Optimization and investment decisions of electrical motors’ production line using discrete event simulation

(1)

IN

DEGREE PROJECT MECHANICAL ENGINEERING, SECOND CYCLE, 30 CREDITS

STOCKHOLM SWEDEN 2020,

Optimization and investment decisions of electrical motors’

production line using discrete event simulation

ELLEN BURKHARDT

KTH ROYAL INSTITUTE OF TECHNOLOGY

SCHOOL OF INDUSTRIAL ENGINEERING AND MANAGEMENT

(2)

I

Optimization and investment decisions of electrical motors’ production line using discrete event simulation

Ellen Burkhardt

Degree Project in Production Engineering and Management KTH Royal Institute of Technology

Stockholm, Sweden 2020

(3)

II

Abstract

More dynamic markets, shorter product life cycles and comprehensive variant management are challenges that dominate today's market. These maxims apply to the automotive sector, which is currently highly exposed to trade wars, changing mobility patterns and the emergence of new technologies and competitors. To meet these challenges, this thesis presents the creation of a digital twin of an existing production line of electric motors using discrete event simulation. Based on a detailed literature research, a step-by-step establishment of the simulation model of the production line using the software Plant Simulation is presented and argued. Finally, different experiments are carried out with the created model to show how a production line can be examined and optimized by means of simulation using different parameters.

Within the scope of the different experiments regarding the number of workpiece carriers, number of operators as well as buffer sizes, the line was examined concerning the increase of the output. Furthermore, the simulation model was used to make decisions for future investments in additional XXX machines. Four different scenarios were examined and optimized. By examining the different parameters, optimization potentials of XXX% in the first scenario and up to XXX% in the fourth scenario were achieved.

Finally, it was proven that the developed simulation model can be used as a tool for optimizing an existing production line and can generate useful investment information.

Beyond that, the development of the simulation model can be employed to investigate further business questions at hand for the specific production line in question.

Keywords

Industry 4.0, Cyber-physical systems, Digital Twin, Simulation, Simulation process, Electro mobility, Verification and Validation, Plant Simulation

(4)

III

Sammanfattning

Mer dynamiska marknader, kortare produktlivscykler och omfattande varianthantering är utmaningar som dominerar dagens marknad. Dessa maximer gäller bilindustrin, som för närvarande är mycket utsatt för handelskrig, förändrade rörlighetsmönster och framväxten av ny teknik och nya konkurrenter. För att möta dessa utmaningar innebär denna avhandling skapandet av en digital tvilling av en befintlig produktionslinje av elmotorer med diskret händelsesimulering. Baserat på en detaljerad litteraturforskning presenteras och argumenteras en steg-för-steg-etablering av simuleringsmodellen för produktionslinjen med hjälp av programvaran Plant Simulation. Slutligen utförs olika experiment med den skapade modellen för att visa hur en produktionslinje kan undersökas och optimeras med hjälp av simulering med hjälp av olika parametrar.

Inom ramen för de olika experimenten när det gäller antalet arbetsstyckesbärare, antalet operatörer samt buffertstorlekar undersöktes linjen om ökningen av produktionen.

Dessutom användes simuleringsmodellen för att fatta beslut för framtida investeringar i ytterligare hårnålsmaskiner. Fyra olika scenarier undersöktes och optimerades. Genom att undersöka de olika parametrarna uppnåddes optimeringspotentialer på XXX % i det första scenariot och upp till XXX % i det fjärde scenariot.

Slutligen bevisades det att den utvecklade simuleringsmodellen kan användas som ett verktyg för att optimera en befintlig produktionslinje och kan generera användbar investeringsinformation. Utöver detta kan utvecklingen av simuleringsmodellen användas för att undersöka ytterligare affärsfrågor till hands för den specifika produktionslinjen i fråga.

Nyckelord

Industri 4.0, Cyber-fysiska system, Digital tvilling, simulering, simuleringsprocess, elektromobilitet, verifiering och validering, växtsimulering

(5)

IV

Acknowledgements

At this point, I would like to thank all those who have supported and motivated me during the preparation of this master thesis.

First, I would like to thank Filmon Yacob, who supervised and reviewed my master thesis. I would like to thank him for his helpful suggestions and constructive criticism during the preparation of this thesis.

I would also like to thank the Robert Bosch GmbH, which enabled me to write my thesis in the company. Special thanks go to Dr. Marc Knapp, Fabian Renninger and Richard Scheer, who supported me with a lot of patience, interest and helpfulness. I would like to thank them for the numerous interesting debates and ideas that have contributed significantly to the fact that this thesis is available in this form.

Finally, I would like to thank my family and friends, who always supported me not only during the thesis but also during my entire studies. They always had an open ear for me and supported me throughout my whole studies.

(6)

V

Blocking note

The following sections of this master's thesis:

• Chapter 3.1.2

• Figure 16

• Figure 17

• Figure 18

• Figure 19

• Figure 20

• Figure 21

• Figure 22

• Figure 23

• Figure 24

• Figure 25

• Figure 26

• Figure 27

• Figure 28

• Figure 29

• Table 5

• Figure 30

• Chapter 4.5.2 including all subchapters

• Chapter 4.5.3 including all subchapters

• Chapter 4.5.4

• Figure 40

• Chapter 4.5.2

• Appendix B

• Appendix C

• Appendix D

• Appendix E

• Appendix F

are confidential information of the Robert Bosch GmbH. These sections are therefore only accessible to the supervisors and examiner of the Audit Committee for auditing purposes.

Publications and reproductions of the relevant sections are not permitted without the expressed permission of the company.

This blocking notice is valid for 3 years from the date of submission of the thesis to the examination office.

(7)

II

Master’s thesis statement of originality

I hereby confirm that I have written the accompanying thesis by myself, without contributions from any sources other than those cited in the text and acknowledgements.

This applies also to all graphics, drawings, maps and images included in the thesis.

____________________________________________

Ellen Burkhardt

Stockholm, 01.08.2020

(8)

III

Author

Ellen Burkhardt (egbu@kth.se)

School of Production Engineering and Management KTH Royal Institute of Technology, Stockholm, Sweden

Place of the thesis

Robert Bosch GmbH, Schwieberdingen, Germany

Examiner

Daniel Tesfamariam Semere (danielts@kth.se)

KTH Royal Institute of Technology, Stockholm, Sweden

Supervisors

Filmon Yacob (filmona@kth.se)

KTH Royal Institute of Technology, Stockholm, Sweden Dr. Marc Knapp (Marc.Knapp@de.bosch.com)

Robert Bosch GmbH, Schwieberdingen, Germany Fabian Renninger (Fabian.Renninger@de.bosch.com) Robert Bosch GmbH, Schwieberdingen, Germany

(9)

IV

List of abbreviations

AB Agent-based

VDA Association of the German Automotive Industry

CAx Computer-aided x

CPS Cyber-physical systems

DE Discrete event

DS Dynamic Systems

ERP Enterprise resource planning GAWizard Genetic Algorithm Wizard

HV High voltage

Init Initialization

IMG Integrated motor-generators

OP Operation

OEE Overall Equipment Effectiveness PouP Point of use Provider

PwC PricewaterhouseCoopers PLM Product Lifecycle Management PPC Production planning and control

sec Seconds

SD System dynamics

V&V Validation and verification

MES Warehouse management and manufacturing execution systems

(13)

VIII

List of figures

Figure 1: Components of Industry 4.0 ... 6

Figure 2: Concept of a CPS ... 7

Figure 3: Development of the application areas of the digital twin ... 10

Figure 4: Main aspects of the system description ... 12

Figure 5: The steps from a system to a model . ... 13

Figure 6: Abstraction levels of paradigms in Simulation Modeling ... 14

Figure 7: Visualization of the moment of event in a discrete event simulation ... 15

Figure 8: Procedure model for the simulation with Verification and Validation ... 17

Figure 9: Genetic algorithm cycle ... 25

Figure 10: Applications of V&V in a simulation study - Differences between V&V ... 27

Figure 11: System development phases of the simulation ... 30

Figure 12: User interface of Plant Simulation ... 33

Figure 13: Two- and three-dimensional visualization in Plant Simulation ... 34

Figure 14: The integrated motor-generator ... 36

Figure 15: Overview of production process ... 37

Figure 16: Conceptual model of the production line ... 40

Figure 17: Formal model of the production line ... 41

Figure 18: Exchange of the OPXXX ... 42

Figure 19: Source code of the entrance method of OPXXX ... 43

Figure 20: Extensions of XXX, OPXXX, OPXXX ... 44

Figure 21: Sequence change of OPXXX ... 44

Figure 22: Implementation of OPXXX ... 45

Figure 23: Extract of the Setup Matrix OPXXX ... 46

Figure 24: Source code of the exit method of OPXXX to define the destination of worker ... 47

Figure 25: Example of a control method within a robot ... 48

Figure 26: Loading method to control the robots ... 48

Figure 27: Executable model of the simulation in three-dimensional view ... 50

Figure 28: Source code of the initialization of the XXX machines ... 51

Figure 29: Method of XXX_1 ... 53

(14)

IX

Figure 30: Comparing the percentage output of real values and the simulation ... 58

Figure 31: Mean output of simulation model ... 59

Figure 32: Preliminary study of model 1 ... 63

Figure 33: Evaluation of the operator workload ... 64

Figure 34: Preliminary study of the workers within the different scenarios ... 71

Figure 35: Optimization of the workers within the different scenarios ... 72

Figure 36: Growth in output due to increased number of operators ... 73

Figure 37: Increased output due to buffer optimization ... 76

Figure 38: Cycle time evaluation of the XXX machines ... 77

Figure 39: Cycle time evaluation of the XXX machines with XXX workers ... 78

Figure 40: Overview of the achieved optimization ... 80

Figure 41: Overview of investment ... 82 Figure 42: Source code of XXX_2 ... C-11 Figure 43: Source code of XXX_3 ... C-14 Figure 44: Source code of XXX_4 ... C-19 Figure 45: Source code of XXX_4.1 ... C-21 Figure 46: Detailed study of the work piece carriers in the actual situation... E-24 Figure 47: Preliminary study of the work piece carriers in scenario I ... F-25 Figure 48: Detailed study of the work piece carriers in scenario I ... F-25 Figure 49: Preliminary study of the work piece carriers in scenario II ... F-26 Figure 50: Detailed study of the work piece carriers in scenario II ... F-26 Figure 51: Preliminary study of the work piece carriers in scenario III ... F-27 Figure 52: Detailed study of the work piece carriers in scenario III ... F-27 Figure 53: Preliminary study of the work piece carriers in scenario IV - variant I ... F-28 Figure 54: Detailed study of the work piece carriers in scenario IV - variant I ... F-28 Figure 55: Preliminary study of the work piece carriers in scenario IV - variant II ... F-29 Figure 56: Detailed study of the work piece carriers in scenario IV - variant II ... F-29

(15)

X

List of tables

Table 1: Example of a plan of experiments ... 22

Table 2: Values for the certainty z in the calculation of the sample size ... 24

Table 3: Scoring of discrete event simulation tools ... 31

Table 4: Overview of scenarios ... 54

Table 5: Extract of trace test ... 57

Table 6: Plan of experiments ... 61

Table 7: Overview of preliminary studies ... 62

Table 8: Intended and optimized buffer capacities of the actual situation ... 67

Table 9: Range of work piece carriers based on the preliminary studies ... 69

Table 10: Optimum number of work piece carriers for the different scenarios ... 70

Table 11: Cost comparison of amount of worker ... 74

Table 12: Buffer optimization of the target situation ... 75

Table 13: Demand for stators in the coming years per 30 days ... 81 Table 14: Overview of the Plant Simulation functions ... A-1 Table 15: Overview of the individual operation steps... B-4 Table 16: Validation of the simulation model ... D-22

(16)

XI

List of appendices

Appendix A ... A-1 Appendix B ... B-4 Appendix C ... C-6 Appendix D ... D-22 Appendix E ... E-24 Appendix F ... F-25

(17)

1

1. Introduction

Today, companies have more and more possibilities to adapt to the requirements of their customers faster and more agile thanks to the progress in digitalization. In this context, companies have various tools at their disposal for enhanced planning of resources and the consequent cost reduction. This includes among others the use of software for the simulation of production areas and processes as well as the use of digital twins.

However, the potential applications are quite diverse. Companies expect digital twins to optimize production lines or routes of deployment or simply to improve the product application. Although digital twins were still being questioned in 2011 with "Is this science fiction?" (Tuegel, Ingraffea, Eason, & Spottswood, 2011, p. 1), their large-scale business application has already been a reality for several years.

For instance, studies on IT trends have already dealt with the topic of the digital twin in 2017 and ranked it among the "Top Trends 2017” (Klostermeier, Haag, & Benlian, 2020, p. 1;

Panetta, 2016). Since then, the importance of the digital twin in companies has increased continuously but even in 2020, the use of digital twins is still not part of the everyday business. Nevertheless, further forecasts show that the use of digital twins in companies will continue to grow rapidly within the upcoming years (Geissbauer, Schrauf, & Morr, 2019, p. 21).

Among others, the Robert Bosch GmbH is observing those developments as well.

Consequently, the application of digital twins is increasing in conjunction with the use of simulation in production areas. This is due to the newly emerging challenges caused by more dynamic markets, shorter product life cycles, more complex production processes and more extensive variant management. In response, many corporates employ simulation to forecast and quantify those challenges (Robert Bosch GmbH, 2020a).

Therefore, this thesis deals in the following with the implementation of a digital twin using simulation software at the Robert Bosch GmbH.

(18)

2

1.1 Problem definition

This thesis critically assesses the simulation of a production line. Currently, the optimal resources such as the number of workers, amount of work piece carriers and the optimal buffer sizes to produce the electric motors are unknown. Thus, the optimal capacity and setup of the production line needs to be derived. Moreover, the need for a further expansion of the line was not investigated. Here the problem relates to the best possible expansion stage of the line to generate the highest output. In order to optimize and derive the maximum output, this thesis applies and examines the simulation of the production line.

1.2 Research Question

Based on the problem definition the following research question is being examined in the subsequent thesis:

1. How can the investment strategy for the ramp-up of a series production line in electro mobility be defined, based on material flow simulation?

Within this research question, one also observes further sub-questions to support the result of the primarily research question. These questions observe the following:

a. What is the optimal number of work piece carriers to achieve the highest output?

b. How many workers should work within the production line to operate the line sufficiently?

c. What are the optimum buffer sizes within the production line to generate the highest output?

d. Which output can the different investment scenarios generate?

1.3 Objective

The aim is to create a digital twin to investigate how far a simulation tool can support different investment strategies for a series production line. In doing so, one investigates the possible performance of the production line with the help of simulation to guarantee the ramp-up of the series production. This should lead to a reliable data basis and first empirical evidence for a real ramp-up of the simulated production line.

(19)

3 Moreover, the goal is to compare and analyze the costs and benefits of different investment strategies by using simulation in the context of production. For this purpose, the development of an optimized production system of serial production for the stator, which is the subject of the analysis, is pending in different scaling scenarios. The focus is given to the utilization of the production capacity as well as the adherence to the ramp-up of the series production. Therefore, the tool Plant Simulation is used as a simulation software for the evaluation and verification of concepts and methods.

1.4 Delimitation

There is a restriction of investment and production planning for the simulation study. One limitation at hand is the number of available machines and the capital available for further investments. As a further delimitation, the material supply of the line and the removal of the final product from the simulation is excluded.

Furthermore, it needs to be considered that the process sequences and the design of the line's equipment are unchangeable. Moreover, there are further restrictions regarding the estimated values for calculation. Since the line is still in sample production, only empirical values are assumed for calculation purposes.

Another important delimitation is that the line must first run up to provide a certain output.

This means that these losses are considered when starting the line, but longer pauses, such as weekends, are not considered when conducting experiments to avoid these run-up losses.

1.5 Structure of the work

The second chapter of the thesis develops the theoretical foundation for the subsequent empirical analysis. Herein, Industry 4.0 and simulation are being discussed in order to evaluate the suitability of simulation to be employed in production planning. In addition, there is a presentation of the simulation software Plant Simulation. Here the reader acquires knowledge of the mode of operation, in order to follow the sequence of the software during the practical application of simulation.

(20)

4 Subsequently, the third chapter defines the production line, which is the subject and center of this study. Firstly, one introduces the stator and its components as a part. Consequently, details of the processes and the operation of the production line are being analyzed. Thus, this chapter intends to give the reader a basic understanding of the object of investigation.

The fourth chapter describes the implementation of the simulation model. This chapter illustrates the individual development steps of the simulation model including optimization experiments. First, it is about the development of the model according to the simulation process to represent the line as a simulation model. Based on the underlying modelling, the relevant parameters of the optimal production line are being defined and derived by various simulation experiments. Based on the experiments, the research questions are finally answered and recommendations for further investments are given.

Finally, within the fifth chapter the author translates the findings in a business relevant recommendation on how far the application of simulation can support the analysis of the aforementioned business and research questions. In addition, this chapter provides an outlook on further applications of simulation.

(21)

5

2. Theoretical foundations of Industry 4.0 and simulation

As a basis for the scientific treatment of the research problem, the first step is to examine the theoretical foundations. The hereby acquired knowledge provides a methodical XXXwork for the practical solution of the research problem. Consequently, the main topics of the following chapter are Industry 4.0 and Simulation. At first, a first understanding of the term Industry 4.0 as well as important sub-areas is given. The theoretical part concretizes the so-called simulation and explains the most important terms. This discussion is followed by a description of the procedure for a simulation study and the necessary verification of assumptions and validation of the work steps. At the end of the chapter, one presents the simulation software Plant Simulation which is used in the context of this thesis.

2.1 State of the art research

The term "Industry 4.0" clusters the fundamental key trends of recent years, aiming to achieve complete penetration of industry, its products and its services while simultaneously networking the products and services by the means of cyber-physical systems (CPS).

Targeted selection of important information is the key by exploiting resulting data for improving production processes and optimizing manufacturing by using simulation (Bauernhansl, Hompel, & Vogel-Heuser, 2014, p. 249).

2.2 Industry 4.0

For the first time, the term "Industry 4.0" found public attention at the Hannover fair 2011 in Germany. An expert group from industry, science and politics introduced the term to describe the fourth industrial revolution, a revolution driven by the potential of the internet and the connectivity it offers (Kagermann, Wolf-Dieter, & Wahlster, 2011). Since then, companies have used the term Industry 4.0 as a controversial buzzword, although no clear definition exists so far (Siepmann, 2016, p. 22). However, the term Industry 4.0 can be considered as the fourth industrial revolution, a new phase in the organization and management of the entire value chain over the life cycle of products. This cycle relates to increasingly individualized customer requirements and extends from the idea of ordering a

(22)

6 product, through production and delivery, to its recycling. The basis is the availability of all relevant information in real time by networking all the instances involved in value creation and the ability to derive the optimum value-added flow at any given time from the relevant data. By connecting people, objects and systems, dynamic, real-time optimized and self- organized, cross-company value-added networks are created, which can be optimized according to various criteria such as costs, availability and resource consumption (Plattform Industrie 4.0, 2014, p. 1).

Within this XXXwork, Industry 4.0 enables end-to-end digitization and networking of all actors involved in value creation. So-called cyber-physical systems (CPS), recording data by means of sensors and processing them with embedded software in order to influence real processes by means of actors, plays central role. Figure 1 graphically represents the components of Industry 4.0 to build up a basic understanding of the topic.

Step 1:

Cyber-physical system (CPS)

Block 1 Ubiquitous Computing (Ubiquitous information processing through

embedded hardware and software)

Block 2

Internet of things and services (IoTS)

Block 3 Cloud Computing

Step 2:

Cyber-physical production system (CPPS)

Block 1

Machine-to-machine communication (M2M communication)

Block 2

Human-machine interaction (MMI)

New corporate vision Strategy adjustments New business models and processes

Step 3:

Industry 4.0

Figure 1: Components of Industry 4.0. Adapted from Siepmann, 2016, p. 22.

Among the Buzzword Industry 4.0 are therefore many components as shown in Figure 1. In the following, this thesis will refer specifically to the topic of CPS and explain further sub- items to create an understanding for this thesis.

(23)

7 2.2.1 Cyber-physical system (CPS)

The term CPS systems first appeared in the USA in 2006 (Lee, 2006). Rajkumar, Lee, Sha, and Stankovic (2010) define the term in their work "Cyber-physical Systems”, as follows:

“Cyber-physical systems (CPS) are physical and engineered systems whose operations are monitored, coordinated, controlled and integrated by a computing and communication core.” (Rajkumar et al., 2010, p. 731)

Thus, examples of CPS can be found in the integration of calculations with physical processes.

Within these systems, embedded computers and networks observe and control the physical processes. Mostly feedback loops are used, where physical processes influence the calculations and vice versa (Lee, 2008, p. 363). Unlike conventional embedded systems, which are usually self-contained, CPS are networks of interacting elements with physical input and output simulating the structure of a sensor network (Chang, Gao, Lei, Wang, & Wu, 2015, p. 1).

System

Physical world Cyber world

Virtual influences (other CPS, HMI) Environmental Influences

Part status

Machine status

Sensors

Planning and controlling

• current

• self-organized

• versatile

• reactively

• real-time capable ...

Data bases self-

optimizing model

Nominal data

Real data

Process status Info. exchange

Figure 2: Concept of a CPS. Adapted from Imkamp et al., 2016, p. 326.

(24)

8 Figure 2 shows how the CPS connects the real "physical world" and its processes with the information-processing (virtual) objects and processes of the "cyber world". For this reason, models are indispensable for the virtual representation of the physical processes, its components, and their interaction within the CPS. These models are challenged to describe the real processes and components sufficiently enough to provide a suitable representation of the system's functions and for its planning and control. However, the complexity must also be limited to produce a model at reasonable costs and that can be managed by information technology. This results in using sensors to link the cyber and real world (Figure 2) in order to transfer the actual state from the “physical world” as real data to the "cyber world". The aim is to derive process information, in which is stored in databases and the basis for the models to adapt it to real circumstances. This idea forms the foundation for describing complex interactions and thus preventing process deviations or enabling a real-time-capable reaction. The central challenge here is to collect the "right" measurement data at the "right"

place and at the "right" time and to handle the collected data (Imkamp et al., 2016, p. 326).

However, the question arises in which areas CPS could be employed. Herterich, Uebernickel, and Brenner (2015) present a study in the field of industrial services and identified the following seven different applications for CPS in the manufacturing industry:

• Better use of operational performance data for future development

• Optimization of device operation based on historical data

• Controlling and managing devices remotely

• Predictive maintenance and service

• Use of remote diagnosis to replace field service activities by remote maintenance

• Optimization and improvement of existing service processes

• Marketing information and data as services (Herterich et al., 2015, p. 325)

In consequence, it can be concluded that CPS, through its diverse applications, will support the manufacturing industry in the future and help companies participate in the fourth industrial revolution.

(25)

9 2.2.2 Digital Twins

The concept of the "digital twin" plays an essential role in networking the product life cycle and its value chains. The German Informatics Society defines a digital twin as the digital representation of things from the real/physical world, including all geometry, kinematics and logic data. These can be physical as well as non-physical objects such as services with their real existence being of minor importance. (Kuhn, 2017, p. 440).

Despite its definition, the meaning of the digital twin is still ambiguous to a certain extent. In several studies, the term is seen as a physical and functional model providing benefits at every stage of the product life cycle (Boschert & Rosen, 2016, p. 59; Tao et al., 2018, p. 3566).

However, other studies take the more comprehensive, fundamental view of a model being a digital twin as long as its representation is a system or a part of a system (Kuhn, 2017, p. 440). In the context of this thesis, one uses the last-mentioned view in order to distinguish more clearly between the concepts of a digital twin and a CPS.

Moreover, the use of digital twins is and will be increasing in the following years. According to the annual study on IT trends conducted by Capgemini Consulting (2020), digital twins have become significantly more important in recent years. Yet they are still hardly used in day-to-day business. This could change in the coming years, as around a quarter of those surveyed are already involved in planning and implementation (Capgemini, 2020, p. 28). In addition, the Digital Product Development 2025 forecast of the management consultancy PricewaterhouseCoopers (PwC) supports this hypothesis. This forecast predicts a 25%

increase in the use of simulations within the next three years (Geissbauer et al., 2019, p. 21).

Using digital twins in the production environment makes it possible to realize a comprehensive exchange of information. Digital twins allow the virtual planning and simulation of a product's manufacture. All production steps are stored in the digital twin and even before production begins, a digital twin of the production line will be virtually tested and optimized (Kuhn, 2017, p. 440). Thus, one first creates a virtual image, which allows simulating all-important functions digitally. Then this virtual image can be tested in the simulation and put it into virtual operation. In this way the startup time can be drastically reduced by simulating the digital twin (Wegener, 2015, p. 7). In addition, one can implement

(26)

10 the digital representations in simulation models, which ideally realistically represent the influence on the work piece, as it would occur in practice during test runs. As a result, one gradually more carries out tests virtually, which brings the decisive advantage of early fault detection and optimization possibilities as well as time and cost savings. Furthermore, the combination with simulation models makes what-if analyses in virtual space possible, allowing to map risky scenarios without affecting the real processes. Only when one finds an optimal solution, it will be converted in the real production (Kuhn, 2017, p. 444).

In this way, the "digital twin" has become a synonym throughout the time, referring to a variety of simulation tools for machine or plant simulation (Drath, 2018, p. 6). Thus, Figure 3 shows the continuous development of the digital twin in the context of simulation applications.

Figure 3: Development of the application areas of the digital twin. Adapted from Klostermeier et al., 2020, p. 5.

2.3 Simulation

The term "simulation" derives from the Latin "simulare" and one often equates it with the term "imitation" or "deception". In a figurative sense, one uses the term nowadays in relation to knowledge production as the reproduction and imitation of complex, dynamic systems and processes to an experimental model (Lenhard, Küppers, & Shinn, 2006, pp. 3–4).

As the definition shows, simulation refers to the terms system and model. Therefore, this chapter focus further on the two terms.

Individual applications

•1960+

Simulation Tools

•1985+

Simulation- based system design

•2000+

Conecpts of digital twins

•2015+

(27)

11 2.3.1 System

In the middle of the 20th century, Hall and Fagen (1956) already established a definition for systems which is still used nowadays. According to this definition, a system consists of several elements and their properties that interact with each other. Elements are physical objects such as components and machines, as well as abstract objects such as orders. Of great importance for a system are the relationships between the elements. Without them, the elements would merely be a loose collection of objects (Hall & Fagen, 1956, p. 18).

A small example will illustrate this statement: Machines in a production line cannot produce a functional product when being considered individually. Only the linking of the individual machines forming a chain results in a system. The system aspects in Figure 4 describes generally any system:

• Attributes are the basis for the input and output variables as well as the conditions of the system.

• The functions in a system are responsible for linking input and output variables. They describe the transition from the input variable to the output variable.

• Within the system, hierarchy represents subsystems.

• The structure of a system describes the system components and subsystems as well as their relationships to each other (Košturiak & Gregor, 1995, pp. 2–3).

(28)

12

State of the system:

- Processing - Downtime - Waiting ...

Input - Orders - Material

- Energy Output

- Products - Scrap ...

Attributes

Rod Sawing

section

Function

Sawing machine with the function sawing

Function

Hierarchy

Factory

Workshop Cell 2 Cell 1

System boundary Environment

Structure

- Layout - Shift plan ...

- Machines - Workpieces ...

Relations

Elements

Figure 4: Main aspects of the system description. Adapted from Košturiak & Gregor, 1995, p. 2.

Another important component of a system is its system boundary. It defines the transition of the system to the environment and represents a boundary of the system. At the system boundary, an exchange of material, energy and information between the system and its environment can take place via defined interfaces. Depending on the direction of the flow, one speaks of input (into the system) or output (from the system) (VDI 3633, 2014, p. 4).

Hall and Fagen (1956) define the system environment as the set of objects whose changing properties influence of the system or whose properties are influenced by the system. Based on this, the question arises when an element is considered as a part of the system or a part of the environment, since in both cases it interacts with the other elements. The authors conclude that one can never answer this question unambiguously as the sum of all things constitutes a system. What matters is the purpose of the system analysis. If elements and their properties are necessary for the investigation, they must be included in the system (Hall

& Fagen, 1956, p. 20).

(29)

13 2.3.2 Model

A model is a simplified replication of a planned or existing system with all its processes, which represents the real system in a simplified form. In this context, the differences between the model and the real system with respect to its relevant characteristics for the investigation lie within a given tolerance range (VDI 3633, 2014, p. 4). Concluding from this the investigations are usually not carried out directly on the system so that first a model needs to be developed in the steps of model building.

Systemanalysis Model building Implementation Experiments

and analysis Analysis and

abstraction

Data preparation

Feasible model

Figure 5: The steps from a system to a model. Adapted from Gutenschwager, Spieckermann, Rabe, & Wenzel, 2017, p. 19; VDI 3633, 2014, p. 20.

Figure 5 shows simplified the steps of model building - system analysis, model building, implementation, and experiments and analysis:

1. At the beginning, one need to analyze the modelled system. This involves identifying the system elements and their attributes, the existing relationships within the system and the system environment.

2. The analysis results in a detailed system description. Since a 1:1 implementation leads to high effort and a comprehensive model, one carries out an abstraction of the initial model. Thus, the resulting model is limited to the information and properties relevant for the problem definition and kept as accurate as necessary.

3. If the model is not a physical model or - as in the context of this thesis - a computer model, the implementation of the model into the simulation environment follows.

(Rabe, Spieckermann, & Wenzel, 2008, pp. 5–6; VDI 3633, 2014, pp. 22–32).

Since the model building is part of a simulation study, this section is only a rough summary of the process. In section 2.5 one describes modelling in the context of a simulation study in more detail.

(30)

14

2.4 Simulation paradigms

Based on their properties and behavioral characteristics, it is possible to differentiate systems and analogous simulation models according to various criteria. Within this context, simulations are assigned to different paradigms. There are major paradigms in simulation that offer possibilities to facilitate the modeling of material flows. Figure 6 shows these paradigms in simulation: System dynamics (SD), discrete event (DE) and agent-based (AB).

SD and DE are traditional paradigms whereas AB is a relatively new one. Dynamic Systems (DS), on the other hand, is a little aside, as it is used for modeling and designing "physical"

systems (Borshchev & Filippov, 2004, p. 3).

Discrete Event (DE)

• Entities

• Flowcharts and/or transport networks

• Resources

Agent Based (AB)

• Active objects

• Individual behavior rules

• Direct or indirect interaction

• Environment models

System Dynamics (SD)

• Levels (aggregates)

• Stock-and-Flow diagrams

• Feedback loops

Dynamic Systems (DS)

• Physical state variables

• Block diagrams and/or algebraic-differential equations

Discrete Continuous

Low Abstraction More Details Micro Level Operational Level Middle Abstraction Medium Details Meso Level Tactical Level High Abstraction Less Details Macro Level Strategic Level

Figure 6: Abstraction levels of paradigms in Simulation Modeling. Adapted from Borshchev & Filippov, 2004, p. 3.

Figure 6 provides an overview of the different paradigms and classifies the abstractions of the different paradigms along the y-axis. The axis moves from a low abstraction, which finds its application in the operational area, to a high abstraction, which is on a strategic level. If one divides the paradigms according to their abstractions, dynamic systems or "physical"

modelling is at the bottom of the diagram. In contrast, system dynamics, which deals with aggregates, is at the highest level of abstraction. Both paradigms are on the side of

(31)

15 continuous simulation. On the other side of discrete simulations, discrete event modeling and agent-based modeling are integrated. Concluding from the diagram, discrete event modeling is used for low to medium abstraction. However, agent-based modeling is used at all levels of abstraction. In this way, one can use agents to model objects of very different types and sizes and intelligence levels. For example, on the "physical" level, agents can be pedestrians, cars or robots, on the middle level, they can be customers and on the highest level, agents can be competing companies (Borshchev & Filippov, 2004, pp. 3–4).

Consequently, discrete event simulation is used to investigate complex systems such as production areas. Since the context of this thesis refers to a simulation of a production line, one explains the discrete event simulation in more detail.

2.4.1 Discrete event simulation

The discrete event simulation is particularly suitable for the analysis of dynamic systems. It consists of a sequence of discrete points in time at which a certain event occurs. However, the time between two events is not relevant for the system (Waldmann & Helm, 2016, p. 77).

Figure 7 graphically represents the principle of a discrete event simulation.

0 t1 t2 t3 t4 t5 t6 t

ti – moment of event

Figure 7: Visualization of the moment of event in a discrete event simulation. Adapted from Waldmann & Helm, 2016, p. 17.

According to White and Ingalls (2015, pp. 1741-1743) a discrete event simulation consists of the following eight elements:

1. Inputs, Outputs, and State: Inputs are actions that are entered into a system in order to change the state within the system resulting in changes of the output.

2. Entities and Attributes: One realizes Inputs by the arrival of dynamic entities. An entity is a moving object with attributes, which can cause changes within the system.

(32)

16 3. Activities and Events: Activities are processes and logic. Events are conditions that occur at a certain point in time and lead to a change of state of the system. In the system, an entity interacts with activities to create events.

4. Resources: Resources represent everything that has a limited capacity. Common examples of resources are workers, machines, nodes in a communication network and traffic junctions.

5. Global Variables: With global variables assign values in a simulation. These values are available at any time to the whole simulation and can be used to track almost anything of interest

6. Random Number Generator: A random number generator is a software routine that generates an independent random number evenly distributed between zero and one.

7. The Clock and Calendar: The clock is a global variable that represents the value of the current simulation time. The calendar is a list of events that lie in the future.

8. Statistics Collectors: Statistics collectors collect statistics about certain conditions of the simulation or values of global variables (White & Ingalls, 2015, pp. 1742–1746).

2.5 Simulation process

In the literature, there is a multitude of well-known procedure models for simulation studies.

One of the most frequently quoted models is that one from Rabe et al. (2008) in their work

“Verifikation und Validierung für die Simulation in Produktion und Logistik”. This model brought together different approaches and developed them further. Compared to other process models, the simulation process model is characterized by the consistent introduction of phase results and the separate treatment of model and data (Rabe et al., 2008, p. 5). Furthermore, it is recommended in the VDI guideline 3633, as well. Thus, one applies this XXXwork in the context of this thesis in order to explain the individual steps of a simulation study as shown in Figure 8.

(33)

17

Target Description

Task specification

Conceptual model

Raw data

Formal model

Prepared data

Executable model

Simulation results System analysis

Model formalization

Implementation Data collection

Data preparation

Definition of tasks

Experiments and analysis

Verification & validation of data & models

Figure 8: Procedure model for the simulation with Verification and Validation. Adapted from Rabe et al., 2008, p. 5.

(34)

18 For a better understanding is the model divided into phases and phase results. Ellipses represent the phases whereas rounded squared demonstrate the phase results. It is noticeable that the phases "Data acquisition" and "Data preparation" are separated from the phase results "Raw data" and "Prepared data”. Resulting that one can carry out those phases independently of the modelling process regarding their content, time and the persons to be involved. The only condition is that data preparation requires raw data, that data procurement uses the results of the task specification and that prepared data must be available for the use of the executable model (Rabe et al., 2008, p. 6). To conclude the process model, sequential steps represent the individual steps in the model. However, this is not necessarily the case in practical application since iteration loops can occur between the individual steps. For example, if in the further development of the study one realizes that the previously defined assumptions and decisions are not target-oriented or sufficient and therefore one need to make corrections (Gutenschwager et al., 2017, pp. 142–143).

2.5.1 Target description and definition of tasks

Basically most simulation models start with a "problem formulation", "target description" or

"problem description" (Banks, Carson, Nelson, & Nicol, 2010, p. 16; Birta & Arbez, 2019, p. 35; Rabe et al., 2008, pp. 5–6). Usually, this document summarizes and elaborates important information, such as:

• Initial situation including problem definition and purpose of the study

• Project scope including essential elements as well as a rough functionality of the model and essential goals of the simulation

• Constraints such as the timetable and the financial XXXwork (Gutenschwager et al., 2017, p. 147)

In addition, the first possible solutions are already evaluated and discussed during this phase. The aim of the elaboration process is to ensure that the problem description document serves as a useful basis for the execution of a modeling and simulation study and that the problem is clearly understood (Banks et al., 2010, p. 16; Birta & Arbez, 2019, pp. 35–

37).

(35)

19 Based on the target description, one carries then out the task specification. This is the first analysis step within a simulation study and the first phase of the model represents it (Rabe et al., 2008, p. 6). It contains the description of the future system and, if necessary, concretizes and supplements the target description. The most important information is the delimitation of the system and its level of detail. This specification has a decisive influence on the modelling that will take place later and therefore needs to be carried out carefully based on the objective description. Further components of the task specification are the system properties under investigation and the project or experiment plan in order to meet the time schedule (Gutenschwager et al., 2017, pp. 148–149). The task specification then serves as a starting point for the further work steps of the simulation study describing the solvable problem. In addition, this document checks at the same time if the problem solving is feasible with the provided resources and within the planned time and cost XXXwork (Rabe et al., 2008, p. 47).

2.5.2 Model building

One can summarize the concept of model building as the three phases of system analysis, model formalization and implementation from the model in Figure 8.

During the phase of system analysis, one develops the conceptual model. It provides a documentation of the future simulation model with its objectives, inputs, outputs, elements and relationships, assumptions and simplifications. This creates a transition from the description of the task to the solution of the task. It also describes the scope of the model as well as the required level of detail (Rabe et al., 2008, p. 48).

The model formalization draws its conclusions from the concept model. This is a simplified representation of the system through abstraction and reduction. According to Birta and Arbez (2019, p. 38), when creating the concept model, one initially moves to a higher level of abstraction than in the problem description. If necessary, one adds additional detail, embeds formalisms and improves the precision and completeness of the collected information. Moreover, the concept model can be a collection of sub models that deal with different aspects. Overall, it serves as a bridge between problem description and the simulation model (Birta & Arbez, 2019, pp. 38–40).

(36)

20 Implementation is the next phase in the model, resulting in an executable model or simulation model (Rabe et al., 2008, p. 5). It is of high importance for the experimental work and consists of a two-step process. First, the executable or also simulation model is developed from the formal model and then, in the second step, the simulation program is designed. Thus, the simulation model is a representation of the model, which already considers the syntax for further programming. The simulation model is the penultimate stage of the development process, which begins with the identification of the problem that leads to the decision for the formulation of a modeling and simulation project (Birta & Arbez, 2019, pp. 39–40).

However, there are two categories when creating the simulation model. On the one hand, there are the basic implementation questions and on the other hand, there is the question which type of the experiment should be used and how it relates to the realization of the project. During the implementation, one confronts the tasks of data acquisition, initialization or assignment of the observation interval. Normally, the software environment provides programming functions to accomplish the tasks. The second category, which includes the implementation of the project, considers functions for visualization and animation but also for data analysis or optimization. In general, it can be said that the extent to which a particular modeling and simulation project requires services of these categories can vary greatly (Birta & Arbez, 2019, pp. 40–41).

2.5.3 Data analysis

Another component of a simulation study is the data generation. This includes the steps of data collection and data preparation running in parallel. However, since one usually needs much time for the data collection, it is necessary to begin as early as possible with this step (Banks et al., 2010, p. 18).

Hence, the first step of data analysis is the collection of the right data. First, it is necessary to determine which information are required for the simulation study based on the task specification. The term of this analysis is objective information identification. The investigation follows this step according to the data needed to meet the information requirements and according to the sources that can provide it. Once this process is complete,

(37)

21 the actual data collection takes place. For this purpose, one draws up a survey plan. Here the actual data sources are selected (Gutenschwager et al., 2017, pp. 159–161).

During data collection, business experts and IT managers provide the data and decide about the data sources. In this process, the task specification and the concept model determine the type and scope of the data (Rabe et al., 2008, p. 52). Possible data sources are, among others, current production systems, which use numerous information systems to control and monitor production. Frequently used systems are production planning and control (PPC) (including master data of parts) and enterprise resource planning (ERP) systems (including demand forecasts, replenishment lead times) as well as production data acquisition (including order and machine data). Warehouse management and manufacturing execution systems (MES) are also possible data sources for the simulation study. Another source of data is the family of computer-aided x (CAx) systems. These systems provide, for example, factory layout data, process times and work plans. In this way, the collected data form the result of the work step as raw data (Gutenschwager et al., 2017, pp. 161–163).

Once the data are collected, they need to be available for the simulation study. Known as data preparation, this step transfers the data into a form that one can use it in the executable model for the experiments. Typical methods of this step are data compression by forming static distribution functions based on the collected data set, data transformation and data filtering (Rabe et al., 2008, p. 52).

2.5.4 Experiments and analysis

The simulation process is now entering the final operational phase. Here one conducts experiments and analysis, which results in the simulation results (Rabe et al., 2008, p. 5).

When planning the simulation experiments, one needs to take care to define the variation of parameter values in such a way that as few simulations runs as possible achieve the simulation goal (VDI 3633, 2014, p. 35). Consequently, one enters these parameters into a table. Table 1 represents an example of such a plan of experiments.

(38)

22

Table 1: Example of a plan of experiments

Run Parameter Results (produced parts)

x1 x2 x3 Replication 1 Replication 2 Replication 3 … Replication n

1 1 12 B1

2 1 12 B2

3 1 12 B3

4 1 14 B1

5 2 14 B2

6 2 14 B3

7 2 12 B1

8 2 12 B2

9 3 12 B3

10 3 14 B1

11 3 14 B2

12 3 14 B3

Note. Adapted from Gutenschwager et al., 2017, p. 175.

Although one only considers three factors in Table 1, each with two or three characteristics, the experiment already comprises 12 scenarios. This shows the importance of well-thought- out experiment planning. If one considers the parameters and their values too carelessly, the scope of the experiment increases abruptly. In the example presented here, a further parameter with three values would already triple the scenarios to 36. From this, one can conclude that the creation of the plan of experiments is significantly dependent on the parameter definition. In the following, one presents some methods for these determinations.

First, the description of the objective must be considered since it defines the type of the required investigation. One makes a distinction between:

• What-if- and

• How-to-achieve analysis.

In what-if analyses, the task definition determines the investigation of parameter combinations. Thus, the effort of experiment structuring can be limited significantly. In this

(39)

23 way, the task definition clearly outlines what parameters one can change in order to lead directly to the investigated parameter combinations. This results in the respective parameter combinations, which one records in writing in the plan of experiments (Gutenschwager et al., 2017, pp. 176–177).

With the how-to-achieve analyses, there is greater freedom in the scope of the investigation poses, since one does not know the simulation runs. However, this also poses a great challenge in planning. An intelligent selection of the investigated parameter configurations can significantly influence the number of runs. Thus, this leads to the problem in experiment planning that there are, in principle, an infinite number of possible investigations of scenarios. Here, the experience and system knowledge of the planner play a decisive role.

The planner can limit the parameter values to a range in which he expects the required target states. In general, the aim of the study is to find a parameter configuration that achieves pre- defined target values or obtains the best possible value for a target value (Gutenschwager et al., 2017, p. 177).

In addition to this subjective approach, one can also apply a systematic approach. Here, one makes a distinction between methods that determine the parameters statically or dynamically. The fact that one fixes the parameter combinations before the experiment starts, characterizes the static methods. With a dynamic approach, one only determine the investigated scenarios subsequently during the execution of the experiment (Gutenschwager et al., 2017, 177).

2.5.4.1 Calculation of replications

However, since simulation models are subject to a stochastic influence, the results of a one- time simulation run would not be meaningful. To achieve meaningful results, one need to repeat a simulation run several times under the same conditions but with different starting points for random number generation. These repetitions are called replications (Rabe et al., 2008, p. 12). Therefore, a further important step in the preparation of the experiment is the calculation of the necessary number of replications in order to generate valid results. Since a replication stands for an excerpt from the real system and is not exactly predictable due to the stochastic influence, one can regard it as a sample of the system behavior in analogy to

(40)

24 statistics. This makes it possible to use statistical methods to calculate the required number of replications. In this context, one uses the following formula ( 1 ) for calculating the necessary sample size n and therefore one also uses it in the context of this thesis (von der Lippe, 2011, pp. 3–5).

𝑛 ≥ 𝑧²𝜎² 𝑒²

( 1 )

In the formula, the expression z defines the certainty or probability of error. It usually builds on a normal distribution. Therefore, one can assume the z-values as follows:

Table 2: Values for the certainty z in the calculation of the sample size

Certainty Probability of error z-value

90% 10% 1.64

95% 5% 1.96

99% 1% 2.58

Note. Adapted from von der Lippe, 2011, p. 3.

Further, σ represents the standard deviation of the observed target value. The difficulty here is that one does not know the standard deviation in a simulation study. Therefore, one need to choose an estimated or assumed value for σ and, if necessary, adjust it if new information becomes available. The quantity e represents the desired absolute accuracy of the target value. For example, if e = 10 sec in an examination of the throughput time in a production line, then e indicates that a maximum deviation of ± 10 sec from the mean value is desired in this case (von der Lippe, 2011, pp. 3–4). Eley (2012) provides a derivation of the formula shown.

Once one has determined the number of necessary replications and the plan of experiments, one can start carrying out the experiments.

2.5.4.2 Genetic algorithm

Due to the size of the experiment plan as well as the necessary number of replications of an experiment, it often happens in practice that one has to conduct a large number of simulation runs. To avoid this issue in order to reduce the time of the simulation runs, one can use

(41)

25 genetic algorithms (Bangsow, 2015, p. 277). According to Sivanandam and Deepa (2008, p.29) a genetic algorithm is a problem solving tool that uses genetics for problem solving.

Thus, the genetic algorithm deals with the possible solutions of a certain population.

Chromosomes represent the solutions as an abstract representation. By applying reproductive operators to the chromosomes, one should find the best solution through mutations and recombination. In order to find the correct selection, each chromosome has an associated value, which then the genetic algorithm compares with the so-called fitness value. One should choose the fitness value in such a way that its evaluation determines how good the candidate solution is. Depending on the definition, the optimal solution is the one that maximizes or minimizes the function of the fitness value. Once one correctly defines the reproduction and fitness value, one formulates a genetic algorithm according to the same fundamental structure. This starts the generation of the initial population of chromosomes.

Based on this, the algorithm develops a gene pool to define the further search spaces and to find the optimal combination of chromosomes. The genetic algorithm then goes through an iteration process, so that the population is further developed and optimized. After passing through the different iteration processes, the genetic algorithm determines which combination of input values provides the best fitness value (Sivanandam & Deepa, 2008, pp. 29–30). The following circle in Figure 9 simplified embodies the genetic algorithm:

Population (Chromosomes)

Evaluation (Fitness Value)

Selection Genetic Operations

Decoded String

Reproduction Mate

Offspring

New generation

Parents

Calculation/

Manipulation

Figure 9: Genetic algorithm cycle. Adapted from Sivanandam & Deepa, 2008, p. 31.

Optimization and investment decisions of electrical motors’ production line using discrete event simulation

Optimization and investment decisions of electrical motors’

production line using discrete event simulation

ELLEN BURKHARDT

Optimization and investment decisions of electrical motors’ production line using discrete event simulation

Ellen Burkhardt

Degree Project in Production Engineering and Management KTH Royal Institute of Technology

Stockholm, Sweden 2020

Abstract

Keywords

Sammanfattning

Nyckelord

Acknowledgements

Blocking note

Master’s thesis statement of originality

____________________________________________

Author

Place of the thesis

Examiner

Supervisors

Table of contents

List of abbreviations

List of figures

List of tables

List of appendices

1. Introduction

1.1 Problem definition

1.2 Research Question

1.3 Objective

1.4 Delimitation

1.5 Structure of the work

2. Theoretical foundations of Industry 4.0 and simulation

2.1 State of the art research

2.2 Industry 4.0

2.3 Simulation

2.4 Simulation paradigms

2.5 Simulation process