Available online at www.

scie
A encedirect.com
m
A
Available online at www.scie
encedirect.com
m
Scie
enceDir
rect
Scie
Available enceDir
Availableonline rect
onlineatatwww.sciencedirect.com
www.sciencedirect.com
Procedia Man
nufacturing 00 (20019) 000–000

ScienceDirect
ScienceDirect 
Procedia Man
nufacturing 00 (20019) 000–000 www.elsevier..com/locate/proceedia
www.elsevier..com/locate/proceedia
Procedia Manufacturing 28 (2019) 18–23
Procedia Manufacturing 00 (2017) 000–000
www.elsevier.com/locate/procedia
Internatioonal Conferrence on Ch
hangeable, A
Agile, Recon
nfigurable and
a Virtual Production
n
Internatioonal Conferrence on Ch
hangeable, A
Agile, Recon
nfigurable and
a Virtual Production
n
Automat
A ted prodduction data d feeddback forr adaptiv ve work planning g
Automat
A ted prodduction data d feeddback forr adaptiv ve work planning g
and d produc ction conntrol2017, MESIC 2017, 28-30 June
and d producction con
Manufacturing Engineering Society International ntrol
Conference
2017, Vigo (Pontevedra), Spain
Berend Deenkenaaa, Marc-André
B M Dittrichaa, Sören Wilm
msmeieraa *
B
Berend Deenkena , Marc-André
M Dittrich , Sören Wilmmsmeier *
Costing models
G a
forLeibniz
Gottfried Wilhelm capacity optimization
Universiity Hannover inEngineering
- Insstitute of Producttion Industry 4.0:
and Machine Trade-off
Toools,
G a An derrityUniversität
Gottfried Wilhelm Leibniz Universi Hannover 2, 300823
- InsstituteGarbsen, Germany
Gtion Engineering and Machine Toools,
of Product
between usedAncapacity derr Universität 2,and operational
300823 Garbsen, Germany
G efficiency
Absttract A. Santana , P. Afonso , A. Zanin , R. Wernke
a a,* b b

Absttract a
University of Minho, 4800-058 Guimarães, Portugal
Highher customer’s eexpectations annd new technolo ogical developm ments have resu ulted in an increased complexi xity of manufactturing
b
Unochapecó, 89809-000 Chapecó, SC, Brazil
syste
High ems.customer’s
her Consequenntly, work plan
eexpectations nning
annd new and produ
technolo uction
ogical control
developm hhave
mentsbecome more
multedcrucial
have resu than
in an incr neased
ever complexi
to the suuccess
xity of compturing
of manufact anies.
Adap ptive Consequen
systeems. work plaanning
ntly, workandplanprooduction
nning andcontro
produ ol approaches
uction oofferbecome
control hhave high pote
mentialcrucial
more with than
resnpect
evertotoflexibiility and
the suuccess of companies.
lexity
mana
Adap agement.
ptive workHowe ever, the
plaanning andappro oaches find
prooduction litt
contro tle approaches
ol use in pracctice,
ooffer since theential
high pote automatic
with acqresquisition of allility
pect to flexibi potentially
and comprellevant
lexity
infor
mana rmation
agement.with
Abstract Howe addditional
ever, the sensor rs is cost-intens
approoaches siveuse
find litttle andinnot fectice,
prac easiblesince
in many
they automatic
cases. At theacq same time,
quisition alle potentially
of the potential of rel
allready
levant
exist
inforting
rmationproduction
with ad ndditional
data, that is stored
sensor srs is cost-intens
in Manusive facturing
and notExecu
fecution
easible Systems
in many y (MES),
cases. Atremain
the ns untapped.
same time, theT
This paper of
e potential preseents a
allready
meth
existhod
Under for
ting the automat
production
concept nted production
data,
of that sdata4.0",
is stored
"Industry feedback
in Manu that
tfacturing
production guarantees
Execucution
processes will beupdate
a systematic
Systems u (MES),
pushedof remain
produ
to be ction
ns plan data
untapped.
increasinglyTa only
This basedprese
paper onents
interconnected,MESa
data.
methhod Theforfunctiona
automat ality
ted of the
production me thod
data is validat
feedback ed
that
t exemplarily
guarantees y aat a thread
systematic manufacturer.
mupdate
u of
information based on a real time basis and, necessarily, much more efficient. In this context, capacity optimization produ T he results
ction plan reve
dataeal
a that
only an
based ME
on S can
MES
provi
data.ideThea sufficient t database
functionaality of thefor meadaptive
athod is validat
work planning
p exemplarily
ed and pyproduction cont
at a thread mtrol approaches
manufacturer. Ts.heMoreover, thhe
results reve ealdeveloped
that an ME method
mS can
goes beyond the traditional aim of capacity maximization, contributing also for organization’s profitability and value.
can
provibe
bideapplied to idtentify
a sufficient suitable
database aworkstations
w
for adaptive workan nd
p measuremen
planning and nt categoriescont
pproduction forrtrol
additional
approachessens.sor implementa
Moreover, ation.
thhe developed method
m
Indeed, lean management and continuous improvement approaches suggest capacity optimization instead of
can be
b applied to identify suitable workstations
w annd measuremennt categories forr additional sensor implementaation.
maximization.
© 20019 The Author The rs. study
Publishedof capacity
byy Elsevier optimization
B.V. and costing models is an important research topic that deserves
© 2019
This
© 20019 The
contributions
is an Authors.
open
The from
acce
Author Published
both the
essPublished
rs. article unde by Elsevier
practical
byyr Elsevier and
the CC BY-NB.V.
theoretical
B.V.NC-ND license perspectives. Thisecommons.org/l
((https://creative paper presents and discusses
licenses/by-nc-n a mathematical
-nd/4.0/)
This is an
model foropen accessmanagement
capacity article under the CC BY-NC-ND
based on different license
costing (https://creativecommons.org/licenses/by-nc-nd/4.0/)
models (ABC and TDABC). A generic model hasable
been
This-review
Peer- under
is an open acce reess
esponsibility
article unde off rthe
thescientific
CC BY-N coommittee
NC-ND of thee((https://creative
license International Conference
Cecommons.org/l on Changeable,
Clicenses/by-nc-n
Agi-nd/4.0/)
ile, Reconfigura
Peer-review under responsibility of the scientific committee of the International Conference on Changeable, Agile, Reconfigurable
developed
and
Peer-Virtual
V and it
Producti
-reviewProduction.
and Virtual
was
ion. used to analyze idle
under reesponsibility off the scientific co capacity and to design strategies
ommittee of thee International Conference
C towards the maximization
on Changeable,
C of organization’s
Agiile, Reconfiguraable
value.
V TheProducti
and Virtual trade-offion. capacity maximization vs operational efficiency is highlighted and it is shown that capacity
optimization
Keyw might
words: Adaptive hide operational
pproduction inefficiency.
planniing; Digital manu
ufacturing system
m; Production dataa
Keyw
© words:
2017 Adaptive
The pproduction
Authors. planni
Published bying; DigitalB.V.
Elsevier manu
ufacturing system
m; Production dataa
Peer-review under responsibility of the scientific committee of the Manufacturing Engineering Society International Conference
2017.

Keywords: Cost Models; ABC; TDABC; Capacity Management; Idle Capacity; Operational Efficiency

*1.Corresponding
CIntroduction autthor. Tel.: +49 5111 762-18285; fax x: +49 511 762-5 115.
* E-mail
E W
Wilmsmeier@ifw.
address:aut
Corresponding
C thor. Tel.: +49 5111 uni-hannover.de
762-18285; faxx: +49 511 762-5 115.
E-mail
E cost
The address: W
Wilmsmeier@ifw.
of idle capacity is a uni-hannover.de
fundamental information for companies and their management of extreme importance
2351-
in -9789 ©production
modern 2019 Thhe Authors. Publis
systems. shed
In by Elsevier
general, it isB.V.
defined as unused capacity or production potential and can be measured
This is
i an open
2351--9789 access
© 2019 s article
Thhe under
Authors. theeshed
Publis CC BY-NC-ND DB.V.
by Elsevier license (https://ccreativecommonss.org/licenses/by-nc-nd/4.0/)
in several
Peer-r
ways: tons of production, available hours of manufacturing, etc. The management offigurable
the idle capacity
ireview
This is under
an open resp
accesssponsibility of thee scientific
article under comm
CC BY-NC-ND Dittee of the
license Internnational
(https://c Conferen
creativecommons nce on Changeablle,
s.org/licenses/by- Agile, Reconfi
nc-nd/4.0/) and Virttual
* Paulo
Produ Afonso.
uction. Tel.: +351 253 510 761; fax: +351 253 604 741
Peer-r review under respponsibility of thee scientific committee of the Internnational Conferen nce on Changeablle, Agile, Reconfi
figurable and Virttual
E-mail
Produ uction.address: psafonso@dps.uminho.pt

2351-9789
2351-9789 ©©2017
2019The
TheAuthors.
Authors. Published by Elsevier
Published B.V. B.V.
by Elsevier
Peer-review underaccess
This is an open responsibility of the scientific
article under committee oflicense
the CC BY-NC-ND the Manufacturing Engineering Society International Conference 2017.
(https://creativecommons.org/licenses/by-nc-nd/4.0/)
Peer-review under responsibility of the scientific committee of the International Conference on Changeable, Agile, Reconfigurable
and Virtual Production.
10.1016/j.promfg.2018.12.004
 Berend Denkena et al. / Procedia Manufacturing 28 (2019) 18–23 19
2 Author name / Procedia Manufacturing 00 (2019) 000–000

1. Introduction

Rising customer’s expectations and technological developments have led to an increased complexity of
manufacturing systems. Additionally, companies face stochastic disturbances and cyclic demands, which results in
an unbalanced utilization of manufacturing capacity. Thus, effective and efficient production planning and control
have become a central advantage in competition. While production planning defines the required production
technologies and strategies as well as the sequence of the production steps, production control is concerned with
reoccurring activities in production, like order release or machine tool allocation, and short-term rescheduling due to
unplanned machine breakdowns or deviations from planned times. However, the separation of work planning and
production control has proven to be too rigid and inefficient in case of single-part and small-series production due to
the following reasons [1]-[4]:

 Generation of unfeasible work plans by neglecting current conditions in production (e.g. machine utilization, tool
condition)
 Limited decision-making scope for production control due to limited process alternatives
 Lack of systematic evaluation of routing alternatives due to manual rescheduling by the production planner
 Focus on individual optimization criteria in production planning and control

A consequence of a separated production planning and control are, for example, extended throughput times,
nontransparent decision-making and a low ability to react in the event of faults and unplanned events. In order to
overcome these challenges, Denkena et al. developed a method for an automatic generation of alternative work plans
and an integrated production planning and control [5]. Prototypical implementations have shown that the throughput
time can be reduced by approximately 4% compared to the use of linear work plans under laboratory conditions [6].
However, the approach has not been implemented into practice. The main reason for this can be seen in the
required effort to maintain a valid database, which is necessary in order to achieve a high level of planning accuracy
and acceptance, in a constantly changing production environment. Examples for changes, which occur on a daily
basis in manufacturing, are amendments of drawings, new tools, machine malfunctions, stochastic deviations of the
throughput time (TPT) or staff shortages. The automatic collection of all potentially relevant information with
additional sensors is cost-intensive and not feasible in most cases. On the other hand, the potential of already
existing production data, that is stored in Manufacturing Execution Systems (MES), remains often untapped.

2. State of the art

The technical progress in data storage technology and the associated reduction in costs for the collection and
storage of data stocks have significantly increased the quantity and availability of production data in production
companies [7]. Due to the amount and complexity of the data, manual analysis is usually not effective and efficient.
In the literature, methods for an automated analysis of data are discussed under the generic term Knowledge
Discovery in Databases (KDD) [8][9]. The term refers to a process for obtaining (statistically) valid, so far
unknown, potentially useful and understandable knowledge. According to Fayyad et al., "KDD is the nontrivial
process of identifying valid, novel, potentially useful, and ultimately understandable patterns in data" [8]. Different
models of KDD process can be found in the literature. Most widely used is the model by Fayyad et al., depicted in
Fig. 1 (a). Albeit data mining is just one step of the KDD process, both terms are used partly as synonyms [7][9]. A
more detailed description of the KDD process with direct reference to the context of manufacturing is given by
Choudhary et al. [11].
In addition to this general approach of handling large amounts of data, there are also concrete methods for
adapting planning data in the literature. Monostori et al. [12] give an extensive overview about the use of cyber-
physical systems in production planning. Geiger and Reinhart [13] pointed out that actual production data is crucial
to the success of adaptive planning methods. Thus, they developed a Radio Frequency Identification (RFID) based
approach to obtain accurate data from the shop floor. For this purpose, components and production resources (e.g.
20 Berend Denkena et al. / Procedia Manufacturing 28 (2019) 18–23
Author name / Procedia Manufacturing 00 (2019) 000–000 3

machines) are equipped with RFID transponders and enriched the main database with relevant product and resource-
specific data (e.g. target work plan, times and costs, planned job sequences and machine utilization). After the data
acquisition and transformation, the data sets are segmented into clusters of variances between target and actual
values of TPT components (e.g. process-time, setting-up time, wait time, transport time). Aiming to identify
correlations between TPT deviations and product specific attribute, binary decisions trees are set up. Information
about the current condition of the production are also taken into account [13][14].
However, the method by Geiger and Reinhart has some drawbacks with respect to its practical implementation.
First, RFID transponders are used to collect component-related data. The integration of such transponders is not
feasible for some components and, as mentioned above, for many SME. Second, it can be assumed that the
deviations of the target values from the actual values are subject to continuous fluctuations, which is why clustering
must be repeated continuously. Since the decision tree for assigning the correction values from the clusters to the
TPT-components depends on the result of clustering, the subsequent method steps must also be executed for each
recalculation. The calculation effort can quickly become too high to guarantee a process-parallel update [14].
In contrast to the aforementioned approach, the following section presents a method that guarantees a systematic
update of production plan data based on typical MES data only and with little computing effort. Thus, the method
allows a process parallel execution.

3. Method for automated production data feedback

3.1. Data preparation

The method for automated production data feedback for adaptive work planning and production control is shown
in Fig. 1 (b) and based on the model by Fayyad et al. [10]. In the first step, the production planner identifies
potentially relevant influencing factors for time delays in the production process (e.g. machine tool in use,
processing step to be executed, component characteristics, and responsible employees) and exports the
corresponding historical data records from the MES database. This manual and company-specific selection ensures
that the scope of the analysis meets the conditions in the company (e.g. the selection of suitable attributes describing
the product, such as diameter or length) and company-specific agreements (e.g. regarding the handling of employee-
related data).
Next, available data is adjusted to avoid the identification of incorrect correlations and overlooking of additional
correlations. For this purpose, incomplete data records are removed, outliers are identified by statistical analyses,
e.g. box-plots, and excluded for further analysis. A study with an industry data set confirmed the common rule in
literature, according to which the whiskers of the box plot should correspond to 1.5 times the interquartile range.

Fig. 1. (a) KDD-process by Fayyad et al. [10]; (b) Method for automated production data feedback.
 Berend Denkena et al. / Procedia Manufacturing 28 (2019) 18–23 21
4 Author name / Procedia Manufacturing 00 (2019) 000–000

Fig. 2. (a) Correlation matrix with N potential factors of influence; (b) Principle of production data feedback (example for three significant
influencing variables).

3.2. Data transformation and data mining

The dependencies between the identified potential influencing factors are examined through a correlation
analysis. The focus is on the influence on the time difference between planned and actual time values. A separate
analysis is carried out for each TPT component. This means that the correction values represented by the mean
deviation between target and actual value can be calculated for each TPT component. In this study, a Pearson-
Correlation is used to identify potential dependencies. The total number of selected influencing factors N determines
the dimension of the correlation matrix (see Fig. 2 (a) with an N×N correlation matrix). To interpret the effect size r
of each factor combination Cohen provides the limits shown in table 1[15].
For factors that cannot be examined within the Pearson-Correlation, since nominal and metric scale levels have to
be compared (e.g. the dependency of the time deviations from workstation or work step), the error reduction measure
ƞ² is used. This represents a measure of the achievable percentagewise improvement in prediction by including the
variable examined. As a result, ƞ² can provide indirectly information about the strength of the correlation to the time
deviation. In literature, there are different approaches for assessing the intensity of dependency to calculated values
of ƞ². While Richardson discusses various limit values, Cohen suggests the levels shown in table 1, which will also
be used within this paper [15][16].

Table 1. Overview of correlation classes.

Correlation strength Effect size Error reduction measure
No correlation r < 0.1 ƞ < 0.1
Small correlation 0.1 ≤ r < 0.3 0.1 ≤ ƞ < 0.25
Medium correlation 0.3 ≤ r < 0.5 0.25 ≤ ƞ < 0.4
Strong correlation r ≥ 0.5 ƞ ≥ 0.4

3.3. Interpretation

In the last step, the identified patterns are interpreted. All influencing factors with a medium or strong correlation
are regarded as significant and used for production data updates. Fig. 2 (b) shows exemplarily the result for
identifying three significant influencing factors. For each possible combination of influence factors, the mean
percentage of all available time deviations is calculated and these values are uniquely assigned to the corresponding
factor combination. The method is not subject to any dimensional restrictions.
22 Berend Denkena et al. / Procedia Manufacturing 28 (2019) 18–23
Author name / Procedia Manufacturing 00 (2019) 000–000 5

Fig. 3. (a) Results of Pearson-Correlation; (b) Excerpt of resulting workstation-workstep-matrix for setup operations.

The steps presented do not have to be executed completely with each method call. The correlation analysis must
be performed during the first method execution. Afterwards it can be repeated manually if fundamental changes are
made to the production environment or the product structure. The same applies to the selection of potential
influencing factors. The other steps can be fully automated and require very little computing effort, which guarantees
process-parallel data preparation and update.

4. Results of exemplary method application

In the following chapter, results of an exemplary application of the method based on actual data sets from a threat
component manufacturer are presented. The analysis was carried out for setup and processing times. First, the MES
data for completed orders from a two year period were exported via an SQL query and outliers were determined and
removed according to equation (1) and (2). After incomplete data records were filtered out, a total of 19,854 useful
data records were available for the subsequent correlation analysis, of which 9,357 related to setup operations and
10,497 to processing operations.
In Fig. 3 (a) the results of the Pearson-Correlation are shown. Taking into account the significance limit, as
defined in chapter 3.3, the Pearson-Correlation provided no relevant correlations for setup and production processes.
The additional investigation on the influence of the work steps and workstations using ƞ showed for setup
operations the work step under consideration has an average influence (ƞ = 0.30) and the workstation used has a
strong influence (ƞ = 0.48) on the deviation from target to actual setup time. With respect to the processing operation
only a minor effect of the work step (ƞ = 0.20) and a medium influence of the workstation (ƞ = 0.30) was calculated.
As a result, both factors are relevant for setup operations in relation to the significance limit mentioned. For
production operations, however, the time deviation depends mostly on the selected workstation.
Based on the results of the correlation analysis, the feedback function was defined. In the case of processing
operations, this corresponds to a simple list, which assigns the average percentage of processing time deviations to
the available workstations. The results can be summarized in a matrix. Fig. 3 (b) depicts exemplarily an excerpt of
the workstation-workstep-matrix for setup operations.

5. Summary and Outlook

The successful application of adaptive work planning and production control requires extensive data acquisition.
However, the implementation of additional sensors or tracking devices is not affordable for most companies,
especially for SME, and sometimes technically not feasible. The data basis of an MES can fill this gap. Therefore, a
method for automated production data feedback based on the KDD process was presented in this paper. A major
advantage of the method over the state of the art is that only the mean values of the previously determined relevant
 Berend Denkena et al. / Procedia Manufacturing 28 (2019) 18–23 23
6 Author name / Procedia Manufacturing 00 (2019) 000–000

influencing factor combinations have to be recalculated for new production data. The more complex correlation
analysis can also be repeated if necessary, but without significant changes in the production environment or the
introduction of new products there is no need for action.
The exemplary method execution at a thread manufacturer has shown that the method can be implemented in
practice with little effort. As initially assumed, the MES can provide a sufficient database to make complex and
expensive sensor retrofits for the introduction of adaptive work planning and production control obsolete. In
addition, it was shown that often there are only a few influencing factors on target/actual time deviations, whereby
the complexity of the method is further reduced. For example, the time deviation of processing operations depends
only on the selected workstation. In addition, the work step was identified as an influencing factor for setup
processes. As a result, the data feedback for setup times could be limited to a simple workstation-workstep-matrix.
Moreover, the presented method can be used to support a digitization strategy. For this purpose, additional
measurement categories are included into the analysis and sensors based on the results selected.
The next step is to examine to what extent the use of the method in the planning and control phase has a positive
effect on the achievement of the company's objectives. This investigation is part of an ongoing research project and
will focus on the effect on usual performance parameters, such as throughput time, reject rates and costs.

Acknowledgements

The results presented in this paper were obtained within in the Collaborative Research Center 653 “Gentelligent
Components in Their Lifecycle”. The authors would like to thank the German Research Foundation (DFG) for its
financial support and the Bornemann Gewindetechnik GmbH & Co. KG as well as the Fauser AG for their
cooperation.

References

[1] Denkena, B.; Lorenzen, L.-E.; Schmidt, J: Adaptive process planning. Production Engineering 6:1, pp. 55-67, 2012
[2] Phanden, R. K.; Jain, A.; Verma, R.: Integration of process planning and scheduling: a state-of-the-art review. International Journal of
Computer Integrated Manufacturing 24:6, pp. 517-534, 2011
[3] Schuh, G.; Potente, T.; Hauptvogel, A.: Methodology for the evaluation of forecast reliability of production planning systems. Proceedings
of the 47th CIRP Conference on Manufacturing Systems, 28.-30. April, Windsor, pp. 469-474, 2014
[4] Kreutzfeldt, J.: Planen mit Bearbeitungs-alternativen in der Teilefertigung. Dr.-Ing. Dissertation, Leibniz Universität Hannover, 1994
[5] Denkena, B.; Dittrich, M.-A.; Winter, F.: Adaptive process planing and control, In: Denkena, B.; Mörke, T.: Cyber-Physical and
Gentelligent Systems in Manufacturing and Life Cycle. Academic Press, London, 2017
[6] Schmidt, J.: Integrierte Arbeitsplanung und Fertigungssteuerung auf Basis von Zustandsinformationen, Dr.-Ing. Dissertation, Leibniz
Universität Hannover, 2015
[7] Beierle, C.; Kern-Isberner, G.: Methoden wissensbasierter Systeme. Grundlagen, Algorithmen, Anwendungen. 4. Aufl., Vieweg + Teubner,
Wiesbaden 2008
[8] Fayyad, U.; Piatetsky-Shapiro, G.; Smyth, P.: From Data Mining to Knowledge Discovery in Databases. AI Magazine 17:3, pp. 37-54, 1996
[9] Ester, M.; Sander, J.: Knowledge Dis-covery in Databases. Techniken und Anwendungen. Springer Verlag, Berlin, Heidelberg, New York
2000
[10] Fayyad, U.; Piatetsky-Shapiro, G.; Smyth, P.: The KDD process for extracting usefull knowledge from volume of data. Communications of
the ACM 39:11, pp. 27-34, 1996
[11] Coudhary, A. K.; Harding, J. A.; Tiwari, M. K.: Data mining in manufacturing: a review based on the kind of knowledge. Journal of
Intelligent Manufacturing 20, pp. 501-521, 2009
[12] Monostori, L.; Kádár, B.; Bauern-hansl, T.; Kondoh, S.; Kumara, S.; Rein-hart, G.; Sauer, O.; Schuh, G.; Sihn, W. ; Ueda, K.: Cyber-
physical systems in manufacturing. CIRP Annals - Manufac-turing Technology 65, pp. 621-641, 2016
[13] Reinhart, G; Geiger, F.: Adaptive scheduling by means of product-specific emergence data. Proceedings of the 2011 IEEE International
Conference on Industrial Engineering and Engineering Management, pp. 347-351, 2011
[14] Geiger, F.; Reinhart, G.: Knowledge-based machine scheduling under consideration of uncertainties in master data. Production Engineering
10:2, pp. 197-207, 2016
[15] Cohen, J.: Statistical power Analysis for the behavioral sciences. Lawrence Erlbaum Associates, Hillsdale NJ, 1988
[16] Richardson, J. T. E.: Eta squared and partial eta squared as measures of effect size in educational research. Educational Research Review
6:2, pp. 135-147, 2011