volume determined
needs of particular location and the configuration of
by the PRC Analysis
analyzed windows. For example, for a small plot with
software.
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
Figure 3 typology.
Designation of the
impact area of the METHODOLOGY
designed building. Within the framework of the developed method of in-
Range of shadow solation analysis of premisses in adjacent buildings,
cast on the spring three stages can be distinguished:
and autumn
equinoxes of an • Initial analysis of insolation of windows in the
extruded area of existing state;
the plot. There is also a software dedicated to this type of anal- • Analysis of the impact of the proposed form
ysis of local lighting regulations. PRC Analysis created on the surrounding buildings;
by Jacek Markusiewicz (2014) enables the analysis of • Determination of a Maximum Building Vol-
windows and whole apartments for compliance with ume or conduct of a custom optimization pro-
the local laws of insolation and occlusion. After a pre- cedure.
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 Insolation analysis
Building Volume (MBV), but the calculation that in- The first step of this process is to prepare a three-
cludes all of the potentially affected windows is very dimensional model of the building surrounding,
time consuming and involves preliminary windows which includes the whole impact area of the pro-
selection in order to permit the software to deter- posed building form. The upper dimensions of the
mine the solution (Figure 2). Because of this it is nec- building should be established in order to check how
essary to conduct a series of preliminary analysis us- far the shadow of the building during the spring and
autumn equinoxes reach (Figure 3), because it will
SHAPE, FORM AND GEOMETRY | Grammars and Concepts - Volume 2 - eCAADe 34 | 373
determine the windows which insolation conditions The procedure was created in a parametric en-
are potentially a subject to change. vironment. Energy Plus weather data file is used to
The next step is to determine which windows determine the direction of sun rays for the analyzed
belong to apartments or care and education institu- time periods. The intersections of sun rays with build-
tions, because their lighting conditions are particu- ing blocks are checked. Rays illuminating the win-
larly described in local building regulations. Other dow are grouped into two to create time intervals.
windows are not evaluated in this procedure. It is Depending the adopted accuracy of the calculations,
only necessary to model the surrounding buildings they can amount to a minute or more. The number of
to take into account their impact on the lighting con- intervals is counted and compared with the value of
ditions that are subject to evaluation (Figure 4). 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
To satisfy the requirements of building regulations cannot exceed in no case.
sufficiently, but not restrict the building capabilities In the second step, the same windows are
of the plot too much, it is necessary to precisely es- checked again, but this time taking into account the
tablish the analysis outlines. Since the building reg- proposed building. The intersections of rays, that
ulations specify that the insolation requirements de- until now were not obstructed by any building, are
pend on the size of the apartment, it is necessary to checked if they intersect with the new building vol-
discernment which windows belong to which apart- ume. The windows, that were previously correctly il-
ment and group them accordingly. In multi-room luminated are checked whether their sun exposure
apartments it is sufficient to meet the requirements is not deteriorated to an incorrect level. If it is true,
for only one room and one-room apartments do not they will be included in the optimization procedure
have to meet them at all. For this reason, it is neces- as boundary conditions.
sary to determine an internal functional arrangement The procedure allows to increase the accuracy of
of residential buildings located within the building's window analysis by adding more control points to
impact zone. the line located on the sill level on the inner face of
Because the regulations do not clarify to which the wall. Depending on the computational capabil-
extent the windows should be illuminated, there is ities, it is possible to analyze only the center point
a large scope for interpretation of these provisions. of each line or include additional points such as end
It can be assumed that illumination should be pro- points and additional points between these.
vided on the entire plane of the window or only on The result of the insolation analysis process is the
its selected points or edges. Depending upon which division of windows into three groups:
interpretation is accepted, it is possible to get com-
pletely different boundary conditions for the process • Group 1 - windows that have correct insola-
of optimization and hence, obtain different results of tion level and are not influenced by the de-
possible spatial building form. signed building;
374 | eCAADe 34 - SHAPE, FORM AND GEOMETRY | Grammars and Concepts - Volume 2
• Group 2 - windows that are insufficiently il- tion of the building at every stage of the design pro-
luminated in the current state, without tak- cess and at the same time meet all the requirements
ing into account the designed building. There arising from the context and regulations. The ability
shall be no sun rays obscured by its volume. to use the described analytical method as part of the
These windows define surfaces that shall to be form finding process with the use of the optimization
exceeded by the designed building; procedure gives the designer more freedom to shape
• Group 3 - windows that were correctly illumi- the form and still meet the legal requirements.
nated in the current state and the designed
building affects their level of insolation. Sun Optimization procedures
rays obscured by the new buildings are in- Another possibility is to develop a bottom up ap-
volved in the optimization procedure. 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
Maximum Building Volume the proposed forms are not worsening the insolation
The procedure allows the determination of a Maxi- conditions of any windows below the legally accept-
mum Building Volume. The outline of the plot ex- able level. The final form depends on the configura-
truded to a maximum height is trimmed so as to tion of input parameters and their ranges. Optimiza-
meet the insolation regulations of premisses in ad- tion procedures are prepared with the use of genetic
jacent buildings. This procedure takes place in two algorithms.
stages. For windows belonging to Group 2, all of the
extruded sun rays blocked by the new building are
RESULTS
subtracted from the volume (Figure 5).
As an example of the principles of the proposed
Figure 5 method, a plot located in the center of Warsaw was
MBV delimination - selected. In the impact area of the designed build-
stage 1. 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
In the next step sun rays, which are needed for a min- way that in order to illuminate a particular window in
imum illumination of the windows from the Group 3, the corresponding period of time, the entire obscur-
are selected in such a way that the volume subtracted ing volume is cut above the plane of the sun ray inci-
from the input form is minimized (Figure 6). dent on the windowsill of analyzed window. There-
This procedure for determining the MBV is deter- fore it is only necessary to analyze the windows lo-
ministic, because its objective is to obtain the largest cated at the lower floors of the buildings, assuming
possible volume. From the initial form the smallest that the functional layout is the same on each floor.
total volume designated by extruded sun rays is sub- If some floor has a different functional arrangement
tracted. It does not allow, however, to control which or other internal division of apartments, then these
parts of the form are more valuable then other and windows should also be included in the analysis.
protect them from being deleted. For the designer After removal of the windows that are not re-
it is crucial to be able to control the shape and loca-
SHAPE, FORM AND GEOMETRY | Grammars and Concepts - Volume 2 - eCAADe 34 | 375
Figure 6
MBV delimination -
stage 2.
376 | eCAADe 34 - SHAPE, FORM AND GEOMETRY | Grammars and Concepts - Volume 2
culations is reduced to 10 minutes, the number of not permitted in the building regulations, therefore
windows that meet the requirements decreases to the Maximum Building Volume that is not taking into
49, and the number of underexposed windows in- account other provisions and local factors is useful in
creases to 33. This is due to adopted rounding of the design process but not categorically conclusive.
values. If a calculation is conducted with an accu- It is possible to locate other forms, beyond the MBV
racy of one minute, the results will be the most reli- outline determined with the use of individual opti-
able, but the time of calculations will increase signif- mization procedures based on the results of insola-
icantly. For such precision of calculations the num- tion analyzes.
ber of windows illuminated properly equals to 56, the One of the possibilities of implementation of the
number of underexposed windows - 25 and windows optimization procedure is the search for such an ar-
demanding optimization - 26. rangement size and location of a rectangular prism
The end result of this procedure is the determina- on the plot in order to maximize its cubic capacity,
tion of the Maximum Building Volume, which is rep- when there are no underexposed windows (Figure
resented by the impassable building envelope, set 9). The number of sun rays obscured by the body
within the framework of building regulations con- of the building falling on initially underexposed win-
cerning insolation of premisses in adjacent buildings dows plus the amount of time needed to illuminate
(Figure 8). In order for this volume to treated as final, the initially correctly isolated windows to the mini-
other legal conditions and the ones resulting from mum required level ofter adding the designed build-
the architectural and urban context should be also ing are minimized. This procedure may also involve
taken into account. more than one a rectangular prism or address an ar-
rangement of any other geometric shapes or search
Figure 8
for proper parameters of on arbitrary building shape.
Maximum Building
The initial shape of the building is the entire surface
Volume obtained
area of the plot extruded to a height of 82 meters. The
after the insolation
result of this optimization process is the body of the
analysis process.
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-
The resulting Maximum Building Volume is heavily mum Building Volume. Studying typologically differ-
cut, despite the fact that in the process of creating ent geometries can lead to obtaining forms well fit-
its goal was to reduce the volume as little as possi- ted to the context and fulfilling building rules.
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
CONCLUSIONS
The paper presents a method for finding an optimal
possible configuration of sun rays needed to illumi-
building volume and its location on the plot in rela-
nate the windows. It is possible to designate a num-
tion to the insolation conditions of buildings which
ber of other configurations that also comply with the
are located in the impact area of the designed build-
regulations, but their arrangement will allow the lo-
ing. The proposed method refers to the national rules
cation of other volumes on the site , that does not
governing the conditions of insolation of rooms and
fall within the designated maximum volume. Plac-
apartments in adjacent buildings. These regulations
ing the building on the entire surface of the plot is
SHAPE, FORM AND GEOMETRY | Grammars and Concepts - Volume 2 - eCAADe 34 | 377
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
SHAPE, FORM AND GEOMETRY | Grammars and Concepts - Volume 2 - eCAADe 34 | 379
ies can be found of a 3D façade replacing the con- placement of a rectangular fenestration on a planar
ventional building façade. Nevertheless, the interest building envelope. These particular facades yield a
of more complex shapes and patterns applied to the range of parameters such as the size, number, and
building façade is growing significantly in contem- the distribution of the windows as well as the thick-
porary architecture due to technological and fabrica- ness and the material of the wall with external shad-
tion advancements (Rahimzadeh et al. 2013). In addi- ing devices (Hassaan et al. 2016; Echenagucia et al.
tion, while buildings which do feature a complex 3D 2015). For example, Echenagucia et al. (2015) stud-
façade have been erected in some urban areas, most ied the exterior wall of an open space office's plan-
of these buildings are very experimental. Therefore, a ner in an urban and non-urban context at four dif-
systematic exploration of 3D façades and an investi- ferent locations in Europe. In contrast to parametric
gation of the benefits of these emerging façade pat- modelling, a new approach suggested by J. Wright et
terns are highly significant. al. (2014) generates façade patterns by dividing the
The aim of this paper is to suggest a methodol- surface into small equal rectangular cells and deter-
ogy to generate an enclosed 3D façade unit, which mines the optimal number of windows and distribu-
is near optimal in terms of daylighting. The gener- tion through multi-objective optimization based on
ated 3D facade unit consists of mainly three compo- energy performance and capital cost.
nents: a transparent window allows sunlight to pene-
trate into the indoor space; an opaque wall functions Assessment of a building façade perfor-
as a shading device, and BIPV harvests solar energy. mance
To achieve this goal, 3D façade units are generated In this study, we measured the amount of natural
in two phases: first a basic 2D shape is generated light using the "Useful daylight luminance" (UDI) pre-
and, next, it is expanded into a 3D façade unit. Sub- dictive method. The UDI method is first coined by
sequently, materials are applied for daylight simula- Nabil and Mardaljevic (2005), and divides annual day-
tion. By performing all assignments parametrically, light illuminance at the workplace into three bins.
an evaluation of daylight simulation can take place The first bin includes areas that receive under 100 lux,
in order to suggest an optimal façade. which is not suitable and thus demands additional ar-
tificial lightning; the second bin corresponds to the
LITERATURE REVIEW range of 100 to 2000 lux, which is suitable for work
A vast number of peer-reviewed studies could be activity; the third bin includes illuminance that ex-
found regarding the building façade. In this section, ceeds 2000 lux and which results in potentially vi-
we have categorized them into three main parts as sual discomfort (Nabil and Mardaljevic 2005). This
a generation, performance assessment, and the opti- method is more realistic than the conventional "day-
mization. light factor approach" which only considers a single
overcast sky. When natural lighting cannot supply a
Parametric façade design sufficient amount of light into the indoor space, ar-
Techniques for generating building façades have tificial lighting would be required in the space. To
been investigated by a number of authors. One of decrease this electrical demand, building integrated
the most common methods is parametric modelling, photovoltaic panels could be attached at the outer
which is highly effective to automate the generation side of the façade for electricity harvesting. This prac-
of a large set of architectural design instances by the tice is one of the sustainable features in the building
combination of pre-defined design parameters (Tur- domain which could potentially cover more than half
rin et al. 2011). In designing a building façade, de- of the daily energy needs (Berkel et al. 2014). Man-
signers and researchers are highly concerned by the dalaki et al. (2012) examine the thirteen most com-
380 | eCAADe 34 - SHAPE, FORM AND GEOMETRY | Grammars and Concepts - Volume 2
Figure 1
Facade 3D unit
generation and
optimization
system diagram.
monly used types of fixed shading devices as a PV natives for performance assessments. The optimal
panel for an office building. Among them, the single building façade design can be achieved effectively
inclined canopy showed the most efficiency when by means of building performance simulation cou-
comparing the area of PV with the generated elec- pled with an evolutionary algorithm tool (Turrin et al.
tricity. Vartiainen et al. (2000) analyzed the optimal 2012; Evins et al. 2011) (Figure 1). Especially, the tech-
size and orientation of a single rectangular fenestra- nique of multi-objective optimization is highly prac-
tion in a fully covered PV integrated building façade tical (Wang et al. 2005) in that it provides visual infor-
unit. A very low percentage of window area, ten to fif- mation of the trade-offs between contrasting design
teen percent of the whole façade, proved to be ideal objectives (Mela et al. 2012). Several studies have
when considering energy harvesting through PV and focused on the window-to-wall ratio (WWR) and en-
the replacement of artificial lighting by daylight in ergy performance (Goia et al. 2013; Echenagucia et
the specific location of Europe. al. 2015); while other researchers optimize window
size and external shading types by the means of ge-
Façade Optimization netic algorithms (Torres and Sakamoto 2007). Yi and
In order to achieve a better performance with re- Malkawi (2009) investigated the overall building form
spect to daylighting, the building façade needs to controlled by a hierarchical relation of geometry and
be optimized. There is a large pool of variables form optimization and the method was able to find
that controls the design and the overall performance a particularly complex shape rather than the simple
of the building façade, which could be effectively boxy one.
controlled parametrically to yield a number of alter-
SHAPE, FORM AND GEOMETRY | Grammars and Concepts - Volume 2 - eCAADe 34 | 381
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.
382 | eCAADe 34 - SHAPE, FORM AND GEOMETRY | Grammars and Concepts - Volume 2
Subsequently, materials are applied to the spatial (Roudsari et al. 2013).
space frame wires resulting from the Delaunay trian- An adequate amount of daylighting requires an
gulation (Figure 2 F). appropriate window-to-wall ratio, while solar energy
Three main material types are being considered: harvesting increases when the PV surface area ex-
glazing, wall, and a totally opaque PV panel. For the pands. These two characteristics are highly depen-
daylight simulation, these three materials can be cat- dent on the direction of the façade and the loca-
egorized based on two general properties, whether tion of the building. Furthermore, these two objec-
opaque (the PV panel and the wall) or transparent tives contrast with one another: when designers set a
(the glazing). This binary surface option serves as an- goal to maximize the amount of electricity from BIPV,
other parameter for the optimization fitness function. it will affect the size of the window, consequently
This method can be applied to any type of geom- deteriorating the daylighting potential. Therefore,
etry with the appropriate base 2D tessellation and a we employed evolutionary computation for multi-
sufficient number of additional vertices on each fa- objective optimization, using the Octupus plug in.
cade unit. In the case of a conventional façade de-
sign, with a traditional rectangular window within Optimization strategy and Fitness function
the façade unit, it is sufficient to consider four addi- The objective function maximizes the area which is
tional points with a zero extrusion length. correspondent to the range of UDI100-2000 in the
given space, while also maximizing the amount of an-
Parametric modeling and Multi-objective nual energy which obtained by BIPV on the façade
optimization unit.
We used the Grasshopper parametric modeling tool The evolutionary algorithm inputs are classified
as a platform for the entire process, including both into two main categories. The first category of inputs
design exploration and simulation. are the extrusion lengths of the additional points and
the binary material selection of the triangular faces
Figure 3 that are generated from the Delaunay triangulation.
Testing room is in In our case study, four additional points make ten
the middle of a 3 x 3 triangular façade geometries in 3D space, thus ten
units. In this combinations of façade material and four extrusion
picture, planar lengths, or a total of fourteen input genes for the op-
façade with timizing algorithms.
rectangular unit is In the multi-objective search, HypE mutation
simulated. 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.
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
Simulation environment 1012.5 kWh annual electricity was generated from
A square division is selected for the basic tessella- the opaque façade area.
tion of the building façade, and we added identical For the second case study, we created a 3D
one-person office rooms to each square unit. The façade through extrusion of the optimization strat-
building is located in Incheon, Korea, and the test- egy and the fitness function. The population size was
ing façade is facing the south side. A typical office set to 50 and 25 generations have been conducted
room is selected for the daylight simulation as a case for the optimization. General settings of parameters
study. The room dimension is 6m by 4.2m, depth are reported in Table 1.
and width respectively, and 3.2m in height (floor to
ceiling). The reflectiveness of the materials taken Table 1
from Nabil and Mardaljevic (2005)'s experiment en- Multi objective
vironment are wall 0.7, ceiling 0.8, floor 0.2 and the optimization
window transmittance is 0.76 (Nabil and Mardalje- settings.
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 A constraint was introduced with each evaluation
the case room as shown in Figure 3. According to and generation of the façade unit that surrounding
the basic tessellation, the additional eight units are unit shapes are identical to the case room façade de-
located one at the east and on at the west side of sign. And extrusion lengths are limited to maximum
the case room and three positioned above as well 2.0 m. Other factors not considered is the façade con-
as three at the lower level. Consequently, nine units struction, e.t. size and type of mullions.
having the identical façade geometry are generated
3 x 3 where the center of the middle level is present-
ing the testing room. RESULTS
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 AND NON APPARENT REPETITION', 29th European
optimization. Photovoltaic Solar Energy Conference and Exhibition,
When the optimization method we created was pp. 366-372
David, M, Donn, M, Garde, F and Lenoir, A 2011, 'Assess-
applied, we found that our pareto-front solutions
ment of the thermal and visual efficiency of solar
have significantly higher value in both daylighting shades', Building and Environment, 46(7), pp. 1489-
and electricity generation in comparison to the con- 1496
ventional flat façade where the achieved UDI in- Echenagucia, TM, Capozzoli, A, Cascone, Y and Sassone,
creased by 9% in the best case of daylighting per- M 2015, 'The early design stage of a building enve-
formance and a growth in electricity harvesting of as lope: Multi-objective search through heating, cool-
ing and lighting energy performance analysis', Ap-
much as 50% in the best case electricity generation.
plied Energy, 154, pp. 577-591
The advantage of implementing of multi- Evins, R, Pointer, P and Vaidyanathan, R 2011 'Multi-
objective optimization in this particular case were objective optimisation of the configuration and
twofold. First the number of unique solutions yielded control of a double-skin facade', Proceedings of Build-
in pareto-front. Even though the performance is al- ing Simulation 2011: 12th Conference of Interna-
most identical, significantly different design alterna- tional Building Performance Simulation Association,
pp. 1343-1350
tives could provide important information to design-
Goia, F, Haase, M and Perino, M 2013, 'Optimizing the
ers in early design stage. The second adventage is, configuration of a facade module for office buildings
the comprehensive feedback on the performance of by means of integrated thermal and lighting simu-
the optimal solutions. The Designers can achieve lations in a total energy perspective', Applied Energy,
their set goals for their façade designs by consider- 108, pp. 515-527
ing the best trade-offs between the achieved UDI Gonzalez, J and Fiorito, F 2015, 'Daylight Design of Office
Buildings: Optimisation of External Solar Shadings
and the generated electricity not completely disre-
by Using Combined Simulation Methods', Buildings,
garding one of the trade-offs. 5(2), pp. 560-580
Furthermore, the suggested method shows Hassaan, A, Mahmoud, A and Elghazi, Y 2016,
great flexibility and compatibility of generating any 'Parametric-based designs for kinetic facades to op-
façade design with the same process. The method timize daylight performance : Comparing rotation
could be used effectively in the early stages of a de- and translation kinetic motion for hexagonal facade
patterns', Solar Energy, 126, pp. 111-127
sign of an office building and assess the daylighting
Krarti, M, Erickson, PM and Hillman, TC 2005, 'A simplified
performance and renewable energy generation pre- method to estimate energy savings of artificial light-
diction in a given location. ing use from daylighting', Building and Environment,
3D shaped building façades showed an advan- 40(6), pp. 747-754
tage of energy harvesting and daylighting perfor- Li, D.H. and Tsang, E.K. 2008, 'An analysis of daylight-
mances, however energy consumption was not ad- ing performance for office buildings in Hong Kong',
Building and Environment, 43(9), pp. 1446-1458
dressed in this research. Future works will extend the
Mandalaki, M, Zervas, K, Tsoutsos, T and Vazakas, A
scope of this study by adding energy efficiency pa- 2012, 'Assessment of fixed shading devices with in-
rameters into the façade generation method. tegrated PV for efficient energy use', Solar Energy,
86(9), pp. 2561-2575
Mela, K, Tiainen, T and Heinisuo, M 2012, 'Comparative
REFERENCES study of multiple criteria decision making methods
Bader, J and Zitzler, E 2011, 'HypE: an algorithm for fast for building design', Advanced Engineering Informat-
hypervolume-based many-objective optimization.', ics, 26(4), pp. 716-726
Evolutionary computation, 19(1), pp. 45-76 Nabil, A and Mardaljevic, J 2005, 'Useful daylight illu-
Van Berkel, T, Minderhoud, T, Piber, A and Gijzen, G minance: a new paradigm for assessing daylight in
2014 'DESIGN INNOVATION FROM PV-MODULE TO buildings', Lighting Research and Technology, 37(1),
BUILDING ENVELOPE: ARCHITECTURAL LAYERING pp. 41-59
SHAPE, FORM AND GEOMETRY | Grammars and Concepts - Volume 2 - eCAADe 34 | 387
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
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
stereotomic rules and interdependencies of the orig- Each trimming plane is an offset of the plane defined
inal Abeille vault. by vectors Vu+ and Vv+ at the centroid Cn of each
field (Figure 2d).
Abeille-based TIA The Abeille-based TIA geometric construction
The starting point of the geometrical construction method is, thus, based on the four main variables:
method is formulation of a point grid on a given sur- the point grid distribution of the initial surface, curva-
face. For each grid field defined by vertices ABCD four ture of the initial surface in U and V directions, angle
points at vertices mid-span are derived: Pt0, Pt1, Pt2 of plane rotation, and the position of the trimming
and Pt2. Points Pt0 and Pt2, and Pt1 and Pt3 laying planes. Through establishing the interdependences
at opposite sides of the field define vectors in U and between the variables the construction method is
V direction respectively. Thus, each mid-point Pt that further parametrized, and the interlocking properties
does not lie at the surface boundary defines a pair of assessed.
vectors in opposite directions, Vu+ and Vu-, or Vv+ First geometrical interdependency established is
and Vv-. between the curvature of the initial surface in U and
A plane Pl is defined at each mid-point Pt based V directions , and the plane rotation angles. The val-
on the vector between the mid-point and adjacent ues of these variables are inversely proportional: the
vertex, and the sum of the vector pairs at that mid- higher the curvature, the lower is the planes rotation
point, referred to as Vδu or Vδv (Figure 2a). The mag- angle, and vice versa. In this way the elements within
nitude of vectors Vδu and Vδv reflects the surface cur- a planar assembly are keyed in, while elements within
vature in U and V directions respectively at the given an area with a high curvature degree do not have
mid-point. Each Pl plane is rotated around the first awkward configurations (Figure 3).
axes for an angle α. The rotation direction is opposite Second geometrical interdependency estab-
to the rotation direction of the adjacent plane (Fig- lished is between the curvature of the initial sur-
ure 2b). The intersection of the four rotated planes face in the U and Vi directions, and the point grid
defines a tetrahedron (Figure 2c). density, and is directly proportional. Through this
Finally, pairs of trimming planes are defined. parametrization high curvature defines high grid
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.
Figure 2
Al Bahar Towers,
The folding shading
system, opens and
closes according to
sun's position.
Figure 4
DIVIDING A SQUARE
However, with origami, it is surprisingly easy to
The Syddansk
achieve different divisions of a squared module. One
Universitet
of the ways of using geometric shapes is by splitting
communications
a shape up into simpler shapes or splitting it in a sys-
and design building
tematic way to make new shapes. At first, Simple divi-
in Denmark
sions or less complex pattern can be used to actually
features a
divide the square horizontally or vertically, creating a
climate-responsive
grid.
kinetic facade.
Every step adds as many lines, cease or tessella-
PARAMETRIC MODELING tion as possible which do not depend on each other
Parametric design refers to a practice of digitally but previously existing lines. This means that you
modeling a series of design variants whose relation- can create different 3D configuration from the same
ships to each other are defined through one or sev- grid divisions. Pattern sequences are used to split the
eral mathematical relationships (parameters). Para- square into number of triangular parts. It is accom-
metric models can automatically adjust geometric plished by connecting lines between intersections,
models in response to dynamic real-time data (Davis corners of the square, or from the center of each side
et al. 2011), to produce a flexible forms that can be to the two corners of the opposite side as shown in
animated (Leach 2009) . It is considered an essential Figure 5.
tool during this research due to the complexity of the
folding operations which was made to design a flexi- METHODOLOGY
ble form that when collapsed would be able to open To integrate all characteristics of architectural design,
diaphragm enabling daylight to enter the space and a comprehensive methodology is required to con-
when expanded would form a planar flat surface. Ini- sider the aesthetic qualities of the façade and the
tially, the parametric model began with a square. The performance behavior of the skin. This paper ex-
folds were composed by finding relations and con- plores origami-based Kinetic façades design to con-
nections between the vertices, mid points of edges trol daylight uniformity through three phases. The
of the square to its center point. The relations of lines first phase explores kinetic behaviour of origami ge-
and points changes in order to open and close the ometries through paper maquettes to compare dif-
diaphragm. The main parameters of each Origami ferent simple folding configurations to understand
unit form were the motion of the folds. Different the movement at global scale. The second phase use
combinations of rotations and translations led to dif- Parametric modelling tools to model different kinetic
The parametric model provides a constant feedback terms of SDA and ASE for different Origami modules.
on the explorations of the proposed scenarios of The best possible solutions were found by reaching
Origami patterns to be animated for daylighting sim- the balance point between SDA and ASE. It aims to
ulation. The main tested parameters are the form maximize the sDA value while minimizing the effect
of Origami folds, the module size and motion sce- of ASE to get rid of direct sun exposure.
narios. The simulation includes 10 explored origami It was found that the large Origami module
screens with variations of 3 module sizes and 6 mo- (1.5x2m), resulted in a relatively high percent of HSE
tion steps resulting in (180) cases tested on two sim- and over-lit portion of space while the small modules
ulation stages. First measuring illuminance at four (0.5 x 0.5 m) results in a partially day-lit space. The in-
times of the year, March 21, June21, September and termediate sized modules (1m x 1m) could relatively
December 21, taken every hour of the working day get rid of direct sun exposure while maintain appro-
from 8:00 am to 6:00 pm (7200 runs) to be able to priate Day-lit percentage as shown in Figure 13. In
compare different configurations. addition, this size of apertures maintains a minimum
level of contact with outside.
RESULTS All the kinetic folded tested cases resulted in
The process of parametric simulation for the Ten acceptable daylighting performance. Acceptable
tested cases took about seven days to be completed daylight performance was mostly achieved at fully
and written in spreadsheets using SpeedSim-for- opened cases and partially opened in winter time.
DIVA plugin along with grasshopper (Wagdy 2015). Unacceptable HsDA results always appears in De-
The daylighting performance results is represented in cember at 4:00 & 5:00 pm.
CURVED FOLDED DESIGN DEVELOPMENT chitectural orders (see Figure 6) in addition having a
The goal was to investigate diverse shapes that could hexagonal section better illustrates the idea of a cir-
work as both structural and ornamental architectural cular section, which is specific to the concept of col-
elements, such as the columns used in the classical umn. The design needed to include new character-
orders. We intended to use sheet metal that can be istics compared to the classical or even contempo-
very effective from a structural point of view but also rary columns, hence the idea of transparency arose.
generates pleasing reflections when curved. Sev- This was achieved by splitting the column, and later
eral curved crease folding column designs were de- on also the option of using inner light that accentu-
veloped at model scale (see Figure 5), each with its ates the shape of the column in dark environments
own advantages and disadvantages regarding differ- and further enhances the ornamental aspect of this
ent aspects of aesthetical, practical and structural na- structural element (see Figure 7).
ture. The chosen design (see Figure 7) references
the entasis and the fluting found in the classical ar-
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).
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.
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.
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
Figure 3 Limitations
Almost 100 chairs During several informal sessions we collected feed-
combined back for Michelangelo. Designers appreciate the
randomly from only web-based platform for providing easy scripting and
4 source designs. 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
Figure 3 shows a variety of chair models with parts plans created for the office building would not com-
combined from four incomplete grammars. The im- ply with a different one. A different boundary poly-
age was generated by a grammar which contained gon and different placement of vertical structures like
only axioms for wooden chairs, but no rules were columns, pipes and elevators would be the main is-
given. All modeling decisions were left to the cloud. sues. A dedicated rule would be necessary for a uni-
Variance implied by the high number of possible versal procedural approach to floor planing.
combinations is the first important use-case of the The current scope of Michelangelo are furniture
presented method. items. Our main goal is to improve the available de-
Instancing of complex model hierarchies is the formation rules to allow a wider range of forms. The
most powerful use-case of the presented approach. current set is limited to simple taper, twist and shear
Procedural models, if defined well, are able to fit rules.
automatically into different space volumes. Figure
4 shows an office building with 2500 m2 on three
floors populated purely by procedurally generated FUTURE WORK
content. Shape grammars were in particular helpful We hope that more out-of-house artists will try out
with furniture layout and design due to extensive in- Michelangelo. As the cloud will grow there will be
stancing of similar objects. The most frequent items new challenges for keeping the system consistent
are 240 desks and office chairs. A part of them is visi- and focused on the goals given in the axioms. The
ble on the floor plan of the top floor on Figure 5. De- topic of cross-reviewing grammars is part of a larger
tailed variations of an office are presented in Figure 6 research package planned for the next months. We
and Figure 7. It is mainly the arrangement of furniture see a large potential in turning the cloud intelligent.
that changes in the examples. By observing user behavior and collecting feedback
Evaluation of the whole building axiom was per- on the produced models the cloud should be able to
formed in less than two minutes. The generation pro- learn and provide a highly customized experience:
cess took 47,284 derivation steps controlled by 329 • Learn new concepts and select rules better. If
rules which were combined from 27 grammars. The there are many rules without an explicit pred-
Figure 5
Interiors of offices
in the top floor.
Figure 7
More office layouts.
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
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.
Casa da Música's exterior walls are all built of concrete pressive and intentional aesthetics, making the poly-
cast in place [3]. The construction of the walls was hedral nature of the project (Figure 2) change its own
phased, because concrete has a limit on its height for stereotomy semantics.
proper setting. Because concrete is a ductile mate-
rial while it is setting, horizontal lines appear natu- EMP Museum
rally in the top height of the concrete layers, regard- Gehry's EMP Museum (1995-2000) is covered in sheet
less of formwork. The architects seem to have had a metal in a double curvature configuration. Sheet
different purpose for the appearance of the building, metal does not admit double curvature, so the design
denying its relationship to gravity and its horizontal had to be rationalised into developable strips. Break-
and vertical related vectors. Standard formwork pan- ing down a curved surface to smaller parts which
els were chosen, creating a rectangular array. This bear a smaller curvature is an approach that can be
array of panels is aligned with a roof edge and from traced back to the earliest records on stereotomy
there, it develops around the facets of the building practices. The first forms of stereotomy relied in
creating oblique directions which are aligned in an shapes easily projectable to a plane, featuring vari-
intrinsic surface analysis but completely unaligned as ation in only one dimension, such as arches and bar-
seen from the outside. A conflict is generated be- rel vaults. The method for materialising this kind of
tween the construction lines and the falsework (Fig- designs was mainly "par équarissement"; this process
ure 1), in a tension that clearly marks a stand from relies in having two or more projections drawn in
the architect's point of view: the division lines appar- a block of stone and then having the mason carve,
ent in the surface ought to be oblique. The intention progressively from both sides until these meet. As
of organising a prefabricated, mass produced false- the discipline created more complex shapes such as
work, much like in the architecture of Tadao Ando, the skewed arch or arriére-voussure the transferring
is an expression of stereotomy for a subjacent idea "par équarissement" revealed its shortcomings as the
of organisation is implied by the apparent junctions. bounding box would be too big or the orthogonal
These false work imprinted junctions are key to ex- projections not sufficient to describe the complexity
press the design intent in the architecture, gaining of the angled shape.
prominence above the construction lines, for its ex-
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?
Table 1
List of English
Words for the art
related
word-filtering
process of the
feeds.
Table 2
List of some
Singapore
sub-administrative
boundaries with the
scores performed in
the analysis related
to their cultural
planning.
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".
Figure 1
Distribution pattern
of high tech
companies, based
on geolocated
company data from
CrunchBase.
(Startup companies
in cyan and
headquarters in
dark blue.)
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.
sign goal for more efficient building use was achieved decreases at night. This implies that human activi-
(see Figure 5). 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
We can see, for example, that geo-located Twitter or the overall contextual polarity of a document.
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
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- Johnson, K 2014, Innovating the City: Challenges and Op-
tricts", an important topic of the 21st century. 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-
ACKNOWLEDGMENTS alytics., Springer
We thank C. Sun, M. Dennis, and S. Williams for dis- Lynch, K 1960, The Image of the City, Cambridge, Mass.:
cussions and paper preparation,C. Haung for help in MIT
getting the user interface to run, J. Chen, C. Summit, Lynch, K 1981, A Theory of Good City Form, Cambridge,
Mass.: MIT
M. Lin and J. Nawyn for helpful discussions and com-
Russell, M 2011, Mining the Social Web: Data Mining Face-
ments, members of Media Lab Changing Place group. book, Twitter, LinkedIn, O
In memory of W. J. Mitchell. Simmie, J 2011, Innovative Cities, London: Spon
[1] http://www.brookings.edu/˜/media/Programs/metro
/Images/Innovation/InnovationDistricts1.pdf
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
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.
Figure 3
Confidence interval
for models
predicting the
number of
employment based
on different spatial
features.
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
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.
Figure 4
A Time Historical
Image of Pedestrian
Spots (THIPS).
pedestrians move to another space. We defined the In a THIPS, a pedestrian is regarded as a cluster
vertical section as the 'THIPS section', and the line on of connected pixels with the same label. First, the
movies as the 'THIPS line'. algorithm calculates the distance of every image by
On that account, as illustrated in Figure 5, we subtraction and arranges the pixels. Second, it as-
classified THIPSs as follows: 'time historical image of sesses the value of each pixel as 0 or 1 with a thresh-
pedestrian horizontal spots (THIPS_H)', which is de- old; that is, whether a pedestrian exists or not, re-
noted by a horizontal line on movies; 'time historical spectively. Then, the algorithm defines the cluster
image of pedestrian vertical spots (THIPS_V)', which of a pedestrian in a THIPS based on a labelling al-
is denoted by a vertical line on movies and 'time his- gorithm that examines neighbor pixels against a la-
torical image of pedestrian diagonal spots (THIPS_D)', belled pixel. Furthermore, a morphological opera-
which is denoted by a diagonal line on movies. 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-
ALGORITHM FOR THE ESTIMATION OF tion. Therefore, the algorithm introduces a morpho-
SPATIOTEMPORAL DISTRIBUTION OF logical operation to integrate divided clusters and to
PEDESTRIANS remove noise.
We interpreted the information of pedestrians on The morphological operation works on dilating
THIPSs to estimate the spatiotemporal distribution of or eroding clusters on images; it is usually used a cou-
pedestrians. We propose an algorithm and describe ple of times. In addition, the 'opening operation' di-
each method of calculation. lates after eroding clusters, and the 'closing opera-
different microspaces, as shown in Eq. (1). As a result, CLEANING NOISE BY DISTANCE OF CLUS-
we understand the direction whether Δt is larger than TERS BETWEEN THIPSS
zero or not. To estimate the travel direction of a unique pedes-
K|y= − K|y= = ∆u = (∆x, ∆t) (1) trian, we have to be able to distinguish a particular
cluster when compared with clusters of other THIPSs;
Thus, we assess whether the moving direction, which
those do not always exist in similar positions. Con-
is estimated by the difference between centroids of
sequently, we cleaned irrelevant clusters of other
clusters, is correct or not and verify the accuracy of
THIPSs due to the large distance between them. We
our algorithm to estimate the moving direction of
focused on almost not changing the position of clus-
pedestrians. In conclusion, we achieved compara-
ters in case the intervals of THIPS lines were infinites-
tively high validity of the algorithm in Table 2.
imal. It would be the closest distance identified with
Table 2 same pedestrian, calculating the distance of clusters
Correct Estimation between THIPSs by the Ward method. Hence, it is
of the Direction of likely to be further than the distance between clus-
Pedestrians. 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.
Figure 11
The Matching
Conclusion
between
Clusters(upside:
Source Layer,
underside:
Compared Layer).
Table 3
The Estimation
Result of Traffic
Volume of
Pedestrians in
Watch Cameras.
LABELLING CLUSTERS BETWEEN THIPSS ally performs well. The result of matching clusters is
The source layer and the compared layer are inte- shown in Figure 11.
grated on the basis of the matching results so as to
estimate the travel direction of pedestrians. Data EXPERIMENT
are structured as in Section "Data Structure" and also I estimated the traffic volume of pedestrians over the
held in the list structure. area illustrated in Figure 12 by this algorithm. The re-
sult is shown in Table 3. It describes no estimated
VERIFICATION OF THE ALGORITHM value is far from the correct number of pedestrians.
Although the influence that some noises are In addition, during times of low pedestrian flow, the
matched to far cluster is ascertained, the perfor- estimated value was almost equivalent to the actual
mance of labelling algorithm between THIPSs gener- count as shown in the results of cameras 3 and 7.
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).
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.
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
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-
Figure 3
Pedestrian tracks
recorded by
students in outdoor
studies.
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
Figure 1
Unity3D Timeline UI
component.
Figure 3
3D Studio Max
custom properties
metadata.
and photogrammetry. The images show that results tas-de-charge and vault ribs springing from the aba-
for the photogrammetric study at Chester were un- cus were more focussed, and consequently the wider
reliable, whereas the laser scanned version produces context was ignored. The data was then processed
an accurate rib profile for tracing. It is important to alongside the laser scanned version of the pier. Data
state that the data captured at Chester was intended from Exeter in Figure 6 shows that the laser scanned
for the large-scale investigations only, and whilst the tracing of rib profiles was still more accurate than
laser scan data also proved adequate for more de- the photogrammetric survey. This is particularly no-
tailed investigations, we realised that for photogram- ticeable for the upper rib cut, where the laser scan
metry to be effective a more strategic approach was data shows all the intricacies of the rib profile, while
required. the photogrammetry data shows the general shape
At Exeter, based on our previous findings, we of the rib profile, but the details of the ribs are lost.
selected a single pier and produced a photogram- However, when we compare the profiles of the main
metric survey of it. As a strategy, the images of the pier below the abacus line, we can see that the pho-
togrammetry version is comparable to the laser scan laser scanning we can state that the standard rib cuts
version (Figure 6). This is because of two factors, matched those of the tas-de-charge blocks in the se-
firstly the pier shafts are lower than the ribs and sec- lected features at both Chester and Exeter. This sup-
ondly the pier itself features in the majority of digi- ports Willis's theory that the overall rib profile is dis-
tal images used for photogrammetry, therefore it is torted at the springing, probably, according to his in-
documented much more thoroughly than the ribs. terpretation, for ease of constructive practice when
This tells us that photogrammetry could be a viable this construction shortcut would be invisible to the
method of investigating tas-de-charge distortions if viewer.
we could get close enough to the ribs using scaffold-
ing or a basket crane, however, as we have the use of RECOMMENDATIONS AND REFLECTIONS
a laser scanner this method is still the most successful The surveys captured at Chester Cathedral were part
in terms of both results and logistics. of a pilot study, whereas those captured at Exeter
Based on the results obtained primarily from
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.
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.
titative ones, the simplified assumptions still ensure Aydin, S and Schnabel, MA 2015 'Fusing Conflicts Within
a good reliability of the simulation and useful feed- Digital Heritage Through the Ambivalence of Gam-
backs to the different actors involved in the decision- ing', Proceedings of CAADRIA 2015, Daegu, pp. 839-
848
making process. Future work within this research
Batty, M 2001, 'Agent-Based pedestrian modeling', Envi-
scope will focus on additional testing and calibration ronment and Planning B: Planning and Design, 28(3),
of the model, in order to improve adherence to real pp. 321-326
phenomena and more accurate estimation of users' Boeykens, S 2011 'Using 3D Design Software, BIM and
behaviour impact on the built heritage artefact. Game Engines for Architectural Historical Recon-
struction', Proceedings of CAAD Futures 2011, Liege,
pp. 493-509
REFERENCES Kalay, YE 2008, 'Preserving Cultural Heritage through
Affleck, J and Kvan, T 2005 'Reinterpreting Virtual Her- Digital Media', in Kalay, YE, Kvan, T and Affleck, J
itage', Proceedings of CAADRIA 2005, New Delhi , pp. (eds) 2008, New Heritage: New Media and Cultural
169-178 Heritage, Routledge, New York, pp. 01-10
INTRODUCING 3D DIGITAL RECONSTRUC- select key aspects of previous projects about digital
TIONS OF LOST BUILDINGS: WHAT, WHY reconstructions of lost buildings. References (con-
AND HOW ference papers, journal articles, books and websites)
An analysis of several sources, supported by authors' have been checked from areas of intersections be-
experiences in this field, have allowed to identify and tween digital technologies with architecture and ar-
chaeology. Moreover, websites of specific research
of Trajan's Forum in Rome includes an archaeologi- mation portal, designed to allow updating and alter-
cal investigation and a sophisticated modeling work native interpretations.
(Packer 2001). A similar purpose has led the digital
reconstruction of the Empire Exhibition held in Glas- Documentation
gow (Scotland) in 1938 and subsequently dismantled The model allows producing architectural graphics
[5]. and other information for historical studies, conser-
vation, restoration or adaptive use. Rendered images
Information Management Systems and animations of the digitally reconstructed portion
Recent work in the field of archaeology, cultural her- of the cornice of the Chicago Stock Exchange Build-
itage and architectural history tends to move be- ing document the relationship of the terra cotta en-
yond digital 3D modeling into more specific tech- velope and the steel frame that resulted from the 3D
nologies for knowledge organization. This includes digital reconstruction phase (Figure 5). The modeling
data management, ontology engineering, multime- developed effectively as a heuristic process, which re-
dia systems, immersive environments, virtual and vealed how an entire row of terra cotta blocks had
augmented reality and semantic information mod- been damaged during construction because of lack
eling (GIS and BIM). Thematic modeling integrated of coordination between the terra cotta production
with GIS is becoming common practice in archaeol- and the steel design. When the digital terra cotta
ogy. Human interfaces, interaction and usability have blocks were positioned on the model of the frame, it
become an essential component of reconstructions. appeared that all blocks below the coping had been
The 3D model can also be enriched with information chiselled on the back at the same level, which was
pertaining to any knowledge domain using semantic confirmed by visual inspection of the fragments. The
web technologies. The model can become an infor- portion removed from each piece lined up perfectly
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. personal experiences in the field. The big amount
of research about 3D digital reconstruction of lost
CONCLUSIONS AND FUTURE DEVELOP- buildings, developed in a span of over 30 years, pre-
sented a wide variety of theoretical positions, meth-
MENTS
ods and tools influenced by disciplinary areas such as
This piece of writing presented a first critical framing
architecture and archaeology. However, counter this
of 3D digital reconstruction of lost buildings based on
amount and variety of research, we found the almost
the result of an examination of example projects and
Figure 1
MULTIRAMA
method application
process (Ozer et al.
2016).
Figure 3
A comparison of
the real/real (left)
and the virtual/real
(right) of the scene
of Parion Theater.
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
Figure 7
MULTIRAMA Parion
v1.0 Beta Interface.
The outcome of the research and the benefit to the new database for historic sites around the world, also
related fields are summarized below: some Turkish sites will be included such as Cappado-
cia. The second project is named as "Using Virtual Re-
• Use of a low-cost, holistic method utilizing AR
ality (VR) and Augmented Reality (AR) Technologies
technologies to represent digital heritage,
in Architectural Education", which will focus on in-
• Reviving the context connection with the ar-
tegrating Cultural Heritage Visualization to architec-
chitectural heritage since architects usually
ture education. This study aims to integrate "Cultural
documents the heritage via 2D drawings,
Heritage Preservation and Documentation" and "Dig-
• An easy interface and visualization tool for ar-
ital Design", through the development of a low-cost,
chaeologists to analyze their works in a 3D set-
holistic method that utilizes VR / AR technologies to
ting instead of 2D drawings,
represent Digital Heritage. This project also aims to
• Providing tourists a different perspective
visualize the AR on site (see Figure 8).
about the ancient architecture while visiting
archaeological sites,
• Suggesting an easily-usable, understandable ACKNOWLEDGEMENT
and informative medium with an user-friendly This study was made during post-doctoral research
interface running on a tablet / smartphone in ARC Group (MIT), with a collaboration of authors
for the children and students interested in ar- in 2015. We would like to thank Nikolaos Vlavianos,
chaeology, Omer Ozturk, Mustafa Yildizli and Yichen An for
• Contributing to the presentation and preser- their contributions in the field work and modelling.
vation of Turkey's cultural heritage by docu- We also would like to thank the rest of ARC Group
mentation, (MIT) previously collaborated in Multirama project;
• Providing a means for the reconstruction of Woong-ki Sung, Daniel Tsai and Howard Burns.
the historical architecture. The project presented in this paper was funded
by the postdoctoral research grant of TUBITAK 2219
fellowship and Istanbul Kemerburgaz University. We
FUTURE WORK would like to give sincere gratitudes to Parion Ar-
There are two future studies which will follow this chaeological Expedition Team, Prof. Dr. Vedat Keles
project. First is an MIT- Singapore University col- and Asst. Prof. Dr. Ertug Erturer for providing the
laboration which will be focusing on structuring a study environment and for their invaluable support.
Knowledge acquisition through expert in- constraint as it plays an important role in identifying
terviews architectural styles.
An expert interviewed was a researcher from the Na-
tional Research Institute of Cultural Heritage with ex- Ontological approach to well-indexed set of
perience of 19 years. From interviewing the expert, cases
we had two findings. First is, since we do not have The ontological structure suggested in this paper
access to Chinese and Japanese architecture data in is designed for digital architecture education. As
a given amount of time, we should narrow down the shown in Figure 2, a 'wooden architectural heritage' is
case library scope to the existing national treasure divided into three classes which are context, compo-
architectures from 'after middle age architecture' in- nent, and style. The context class includes informa-
cluding early, middle, and late eras of Goryeo Dy- tion that affects the style of the building such as its
nasty and early, middle, and late eras of Joseon Dy- period when built, location of its existence, and us-
nasty. Second is, as the experts do, we should incor- age or the function. The components class includes
porate the location of the architecture into our sys- four of the main parts of a building because Ko-
tem as one of the constraint along with the era and rean architectural heritages are mainly wooden and
the usage of the architecture; therefore, the user in- have a precast structure. The four parts represen-
put information was revised to era, usage, and loca- tative at each of their vertical placements are base,
tion. For the era, the experts refer to the architectural bracket, frame, and roof. For example, on the bot-
styles appeared during the same era and estimate the tom of the building is the base component that in-
architectural style of its golden age. As for the loca- cludes elements such as basestones, steps, and sty-
tion, the experts refer to the architectural styles ap- lobates. Likewise, on the top of the building is the
peared near the site of excavation. To summarize, roof component that includes elements such as dori,
we learned that we should narrow down the case li- rafter, and rooftile. Lastly, the style class includes all
brary to existing national treasure architectures and possible components' types.
that we should include the location information as a
considerforstylobate). Once the cases are narrowed system will search for the cases that meet the require-
down from the stylobate selection, then the cases will ments. For example, historical cases with the same
be asserted as (status considerforbasestone). era, usage, and location are considered. However,
on the interface is the complete list of all the types
IMPLEMENTATION under the architectural component. The panel lo-
Using the built cases, tasks, rules, and control meth- cated under the phrase "RECONSTRUCT YOUR BUILD-
ods, we designed the system GUI and implemented ING HERE" is the 'Instruction' panel where the system
into a system using JESS for inference engine and and the user can communicate especially to explain
JAVA swing for interface design. Our system, Logi- why the selected type is wrong. Below that is the
Rec, is divided into seven tasks (see Figure 4). The 'Workspace' panel which displays all the choices for
'User Inputs' panel on top left is the panel that re- the building components. The 'Check' button veri-
trieves the initial information for the system to run fies whether if the selection is appropriate. On the
such as the building name, the era, the usage, and the bottom is the 'Information' panel which provides ex-
location. When the 'Submit' button is pressed, the planation for the building component with an image.
Figure 4
Tasks and Rules.
Table 2
Evaluation matrix
and examples of
users’ remarks.
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',
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.
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.
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.
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.
INTRODUCTION AND BACKGROUND situated in Saudi Arabia, well before the building was
As a part of the exhibition series World Architecture at inaugurated.
the Danish Architecture Centre [10] (DAC) in Copen-
hagen, the exhibition 'World Architecture - Snøhetta' Architectural representation and experi-
[13] opened to the public in June 2015. The exhibi- encing architectural space
tion, created by Snøhetta [12] and Danish Architec- It is not enough to see architecture; you must expe-
ture Centre, ran until September 27th 2015 and had rience it (Rasmussen 1959). There is a discrepancy in
13.000 visitors. Experimental use of virtual reality was the way we sense architecture and its representation.
implemented in the exhibition to demonstrate state For an architect it is important to be able to precisely
of the art communicative methods and tools as well express oneself and make sure that visions and inten-
as expand the range of media used to present archi- tions are fully understood. Architectural represen-
tecture in exhibitions. tations are primarily drawings or images, previously
The experimental VR model let the audience ex- handmade and nowadays mostly computer gener-
perience the vast space of the Central Hall in King Ab- ated from 3D models, but it is still 2D representations
dulaziz Centre for World Culture [11] in a 1:1 scale as presented either on paper or on screen. Although of-
if they were actually present in the building complex ten supplemented by scale models to touch, the 2D
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,
Figure 5
KACWC position B.
itive with 31% of survey respondents experiencing dered from within Unity 3d game engine in a rather
the realism to be "Fantastic" and another 43% respon- blurry resolution!) displayed in GearVR on a Samsung
dents experiencing the realism to be "Good". Degree Galaxy S6 [7] mobile. The comparison was interesting
of vision and comfort were experienced as "Fantastic" because the high quality VR model used static points
by 53% and 44% of survey respondents respectively. of views, and did not utilised the possibility of move-
See figure 6. ment. The same static point of view in a pre-rendered
Visitors used Oculus DK2 to view the looping VR stereoscopic 360° rendering can be rendered directly
model from the exhibition and had the opportunity from the render software without the need of time
to compare it with a pre-rendered stereoscopic 360° consuming optimising for real-time rendering in a
image of the same VR model (in this example ren- game engine.
Figure 6
Experience of the
room with Oculus
Rift DK2.
Difference in experience of realism and image quality ished stone, glass and concrete, only the rammed
was noticed though including the smaller view frus- earth (walls) was generally confused with other ma-
tum in Gear VR (100°) compared to Oculus DK2 (120°). terials like cork, wood or brick. One exception was a
See figure 7. participant confusing the polished stone floor with a
With the use of high quality renders, the GearVR carpet.
does have great potential - it is cheap, portable, and Light. Many participants described in detail different
you can store many 360-renderings on the mobile. light settings, contrasts and emphasized the natural
or artificial appearance.
Walk-and-talk One architect commented on the different light
The walk-and-talk interviews were conducted during setups in the two positions in the VR model:
two days in November 2015. (Pos. B) "I can see they have been working with
Sixteen questions in three categories were in- artificial lighting in the ceiling. I am not particularly
cluded in the interview guide, asking for comments mad about it, if I may say so. I think that the skylight
ranging from purely descriptive to emotional and in- at the end of the room is impressive though, but it
tuitive reflections. feels like some daylight is missing, compared to the
Participants were architects (4) and users of ar- other place I was before.
chitecture (5) and the two groups had very different (Pos. A) Here there are the same ceiling lights
approaches to VR. The architects reflected mostly on as in the other place, but here it is much lighter in
the representation of architectural space whereas the a pleasant way. It is because I can see that the light
architect-users reflected on the virtual space and not comes from above. There is a place by the stairs
the representation of space. where some light is, it could be daylight."
Mass and volume. All participants described the dif- In the final focus group discussion her conclusion
ferent components of the space, especially the con- was:"I think what worked best in VR was the repre-
trast between organic volumes emerging from regu- sentation of the light. [...] You got a great feel for
lar straight ceilings and walls. Most participants also the character of light in VR, and you could experience
commented on the large number of columns scat- the various light settings, daylight, reflected light and
tered in the hall. various types of artificial lighting. It is difficult to rep-
Materials. The participants perceived most of the resent light in an image. Here I really think it [VR] has
materials correct. They identified wood, metal, pol- potential." (AN, Interview C)
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
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
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.
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.
the documentation, especially when numerous top- can only modify her or his own annotations but not
ics were discussed, lacked precision, structure and a annotations created by a remote user. In tight cou-
good degree of self explanation / visual reference. pling mode however, you can modify all annotations.
This is, why we implemented two methods to ease The absolute scale of the annotations is determined
the documentation: Semi-automated documenta- locally so that labels are readable from any location
tion in an office document and semi-automated doc- and on any type of device.
umentation in the BIM model. Both methods overlap
to a certain degree. Figure 3
Office document
BIM / VR meeting documentation in office with automatically
documents taken snapshot
Any time during a team discussion in VR, the partic- from the CAVE and
ipants can take 2D snapshots of the current 3D view manually added
(see figure 3). They can also make annotations in comments.
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 A comprehensive video recording plugin allows to
viewer. Annotations which are far away are rendered record live 2D or 3D videos from a VR session.
larger so that they are still readable from the users Through a tablet user interface, the user can chose
position. In collaborative sessions, it depends on the between different frame rates, encoders and resolu-
current collaboration mode whether any participant tion and other encoder specific parameters. In a col-
can modify and annotate. In Master-Slave mode, one laborative session, videos are only recorded in the lo-
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
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.
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.
tion process to be customised to the needs of the de- tributes are created to store the number of ver-
signer. The customised rules can take into account a tices, perimeter, area, and normal vector. For poly-
variety of factors, including the scale of the project, gons, specific shapes are recognised, and a series
the architectural design language, and the type of of Boolean attributes are created. These include is-
construction. Through an iterative process of tweak- convex, is-rectangular and is-stable. A stable polygon
ing rules, designers will over time be able to develop is defined as one whose lowest edge is horizontal.
libraries of personalised rules. Lines and polygons are also categorised according to
In step 4, the transitional IFC model is down- their angle of inclination, into one of four categories:
loaded and imported into the BIM application, where is-horizontal (0°), is-sloping (1 to 45°), is-leaning (46 to
the IFC elements can be replaced with more complex 89°), and is-vertical (90°).
native BIM objects. There is therefore no expectation The relationships between lines and polygons
that this materialisation process will create the whole are analysed for a given tolerance. For each type
model. It is more likely that the process creates a BIM of relationship, attributes are created with the at-
model that can be readily populated with higher res- tribute value being a list of IDs of the entities in-
olution information of the building project. volved in the relationship. Contact relationships are
any lines or polygons that touch each other in some
Geometric DXF Model to Decorated Model way. Attributes are created for various types of con-
The DXF model is assumed to contain highly simpli- tact between points, edges, and faces. For poly-
fied representations of a building following a set of gons, two additional types of relationships are co-
modelling conventions, using only planar polygons planarity and containment. Co-planarity refers to all
and lines. Building elements such as walls, slabs, polygons that are co-planar to a given polygon, with
doors and windows are represented using polygons, a single co-planar attribute being created for each
while building elements such as columns and beams polygon. Containment refers to a relationship be-
are represented using lines. The analysis step first tween co-planar polygons, where the interior of one
assigns unique IDs to the entities in the model, and polygon completely contains another polygon. At-
then analyses the individual entities and their topo- tributes are created for both contained-by and con-
logical relationships. tains.
For lines, attributes are created to store the
length and the direction vector. For polygons, at-
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, In the templating step, the decorated model is con-
six windows; three on each level, one door on the first verted into a transitional IFC model. This step starts
level and a roof at the top level. by identifying all the floors as horizontal polygons
In the analysis step, the DXF model imported with upward pointing normal, of which there are two.
from Rhino is analysed and a decorated model is gen- Since these floors are not co-planar, two separate
erated with additional attributes. For this example, IFCBuildingStories are generated: a ground storey and
there are no columns and beams, and as a result the a first storey. The remaining entities in the geometric
model only contains polygons. model are then categories into the different building
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-
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.
BIM | Concepts - Volume 2 - eCAADe 34 | 597
sible by the structural system framework, digital processes. It is structured based on a set of mod-
project documentation would facilitate adaptabil- ules devised to aid in the process of modeling, simu-
ity, while overcoming structural and technical chal- lation, leading to precise fabrication, and assembly of
lenges. Given that parts and components of the components. Commonly, a BONE Structure residen-
structure are assembled using bolts and fasteners, tial home project would include between 200 to 400
and all of these connections are documented within components in variable quantities, for a total in the
the comprehensive 3d model, variation in building range of 8000 to 15000 individual steel parts. These
envelope can be achieved effectually. while in the components are divided into twelve families, based
case of pre-construction design, the placement al- on the type of material and manufacturing process.
gorithm tolerates configuration of special parts and In addition to just-in-time virtual tracking of
components, by replacing standard ones, yet post- components during design, the application of Dig-
occupancy adaptability can be achieved by disas- ital Prototyping informs the fabrication and assem-
sembling components, then reassembled following bly of parts. Autodesk Inventor generates fabrication
other configuration in response occupants needs. To data to feed different Computer Numerical Control
elaborate on customized design specifically, figure (CNC) machinery involved in the process with precise
3 represents a project that dictated the design and instruction. In addition to standardized parts, many
fabrication of special parts to suite the design, while of the projects involve the design and production
keeping the principles; materials, process, and stan- of special parts for the sake of customized housing.
dardized connection details of the BONE Structure These parts are designed, tested, optimized, simu-
system. The flexibility in design and optimization in lated, and fabricated by simply manipulating param-
production are embedded in the process based on eters of the standardized parts. This also leads to sim-
the digital strategy included particularly in the coded plifying the workflow from design to manufacturing,
algorithms. 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
Digitally supported fabrication, and assem- instructions. These instructions document quantities
bly of families and hardware components, steel parts,
The focus of the applied integrated digital strategy fasteners and welding instructions, if applicable, as
is directed towards a multi-level modeling environ- well as precise sequential assembly procedures as
ment that benefits abundantly from Computer Aided shown in figure 4. Accordingly, the process becomes
Design/Computer Aided Manufacturing (CAD/CAM) a predefined systematic practice that leverages qual-
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
Figure 2
Three eras of CAD.
Adapted from (Aish
2013). Sizes of the
grey arrows refer to
current usage of
each modeling
method.
proaches. Figure 5 presents the main diagram, which sign and construction. OOD focuses on standardiza-
shows the discussed concepts together. The figure tion of design, fabrication and assembly, as AAD en-
affirms that AAB is not only connecting design tools ables non-standard procedures in them.
and design methods but also connecting two sepa- As a consequence of industrial revolution, in-
rate digital design approaches called Object-Oriented dustrial processes are commonly based on mass-
Design (OOD) and Algorithm-Aided Design (AAD). This production, on which the OOD approach relies on.
connection opens many new possibilities to CAD by Logics of form-making have been derived from in-
enabling many design processes that have not been dustrial construction standards (Denari 2012). Even
possible before. AAB tools enable designers to use current BIM software are based on these standards.
object-oriented and algorithm-aided design meth- AAD relies on mass-customization, where the effi-
ods in the same design process. In other words, script ciency can emerge in structures and construction
based design methods, such as modeling, analysis, even if all building objects are singular (ibid). Cus-
simulation and optimization, can be combined with tomization was challenging before because there
manual object-oriented design methods. This opens were no tools to fabricate custom objects. However,
new possibilities for collaborative design. technological enhancements in building component
These two digital design approaches have been production have enabled many fabrication processes
rather separate until recent tool developments. Both that were not possible before.
of these sides focus on different aspects of digital 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.
• every relationship set in the submodel is sub- into multiple elementary submodels and a remaining
sets of any relationship set in the original link model, consisting only of relationship sets.
model In order to split a model M = (O, A, R) into
n submodels M1 , M2 , . . . , Mn , the following steps
Thus, a model M ? = (O? , A? , R? ) is a submodel must be conducted:
of another model M = (O, A, R) if and only if the 1. Partition O according to an equivalence relation
following conditions are satisfied: Rsplit ∈ O × O into n partitions O1 , O2 , . . . , On
2. Partition each A ∈ A into n partitions
O? ⊆ O (6)
A1 , A2 , . . . , An , such that each partition contains
the object-value-tuples with the objects beeing from
∀A? ∈ A? ∃A ∈ A : A? ⊆ A (7)
the respective object set partition:
∀R? ∈ R? ∃R ∈ R : R? ⊆ R (8) ∀i ∈ {1, . . . , n}∀(o, v) ∈ Ai : o ∈ Oi . (9)
By selecting a subset of the objects in an information Each attribute function Ai is then the restriction of A
model O? ∈ O it is possible to also select the sub- to the respective object set partition Oi : Ai = A|Oi .
sets of attribute functions and relationship sets con- 3. Partition each relationship set R ∈ R into n
cerning only the selected objects, such that the se- partitions R1 , R2 , . . . , Rn , and a reminder R0 , with
lection itself is a valid information model according to each object related through a relationship set Ri (ex-
our definition. More generally, based on a partition of cept the reminder R0 ) being in the object set Oi of
its object set, a given information model can be split
parts, since the complex visualization techniques are The current prototypical implementation of a config-
to date only described in an informal way. The formal urable visualization framework lacks functionality in
description of the visualization process allows for a the area of complex visualization and might be ex-
precise formulation of the model yielding a threefold tended based on the formal model described here
benefit. in the future. Third, it allows for further studies with
First, it fosters a better understanding of visual- a quantitative approach, where methods to bench-
ization processes in the context of construction spe- mark visualization techniques could be based upon.
cific environments and developments. Other ap- We acknowledge the support of the German
proaches focus ether mainly on the data or on the vi- Academic Exchange Service in the form of a Travel
sualization side, while the approach presented here Grant which enabled the attendance of this confer-
seeks to connect both sides with an integrative ap- ence.
proach.
One recent development in the construction REFERENCES
area is the multimodel, which attempts to connect Brunetti, JM, Auer, S, Garćia, R, Klímek, J and Nevčaský,
growing amounts of information while it maintains a M 2013 'Formal Linked Data Visualization Model',
separation of concerns by keeping the diverse mod- 15th International Conference on Information Integra-
els separated. A generalized notion of the multi- tionand Web-based Applications & Services (iiWAS), Vi-
model may serve as a general structuring principle enna
Chen, PPS 1976, 'The Entity-Relationship Model: To-
for large information stocks and as such may be used
ward a Unified View ofData', ACM Transactions on
in other areas than visualization. Database Systems, 1(1), pp. 9-36
Second, the formal model may serve as a base Chi, EH 1999, A Framework for Information Visualization
for the implementation in a software application. Spreadsheets, Ph.D. Thesis, University of Minnesota
Figure 11
Workflow for
Prototype 3.
Figure 1
Six types of
building element
relations.
INTRODUCTION: WHY ARE CHANGES IN resents the current state of the building has to be as-
BUILDINGS OBSERVED? sessed. The currently fastest and most precise way
The monitoring of the progress on building sites, the to document the geometry of an existing building is
control of geometric deviations between planned by means of Point Clouds, which are generated by
and as-built geometry and the check of planarity of 3d laser scanners or photogrammetric devices.Point
building elements during construction planning are clouds consist of millions of unrelated coordinates.
essential tasks in AEC. Further applications are lo- This is diametrically opposed to BIM, the current state
cated in the area of Facility Management and plan- of the art paradigm for architectural planning and fa-
ning of renovation tasks, where the detection of dif- cility management, which represents relational ob-
ferences between a documented state and the cur- ject oriented data in a hierarchical structure.
rent state are a prerequisite for decision making and Hence, finding differences between these two
design, and where the quality of a 3D model that rep- types of datasets first of all necessitates to represent
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
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
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.
Table 1
Relevant tasks for
collaborating
taking into account
sensitive spatial
dimensions.
then to determine ambiance fostering the unfolding tive and topological requirements previously set by
of each <Activity Type>. Each <Ambiance> has spe- the designer based on the requested program.
cific requirements (see Figure 2).
A space is governed by a set of design require- CONCLUSION AND FUTURE WORK
ments depending on the <Space Type> in which it This study presents a review of architectural space
belongs, following a topological preference or ac- as a key concept to reveal a sensitive dimension of
cording to ambiance required by the <Activity> that architecture in current BIM practices during design
will be housed in. Qualitative requirements may phases. Observing recent research having dealt with
also be associated with a <Space Type>. Thus each space data modelling has allowed identifying the lim-
<Space> must obey the common requirements for its of these models to integrate sensitive spatial data.
its <Space Type>. Requirements related to the space Exploring existing BIM tools has enabled this study
<Topology> define its relations with the remaining to determine missing tasks allowing designers to de-
building spaces. E.g. Communication relation or sign and co-produce spaces, to coordinate and also
closeness relation between two spaces, openness re- to communicate about spaces. In particular, about
lation or extension relation with the outside, etc. All spaces sensitive aspects. As shown in Figure 1, only a
these requirements are variable, unlike regulatory re- small number of tasks are supported, even partially,
quirements. They may vary from a project to another by tested tools. We noticed that tasks 6, 7, 9, 10, 11
according to the climatic and geographical condi- and 12 are not addressed by all of existing BIM tools.
tions of the project, etc. and especially according to Visibly, dRofus has the greatest ability to enable re-
the future user preferences.Our first space model is maining tasks.
illustrated in Figure 2 using Express-G diagram sym- A future work will include improving this first
bols. In this model, we tried to take into account sen- model. The final global model will be used to imple-
sitive spatial data highlighted in this paper. Based ment a decision support method allowing designers
on defines core concepts, a control protocol is pos- to control the spatial quality of the designed build-
sible to detect the spatial quality issues over qualita- ing, during cooperative work. DRofus seems to con-
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
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.
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.
in using the room as the principle unit of design and ceives a link to an instance of topoBIM pre-loaded
analysis. Each room model contains semantic infor- with a set of room models specific to a given de-
mation and geometry for the room itself and asso- partment in the hospital (see Figure 3). A collection
ciated furniture and equipment, allowing users of of rooms can then be selected from this list by the
the software to inform design proposals with plau- user and loaded into the 3D viewer. Each room ap-
sible assumptions about the requirements for stan- pears as a simplified volume with associated equip-
dard room types. The following is an outline of the ment and furniture (see Figure 4). Multiple 3D lay-
proposed topoBIM workflow. outs are available in the interface. Directed connec-
Each participant in the consultation process re- tions can be added by clicking on two rooms in the
pertise: for example a surgeon can create and review participation can be enhanced when these applica-
only connections related to the circulation route for tions are designed in a way that promotes the overall
surgical staff in a particular section of the hospital, transparency of the design process, a goal that can
without the distraction of considering overlapping be accomplished in part through the use of human-
and adjacent activities within the building. readable (text-based) file formats, parsimonious user
Currently there is a gap in the BIM-based design interfaces and modularity of functionality and de-
process between the BIM authoring environment on sign. The intention is to enable additional oppor-
one hand, and on the other standard data formats tunities for designers to apply their creativity and
such as COBie and IFC. Both the proprietary author- intuition in problem-solving, rather than relying on
ing software and the open data standards are restric- the computer to automate the process of evaluating
tive in their own way to stakeholder participation, building data in early design.
and can benefit from the addition of modular soft- The purpose of topological representation in this
ware applications which add specific functionality to project is to isolate a specific element of the BIM data
the BIM design process. We propose that stakeholder model for use by a particular stakeholder. This ap-
CONCLUSIONS REFERENCES
This project has focused on the collection of stake- Bastian, M, Heymann, S and Jacomy, M 2009 'Gephi: an
holder knowledge in early design rather than anal- open source software for exploring and manipulat-
ysis of this information. As discussed earlier there ing networks', Proceedings of the International Con-
ference on Web and Social Media (ICWSM), pp. 361-
is a long and well-established research trajectory in
362
building performance analysis based on topological Biswas, T, Wang, T-H and Krishnamurti, R 2013, 'From de-
representation, and we have deliberately not incor- sign to pre-certication using building information
porated analysis in topoBIM. Following the principles modeling', Journal of Green Building, 8(1), pp. 151-
of simplicity and modularity described above, our 176
aim in this project has been to provide only the func- Coates, P, Derix, C, Lau, T, Parvin, T and Puusepp, R 2005
'Topological approximations for spatial representa-
tionality needed to collect information that might
tions', Proceedings of the Generative Arts conference,
not otherwise be recorded and store this information Milan
within the BIM model. We anticipate that the analysis Derix, C 2010 'Mediating spatial phenomena through
and use of this information will become the subject of computational heuristics', Proceedings of ACADIA
a future project. 2010, New York, pp. 61-66
In future work we will investigate the application Eastman, CM 1973, 'Automated space planning', Articial
Intelligence, 4(1), pp. 41-64
of topoBIM at multiple scales. Previous research has
Hillier, B, Hanson, J, Peponis, J, Hudson, J and Burdett, R
demonstrated the validity of topological representa- 1983, 'Space Syntax', Architects Journal, 178(48), pp.
tion for a range of scales including design for manu- 67-75
facture and assembly (DfMA), the whole building and Jo, JH and Gero, JS 1998, 'Space layout planning using
the city. Although our current functional prototype an evolutionary approach', Articial Intelligence in En-
focuses on the scale of a single building oor or de- gineering, 12(3), pp. 149-162
Kassem, M, Brogden, T and Dawood, N 2012, 'BIM and 4D
partment, it can be made to work with these addi-
Planning: A holistic study of the barriers and drivers
tional scales with minor modification. We will also in- to widespread adoption', Journal of Construction En-
vestigate in future work the modification of existing gineering and Project Management, 2(4), pp. 1-10
BIM data formats to include the possibility of captur- Krishnamurti, R, Toulkeridou, V and Biswas, T 2014, 'Com-
ing topological information such as that produced by municating semantics through model restructuring
our application. One exciting possibility to explore in and representation', in Kensek, K and Noble, D (eds)
2014, Building Information Modeling: BIM in Current
future work is the integration of topological user an-
and Future Practice, Wiley
notations in a future version of the IFC. Langenhan, C, Weber, M, Petzold, F, Liwicki, M and Den-
TopoBIM has been developed as a collaboration gel, A 2011 'Sketch-based methods for researching
between researchers at the University of Sheffield building layouts through the semantic fingerprint
School of architecture and London architecture firm of architecture', Proceedings of the 14th International
Bryden Wood Limited. The use of topological rep- Conference on Computer Aided Architectural Design
Futures, Liege
resentation for knowledge capture was developed
Liggett, RS and Mitchell, WJ 1981, 'Optimal space plan-
in a series of projects at Bryden Wood Limited, the ning in practice', Computer-Aided Design, 13(5), pp.
most recent of which is described in this paper. Fund- 277-288
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.
INTRODUCTION: POINT CLOUDS IN ARCHI- hard- and software. Their proliferation is hence di-
TECTURAL DESIGN rectly linked to the advancements in computational
The last years has seen a massive proliferation of 3D power. Point Clouds are a relatively new geomet-
scanners in all building related disciplines[1][2]. Af- rical class, which was only introduced in the 2000s
fordable scanners are today available for all sorts of (Rusinkiewicz and Levoy 2000). Since then only spe-
applications from airborne scans with drones, over cialist software was able to handle point clouds with
high-resolution indoor scans to handheld scanners, reasonable speed. The current market of point cloud
which provide real-time data. 3D capturing technol- software is oriented towards professionals from en-
ogy is becoming increasingly faster and mobile (see gineering and land surveying industries. The leading
Figure 1) and provides a new link between the build software as those from 3D laser manufacturers, such
environment and its representation in architectural as FARO scene [3] or Leica Cyclone [4], from CAD pro-
design. ducers, such as Bentley Pointtools [5] or Autodesk Re-
Cap [6]. Design and engineering processes can cur-
Point clouds as a geometrical class rently not take place in Point Clouds, which neces-
The output of all 3d capturing devices are point sitates a transformation of these into CAD formats
clouds. Their sheer size poses a challenge to the cur- by means of software, such as FARO Point Sense [7],
rent computational infrastructure and processes of Clearedge Edge Wise[8], Trimble Real Works[9], Sca-
design.Point clouds are computationally heavy and lypso [10], PointCap [11]. The mentioned tools are
the processing of them is demanding in terms of expensive: a Faro Scene licence costs e.g. around
11.000€ [12]. The functionality of these tools is ma- be read by some architectural modelling programs,
ture and they have e.g. their own advanced graphic such as Recap for Autodesk Revit or E57 and XYZ
pipelines. All of this software is however conceived as point cloud formats for Archicad. These design en-
processor of point cloud datasets before their inges- vironments do not foresee a further interaction with
tion into stakeholders design software. For this task the point clouds. Users have currently hardly any
a set of generic workflows is offered, which allows means to manipulate the point clouds during the de-
specialists to clean, subsample and transform and to sign process. This results for instance in retrofitting
some extend analyse point clouds and their quality projects in a situation, where the core of the design,
(Tamke et al. 2016). The outcome of these processes the representation of the building to be designed, is
is either a digital model, in a format that is directly in fact not accessible by the architectural designer.
editable in a CAD package (e.g. polygon, Nurbs or This results in an overly complex workflow, where
BIM formats) or a processed point cloud, which can even simple tasks, such as the transformation of point
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.
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