Download as pdf or txt
Download as pdf or txt
You are on page 1of 6

This article has been accepted for publication in a future issue of this journal, but has not been

fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TIA.2017.2732340, IEEE
Transactions on Industry Applications

Cloud-based control strategy: downstream defect


reduction in the production of electric motors

D. Coupek, A. Lechler, A. Verl


Institute for Control Engineering of Machine Tools and Manufacturing Units (ISW),
University of Stuttgart
Stuttgart, Germany
Daniel.Coupek@isw.uni-stuttgart.de

Abstract—Classical computing is shifted gradually into the engines cannot be transferred directly to electric motor
cloud, offering completely new possibilities in information usage, production [11]. Challenges of the electric motor production
computing power and application of learning algorithms. In this and strategies to overcome them are part of the research carried
paper, a cloud-based architecture of control systems is out in the European funded project MuProD [12], while the
investigated showing the benefit for multi-stage production focus is on the permanent magnet synchronous motors
systems. All sequential manufacturing and assembly processes (PMSM) that are widely used in electric drive trains [13,14].
are connected via a cloud-based architecture, which allows using
information from a previous production step in one of the This paper describes how cloud computing is used to
subsequent steps for downstream deviation compensation. This enhance the multi-stage production line of rotors in order to
strategy is applied to the rotor production of electric motors in increase the product quality while decreasing the amount of
the automotive industry, as the current production shows high scrap parts. Data acquisition distributed among the production
defect rates due to the lack of adequate sensor signals and line provides the cloud with raw data for optimization tasks
optimization algorithms. The magnetization process of that require high computational effort. The results are
permanent magnets is executed in saturation, so that the transferred back to local control units in the corresponding
generation of deviations cannot be avoided by process control or process stage where the assembly takes places. Section 2
process optimization. Instead, the variance in the magnetization introduces briefly the production line and the rotor type under
signal must be compensated in a downstream process, here the analysis. Section 3 discusses the cloud-based control
rotor assembly stage. Project results show how such a cloud-
architecture in general while Section 4 focuses on its
based architecture can increase the product quality while
decreasing the amount of scrap parts in a real industrial
application for downstream defect compensation and presents
scenario, consequently saving valuable resources like energy and preliminary results. Finally, conclusions and future research
raw materials. Reduction of deviations is crucial for this activities are discussed.
emerging industrial sector as electric motor production for
vehicles is moving towards mass production in the future. II. PRODUCTION LINE OF ELECTRIC MOTORS
Keywords—Control, Production, Defect, Cloud Computing, A. Multi-stage production system
Electric motor, Optimization
Electric motors consist of the three main components rotor,
stator and housing and they are manufactured in multistage
I. INTRODUCTION production systems consisting of about 50 processes. A very
The transportation sector can improve its energy efficiency simplified model of this multi-stage production system with
by changing the travel behavior and the vehicle fleet, with focus on the rotor production is indicated in Fig. 1.
respect to the amount of cars and the technology used [1]. In
Rotor Line
2015, the transportation and traffic sector caused 16% of the
global emissions [2]. In order to reduce emissions, the trend is M1,1 M2,1 M3,1 M4,1
going towards zero-emission vehicles using electric motors [3].
The main drawbacks of electric vehicles, namely low range and M1,2 M2,2 M3,2 M4,2
M5 M6
high costs, are being overcome by the development of new

battery systems [4,5]. Current studies show the huge potential


of electric motors replacing combustion engines (petrol and M1,N M2,N M3,N M4,N M7 M8 Motor
diesel), starting with medium-sized cars [6,7]. Governmental
regulations and targets for increasing the amount of electric Stator Line Scrap
vehicles accelerate this development [8–10]. The increasing Housing Line
need for electric motors in the automotive sector requires
efficient mass production, considering that optimized methods Fig. 1 Simplified scheme of the multi-stage production system for electric
and machines from current production lines for combustion motors in the automotive industry, with focus on the rotor production.

0093-9994 (c) 2017 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TIA.2017.2732340, IEEE
Transactions on Industry Applications

The rotor of a PMSM is composed of a number of SP 0°


Magnetization
laminated steel stacks, which contain a certain number MP of
embedded permanent magnets. Common values of the rotor
parameters are SP = 5 and MP = 24. The process stages
necessary to produce a rotor of this type are:
270° 90°
 M1: loading of the stacks from suppliers on the pallet.
 M2: magnet assembly on the stacks by pick and place.
 M3: stack magnetization process and inspection. (a)
180°
(b)
1 12
magnet number
24

 M4: heating station (glue drying).


Fig. 2 Space-resolved inspection of the magnetic field of one stack (a) and
 M5: rotor assembly machine for SP stacks. the reduced magnetization vector (b).
 M6: rotor balancing station.
 M7: motor assembly of stator, rotor and housing. III. CONTROL ARCHITECTURE
A classical control architecture for production systems is
 M8: EOL quality inspection of the assembled motor.
divided into several layers. It starts from the lowest field layer
After the motor is assembled, it undergoes a final quality where sensors and actuators are located and goes up over
control at the end of line (EOL) inspection station. Active and machine control and several other layers to the enterprise layer.
passive EOL testing for quality inspection is very common in Each of these layers has its own control and data management
current production lines of electric motors [15]. The main system with different data models and interfaces. The
drawback of EOL is the late and off-line inspection at the final connection between these layers and the use of data in another
stage of the manufacturing chain, where defects of the controller requires a high effort of configuration, if it is
production chain have already been accumulated [16]. possible at all.
Consequently, a defective workpiece is machined wasting time,
money and energy resources for creating a final product that is A. Cloud computing
out of tolerances and has to be recycled or scrapped [17]. Considering existing mechanisms of internet technologies
Therefore, feedforward control methods [18] are considered like cloud computing, it offers many advantages for production
for a downstream compensation of deviations generated at systems [22]. Cloud computing in general could be defined as
previous process stages within MuProD. Further details about an infrastructure that offers computation power for services
the motor composition and the production system can be found with a minimum amount of manual configuration effort [23].
in previous works of the authors [19–21]. The focus of this These technologies are increasingly becoming an integral part
research is to obtain a uniform magnetization profile of the of modern production systems in the digital manufacturing
rotor by improving the assembly process through effective data environment [24], which are generally referred to as cyber-
handling and optimization algorithms. The rotor assembly can physical systems (CPS) [25]. They are defined as the
be influenced by optimizing the stack rotation and the stacking integrations of computation and physical processes. In CPSs,
order. The optimization problem must be solved within the embedded computers and networks monitor and control
cycle time of the production system, which is 60 seconds in the physical processes, usually with feedback loops where physical
case under analysis. The basis of those algorithms is the space- processes affect computations and vice versa. Cloud computing
resolved inspection at stage M3, which is therefore described infrastructures offer following main characteristics [26,27]:
briefly in the following section.  On-demand self-service: this is the provision of
computation power depending on actual needs.
B. Space-resolved inspection
After the magnetization process in stage M3, the stacks are  Broad network access: data and services are
inspected in order to detect deviations from the magnetization accessible over standard mechanisms from any
set point. A hall sensor is located next to the rotating stack device.
yielding a space-resolved (0°-360°) distribution of the  Resource pooling: memory, network bandwidth and
magnetic field B(φ) with φ=0..2Π (Fig. 2, (a)). For the processing is shared.
downstream compensation methods described later, the
complexity of this high-resolution signal is reduced to obtain  Rapid elasticity: Resources are provisioned and
one value per magnet resulting in the discrete signal B(k) with released on demand.
k=1..24 (Fig. 2, (b)). This is done by integrating the signal over  Measured service: the need of resources of a service is
the magnet size. The resulting value B(k) is an indicator for the measured and could be analyzed.
magnetic strength of the corresponding magnet k and allows
the identification of defective magnets, e.g. caused by With regard to security aspects, there are different kinds of
variations of the raw material properties. This magnetization clouds like public, private community and any kind of hybrid
vector has to be computed and stored for every single stack in a clouds. The most relevant service model of cloud computing
central control unit. infrastructures for production systems is Software as a Service

0093-9994 (c) 2017 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TIA.2017.2732340, IEEE
Transactions on Industry Applications

(SaaS) [28]. For SaaS the application is running on the cloud C. Information Sharing Platform
infrastructure and is accessible from every client or other A first implementation of such a cloud-based software
service. The service access is independent of any hardware or application called Information Sharing Platform (ISP) was
operating system. designed and realized within MuProD [29]. The ISP is the
The decision whether a cloud-based architecture increases central element for the collection, storage, and analysis of data
the productivity or not depends on the way information is obtained from a production line from different sensor systems
generated and in which way it has to be available. In addition, and machine controls. Fig. 4 shows the control architecture of
the amount of needed computing power has to be taken into this cloud infrastructure with the corresponding interfaces to
account. Cloud-based architectures are convenient if one of the the machines and the operators. The ISP consists of three main
following criteria is fulfilled: services, the collector, the data storage and the ISP viewer
(graphical user interface, GUI).
 Information during production is generated at
different location or time, The collector obtains measured values from the machine
controls or sensor systems and sends them to the data storage
 Information is needed from different parts and component. Meta files and raw data are stored in the database.
previous process steps, The ISP viewer controls the single components (collector, data
store) and allows to plot the measured data online in real time
 Calculations need high computation power,
or to display the stored data from earlier measurements offline.
 Calculations are acyclic. Moreover, online data analysis and knowledge extraction
methods are controlled and configured in the ISP viewer.
B. Cloud-based assembly service Operations on the data can be performed either by using intern
analysis methods of the ISP or by exporting the data to other
In the present production process for electric motors, three tools and services using offline import and export functions
of the criteria mentioned above are fulfilled. First, the that support several file formats like txt, csv, xml, etc.
measurement takes place in an upstream inspection stage (M3)
while the information is used in the downstream assembly The ISP can be used as fully distributed system on various
process (M5). Second, solving the optimization problem computers. The data is transferred via a local network or the
requires high computation powers and storage capacities. internet. For example, the collector can be installed on the
Third, calculations are acyclic. control frontend of a machine tool for data collection. The data
is stored and processed on a cloud server, which is connected
The information about magnetization of the single to several machine tools or sensor systems and the user can
permanent magnets is generated at an early production process, access the data via the network from each client. This
the magnetic field inspection stage (M3), and used in a later application is the setup for a complete production line. Data
process step. The data acquired by the sensor system is from several machine tools and sensor systems can be stored in
communicated and stored in a cloud database for each the cloud database. The knowledge gained from one machine is
individual stack. There, the data reduction takes place to obtain present for all machines and different users can access the data
a reduced magnetization vector. During combination and at their local clients. The GUI of the ISP is shown exemplarily
assembly of the rotor in stage M5, the calculation of for the analysis of a stack magnetization profile in Fig. 5. The
optimization algorithms is done in the cloud as a service. The upper screen shows the raw data acquired from the hall sensor
resulting stack rotation and assembly order is sent to the local while the lower screen shows the processed and reduced
controller of the assembly station for execution. Fig. 3 shows magnetization vector.
schematically the integration of cloud-based control into the
production line of electric motors for the rotor assembly. The
user operator
production of a number of SP stacks results in parallel
production lines and consequently in parallel data acquisition
streams. viewer
ISP
Cloud based Optimal assembly analysis offline txt
Measurement methods export
control policy storage csv
meta files
Stack 1
offline xml
M1 M2 M3 M4 data base

import
collector

Stack SP M5 M6 mat
M1 M2 M3 M4 M7 M8 Motor software
online
interface online export
import
Stator Line
Scrap
Housing Line machine data machine/
data user
control generator sensor
Fig. 3 Integration of cloud-based control into the multi-stage production
system of electric motors. Fig. 4 Structure of the ISP with software components and interfaces (blue)
to external users or applications (green).

0093-9994 (c) 2017 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TIA.2017.2732340, IEEE
Transactions on Industry Applications

Cloud-based control

Optimization
Database algorithm
magnetic optimal
profiles B(φ) rotation and
stack order rotor
assembly

pos. neg.
deviation deviation uniform magnetic
field of assembled
rotor
stack buffer

Fig. 6 Sequential rotor assembly using cloud services.

If a brute-force algorithm is used to search the complete


Fig. 5 GUI of the ISP with raw magnetization data (above) and reduced solution space for the global optimum, the number of possible
discrete data points (below) for one rotor stack.
rotations πrot and consequently the calculation time grows
The application of this tool for downstream defect deviation exponentially with the amount of stacks needed for one rotor:
within the multi-stage rotor production line is described in the S P 1
 MP  .
following section for two optimization strategies.  rot    (1)
 2 
IV. DOWNSTREAM COMPENSATION Table 1 shows the number of possible combinations of
stack rotations and the averaged computation time for different
Deviations in the magnetic field of single rotor stacks occur numbers of stacks. Table 2 shows the number of permutations
in the previous process stages (upstream) by variability of the πord=SP! and computation time for finding the optimal stacking
magnetization process and the stack properties (stage M3). In order of the corresponding stacks. The total time is the sum of
addition, handling of the magnets in the pick and place process both calculation times. A local control unit, e.g. the numeric
(M2) can cause cracks and other defects that reduce the control of a machine tool, cannot handle this kind of
magnetic field intensity. Downstream compensation means that optimization task for rotors consisting of a high number of
those deviations are compensated by applying an optimal stacks. Instead, more computation power is required, which can
assembly strategy πopt in the downstream assembly stage (M5), be provided by the cloud based control architecture [32].
where SP stacks are assembled to form one rotor. The effects of Parallelization of the optimization problem can be achieved by
downstream compensation methods on the overall system distributing the possible permutations on several cores of the
performance that are induced by additional buffers or cloud. This decreases the computational effort in a linear way.
decoupling stations are discussed in [30,31] for the electric As the complete solution space is analyzed, the global optimum
motors production system described above. In the following, is found yielding the most uniform magnetic field of the
the two approaches of downstream compensation sequential assembled rotor. The variance of the magnetic flux across the
and selective assembly are presented with respect to the cloud- circumference of the rotor is reduced by 65% compared to the
based control architecture described in the previous section. current assembly.

A. Sequential assembly
TABLE I. EXPERIMENTAL ESTIMATION OF THE COMPUTATIONAL TIME
The sequential assembly strategy finds online the optimal OF ROTATION OPTIMIZATION FOR DIFFERENT SP.
combination of the SP stacks in one batch that are needed to
SP MP Permutations [-] Computation time [s]
form one rotor [12]. After the magnetization process, the 3 24 144 0.001
complete batch is inspected, so that SP magnetic profiles B(φ) 4 24 1 728 0.012
are available. The collector of the ISP acquires the 5 24 20 736 0.156
measurements sequentially in real time and sends it to the data 6 24 248 832 1.960
storage service. The data is stored in the cloud database as raw 7 24 2 985 984 25.089
8 24 35 831 808 319.993
data and also as reduced data B(k). Based on those reduced
magnetization profiles the sequential assembly optimization
TABLE II. EXPERIMENTAL ESTIMATION OF THE COMPUTATIONAL TIME
algorithm calculates the best rotation and stack order to gain OF STACKING ORDER OPTIMIZATION FOR DIFFERENT SP.
uniformity and reduce variability of the output field intensity.
The algorithm imposes an angular misalignment α between the SP MP Permutations [-] Computation time [s]
3 24 6 0.0001
stacks with respect to a reference axis and finds an optimal 4 24 24 0.0002
sorting policy by minimizing a cost function. Those 5 24 120 0.0009
calculations are performed as cloud service, while the resulting 6 24 720 0.0057
assembly policy πopt is transferred to the assembly control as 7 24 5 040 0.0423
new set point, a vector of scalar values (Fig. 6). 8 24 40 320 0.3601

0093-9994 (c) 2017 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TIA.2017.2732340, IEEE
Transactions on Industry Applications

B. Selective assembly Cloud-based control


Selective assembly consists in measuring the key quality
feature Classi- optimal
characteristics of each sub-component and sorting the Database extraction fication assembly
components into clusters according to the measurement
outcome (sorting policy) [33–35]. Existing approaches use
magnetic
clustering algorithms to create groups of mating parts. In profiles B(φ)
[36,37] the impact of reducing the dimension of the feature cluster Ci
vector on clustering and classification processes is analyzed C1
rotor
and methods are proposed to reduce the feature dimensions. In assembly
order to improve the product quality, the assembly station is C2
allowed to assemble components only according to assigned pos. neg.
component cluster matching (assembly and matching policy). deviation deviation C3
uniform
By selective assembly, high precision assemblies can be magnetic field
produced from low precision components, at the cost of classification execution of
increasing the complexity of the system management, the of stacks offline policy
work-in-progress and of decreasing the production rate of the
system [38,39]. Fig. 8 Selective rotor assembly using cloud services.

After magnetizing a single stack, its magnetic profiles B(φ)


and B(k) are stored in the cloud where relevant signal features V. CONCLUSION AND OUTLOOK
are extracted. Based on those features, the stacks are classified This paper is based on [44] and discusses the application of
into the clusters Ci. For classification, specific artificial neural a cloud-based control architecture in the production line of
networks, self-organizing maps (SOM), are suitable [40]. SOM rotors for the automotive industry. Compared to numeric
allow to identify signal structures that are not obvious for controls of machine tools the cloud offers higher storage space
human operators and can therefore classify unknown effects or and computational power that is used for improving the
deviations [41]. Similar signals are grouped in similar classes, assembly process. Downstream compensation methods use
yielding a topological map. The SOM are implemented as a data acquired in upstream process stages (magnetization of
cloud service and they are trained with training data (several single stacks) for optimizing the downstream rotor assembly
thousands of stacks) before being applied for classification process. In this paper, sequential and selective assembly and
tasks of real production data. The cloud architecture allows to their linkage to the proposed cloud-based controller
run offline training without any connection to the production architecture are presented as two possible strategies.
system as well as online classification tasks of sensor data. Experimental results within the project MuProD based on
Based on the identified classes, an offline assembly and rotors from industrial fabrication have shown that the
matching policy is derived and stored as rule base in the cloud application of the proposed strategy yields a significant
application. These rules determine which stacks have to be increase of 16,5% in the production rate of conforming engines
assembled in order to obtain a uniform magnetic field of the of the system. At the same time, the variance of the magnetic
final rotor. Several approaches for finding the optimal flux across the circumference of the rotor is reduced by 65%
assembly policy have been investigated, where fuzzy inference compared to the current assembly. This paper describes briefly
systems [42] yield promising results. Preliminary experimental the ISP, which is a first implementation of the presented cloud-
results (Fig. 7) show that using a fuzzy controller for the based control architecture that was realized within MuProD.
assembly of two stacks can significantly reduce the variability The data acquisition and process optimization algorithms were
of the magnetic field compared to random assembly currently implemented as services. The ISP was used in an experimental
applied in industry [21]. The fuzzy system uses a rule base that setup at laboratory scale as distributed system for the rotor
is stored in the cloud application (Fig. 8). Due to high assembly optimization.
computational effort and special libraries [43], those algorithms Future work will focus on the implementation of the
run as cloud service instead as local optimization task on the solution in an industrial production system and it will quantify
control unit. The assembly policy πopt is finally transferred to the potential of improvement experimentally. As the strategies
the assembly control unit for execution of the assembly. and the software are modular and flexible, a generalization and
application to other industrial applications is possible and will
be investigated in future research.

ACKNOWLEDGMENT
The research leading to these results has received funding
from the European Union Seventh Framework Program
(FP7/2007-2011) under grant agreement n° 285075 in the
project MuProD. The authors would like to thank the
Mechanical Engineering Department of Politecnico di Milano
and the Robert Bosch GmbH for the support in this research.
Fig. 7 3D FEM simulation of random rotor (left) and optimized rotor (right).

0093-9994 (c) 2017 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TIA.2017.2732340, IEEE
Transactions on Industry Applications

REFERENCES [22] Monostori, L., Kádár, B., Bauernhansl, T., Kondoh, S., Kumara, S.,
Reinhart, G., Sauer, O., Schuh, G., Sihn, W., Ueda, K., 2016. Cyber-
[1] Leal Filho, W., Rath, K., Mannke, F., Vogt, J., Kotter, R., Borgqvist, M., physical systems in manufacturing. CIRP Annals - Manufacturing
Myklebust, B., van Deventer, P., 2015. Fostering Sustainable Mobility in Technology 65 (2), 621–641.
Europe: The Contributions of the Project “E-Mobility North Sea [23] Verl, A., Lechler, A., Wesner, S., Kirstädter, A., Schlechtendahl, J.,
Region”, in: Leal Filho, W., Kotter, R. (Eds.), E-Mobility in Europe. Schubert, L., Meier, S., 2013. An Approach for a Cloud-based Machine
Springer International Publishing, Cham, pp. 3–25. Tool Control. Procedia CIRP 7, 682–687.
[2] International Energy Agency (IEA), 2015. Co2 emissions from fuel [24] Schlechtendahl, J., Kretschmer, F., Sang, Z., Lechler, A., Xu, X., 2015.
combustion 2015. Organization For Economic, Paris. Extended study of network capability for cloud based control systems.
[3] Greene, D.L., Park, S., Liu, C., 2014. Analyzing the transition to electric Robotics and Computer-Integrated Manufacturing.
drive vehicles in the U.S. Futures 58, 34–52. [25] Lee, J., Bagheri, B., Kao, H.-A., 2015. A Cyber-Physical Systems
[4] Catenacci, M., Verdolini, E., Bosetti, V., Fiorese, G., 2013. Going architecture for Industry 4.0-based manufacturing systems.
electric: Expert survey on the future of battery technologies for electric Manufacturing Letters 3, 18–23.
vehicles. Energy Policy 61, 403–413. [26] Foster, I., Zhao, Y., Raicu, I., Lu, S. Cloud Computing and Grid
[5] Kurfer, J., Westermeier, M., Tammer, C., Reinhart, G., 2012. Production Computing 360-Degree Compared, in: 2008 Grid Computing
of large-area lithium-ion cells – Preconditioning, cell stacking and Environments Workshop, Austin, TX, USA, pp. 1–10.
quality assurance. CIRP Annals - Manufacturing Technology 61 (1), 1– [27] Mell, P.M., Grance, T., 2011. The NIST definition of cloud computing.
4. National Institute of Standards and Technology, Gaithersburg, MD, 7 pp.
[6] Bradley, T.H., Frank, A.A., 2009. Design, demonstrations and [28] Huang, K.-C., Shen, B.-J., 2015. Service deployment strategies for
sustainability impact assessments for plug-in hybrid electric vehicles. efficient execution of composite SaaS applications on cloud platform.
Renewable and Sustainable Energy Reviews 13 (1), 115–128. Journal of Systems and Software 107, 127–141.
[7] Plötz, P., Schneider, U., Globisch, J., Dütschke, E., 2014. Who will buy [29] Coupek, D., Friedrich, J., Verl, A., Lechler, A., 2014. Modular
electric vehicles?: Identifying early adopters in Germany. Transportation Information Sharing Platform for Real-Time Data Processing and
Research Part A: Policy and Practice 67, 96–109. Knowledge Extraction in Modern Production Systems. Proceedings
[8] European Commission (EC), 2011. Interim Assessment of the Research (CD) of the “6th International Scientific Conference.
PPPs in the European Economic Recovery Plan, Energy Efficient [30] Colledani, M., Coupek, D., Verl, A., Aichele, J., Yemane, A., 2014.
Building, Factories of the Future, European Green Cars Initiative. Design and Evaluation of In-line Product Repair Strategies for Defect
[9] IEA - International Energy Agency, 2016. Global EV Outlook 2016: Reduction in the Production of Electric Drives. Procedia CIRP 21, 159–
Beyond one million electric cars, Paris, 51 pp. 164.
[10] International Council on Clean Transportation (ICCT), 2012. [31] Colledani, M., Yemane, A., Coupek, D., Lechler, A., 2015. Quality-
International Perspective on Electric Vehicles. oriented Design of Rotor Assembly Strategies for Electric Drive
[11] Kampker, A., Deutskens, C., Nee, C., Wowreczko, D., 2012. Neue Production Systems. Procedia CIRP 36, 141–146.
Reserven in der Produktion: Herausforderungen und Lösungsansätze für [32] Keinert, M., Verl, A., 2012. System Platform Requirements for High
die Fertigung von Elektromotoren in Fahrzeugantriebssträngen. WZL Performance CNCs, in: Nylund, H. (Ed.), Proceedings of the FAIM
der RWTH Aachen. http://www.industrieanzeiger.de/automotive-und- 2012. 22nd International Conference on Flexible Automation and
elektromobilitaet/-/article/32571342/37441702. Intelligent Manufacturing, June 10th-13th, 2012, Helsinki, Finland.
[12] Coupek, D., Verl, A., 2014. Ausschussreduzierung und Tampere University Of Technology, Finland, pp. 837–843.
Defektkompensation in mehrstufigen Produktionssystemen, in: Brecher, [33] Colledani, M., Ebrahimi, D., 2013. Performance Evaluation of Selective
C. (Ed.), Effiziente Produktion, Als Ms. gedr ed. VDI-Verl., Düsseldorf, and Adaptive Assembly Systems. SMMSO 2013.
pp. 82–91. [34] Pugh, G.A., 1992. Selective Assembly with Components of Dissimilar
[13] Butov, A., Herrmann, F., Merkert, M., Nägele, F., 2014. Modulares Variance. Computers and Industrial Engineering (23), 487–491.
Fertigungskonzept für E-Motoren. ATZextra (11/2014). [35] Tan, M.H., Wu, C.F.J., 2012. Generalized selective assembly. IIE
[14] Dorrell, D.G., Knight, A.M., Popescu, M., Evans, L., Staton, D.A. Transactions 44 (1), 27–42.
Comparison of different motor design drives for hybrid electric vehicles, [36] Liu, H., Motoda, H., 1998. Feature Extraction, Construction and
in: 2010 IEEE Energy Conversion Congress and Exposition (ECCE), Selection: A Data Mining Perspective. Springer US, Boston, MA.
Atlanta, GA, USA, pp. 3352–3359. [37] van der Heijden, F., 2004. Classification, parameter estimation, and state
[15] Butov, A., Verl, A. Comparison of end of line tests for serial production estimation: An engineering approach using MATLAB. Wiley,
of electric motors in hybrid truck applications, in: 2014 4th International Chichester, West Sussex, Eng., Hoboken, NJ.
Electric Drives Production Conference (EDPC), Nuremberg, Germany, [38] Colledani, M., Ebrahimi, D., 2012. Optimal Process Shift Design in
pp. 1–4. Selective and Adaptive Production Systems. Procedia CIRP 3, 79–84.
[16] Klippel, W., 2011. End-Of-Line Testing, in: Grzechca, W. (Ed.), [39] Iyama, T., Mizuno, M., McKay, K.N., Yoshihara, N., Nishikawa, N.,
Assembly Line - Theory and Practice. InTech, pp. 181–206. 2013. Optimal strategies for corrective assembly approach applied to a
[17] Colledani, M., Copani, G., Tolio, T., 2014. De-manufacturing Systems. high-quality relay production system. Computers in Industry 64 (5),
Variety Management in Manufacturing Proceedings of the 47th CIRP 556–564.
Conference on Manufacturing Systems 17 (0), 14–19. [40] Kohonen, T., 2001. Self-Organizing Maps, Third edition ed. Springer
[18] Abellan-Nebot, J.V., Liu, J., Romero Subirón, F., 2012. Quality Berlin Heidelberg, Berlin, Heidelberg.
prediction and compensation in multi-station machining processes using [41] Aroui, T., Koubaa, Y., Toumi, A., 2009. Clustering of the Self-
sensor-based fixtures. Robotics and Computer-Integrated Manufacturing Organizing Map based Approach in Induction Machine Rotor Faults
28 (2), 208–219. Diagnostics. Leonardo Journal of Sciences 2009 (15), 1–14.
[19] Coupek, D., Verl, A., Aichele, J., Colledani, M., 2013. Proactive quality [42] Mamdani, 1977. Application of Fuzzy Logic to Approximate Reasoning
control system for defect reduction in the production of electric drives, Using Linguistic Synthesis. IEEE Trans. Comput. C-26 (12), 1182–1191.
in: 2013 3rd International Electric Drives Production Conference [43] Sivanandam, S.N., Sumathi, S., Deepa, S.N., 2007. Introduction to fuzzy
(EDPC), Nuremberg, Germany, pp. 1–6. logic using MATLAB. Springer, Berlin, New York.
[20] Coupek, D., Verl, A., Lechler, A., Aichele, J., Junker, S. Defect [44] Coupek, D., Lechler, A., Verl, A., 2016. Cloud-based control for
reduction in the production of electric drives by downstream downstream defect reduction in the production of electric motors,
compensation and space-resolved inspection, in: 2014 4th International in: Proceedings. 2016 International Conference on Electrical Systems for
Electric Drives Production Conference (EDPC), Nuremberg, Germany, Aircraft, Railway, Ship Propulsion and Road Vehicles & International
pp. 1–8. Transportation Electrification Conference (ESARS-ITEC), Toulouse,
[21] Coupek, D., Gülec, A., Lechler, A., Verl, A., 2015. Selective rotor France. 02.11.2016 - 04.11.2016. IEEE, pp. 1–6.
Assembly Using Fuzzy Logic in the Production of Electric Drives.
Procedia CIRP 33, 550–555.

0093-9994 (c) 2017 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.

You might also like