Simulation-based Optimization of Energy Efficiency in Production (Forschung zur Digitalisierung der Wirtschaft | Advanced Studies in Business Digitization) 365832970X, 9783658329709

Citation preview

Anna Carina Römer

Simulation-based Optimization of Energy Efficiency in Production

Forschung zur Digitalisierung der Wirtschaft | Advanced Studies in Business Digitization Series Editors Volker Nissen, FG WID, Technische Universität Ilmenau, Ilmenau, Thüringen, Germany Udo Bankhofer, FG Quantitative Methoden der, TU Ilmenau, Ilmenau, Thüringen, Germany Dirk Stelzer, Fakultät für Wirtschaftswissenschaf, Technische Universität Illmenau, Illmenau, Thüringen, Germany Steffen Straßburger, Fraunhofer Institute for Factory Operation and Automation, Ilmenau University of Technology, School of Economics Science, Ilmenau, Thüringen, Germany

Digitalisierung verändert Unternehmen und Behörden, unser Privatleben und unsere Gesellschaft. Organisationen, die wettbewerbsfähig bleiben wollen, müssen die dadurch notwendigen Veränderungen aktiv mitgestalten. Die Rolle der IT muss sich in diesem Zusammenhang vom Kostenfaktor zum Werttreiber wandeln. Weiterentwicklungen der IT-Kernkompetenzen sind nötig. In dieser Reihe werden wissenschaftlich fundierte und praktisch relevante Erkenntnisse zur Digitalisierung veröffentlicht. Praktiker finden darin vielfältige Anregungen für die Gestaltung und Verbesserung von Strukturen, Prozessen und IT-Systemen. Wissenschaftler erhalten Impulse für die Beschreibung, Erklärung, Prognose und Gestaltung der Digitalisierung. Digitalization is changing businesses and governments, our private lives and our societies. Organizations that want to remain competitive must actively shape the changes that are necessary as a result. In this context, the role of IT must change from a cost factor to a value driver. Further developments of the IT core competencies are necessary. This series publishes scientifically sound and practically relevant findings on digitization. Practitioners will find many suggestions for the design and improvement of structures, processes and IT systems. Scientists receive impulses for the description, explanation, prediction and design of digitisation.

More information about this series at http://www.springer.com/series/16416

Anna Carina Römer

Simulation-based Optimization of Energy Efficiency in Production

Anna Carina Römer Weinstadt, Germany

Zugleich Dissertation, Fakultät für Wirtschaftswissenschaften und Medien, Technische Universität Ilmenau Erstgutachten: Prof. Dr. Steffen Straßburger, Technische Universität Ilmenau Zweitgutachten: Prof. Dr. Rainer Souren, Technische Universität Ilmenau Drittgutachten: Prof. Dr. Thomas Schulze, Universität Magdeburg Tag der Abgabe: 22. Januar 2020 Tag der wissenschaftlichen Aussprache: 15. Juli 2020

ISSN 2662-4788 ISSN 2662-4796 (electronic) Forschung zur Digitalisierung der Wirtschaft | Advanced Studies in Business Digitization ISBN 978-3-658-32970-9 ISBN 978-3-658-32971-6 (eBook) https://doi.org/10.1007/978-3-658-32971-6 © The Editor(s) (if applicable) and The Author(s), under exclusive license to Springer Fachmedien Wiesbaden GmbH, part of Springer Nature 2021 This work is subject to copyright. All rights are solely and exclusively licensed by the Publisher, whether the whole or part of the material is concerned, specifically the rights of translation, reprinting, reuse of illustrations, recitation, broadcasting, reproduction on microfilms or in any other physical way, and transmission or information storage and retrieval, electronic adaptation, computer software, or by similar or dissimilar methodology now known or hereafter developed. The use of general descriptive names, registered names, trademarks, service marks, etc. in this publication does not imply, even in the absence of a specific statement, that such names are exempt from the relevant protective laws and regulations and therefore free for general use. The publisher, the authors and the editors are safe to assume that the advice and information in this book are believed to be true and accurate at the date of publication. Neither the publisher nor the authors or the editors give a warranty, expressed or implied, with respect to the material contained herein or for any errors or omissions that may have been made. The publisher remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. Responsible Editor: Anna Pietras This Springer Gabler imprint is published by the registered company Springer Fachmedien Wiesbaden GmbH part of Springer Nature. The registered company address is: Abraham-Lincoln-Str. 46, 65189 Wiesbaden, Germany

Preface

This book was written in the context of my work as an external PhD student at the Department Information Technology for Production and Logistics at the Technical University of Ilmenau. I would like to thank my doctoral supervisor, Professor Steffen Straßburger, the head of the department, for supporting my academic research, for the stimulating discussions, and critical examination of my work. My special thanks go to the Johannes Hübner Stiftung Giessen for the financial support of the doctorate, as well as for the constructive support and the expert advice. I would also like to thank the Thüringer Graduiertenförderung for the doctoral scholarship in the start-up phase of my research. I would like to mention the very good cooperation with the Robert Bosch GmbH. In this context, a special thank you goes to Mr. Stefan Regert, Mr. Christian Kau, and Mr. Fabian Bucksch for providing the practical data, for the interesting discussions, and the numerous helpful proposals. I am grateful to my family, who has always supported me incredibly. I would like to thank my parents-in-law, Karin and Matthias Römer, who always and immediately helped when there was need. From the bottom of my heart I would like to thank my parents—Regine and Wolfgang Gollub—who have always accompanied me with love and patience during my studies, in my professional life and in the creation of this work. They have given me constant and unconditional support and encouraged me in my decisions over all the years. My last and most heartfelt thanks go to my husband Fabian and my children Max and Jonathan, who kept me free of obligation and supported me in an invaluable way whenever it was necessary. I am grateful for their understanding

v

vi

Preface

and patience, especially during the last very intensive months when finalizing this work. January 2020

Anna Carina Römer

Zusammenfassung

Die vorliegende Arbeit beschäftigt sich mit der Integration von Energieaspekten in die Produktionssimulation. Ziel der Arbeit ist es, die Energieverbräuche von Produktionsanlagen realitätsnah in einem Simulationsmodell abzubilden, um diese anschließend für die simulationsbasierte Optimierung der Energieeffizienz nutzen zu können und damit eine umfassende Prozessqualität im Hinblick auf den optimalen Ressourceneinsatz des Faktors Energie im Produktionsprozess sicherzustellen. Hierzu wird zunächst ein hybrider Simulationsansatz entwickelt, der verschiedene Simulationsparadigmen in einem Modell kombiniert. Die Hybridisierung von Simulationsmodellen bietet dem Modellersteller eine große Flexibilität bei der Erfassung von Problemen, die sich gleichzeitig auf diskrete (Materialfluss) und kontinuierliche (Energiefluss) Strukturen beziehen. In einem zweiten Schritt wird das Simulationsmodell für Optimierungsexperimente genutzt. Die Grundidee hinter diesem Ansatz ist es, durch mehrere Iterationen, die verschiedene Systemkonfigurationen simulieren, eine optimale Lösung für die zu variierenden Optimierungsparameter zu finden. Die Simulation wird durch die Optimierung gestartet, liefert die Ergebnisdaten und bildet die Grundlage für eine Beurteilung des dynamischen Verhaltens des abgebildeten Produktionssystems. Auf diese Weise lassen sich optimale Parameterkonfigurationen im Hinblick auf die gestellte Zielfunktion unter Nutzung des Simulationsmodells ermitteln. Die in dieser Arbeit entwickelte Methodik gliedert sich damit in zwei Module, ein Simulations- und ein Optimierungsmodul. Das Simulationsmodul ist in drei Komponenten unterteilt, eine Materialfluss-, eine Maschinen- und eine Energiekomponente. Damit lassen sich die Produktionsprozesse auf allen notwendigen Ebenen in einem Modell abbilden. In Anlehnung an das in der Industrie übliche Vorgehen bei der Erstellung eines Produktionssimulationsmodels wird eine

vii

viii

Zusammenfassung

Materialflusskomponente definiert, in der die Prozesse der Produktion diskret und prozessorientiert dargestellt werden. Das Materialfluss-Modell umfasst alle elektrischen Verbraucher, die unmittelbar an den Produktionsprozessen beteiligt sind. Um die Abläufe der Arbeitsschritte der einzelnen Maschinen exakt abzubilden, wird eine agentenbasierte Maschinenlogik definiert, die unter Nutzung von Zustandsdiagrammen die Prozesse darstellen kann. Dadurch lassen sich Maschinenzustände, Zustandsänderungen sowie Zustandsdauern unterschiedlich komplexer Maschinentypen realitätsnah modellieren. Die Verbindung zum Energieverbrauchsverhalten wird über die Energiekomponente des Simulationsmoduls hergestellt. Die Energiekomponente beinhaltet die Modellierung kontinuierlicher Energielastprofile, die den einzelnen Maschinenzuständen zugeordnet werden und bei den Simulationsläufen entsprechend des Maschinenschaltverhaltens zur Ermittlung des Gesamtenergieverbrauchs der Produktionsprozesse genutzt werden. Um die energetischen Aspekte in der Produktion nicht nur zu modellieren, sondern auch für Optimierungsszenarien zu nutzen, werden lexikographisch geordnete Zielfunktionen abgeleitet, die im Rahmen von simulationsbasierten Optimierungsexperimenten ideale Parameterkonfigurationen für den energieeffizienten Betrieb der Produktionslinien ermitteln. Der Schwerpunkt der Optimierung liegt dabei auf der Reduzierung des Gesamtenergieverbrauchs durch die Vermeidung nicht-wertschöpfender Maschinenzustände. In diesen Phasen verbrauchen die Produktionsanlagen unnötig Energie, werden aber nicht aktiv zur Wertschöpfung eingesetzt. Die Gesamtverbrauchsoptimierung zeigt auf, dass Unternehmen dieser Ressourcenverschwendung durch ein effizientes Schalten der Anlagen entgegenwirken können, ohne große finanzielle Investitionen in neue Technologien tätigen zu müssen. Neben der Optimierung des Gesamtenergiebedarfs beinhaltet die Methodik die Möglichkeit, im Rahmen einer Lastspitzen-Optimierung die Maschinenstarts innerhalb eines definierten Zeitraumes so zu verändern, dass auftretende Spitzenlasten reduziert werden. Die entwickelte Methodik wird zunächst in einer fiktiven Fallstudie implementiert und verfeinert, bevor sie am Beispiel eines Automobilzulieferers in der industriellen Praxis Anwendung findet. Im Rahmen dieser Praxisversuche werden verschieden hoch aufgelöste Energiedaten getestet. Die praktische Anwendung der Methodik zeigt, dass es möglich ist, ein hybrides Simulationsmodel zur Darstellung des Energieverbrauchsverhaltens in der Produktion auf Basis historischer Verbrauchsdaten aufzubauen und in Kombination mit Prognosezahlen auch die zukünftigen Energieverbräuche mit den anstehenden Spitzenlasten und nicht wertschöpfenden Produktionsphasen sehr genau abzubilden. Aus den durchgeführten

Zusammenfassung

ix

Optimierungsversuchen ergeben sich damit Handlungsvorschläge zur energieeffizienten Steuerung von Maschinen, die in Produktionssituationen ähnlich des fiktiven Beispiels zu Energieverbrauchsreduzierungen von 10 % und in stark verketteten Produktionslinien zu Einsparungen von etwa 6 % führen.

Summary

This book presents a methodology for the integration of energy aspects into production simulation. The aim of this work is to realistically represent the energy consumption of production plants in a simulation model in order to be able to use it for the simulation-based optimization of energy efficiency and thus to ensure a comprehensive process quality with regard to the optimal use of the factor energy in the production process. For this purpose, a hybrid simulation approach is developed, which combines different simulation paradigms in one single model. The hybridization of simulation models offers the model creator great flexibility in the detection of problems that are simultaneously related to discrete (material flow) and continuous (energy flow) structures. In a second step, the simulation model is used for optimization experiments. The basic idea behind this approach is to find an optimal solution for the optimization parameters being varied through several iterations to simulate different system configurations. The simulation is started by the optimization, delivers the result data and forms the basis for an evaluation of the dynamic behavior of the mapped production system. In this way, optimal parameter configurations can be determined with regard to the set target function using the simulation model. The methodology developed in this work is thus divided into two modules, a simulation and an optimization module. The simulation module is divided into three components, a material flow, a machine and an energy component. This allows the production processes to be depicted on all necessary levels in one simulation model. Following the standard procedure in industry for the creation of a production simulation model, a material flow component is defined in which the production processes are represented discretely and process-oriented. The material flow model includes all electrical consumers that are directly involved in the

xi

xii

Summary

production processes. To map the work steps of the individual machines as accurate as possible, an agent-based machine logic is defined, which can represent the processes using status diagrams. This allows a realistic modeling of machine states, state changes and state durations of machine types of varying complexity. The connection to energy consumption behavior is made via the energy component of the simulation module. The energy component includes the modeling of continuous energy load profiles, which are assigned to the individual machine states and are used in the simulation runs according to the machine switching behavior to determine the total energy consumption of the production processes. In order not only to model the energy aspects in production but also to use them for optimization scenarios, lexicographically ordered objective functions are derived, which determine ideal parameter configurations for the energy-efficient operation of the production lines in simulation-based optimization experiments. The focus of the optimization is on the reduction of the total energy consumption by avoiding non-value-adding machine states. In these phases, the production lines consume energy unnecessarily, but are not actively used to create value. The overall consumption optimization shows that companies can counteract this waste of resources by efficiently switching the machines without having to make large financial investments in new technologies. In addition to the optimization of the total energy demand, the methodology includes the possibility to change the machine starts within a defined period of time in order to reduce peak loads. The developed methodology is first implemented and refined in a fictitious case study before it is applied to the example of an automotive supplier in industrial practice. Within the scope of these practical tests, energy data with different resolutions is tested. The practical application of the methodology shows that it is possible to build a hybrid simulation model for the representation of energy consumption behavior in production on the basis of historical consumption data and, in combination with forecast figures, to very accurately represent future energy consumption with upcoming peak loads and non-value-adding production phases. The conducted optimization experiments thus result in proposals for action for the energy-efficient control of machines, which in production situations similar to the fictitious example lead to energy consumption reductions of 10% and in strongly interlinked production lines to savings of about 6%.

Abbreviations

ABS ATSA BCVTB CNC CSS DES DESS DEVS DS DTSS EIE EnPI ERP GA GEA GHG HSM IP IS IT FSM KPI LP LPC MES MILP

Agent-based Simulation Adaptive Thermo-Statistical Simulated Annealing Building Controls Virtual Test Bed Computerized Numerical Control Continuous System Simulation Discrete Event Simulation Differential Equation Specified Systems Discrete Event Specified Systems Dynamic Systems Discrete Time Specified Systems Energy-Intensive Enterprises Energy Performance Indicator Enterprise Resource Planning Genetic Algorithm Global Energy Assessment Greenhouse Gas Hybrid Systems Modeling Interaction Points Information Systems Information Technology Finite State Machine Key Performance Indicator Linear Programming Load Profile Clustering Manufacturing Execution System Mixed Integer Linear Programming Model

xiii

xiv

MS M&S MTTF MTTR NLP ODE OEE OM OR OTC PDE PLC PSO QP RNN ROI SA SCM SD

Abbreviations

Management Science Modeling and Simulation Meantime to Failure Meantime to Repair Nonlinear Programming Ordinary Differential Equation Overall Equipment Effectiveness Operations Management Operations Research Over the Counter Partial Differential Equation Programmable Logic Controller Particle Swarm Optimization Quadratic Programming Recurrent Neural Network Return of Investment Simulated Annealing Supply Chain Management System Dynamics

Symbols

Symbol Unit

Description

E

Joule, kWh energy

E state

kWh

ξ

energy required to reach an operating state (standby, idle, run time, operating) randomness of a system

f

objective function

F

N, kg*m/s2 force

g

m/s2

gravity

h

m

height

I

A

electric current

k

kJ/cm3

machine specific constant

m

kg

mass

m iop

kW

energy consumption of machine operation (standby, idle, warmup, …) of machine i

n

production volume

ηi

degree of efficiency

ηsystem

overall efficiency of a system

Θ

feasible region or constraint set

P

kW

power

ρ

bar

pressure

ρ am

bar

ambient state pressure

Rn s

geometrical space of decision variables mm

distance

xv

xvi

Symbols

Symbol Unit

Description

t

sec

time

t warmup

sec

warmup time

T

°C, K

temperature

T am

°C

ambient state temperature

U

V

voltage ms−1

ν

m/s,

v˙

cm3 /sec

material processing rate

W

J, Ws

work

speed

x

decision variable

X

design space of an optimization problem

Contents

1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1.1 Initial Situation and Problem Definition . . . . . . . . . . . . . . . . . . . . . . 1.2 Objectives and Work Structure . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1.3 Classification in the Fields of Information Systems and Research Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

1 1 3 6

2 Simulation-Based Optimization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.1 Modeling and Simulation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.1.1 Continuous and Discrete Simulation . . . . . . . . . . . . . . . . . . 2.1.2 Hybrid Modeling and Simulation . . . . . . . . . . . . . . . . . . . . . 2.2 Optimization Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.2.1 Terminology in Optimization . . . . . . . . . . . . . . . . . . . . . . . . . 2.2.2 Exact and Heuristic Optimization Methods . . . . . . . . . . . . 2.3 Combination of Simulation and Optimization Methods . . . . . . . . . 2.3.1 Coupling and Interaction Modes of Simulation and Optimization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.3.2 Objectives and Challenges of Simulation-based Optimization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.3.3 Optimization Packages Interfaced With Simulation Software . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

9 10 14 19 24 25 26 28

3 Energy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.1 Development of the Global Energy Market . . . . . . . . . . . . . . . . . . . 3.2 The Concept of Energy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.2.1 Work, Energy and Power . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.2.2 Energy Efficiency . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

39 40 41 41 49

30 32 35

xvii

xviii

Contents

3.2.3 Energy Consumption of Manufacturing Processes and Systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.3 Energy Costs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.4 Energy Saving Potentials and Energy Efficiency in Manufacturing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.4.1 Energy Saving Potentials in Manufacturing . . . . . . . . . . . . 3.4.2 Energy Performance Indicators . . . . . . . . . . . . . . . . . . . . . . . 4 State of the Art . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4.1 Selection and Evaluation of Relevant Research Approaches . . . . 4.1.1 General Limitations for the Selection of Research Approaches . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4.1.2 Studies on the Optimization of the Energy Demand in Production Systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4.1.3 Studies on Simulation-Based Modeling of the Energy Demand in Production Systems . . . . . . . . . . 4.1.4 Studies on Simulation-Based optimization of Energy Efficiency in Production . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4.2 Comparison of Results and Discussion . . . . . . . . . . . . . . . . . . . . . . . 4.3 Derivation of Research Demand . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5 Development of a Simulation-based Methodology for Energy Efficiency Optimization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5.1 Concept Objective and Requirements . . . . . . . . . . . . . . . . . . . . . . . . 5.2 Conceptual Framework . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5.3 Description of the M&S Approach . . . . . . . . . . . . . . . . . . . . . . . . . . 5.3.1 Production Flow Component . . . . . . . . . . . . . . . . . . . . . . . . . 5.3.2 Machine Component . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5.3.3 Energy Component . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5.3.4 Interaction Point Definition . . . . . . . . . . . . . . . . . . . . . . . . . . 5.4 Description of the Optimization Approach . . . . . . . . . . . . . . . . . . . 5.4.1 Energy Optimization Scenarios . . . . . . . . . . . . . . . . . . . . . . . 5.4.2 Optimization Objectives . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5.4.3 Optimization Parameters and Constraints . . . . . . . . . . . . . . 5.4.4 Interdependencies between the Simulation Model and the Optimization Process . . . . . . . . . . . . . . . . . . . . . . . . 5.5 Software Evaluation and Selection . . . . . . . . . . . . . . . . . . . . . . . . . . 5.5.1 Overview of Multi-method Simulators . . . . . . . . . . . . . . . . . 5.5.2 Software Selection . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5.6 Prototypical Implementation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

50 57 61 61 63 67 68 71 72 75 80 83 87 91 92 94 94 97 100 104 106 109 111 116 117 119 121 122 123 126

Contents

xix

5.6.1 Description of the Fictional Production Scenario . . . . . . . 5.6.2 Simulation Model Details . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5.6.3 Adaptions of the Simulation Model in Preparation for the Simulation-based Optimization of Energy Efficiency . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5.6.4 Depiction of the Optimization Potential . . . . . . . . . . . . . . . 5.6.5 Evaluation of the Optimization . . . . . . . . . . . . . . . . . . . . . . . 5.6.6 Conclusions for the Practical Application of the Methodology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

126 128

132 136 145 152

6 Experimental Validation of the Methodology . . . . . . . . . . . . . . . . . . . . . 6.1 Presentation of the Practical Area of Application . . . . . . . . . . . . . . 6.1.1 Presentation of the Data Basis . . . . . . . . . . . . . . . . . . . . . . . . 6.1.2 Data Acquisition, Processing, and Validation . . . . . . . . . . . 6.2 Simulation Model Details . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6.3 Optimization Experiments . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6.3.1 Optimization Parameter Adaptations . . . . . . . . . . . . . . . . . . 6.3.2 Simulation-based Optimization using the 1-Min Resolution Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6.3.3 Simulation-based Optimization using the 1-sec Resolution . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6.3.4 Examination of the Necessity of Combined Simulation in Practice . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6.3.5 Summary and Evaluation of Simulation-based Optimization Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6.3.6 Financial Evaluation of the Increase in Energy Efficiency . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6.4 Recommendations for Action for Bosch Derived from the Simulation-based Optimization Methodology . . . . . . . . . 6.5 Evaluation of the Practical Applicability of the Simulation-based Optimization Methodology . . . . . . . . . . .

155 156 156 159 163 168 168

7 Summary and Outlook . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7.1 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7.2 Critical Appraisal of the Methodology . . . . . . . . . . . . . . . . . . . . . . . 7.3 Outlook and Future Work . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

195 195 197 204

References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

207

169 174 178 184 188 190 191

List of Figures

Figure 1.1 Figure 1.2 Figure Figure Figure Figure Figure Figure Figure Figure Figure

2.1 2.2 2.3 2.4 2.5 2.6 2.7 2.8 2.9

Figure 2.10 Figure 3.1

Figure 3.2 Figure 3.3 Figure 3.4 Figure Figure Figure Figure

3.5 3.6 3.7 3.8

Objectives and work structure . . . . . . . . . . . . . . . . . . . . . . . . Methodological system for IS and classification of this work . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Ways to study a system . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Stages of a simulation study . . . . . . . . . . . . . . . . . . . . . . . . . . Discrete and continuous time base in a simulation . . . . . . . Overview of simulation paradigms . . . . . . . . . . . . . . . . . . . . Cyclic and parallel interaction mode . . . . . . . . . . . . . . . . . . . Stages of a hybrid simulation study . . . . . . . . . . . . . . . . . . . . Local and global minima of an objective function . . . . . . . . Overview of optimization methods . . . . . . . . . . . . . . . . . . . . Different cases of sequential and hierarchical coupling of simulation and optimization . . . . . . . . . . . . . . . . . . . . . . . . Interactions between optimization and simulation model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Development of the global primary energy consumption to 2008 and three GEA pathways to 2030 and 2050 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Change in primary energy demand, 2016–2040 (Mtoe) . . . Global growth in energy consumption by sector . . . . . . . . . Producer price indices for electricity on the example of Germany . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Energy conversions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Energy balance . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Energy efficiency levels . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Energetic correlations in manufacturing companies . . . . . . .

5 7 11 12 14 17 22 23 26 29 31 33

41 42 43 44 45 47 50 51

xxi

xxii

List of Figures

Figure 3.9 Figure 3.10 Figure 3.11

Figure 3.12 Figure 3.13 Figure 3.14 Figure 4.1 Figure 5.1

Figure 5.2 Figure 5.3 Figure 5.4 Figure Figure Figure Figure

5.5 5.6 5.7 5.8

Figure 5.9 Figure 5.10 Figure 5.11 Figure 5.12 Figure 5.13 Figure 5.14 Figure 5.15 Figure 5.16

Classification of machine states according to time and optimization aspects . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Measured consumption profile of a production profile of a machine . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Representation of exact consumption profiles and consumption profiles of a machine determined by mean value formation . . . . . . . . . . . . . . . . . . . . . . . . . . . . Coherence of operational machine and energy state . . . . . . Application of the LPC methodology . . . . . . . . . . . . . . . . . . Example of electricity composition and sample load profile . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Convergence of disciplines in the context of energy efficiency in producing companies . . . . . . . . . . . . . . . . . . . . . Conceptual framework of the simulation-based optimization and interfaces to the real production system . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Overview of the hybrid simulation approach for the energy consumption model of a production . . . . . . . Specification of the module structure . . . . . . . . . . . . . . . . . . Example visualization of a production flow in AnyLogic . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Production flow and planning parameters . . . . . . . . . . . . . . . Production flow component . . . . . . . . . . . . . . . . . . . . . . . . . . Different complexity types of the machine logic . . . . . . . . . Machine logic and power consumption profile with varying production quantities . . . . . . . . . . . . . . . . . . . . . Production process parameters . . . . . . . . . . . . . . . . . . . . . . . . Machine component . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Machine-state dependent summation of single energy states to a total energy consumption . . . . . . . . . . . . . . . . . . . Energy Component . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Energy profile creation depending on the events in the material flow and the resulting machine behavior . . . Effects of events triggered in the machine component on the material flow . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Effect of machine failures triggered in the machine component on the material flow . . . . . . . . . . . . . . . . . . . . . . . Conceptual structure of the simulation module . . . . . . . . . .

53 54

55 56 58 60 69

95 96 98 99 99 100 101 102 103 103 104 106 107 109 110 111

List of Figures

Figure 5.17 Figure 5.18 Figure 5.19 Figure 5.20 Figure 5.21 Figure 5.22 Figure 5.23 Figure 5.24 Figure 5.25 Figure 5.26 Figure 5.27 Figure 5.28 Figure 5.29 Figure 5.30 Figure 5.31 Figure Figure Figure Figure Figure

5.32 5.33 5.34 5.35 5.36

Figure 5.37 Figure 5.38 Figure Figure Figure Figure Figure

5.39 5.40 5.41 5.42 5.43

Figure 6.1

xxiii

Machine logic and power consumption profile with varying idle times . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Energy savings through idle state avoidance . . . . . . . . . . . . . Consumption peak avoidance in production lines . . . . . . . . Status and peak load optimization of a four-machine scenario . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . The process of simulation-based optimization . . . . . . . . . . . Conceptual model for the simulation-based optimization of energy efficiency . . . . . . . . . . . . . . . . . . . . . . AnyLogic’s hybrid simulation engine architecture . . . . . . . . Integration of MATLAB libraries . . . . . . . . . . . . . . . . . . . . . . Fictional production scenario of a die-cast part processing line . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Machine delay calculation logic . . . . . . . . . . . . . . . . . . . . . . . Starting and ending of production shifts . . . . . . . . . . . . . . . . Machine set up for the depiction of realistic energy load profiles . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Definition of internal state transitions . . . . . . . . . . . . . . . . . . Implementation of the individual simulation focuses in an overall model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Implementation of the retention time in the machine logic . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ‘Select machine function’ logic . . . . . . . . . . . . . . . . . . . . . . . Throughput time estimation . . . . . . . . . . . . . . . . . . . . . . . . . . Prepare machine functionality . . . . . . . . . . . . . . . . . . . . . . . . Objective setup of total energy consumption optimizer . . . . Parameter and model time setup for the optimization experiment . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Optimization experiment visualization in AnyLogic . . . . . . Insertion of the offset time for shifting of machine starts . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Depiction of the offset effect for one machine . . . . . . . . . . . Setup of the peak optimization experiment . . . . . . . . . . . . . . Results of the total consumption optimization . . . . . . . . . . . Results of the peak consumption optimization . . . . . . . . . . . Peak optimization experiment in combination with the use of mean values for the energy consumption . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Bosch production lines 1 and 2 . . . . . . . . . . . . . . . . . . . . . . .

112 114 114 116 119 120 124 125 126 129 130 131 131 132 133 134 135 137 138 140 142 143 143 145 146 147

150 157

xxiv

Figure 6.2

Figure 6.3 Figure 6.4 Figure 6.5 Figure 6.6 Figure 6.7 Figure 6.8 Figure 6.9 Figure 6.10

Figure 6.11 Figure 6.12 Figure 6.13 Figure 6.14 Figure 6.15 Figure 6.16 Figure 6.17 Figure 6.18 Figure 6.19 Figure 6.20

List of Figures

Comparison of load profiles of three different product variants on machine 1 (of line 1) for a batch of 80 pieces/ 50 minutes per variant . . . . . . . . . . . . . . . . . . . . . . . . Load profile extract from the finish machine of line 2 . . . . Detailed load profile view (L2_finish) . . . . . . . . . . . . . . . . . . Machine logic and switching options of the Bosch production machines . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Simulation model of the Bosch production lines 1 and 2 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Comparison of data resolutions of L1_load over eight hours/one shift . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Optimization parameter setup of the production machines . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Results of the total consumption optimizer experiment run (data resolution 1-min) . . . . . . . . . . . . . . . . . . . . . . . . . . . First iterations of the total consumption optimizer experiment run (1-minute resolution of the energy data) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Peak consumption optimizer experiment run (1-min resolution) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Effects of the offset parameter on the power consumption peaks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Depiction of the offset parameter effect on the machine starting time . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Energy consumption profile of L1_load with one data record per second . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Energy consumption profile of L1_load with one data record per minute . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Results of the total consumption optimization experiment run (1-second resolution) . . . . . . . . . . . . . . . . . . Result of the peak consumption optimizer experiment run (1-second resolution) . . . . . . . . . . . . . . . . . . . . . . . . . . . . Depiction of the determined offset in the visualization of the energy flow at the example of six machines . . . . . . . Required machine setup to use energy consumption profiles . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Required machine setup to use calculated mean values . . . .

158 161 162 163 165 167 169 170

171 172 172 173 174 175 176 177 177 179 180

List of Figures

Figure 6.21

Figure 6.22 Figure 6.23 Figure 6.24

Figure 6.25

xxv

Visualization of the energetic profiles of two example machines and the total energy consumption of both lines using table functions and mean values . . . . . . Optimization experiment run results using mean values . . . Peak optimizer experiment run for the mean value scenario . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Comparison of the machine state time shares of the optimization scenarios in different data resolutions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Results of the simulation run (1-sec resolution optimization scenario) with the optimal parameter set of the 1-min resolution optimization experiment . . . . . . . . .

182 183 183

186

187

List of Tables

Table 2.1 Table 3.1 Table Table Table Table

4.1 4.2 4.3 5.1

Table 5.2 Table 5.3 Table 5.4 Table Table Table Table

6.1 6.2 6.3 6.4

Table 6.5 Table 6.6 Table 6.7

Comparison of DES, ABS and SD . . . . . . . . . . . . . . . . . . . . . Energy efficiency levels and their optimization parameters . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Summary of the literature analysis for intersection I . . . . . . . Summary of the literature analysis for intersection II . . . . . . Summary of the literature analysis for intersection IV . . . . . Process parameter of the die-casting production machines . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Optimization parameters for the reduction of the overall energy consumption . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Comparison of scenario key figures . . . . . . . . . . . . . . . . . . . . . Comparison of the use of energetic load profiles and mean values for the energy consumption . . . . . . . . . . . . . Overview of parts and carriers in the production line . . . . . . Assignment of switching options to machine states . . . . . . . . Overview of created simulation models . . . . . . . . . . . . . . . . . Comparison of the reference and the optimization scenario (1-min resolution) . . . . . . . . . . . . . . . . . . . . . . . . . . . . Comparison of the reference and the optimization scenario (1-second resolution) . . . . . . . . . . . . . . . . . . . . . . . . . Comparison of the reference scenarios at 1-min and 1-sec resolution . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Comparison of the mean value and the optimization scenario . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

18 62 85 86 87 127 139 148 149 159 164 166 173 178 178 184

xxvii

xxviii

Table 6.8

Table 6.9

Table 6.10 Table 6.11

List of Tables

Comparison of the idle and standby optimizer parameter configurations in the optimization scenarios at different data resolutions . . . . . . . . . . . . . . . . . . . . . . . . . . . Comparison of the parameter configurations in the mean value and the optimization scenario (1-min resolution) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Price components of the electricity supply contract . . . . . . . . Evaluation of the proposed simulation-based optimization methodology . . . . . . . . . . . . . . . . . . . . . . . . . . . .

185

188 189 192

1

Introduction

“To raise new questions, new possibilities, to regard old problems from a new angle re-quires a creative imagination and marks the real advances in science.” Albert Einstein

1.1

Initial Situation and Problem Definition

The importance of the energy and commodity markets has steadily increased since the first oil crisis in the 1970s. A price increase of nearly 80% for fossil fuels from 2000 to 2014 as well as debates on climate change and the exploitation of finite resources have triggered a mind change among politicians, entrepreneurs, and scientists [DLR2004, p. 3; St2014, p. 1]. The sustained use of energy and other resources has become a basic requirement for a company to competitively perform on the market. Based on the energy management as well as energy data acquisition systems, companies nowadays record energy data and try to increase their performance in the course of their continuous improvement processes. “A systematic analysis of materials and energy flows indicates significant potential savings for process integration, heat pumps, and cogeneration” [Jo+2012, p. 48]. The design of production processes therefore requires not only the consideration of logistical and technical production conditions but also the consistent optimization of resource Electronic supplementary material The online version of this chapter (https://doi.org/10.1007/978-3-658-32971-6_1) contains supplementary material, which is available to authorized users.

© The Author(s), under exclusive license to Springer Fachmedien Wiesbaden GmbH, part of Springer Nature 2021 A. C. Römer, Simulation-based Optimization of Energy Efficiency in Production, Forschung zur Digitalisierung der Wirtschaft | Advanced Studies in Business Digitization, https://doi.org/10.1007/978-3-658-32971-6_1

1

2

1

Introduction

consumption, as the production is directly linked to the use of resources and immediately affected by shortages and changes [Ne+2008, p. 2]. As the use of simulation technology has become a common tool to model, analyze, and assess dynamic production processes [St+2006, p. 394], the consideration of energy-related issues in the context of simulation is becoming a more frequent subject in scientific discussions. The integration of auxiliary and ancillary processes to support a holistic view of the value chain in a production unit requires new or modified simulation approaches [SWM2014, p. 71]. Existing approaches mainly focus on the consideration of resource consumption variables based on metrologically collected data on operating states, which are considered to be constant over a certain period of time. The energy consumption of machines is defined to be status-based and can be modeled and simulated using discrete event simulation (DES) approaches. Additionally, the system borders within the different scientific approaches to optimize the overall energy consumption of a production are defined in various ways and range from the consideration of machines that are directly involved in production processes to the consideration of compressed air connections and peripheral facilities. The prevailing majority of approaches does not consider all consumers that contribute a share to the overall energy consumption for the model definition, as the efforts for the data acquisition to model the operating states are too high. For the modeling and simulation of highly dynamic production processes DES approaches do often not provide a sufficient model accuracy to simulate such processes because they use quasi-static operating states. The latest scientific publications state the approach of hybrid simulation as a possible solution for a realistic representation of highly dynamic processes [SP2014, p. 109; Ba+2017, p. 67; SKS2017, p. 440; Pe+2017, p. 3792]. Using hybrid simulation could comprise the utilization of continuous simulation approaches, which are generally used for the modeling of physical processes, such as the behavior of liquids and gases, for the depiction of the total energy demand and combining it with a DES approach for the modeling of material flows and supporting logistic processes. By merging both models, the complex interactions between the material flow and the energy usage in production can be simulated closer to reality, especially the depiction of energy consumption peaks could become possible. An essential step towards reducing energy consumption in production is the optimization of the energy use of non-value-adding production phases. In these phases, the production equipment unnecessarily consumes energy but is not actively used to produce and create value. Companies can address this waste of resources without the need for large investments. Optimizing non-valueadding machine times thus harbors an enormous potential, whose development in industrial practice currently lacks the necessary planning tools. To capture all

1.2 Objectives and Work Structure

3

interdependencies in a manufacturing system, a holistic perspective is of high importance to analyze the total energy consumption in production. A realistic presentation of the energy consumption and its dynamics within the production system requires a significant visualization as the pure process description in form of mathematical models is rather a research topic than industrial practice. Using the simulation as the only tool does not lead to the optimization of the production system. The simulation, however, allows the accurate evaluation of different variants considering dynamic influences, i.e., time-dependent and stochastic aspects [St+2006, p. 394]. To be able to evaluate optimization potentials in production processes computer-based, it requires the coupling of the simulation model with an optimization tool. The simulation of energy use without integrated optimization algorithms can be considered as a tool to display material and energy flows as well as machine states during run time. But due to highly complex and dynamic processes in nowadays production systems, manual process optimizations carried out by single persons are not goal-oriented when aiming towards an energy efficient production in all relevant fields of action. Therefore, a holistic approach combining hybrid simulation with automated optimization algorithms is required to recognize systematic improvements in producing companies.

1.2

Objectives and Work Structure

This thesis focuses on the development of an application oriented and simulationbased procedure to allow for an optimum use of energy resources in manufacturing systems. Firstly, it needs to be examined, which different modeling approaches can be used to depict the dynamic energy consumption behavior of production processes and how an application-specific classification of the modeling and simulation approaches is necessary and feasible in practice to depict and evaluate the energy consumption behavior of production machines. In a second step, the considered simulation approaches need to be merged with optimization algorithms to determine the optimum use of energy in production without compromising production flexibility or output amount, nor influencing process and product quality. The objective of this thesis is to develop a methodology which allows the application-oriented and simulation-based optimization of energy efficiency in production to ensure a comprehensive and steady process quality regarding the optimal use of the factor energy in the production planning process. The applicability of the methodology is first tested using fictional case studies and subsequently verified using a practical example.

4

1

Introduction

The development of the methodology should be adequate to answer the following three primary research questions: Q1. Q2.

Q3.

How can the energy consumption of a production system be depicted in a simulation model that can also be used for optimization scenarios? How can an energy efficiency optimization of a production system be executed without causing any restrictions on production flexibility, without influencing the quality or the output of the production? How can the profitability of the energy optimization methodology be rated considering the various fields of application?

The structure of this thesis is shown in Figure 1.1. The introduction chapter is followed by an overview on the technical background in the context of simulation and optimization technologies (chapter 2). Besides the discrete event and the continuous simulation methods, the hybrid simulation approach will be discussed. Subsequently, optimization methods in general and the complexity of optimization algorithms will be checked, before the scientifically proven concepts for the combination of simulation methods and optimization algorithms are elaborated. Chapter 2 ends with a state-of-the-art-review on optimization software packages. Chapter 3 comprises an elaboration of the scientific basics on energy use in manufacturing, energy efficiency related topics as well as on energy pricing for industrial customers and energy saving potentials. In chapter 4, the state of the art of simulation-based optimization in production is given. Besides a literature review on existing publications, the chapter contains a definition of the research demand as well as the definition of the requirements for the methodology of this project. In chapter 5, the development of the simulation-based methodology for the energy efficiency optimization is described, including the conception of the methodology, the software selection for the use in the practical part of this thesis, as well as the modeling approaches, and the drafting of the fictional case studies used for a first validation. This chapter also includes a section on multi-method supporting simulation software tools and the software selection for the practical implementation of the methodology. Subsequently, chapter 6 is opened with a presentation of the practical area of application, followed by the transfer of results from chapter 5 on the use case. Chapter 6 closes with the validation of the simulation methodology. The thesis concludes with a summary, the answering of the research questions, a concept evaluation and an outlook in chapter 7.

1.2 Objectives and Work Structure

5

Analysis

Elaboration of scientific basics on „simulation and optimization techniques“

• Comprehensive inspection of all factors throughout the entire production process required • New approach by combining theories and using them in different application following an integrative approach

Elaboration of scientific basics on „energy“

• Energy use in manufacturing • Energy efficiency • Energy costs • Energy savings

• Comprehensive inspection of all factors throughout the entire production process required

• Studies on the optimization of the energy demand • Studies on the simulation-based modelling of the energy demand • Studies on the simulation-based optimization of the energy efficiency

• Summary of the current state-of-theart in the field of simulation-based optimization of energy efficiency • Definition of the research demand • Definition of requirements for the combined simulation approach as existing approaches do not meet requirements

Chapter 6

Chapter 5

Chapter 4

Chapter 2

Result

• Simulation methods • Simulation in the context of production systems • Combined simulation • Optimization methods • Combination of simulation and optimization

Chapter 3

Content

State of the art review on “Simulationbased optimization of the energy efficiency in production”

Development of a simulation-based methodology for energy efficiency optimization including: • the development of a conceptual framework, • a detailed description of all required components as well as • a prototypical implementation using a fictional use case scenario.

Experimental validation of the developed methodology in a practical area of application including: • the transfer of results of the fictional case study on the real business case and • an extensive and critical validation of the methodology.

Chapter 7

Figure 1.1 Objectives and work structure

6

1.3

1

Introduction

Classification in the Fields of Information Systems and Research Methods

Information Systems1 (IS) deals with the conception, development, implementation, maintenance, and use of computer-aided information systems applications across the economy [SM2009, p. 1]. The research methods in the fields of Information Systems are subdivided into two complementary but distinct paradigms: the behavioral science and the design science [BKN2009, p. 25; HC2010, p. 10]. Whereas the behavior-science paradigm attempts to develop and evaluate theories explaining or predicting individual or organizational behavior based on natural science research methods such as principles and laws, the design science paradigm with its origin in the fields of engineering is basically a problem-solving paradigm. “It seeks to create innovations that define ideas, practices, technical capabilities, and products through which the analysis, design, implementation and use of information systems can be effectively and efficiently accomplished” [HC2010, p. 11]. As described in section 1.2, this thesis is divided into a theoretical part covering the basic terms and theories on energy, simulation and optimization (chapters 3 and 2) as well as a state of the art literature research (chapter 4), and a practical part covering the development and the validation of a simulation methodology using fictional case studies and a business case (chapters 5 and 6). Following the methodological system for information systems according to Wilde and Hess [WH2007, p. 284], this thesis covers the areas of conceptual-deductive as well as formal-deductive analysis, reference modelling, and simulation. It can therefore be assigned to the field of constructive-quantitative methodologies (Figure 1.2) and accordingly to the design science paradigm. The theoretical chapters 3 and 2 are authored following an extensive and systematical literature review of relevant nationally and internationally published professional standards, scientific and technical books, standard compilation books, as well as scientific papers and publications to elaborate the state of research in the areas mentioned. Chapter 4 describes the convergence of disciplines of Operations Research (OR), Operations Management (OM) and Modeling and Simulation (M&S) in the context of energy efficiency in producing companies and induces the limitations and criteria for the literature research and analysis towards the research 1 The

term ‘Information Systems’ will be used as a translation for the German term ‘Wirtschaftsinformatik’ in this thesis. The author is aware of the fact, that the disciplines of IS and WI do not purely overlap but are handled as sister disciplines in literature [HS2008, p. 2]. In this document, the design-science orientated definition of the WI will be followed, referring to the authors Herzwurm and Stelzer for further definitions [HS2008].

1.3 Classification in the Fields of Information …

7

formal deductive analysis

laboratory experiment qualitative cross-sectional analysis

classification of this book in the fields of information systems

simulation

quantitative

reference modelling conceptual deductive analysis

degree of formalization

prototyping

qualitative

quantitative cross-sectional analysis

case study action research

behavioral science

argumentative deductive analysis

design science paradigm

Figure 1.2 Methodological system for IS and classification of this work [following WH2007, p. 284]

objective of this thesis. Used search terms, key word combinations, and data bases are described in section 4.1.1. Following a concept-centric structure for the literature review in sections 4.1.2 to 4.1.4, including concept matrixes to communicate major findings in the considered research approaches, the derivation of the research demand in section 4.3 and the definition of objectives and requirements in section 5.1 form the basis for the conceptional deductive development of the case studies and the resulting conceptual reference model in the sections 5.2 and 5.6. As described in Klinger and Wenzel, a reference model includes a systematic and general description of a defined area of application including all characteristics relevant to a given task. In the field of simulation, reference models are used as design scheme for simulation models to reduce the efforts of the simulation model creation process [KW2000, p. 13]. Remaining in the field of constructive quantitative research methods, the reference model is used as a basis to model the energy consumption in a hybrid simulation approach (in section 5.6 for the fictional case studies and in section 6 for the industrial case). The discrete-continuous simulation model in combination with the optimization algorithms are then being validated and evaluated (sections 5.6.5 and 6.4) in a formal-deductive practice. In-depth details on research methods in information systems can be found in the publications of Wilde and Hess[WH2006; WH2007], and Wenzel [We2000] and will therefore not be explained in this book any further.

8

1

Introduction

Due to the interdisciplinarity of the research topic and for the sake of completeness, a classification of the work in the field of business administration is made in addition to the classification in the field of Information Systems. Business administration is, according to the new literature, divided into the areas of production management, finance, marketing, business management, corporate accounting, etc. The focus of the production economy is the process of object transformation, i.e., the qualitative, quantitative, spatial or temporal change of objects, for the purpose of service provision [DS2010, p. 3]. While spatial and temporal changes are summarized as logistics processes, the production of a company deals with qualitative changes of the input variables which aim at a value increase. The main task of the production economy can thus be summarized in the business analysis of production systems. Of particular importance for the analysis of production systems is the close relationship with the neighboring disciplines, which are engineering, computer science, OR and some others. Production systems usually have a large number of technical design parameters that are examined with the aid of the relevant engineering sciences (mechanical engineering, process engineering, electrical engineering). The collection, processing and provision of decision-relevant information takes place with the aid of methods from computer science [DS2010, p. 6]. The topic of the simulation-based optimization of energy efficiency in production thus lies at the interface of different research areas, which are brought together in this book to find practical solutions.

2

Simulation-Based Optimization

There are things we can do to combat complexity. We must use all the resources at our disposal to fight the complexity dragon. At the moment, he’s winning. J AMES O. H enriksen (Henriksen at the Titan’s Talk of the Winter Simulation Conference 2006 [He2006, p. 2].)

“Simulation refers to a broad collection of methods and applications to mimic the behavior of real systems, usually on a computer with appropriate software” [KSZ2002, p. 3]. The use of simulation is a proven decision support tool in operational practice for production planning and control, whenever a complex system of target figures, control parameters and disturbance variables are present, and the number of system components results in a complex system behavior over time. It is hardly feasible to handle this complexity by analytical methods and to give tractable mathematical formulations [MW2011, p. 7]. Through the simulation, it becomes possible to depict system-inherent causal relationships and to calculate the result variables based on the dynamic runtime behavior. Simulation is a high value tool but is not fully sufficient by itself. “An extra step is needed—a step that joins simulation and optimization” [Ap+2004, p. 76]. This thesis uses modeling and simulation techniques as a tool to depict the energy consumption as well as its complex interactions with material and production workflows in manufacturing. Modeling these different aspects in one approach requires the use of hybrid simulation (section 2.1.2). The individual simulation paradigms have their own graphical approaches to represent models, “but they do not have obvious capabilities to depict the hybridization elements” [Br+2019, p. 727]. For this reason, one focus of this thesis is placed on the conceptual model, which will be used in © The Author(s), under exclusive license to Springer Fachmedien Wiesbaden GmbH, part of Springer Nature 2021 A. C. Römer, Simulation-based Optimization of Energy Efficiency in Production, Forschung zur Digitalisierung der Wirtschaft | Advanced Studies in Business Digitization, https://doi.org/10.1007/978-3-658-32971-6_2

9

10

2

Simulation-Based Optimization

chapter 5 to explain the relationship of the individual model elements of different simulation methods in detail. To identify the optimal production setup, optimization will be used to increase the energy efficiency in production (section 2.2). “Advances in the field of metaheuristics -the domain of optimization that augments traditional mathematics with artificial intelligence and methods based on analogs to physical, biological or evolutionary processes- have led to the creation of optimization engines that successfully guide a series of complex evaluations with the goal of finding optimal values for the decision variables” [Ap+2004, p. 76]. Thus, optimization engines appear to be the ideal tool for dealing with the issue of increasing resource efficiency in production plants. Therefore, chapter 2 briefly describes relevant terms and definitions in simulation and optimization as well as possible combinations of the two. Reference to advanced literature will be made in the appropriate places of the text.

2.1

Modeling and Simulation

In a broad sense, simulation refers to the process of constructing, using and verifying a virtual, mostly computer-based experiment or reproduction of a system to gain insights on the system that are transferable to reality [Ba+2005, pp. 3–4; Du2018, p. 13]. A simulation study aims at the abstract and consistent imitation of a real-world system with its dynamic processes in an experimental computer model that contains a variation of parameters and structures, constituting a system’s behavior over time [Ba1998, p. 3; VD2014a, p. 3] to conduct “experiments with this model for the purpose either of understanding the behavior of the system or of evaluating various strategies (within the limits imposed by a criterion or set of criteria) for the operation of the system” [Sh1975, p. 2]. A system consists of a group of objects, which are characterized by attributes, that interact with each other and have interdependencies towards the accomplishment of a given purpose within defined system boundaries [Ba+2005, p. 8]. A simulation allows to run through numbers of scenarios and various system settings that cannot easily be altered in a real production as this would be too costly and disruptive. The process of incrementally and interactionally varying model parameters and comparing the system’s behavior against a starting situation is referred to as simulation in the narrow sense [Du2018, p. 13]. Assumptions about the systems can be considered, as they are sometimes necessary to develop a model [Go2015, p. 15]. Consequently, different ways to study a system can be followed (Figure 2.1). To perform a simulation, it is required to have a look at a typical simulation life cycle and its stages. Every simulation study starts with the problem formulation

2.1 Modeling and Simulation

11

Figure 2.1 Ways to study a system [La2007, p. 4]

of a (real-world) problem and the objectives that are to be addressed and ideally ends with the solution implementation. The paths between these two points are described differently in literature. They mainly vary in the order and level of detail of the individual steps. A similar list of required steps can be found in all sources, such as Balci, Banks, Law, Robinson and Shannon [Ba1998, pp. 15– 18; Ba+2005, pp. 11–16; Ba2012a, pp. 874–882; La2003, pp. 66–69; Ro2008, pp. 279–281; Sh1998, pp. 9–13]. As shown in Figure 2.2, this thesis follows the stages of the simulation lifecycle according to Robinson and Eldabi et al. [Ro2008, pp. 279–281; El+2018, p. 1501]. A special focus is put on the conceptual model, as the conceptual modeling is an extremely important step in a hybrid simulation study1 . “The development of a conceptual model is a purely mental activity that involves more art than science and requires that the complexity of the physical system be reduced and controlled, keeping in mind a possible operational formalization of the model” [TB2017, p. 23]. All steps of the simulation lifecycle are known and will be applied in section 5.6 and chapter 6 during the prototypical implementation 1 The term hybrid simulation study as well as the reasons for focusing on the conceptual model

in this context will be explained in detail in section 2.1.2.

12

2

Simulation-Based Optimization

and the practical use case, but will not be described in detail at this point. The reader is referred to further literature2 .

Real world problem

Conceptual Model Validation

Operational Validation

Conceptual Model

Modelling and general project objectives Solutions/ understanding Experimental accepts factors

Scenario Verification

Computer model

Model content: scope and level of detail

provides

Responses

Model Verification

Figure 2.2 Stages of a simulation study. Adapted from [Ro2008, p. 280; El+2018, p. 1501]

Robinson defines conceptual modeling as “a non-software specific description of the simulation model that is to be developed, describing the objectives, inputs, outputs, content, assumptions, and simplifications of the model” [Ro2004, p. 65]. In a hybrid simulation study, the type of hybridization and the links between the single sub-models used, have to be described, as well as the combination of modeling techniques used (e.g., system dynamics and discrete-event simulation, or system dynamics and agent-based simulation, or a combination of all three of them) [El+2018, p. 1501]. In the panel discussion on hybrid simulation at the Winter Simulation Conference 2018 [El+2018], Eldabi et al. stated, that conceptual modeling for hybrid simulation is not well developed. Individual simulation methods all have their own approaches for conceptual models, but hybridization elements are not modelled so far. Questions to answer with a conceptual model include how the sub-models are interrelated, what information is exchanged, and how process flow models, state charts, and stock flow models can be combined in one approach. The overarching requirement for model conception is “to keep the model as simple as possible to meet the objective of the simulation study” [Ro2010, p. 20] 2 Brailsford et al. describe a life-cycle based framework for hybrid simulation based on the stages of a simulation study in [Br+2019, pp. 723–725], so do Eldabi et al. in “Hybrid Simulation Challenges and Opportunities: A life-cycle Approach” [El+2018, p. 1501–1503].

2.1 Modeling and Simulation

13

as simple models are developed quicker, are more flexible, require less data and can be executed faster. There are several options to classify models3 of which three are depicted here. Models can be static or dynamic, deterministic or stochastic, and discrete or continuous. While static models are used to constitute a system in which time is no longer a factor, dynamic models depict a system as it evolves in the course of time. As the experimentation with static models only depicts a point behavior4 , it can only be used very limited for computer simulation [KSZ2002, p. 9; Ör2009, p. 19; MC2007, pp. 3.8–3.9]. Therefore, static models are not relevant for this thesis. The second dimension differentiates models according to the randomness of a system. A deterministic model does not involve any probabilistic components, the behavior is completely predictable, and the output is determined. If a system has random input components, it is referred to as a stochastic system. Manufacturing and inventory systems are generally described in a stochastic simulation model [Ba+2005, p. 11–12; La2007, p. 5–6]. The third dimension distinguishes between the system dynamics. Systems are hybrid by nature, only a “few systems in practice are wholly discrete or wholly continuous; but since one change predominates for most systems, it will usually be possible to classify a system as being either discrete or continuous” [LK2000, p. 3]. For discrete models, the state variables change only at separated points in time (time-stepped) or on the occurrence of state-affecting events (event-stepped) [Ba1998, p. 8; He2008, p. 18] while a continuous model concerns the modeling over time with the state variables changing continuously with respect to time, having an infinite number of states [La2007, p. 70]. Depending on the object of study, both, the continuous time-advancement and the discrete event state jumps might be important for the simulation purpose [Lu2002, p. 3]5 . Depending on the type of model performed, simulations can be subdivided into discrete, continuous or hybrid approaches. After a short introduction on discrete and continuous simulation in section 2.1.1, the hybrid 3 The

‚pieces of a simulation model‘ (e.g., entities, attributes, (global) variables, resources, cues, events, etc.) are assumed to be basic knowledge and will not be described here. The interested reader is referred to Kelton, Sadowski and Randall [KSZ2002, pp. 24–29] as well as to Banks et al. [Ba+2005, pp. 8–9], who have worked out the basics in great detail. 4 “Most mathematical and statistical models are static, in that they represent a system at a fixed point in time. Consider the annual budget of a firm. The budget resides in a spreadsheet. Changes can be made in the budget and the spreadsheet can be recalculated but the passage of time is usually not a critical issue” [Ba1998, p. 7]. 5 Readers are referred to ‘Banks’ Handbook of Simulation [Ba1998, pp. 6–13, pp. 31–36 and pp. 47–49], Discrete-Event System Simulation [Ba+2005, pp. 3–12] and LAW’s Simulation Modeling and Analysis [La2007, pp. 1–6, pp.] for further reading on the definitions of system, model and simulation in general.

14

2

Simulation-Based Optimization

character of systems combining those two system dynamics will be focused on in section 2.1.2.

2.1.1

Continuous and Discrete Simulation

The time progress of simulation models can either be continuous or discrete (Figure 2.3). “In most simulations, time is the major independent variable. Other variables included in the simulation are functions of time and are the dependent variables. The adjectives discrete and continuous when modifying simulation refer to the behavior of the dependent variables” [PO1999, p. 20].

Figure 2.3 Discrete and continuous time base in a simulation [Ru2018, p. 8]

Continuous simulation models have dependent state variables that are defined as continuous functions of time. The state variables in continuous models can be differentiated according to the underlying equation system. The most used equation systems consist of ordinary differential equations (ODEs) and explicit functional forms [Wi1998, p. 181; Pr1998, p. 43]: x˙ = f (x, t) y = g(x, t)

(2.1)

The state variables can also be represented by a system of difference equations as:

2.1 Modeling and Simulation

15

yn+1 = ay n + bun

(2.2)

Does the simulation purpose require the parallel consideration of temporal and spatial processes, partial differential equation (PDE) systems are used [Wi1998, p. 15]. The independent variable in continuous simulation models is typically the time. For the state variables that are described by explicit functional forms or differential equations, a specified value for an initial point in time is defined, using these values as an input to generate new values at the next point in time [Pr1998, p. 43]. The continuous simulation has its origin in the “Industrial Dynamics” published by Forrester in the late 1950’s to investigate the dynamics of industrial processes. „Industrial Dynamics is the study of information feedback characteristics of industrial activity to show how organizational structure, amplification (in policies), and time delays (in decisions and actions) interact to influence the success of the enterprise. It treats the interaction between the flows of information, money, orders, materials, personnel, and capital equipment in a company, an industry, or a national economy“ [Fo1961, S. 13]. Due to its generality and its applicability in other research areas, Industrial Dynamics was further developed and later renamed into System Dynamics (SD) [Sc2004, p. 34]. In SD models, a distinction is made between stock and flow parameters. The depiction of stocks, feedbacks and flows in stock- and flow-diagrams is of central importance in the SD concept. The stock-and-flow diagrams are formulated in a system of nonlinear ordinary differential equations that calculate the change of each variable through integration over time [St2000, p. 903]. Discrete simulation models have dependent state variables that change instantaneously at distinct points in time, the so-called event-times. Pritsker distinguishes discrete simulation models according to their world view describing the system [Pr1995, pp. 52–60]. The event-oriented world view requires the definition of all events that can possibly occur in a system and change its state. The state of the model remains constant between single events, and the dynamic of a system is portrayed by simulating time from one event to the next. The approach is therefore very efficient, as only the state changes in occurrence of events are modeled [Wa2009, p. 11]. The event-oriented view functions as a basis for further world views, the activity-oriented or activity scanning and the transaction-oriented world view. For the activity scanning view the conditions for an event to start or end are described, but the event’s start as well as the end is not scheduled. The conditions for each activity are tested while the simulation moves from event to event. “To ensure that each activity is accounted for, it is necessary to scan the entire set of activities at each time in advance” [PO1999, p. 25]. As a result, models using the activity scanning approach cannot be implemented efficiently. Both,

16

2

Simulation-Based Optimization

the activity-oriented and the transaction-oriented view find little or no practical relevance anymore [Pa1991, pp. 25 and 29]. The process interaction view summarizes logically connected activities which occur in defined pattern to processes through which the entities in a model flow. The interactions in a model are thus not described by events but through processes [PO1999, p. 26]. A sub-class of the process interaction approach is the transaction flow, which differentiates transactions (dynamic objects) and stationary objects (stations). While the event-oriented worldview was basically used during the first two decades of simulation, focusing on the events themselves as the main modeling event, the process-oriented world view provides a more natural way to describe a system which is easier to model [Pe2017, pp. 83–84]. The event-discrete time advance is also used for agent-based simulation (ABS) models [NM2007, p. 11]. In the 1980s, the process-oriented view gradually replaced the event-oriented world view, therefore most DES systems are process-oriented by default. Nevertheless, the event-oriented time progress forms the core of most discrete-event simulators [KSZ2002, p. 37]. “All discrete event simulation systems implement their internal logic using this basic modelling approach, regardless of the world view that they present to the user” [Pe2010a, p. 212]. The process-oriented worldview is nowadays often implied when the term DES is used. This assumption also applies for this work. The simulation concepts used in this work are based on the classification described above. For completeness, it should be mentioned at this point that other classification options for simulation models can be found, in Law and Kelton [LK2000], Banks et. al [Ba+2005], and Wenzel [We2000]. Zeigler, Praehofer and Kim describe different modeling formalisms according to the presentation of the time base and the value of state variables. They describe three different basic formalisms6 , the Differential Equation System Specifications (DESS), the Discrete Time System Specifications (DTSS) and the Discrete Event System Specifications (DEVS) [ZPK2010, p. 6, p. 137 ff.]. The interested reader is referred to the further literature, since in-depth explanations would go beyond the scope of this work.

6 Besides the three basic formalisms DESS, DTSS and DEVS and for the sake of completeness,

the Discrete Dynamic Systems has to be mentioned as well as a specification of the DEVS models. The interested reader is referred to the detailed explanations in [ZPK2010, p. 6 and p. 137 ff.].

2.1 Modeling and Simulation

17

Based on the above-mentioned classification, three major modeling paradigms can be found in today’s simulators7 , ‘System Dynamics’, ‘Discrete-Event Simulation’ and ‘Agent-Based Simulation’. These three modeling paradigms are used as the basis for this thesis (Figure 2.4). As described earlier, SD is based on a continuous approach, while DES and ABS follow the discrete event view [BF2004, p. 3].

Figure 2.4 Overview of simulation paradigms [BF2004, p. 3]

SD typically addresses problems on a high level of abstraction, e.g., economic or social systems and use stock and flow elements as well as balancing or reinforcing feedback loops to describe cause-effect-relationships and thus the dynamic systems behavior [Fo1968, pp. 402–404]. DES is used in research and industrial practice for different purposes, generally showing entities, which are passive objects like parts, tasks, or documents, going through “the blocks of the flowchart where they stay in queues, are delayed, processed, seize and release resources, 7 Dynamic

Systems (DS) as a fourth simulation paradigm is not of interest for this thesis, as it is used mainly for physical state variables [BF2004, p. 3]. The interested reader is referred to the further literature at this point [Di2014, pp. 11–33 and pp. 85–137; Ka1993; BF2004, p. 5].

18

2

Simulation-Based Optimization

split, combined, etc.” [BF2004, p. 6], following the process-oriented world view. ABS is a relatively new simulation approach, which is used to model complex systems of interacting agents for a broad spread of disciplines such as supply chains, modeling the behavior of the stock market or the spread of diseases [MN2014, p. 11; Ta2014, p. 2]. Wooldridge defines an agent being an “computer system that is situated in some environment, and that is capable of autonomous action in this environment in order to meet its delegated objectives” [Wo2009, p. 21]. Therefore, every agent is a modular and identifiable individual, having an internal state and a set of rules and behavior patterns representing its current situation within the model [NM2007, p. 24; He+2011, p. 2791]. The major differences between DES, ABS and SD are summarized in Table 2.1. Table 2.1 Comparison of DES, ABS and SD. Following [RS2011, p. 32], [La2000, p. 16] and [SM2003, p. 7] DES

ABS

SD

Perspective

analytical, emphasis on detail complexity

analytical, emphasis on detail complexity

holistic, emphasis on dynamic complexity

Resolution of model

mesoscopic/microscopic

microscopic

macroscopic

Basic building blocks

individual but non-interacting entities, process flow blocks

individual interacting agents, state charts

stock and flow elements and feedback loops

Unit of analysis

rules

rules

structure

Modeling efforts

high

high

low

Time increments

variable

variable

constant

State changes are caused by

events regarding location and state changes of the objects

events regarding location and state changes of the objects

simulation time advance

Mathematical formulation

logic

logic

integral equations

However, if systems are to be analyzed that contain both, discrete and continuous components that are equally relevant to the overall system behavior, hybrid modeling and simulation approaches are required for a realistic description of the actual system. The hybrid modeling approach overcomes the weaknesses of the

2.1 Modeling and Simulation

19

individual model techniques and provides a structure to study different aspects of a real-world problem.

2.1.2

Hybrid Modeling and Simulation

Tolk defines hybrid—biologically and technically speaking—as “the result of merging two or more components of different categories to generate something new, that combines the characteristics of these components into something more useful. A mule is a biological hybrid, the crossbred of a donkey and a horse with better endurance and a longer useful lifespan than its parents. Crops grown from hybrid seeds produce plants of high quantity and quality. A hybrid car combines the advantages of gasoline engines and electric motors. Hybrid golf clubs combine the characteristics of wood and iron. Hybrids take two—or more— components and create something better” [Mu+2017, p. 1640]. The reason for mixing methods is that real world problems are usually very complex and neither completely event-discrete, nor completely continuous. “They require different methods to address the multiplicity of dimensions of a problem. Additionally, all methods have different strengths and weaknesses so mixing methods can overcome the limitations of one method” [Mu+2017, p. 1638]. There is no consensus in literature or throughout the hybrid simulation community on one precise definition of hybrid simulation. “Furthermore, new terms have been introduced which, arguably, have the same meaning, e.g., multi-method simulation, multiparadigm modeling, cross-paradigm simulation, mixed-modeling and combined simulation” [Mu+2017, p. 1631]. All approaches describe a combination of two or three of the most commonly applied simulation techniques DES, ABS and SD in connection with the occurrence of the term hybrid simulation or one of its derivates. Hybrid simulation is usually applied in the implementation stage of a simulation. At this point, it is important to differentiate hybrid simulation from hybrid systems modeling (HSM), which describes the combination of simulation techniques “with methods and techniques from disciplines such as Applied Computing, Computer Science, Systems Engineering, and OR” [MP2018, p. 1430], to name for example problem structuring methods, forecasting, classical optimization techniques, process mining, data mining, and machine learning. HSM is not only applied in the implementation phase of the simulation study life cycle but can also be applied in the conceptual modeling phase, the model verification and validation phase, as well as in the experimental stages8 . 8 Compare

here the full paper of Mustafee and Powell for further reading [MP2018].

20

2

Simulation-Based Optimization

The idea of combining different simulation paradigms is far from new as there are references from the 1960’s and earlier. “Over the years ‘hybrid simulation’ has meant a number of things: models that are simultaneously implemented on both analogue and digital computers, or models that contain both discrete and continuous variables, or even models that combine simulation with an analytical method such as optimization” [Br+2019, p. 2]. Examples for the combined use of DES and SD can be found in Zeigler’s and Oeren’s “futuristic simulation environments which support flexible adoption of multiple perspectives” [ZO1986, p. 708], Pritsker’s approach to simulate the loading and unloading processes of oil tankers [Pr1995, pp. 354 ff.], or Mosterman with his models, having continuous and discrete proportions that can be switched between to adjust the level of detail [MB1997; Mo1999]. Even though DES and SD have been established methods for more than 60 years, “[…] the combination of continuous-time simulation and DES is still a challenging area of research” [BFG2015, p. 139]. The youngest of the three simulation paradigms, ABS, has become popular within the simulation community in the last decade. Several approaches combining ABS and SD9 or ABS and DES10 pairwise as well as ABS, DES and SD11 together in one approach can be found in literature. ABS has proven to be a useful tool to model autonomous agents and their interactions in complex systems to illustrate the agent’s behavior in detail [Kh+2015, p. 1491]. This thesis follows the definition of hybrid simulation according to Brailsford, who states that “hybrid simulation is one single conceptual model 9 Examples

on the combined use of ABS and SD can be found in Djanatliev et al., who develop a combined approach to assess health care technologies to learn about impacts of new products before product releasement [Da+2014], in Milling, who is building a SD model to analyze diffusion patterns in combination with an ABS to model the occurrence of innovations and the adaptation over generations [Mi2002], as well as in Schieritz and Grössler who combine SD and ABS for reducing the a priori complexity of supply chain models [SG2003]. 10 The combination of ABS and DES is used in Dubiel and Tsmihoni to model human agents traveling freely through a DES environment [DT2005], in Fakhimi et al. for analyzing sustainable planning strategies for emergency medical service [Fa+2014], in Liraviasil et al. to simulate efficient manufacturing processes under change [Li+2015], and in Nguyen et al. for the strategic aircrew manpower planning [Ng+2017]. Khedri Liraviasl et al. use ABS and DES to model assembly systems in production for depict interchangeable and reconfigurable production layouts [Kh+2015]. 11 Examples on the combined use of all three simulation techniques are found in Djanatliev and German to predict the effects of medical products [DG2013] and in Konstantinos and Angelopoulou who present a modeling framework for combining DES, SD and ABS approaches [MA2019].

2.1 Modeling and Simulation

21

that, when implemented in a computer software, uses more than one simulation paradigm” [Mu+2017, p. 1634]. Consequently, and in contrast to the definition of some other scientists, the combination of DES and ABS is already seen as a hybrid, even though both simulation approaches are based on a discrete time progress. Brailsford’s approach is one of the few that emphasizes, that the hybridization of simulation paradigms occurs in one model rather than by combining different simulators. While Tolk differentiates between “hybrid M&S studies, in which various modeling paradigms are applied in an orchestrated set of simulation tools, and hybrid M&S systems, where several such paradigms are used within one simulation” [Mu+2017, p. 1640], other scientists, e.g., Djanatliev, argue that in a hybrid model, the simulation paradigms need to have a different resolution according to their time advance to be called ‘hybrid’ at all [Mu+2017, p. 1636]. Following this logic, the term ‘hybrid simulation’ would indicate the combined use of continuous and discrete simulation approaches, while the combination of different simulation paradigms without a distinction between continuous and discrete should be referred to as multi-paradigm simulation [Mu+2017, p. 1636]. Chahal even defines a spectrum for modeling, analysis, and synthesis of hybrid systems, having hybrid approaches which constitute the extension of continuous systems to model discrete events on the one end and “on the other end […] discrete models extended to represent the behavior of continuous models” [Ch2009, pp. 28–29]. In between those two extrema, mixed discrete and continuous approaches, combining complementary aspects of both techniques, can be found [Ch2009, p. 29]. The following definitions are applied in this work: • Hybrid Simulation Hybrid simulation represents the generic term for the use of at least two different simulation paradigms in one simulation model. The simulation paradigms may have the same time progress. • Combined Simulation Combined simulation represents a sub-form of hybrid simulation in which the simulation paradigms used follow a different time progress. Thus, the combination of DES and SD or ABS and SD is called combined. • Hybrid System Modeling Hybrid systems modeling describes the combination of simulation techniques with methods from Applied Computing, Computer Science, Systems Engineering, Artificial Intelligence, and OR.

22

2

Simulation-Based Optimization

Independently of the naming of the combination of different simulation paradigms, the way information is exchanged over time progress between sub-models has to be defined. Generally, there are two modes of interaction for sub-models, they can perform cyclic or parallel interaction [Ch2009, p. 45]. For the cyclic interaction modes (Figure 2.5a), sub-models are run separately and do not exchange any information during runtime. Instead, the interaction takes places after completing a simulation run, when discrete outputs are used to feed the continuous sub-model and vice versa. In the parallel interaction mode, discrete and continuous models are run simultaneously and exchange information during runtime (Figure 2.5b), thus continuously and discrete changing elements affect each other directly [Ch2009, p. 46].

Figure 2.5 Cyclic and parallel interaction mode [Ch2009, p. 45 and p. 47]

“Variables whose values are changed or influenced by variables of the other model and variables which replace or influence the values of variables of the other models during hybrid simulation […] [are] named […] interaction points” [Ch2009, pp. 53–54]. These interaction points (IP) function as interfaces between the sub-models and, in combination with the interaction mode, combine them into a holistic model. Chahal and Eldabidefine a three-stepped procedure to build hybrid models, which can easily be integrated in the typical lifecycle of a simulation study: (1) problem identification and justification to use hybrid approaches, (2) identification of interaction points of SD and discrete simulation paradigms, and (3) identification of a mode of interaction for the models [DG2015, p. 1610, Ch2009, p. 48] (Figure 2.6). Prior to model implementation, in the conceptual phase of the modeling process, the modeler must think about the nature of the system to be able to find a fit between the modeling paradigms, the modeled system, and the problem (justification to use hybrid approaches). For this, the level of abstraction as well as the views on a system must be defined, followed by a linking of paradigms to the levels of abstraction.

2.1 Modeling and Simulation

Real world problem

23

Conceptual Model Validation

Conceptual Model

Operational Validation

Modelling and general project objectives determine Experimental factors

Solutions/ understanding

accepts

Model content: scope and level of detail

Scenario Verification

Justification to use hybrid approaches

Sub-model definition

provides

Responses

Model Verification

Definition of discrete and continuous interaction points Identification of mode of interaction for sub-models

Computer Model

Figure 2.6 Stages of a hybrid simulation study. Adapted from [Ro2008, p. 280; El+2018, p. 1501; Ch2009, p. 48]

In combined simulation models, three basic forms for linking discretely changing and continuously changing state variables can occur [Pr1995, pp. 61 ff.]: (1) A discrete event can cause a discrete change in the value of a continuous state variable. (2) A discrete event can change the relationships that determine the behavior of a continuous state variable at a particular time. (3) Reaching a defined limit by a continuous state variable can trigger a discrete event. The interaction of combined ABS-DES models is different. While the behavior of entities in DES is determined by rules which may range from if-then conditions to highly complex algorithms, agents are autonomous individuals which have the ability to act independently, to learn from the past and to adapt their behavior for the future. Their behavior is determined by learning algorithms and the abstract presentation of the relationships between the system states and the individual agent. Agents communicate by passing signals or messages from one to another and they can pass messages to their environment [Br2014, p. 1541]. Thus, either messages or occurring events affect the behavior of agents and entities in a combined model.

24

2

Simulation-Based Optimization

For a straight forward model development, especially when combining different simulation paradigms, a structured concept helps to achieve sustainable results. Therefore, the development of a conceptual model makes up the basis for robust and realistic computer model that can be used for simulation experiments12 . Summarizing, it can be said that hybrid simulation is used in various forms to assess problems from different dimensions and high complexity. The hybridization of simulation models based on mixing simulation paradigms gives a broader flexibility to the modeler, e.g., to capture problems that refer to discrete and continuous structures at the same time. An example for such a problem is the depiction of energy consumption in production systems. While the representation of the energy flows requires the use the SD paradigm, events in the production flow, as for example the flow of parts through the machines, causing changes in the energy consumption of production machines, require the use of the DES paradigm. The hybrid simulation allows the depiction of discrete and continuous processes in one single model.

2.2

Optimization Methods

Optimization is “the process of searching for the best value that can be realized or attained. In mathematical programming, this is the minimum or maximum value of the objective over the feasible region” [Op2013, p. 1092]. Due to its general formulation, optimization can be seen as a cross-application, which occurs in the most diverse contexts [Li1992, p. 1]. To name just a few examples: in the process industry, mixture problems are often considered, in which optimal mixing conditions must be determined considering various rules and restrictions [SM2009, p. 22]. While the supply chain management (SCM) looks at optimization tasks concerning the entire value chain through demand-based deliveries, faster adaptation to changes in the market, reduction of stocks and costs in the logistics chain as well as shorter order processing times, optimization problems in the area of transport and traffic are the performance increase of networks and routes, the optimal fleet assignments and crew deployment planning or the optimization of dynamic traffic situations, such as traffic densities in road traffic [SM2009, p. 23]. The scope of optimization is infinite, ranging from portfolio optimization analysis in finance, determining optimal energy costs and planning decisions in energy

12 The

interrested reader is referred to Eldabi et al. for further reading on representation methods for a hybrid simulation conceptual models [El+2016, pp. 1397–1398].

2.2 Optimization Methods

25

and water production, to total revenue optimization in revenue management or the optimal resource allocation in health management. This chapter summarizes the basics of optimization methods starting with an introduction on the common terminology in the field of optimization techniques. Since there is no uniform classification of optimization methods in the literature, the classification followed in this book is presented in section 2.2.2.

2.2.1

Terminology in Optimization

Optimization is seen as a basic tool in the field of Operations Research, concerning the development of solutions and algorithmic techniques for solving mathematical problems under consideration of constraints or boundary conditions which are limiting the allowed range of independent variables [Ba2012b, p. 3; PR2002, p. XV; SM2009, p. 6]. An optimization model is a formal representation of a planning problem that, in its simplest form, contains at least one alternative set and one objective function that evaluates it. It is developed in order to be able to determine optimal or suboptimal solutions with suitable methods [Do+2015, p. 4]. Once an optimization problem and the variables influencing the optimization problem, the input parameters, and, if necessary, constraints setting bounds for input parameters are defined, the way of evaluating the performance of the problem needs to be found. The performance measure is the objective function and the range of its possible values is the solution space [Fu2015, p. 1; Li1992, p. 9]. The general parametric optimization problem where the objective function is to be minimized is min f (x), x ∈ X ⊆ Rn x∈

with

f x  X Rn

(2.3)

objective function decision variables feasible region or constraint set design space of the optimization problem geometrical space of decision variables

Mathematical optimization is generally determining a minimum or maximum of the objective function. Both extremes are summarized using the term optimum. There is an equivalence relationship (principle of duality) between the

26

2

Simulation-Based Optimization

minimization and maximization tasks, therefore a minimization problem can be transformed into a maximization problem [GB2018, p. 3]: max f = −min(− f ) and min f = −max(− f )

(2.4)

The objective function will generally reach its optimum in a single point, the global optimum [TRP2013, p. 650]. In addition, there are points in the immediate vicinity of which the objective function only assumes larger values than in this point itself (Figure 2.7). These points are called local minima [Sc2016, p. 424; JH2015, p. 1782].

F(x)

local minimum global minimum x Figure 2.7 Local and global minima of an objective function [Sc2016, p. 424]

2.2.2

Exact and Heuristic Optimization Methods

Optimization methods can be categorized according to various criteria13 . Due to the rapid development of algorithms and the variety of disciplines in this area, 13 Yang and Koziel present a good overview of possible classifications, including gradientbased and gradient-free algorithms, trajectory-based and population-based algorithms, deterministic and stochastic algorithms, as well as a classification according to the consideration of randomness [YK2011, pp. 4–6]. Note that several different classification systems

2.2 Optimization Methods

27

no uniform classification can be found in literature. More than 1000 optimization programs can be found worldwide. “In mathematics we distinguish between exact methods that deliver a totally definite answer without uncertainty and heuristic methods that deliver an approximate answer” [Ba2012b, p. 5]. Exact methods determine the optimal solution of an optimization problem for a given period of time with exponentially increasing computational efforts with increasing numbers of variables or boundary conditions [PH2010, p. 401]. Exact optimization methods are very computationally intensive and therefore only suitable for a limited system complexity. Additionally, exact methods are not very robust, which means that every time the problem changes, the algorithm needs to be adjusted [MF2004, p. 55]. The most basic exact method is the complete enumeration, which lists all possibilities and selects the best one [Ba2012b, p. 5]. By breaking up the solution space into sections and furthermore creating hierarchical structures (branches) for these sections, a search tree for the branch-and-bound algorithm is generated. “The bound part of the method uses a problem-specific method to compute an upper and lower bound upon the goal function value” [Ba2012b, p. 6]. Thus, large parts of the solution space can be easily identified as unsuitable for the optimum solution of the optimization task. Exact methods are also called mathematical programming and make up the only kind of optimization accepted by the mathematical optimization science [Ca2013, p. 7]. Besides exact methods, heuristic optimization methods exist. They “describes a class of procedures for finding acceptable solutions to a variety of difficult decision problems, that is, procedures for searching for the best solutions to optimization problems” [LM2013, p. 695]. Through a limitation of the size of the solution space, heuristic methods use a defined search-strategy to allow the search for a relatively good solution depending on the available computing time [PH2010, p. 401]. Heuristic methods are usually problem-oriented and do not guarantee to find the mathematical optimum. Additionally, they do not give an estimate of how far the solution found is from the actual optimum, but they use the known properties to quickly generate good solutions [SM2009, p. 13]. Heuristic optimization methods can be split up in optimization procedures specifically tailored to OR problems and metaheuristics which can be used universally to solve multiple optimization problems [So2018, p. 58]. While heuristics exploit a specific aspect of a problem and only apply to this aspect14 , metaheuristics are general exist, which will not be further discussed here for reasons of space. For this work, the differentiation of exact and heuristic methods will be made and within the heuristic methods, the classification of trajectory-based and population-based algorithms will be discussed shortly. 14 „For example, when solving a linear programming problem by the simplex algorithm, a heuristic is often used for choosing so-called entering and leaving variables“ [CT2018, p. 22].

28

2

Simulation-Based Optimization

problem-independent principles and schemes for the development and control of heuristic processes which often reproduce natural processes, such as genetic algorithms (GA), simulated annealing (SA), or ant colonialization [SG2013, p. 960; SM2009, p. 13; CT2018, p. 22]. To do so, the individual metaheuristics have adjustable parameters, that allow the search process to be adapted to a particular class of problems, requiring a suitable calibration of the method [Ra2008, p. 61]. Metaheuristics are split up in trajectory-based and population-based algorithms. While trajectory-based algorithms start with one possible solution and improve this successively, population-based algorithms develop several solutions simultaneously within each iteration and improve them in parallel [So2018, p. 58; Go2015, p. 93]. Population-based algorithms are subdivided into evolution-based and swarm-based algorithms. Evolution-based methods are inspired by the laws of natural evolution, e.g., GAs emulate the Darwinian evolution theory. Swarmbased algorithms reflect the collective behavior manifested by animals, to mention the Particle Swarm Optimization (PSO) Algorithm, depicting the behavior of birds or the Ant Colonialization Algorithm, the Firefly Algorithm or the Bee Colony Algorithm, imitating the social collective behavior of insects. Figure 2.8 provides an overview of existing optimization methods according to the classification previously described. Single algorithms will not be discussed any further for reasons of space. The interested reader is referred to more extended reference literature15 for deeper professional involvement. The algorithms being relevant for this work are discussed in more detail in section 2.3.3 in the context of their use in different optimization software solutions.

2.3

Combination of Simulation and Optimization Methods

Following Fu, “the two most powerful operations research/management science (OR/MS) techniques are simulation and optimization” [Fu2015, p. V]. The incentive to couple simulation and optimization can be considered from two 15 While

Cavazzuti [Ca2013, pp. 77–102] and Suhl and Mellouli [SM2009, pp. 33–134] provide a good overview of the exact optimization methods, a detailed description is given by Bhatti [Bh2000] and Marti, Pardalos and Resende [MPR2018]. Yang [Ya2018, pp. 125– 194], Domschke et al. [Do+2015] and Michalewicz and Fogel [MF2004, pp. 35–111] present background information on the mentioned heuristics of OR. Chopard and Tomassini give an accessible introduction to metaheuristics [CT2018], additionally, Shaheen et. al provide an excellent overview on latest metaheuristics in [Sh+2018, pp. 215–231], and Siarry conveys very extensive and deep explanations of the individual trajectory- and populationbased metaheuristics [Si2016].

2.3 Combination of Simulation and Optimization Methods

29

Figure 2.8 Overview of optimization methods. Following [Sh+2018, p. 218; So2018, p. 56]

perspectives. “From the simulation perspective, it is motivated by the desire to compare the effects of different decision variables on the output of a complex simulation model; from the optimization perspective, one might be interested in accounting for randomness in a deterministic model to better capture the real-life system being modeled” [JH2015, p. 1781]. The combination of simulation and optimization is often used when it comes to optimization tasks of complex systems for which the definition of an analytical optimization model cannot be developed with justifiable efforts and their use seems impractical due to simplifying assumptions that distort the core of the actual problem [Do+2015, p. 233]. This chapter briefly summarizes how simulation and optimization methods can generally be used in combination and which objectives are followed by

30

2

Simulation-Based Optimization

simulation-based optimization16 , as well as challenges occurring when combining both techniques. The section closes with a summary on optimization packages interfaced with simulation software.

2.3.1

Coupling and Interaction Modes of Simulation and Optimization

As described in section 2.1, the process of incrementally and interactivly varying model parameters and comparing the system’s behavior against a starting situation is referred to as simulation. Simulation supports the analysis of complex system relationships, illustrates the effects of different environments on a system, enables the testing of options in a model before process changes are brought into the real system, and allows the analysis of dynamic processes in case of time adjustments [Bi+2004, p. 14]. The complexity of the optimization problems often inhibits a solution by means of exact procedures. Therefore, heuristic optimization methods are frequently applied, based on the combination of metaheuristics with simulation models [VS2010, p. 93]. There is no common proceeding in industry for modeling, simulation and optimization coupling so far [VD2016a, p. 4]. Therefore, the coupling of simulation and optimization is not limited to one predominant method. It can be classified according to different criteria. Decisive for the type of coupling are the interdependence of the simulation and optimization procedures used, the relation of subordinate or superordinate order as well as the temporal sequence of the calculations [MK2011, p. 42; VD2016a, p. 3]. Generally, two types of hierarchical and two types of sequential coupling can be distinguished. Respectively, this results in four coupling options for simulation and optimization, which are shown in Figure 2.9. The sequential coupling requires the complete execution of each method. The results of the execution phase of the first method form the input variables for the execution of the following method. Besides the definition of the input variables for the second method, the execution of phases is working independently [VD2016a, p. 3]. If the simulation results are used as starting values for the optimization, there is usually a fully formulated optimization problem existing besides the simulation model which includes unavailable parameters. These parameters are determined during the simulation run and are made available for the optimization (Figure 2.9, 16 The terms “simulation-based optimization”, “simulation optimization”, and “optimization via simulation” are not clearly differentiated in literature. This book consistently uses the term “simulation-based optimization”, except for direct citations of reference literature.

2.3 Combination of Simulation and Optimization Methods

31

sequential coupling case a). In the case of upstream optimization (Figure 2.9, sequential coupling case b), the simulation serves to check the feasibility of the proposed solution. Often, it is not possible to formulate the causal conditions of a complex system completely analytical in form of conditions and restrictions. To analyze and optimize this system, a simple optimization model is created, which allows the calculation of an optimum, but which does not cover all facets of reality [MK2011, p. 44; VD2016a, p. 3]. With the help of the subsequent simulation, which depicts causal relationships by depicting relevant behavioral rules, the optimization solution can be tested for feasibility.

Sequential Coupling b)

a)

Simulation

Optimization

Optimization

Simulation

Hierarchical Coupling c)

d)

Simulation

Optimization Optimization

Simulation

Figure 2.9 Different cases of sequential and hierarchical coupling of simulation and optimization [VD2016a, pp. 3–4]

While for the sequential coupling the results of one method are available before the following one, there is a dominant method in the hierarchical architecture that controls the other method. The dominant method calls the other method as a subcomponent during its execution (Figure 2.9, scenarios c and d). In a simulation experiment, decision points may exist, at which a solution must be selected from alternatives to determine the further course of the experiment. Optimization methods are then used to make this selection of an alternative. The simulation model transfers the current state as an input parameter to the optimization model, which returns a value that serves as a decision parameter for the simulation model to determine the further progress of the simulation run [VD2016a, p. 4]. If the optimization is the dominant method, the simulation is started by the optimization. The simulation then forms the basis for an assessment of the dynamic behavior of the depicted system. The simulation is used to calculate the objective function value, the optimization procedure represents the alternative search [MK2011,

32

2

Simulation-Based Optimization

p. 43]. In literature, most authors refer to the coupling form shown in case d) as the classical simulation-based optimization or simulation optimization. Therefore, this work is as well based on this type of coupling of simulation and optimization when talking about simulation-based optimization. In addition to sequential and hierarchical coupling, another way of categorizing the mode of coupling is to look at how the interaction between optimization and simulation is implemented [VD2016a, p. 4]. The method’s interaction mode can be totally independent, meaning the two software tools run independently and the data exchange between them is done manually. Alternatively, the interaction mode can be based on a common data exchange format. The tools still act independently without any communication during runtime but the data exchange after every completed run is executed automatically. A third option of interaction is the inclusion of the other method respectively by a software-based coordination mechanism, for example using a communication or control platform to parameterize, control, and synchronize bidirectional interactions simulation and optimization. The fourth implementation level is the direct communication of simulation and optimization tools as well as the mutual coordination through builtin interfaces. [VD2016a, p. 4]. “Researchers have proposed and developed many different methods that attempt to optimize a simulation by searching through the space of possible input-factor combinations […] with the results from simulating earlier configurations being used to suggest promising new directions to […] better system performance” [La2007, p. 658]. The commercial and freely available software systems already provide optimization components that can efficiently support this search process. However, the concrete implementation of the interfaces between simulation and optimization as well as the application of tools and the interpretation of the results remain mainly the tasks of the user [VD2016a, p. 5].

2.3.2

Objectives and Challenges of Simulation-based Optimization

The simulation-based optimization has the objective to improve a process by checking whether a change in values of variables has an influence on the process and which process inputs are most influential on the process outputs of interest. The basic idea behind this approach is to find an optimal solution by using multiple replications simulating different system configurations [Fu2013, p. 1418]. The simulation is started by the optimization, produces the result data, and forms

2.3 Combination of Simulation and Optimization Methods

33

the basis for an assessment of the dynamic behavior of the mapped production system. The simulation starts with an initial parameter setting. On the basis of these initial input data, the simulation model is executed for several iterations and depending on existing stochastics in the model also with a defined number of replications. The results are passed back to the optimization tool to generate additional, ideally better parameter configurations based on the optimization algorithm. As shown in Figure 2.10, this process is repeated until the termination criterion is reached [La2007, p. 658].

Optimization Package

Simulation Model

Start

is stopping rule satisfied?

No

Yes

Specify (additional) system configuration

Report solution

Simulate specified system configuration

Stop

Report Simulation Results (Objective-Function Values)

Figure 2.10 Interactions between optimization and simulation model [La2007, p. 659]

For the simulation-based optimization, the objective function as well as the constraints can either be linear or nonlinear, while the decision variables can be continuous or discrete by nature. In most cases, the objective function f and/or the constraint functions g include randomness, which leads to the fact, that they cannot be evaluated exactly [JH2015, p. 1781; An1998, p. 308; Fu2015, p. 2]. The objective function f and the constraints  may be written as f (x) = E f (x, ξ ), and  = {x : E g(x, ξ ) ≥ 0}

(2.5)

34

with

2

ξ x f (x, ξ ) g(x, ξ )

Simulation-Based Optimization

randomness of the system decision variables output of the simulation for the objective for one replication constraint function for one replication

While the optimization of systems with many variables is complex anyway, simulation-based optimization adds the complexity of dealing with estimates for f (x) for any given x [JH2015, p. 1783; Fu2013, p. 1418]. It is hardly possible to evaluate the performance of a particular experiment design exactly. “Because we have estimates, it is not possible to conclude with assurance that one design is better than another, and this uncertainty frustrates optimization algorithms that try to move in improving directions. In principle, one can eliminate this complication by making so many replications, or such long runs, at each design point that the performance estimate has essentially no variance” [Ba+2005, p. 412]. This is highly time and capacity consuming, as the execution time of a simulation is highly dependent on the number of system configurations that have to be tested [La2007, p. 659]. When trying to come up with algorithms that reliably identify high-quality solutions, simulation-based optimization is forced to make compromises such as guaranteeing a prespecified probability of correct selections, ensuring asymptotic convergence to the optimal solution, avoiding algorithms that can be misled by sampling variability or the use of robust heuristics that include randomness as a part of their search strategy as they are less sensitive to sampling variability [Ba+2005, p. 412]. Despite the mentioned problems, the simulation-based optimization also brings advantages. Since the simulation-based optimization is not subject to specific restrictions regarding the form of the objective function, any number of targets can be specified. Additionally, due to the so-called “any-time characteristic”, the simulation-based optimization algorithms, which are generally iterative in nature, determine a result, even in the case of insufficient computation time and premature termination of the calculation run as the best result detected up to this time is used [Vö+2003, p. 37]. Further, the consideration of important constraints in the underlying simulation model of the simulation-based optimization is a common tool to depict special process conditions, which is—in most cases—easier than the mathematical formulation of systems of inequations [Vö+2003, p. 38]. Based on the availability of fast computers and improved optimization heuristics, most simulation software vendors have nowadays integrated optimization packages in their simulation software [La2007, p. 658]. According to Bayraskan in Fu et al. [Fu+2014, p. 3698] “recent developments in the software side are

2.3 Combination of Simulation and Optimization Methods

35

encouraging, with implementations of specific or general algorithms. However, for the success of simulation optimization in practice, software that (i) connects data to models, (ii) simulates and optimizes models using effective solution algorithms, and (iii) provides the users with ways to assess the quality of the obtained solutions need to be further developed.” The following section 2.3.3 gives a short overview of the state of the art of software packages including both, simulation and optimization.

2.3.3

Optimization Packages Interfaced With Simulation Software

The combination of simulation and optimization software is required in situations, when the modeled system is too complex to use simulation to evaluate the systems performance for each set of possible input parameters. As described already in the previous section, the technique used to optimize the systems performance needs to be robust enough to locate an optimal set of parameters and it is desirable that the used technique is reasonably efficient [An1998, p. 329]. Law has summarized the key requirements that an optimizing package for the integration in a simulation software should meet [La2007, pp. 659–660]. While the quality of the solution and the amount of execution time to get to the solution are seen as the most important features, a dynamic display of information during the execution, the consideration of linear and nonlinear constraints, stopping rules and confidence intervals for expected values of the objective function, as well as the possibility to make more replications for higher variance configurations in order to precisely estimate the objective function should be part of the software. There are several different commercial optimization packages available for simulation software products, such as AutoStat, Evolutionary Optimizer (Extend), RISKOptimizer, OptQuest, SimRunner2, and WITNESS Optimizer17 . Nearly all of them use metaheuristics as for problems with large solution spaces, neither ranking and selection methods nor enumeration are the optimization technique of choice. “Since it becomes difficult to use a variable number of replications, as needed in ranking and selection, with metaheuristics, one usually uses a large, but fixed, pre-determined number of replications (samples) in evaluating the function at any point in the solution space” [Go2015, p. 89]. Metaheuristics are based 17 Details on the commercial software packages as well as on supported simulation software products can be found on the vendor’s websites [Au2019a; Au2019b; Pa2019; Op2019; Pr2019; La2019].

36

2

Simulation-Based Optimization

on numeric function evaluations and can therefore be easily used in combination with simulation, even though they do not guarantee to produce globally optimal solutions [Go2015, p. 90]. While AutoStat, Extend Optimizer, RISKOptimizer, and SimRunner2 use evolution strategies and/or genetic algorithms for their search procedures, OptQuest is based on a combination of scatter search, tabu search as well as neural networks, and WITNESS Optimizer uses simulated annealing and tabu search [La2007, p. 660; FGA2005, p. 88]. However, the literature on the performance of those optimization packages is rather poor [AGA2016, p. 1]. The most discussed optimizers are OptQuest and WITNESS Optimizer18 . OptQuest uses scatter search as its primary search procedure and combines it with tabu search as well as neural networks. Scatter search can be applied to problems with continuous and discrete variables, as well as on one or multiple objective optimization. Scatter search shows high-quality outcomes for hard combinatorial optimization problems and supports the use of additional heuristics to check selected reference points for improved solutions [La2011, pp. 4–5; MCP2018, p. 717]. The algorithm selects the best solutions as a starting point for the next application until a defined number of iterations or one of the three stopping rules is reached. As described by Law, scatter search stops the search process when a user-specified number of configurations is reached, when there is no improvement in the objective function (Automatic Stop), or when the number of non-improving configurations makes up five percent of the user-specified number of configurations [La2007, p. 661]. Tabu search in forms of a modified neighborhood search is included in OptQuest to keep track of relevant solutions in tabu lists and to prohibit the reinvestigation of already evaluated solutions to reach a global optimum [Am+2016, p. 366; Go2015, p. 95]. Neural networks are used as a prediction model to shorten the search process by avoiding the evaluation for reference points with a predictable low-quality value [Es+2011, p. 2364]. OptQuest allows linear and nonlinear constraints (such as space restrictions, workforce allocations or budget limits) on decision variables as well as on output variables and the dependence on a variance estimate for the number of replications for a particular configuration. This provides a higher statistical guarantee that the best simulated system configuration will be returned [La2007, p. 661]. 18 Following, OptQuest and WITNESS Optimizer as well as the used search procedures are described very briefly. The interested reader is referred to Fu, Glover and April [FGA2005], Glover, Laguna and Martí [GLM2000], Law [La2007, pp. 659–666], Hindelsberger and Vidal [HV2000], Eskandari et al. [Es+2011], and Gosavi [Go2015, pp. 71–120 and pp. 343–348] for further reading on OptQuest and WITNESS Optimizer as well as on scatter search, tabu search, and simulated annealing.

2.3 Combination of Simulation and Optimization Methods

37

WITNESS Optimizer uses Adaptive Thermo-Statistical Simulated Annealing (ATSA) as a primary search procedure. ATSA is a combination of simulated annealing and tabu search, which is able to modify its search strategy accordingly by learning from its experience of the problem domain [Es+2011, p. 2365]. The WITNESS Optimizer has three stopping rules. Firstly, it stops after a defined maximum number of configurations has been reached. Secondly, the algorithm is stopped when a number of consecutive configurations for which we see no improvement in the values of the objective function is reached and thirdly, the optimizer stops when all feasible configurations have been run through. WITNESS Optimizer calculates feasible configurations, considering constraints formulated for an optimization problem. Additionally, it includes a mechanism to evaluate “the replication-to-replication variability of the objective function for a particular configuration” [La2007, p. 661]. In addition to the embedded optimization solutions previously described, there are also software solutions with a so-called open architecture, whose core cannot be changed, but which can be extended significantly by the use of toolboxes. Matlab is one example of such a software solution [Be2010, p. 3]. The Matlab toolboxes are understood to be a collection of predefined subprograms to expand the Matlab functionalities19 . They are on the one hand offered by MathWorks itself to supplement the software but are also freely available online in the form of non-commercial toolboxes. MathWorks provides the option of including an Optimization Toolbox™ for finding parameters for minimizing or maximizing objective functions and to solve linear programming (LP), mixed-integer linear programming (MILP), quadratic programming (QP), nonlinear programming (NLP), constrained linear least squares, nonlinear least squares, and nonlinear equations [Th2019a]. Additionally, a global optimization toolbox is available, which allows for the use of any optimization algorithm including pattern search, genetic algorithms, particle swarm optimization, simulated annealing, multi-start optimization, and global search algorithms. The Global Optimization Toolbox can solve optimization problems “where the objective or constraint function is continuous, discontinuous, stochastic, does not possess derivatives, or includes simulations or black-box functions” [Th2019b]. 19 The interested reader is referred to the further literature on the Optimization Toolbox™ at this point, since only a very briefly overview of the functionalities can be given in this work. The MathsWorks, Inc. offers a very detailed manual on the optimization functionalities of the toolbox [Th2019a], in addition there are further explanations in Angermann et al. [An+2017, pp. 241–286] and Benker [Be2010, pp. 213–226]. Ait El Cadi, Gharbi and Artiba provide a comprehensive comparison of the functionalities and the performance of OptQuest and the Optimization Toolbox™ [AGA2016].

38

2

Simulation-Based Optimization

Summarizing, various software solutions that include optimization packages interfaced with simulation software are available on the market. As companies need to continuously improve their processes in order to remain competitively on the market, the use of simulation software in combination with integrated optimization possibilities represents a tool to model, assess and enhance dynamic production processes in manufacturing. The simulation-based optimization therefore seems to be a possible solution to model and optimize the energy consumption in the production systems and to constitute a decisional support system for production planning and control in order to achieve an optimum use energy in production processes.

3

Energy

Energy is essential for human development and energy systems are a crucial entry point for addressing the most pressing global challenges of the 21st century, including sustainable economic and social development, poverty eradication, adequate food production and food security, health for all, climate protection, conservation of ecosystems, peace and security. [Jo+2012, p. 4]

Nowadays, producing companies are facing various economic and social changes. Environmental driven topics, such as global warming, CO2 emissions and resource depletion have become strategically relevant aspects. The sustained use of energy and other resources has become a basic requirement for a company to competitively perform on the market. Chapter 3 briefly describes the development of the global energy market (section 3.1), relevant terms and definitions in the context of energy in section 3.2 and energy efficiency (section 3.2.2) as well as the meaning of energy consumption in manufacturing systems in section 3.2.3. The correlations in the area of energy costs are explained in section 3.3, followed by a depiction of energy saving potentials in production plants in section 3.4. Reference to advanced literature will be made in the appropriate places of the text.

© The Author(s), under exclusive license to Springer Fachmedien Wiesbaden GmbH, part of Springer Nature 2021 A. C. Römer, Simulation-based Optimization of Energy Efficiency in Production, Forschung zur Digitalisierung der Wirtschaft | Advanced Studies in Business Digitization, https://doi.org/10.1007/978-3-658-32971-6_3

39

40

3.1

3

Energy

Development of the Global Energy Market

The development of the global energy market is directly connected to changes in the world’s population structure, as well as the economic growth and geopolitical risks. The trend of the world’s energy consumption has enormously increased and changed considering the shares of different energy types (Figure 3.1). The Cambridge University’s International Institute for Applied Systems has published the Global Energy Assessment (GEA) including prognoses for the global energy consumption development until 2050. The global economy is expected to develop at an average growing rate of 3,4% per year for the next 20 years [BP2017, p. 6]. The globalization of the international markets is a promising future growth perspective especially for national economies in the areas of the emerging countries. The global energy needs are expected to expand by 30% between today and the year 2040 [In2017b; BP2017, p. 11; Jo+2012, p. 9]. “The largest contribution to demand growth—almost 30% [of this]—comes from India, whose share of global energy use rises to 11% by 2040” [In2017a, p. 1]. Emerging countries in Asia will account for two-thirds of the world’s energy growth, followed by Africa, the Middle East and Latin America (Figure 3.2). The industry sector accounts for about 30% of the global final energy use provided that cement, iron and steel, chemicals, aluminum and paper make up more than half of the industry’s global demand [Jo+2012, p. 516]. Whereas the energy intensity of industrialized countries has shown a steady decline of energy consumption due to energy efficiency improvements, emerging countries as China and India have high growth rates in the production of energy intensive materials [Jo+2012, p. 48]. Overall the demand for energy of the industry sector steadily increased ever since and following the BP Energy Outlook, the industry sector will remain the largest market for final energy consumption, but the demand growth will, driven by efficiency improvements, slow down from 2,5% to about 1,5% per year (Figure 3.3) [BP2017, p. 17; Po2011, p. 131]. The development of the worlds need for energy clearly shows the strong need of appropriate methods to optimize the energy efficiency in producing companies to decrease energy costs, particularly against the background of continuously rising energy prices1 (Figure 3.4).

1 Current

price data on energy price trends for Europe can be seen in the price publication of the Federal Statistical Office DESTATIS published monthly.

3.2 The Concept of Energy

41

Figure 3.1 Development of the global primary energy consumption to 2008 and three GEA pathways (“GEA explored 60 alternative pathways of energy transformations toward a sustainable future that simultaneously satisfy all its normative social and environmental goals of continued economic development, universal access to modern energy carriers, climate and environment protection, improved human health, and higher energy security (see Section TS-2.6). The pathways were divided into three different groups, called GEA-Supply, GEA-Mix, and GEA-Efficiency, representing three alternative evolutions of the energy system toward sustainable futures. The pathways within each group portray multiple sensitivity analyses about the ‘robustness’ of the three different approaches in mastering the transformational changes needed to reach more sustainable futures.” [Je2012, p. 71].) to 2030 and 2050 [Jo+2012, p. 9]

3.2

The Concept of Energy

3.2.1

Work, Energy and Power

“Energy is one of the most basic scientific concepts and is a unifying principle that has applications in all scientific disciplines” [My2006, p. 68]. Various forms of energy exist, generally they are classified according to their source, e.g., nuclear, chemical, solar, geothermal or electrical energy. Energy can be stored within a system in three general forms the inner energy, the potential energy and the kinetic energy. While potential energy is the stored energy, resulting from the position of a system of objects that exert forces on one another, kinetic energy is understood as the energy of motion [My2006, pp. 68–69; MVG2010, pp. 28–29]. When a system changes its energetic state from one to another, this transition is described

42

3

Energy

Figure 3.2 Change in primary energy demand, 2016–2040 (Mtoe) [In2017b]

using the process variable work [CHR2002, p. 235]. Work W is defined as force F applied on an object causing it to move a distance s in the direction of the applied force [MVG2010, p. 27]: s2 W =

F(s)ds

(3.1)

s1

The SI-unit for W as well as for E is Joule [J]. The usage of this standardized unit underlines the equivalence of the variables energy and work [Mü+2009, p. 68; VD2014b, p. 8]. 1J =1N ∗ m = 1

kg ∗ m 2 = 1W ∗s s2

(3.2)

The work of a process related to a period of time is the power P measured in kilowatt [CHR2002, p. 267]: P=

W [kW h] = [kW ] = t [h]

(3.3)

3.2 The Concept of Energy

43

Figure 3.3 Global growth in energy consumption by sector [BP2017, p. 16]

The units kWh and MWh are used more often in the context of producing companies as well as in workaday life [Mü+2009, p. 68]. The common calculation of the unit conversions is: 1 kW h = 3600 J = 0,001 M W h To understand the basic correlations between the different forms of energy and the energy usage in manufacturing systems, the relevant principals of thermodynamics will be explained subsequently. Following the first law of thermodynamics (law of conversation of energy) energy in a closed system can be transformed from one to another, but it can neither be created nor get lost [He2012, p. 26].

44

3

Energy

Figure 3.4 Producer price indices for electricity on the example of Germany (Figure 3.2 is using the energy price of 2010 as a reference year, taking the 2010 year price value as 100%.) [St2017, p. 15]

The total energy E of a system is the sum of the inner energy U, the kinetical energy Ekin and the potential energy Epot [AK2014, p. 2]. E = U + E kin + E pot

(3.4)

m ∗ ν2 2

(3.5)

with E kin =

and E pot = m ∗ g ∗ h

(3.6)

Nonetheless, in practice it is often spoken about energy loss, energy consumption and energy wasting. In a strictly physical sense and according to the second law of thermodynamics, energy cannot decrease for an isolated system but with regard to its intended purpose, the usable share is reduced. Especially when transforming thermal energy into electrical energy, the degree of efficiency η is low and energy conversions are not always completely reversible [He2012, p. 18]. Generally, energy consists of two parts, exergy (EE ) and anergy (EA ) including the option that one of them can be zero.

3.2 The Concept of Energy

45

E = EE + EA

(3.7)

EE and EA can both be described as state variables. Whereas exergy is the maximum useful work that can be taken from a reversible process, anergy is the share, that cannot cause work (Figure 3.5). “The concept of exergy allows for the comparison of energies given in different forms such as electrical energy, heat provided at different temperatures, or energy stored in chemical bonds. The quantity exergy is only well defined when a reference state, the ambient state, is given” [KS2015, p. 11]. This state defines the pressure ρ and the temperature T, as well as the concentration of chemicals in the atmosphere. The conversion process of primary energies Eprim into useable energy Eend like electric power involves a degree of energy loss.

Figure 3.5 Energy conversions [KS2015, p. 12]

Using the first-law efficiency η1 , the ‘energy loss’ in the transformation process can be defined as the ratio between final and primary energy: η1 = with

E end ; 0 < η1 < 1 E prim

E prim E end

primary energy useable energy

(3.8)

46

3

Energy

If the energy conversion system contains several conversion steps, the degrees of efficiency are multiplied. Therefore, the overall efficiency of a system ηsystem might often be very small. ηsystem = η1 ∗ η2 ∗ · · · ∗ ηn =

n 

ηi ; 0 < ηi < 1

(3.9)

i=1

To understand the energy efficiency, it is required to look at the number of conversions during the energy production process (Figure 3.6). The energy stored in natural energy resources such as fossil fuels is named primary energy2 [VD2014b, p. 11]. It needs to be refined and cannot be used directly by end users. The refining process creates the so called secondary energy (e.g., diesel, fuel oil, electric power, …) as well as conversion losses in form of unused waste heat and substance losses [He2012, p. 20; Er2014, p. 45]. The energy delivered to the actual consumer is called final energy3 . This conversion step again causes energy losses when transmitting, distributing, and wiring the energy. The useful or net energy that is consumed by the actual machine is again lossy, as for example waste heat is produced by the consuming device. “The overall net energy output is about ½ of the final energy consumption, leading to a rough estimate of the ratios of primary energy to final to net of 3:2:1” [KS2015, p. 3]. Depending on the level of refinement, the efficiency of the transformation process can be only 5%, e.g., when transforming primary energy into compressed air [He2012, p. 20]. Thus, the energy conversion chain for energy used in manufacturing systems should be as short as possible, to be most efficient [Mü2009, p. 77]. Erlach has summarized the main types of energy losses, that occur along the energy conversion process [Er2014, p. 46]: • over sizing (power/performance of machines) • use of standby mode • distribution losses during transport and storage 2 Besides fossil fuels, nuclear fuels and renewable energies are primary energy sources as well.

The efficiency for the conversion of primary energy into electrical energy is assumed to be η = 0.33 for nuclear power and η = 1.0 for renewable energy sources [KS2015, p. 2]. As this thesis aims on the optimization of the final energy in manufacturing facilities, a differentiation for the efficiency calculation depending on the efficiency of the different primary energy sources will not be considered. 3 The VDI Guideline 4661 contains detailed descriptions on the balancing process for energy balances of technical systems as well as for economic areas [VD2014b, p. 29 ff.].

3.2 The Concept of Energy

47

Figure 3.6 Energy balance. Adapted from [Pe2010b, p. 22]

• • • •

dissipation due to missing energy recovery operation mode of machines by workers efficiency losses/ conversion losses losses due to wrong energy acquisition type.

One of the most important forms of energy used in industrial production is electrical energy. It is a high-quality exergy that can be converted into almost all other forms of energy. The working capacity of electrical energy is based on moving electric charges and their electric and magnetic fields. While negatively charged electrons and anions repeal each other due to the same charge, they attract positively charged protons and cations. Electric charges generate electric fields, moving electric charges additionally generate magnetic fields [Mü+2009, p. 83]. The electrical work W is determined, calculating the integral over the current power P. Thus, the current value of the power is always time-dependent [Bö2013, p. 769]: t2 W =

P dt

(3.10)

t1

The current power in the direct current circuit is defined as the product of voltage U and current I [Bö2013, p. 310]:

48

3

P =U×I with

U I

Energy

(3.11)

voltage in [V] electric current in [A]

Therefore, the electrical work can be written as: t2 U × I dt

W =

(3.12)

t1

W = U × I × t The power in the alternating current circuit can be calculated in the same way as the power in the direct current circuit for values that coincide in time, since current and voltage have the same phase position at the effective resistance [Bö2013, p. 311]. Thus, the current power p is: p =u×i

(3.13)

According to DIN 40110, the arithmetic mean values of current and voltage are defined as [DIN2002]:

i=

1 T

T i(t) dt

(3.14)

0

u=

1 T

T u(t) dt 0

If the rectified value i or u is zero, an alternating current or an alternating voltage is present [HGM2013, p. 52]. An alternating current i(t) causes a timedependent power at an ohmic resistance. The current intensity I of a direct current that produces the same power is called the effective value of the alternating current [He+2018, p. 69]. Taking into account the current value of u(t) and i(t), their connection to the effective values U and I, as well as the angular frequency and

3.2 The Concept of Energy

49

the zero-phase angle, it can be derived4 that the arithmetic mean of the current −

power p represents the effective power Pe f f [He+2018, p. 73]:

Pe f f

−

1 = p= T

T u × i dt

(3.15)

0

The measurement of the electricity consumption of the production machines is in fact the measurement of the effective power. The explanation of the terms apparent power and reactive power is omitted as these are not considered in the course of the work. The interested reader is referred to [HMW2012, pp. G25–G27].

3.2.2

Energy Efficiency

As already stated in the previous section (equation 3.8) efficiency can be defined as the ratio of an output to the necessary input of a system. Transferred to the energy industry understanding, energy efficiency in production means keeping the energy used to produce goods as low as possible. “An exergy analysis (second law of thermodynamics) reveals that the overall global industry efficiency is only 30%” [Jo+2012, p. 516]. The more efficient final energy is used, the more it reduces the amount of primary energy required. Energy efficiency in manufacturing systems can be divided into several different levels, starting with the process level efficiency (Figure 3.7). The process level efficiency considers energy losses by the physical mechanisms that are involved in the process itself. On the next higher level, the machine efficiency level, the energy use of the machine for the process itself as well as for peripheral aspects of the process are being looked at. “Finally, one has to consider the production line level and the entire factory level, where energy efficiency is mainly a function of production planning” [Fy+2013, p. 629]. As the percentage of energy spent on the process itself is very small compared to the share spent on machine level to perform processes, for coolant pumps or lubrication supply, the optimization of the process level energy use will not be the main focus of this book. 4A

detailed derivation is not provided here. The interested reader is referred to the detailed descriptions in the literature, especially in Hering et. al [He+2018, pp. 68–74], Hering, Gutekunst and Martin [HGM2013, pp. 52–57] and Niedrig and Sternberg [NS2012, pp. B166–B170].

50

3

Figure 3.7 Energy efficiency levels [Fy+2013, p. 629]

Energy

factory level line level

machine level

process level

This thesis will concentrate on the optimization of the energy efficiency on machine, line and factory level, as “the energy resources, optimization potential of the overall process or the total system can be exploited only through a holistic view of the complex interactions of individual processes and structures of a factory” [Fy+2013, p. 632].

3.2.3

Energy Consumption of Manufacturing Processes and Systems

Industrial production5 is generally defined as the creation of material and immaterial outputs intended for sale by transforming inputs via special technical procedures into goods of higher value [Zä2001, p. 1; DS2010, p. 4]. Production processes require a substantial amount of input factors in the form of raw materials and energy and the use of technical machines for their transformation processes. The industrial sector is responsible for one third of the worlds final energy consumption. The energy-intensive industries consume nearly half of this third. Besides the importance of energy for companies in the industrial sector as a competitive factor due to changing energy prices, tightened regulations on greenhouse gas (GHG) emissions as well as an expected shortage on primary fossil fuels, there has been a dramatic increase in dynamism and complexity in the energy sector in recent years [Po2011, p. 136].

5 In

this thesis, the terms ‘manufacturing’ and ‘production’ are used synonymously and both refer to the given definition of industrial production unless otherwise noted. The definitions may vary in case cited authors use the terms in a special context or with a differing definition from the one used here.

3.2 The Concept of Energy

51

The use of energy in manufacturing systems starts with the availability of the final energy at the system boarders of the company. The energy flow through the company is divided into different segments, which characterize the conversion steps of the industrial energy system. It includes all assistive devices that are required for the energy import, the energy conversion and distribution, the energy application, as well as for the energy recovery6 and energy losses (Figure 3.8). The energy import includes the selection and optimization of the delivered final energy products to the company. The provision of net energy can involve transformation processes as heat converting, electricity or compressed air converting as well as the further distribution within the company to the point of use [Po2011, p. 137]. The main consumers of energy in a manufacturing company can be divided into electronic, steam and hot water using applications. Main applications using electric energy are electronic power drives of production machines, compressed air pressure supply units, or the room lighting. Producing companies often create waste heat during their production processes which can be recovered and reused [Ha2013, p. 15].

Figure 3.8 Energetic correlations in manufacturing companies [Ha2013, p. 13] 6 The

process of energy recovery describes the use of energy that would have otherwise been characterized as a loss for covering, e.g., a heating demand [RW2008, p. 15].

52

3

Energy

The general understanding of the energetic correlations in manufacturing companies is necessary to be able to model, simulate, and optimize the energy consumption of a production system. The focus of this thesis is centered on the optimization of the use of electric energy in the main production processes. Peripheral processes, which are not directly linked to the production, e.g., heating, lighting, ventilation, and climate control of offices, are not considered in this book7 . Besides the general overview of energetic correlations in production, the consideration of the energy consumption behavior of production machines is required. The consumption behavior of machines in industry is usually not constant but highly dynamical depending on the current operating state of the machine. Machine processes are built of several energy consuming process steps causing a specific electrical load profile when the machine is in an operating state. By classifying different operational machine states, the definition of load profiles becomes possible. Generally, the following operating states can be found in literature [We2010, p. 62; Be+2011, p. 1063; Th2012, p. 21]: • OFF—no energy consumption, the main switch is off • WARMUP—energy use for start-up process after machine has been switched off or remained in standby, often causes a peak demand • IDLE—relatively constant energy use after completed start-up • SETUP—positioning and loading before actual processing • PROCESSING/PRODUCING—actual production process is running • STANDBY—machine has a reduced consumption rate • FAILURE—the failure state generally also consumes energy Operational machine states can be divided into time-constant and time-variable machine states. While the machine undergoes a technically necessary and timed start-up plan at time-constant conditions, such as a machine startup after off-mode, the retention time and the energy requirement of a machine in a time-variable machine state, such as the producing state, are dependent on the production task [We2010, p. 61]. While the time-constant states are usually technically necessary, the time-varying states can be differentiated into value-adding and non-valueadding machine states (Figure 3.9).

7 The reason for this lies in the height of the actual share of total energy consumption, which is

usually only between two and 20 percent depending on the production machine. The majority is thus attributable to the production machine itself [Mü+2009, p. 26].

3.2 The Concept of Energy

53

operational machine state

time-constant states

technically relevant

value-adding

warm up state

producing state

setup state

time aspect

time-variable states

non-value-adding

optimization aspect

idle state stand-by state

Figure 3.9 Classification of machine states according to time and optimization aspects

As non-value-adding machine states, this work refers to conditions that can be reduced to a minimum or even be avoided, by an efficient production scheduling. Unnecessary lingering of machines in standby mode or reaching the idle state significantly before the start of production are just two examples for states that consume energy but do not add value to the finished product and should therefore be eliminated. The power consumption profiles of the production machines can thus be represented as a juxta positioning of different machine states (Figure 3.10). In principle, there are various possibilities to model the energy consumption behavior of machines for simulation and optimization purposes. The energy consumption can be represented in the form of mathematical functions or value tables. Table functions can be created only when the required measuring equipment is installed at each production machine. The use of measuring equipment allows the collection of exact energy consumption data per machine, generally the effective power8 for definable time intervals is documented. The data is then used in forms of table functions in the simulation software for a realistic depiction of the consumption behavior of production machines. A third option is the use of state-based values which are obtained by averaging the energy consumption values over the duration of a machine state. The calculated mean value is used in the simulation as a consumption value for the entire duration of the machine state. Thus, a realistic representation of energy consumption behavior is lost, but at least allows for the consideration of machines in a simulation study whose exact consumption behavior cannot be determined for technical or organizational reasons. 8 The

explanation of the correlations and the derivation from the effective power is given in section 3.2.1.

54

3

Energy

Figure 3.10 Measured consumption profile of a production profile of a machine

While the use of exact data allows the representation of extreme values, such as consumption peaks, those extrema are completely lost due to mean value formation (Figure 3.11). Another difference is the underlying simulation technique for the case of the energy data simulation. While the status-based mean values can be represented by using DES, the use of continuous simulation techniques such as SD is required for the simulation of exact power consumption profiles. For a mathematical calculation, the energy consumption of a production machine is generally split up into a fixed and a variable share. The constant consumption rate of a machine includes the energy which is used for all units that enable the operating state (e.g., control units, coolers, …). While the machine is in off-state, the electricity demand can be assumed to be zero. Depending on the complexity of the machine, a start-up process from off or standby mode into idle or producing may be required. This warmup requires a defined amount of time, twar mup , and has a machine and state-specific power consumption profile [Be2017, p. 135]. After finishing the warmup process, the machine can either go in producing or idle state. Once the manufacturing machine goes from idle into

3.2 The Concept of Energy

55

Figure 3.11 Representation of exact consumption profiles and consumption profiles of a machine determined by mean value formation [RRS2018, p. 76]

producing state, the additional electrical power requirements P are proportional to the quantity of input material being processed [Be2017, p. 135; GDT2006, p. 624]: P = PI DL E + k ∗ v˙ with

PI DL E k v˙

(3.16)

power required in the “idle” state in kW machine specific constant in kJ/cm3 rate of material processing in cm3 /sec

While PI DL E comes from the equipment required to support the production process, k comes from the process physics9 .

9 „For example, for a cutting tool, P

0 comes from the coolant pump, hydraulic pump, computer console and other idling equipment, while k is the specific cutting energy which is closely related to the work piece hardness and the specifics of the cutting mechanics. For a thermal process, Po comes from the power required to maintain the furnace at the proper temperature, while k is related to the incremental heat required to raise the temperature of a unit of product“ [GDT2006, p. 624].

56

3

Energy

In combination with the definition of work as the power consumption over a certain period of time, the following equation for the energy consumption for an entire production can be defined with the state-based power demand P, a cycle time t and the production volume n [Mü+2009, p. 137]:    W = P ∗ tcycle ∗ n + ttotal − n ∗ tcycle ∗ PI DL E

(3.17)

The equation depicts the significant impact of the idle-state on the total energy consumption, underlining the relevance of non-productive machine times for the total energy consumption of a production. As the mathematical calculation of energy consumption values is quite complex, it is more common to work with measured electrical load profiles of production machines when it comes to improvements of production systems in the context of energy efficiency. For the load profile analysis, power consumption profiles and—parallel to the measuring period—machine scheduling data, such as exact product arrivals at each machine and produced quantities are required. The assignment of the measured, time-based energy profiles is done extracting profile sections corresponding to the operative machine states. The procedure of the matching of machine state with its energy data will be described in detail in the practical part of this thesis in section 5.6 and chapter 6 (Figure 3.12).

definition of the coherence / transformation logic

energy load data acquisition, measuring

energy consumption (actual / per period)

visualization and interpretative analysis

coherence to energy state

operational state

material flow

Figure 3.12 Coherence of operational machine and energy state (adapted from [WKD2012, p. 64]).

3.3 Energy Costs

57

Often, production specific data is not available due to missing manufacturing execution system (MES) data. Teiwes et al. developed a load profile clustering (LPC) algorithm using only electrical load profiles and processing times to identify different load levels such as processing and waiting times of machines [Te+2018, p. 273]. The five-step approach is based upon a k-means clustering algorithm and aims at a simplification of energy allocation (Figure 3.13). First, the electrical consumption profile of the machine is measured. Thereafter, the single data entries are clustered in different clusters according to the machine states using the k-means algorithm in several iterations. The clustering thus allows the distinction of different load levels. After having defined the clusters, the process time intervals are assigned, followed by product identification for the case that several different products are processed at the machine under consideration. In a fifth step, the results are subject to a plausibility check [Te+2018, p. 274]. Thus, the LPC methods allows the extraction of machine state-based energy consumption information from available load profile data even though further process data is not available. However, the interpretation of energy data is still challenging but measurability and transparency are the most important requirements towards the optimization of energy efficiency in manufacturing systems.

3.3

Energy Costs

The calculation of energy costs in producing companies is highly complex. As the share of energy costs in total production costs differs depending on the industry sector, the products, as well as the processes, general statements regarding the height of the energy cost share for industrial customers can hardly be made [We2010, p. 21]. Additionally, it has to be considered, that main price elements of electricity are taxes and duties, as well as regulated price components that can neither be influenced nor optimized10 . Therefore, the management of energy consumption and every unused kilowatt hour (kWh) make up the best and costeffective, near-term way of energy efficiency optimization [Ge+2015, p. 82 f.; Jo+2012, p. xvi; DS2008, p. 42]. The time frame of energy consumption can also influence the height of energy costs. A high flexibility regarding the energy supply and the energy consumption 10 A detailed explanation of the cost structure and its components for the German electricity prices as well as an example calculation for an electricity bill for a German industrial company can be found in [DI2017, p. 7 ff.].

58

3

Energy

Figure 3.13 Application of the LPC methodology [Te+2018, p. 274]

time can make the energy purchasing at the spot market11 very attractive. The spot market temporarily offers lower or even negative energy prices for times where an energy overload of the power network exists. In this case, the energy 11 “The spot market is the market for purchasing energy carriers at short notice. In Germany power is currently traded on the spot market for the next day or next days (before weekends and/or public holidays). A major part of spot power trading is carried out via the EEX energy

3.3 Energy Costs

59

consuming customer can make money by using a lot of energy to discharge the power network [Ge+2015, p. 83; Sc2016, p. 320]. As most producing companies don’t have the high flexibility required regarding the time period of electricity consumption, the described case will not be considered any further in this work. The focus for optimizing the energy costs will be on the optimization of the used energy amount and the avoidance of high consumption peaks, as far as the production schedule allows this. The energy pricing for industrial customers generally distinguishes between three main price components, the electrical purchase price per kWh, the network charges and the state regulated components [Gr+2015; Ba+2015, p. 3]. Depending on the size of the organization, companies either obtain their electricity from energy suppliers or they use energy traders to trade on the power exchange. For the second case, purchase contracts generally split up into 80% long-term agreements and 20% spot market purchases [Ba+2015, p. 39]. The purchase price per kilowatt hour (kWh) is always depending on the electricity reference amount, the duration of the electricity use, the voltage level as well as the utilization of special rules and exceptions. As a general rule, it can be said that the single kWh is cheaper the higher the reference amount, the voltage level and the duration of usage are and the lower the consumption peaks turn out to be [DI2017, p. 8]. The relevant billing interval for electricity costs is 15 minutes, the measured energy demand is averaged over this time period and results in 96 measured values for a day [Th2012, p. 24]. As described earlier in this thesis, the power consumption time plays a major role for the energy price. The energy supplier generally differentiates between base and peak times. Due to higher demands during the day, suppliers charge higher prices per kilowatt hour and to utilize their networks during the base times as well, energy consumer pay lower prices during the night (Figure 3.14) [Th2012, p. 25]. Besides the actual electricity demand, the guaranteed electric grid capacity influences the prices. Here “the highest electrical power value for a month could be the base of calculation … [but also] cost calculations which take the three highest values over a whole year into account can be found in industrial practice” [Th2012, p. 25]. Considering the state regulated components, the total energy costs sum up to:

exchange located in Leipzig. In some cases, however, even short-term transactions are conducted bilaterally (“over the counter” or OTC) between companies or between companies and brokers” [VD2007, p. 52].

60

3

Energy

energy coststotal = electricit y costswor k + electricit y costs power + electricit y costsstandar d

with

(3.18)

electricit y costs wor k = electrical wor k ∗ price per kW h electricit y costs power = highest power demand ∗ price per kW electricit y costsstandar d =



(electrical wor k ∗ variablestandar d costs per kW h

+ f i xed standar d costs)

While the fixed and variable standard costs are taxes, fees for network access or specific dues and cannot be negotiated, the prices for the electrical work and the power costs are subject to individual negotiation.

Figure 3.14 Example of electricity composition and sample load profile [Th2012, p. 24]

Figure 3.14 shows a sample cost composition as well as a sample load profile for an industrial customer. An analysis of the main energy consumption times and the energy consumption peaks will allow for the discovery of energy-saving potentials and the improvement of the energy efficiency in manufacturing as described in the following section 3.4.1. For the economic evaluation of energy saving measures, most companies use a calculated mixed price. For this purpose, the energy costs of one year specified on the invoice of the electricity supplier are summed up and divided by the energy amount consumed [Mü+2009, p. 97]. If the energy saving measures aim at the

3.4 Energy Saving Potentials and Energy Efficiency in Manufacturing

61

avoidance of peak loads, this method is not sufficient and requires the use of further performance indicators for the correct economic evaluation of implemented optimizations.

3.4

Energy Saving Potentials and Energy Efficiency in Manufacturing

3.4.1

Energy Saving Potentials in Manufacturing

Energy is a balance sheet item. Energy balances can be used as an effective analytical tool to plan and operate energy efficient factories [Mü+2009, p. 71]. Most companies monitor the common performance targets like production outputs, quality as well as time and cost key indicators for their production units. Energy is very rarely used as a performance indicator. Consequently, most companies do not even know how much energy single machines, production lines or factory areas are consuming. Energy saving potentials therefore remain undetected, as the overall understanding for energy requirements and the knowledge about the energetic behavior of the production are missing [He+2008, p. 12; Ha2013, p. 5]. As described in section 3.2.2, different energy efficiency levels (Figure 3.7 on page 44) can be addressed for optimizations. Energy saving potentials can be identified for each level on both, a technical and an organizational view. While optimizations on the technical site12 include the usage of improved technical devices with a higher degree of efficiency as well as the installation of compensation circuits and setups for the recycling of lost heat, organizational optimizations will involve lower investments as they basically aim on the optimization of the load management [We2010, p. 43]. This type of optimization is aiming at the avoidance of unnecessary consumption and the smoothing of energy requirements of machines [VD2007, p. 58], for example by revising excessive process temperatures, by shutting down continuously operating heaters and by turning off machines 12 In the case studies of this work, it is assumed that all considered machines have the optimum technical condition and are running, technically seen, energy-efficiently. The fact that this ideal case is usually not found in the production is known by the author. The practical use cases in this work include only already operating production lines. For the case of planned new or replacement investments for the machine pool, it is recommended to the practice partner to explicitly include the requirement for energy efficiency in the technical specifications for the plant and, in addition to the investment costs, to also consider the operating costs (energy costs, costs for maintenance, etc.) for the profitability analysis. According to Erlach, this is called life-cycle costing and evaluates the costs of an asset over its entire life cycle [Er2014, p. 54].

62

3

Energy

under no load, as well as on the reduction of energy consumption peaks by shifted machine starts. Table 3.1 contains an overview of the approaches for energy saving measurements in manufacturing considered in the fictional cases as well as in the practical use cases in this thesis, sorted by the energy efficiency levels, their optimization parameters, and the potential effects on the energy efficiency level when changing the parameter. Table 3.1 Energy efficiency levels and their optimization parameters Energy Level

Optimization parameter

Effect on Energy efficiency

Process level

– processing time/speed, – process parameters, – process interdependencies (e.g., cooling unit and process speed)

minor effect, as consumed energy on process level is rather small

(technical) Machine level

– correct dimensioning of machine parts (e.g., engines and drives) – replacement of single machine parts

high impact, as new more efficient technologies have entered the market

(organizational) Machine Level

– timing of operational machine states – decrease of peak demands – planned machine runtimes and shutdowns

major impact, as the behavior of main consumption centers is addressed and influenced

Line level

– machine scheduling (line internal) – line-internal dispatching rules – consumption profiles of idle machines

major impact, as the behavior of main consumption centers is addressed and influenced

Factory level

– electricity control policies – machine scheduling (cross-production line) – dispatching rules (cross-production line)

major impact, as the behavior of main consumption centers is addressed and influenced

The technical optimization parameters on the machine level are included for completeness only and will not be addressed any further in the used cases later in this thesis.

3.4 Energy Saving Potentials and Energy Efficiency in Manufacturing

63

For this work, the following focus arises with regard to realization of optimization potentials: • Energy Consumption Optimization via timing of machine states By determining maximum duration times for non-value-adding machine states (on line and factory level), it is possible to achieve an energy-optimal timetable and necessary recommendations for action with regard to the switching behavior of machines during production interruptions. • Energy Consumption Optimization via scheduling of machine starts The exact calculation of machine starts as a function of the machine states and processing times of previous production machines leads to the avoidance of unnecessarily early availability of downstream production machines. This eliminates the lingering of machines in non-productive but energy-consuming machine states and thus increases the energy efficiency. • Energy Cost Optimization via peak consumption avoidance Early or delayed machine starts as well as timed machine state changes can be used to reduce load peaks in the power consumption profiles and thus have a direct impact on the energy cost component “electricity costs for power demand”. The following section briefly describes how optimization measures can be presented and evaluated using key figures.

3.4.2

Energy Performance Indicators

A key performance indicator (KPI) system is supposed to track a development in a respective field of activity. To better illustrate the development of energy consumption and to track the success of optimization measures, energy performance indicators (EnPI) can be defined [Ge+2015, p. 33; DS2008, p. 150]. Generally, a distinction between absolute values and ratios has to be made. While absolute values (e.g., the total energy consumption per year, peak demand, …) are relevant to follow up on energy supply contracts, ratios (e.g., specific energy use per production volume) can facilitate comparisons and statements about energy efficiency [Au2009, p. 48]. A classic key indicator is the yearly energy consumption of a production, tracked from the actual conditions and used to draw conclusions for the future [Sc2006, p. 42]:

64

3

E year = E 0 year + w Fyear ∗ Fyear with

Energy

(3.19)

E 0 year

yearly energy consumption unrelated to production output

w Fyear Fyear

yearly total energy consumption of a production in kW total production output of a year

To calculate with planned numbers, all components can be adapted, and the equation can be extendet to [Sc2006, p. 42]:     E plan = E 0 year ± E 0 plan + w Fyear ± w F plan ∗ (Fyear ± F plan ) = E0 p + w Fp ∗ Fp with

(3.20)

E plan E 0 plan

planned energy consumption added/omitted production unrelated energy consumption

w F plan F plan

total planned energy consumption of a production in kW adapted production output

Besides the yearly energy consumption in kWh, the yearly energy costs are a relevant key indicator (equation 3.18). The EnPIs Eyear and Eplan can be used to track the overall energy consumption, to draw conclusions from the past developments and to predict the effect of energy consumption adjustments in the context of operational and production changes. As the peak demand is an important factor for the energy pricing of a company, it should be tracked either using the 15-minutes-billing interval values tracked by the energy supplier or using an internal energy consumption metering. “In any energetic consideration or assessment, the specific energy demand is the most important characteristic of machines, installations and processes” [VD2014b, p. 16]. Especially for machines and production lines with a high energy consumption, it is useful to track the specific energy consumption Espez as a EnPI for a defined period of time. The energy consumption of the machine or line is divided by the reference quantity produced [Ha2013, p. 40]: E spez =

E used x

(3.21)

3.4 Energy Saving Potentials and Energy Efficiency in Manufacturing

with

E used x

65

consumed energy in defined period of time quantity produced in defined period of time

In order to be able to carry out a holistic cost-related assessment of the energy efficiency measures, it is necessary to take the traditional production KPIs into account to a certain extent, e.g., when evaluating the costs of load optimizations that might lead to production quantity shifts into the night shift or the shortterm build-up of buffer stocks to be able to change machine run times. For a derivation and a detailed description of the classic production KPIs are omitted in this thesis13 . At this point it should be noted that the development of an EnPI system for the practice partner will not be done in the context of this work. The increase in efficiency is only measured by the kilowatt hours actually saved.

13 For

further reading on KPIs please refer to the authors Nyhuis and Wiendahl [NW2012, p. 17 ff.] or Kletti and Schumacher [KS2011, p. 65 ff.].

4

State of the Art

A review of prior, relevant literature is an essential feature of any academic project. An effective review creates a firm foundation for advancing knowledge. It facilitates theory development, closes areas where a plethora of research exists, and uncovers areas where research is needed. W ebster and W atson (Webster and Watson in “Analysing the Past to prepare for the future: Writing a literature review” [WW2002, p. xiii].)

This research focuses on developing an application oriented and simulationbased procedure in order to allow for an optimum use of energy resources in producing companies. For this purpose, energy variables need to be integrated into the decision-making system of production planning and control, in order to be able to affect the energy use of a production depending on the specific state of operation. The accomplishment of this aim requires the consideration of dynamically simulated energy flows in production simulations as well as sufficiently accurate procedures to predict process behaviors. By combining suitable modeling and simulation approaches and known optimization methods, the evaluation of the optimum use of energy would become possible without causing any restrictions regarding the flexibility of production, the process quality nor major changes in production output. Towards this research objective the review of existing papers and scientific publications investigates different areas relevant for the simulation-based optimization of energy efficiency in producing companies.

© The Author(s), under exclusive license to Springer Fachmedien Wiesbaden GmbH, part of Springer Nature 2021 A. C. Römer, Simulation-based Optimization of Energy Efficiency in Production, Forschung zur Digitalisierung der Wirtschaft | Advanced Studies in Business Digitization, https://doi.org/10.1007/978-3-658-32971-6_4

67

68

4

State of the Art

This chapter1 gives an overview of the selection and evaluation process of relevant research approaches in section 4.1, followed by a discussion of results in section 4.2 as well as the derivation of the research demand in section 4.3. Reference to the analyzed publications and advanced literature will be made in the appropriate places of the text.

4.1

Selection and Evaluation of Relevant Research Approaches

The simulation-based optimization of energy efficiency combines concepts from Operations Management, such as the integration of energy efficiency goals in production, Operations Research, e.g., the use of integrated optimization algorithms as well as Modeling and Simulation, describing all techniques for model implementation and execution2 . By using a Venn diagram, Figure 4.1 illustrates the convergence of the involved disciplines. Being embodied in the areas of OM, the integration of energy efficiency goals in the normative, strategic, and operative production management is mandatory for following a holistic and sustainable consistent approach regarding material and energy use throughout all phases of a production [DS2008, p. 97]. While the problem formulation phase of including energy aspects in simulation approaches can be seen in the discipline of OM, the remaining steps of guiding a model builder through a simulation study are based in the discipline of M&S. The success of implementing and executing a simulation model directly depends on how extensively the steps of model conceptualization, data collection, model translation, experimental design, the production runs and analysis and furthermore the reporting and documentation have been accomplished3 [Ba+2005, pp. 12–16; Wa2009, pp. 27–28]. M&S techniques such as the discrete event simulation, continuous system simulation (CSS), other specific classes to simulate dynamic system behavior (e.g., DTSS, DESS, and DEVS) or a combination of different theories leading to a hybrid simulation approach, are typically used [ZPK2010, p. 8]. Located in the discipline of OR, optimization algorithms are required to come 1 Extracts

of this chapter have already been published at the Winter Simulation Conference 2016 [RS2016]. 2 Yilmaz and Ören offer a comprehensive and integrative overview of M&S from different perspectives for further reading [Ör2009, pp. 3–33]. 3 A detailed description of the single simulation study steps can be found in respective literature [Ba+2005, p. 12–16; La2007, p. 66–70].

4.1 Selection and Evaluation of Relevant Research Approaches

69

Studies on simulation-based modeling of the energy demand in production systems (II)

Operations Management (OM)

Integration of energy efficiency goals in production

(II) (II)

M&S techniques for model implementation and execution

Modeling & Simulation (M&S)

(IV) (IV) (I) (I)

Studies on the optimization of the energy demand in production systems (I)

(III) (III)

Use of integrated optimization algorithms

Simulation-based optimization of the energy efficiency in production (IV)

Studies on simulation-based optimization in the context of production systems (III)

Operations Research (OR)

Figure 4.1 Convergence of disciplines in the context of energy efficiency in producing companies

to appropriate business decisions as the complexity of production processes steadily increases. The use of optimization techniques aims at finding mathematical approaches that identify the optimum parameters for a predetermined analytical objective with constraints. Focusing on the convergence of all three described disciplines, only very few studies can be found in recent scientific discussions. Most studies identified as relevant in the selection process of the literature review converge only two of the three disciplines of OM, M&S and OR and can therefore be found in the subsets of intersection in Figure 4.1. • The optimization of the energy demand in production systems (intersection I in Figure 4.1) is sometimes performed without the use of simulation techniques. In the best case, such optimization studies can lead to an improvement of energy consumption even without detailed simulation models depicting all dynamic processes of the system. The overview of all existing machine processes and their decomposition into single operating states which can then be reproduced in form of mathematical models constitute the basis for the use

70

4

State of the Art

of such optimization algorithms. By using different modeling accuracies, the mathematical models of the operating states are combined to depict the complete consumption profile of a production machine. Thus, the optimal solution improving energy efficiency in a production system can be determined. • Studies in the field of simulation-based modeling of the energy demand in production systems (intersection II in Figure 4.1) primarily focus on the consideration of resource consumption based on physically measured operating states, which are expected to be constant over a defined period of time. The energy consumption of resources is assumed to be status based and presented as part of discrete event simulation approaches. Furthermore, some hybrid simulation approaches can be found. The hybrid simulation is here used to merge the DES techniques for material flow simulations with the continuous approach for energy flow simulation to depict the complex interactions between the material flow and the energy demand. • Studies on simulation-based optimization in the context of production systems (intersection III in Figure 4.1) combine simulation techniques with optimization methods to find the best input variable values from among all possibilities without manually evaluating each possibility of a complex large-scale problem. As the simulation-based optimization is well studied and has already been described in section 2.3, publications in this intersection are not considered in this literature review due to the missing focus on energy related topics. • Simulation-based optimization of energy efficiency in production (intersection IV in Figure 4.1) requires the combination of discrete or hybrid simulation approaches with integrated optimization possibilities in order to achieve an improvement of measures in the relevant field of action. Modeling the diversity of energy flows in the context of the whole production system together with an integrated evaluation system can constitute a decisional support system for production planning and control in order to achieve an optimum use of energy in production systems. The aforementioned intersections I and II as well as the convergence of all three concepts (intersection IV) provide the focus of this literature review.

4.1 Selection and Evaluation of Relevant Research Approaches

4.1.1

71

General Limitations for the Selection of Research Approaches

The importance of the energy and commodity markets has steadily increased since the first oil crisis in the 1970’s. Rising fossil fuel prices as well as debates on climate change and finite resources have triggered a mind change among politicians, entrepreneurs, and scientists. The sustained use of energy and other resources has become a basic requirement for a company to competitively perform on the market. There are different options to improve the energy efficiency, from a technical point of view as well as from a processing point of view. Whereas this work is only investigating the increase of energy efficiency through process improvements, Duflou et al. provide an overview of energy saving strategies, focusing on technical improvements, such as the use of energy-efficient components and the investment in innovative production techniques that can be implemented to significantly improve the energy consumption [Du+2012, pp. 589–600]. Within the review of relevant research approaches, a comprehensive research of available publications was conducted and assessed in order to provide an overview of the current state-of-the-art in the field of energy efficiency in production. It is research-based and includes journal articles, conference proceedings as well as theses and dissertations. Relevant literature has been found using the IEEE Xplore Digital Library, the ACM Digital Library, the Web of Science as well as ScienceDirect. As the methodology developed in this thesis is aiming at practical application, the literature review is focused accordingly. To identify the papers that are included in the literature review, the following search terms and key word combinations have been used in English and German language: • (“simulation*based optimization” OR “simulation optimization”) AND “energy efficiency”, • (“simulation*based optimization” OR “simulation optimization”) AND “energy efficiency” AND (production OR manufacturing) • (“simulation*based optimization” OR “simulation optimization”) AND “energy efficiency” AND (“hybrid simulation” OR “combined simulation”) The search results have been refined manually by considering only publications written in English and German language that have been published between January 2000 and December 2017. Additionally, cross-references and frequently quoted sources have been checked and search results have been refined manually by significant publication titles. Thus, the selected search criteria generated 167 relevant publications. The abstracts of all papers have been reviewed under the

72

4

State of the Art

consideration of certain limitations (sub-section 3.1) and thus 86 papers for the full-text review have been selected. Based on the selection process using the aforementioned search terms and key word combinations, the abstracts of the search results have been read and refined by means of the following factors: • The approach focuses on the increase of energy efficiency of production processes by implementing process improvements rather than technological innovations • The approach considers energy consumption in an appropriate level of detail • The approach considers simulation and/or optimization techniques • The definition of system boundaries for the investigated scenarios is at least on the level of a multi-step production machine Research approaches that did not match these criteria based on their abstract have not been considered for further investigation. Papers lacking information in the abstract, have been taken into account for the full-text review and have been sorted out afterwards. The full-text-review again concluded that several publications were not relevant for further investigations. This selection process resulted in the consideration of 25 publications in total. All publications included in the literature review have been analyzed regarding the key aspects of the study, grouped corresponding to the previously defined intersections from Figure 4.1 and evaluated according to the group specific variables subsequently described.

4.1.2

Studies on the Optimization of the Energy Demand in Production Systems

Optimization covers a wide range of problems, aiming at finding the optimal solution for a certain problem, whereas the complexity of the approach depends on the objective function and constraints as well as on the design variables [Ya2018, p. 5]. Several studies on optimization approaches in the context of improving energy consumption of production processes can be found. Fulfilling the mentioned limitations for the paper selection process, seven research papers could be identified as relevant for intersection I. The optimization approaches will be shortly summarized in chronological order according to the year of publication. Devoldere et al. analyze the energy consumption of production systems in their different operating states in the fields of bending presses, milling, and laser cutting machines. Besides the productive operating states, non-productive phases such as stand-by, set-up, and shut down times are considered. Aiming at the

4.1 Selection and Evaluation of Relevant Research Approaches

73

detailed evaluation of the machine’s energy consumption and the ability to analyze process alternatives, time and energy studies are carried out to interpret the energy levels of single machines and their components in the terms of technical and process-based improvement of the resources [De+2007; De+2008]. Devoldere et al. prove in several case studies that the main potential for energy savings can be found in the fields of non-productive operation phases and present results for savings in the laser cutting case of 12% and for bending presses of up to 60%. Dietmair and Verl follow a machine control-based approach to determine the consumption forecasting of milling machines [DV2009]. The basic method for their study is the detailed breakdown of machine processes into individual operating states, which are represented as mathematical models and considered with different modeling accuracies. By additive linking, those single mathematical models are composed to form the production consumption profile. By combining the models with real operational data, Dietmair and Verl calculate reliable forecast values regarding the energy insensitivity of production machines. Thereby they can identify critical machine components, make improvements regarding the machine control or select process alternatives for inefficient machines. Wang et al. present an approach for energy reduction based on an integrated optimization model for batch production load scheduling in energy-intensive enterprises (EIE) without violating existing constraints of the production processes [Wa+2012]. The reduction of energy use during peak hours, the time-shifting of energy use in off-peak times of the electricity tariff as well as reduction of the energy demand through optimal load scheduling are considered in the optimization approach. The formulation of the integrated optimization problem includes batch production loads and power generation scheduling, as the approach is addressing EIEs with self-generation power plants. After analyzing the production load curves and decomposing them into base loads and batch production loads, the operating parameters for rescheduling the batch production are formulated. Subsequently, the power generation scheduling problem is formulated, followed by the creation of the integrated optimization model containing integer variables as well as nonlinear constraints. In order to be able to convert the optimization problem into a mixed integer linear programming model, linearization techniques are introduced. Wang et al. conduct a case study in an iron and steel plant, proving the effectiveness of their approach. Chen et al. analyze the effective control of machine startup and shutdown schedules to optimize the energy consumption considering given productivity requirements for serial production lines with Bernoulli machines and finite buffers [Ch+2013]. Besides using productivity performance measurements, an additional energy performance system is introduced. To calculate the performance measures

74

4

State of the Art

of serial lines, Markovian analysis and a recursive procedure built on aggregation are used. Chen et al. validate their approach in an automotive paint shop line, proving that scheduling the startup and shutdown times of machines lead to significant improvements in energy efficiency. Since 2009, the research group Ecomation is focusing on approaches regarding the control of energy consumption in production facilities and the increase of energy efficiency by using automation techniques [EV2014]. As a part of this research team, Eberspaecher and Verl present a status-based approach focusing on the energy-optimal use of unproductive times in manufacturing. Combining two optimization algorithms, the Dijkstra-algorithm and the A*-algorithm, the most energy-efficient production state for the machine can be found. In order to be able to apply the optimization theory on machine tools, the energy consumption model had to be defined as a graph and was implemented in C# on the control’s operating system. Eberspaecher and Verl prove during a prototypical implementation that the developed consumption graph-based energy optimization approach allows for a switching of energy saving modes during unproductive times to spend them in an energy-optimal state [Eb+2014, p. 48]. Swat presents an approach for the design of energy efficient processes in serial productions which enables the production planner to predict the energy requirements of production processes already in the early planning stages of a production [Sw2015]. By building a business-related energy database, Swat creates the possibility to determine the total energy demand for all combinations of production equipment and machine parameters. Thus, the production planner has the option of selecting alternative process parameters and components to optimize the energy demand. Swat validates his methodology based on the processes of electrochemical machining and honing. The deviation between the predicted and the measured values for the energy consumption of the machine tools was only 4% [Sw2015, p. 95]. Baumann et al. develop an integrated multi-criteria optimization and scheduling platform for energy consumption reductions in the glass tempering industry [Ba+2016]. The platform consists of three components, a thermodynamic process model of all energy-critical steps in glass manufacturing designed using Modelica4 , a scheduling model to determine energy-efficient loading sequences for the furnace using multi-criteria search techniques from discrete optimization and a 4 The development of Modelica was initiated by Hilding Elmqvist in 1996. „The basic idea behind Modelica was to create a modeling language that could express the behavior of models from a wide range of engineering domains without limiting those models to a particular commercial tool […] Modelica is both, a modeling language and a model exchange specification“ [Ti2001, p. 4].

4.1 Selection and Evaluation of Relevant Research Approaches

75

visualization module to present pareto-optimal energy saving solutions graphically to the production planners. Baumann et al. aim at the automated detection of feasible loading sequences and batch sizes, optimized set-up times, as well as the avoidance of peak demands in energy consumption [Ba+2016, p. 124]. Summarizing the research approaches in the field of optimization of the energy demand, two main methodological strategies of the researches can be identified. On the one hand, the effective scheduling and planning of machine start-ups and shutdowns as well as the usage of non-productive times is addressed as the main optimization potential. On the other hand, the energy usage is minimized by analyzing energy consumption of single machines and components and choosing process alternatives for industrial processes.

4.1.3

Studies on Simulation-Based Modeling of the Energy Demand in Production Systems

The simulation-based modeling of energy demand in production is a method for assessing the dynamics of production processes. To keep the production permanently on its optimum energy level, an ideal simulation approach has to consider all relevant energy flows with sufficient accuracy. Fulfilling the abovementioned limitations for the paper selection process, eleven research papers remain relevant for the intersection II. All authors justify the selection of a simulation-based approach to model the energy use by the rising complexity of production processes that can hardly be evaluated in any other way. The reviewed publications aim at creating a pragmatic tool for supporting the production planning and control. The majority of research approaches published in the field of energy flow simulation for production systems involves discrete event simulation. They will be shortly summarized in chronological order according to the year of publication followed by hybrid simulation approaches later in this section. Solding and Petku developed a methodology following the DES approach, focusing on the reduction of energy use as well as on avoidance of energy peaks in energy-intensive industries [SP2005]. They carried out several simulation studies in various industries modeling the state-based consumption behavior of machines by extending conventional simulators through additional programming within the simulation tool. By adding parameters to the simulation model, they created an accurate model for energy reduction, load management measures as well as the decision support for changing and combining different energy carriers.

76

4

State of the Art

Weinert presents a methodology focusing on the time-based structuring of the energy inputs depending on the operating state of the machine [We2010]. Combining the time spend in a certain operating state and the associated specific power consumption profile of a resource, he defines an own classification system. He assumes that similar resources have comparable power consumption profiles and develops a mathematical functional description of the real consumption profiles. By combining the energy profiles to a sequence, the description of a process chain in an appropriate level of detail is reached. The grouping of individually modeled process chains leads to the description of the total energy consumption of a production. Weinert validates the concept in a case study in the field of a mechanical production comparing the forecast quality of his approach to reference measurements. He simulates the production processes using Visual Components 3DCreate and develops a prototype software to model the energy profiles. By introducing several interfaces, he enables the exchange of process data with the simulation software and thus creates a tool to support the modeling of energy use in production systems. Berglund et al. present an approach to improve the production sustainability by measuring and evaluating the concerted effect of process energy from machine operations and the facility energy from building services [Be+2011]. They invent a state-based DES model incorporating processes, process energy, and facility knowledge. Integrating real production data, process data, and facility energy data, Berglund et al. validate their approach in an engine block production. They prove that their approach has the potential to reduce manufacturing energy consumption even though the model generation is highly complex. Wolff, Kulus, and Dreher present attempts to include energy aspects in the material flow simulation [WKD2012]. Assuming that it is possible to model energy consumption as a constant or time-dependent status, they introduce a classification system for the machine states. To ensure that the functionality can be removed at any time for a model, the principle of modularity was adopted for elements of the energy simulation. Thus, the end user is able to switch the energy calculation modules on and off according to his requirements. Taking into account negative effects on the system performance, the energy simulation and the material flow simulation are run in parallel. To realize the import of the energy consumption simulation, three modules are required: a module for parametrization and import, a calculation module as well as a model for statistics and visualization. Conducting pilot studies in the automotive industry, the energy calculation module was validated. For further documentation purposes as well as the presentation and deeper analysis of the results, the researchers provide additional options for

4.1 Selection and Evaluation of Relevant Research Approaches

77

the visualization of the energy demand in form of diagrams, key performance indicators, and statistic tools. As part of the above mentioned Ecomation research team, Haag developed a model-based planning and evaluation methodology to permanently keep the production processes in energetically favorable area [Ha2013]. Using the methods of systems technology to merge the main processes and the production periphery, Haag extends the status-based approach of Dietmair and Verl by considering the time-based dimension and thus converting it from a static to a dynamic model. He integrates the areas of production planning and control and allows the evaluation of planning alternatives already in the early planning stages of production process planning. Haag implements a performance measurement system in the form of a computing system for assessing planning scenarios. He considers production targets (quality key figures, overall equipment effectiveness, and throughput times) to be able to use them as weighted factors for the evaluation of planning scenarios. Haag validates his approach on the example of a cutting production. The modeling and simulation are conducted in Plant Simulation version 10, the parametrization is done in an external database. Energy data, as well as setup times and process data are taken over into the simulation model using an external interface. He proves that his approach is suitable for evaluating the influence of technological and organizational parameters on the overall energy consumption of production as well as for determining the optimal set of parameters. Schlegel, Stoldt, and Putz present an approach integrating energy efficiency analysis in material flow simulations [SSP2013]. Assuming status-based energy consumption patterns in the production, they develop a component model that is enlarged by a software package (eniBRIC). This allows for the consideration of energetically relevant inputs and outputs depending on the operating state of a machine. An evaluation module is used for data aggregation and data analysis. The approach is validated in the automotive industry. Despite the additional efforts for creating the simulation model as well as the increase of simulation time due to parameter variation, the methodology allows a comprehensive analysis of resource consumption, the comparison of process alternatives and the dimensioning of infrastructure facilities. The simple parametrization of eniBRIC allows for the use in different industry sectors. Using the discrete event simulator SIMIO, Cataldo, Taisch, and Stahl present an approach for evaluating the energy consumption of an automotive engine assembly line [CTS2013]. To model the behavior of each production machine, they divide the mechanical functional behavior of a machine into small single

78

4

State of the Art

steps characterized by parameters. Additionally, they implemented control functionalities based on finite state machine (FSM) control algorithms that have been translated into C# and thus integrated into the simulator. The energetic states of the machines have been modeled and integrated the same way. Thus, the energy behavior can be run in the simulator together with the mechanical behavior and the control functionality. Cataldo, Taisch, and Stahl validate their approach through a simulation study with a serial production line made of four machines. They prove that their method allows engineers to analyze the simulated production line and to evaluate the efficiency of the simulated layout. The above-mentioned approaches focus on the consideration of energy consumption based on measured operating states, describing them as constant over a fixed period of time. DES approaches do often not provide a sufficient accuracy for the modeling of highly dynamic production processes. To get a more detailed description of dynamic processes, hybrid simulation models are named as a possible solution in literature [SP2014, p. 111]. The hybrid simulation merges the DES techniques for material flow simulations with a continuous approach for energy flow simulation to model the complex interactions between the material flow and the energy demand. As the topic of hybrid simulation is quite new in the context of production simulation, only four publications have been identified for this literature review. As a part of the “SimEnergy” research project, Peter and Wenzel develop an approach focusing on bidirectional interactions between discrete event production models and continuous energy models [PW2015]. Using a communication platform, the discrete event simulation tool for modeling the material flow is connected with the continuous simulator to depict the energy aspects. The platform allows for the parametrization of the material and the energy flow model. Having a separate plug-in, the data exchange as well as the synchronization time are being controlled. Peter and Wenzel validate their approach using a practical example from the automotive industry sector. Executing several simulation runs, the influence of different shut-off temperatures on the output quantity is examined. It is shown that the interaction between production processes and energy flows through coupled simulation models can be analyzed and evaluated. Peter and Wenzel criticize that adjustments in production control to reduce energy consumption are only implemented in real industry cases, as long as they do not have a negative influence on output quantity [PW2015, p. 543]. Schmidt and Pawletta follow a research approach to describe hybrid production models in a purely discrete event simulation environment [SP2014]. Unlike Peter and Wenzel in their research, the combined consideration of the event discrete and the continuous modeling aspects is not realized through the

4.1 Selection and Evaluation of Relevant Research Approaches

79

coupling of two simulation models executed in their own simulation system but through the use of the Discrete Event System and Differential Equation System Specification (DEV & DESS) approach from the field of systems theory. Schmidt and Pawletta use the discrete event simulator MATLAB/SimEvents without enabling the continuous model library to model their transaction-oriented process chains. To calculate the resource consumption of machines, a newly developed model library is used. It contains specific manufacturing processes and can submit its calculation results in form of state variable vectors to the simulation software. The hybrid simulation model consists of three parts, the material flow component, the state-based controller and a subsystem for displaying time which are all validated using the example of a hardening furnace. The results show that the oven reaches the preset temperature but the set time cannot be met. The approach can be used to validate technical data, to gain information for a proactive maintenance of machines as well as for the reduction of shut down times. In another paper in 2017, Pawletta and Schmidt together with Junglas present a multi-modeling approach, combining several modeling methods, such as discrete event modeling for the material flow, state graphs for the process control as well as continuous models for the process physics, to describe and investigate the dynamic system behavior of a production line [PSJ2017]. The multi-modeling approach subdivides a manufacturing system in three general layers and allows the implementation of different models with varying levels of detail, organized in a library. Thus, it supports the component-oriented and flexible refinement of production line models. Pawletta, Schmidt, and Junglas illustrate their approach using MATLAB/Simulink to model a furnace component and its timerelated energy consumption behavior in a set of models with different levels of abstraction and varying levels of detail. They compare refinement costs, simulation runtimes and the level of accuracy and state that the energy results among the different model accuracies vary by 10–15% [PSJ2017, p. 122]. Schlüter et al. use a bidirectional coupling of material flow and energy models created in a hybrid simulation environment based on MATLAB, Simulink, and Stateflow, supporting continuous and time discrete processes [Sc+2017]. The coupled simulation model communicates with a process control model, the process control receives system and process parameters as well as plan data from the simulation model and returns order and control data to the simulation model for validation. On the basis of the returned configurations, various operating scenarios and the effectiveness of individual optimization measures can be simulated and evaluated. Schlüter et al. validate their approach comparing simulation results to measured reference data from a non-ferrous melting and die-casting plant.

80

4

State of the Art

Summarizing the research approaches of intersection II, two main approaches can be distinguished. Most publications follow a DES-based model including the energy use of machines either as state-based variables or the energy use is included as a state-based consumption with cumulative load profiles. Only a few publications describe a hybrid simulation approach to realistically model the system of highly dynamic production processes. Regardless of the chosen approach, the simulation model generation is described as highly complex and time consuming. Often support software is required in order to manage, aggregate, and evaluate the collected energy data for the simulation models. The system boundaries of the considered production areas are therefore chosen very diversely among the approaches, ranging from the consideration of consumers that are only directly involved in the production process to the consideration of support processes and peripheral equipment.

4.1.4

Studies on Simulation-Based optimization of Energy Efficiency in Production

Studies on the simulation-based optimization of energy efficiency in production converge the three disciplines of OR, OM, and M&S and therefore require a holistic view on producing companies, an understanding of all processes, the relevant process in- and outputs, as well as their existing dynamic interactions. Aiming at the integration of energy efficiency goals in the decision system of a production, an approach combining simulation and optimization studies should be usable as a decision supporting tool in a complex and dynamic industrial environment. Fulfilling the above-mentioned limitations for the paper selection process, only the subsequently listed seven research papers can be identified as relevant. Rager developed a concept for an energy-oriented utilization of identical parallel machines with the objectives of minimizing the number of occupied machines to smooth the use of energy throughout the production time [Ra2008]. By applying heuristic methods for solving the optimization problem based on hybrid evolutionary algorithms, Rager develops a decision model to support the energy-oriented machine scheduling process. He proves the applicability of his approach in a case study of the textile industry. The discrete event simulation software eM-Plant is used for visualization purposes. Depending on the number of production orders, the simulation model execution shows, that energy cost savings between 10 and 20% and load peak reductions up to 30% can be achieved [Ra2008, pp. 120–121].

4.1 Selection and Evaluation of Relevant Research Approaches

81

Lorenz, Hesse, and Fischer present a DES-based approach to simulate and optimize energy consumption in complex automated production lines in the automotive industry using periodic time-expanded networks [LHF2012]. Assuming the energy consumption behavior as being state-based, a consumption profile is assigned to each process of a robot in the simulation. Every ending simulation process as well as waiting processes create an entry into a data table, providing the basis for energy consumption analysis for single robots as well as the whole production line. To reduce peak-loads, a peak-load optimization process is added to the simulation model, calculating the optimal starting time for all processes whenever shifting execution periods is possible before starting the process. The approach is validated in a car body shop showing that the introduction of predetermined waiting periods cause a peak-load drop by nearly 20% [LHF2012, p. 2885]. Aiming at the development of an energy-oriented simulation-based approach that enables every user in a producing company to independently develop and assess simulation studies for the detection of potential energy efficiency improvements, Thiede presents a methodology consisting of ten process steps [Th2012]. Focusing on the general character of his concept, he specifies that the approach should neither be limited to a specific case in production or a particular industrial sector, nor to a specific simulation software. In addition, all energy flows and dependencies in production should be tracked. Thiede uses state-based energy profiles to picture the consumption behavior of a machine depending on its state of operation. To depict the interactions between the energy and the material flows, he follows a discrete-continuous simulation approach. Thiede validates his methodology in case studies from the automotive, the textile, and the electronic industries. He proves the applicability of his approach regardless of the size of production, the industry or the level of training of the executing employees and shows that numerous optimization approaches can be developed and evaluated based on his structured ten step plan. Through the use of the universally applicable optimization library OptQuest™, Thiede includes optimization studies. Heinzl et al. follow an interdisciplinary optimization approach for predicting the impact of energy saving measures by comparing different production plant scenarios [He+2013]. Combining the energy optimization of production processes with separately analyzed aspects of the fields “machine and production system”, “building” as well as the “energy system”, Heinzl et al. use the method of cosimulation to study the energetic interactions of the single subsystems. To cover all energy aspects, several simulation models, which are executable as standalone modeling environments, are developed. Heinzl et al. use different simulation

82

4

State of the Art

tools (e.g., MATLAB, EnergyPlus, and Dymola), to ensure that the special requirements of every sub model are met by choosing the optimum software. To couple the different simulation models, the extensible open-source software platform ‘Building Controls Virtual Test Bed’ (BCVTB) is used [He+2013, p. 306]. It allows the runtime coupling of different simulation software, supports data exchange and hierarchical combination of different modeling semantics. Proving the advantages of their co-simulation approach, the authors create different production scenarios varying the climatic and production conditions. The approach allows the user to make predictions regarding energy optimization measures in production systems. Eberspaecher et al. present an approach combining power measurement data and control signal information with consumption data that has been generated using static and dynamic simulation models [Eb+2014]. In a second step a situation-based optimization is carried out to reduce energy consumption of machine tools. Eberspaecher et al. develop a general machine tool component library to be able to calculate with detailed electric load curves instead of using operating states. They combine their simulation approach to estimate the energy demand on machine and component level with a monitoring approach to allow real time energy demand monitoring. In combination with a component and an operating state optimizer the optimal parameter configuration for an energy-optimal production is found. In 2017, Baumann et al. extend their approach presented in section 4.1.2 by a simulation model to identify energetic losses occurring in manufacturing processes [Ba+2017]. They generate necessary process parameters for the simulation model for each product type by a complex data analysis of production and energy data in MATLAB and then provide them in tables and related files for the simulation model which is modeled using Modelica as a language and SimulationX as a software tool. After a validation process, the model data is exported as a functional mockup unit (FMU) and then integrated in the buildup Advanced Planning and Scheduling (APS) tool chain containing the optimization module. The consideration of the energy demand of the depicted manufacturing processes thus enables the planner to reduce the energy demand and the associated costs of the production process [Ba+2017, p. 67]. Sobottka, Kamhuber, and Sihn present an approach aiming at the development of a new planning tool to increase the energy efficiency of productions using a hybrid simulation and a multi-criteria optimization [SKS2017]. In several steps, they develop an optimization approach based on genetic algorithms to expand the target system of planning to include energy efficiency targets in addition to the classic economic targets. This extension requires a simultaneous mapping of the

4.2 Comparison of Results and Discussion

83

energy and material flow system in a hybrid simulation method, which is only mentioned but not described in detail. So far, the approach has only been tested with a simplified production line, as restrictions to limit the search space for the optimization have to be found for a successful implementation. Summarizing the papers of intersection IV, different approaches to simulate the energy consumption in production systems combined with optimization techniques are presented. While the DES-based approach combined with a peak-load optimizer is suitable to calculate the optimum start and shut down times for robots, the combined simulation approach together with the optimization library OptQuest™ focuses on the detection of potential energy improvements throughout the whole production process. The concepts of simulation model coupling in combination with an operating state optimizer, as well as the creation of an APS tool chain containing an optimization module support the real time monitoring and parameter adaptations to optimize the energy use and the production process costs. Approaches that focus on the continuous presentation of energy consumption in a production simulation together with the discrete material flows and that could optimize in the same software solution have not been found.

4.2

Comparison of Results and Discussion

All publications mentioned in this literature review present new approaches in the research field of energy efficiency in production. Mentioned in all papers, the top motivation for the researchers is the changing importance towards the sustainable use of resources, mainly initiated by debates on global warming, rising energy costs, and resource depletion. For most approaches, the consideration of the energy consumption is generally based on measured operating states, which are considered to be constant over a defined period of time. The defined system boundaries are varying throughout the papers, reaching from the consideration of a multi-step production machine to a whole production facility including technical building services and peripheral production equipment. Regardless of the chosen system boundaries, the efforts of the data acquisition and evaluation as well as the model generation for all optimization and simulation models have been described as complex and time-consuming. The use of continuous simulation approaches for the depiction of a dynamic energy consumption pattern is rarely used. After having analyzed all papers, three main concepts for the reduction of the energy use can be distinguished. Firstly, the optimization of machine control functionalities provides energy saving potentials, e.g., the efficient use of non-productive times or the peak-load avoidance. Secondly, by varying process

84

4

State of the Art

parameters in the production, saving opportunities can be realized without influencing the production output negatively. The third category focuses on early planning phases in production planning. By the evaluation of process alternatives and the appropriate dimensioning of energy efficient components and machines, the production design can be influenced right from the beginning. As the third case is relevant for new planning projects rather than for the optimization of existing productions, it will not be considered any further in this thesis. In general terms, the evaluation methods in the approaches all focus on subareas of the production, following one perspective in order to optimize the energy use. This may lead to incomplete energy policies as well as conflicting goals within the production and may cause problem shifting from one area to another. Only by following an integrated view of production systems, considering all interdependencies between the production equipment, process parameters, relevant inand outputs, an appropriate method for assessing the energy consumption can be established. Most of the studies have been validated using case studies from the automotive industry. Often, developed approaches are very case-specific and not transferrable to the production processes of other industries without high efforts. Several authors criticize that adjustments in production control to reduce energy consumption are only implemented in real industry cases, as long as they do not have a negative influence on output quantities. Although the main focus is on optimizing the energy use without causing any restrictions regarding the flexibility of production, the process quality nor major changes in production output, energy variables need to be integrated into the decision system of production planning and control. Thus, the accurate evaluation of the energy cost savings in relation to additional costs (incurring due to changes in output quantities caused by energy efficient production setups) becomes possible. As it is the case in all approaches of intersection I, the utilization of optimization methods without the visualization of simulation techniques can hardly be used as an actual stand-alone decision support when following a holistic system perspective in production planning and control. A realistic presentation of energy consumption and its dynamics within the production system requires a significant visualization as the pure process description in form of mathematical models is rather a research topic than industrial practice. The simulation of energy use without integrated optimization algorithms can be considered as a tool to display material and energy flows as well as operational machine states during run time. Due to highly complex and dynamic processes in nowadays production systems, manual process optimizations carried out by single persons are not goal-oriented when aiming towards an energy efficient production in all relevant fields of action.

4.2 Comparison of Results and Discussion

85

Therefore, a holistic approach combining simulation techniques with automated optimization algorithms is required to recognize systematic improvements. Tables 4.1, 4.2 and 4.3 summarize all identified research approaches regarding the discussed findings. The following two general conclusions can be drawn: Firstly, no single approach exists covering all criteria in combination to a satisfying extend leaving the research demand for a methodology covering all criteria in combination. Secondly, while the state-based consideration of energy consumption behavior is implemented sufficiently in several approaches, the integration of dynamic power consumption profiles has not been the focus of production simulation developers so far, nor has the combination of simulation and optimization. Table 4.1 Summary of the literature analysis for intersection I

86 Table 4.2 Summary of the literature analysis for intersection II

4

State of the Art

4.3 Derivation of Research Demand

87

Table 4.3 Summary of the literature analysis for intersection IV

4.3

Derivation of Research Demand

The discussed findings underline the complexity of the simulation-based optimization of energy efficiency in production. Resulting from the evaluation of existing research approaches in section 4.2, the following research gaps can be identified:

88

4

State of the Art

• No realization of an approach implementing event-based and continuous simulation in combination with optimization in one software tool without the requirement for challenging interface management and complex data exchange or synchronization demands. • The analyzed approaches mainly lack an easy applicability and usability on different use case scenarios in industry as they do not pursuit a flexible and modular structure. • The approaches analyzed are very case-specific and usually involve only few options regarding parametrization and transfer to other scenarios or use cases. • In most approaches, the interaction of discrete material and continuous energy flows, as well as dynamic dependencies, and complex process structures are not considered in a reasonable level of detail. • Throughout the analyzed approaches, system boundaries of the inspected areas and processes are very broad ranging. An integrative view of processes on production line and factory level to outline conflicts and contradicting requirements and to support the optimization of critical parameters can hardly be found. • Hardly any approach allows for the computer-aided optimization of complex production lines with its interactions and dependencies. Most approaches are limited to machine level evaluation and optimization. The interactions of machines as well as cumulative effects on a production line and factory level are generally not considered in detail. • A comprehensive evaluation, reporting, and visualization of results is required to support a full documentation of improvements and enable a widespread understanding of optimization results. The optimization recommendations must be portrayed comprehensibly to function as an active decision-making tool. Considering the identified gaps, the following research questions can be formulated:

How can the energy consumption of a production system be depicted in a simulation model that can as well be used for optimization scenarios? In this context, it must be clarified how the energy consumption of machines can be integrated into the production simulation, which simulation techniques can be used, and which requirements and limitations arise in combination with optimization techniques.

4.3 Derivation of Research Demand

89

If it is possible to realistically depict the energy consumption in a simulation model, the framework conditions must be clarified to prevent current production factors used for planning from being affected by energy efficiency optimization. Thus, the following research questions have to be answered:

How can an energy efficiency optimization of a production system be executed without causing any restrictions on production flexibility, without influencing the quality or the output of the production? Can energy be integrated as a control parameter for production optimization? Both research questions are closely related to each other. However, answering the first question will automatically lead to the second question. Once the methodology for the simulation-based optimization of the energy efficiency in production has been defined, it is required to clarify options of monetary evaluation of the optimization results. Additionally, it must be thought about the requirement of considering the production periphery for a realistic evaluation of the entire production facility. The following main- and sub-question arise:

How can the profitability of the energy optimization methodology be rated considering the various fields of application? To what extent should the periphery of a production system be included in the simulation-based optimization in order to obtain a quantifiable statement about the energetic behavior of the entire system? Section 5.1 will be used to investigate how an integrated simulation-based methodology for energy efficiency optimization in production can overcome existing inadequacies.

5

Development of a Simulation-based Methodology for Energy Efficiency Optimization

Perfection is achieved, not when there is nothing more to add, but when there is nothing left to take away. Antoine de S aint-E xupe´ry (Antoine de S aint-E xupe´ryaviator, writer and philosopher in „Wind, Sand and Stars“ (original title „Terre des Hommes“ first published in 1939) [Sa2000].)

The identification of research gaps has shown that there is so far no applicable approach in industrial practice that dynamically integrates the energy consumption behavior of machines in classical production simulation and provides recommendations for energy-efficient control of machines during non-value-adding production times. This chapter therefore first defines the objectives and requirements of such an approach, followed by the development of a conceptual framework in section 5.2. Sections 5.3 and 5.4 show how the simulation and optimization components of a methodology for simulation-based optimization of energy efficiency in production can be structured, followed by software selection for the practical implementation of the methodology in section 5.5 and a prototypical implementation using a fictional production example in section 5.6.

Electronic supplementary material The online version of this chapter (https://doi.org/10.1007/978-3-658-32971-6_5) contains supplementary material, which is available to authorized users.

© The Author(s), under exclusive license to Springer Fachmedien Wiesbaden GmbH, part of Springer Nature 2021 A. C. Römer, Simulation-based Optimization of Energy Efficiency in Production, Forschung zur Digitalisierung der Wirtschaft | Advanced Studies in Business Digitization, https://doi.org/10.1007/978-3-658-32971-6_5

91

92

5.1

5

Development of a Simulation-based Methodology …

Concept Objective and Requirements

Subsequent, the identified research demand is brought into specific objectives and requirements, describing the intended state of an energy efficiency optimization methodology for production as well as process steps leading to it. The formulated requirements are used as the basis for the deductive development of the methodology. The main objectives and requirements are • Transparency The methodology shall contain all relevant material and energy flows as well as their dynamic interdependencies to increase transparency in production towards improvements. Additionally, it should outline conflicts and contradicting requirements, to support the optimization of critical parameters and to reduce operating costs of processes. • Accuracy The methodology shall be able to depict production processes in an accurate way to support event-based material flow simulation and time-continuous energy flow simulation. Thus, process structures should be shown in a reasonable level of detail. • Simplicity The methodology shall be implemented in one software tool to provide a holistic view without the complexity of interface management, complex data exchanges, and synchronization requirements. • Expandability The methodology shall be applicable to different production scenarios and is extendible from production sub-systems to factory level. • Parametrization The methodology shall easily be configurable for a comfortable use without the need to consult a simulation expert. To ensure an uncomplicated implementation for a wide range of production cases without causing high efforts for case-specific adaptations, a high level of configuration and parametrization must therefore be supported by the approach. • Modularity The methodology shall be built up modular in order to allow for flexible coupling and fast adaptation of single system components. All relevant functions and systems of a production need to be represented in fast configurable modules. Interactions in the production are depicted using defined interfaces and connecting single modules.

5.1 Concept Objective and Requirements

93

• Optimization The methodology shall include optimization experiments for goal-oriented recommendations in all relevant fields of action. The approach must allow the optimization of complex production lines with its interactions and dependencies and should not be limited to single machine optimization. The interactions of machines as well as cumulative effects on a production line and factory level need to be considered in detail. Additionally, the optimization of the energy efficiency has to be done following an integrative approach, which means that energy costs have to be looked at together with traditional production planning aspects as time, costs and quality. • Predictability The methodology shall support the structuring of data in a way that it is possible to select from the existing database data sets that allow a forecasting of energy requirements for unanalyzed combinations of operating machines, process scheduling tasks and process parameters. • Evaluation The methodology shall include an evaluation tool to provide Energy and Key Performance Indicators (EnPIs and KPIs) and their development over time. The optimization recommendations must be portrayed comprehensibly to function as an active decision-making tool. • Visualization The methodology shall include a dynamic visualization to provide a good overview of parameters, processes, and relevant impact factors. The visualization of results is additionally required to support a full documentation of improvements and to create a widespread understanding of optimization results. • Application The methodology shall include application guidelines that allow a systematic implementation in production. The application guidelines intend to avoid unnecessary repetitions of implementation steps, support the individual implementation phases by means of proven procedures, and minimize the overall implementation time as well as expenses for system and data maintenance. • Usability The methodology shall be readily accessible to ensure broad utilization of the research results. Therefore, the simulation and optimization tool has to be useable without demanding for cost-intensive licenses and implementation specialists.

94

5.2

5

Development of a Simulation-based Methodology …

Conceptual Framework

Based on the defined requirements, this section describes the conceptional framework. For the simulation-based optimization methodology considering the existing production system in a company, the system inputs and outputs, as well as other constituting factors in a production plant. One focus of the concept will be put on the definition of interfaces of the material and energy flows to obtain reliable simulation results. Additionally, optimizations must be integrated in a way they can handle the complexity of hybrid models and allow the automated generation of altered parameters. Conditions and modalities, specific aspects of single modules, and their functional principles will be outlined in detail in sections 5.3 and 5.4. As shown in Figure 5.1, the conceptual framework includes two main modules, a production and energy simulation and an energy optimizer. The simulation module is split up into three subparts: A part for the development of a model of the real production flow, a model for the depiction of production machine behavior and a model of the energy flow in production. The second module includes the optimization tool for performing energy efficiency optimizations using all production model components. Both modules receive the required production related input data from the actual production system to be able to depict production processes in the most accurate and realistic way to perform energy efficiency optimizations. To provide an overview of parameters, processes and relevant impact factors for energy aspects, and to have a full documentation of improvements and optimization results, a visualization tool should be directly linked to the production and energy module as well as to the energy optimizer. The information and the recommendations for action outlined in the visualization tool can then be implemented in the real production.

5.3

Description of the M&S Approach

The modeling and simulation of the material and energy flows in a production requires the use of different simulation paradigms for a realistic depiction of the various aspects1 . Following Djanatliev and German, who propose a methodology targeting to define structured hybrid approaches in domain-specific contexts,

1 Extracts

of this section have already been published at the Winter Simulation Conference 2019 [RS2019].

5.3 Description of the M&S Approach

95

Production and Energy Simulation Production Flow Component

system model of production flow and setups for production planning

Visualization Module

Management system target state

Production system Production machines ERP MES

actual state

production related input data

Machine Component

system behavior model of machines

Evaluation

Energy Component

results and key performance indicators, experiment evaluation

system behavior model of the energy consumption

Energy Monitoring

Visualization Relevant parameters and specific energy data

Optimized planning

modified parameters

feedback

visualization for decisionsupport

Energy Optimizer Optimization Module objectives functions and optimization targets for energy efficiency optimization

Figure 5.1 Conceptual framework of the simulation-based optimization and interfaces to the real production system

the levels of abstractions and the view on subclasses of a system have to be identified [DG2015, p. 12 and p. 16]. So far, this method of model building has only been tested on case studies in the healthcare sector, but it can be adapted and used for the combined energy and production model. Four abstraction levels, requiring different simulation techniques, can be identified. On a macro level, the total energy consumption of the entire production is being relevant. The energy consumption flow is continuous in nature and therefore SD is the appropriate simulation paradigm of choice. Besides it is possible to perform input data updates using DES. The material flow and the production process landscape are presented on the meso level, represented by the process-oriented DES. While the machine behavior is treated as a black box on the meso level, the detailed structure, e.g., the process behavior of the single machines, is presented on the micro level. The behavior of the machines includes, in addition to the machine states triggered and demanded by the material flow, the self-determined change of the machine states in production-free times (e.g., switching from idle to standby or from standby to off mode). ABS state charts are used to model the machine behavior. The agents can act actively on the micro level, the internal level processes are usually passive. The energy consumption process on single

96

5

Development of a Simulation-based Methodology …

Figure 5.2 Overview of the hybrid simulation approach for the energy consumption model of a production. Adapted from [DG2015, p. 1616]

machine level is again continuous in nature and can be represented using the SD paradigm. The interaction of the described levels is presented in Figure 5.2. Material flow and process changes on meso and micro level affect the total energy consumption of the production. Changes on the meso and micro level will directly affect the outcomes on macro level, e.g. the reduction of parts in the material flow on the meso level will result in a lower total energy consumption depicted on the macro level. Interactions between meso and micro levels take place, when the work flow entities (parts) run through the processing by the agents (machines). The agents can cause delays on the meso level when their internal state is not production ready as demanded by the workflow. On the other hand, the workflow can affect the agent’s behavior by not delivering entities and forcing the agent “overthink” their machine behavior due to process changes on meso level, making them behave as efficient as possible with regards to the overall objectives defined on macro level. Thus, the implementation in a simulation requires definition of three blocks, a production flow component, a machine component, and an energy component, all

5.3 Description of the M&S Approach

97

based on different simulation paradigms. The production flow component includes all setups and parameters that are linked to production planning and the production material flow to depict and parameterize the flow of entities through the production machines. The machine component contains all relevant process parameters and the internal process logic of all production machines involved in the manufacturing processes. While the inside of the machines is treated as black box in the material flow model, it is specified in detail in the machine component using the tools of the agent-based simulation, which is event-discrete regarding the time advance. The ideal way to represent the behavior of machines is to treat them as agents and map possible states and transitions with state-graphs. The energy component is used to model the energy consumption behavior of the single production machines over time, requiring a continuous simulation technique in form of System Dynamics. All modules are designed in a flexible way to be used for both, analyzing individual production lines or depicting entire production areas. For the modeling of production processes, the adequate level of detail must be selected. On the one hand, the model should not lose itself in detail and should only depict relevant influencing factors. On the other hand, it must be precise enough to obtain a realistic picture of the energy consumption behavior. The single components of the methodology as well as their interactions are described in detail in sections 5.3.1 to 5.3.4. To have a consistent visualization and description of individual sub-parts of the simulation module, a description of all module components shown in Figure 5.3 is done. All modules receive a unique name, indicating their functionality. Important process and information flows will be visualized in the modules as far as it adds value to the model conception. Parameters and additional requirements that are either prefilled or have to be typed in by the user will be defined per module. To support a holistic understanding of functionalities, all interfaces with involved inputs and outputs as well as occurring feedback loops will be specified for each module.

5.3.1

Production Flow Component

The production flow component describes the process-oriented flow of parts through machine components and machines and comprises relevant production flow and production planning parameters. The structure of the entity-based production flow model includes all main electrical consumers that make a significant and immediate contribution to the

98

5

Development of a Simulation-based Methodology …

Figure 5.3 Specification of the module structure

execution of the production processes. To meet the minimum requirements needed to depict the behavior of the production on the one hand and to keep the model as simple as possible on the other hand, supporting processes such as the transport and storage of material are initially only considered in the model with the time delay they cause, but not for the calculation of the energy consumption as their share of the total energy consumption is low compared to the production processes. The process flow can generally be modeled and visualized using the standard blocks of available simulation software solutions, as shown in Figure 5.4 using the software AnyLogic. The model of the considered production area must have a defined start and a defined ending. Source and Sink symbols define the boarders to upstream and subsequent processes. By using a source, behind which, for example, a delivery schedule can be placed, deliveries from the logistics or other previous production lines can be grouped together, regardless of whether the process under consideration is preceded by another process chain or a simple material supply with stocked or buffered goods. Thus, the material flow and the model boundaries are defined. The modeling of the material flow requires the specification of batch sizes and, associated therewith, the installation of buffers and blockers to ensure compliance of batch sizes in the production line. In connected production flows with multiple processing options, flow dividing and combining elements are implemented with a defined splitting logic.

5.3 Description of the M&S Approach

99

Figure 5.4 Example visualization of a production flow in AnyLogic

The number of parameters and data to be collected for model building is highly dependent on the model purpose and defined investigation targets. The more realworld data is included in the model, the more precise are the simulation results. Following the VDI standard 3633 for production simulation, the required simulation input data for the material flow simulation is divided into system load, organizational and technical data [VD2014a, p. 33].

Figure 5.5 Production flow and planning parameters. Adapted from [VD2014a, p. 34]

Figure 5.5 contains a list of the most important material flow parameters that should ideally be used to fully map a production flow. Using the abovementioned parameters, a realistic image of the production in the production flow component can be created (Figure 5.6).

100

5

Development of a Simulation-based Methodology …

Figure 5.6 Production flow component

5.3.2

Machine Component

The machine component is used to model the process control operations of the production and functions as a control unit for the different production states of the modeled production machines. It also comprises all relevant parameters and required information to assess the system behavior of single production machines. The machine component represents the operational inside of the machines shown in the material flow. For a realistic depiction of the machine behavior, which is the connecting factor for the representation of the corresponding energy consumption behavior, a model of the operational machine states during the different stages of the production processes needs to be included in the machine component. The machine behavior can be depicted using state graphs and is subsequently referred to as “machine logic”. The machine logic includes all possible operational machine states as well as allowed transitions between single states. To make the usage of the machine logic as simple as possible, three basic types will be defined which can easily be customized through various parameters and specified restrictions for each machine (Figure 5.7). Every machine enters its machine logic through the initial state, which does not consume any model time or energy. As described in section 3.2.3, all possible energy consuming operating states need to be considered. Depending on the operation complexity of the machine under consideration, the complexity type of the machine logic is chosen and can be assigned to the machine in the production flow model.

5.3 Description of the M&S Approach

101

Figure 5.7 Different complexity types of the machine logic

The production machines go through different operating states, whose time sequence and duration are affected and determined by technical requirements on the one hand and dependent on production quantities and time tables on the other. An exception is the failure state, which is usually entered unplanned. As machine failures are, unless they have a significant effect on the production performance, often overlooked when modeling a manufacturing system, Rohrer proposes four options to handle machine downtimes. They can either be ignored, which is probably the most chosen variant, they can be included by adjusting processing times, considered as constant values for time-to-failure and time-to-repair, or as a fourth option, statistical distributions2 for time-to-failure and time-to-repair can be used [Ro1998, pp. 525–526]. The energy requirements of the individual operating states of a machine can vary greatly, both in terms of height and curve progression. While some operating states might rather be constant over a longer period of time, others have a highly volatile power consumption profile. Some operating states, e.g., the warm up and setup state, have a defined length which is technically required to set the machine ready for production, while for others the retention time of a machine in that 2 Random failure behavior can be described by different probability distributions. “The exten-

sive use of the Weibull distribution in the interpretation and analysis of failure phenomena is mainly related to the fact that the shape of its failure rate curve depends on a single parameter” [TB2017, p. 51]. The interested reader is referred to the further literature of the authors Bertsche, Schauz, and Pickard and Trivedi and Bobbio [BSP2011, pp. 40–54; TB2017, pp. 46–65].

102

5

Development of a Simulation-based Methodology …

state is dependent on the production task and quantity, e.g., the producing state (Figure 5.8). The required parameters to model the machine behavior in a realistic manner, need to be included in the machine logic.

Figure 5.8 Machine logic and power consumption profile with varying production quantities

The production process parameters can be divided into technical and organizational data (Figure 5.9). To structure the machine behavior, possible machine states and allowed machine state transitions must be defined. Equally important is the definition of machine state restrictions, e.g., a minimum duration of a state. A machine might have to remain in off mode for at least an hour before rebooting the machine for production. These restrictions will be used for the optimization later. As the off state does not have any energy consumption, it might be the most favorable option for non-productive machine times when it comes to minimizing the energy consumption. Restrictions cause this fact to be different because the production break may not be that long, so lingering in standby or idle mode is the more appropriate choice in that case.

5.3 Description of the M&S Approach

103

Figure 5.9 Production process parameters

Machine Component Production Process Parameters Setup times Production/Cycle Times MTTR/MTTF Operational state restrictions Batch sizes …

(1 … n)

Machine logic

operational machine state

Figure 5.10 Machine component

Besides structural data, production data is required to picture the production processes. Important production parameters for a simulation are setup times, production cycle times, as well as data on performance, capacity, and batch sizes. Depending on the chosen method for taking machine failures into account, failure data on Overall Equipment Effectiveness (OEE), Meantime to failure (MTTF), and Meantime to repair (MTTR) are relevant for an exhaustive model. All components of the machine component are shown in Figure 5.10. While the production flow component as well as the energy component are modeled for the entire production system, the machine component has to be modeled for each

104

5

Development of a Simulation-based Methodology …

machine. Thus, the number n of required machine components in the simulation module equals the number of machines in the production flow.

5.3.3

Energy Component

The energy component is used to model the energy consumption behavior of production machines continuously for any moment in time. The modeling of the energy consumption is done using stock and flow elements of the system dynamics library of the simulation software. Every machine has its power consumption profile, which can be split up in sequences and then be assigned to different energy states. As described in section 3.2.3, the energy states are differentiated according to time and optimization aspects and can be split up into three groups, technically relevant operations, value-adding, and non-value-adding operations. For reasons of clarity, it makes sense to depict the energy flow using this grouping, before the consumption is summarized per production line and production area (Figure 5.11).

energy flow value-adding machine operations

energy flow technically required machine operations

total energy consumption

production line

energy flow non-value-adding machine operations

Figure 5.11 Machine-state dependent summation of single energy states to a total energy consumption

This means that the sum of all productive machine times is included in the energy flow which represents the value-adding machine operations, the energy flow for the technically required operations includes the warmup, fast warmup, and setup times of all machines, and the energy flow of the non-value-adding

5.3 Description of the M&S Approach

105

operations summarizes all machine times of standby or idle mode: n i=1

with

m i o per at i on = m 1operation + m 2operation + . . . + m n operation

m ioperation

(5.1)

energy consumption of machine operation of machine i in kW

The single power consumption profiles of the machines sum up to an energy consumption flow of a production line and finally to the total energy consumption of the entire production area. The load profiles of a machine can be created using three different options. They can either be measured and assigned using real consumption data, they can be represented in form of mathematical functions, or it is possible to work with individual energy values, which were formed as an average and are assumed to be constant over the duration of an entire energy state. The use of averaged values instead of the real power consumption profile is going strong at the expense of a realistic representation of the energy consumption but has the advantage of low data acquisition and storage efforts. The creation of mathematical functions to describe the power consumption profile requires a lot of effort in the process of function approximation or physics-based model building3 , but also brings the big advantage of simple storage options with it, since only single function parameters and no big data tables must be stored. Nonetheless, the predictive quality of mathematical functions is highly dependent on the approximation quality. The usage of value tables or table functions causes probably the most accurate prediction quality and the least creation efforts even though the required technical installations for measuring the data as well as the required memory to store the data is higher and the implementation of the data access in the simulation model might be more complicated. While the machine behavior is expressed using the machine logic with its machine states, the energy consumption behavior can be represented adding up different energy states over time. Every machine state has its corresponding

3 An example for physics-based model building can be found in [Si+2018]. Siegel et al. have

modeled the exact power consumption of a fork lift truck based on physical processes (e.g., the acceleration of the vehicle, the constant drive to the place of use as well as braking processes) that occur. He et al. use a combination of physics-based modeling and measured load profiles to estimate the energy requirements of the individual machine components and—in the end— summing them up to the calculate the total consumption of the entire production machine [He+2010].

106

5

Development of a Simulation-based Methodology …

energy state, containing all required information regarding the consumption behavior while the machine is lingering in a certain machine state. For this, continuous variables for the energy state are created by considering the ratio of the passed time in a state and the remaining time. Those variables are on the one hand required to be able to picture the consumption behavior at any moment in time, even in between single events of the event-based part of the simulation model. On the other hand, they are used for the optimized timing of machine startups in the production line, as they allow the calculation of the remaining time until a waiting machine has to be ready for its production process.

Figure 5.12 Energy Component

The energy component enables the depiction of energy-relevant information and thus forms the basis for influencing and optimizing energy-relevant production parameters. Summarizing, the energy component contains the SD energy flow, the energy state as well as the energy consumption parameters (Figure 5.12).

5.3.4

Interaction Point Definition

The three presented sub-models describe different aspects of a production. In order to combine the three models to a holistic production simulation, it is required to define all necessary interfaces. As already indicated in sections 5.3.2 and 5.3.3, the process and the energy component are connected via the machine and

5.3 Description of the M&S Approach

107

the energy state. With the defined system throughput, shift and working schedules, material requirements and production orders, the production flow component functions as a pacemaker for the entire simulation module. The production flow component is directly linked to the machine component. It triggers the machine logic with events which are planned or taking place in the production flow. Planned production orders for example affect the process logic of the production machines, as the machines must be in a production-ready state, as soon as the production order processing time starts. The machine logic is triggered again, when parts actually enter the individual machines of the production line.

events in the material flow component

production start variant A

behavior machine x

setup line for variant B

Producing

(machine component)

Warm up

Off

Set up

no raw material

Producing

Idle

Standby Off

energy profile of machine x (energy component)

time

Figure 5.13 Energy profile creation depending on the events in the material flow and the resulting machine behavior

The delay in material flow a machine causes is not defined in the DES material flow model, as it would be done in a purely process-oriented simulation model but is calculated summing up the process times of the single operating states of the machine logic. Example: Machine A is turned on at 6 am for production start and must go through a 35minute warmup process. Due to raw material supply issues, the raw material is brought to the line late at 6:45 am. The machine has a cycle time of 4 minutes per part and the lot size of the planned production order is 80 pieces. The delay machine A causes in the production flow before the first part is handed over to machine B for processing therefore sums up to 49 minutes instead of the originally planned 39 minutes.

108

5

Development of a Simulation-based Methodology …

Parallel to the machine state sequence running in the machine logic, the load profile of the simulated machines is generated in the energy component. The information about the completion of the machining process on a part must then be given back to the production flow component in order to trigger the next steps of the material flow, such as transferring the part from the machine to a parts buffer or the next machine accordingly. In principle, every event that occurs in the production flow component triggers a change of the machine and the energy state in the other two components (Figure 5.13). Likewise, the processes in the machine component influence the events of the material flow. In the machine logic, all steps of processing every individual part are shown. If the processing time of part x is over, this information is transferred to the production flow component. Subsequently, part x leaves the machine and the next part y is brought to the machine for processing (Figure 5.14). The two components are in a permanent exchange of information through which actions and reactions in form of events are created. While the lack of raw material forces the machine to stop producing and switch to an unproductive state, the change of states initiated by the machine, e.g., in the case of a machine breakdown, will unlikely interrupt the material flow and can postpone material flow events (Figure 5.15). The three sub-models can be combined to form a holistic production model (Figure 5.16). The energy consumption is closely linked to the machine behavior. The interface that has to be created by the modeler clearly consists of the assignment of the corresponding energy load profiles to the machine states. The machine conditions are in turn influenced by the material flow in the production line. Likewise, the behavior of the machines influences the occurrence of events in the material flow, such as the failure of a machine. To ensure a smooth interoperability of the sub-models, all interdependencies must be clearly defined and considered via dynamic parameters, functions, and restrictions in the simulation model. One possibility to realize a hybrid production simulation model and to implement the mentioned interfaces in one software tool is shown in 5.6 in a prototypical implementation.

5.4 Description of the Optimization Approach

109

Figure 5.14 Effects of events triggered in the machine component on the material flow

5.4

Description of the Optimization Approach

In the previous section, the hybrid simulation approach of a combined energy and production simulation has been outlined. In this section, an optimization module is developed which enables the simulation-based optimization of the energy consumption during non-value adding phases in production as well as the elimination of consumption peaks in general, using the simulation model as an

110

5

Development of a Simulation-based Methodology …

Figure 5.15 Effect of machine failures triggered in the machine component on the material flow

evaluation function. With the aim of optimizing the operations strategy for available machines, optimization scenarios are developed in a first step. Afterwards, the development of objective functions, constraints and parameters for the optimization, as well as the required interactions with the simulation module (depicted in section 5.3) are described in this section and demonstrated on the basis of two example models in section 5.6 and chapter 6.

5.4 Description of the Optimization Approach

111

Simulation Module Machine Brick

Production Flow Brick

DES material flow

Planning Parameters

Production Process Parameters Setup times Production/Cycle Times MTTR/MTTF Operational state restrictions Throughput times Batch sizes …

Production Flow Parameters Product data Production orders Maintenance cycles

Waiting times Quantities …

Energy Brick

SD energy flow

Machine process logic

Shift models Machines Transport systems Break arrangements Working schedules Restrictions …

Energy consumption parameters

operational machine state

energy state

> Mathematical functions > Value tables/ table functions > Energy state values

Figure 5.16 Conceptual structure of the simulation module

5.4.1

Energy Optimization Scenarios

This subchapter briefly discusses possible optimization scenarios and their design. As described in section 3.2.3, time-variable operating states can be divided into non-value-adding and value-adding states. While the length of value-adding states is dependent on the production quantity, non-productive states do not exhibit such correlation. They usually occur due to bad planning, unplanned events, or to bridge (short) planned waiting times. The length of non-value-adding states should be subject to a machine state optimization with the overall aim to generally eliminate them. The energy consumed during the lingering of machines in those states is often higher than the energy required for a warm-up after having switched off the machine. Especially during idle states, the machines have a quite high energy consumption level, as all machine parts are kept on a ready-to-produce-level. Since the employee’s awareness of the level of energy consumption of individual machine states has often not been created, machines stay in more energy-intensive conditions than necessary, especially when unforeseen events disrupt the normal flow of production. Only the recording of power consumption profiles leads to the possibility to compare the energy consumption of machine states over time periods to choose low-energy alternatives.

112

5

Development of a Simulation-based Methodology …

Figure 5.17 Machine logic and power consumption profile with varying idle times

Basically, conditions have to be formulated which clearly state under which circumstances which machine state is to be selected, which minimum time durations have to be taken into account in a machine state, and which states have to be run through in order to have the machine ready for production again at a desired time. The occurrence of non-value-adding machine conditions in production offers optimization potential, as it is demonstrated by the following example. The load profile of a production machine has been identified as shown in Figure 5.17. Due to material shortages and malfunctions in the upstream machines of the production line, the observed machine has lingered in an idle state for several hours. As the illustrated state graph shows, the machine can be set into a standby state, from which the system can be brought back into the idle state through a fast warm-up in a relatively short time period. If the length of the failure or production interruption is approximately known, it can now be checked whether it is possible to put the machine in the standby state and to bring back to an idle state in the time for the scheduled production start. tinterr upt ≥ tstandby + twar mup

(5.2)

5.4 Description of the Optimization Approach

with

tinterr upt tstate

113

duration of the production interruption time for lingering in a machine state (warmup, standby)

If the production interruptions are long enough to talk about machine state changes, the energy consumption during the different state combinations has to be compared in the next step. A change of states is worthwhile only if the energy consumption in the idle state is higher than the sum of the consumption of standby and the fast warmup status in the same time period.   tinterr upt × Pidle > tinterr upt − twar mup × Pstandby + twar mup × Pwar mup (5.3) with

Pstate

power consumption in a machine state (idle, warmup, standby)

If the time period t interrupt is long enough to bring the machine into a lowerenergy state for a fixed duration and the total energy consumption is lower than in idle state, this results in potential energy savings during the non-value-adding production times, which can be easily implemented (Figure 5.18). In addition to the standby state, it is also possible to check whether it is more energy-efficient to bring the machine in an off state and through the associated normal warmup process. The normal machine warmup, which is required as soon as the machine has been switched off completely, might be more energy intensive than remaining in standby state, the energy consumptions for the different runtime options need to be compared carefully. Thus, it can be checked for which production interruption length which variant has to be selected to realize optimization potentials. Besides the consequent optimization of machine states, peak load avoidance offers further optimization potential. As described in section 3.3, the network tariff component of the electricity tariff is calculated according to the maximum consumption peaks that occur. The avoidance of power peaks therefore has a direct influence on the guaranteed electric grid capacity and the network charge calculation. To detect and avoid peak consumptions, all electricity consumers in a production must be recorded with their individual consumption values. In addition, it is advantageous to identify non-critical machines that can be taken out of service in case of an imminent peak consumption. Performance peaks can usually be avoided already by a time-delayed scheduling of the main electricity consumers (Figure 5.19).

114

5

Development of a Simulation-based Methodology …

Figure 5.18 Energy savings through idle state avoidance

Figure 5.19 Consumption peak avoidance in production lines

5.4 Description of the Optimization Approach

115

Taking into account the scheduled production orders, the ideal time to start the warmup process can be calculated for each machine involved in the production process: tstar t M i = t pr odstar t M i − t war mupM i − t upstr eam with

tstar t M i t pr odstar t M i t war mupM i t upstr eam

(5.4)

switch on time for machine i planned production start for machine i duration of warmup phase of machine i remaining production time of upstream machines

Especially, for linked production processes with long process times, in which subsequent production steps must wait for the completion of upstream work, it makes sense to calculate machine starts. The consumption peak avoidance requires the control of individual power consumption profiles of machines to influence the overall power load. To use it as a tool to reduce the energy costs of a company, the availability of suitable electrical consumers that can temporarily be switched to different machine states is essential. Frequently, consumption peaks occur especially at the beginning of the week, as all systems in the production hall are switched on for the first production shift of the week. The machines than generally go through energy-intensive ramp-up processes at the same time. A rising of awareness of the workers for the demanddriven start-up of the machines at the right time is necessary to avoid peak loads. This requires the knowledge about durations of start-up phases and consumption profiles of the single machines. In large production companies, such planning usually cannot be done manually. At this point, the use of hybrid production and energy simulations is recommended to generate awareness of the relationships between peak loads, energy consumption patterns and production processes. A clear definition of objective functions as well as the implementation of constraints, such as minimum retention times in certain operating states, energy optimization problems can be defined and solved by a simulation of possible parameter variations. Even ideal machine start times can be determined by recalculations from the machine under consideration along the value chain to the current processing location of the part to be manufactured. Thus, machine state optimizations, peak load avoidance and timed machine start-ups form the basis for sustainably increasing energy efficiency in production processes by lowering

116

5

Development of a Simulation-based Methodology …

the total average energy consumption in a production as well as the maximum consumption level (Figure 5.20).

Figure 5.20 Status and peak load optimization of a four-machine scenario

If more than one objective function exists, it has to be defined which objective is pursued with which priority. This is especially important for the software implementation. It must be considered whether the objective functions are pursued in parallel in one optimization experiment, or if an optimization experiment per objective function is created. For the second option, each of the optimized parameters from experiments with higher priority have to be taken over as fixed variables in subsequent experiments. The exact implications of parallel and separate optimization experiments will be demonstrated using practical examples in sections 5.6.5 and 6.3.

5.4.2

Optimization Objectives

The optimization objectives are generally defined by the general management for the production and often include weighted preferences and prioritized criteria on the performance of single production areas. Optimal decisions are thus dependent on the defined objective function and are optimal only with regard to the underlying objective function itself. Following the scenarios described above, two

5.4 Description of the Optimization Approach

117

objective functions can be formulated to reduce the energy consumption in a production under the constraints that neither the height or the timing of outputs nor the quality of the produced goods should be influenced negatively. Firstly, the total energy consumption can be minimized by focusing on the energy savings through minimizing non-value adding production times of machines. For the minimization, it needs to be checked if all machines are in an efficient energy state regarding the current production task. The total energy consumption can easily be reduced through the avoidance of unproductive machine states which requires the exact definition of mandatory conditions and restrictions that have to come with the objective functions (section 5.4.3) Besides the total amount of the energy consumed, the guaranteed electrical grid capacity provides a starting point for optimizations. Since only a few peak consumptions per year ensure that the guaranteed grid capacity is increased, and so is the grid charge, consumption peaks exceeding the grid capacity should strictly be avoided. Therefore, the energy consumption flow of the production should be smaller than a defined maximum allowed peak value at any time. The maximum allowed peak value is based on the guaranteed grid capacity minus a safety factor to generally avoid exceeding the grid capacity. Threatening load peaks can be detected in advance by simulating production processes and thus precautionary measures can be initiated to avoid them. These precautions may include re-scheduling of energy-intensive warm-up phases of individual machines as well as, for example, the planned shutdown of ventilation and air-conditioning devices in unoccupied or sparsely occupied sections of the production hall over short periods of time. The effect of rescheduling individual machines or entire production lines to avoid peaks loads can be seen in the use cases in sections 5.6.4 and 6.3.

5.4.3

Optimization Parameters and Constraints

The optimization parameters are essential in order to create main control levers with which the energy consumption behavior of the production machines can be designed efficiently over the long term and under any load situation. Therefore, it is indispensable to include adjustable and dynamic parameters and variables for production processes and the production flow as well as planning parameters already when building the simulation model to be able to use them for optimization runs. Since the optimization parameters must be selected in a way that the optimization of the energy consumption is not to the detriment of the production output, it

118

5

Development of a Simulation-based Methodology …

has to be determined very precisely which parameters can be created for the optimization process in the simulation model. Additionally, optimization constraints are required to include given restrictions and limitations of the real production system in the optimization process. The consideration of important constraints is usually done using the simulation model to depict side conditions. Generally, it is easier doing this depiction in the model rather than formulating systems of inequations. At this point, the example of the already mentioned minimum retention time in a machine state is used again. The minimum retention time may be required to complete cooling processes or system shut downs before the machine can be restarted again. In the simulation model, the minimum retention time can be inserted easily into the state graph of the machine logic by dividing a state into a fixed and a variable part. The minimum retention time thus becomes an element of the machine logic that must be run through within the given state changes as soon as the machine switches into this state. The formulation of equations for this restriction is much more complicated. It is necessary to clarify which parameters are included in the formulation of the restrictions, how they can be determined exactly, which range of variation and which units of measure they have. Furthermore, it must be defined under which conditions the state may be left again, etc. This makes the formulation of the restrictions in a mathematical model very time-consuming and complex. The definition of restrictions should always be carried out in parallel with the definition of the optimization parameters, as the value range of the parameters is directly limited by the restrictions. The scheduling of orders for example has a direct impact on the production output of a shift, start times of production orders should therefore not be used as a parameter for increasing energy efficiency, unless there are clear restrictions defined, limiting the range of allowed changes for the order rescheduling by a few minutes forwards/backwards to avoid load peaks for example. The same applies to the use of the optimization parameter “machine state”. Soberly, the idle state is always less energy intensive than the productive state, as well are the standby or even the off mode. Without the definition of restrictions and conditions under which machine state changes may take place, the whole optimization process does not work. Without the clear specification in which order which machine state can be reached, which states have to be passed through to get into the productive state and which minimum retention times must be met in certain states during the non-value-adding machine times, a minimization of the total energy of a production would always lead to putting machines into the off state immediately after completing a production job. A detailed example description for the definition of optimization parameters and constraints will be given in the fictional case study as well as in the practical

5.4 Description of the Optimization Approach

119

example as the non-case-related description is tedious and not goal-oriented. The parameters will be defined according to the optimization goals. For an optimization of the total energy consumption, parameters are needed that offer starting points for changing the machine behavior to a more efficient manner, whereas for a peak optimization, parameters are needed that influence the occurrence of load peaks of individual machines and can delay them for instance.

5.4.4

Interdependencies between the Simulation Model and the Optimization Process

The general interdependencies between the simulation model and the optimization methodology have been described in theory in section 2.3.2 of this book. By simulating different system configurations, the optimization methodology searches for the most influential process inputs and their ideal input values to optimize the process outputs of interest. The simulation-based optimization process is shown in Figure 5.21.

simulation result retrieval Start

feasible solution?

no

pool of unfeasible parameter configurations

yes

Simulate specified system configuration

best parameter configuration

parameter optimized simulation model

no

is stopping rule satisfied?

pool of feasible parameter configurations

specify parameter configuration

update best feasible solution

yes

Optimization Module

Simulation Module

Figure 5.21 The process of simulation-based optimization

The simulation is started by the optimization using an initial parameter configuration, which are random numbers picked of the allowed range for the parameters that have been defined for optimization. Does the model contain stochastic input parameters, the simulation is run a defined number of replications for each parameter set. Is the model free of stochastic variables, the replications are not required.

120

5

Development of a Simulation-based Methodology …

Simulation Module Machine Component (1 … n)

Production Flow Component

DES material flow

Production Process Parameters

Planning Parameters

Setup times Production/Cycle Times MTTR/MTTF Operational state restrictions …

Production orders Shift models Quantities Working schedules …

Production Flow Parameters Waiting times Quantities …

Product data Resource Assignment Maintenance cycles

Machine process logic

SD energy flow

Energy Component

Energy consumption parameters

operational machine state

energy state

> Mathematical functions > Value tables/ table functions > Energy state values

Optimization Parameter

run simulation model retrieve simulation results Feasible Parameter Configurations

Idle Optimizer Standby Optimizer Offset Parameter

Optimization Constraints

Best Feasible Parameter Configuration

Objective Functions Unfeasible Parameter Configurations

Optimization Module

Figure 5.22 Conceptual model for the simulation-based optimization of energy efficiency

To find the optimal solution, the simulation model is run a defined number of iterations. For every iteration the set of parameters is varied, the produced simulation results, not all of them improving or feasible, provide a trajectory to the best solution. Is a parameter set complying with all constraints and restrictions, it is counted as a feasible solution. All feasible solutions are then looked through for the best parameter set, guaranteeing the most efficient production set up for the simulation model. The optimization experiment thus provides the best possible values for the decision variables in regard to a selected objective function. This results in the conceptual model for the simulation-based optimization shown in Figure 5.22. Following, the software selection as well as a prototypical implementation will be described, using a production scenario that has slightly been adapted for

5.5 Software Evaluation and Selection

121

academic use to be able to clearly evaluate the applicability of the developed methodology.

5.5

Software Evaluation and Selection

The modeling and simulation method as well as the optimization algorithm for the improvement of systems is highly dependent on the dynamics of the system examined, the analysis objectives, as well as on the software environment used. This chapter introduces selected simulation systems that are suitable for combined simulation in order to select an appropriate tool for implementing the developed methodology. In principle, it has to be differentiated between simulation systems that enable hybrid simulation in one single software solution, so-called multi-method modeling simulation tools, and simulation systems that implement a hybrid simulation through the coupling of multiple simulation tools. The aim of this work is to implement discrete and continuous simulation in combination with optimization experiments in one software tool without the requirement for challenging interface management and complex data exchange or synchronization demands between different software applications. For this reason, only simulators that are capable of multi-method modeling are considered below. The software most frequently used in production simulation, especially in the automotive industry, is Tecnomatix Plant Simulation. Starting in version 11, Plant Simulation supports a simplified representation of energy consumption in the production by using object attributes for active material flow elements [Ba2015, p. 599]. Thus, the energy consumption for some machine states can be specified as constant values. However, minimum retention times in certain machine states, restrictions on machine state changes, as well as the dynamic machine behavior cannot be modeled by the use of static objective attributes. Plant Simulation is a purely discrete simulation tool, which is not capable of multi-paradigm simulation. As the realistic depiction of the energy consumption behavior of production machines “demands certain flexibility and freedom which cannot be provided by mature, large scale simulation tools like Plant Simulation” [Th2012, p. 97] this tool will not be considered in the context of the development of the simulation-based optimization methodology in this book.

122

5.5.1

5

Development of a Simulation-based Methodology …

Overview of Multi-method Simulators

Different commercial simulation packages enable the development of hybrid or combined models by allowing the combination of several simulation paradigms. “With new advancements in […] simulation software supporting multi-paradigm modeling it is becoming increasingly easier to create hybrid simulations consisting of combinations of DES, SD and ABS” [Ta2014, p. 2]. While some software tools focus on one simulation paradigm but provide extension features for other techniques, other software solutions include multi-method modeling by providing all modeling paradigms in flexible, domain-specific libraries [DG2015, p. 1617]. MATLAB, Arena, ExtendSim and WITNESS Horizon are examples of software solutions that can be extended by different tool boxes for combined modeling while AnyLogic enables combined simulation by merging several models based on different world views in an object-oriented environment [Ru2018, p. 38; He2008, p. 50]. “AnyLogic naturally supports the interaction of stock and flow structures with events, state charts, process flow charts, and agent populations. AnyLogic Simulation engine is a hybrid engine designed for efficient and accurate simulation of continuous dynamics being interrupted by a large number of discrete events” [Bo2013, p. 195]. MATLAB4 is a simulation tool originally developed for continuous models to solve problems via matrix-based programming language [An+2017, p. 69]. Extended with the MathWorks toolboxes Simulink and Stateflow, MATLAB can be used for modeling, simulation and analysis of linear and non-linear systems having a discrete or continuous time advance [AGA2016, p. 4]. Arena is a mainly discrete-event simulator by Rockwell Automation, which offers additional toolsets to model and analyze discrete and continuous systems based on the SIMAN simulation language [AGA2016, p. 4; Ba+2005, p. 110]. ExtendSim and Witness Horizon can also be considered as discrete simulators with additional libraries for discrete, continuous and combined model generation. While Extend combines a block-building approach in combination with an environment to develop new blocks [Ba+2005, p. 111], Witness Horizon offers numerous elements for discrete part-manufacturing in combination with flow elements to simulate fluids in processors, tanks and pipes [Ba+2005, p. 115].

4 MATLAB,

developed by MathWorks, is the abbreviation for Matrix Laboratory, and as the name already indicates, the strength of MATLAB are especially the vector and matrix calculation [An+2017, p. 1].

5.5 Software Evaluation and Selection

5.5.2

123

Software Selection

In addition to the commercial simulators mentioned in the previous section, there are many other software solutions that have been developed at research institutes and universities but are rarely used outside these institutions. AnyLogic and MATLAB are the most widely used software systems in the simulation community for creating hybrid simulation models. Therefore, and due to the reason, that both simulators include options to perform simulation-based optimization, these two simulators are described in more detail in this section. Subsequently, a simulator is selected for the prototypical implementation of the methodology developed in sections 5.3 and 5.4. Many scientists and users claim that AnyLogic is still the only commercial simulation software developed from the beginning to create hybrid models using discrete and continuous simulation paradigms [Br+2019, p. 722; Ga+2018, p. 25; An2019]. AnyLogic allows the use of individual simulation paradigms as well as any combination of paradigms in one model. The software includes different options to connect discrete and continuous model parts5 . Through the use of state charts, the value of continuous variables can be monitored to trigger events causing state transitions [Bo2013, p. 244; He2008, p. 51]. Furthermore, AnyLogic provides dynamic event objects, which can be used to “schedule a discrete event based on a condition associated with the value of some model variable that might be a continuous variable, or based on time” [He2008, p. 52]. If a condition is fulfilled, the event triggers the state change of the model as defined in the event. AnyLogic provides a third way for the modeler to link discrete and continuous model parts by treating continuous components as variables instead of objects. Thus, stock and flow elements of the SD library can be addressed in the actions coded for DES objects, in dynamic events, as well as in agent-based state charts. In reverse, attributes of DES objects can be referenced in the mathematical formulations of the SD variables [He2008, p. 52]. AnyLogic is controlled by an event-discrete engine but includes a numerical solver to perform differential equations to run combined simulations (Figure 5.23). “This offers a potential solution in providing interoperability and compatibility between system time-driven and event-driven simulations within a single environment” [Ga+2018, p. 25].

5 At this point, only a short overview of the linking options of discrete and continuous aspects

in AnyLogic is given. A detailed description can be found in [Bo2013, pp. 241–257].

124

5

Development of a Simulation-based Methodology …

no

AnyLogic Model

threshold crossed or condition met yes

update variable value

continuous variable values

update variable value and invoke a discrete change event

Discrete Engine system of algebraic differential equations

Equation Solver

Figure 5.23 AnyLogic’s hybrid simulation engine architecture [He2008, p. 53]

AnyLogic has a graphical interface and combines two approaches for model generation. On the one hand it is a drag-and-drop solution with configurable building blocks and dialog boxes to simplify and accelerate the modeling process by means of libraries. On the other hand, it is possible to program user-defined settings, integrate functions and test conditions and evaluations with some expertise in java-code writing being required [Ru2018, p. 41; Br+2019, p. 722]. MATLAB was first used in the 1970’s and has developed to a software having a large number of built-in toolboxes and thus attracts a variety of users of different areas from engineering to applied sciences [Tu2019, p. 1]. Completed by the graphical programming environment Simulink, MATLAB/Simulink is a tool for the analysis “of linear and nonlinear systems modeled continuously (continuous-time), discrete (time steps) or hybrid” [AGA2016, p. 4]. Simulink is completely integrated in MATLAB and enables the use of pre-defined blocks from included libraries for discrete and continuous modeling as well as the use of user-defined elements through three included tools, SimEvents, Stateflow, and Simscape (Figure 5.24). By providing blocks for queues, resources, delays, and other elements for visualization and data processing, SimEvents has been designed to simulate DES while Simulink is actually a time-driven simulator. Through the integration of SimEvents, MATLAB/Simulink is equipped with functionalities that enable “an

5.5 Software Evaluation and Selection

125

MATLAB Simulink

SimEvents

Stateflow Simscape

Figure 5.24 Integration of MATLAB libraries [Ru2018, p. 42]

effective co-existence of time-driven and event-driven components in complex hybrid systems” [AGA2016, p. 5]. Stateflow adds building blocks for modeling and simulating decision logicbased state machines and flow charts. Thus, the reaction of systems to events and time-based conditions or external signals can be modeled [Ma2019]. The library of Simscape includes physical blocks and circuit diagram components. Including these features, Matlab/Simulink offers a platform for multi-paradigm modeling of complex systems. Both, AnyLogic and MATLAB, offer the possibility of simulation-based optimization. While the OptQuest optimizer is integrated in AnyLogic, MATLAB accesses a number of heuristics and metaheuristics, that the modeler can adapt and enhance as required (section 2.3.3). Therefore, MATLAB has more extensive analysis options than AnyLogic. Like AnyLogic, MATLAB/Simulink is not a pure building block environment but offers the option of user-specific extensions. MATLAB is rather an open programming environment for which programming knowledge is not only an advantage, but also indispensable for professional handling [Ru2018, p. 43]. Since the model generation and the setup and execution of analysis tools in AnyLogic is possible with little programming knowledge, AnyLogic meets the criteria of simplicity, applicability, and usability from section 5.1 much better than MATLAB. Because of its library with process-oriented building blocks it is also well suited

126

5

Development of a Simulation-based Methodology …

for the easy duplication of material flow processes. Since the use of the methodology in practice strongly depends on the usability of the simulation software and the model generation should be possible without experts and computer scientists, AnyLogic is used for the implementation of the fictional case study (section 5.6) as well as the practical use case (chapter 6).

5.6

Prototypical Implementation

The implementation of the simulation methodology demands a flexible simulation tool which supports multi-paradigm simulation. AnyLogic 8 of TheAnyLogicCompany6 is used for the prototypical implementation7 of the simulation-based optimization methodology.

5.6.1

Description of the Fictional Production Scenario

The prototypical implementation is done using an anonymized production scenario from a foundry, focusing on the final processing steps of pre-fabricated parts. The production scenario comprises a production line for the mechanical processing of large die-cast parts with a total of five machines. First, the parts are machined on one of the two CNC (computerized numerical control) machines. Subsequently, necessary holes and threads are drilled in the drilling machine, followed by a sandblasting process. At the end of the production line, the parts are grouped in lots of twelve parts and cleaned in an industrial washing machine (Figure 5.25).

Figure 5.25 Fictional production scenario of a die-cast part processing line 6 All

simulation models are built up in the Personal Learning Edition AnyLogic PLE Version 8.5.1. 7 Extracts of this section have already been published at the Winter Simulation Conference 2019 [RS2019].

5.6 Prototypical Implementation

127

The production line is operated in a two-shift model, from 6 am to 2 pm and 2 pm to 10 pm, with ten shifts per week. For all five machines, a preventive maintenance schedule exists with weekly occurring services. The produced quantities are highly dependent on the customer’s orders, but a maximum quantity of 60 parts per shift cannot be exceeded. The assignment of the machine logic types is done according to the complexity of the production machines following the suggestion in section 5.3.2. Processing times8 , durations of technically required machine states and other relevant key figures used in the simulation model can be found in Table 5.1. Table 5.1 Process parameter of the die-casting production machines CNC 1

CNC 2

Drilling

Sandblasting Washing

High complexity

High complexity

Medium complexity

Medium complexity

Low complexity

Process time 14 mins

13 mins

4 mins

4 mins

90 mins

Capacity

1 part

1 part

1 part

12 parts

Machine logic type

1 part

Preventive 40 mins/week 40 mins/week 40 mins/week 60 mins/week 120 mins/week maintenance Warmup

31 mins

27 mins

19 mins

23 mins

–

Fast warmup

13 mins

17 mins

–

–

–

Output quantity

Depends on the customers demand → production schedule

Shift model

Two shifts a day summing up to ten shifts per week, operation time 6 am to 10 pm.

The assignment of the power consumption profiles to the individual machine states is done manually. The database for the scenario has been specially created, whereby the data are in principle based on the course of real production data. The data has been adapted to make differences in of machine states changes more clearly and recognizable. This procedure was deliberately chosen to make errors in the construction of the simulation model and the selected optimizations identifiable and transparent. The data available for the case study has a resolution of one measured energy consumption value per minute. The testing of different resolutions does not make any sense for the fictional example due to missing 8 Due

to the rather long process times in the described production line, the simulation model time in AnyLogic is set to minutes.

128

5

Development of a Simulation-based Methodology …

reference profiles and will therefore only be discussed in the practical use case to provide information about the ideal data depth (section 6.1.2). The hybrid model is built up in AnyLogic using the process modeling and the SD library as well as the agent components including state charts. Several variables, parameters and functions are in use to capture relevant interactions between the different model levels.

5.6.2

Simulation Model Details

The material flow through the five machines is depicted using buffer and delay elements from the process modeling standards (process-oriented DES). As the hybrid simulation methodology comes with a flexible calculation of the actual delay the machine causes depending on the machine states the machine is going through, the standard delay object is linked to a machine logic, reflecting the complexity of the machine. Therefore, the delay type of the delay object itself is defined to cause a delay until a ‘stop delay function’ is called. The delay time starts when a part enters the machine. On part enter, a ‘process part function’ which is defined in the machine logic is called by the delay object. At this point, there is a simulation paradigm change in the model. While the entry of the part into the machine is modeled using the process-oriented discrete event simulation paradigm, the simulation of the machine behavior is agent-based. Considering required warmup durations as well as the process duration itself, the delay sums up to the duration of single machine states until the ‘stop delay function’ is called with expiration of the production process duration. The machine delay ends, again there is a simulation paradigm change back to the process-oriented DES, the processed part leaves the machine for the next production step or the finished goods buffer, and the machine remains in an idle state until the processing of the next part starts (Figure 5.26). The production schedule as well as shift times and production shut downs due to maintenance processes are defined in the source element as well as in additional functions closing the machine activity at 10 pm and starting the morning shift again at 6 am using a time dependent event to start and to end the machine activity. The event calls a general function in the main agent that again triggers shift preparation processes as well as shut down functions in the single machine logic types (Figure 5.27). It is assumed that all machines can be switched off completely during the nights and weekends and thus do not consume energy during production free times. The model does not contain stochastic elements, as the fictional case does not require any. However, the simulation components are

5.6 Prototypical Implementation

129

machine delay object

part x

raw material

DES ABS

part x

delay type: delay until ‘stop delay’ is called

part waiting = true

on part enter: call ‘process part function’

machine logic

off warmup duration

stop delay for part x

start delay calculation part x

warmup idle material flow

producing process duration

setup definition duration delay calculation

process part function process time setups

call ‘stop delay’

process variables

simulation paradigm change

Figure 5.26 Machine delay calculation logic

designed to introduce stochastics through the use of failure probabilities, varying delivery schedules, or a mix of products. Often there are machines in production (line) that perform the same processing steps. In the fictional example this applies to the CNC1 and CNC2. Depending on the utilization situation of the production line, it is not necessarily required to use both machines. Therefore, a ‘select output element’ is used to define the part distribution9 . Thus, it is calculated whether it is more efficient to use both machines, e.g., to start the second machine in case only one is running. In this way, it can be determined (in case of poorer machine utilization), how the parts to be produced ideally pass through the production line in an energy-optimized manner. 9 The control of the raw material distribution between the machines will be relevant for the opti-

mization as the CNC machines have different processing times, different energy consumption profiles together with distinct machine state durations.

130

5

Development of a Simulation-based Methodology …

time dependent

events call general function

general functions call specific functions defined in every machine logic

Figure 5.27 Starting and ending of production shifts

Special parameters ensuring, that only one part at a time is processed for CNC1, CNC2, the drilling and the sandblasting machine, as well as the option to use blocker for batch processing for the washing machine are considered depending on the defined batch sizes for the single production process steps. With the assignment of the machine logic type to a machine, the number of energy parameters is defined, following the number of possible machine states. At this point, it is either possible, to use table functions in AnyLogic to work with realistic energy load profiles (Figure 5.28), or the energy consumption parameters can be defined as mean values being used for the whole duration of the single machine state. The total energy consumption of the production line is calculated summing up the energy flows of the single machines over the time period simulated. The current energy flow can be specified with an exact value at any time. To ensure that an energy value is documented for each time unit depending on the selected

5.6 Prototypical Implementation

131

CNC1: Machine Logic Properties

Figure 5.28 Machine set up for the depiction of realistic energy load profiles

model time of the simulation model, each status has a timer that performs this step as long as the machine is in the respective state (Figure 5.29).

Idle

Figure 5.29 Definition of internal state transitions

Through the combination of different simulation aspects in one holistic model, the depiction of all relevant production details to define the basis for production optimization becomes possible (Figure 5.30).

132

5

Development of a Simulation-based Methodology …

Figure 5.30 Implementation of the individual simulation focuses in an overall model

5.6.3

Adaptions of the Simulation Model in Preparation for the Simulation-based Optimization of Energy Efficiency

In order to determine the optimization potential of the energy consumption of the described production line, the simulation model has been extended by applicable restrictions and required parameters concerning the output quantity of the production, allowed deviations from a given schedule, and mandatory retention times for single machine states. The output quantity of the production must not be changed noticeably. For this reason, the optimization proposals cannot be at the expense of the output quantity, following the optimization of the total energy consumption of the production cannot be induced by a reduction of productive times, but only by avoiding nonproductive machine times. Since a delay of more than six minutes in production

5.6 Prototypical Implementation

133

causes a change of the output, delays are permitted only within a time frame of six minutes. The optimization of the production line requires the consideration of restrictions imposed by the production scenario. The CNC1 and CNC2 machines as well as the drilling machine need to stay in off mode for a fixed period of time before they can be restarted again. This defined time is technically required to shut down the machine completely, to empty all compressed air connections and to allow technical components to fully cool down. Since this restriction directly influences the course of the machine logic, it makes sense to incorporate it here in the form of a mandatory status that must be passed through. Therefore, the offstate of the affected machines is split up into a fixed and a variable component. Once, the machines enter the off state, they remain in the fixed off-component until the mandatory retention time has passed (Figure 5.31).

initial state ` off variable

off fix retention time

Figure 5.31 Implementation of the retention time in the machine logic

The machines then switch into the variable off-state from which it is possible to leave any time. In order to cover the case that a minimum retention time can also be specified for the standby state, the logic is implemented not only for the off state, but also for the standby state in the machine logic of the high and the medium complexity machines. For all machines, that do not have any restrictions regarding the switching of states, the retention time is set to zero. Aiming at the efficient use of already running production machines, for the case that several machines are performing the same production step, a ‘select machine function’ is used to check the status of machines and select an already

134

5

Development of a Simulation-based Methodology …

running machine for the processing of new raw material parts instead of starting another machine. The ‘select machine function’ is checking the status of machines before the raw material is assigned for processing (Figure 5.32). In case of the CNC1 and CNC2 the function checks the following cases: if both machines are in off state, the throughput time estimation is used to calculate which machine is available faster to serve the production schedule best in case parts need to be processed. Is CNC1 already producing and new raw material is available, machine CNC2 will be started and vice versa. The last check covers the case that CNC1 and CNC2 are already running and new material needs to be assigned to one of the two machines. Here, the ‘select machine function’ compares the buffer sizes and assigns the material to the machine with the lower buffer stock.

Function

Figure 5.32 ‘Select machine function’ logic

To avoid an early machine start of those machines, that are not required during the first production steps, an automated calculation to determine the remaining time until a machine has to be ready for production is added to the model. By defining a so-called ‘throughput time estimation function’, the time is determined

5.6 Prototypical Implementation

135

subsequent machines have until they must be ready for use depending on the production progress of their predecessors. The ‘throughput time estimation function’ calculates the remaining time of a part in a machine depending on the machine state and remaining required process steps a part has to go through before it is able to leave the machine for the next following machine (Figure 5.33). While the throughput time estimation is done for every machine in the production line, a ‘prepare machine function’ is added to all machines after the initial processing step, which are in the fictional example the drilling, the sandblasting, as well as the washing machine. The ‘prepare machine function’ uses the calculated throughput time of the CNC1 and CNC2 to determine the exact start of the drilling machine, considering that the drilling machine might need to go through a warm up process, which has to be finished when the first parts from CNC1 and CNC2 arrive in the drilling buffer.

Figure 5.33 Throughput time estimation

136

5

Development of a Simulation-based Methodology …

Depending on the state of the machine, the ‘prepare machine function’ calls either the ‘prepare from off function’ when the drilling machine has not been switched on already, it calls the ‘prepare from standby function’ for the case that the drilling machine is in standby state, or it calls the ‘prepare from idle function’ unless the current state is the idle state to calculate the estimated time for the drilling machine to be ready to act (Figure 5.34). By calling the various functions, it is ensured that the machines of the production line are ready on time but are not switched on and consuming energy before the planned use. The exact timing of the machines holds high savings potential, especially for machines that are needed late in the production process. The above described functionalities are added to the simulation model in preparation to define the most efficient production line setup in the optimization experiments. In addition to the optimization potentials mentioned above, there are further approaches on the way to energy-efficient production. Following, two are briefly discussed below but not dealt with in more detail. If several products and product variants are produced on a production line, an energy-optimized order scheduling can be used to generate a further setting lever for optimizations. Since the different products manufactured on the fictional line all produce very similar energy consumption profiles, an energy-optimal order scheduling is not relevant for this case study. The same applies to potential savings that can be realized by planning energy-intensive production processes during the night shift in order to be able to use the more favorable electricity conditions from 10 pm to 6 am. The author is aware of the fact, that these potentials for optimization exist and may be used in practice, but they are not covered in this book.

5.6.4

Depiction of the Optimization Potential

Following the optimization scenarios in section 5.4.1, two main objective functions for the production line are defined. Firstly, the total energy consumption is minimized with a clear focus on the elimination of non-value adding production times. Secondly, occurring power consumption peak values should be minimized. To find the minimum total consumption value Etotal for the total energy consumed by the production line in a four-week cycle, a new optimization experiment in AnyLogic is defined. The objective function is defined in the optimization experiment settings. As it is not possible, to include a stock element from the SD library at this point, a

5.6 Prototypical Implementation

137

Off

Standby

Idle

Figure 5.34 Prepare machine functionality

detour is taken inserting a function ‘totalEnergyUsed’ which returns the totalEnergyConsumption value of the stock element (Figure 5.35). The objective function is: min E total =

 t

(E C N C1 + E C N C2 + E Drill + E Sand + E W ash )

138

5

Development of a Simulation-based Methodology …

Figure 5.35 Objective setup of total energy consumption optimizer

with E C N C1 =



E war mup + E pr oducing + E idle + E standby + E f astwar m + E manual + E f ail

t

E C N C2 =



E war mup + E pr oducing + E idle + E standby + E f astwar m + E manual + E f ail

t

E Drill =



E war mup + E pr oducing + E idle + E standby + E manual + E f ail

t

E Sand =



E pr oducing + E setup + E f ail

t

E W ash =



E pr oducing + E setup + E f ail

 



 

(5.5)

t

To determine the minimum total energy consumption, parameters are introduced, with which the different energy consumption scenarios are designed. With a focus on the non-productive machine states, it needs to be assured that all machines are in the most energy efficient production state allowed by the production tasks. Machines that are currently not producing need to be brought into a lowenergy state for a defined period of time to reduce the overall consumption without a negative influence on the total production output. The optimal duration of nonproductive machine states has to be calculated. The calculation is not trivial, since shutting down a machine can be followed by long and energy-intensive startup times before the machine can return into a productive state again. Thus, the prescribed state sequences of machines do not automatically make it most optimal to completely avoid non-productive machine states. Due to technically necessary warm-up phases after standby and off-state, it may be more appropriate for the overall energy balance to stay in idle state to bridge short production interruptions. The optimum times when a change of state makes sense can be simulated and determined during the optimization experiment. Therefore, the introduction

5.6 Prototypical Implementation

139

of optimization parameters is required for those machines that have non-valueadded states in their machine logic. Machines operating on a low complexity machine logic do not contain standby or idle states and thus do not require optimization parameters. The medium and the high complexity machine logic types comprehend non-productive states (standby and idle), which can be used to bridge short production interruptions. With the introduction of optimization parameters for exactly those states, it is possible to determine the period of time that these non-value-adding states should be used for bridging production interruptions and as of when a shutdown of the machines is the more energy-efficient variant. As an example, the parameter optimizerCNC1idle determines the length of the time period for which it is more energy efficient to remain in idle state than to switch off and later restart the machine for production again. Table 5.2 gives an overview of the required optimization parameters for the total consumption optimization of the fictional case study. Table 5.2 Optimization parameters for the reduction of the overall energy consumption Machine

Idle State Parameter

Standby State Parameter

CNC1

optimizerCNC1Idle

optimizerCNC1Standby

CNC2

optimizerCNC2Idle

optimizerCNC2Standby

Drilling Machine

optimizerDrillIdle

optimizerDrillStandby

For the fictional use case, it is defined that production interruptions of 15 minutes and less will not cause a change of machine states as the machines have warmup and fast warmup phases which exceed 15 minutes. Corresponding to the restrictions given in the practical scenario, it is determined here as well that for production interruptions of more than 60 minutes, the machines must be put into an energy-saving state. It has been mathematically proven for all machines in the production line, that this instruction is reasonable, since keeping the machines in idle state for 60 minutes and more is more energy-consuming than switching them off and running through a warmup again. While the CNC1 consumes 2.064 kWh during 60 minutes in idle state, it requires only 826 kWh for the warmup process to get the machine production ready again. From these specifications, the limiting values for the parameters automatically result and the limits within which the optimization experiment varies the parameters are defined. To be able to compare the energy consumption values of the production line, the optimization experiment is using specified dates to start and

140

5

Development of a Simulation-based Methodology …

Figure 5.36 Parameter and model time setup for the optimization experiment

5.6 Prototypical Implementation

141

stop the simulation. The observation period is set to one month in the experiment properties (Figure 5.36). AnyLogic offers the option to define constraints and restrictions for the optimization experiment. While constraints are tested before the simulation run is started, restrictions are tested afterwards and depending on the outcome of a test the simulation run is counted as feasible or non-feasible solution. The restrictions are important for the test case, since it has to be avoided that any energy savings are achieved at the expense of a lower output quantity. Thus, only parameter variations for an output quantity of 2.100 pieces10 or higher are counted as feasible solutions. Since the defined parameters all have a comparatively large variance range, it makes sense to run through a high number of iterations in order to determine an ideal parameter set for the given objective. As the number of iterations in AnyLogic’s PLE is limited to 500, this maximum allowed number is used. The setup of replications is not required in this scenario because the model does not contain any stochastic elements. When all parameters and setups have been defined the optimization experiment can be started to depict the ideal switching times for the defined parameters. The optimization experiment contains a graphical analysis of the tested parameter variants, showing feasible, infeasible and current parameter sets. After the completion of all iterations, the best parameter set can be directly copied into the simulation model setup to run the simulation model with the optimal parameter setting (Figure 5.37). Besides the minimization of the total energy consumption, the optimization of power consumption peaks is required for the total optimization of the production line. Load peaks are caused by an unfavorable clash of the energy consumption profiles of the individual production machines. At those points, the energy flow P f low has load peaks, which should be minimized. The objective function is formulated as follows: min P f low =



P f lowC N C1 + P f lowC N C2 + P f low Drill + P f low Sand + P f lowW ash



t

with P f lowC N C1 =



P f lowwar mup + P f low pr oducing + P f lowidle + P f lowstandby + P f low f astwar mup + P f lowmanual + P f low f ail

t

P f lowC N C2 =



P f lowwar mup + P f low pr oducing + P f lowidle + P f lowstandby + P f low f astwar mup + P f lowmanual + P f low f ail

t

P f low Drill =



P f lowwar mup + P f low pr oducing + P f lowidle + P f lowstandby + P f lowmanual + P f low f ail

 



t

10 The output quantity of June, July and August has to be between 2100 and 2200 pieces a month to be able to serve the customer’s requests.

142

5

Development of a Simulation-based Methodology …

Figure 5.37 Optimization experiment visualization in AnyLogic

P f low Sand =



P f low pr oducing + P f lowsetup + P f low f ail

t

P f lowW ash =



P f low pr oducing + P f lowsetup + P f low f ail

 

(5.6)

t

For the peak avoidance, the production planning and control allows the rescheduling of single machines between three seconds (0,05 minutes) and three minutes. Therefore, a new optimization parameter, the peak optimization offset, is added to the model to be able to influence the start of production machines. The offset parameter is used to vary the machine state within the given range and has to be added to the machine logic, as it is directly influencing the machine control. To realize the offset, an internal off delay state is added to the off state of the machines. The off delay is entered when the machine gets the message to start the warmup process and thus leaves the variable off state. The off delay is linked to a timeout, meaning the machine leaves the off delay for the warmup process after the timeout duration has passed. The timeout duration is defined as a variable parameter, the peakOptimizationOffset, which is then used for the PeakConsumptionOptimizer experiment (Figure 5.38).

5.6 Prototypical Implementation

143

initial state Off off delay

off variable

off fix retention time

offset time

Figure 5.38 Insertion of the offset time for shifting of machine starts

The wanted effect of the offset usage for a machine in the production line is shown in Figure 5.39. By shifting one machine by three minutes, the consumption flow values sum up differently, and peaks can be avoided.

Figure 5.39 Depiction of the offset effect for one machine

144

5

Development of a Simulation-based Methodology …

Now the peak optimization experiment can be setup as follows: The objective function is the minimization of the maximum peak found in the energy data flow set, the permanent documentation of the current energy flow of the production line. The number of iterations is set to 60, as this covers all iterations possible when varying the offset parameter between zero and five minutes in three-secondsteps (Figure 5.40). In order to illustrate the effect of the offset parameter, it is initially only applied to one machine. In principle, it is possible to determine the offset in parallel for all machines in relation to each other. If, however, the machines are to be switched on manually by the calculated offset, it will be difficult to implement this in practice. The consideration of several offset parameters in practice therefore requires the automated transfer of the offset values to the machine control system. The machine that is looked at, can be selected in the main table by removing the ignore-flag in the parameter settings. The idle and standby optimizer parameters are set to the best feasible solution of the total consumption optimizer experiment, which is run first. The parameters are set to the type fixed will not be varied during the peak optimization experiment. The model time is set to a four-week period. The order of the execution of the defined optimization experiments is dependent on the priorities of the optimization objectives. The objective with the highest priority is run first, as the optimization parameters are then fixed in the following experiment executions. For the case study, the highest priority gets the optimization of the overall energy consumption. The peak reduction is then performed for the optimized total energy consumption scenario without affecting the total consumption value. The methodology can be summarized using the following steps:

Step 1: Step 2: Step 3: Step 4: Step 5:

Modeling of different aspects of the production simulation. Implementation of production and energy data in the model. Definition of required optimization parameters for the simulation. Definition of objective functions and setup of optimization experiments. Execution of the optimization experiment according to the priority of the objective functions. The optimum parameters are copied into the simulation model and the following optimization experiments (as fixed parameters).

5.6 Prototypical Implementation

145

Figure 5.40 Setup of the peak optimization experiment

Step 6:

Step 7:

5.6.5

When all optimization experiments have been executed, the optimum parameter set of all optimization experiments are taken over into the simulation model. Execution of the simulation model with optimal parameters to predict the energetic behavior, total consumption and the peak demands of the production.

Evaluation of the Optimization

This section shortly discusses the optimization results of the fictional production line. To be able to compare the improvements achieved by the optimization experiments, a reference scenario is created. In this scenario, interruptions in the

146

5

Development of a Simulation-based Methodology …

production flow do not trigger any machine state changes. The machines remain in a non-value-adding but energy-consuming machine state for the duration of the disturbance. The reference scenario is thus very similar to the reference scenario of the practical example discussed later in this book. This scenario is reproduced in the simulation model, setting the idle and standby optimizer parameters to 480 minutes which assures, that the machines do not leave the non-productive states during an eight-hour shift. To be able to compare the results, the minimum output quantity of the production line is set to 2000 pieces for a four-week period11 . The total energy consumption sums up to 2.395.082,3 kWh for the production of 2.028 pieces. The maximum occurring peak consumption in the reference scenario is measured with 161,0 kW.

Figure 5.41 Results of the total consumption optimization

To determine the potential for optimization, the optimization experiments are carried out according to the steps mentioned in the previous section (Figure 5.41 and Figure 5.42). Subsequently, the simulation model is started with the determined parameters. 11 As

described earlier in section 5.6.1, maximum of 60 pieces can be produced per shift. On average a month has about 20 working days (40 shifts per months for a two-shift-model) and maintenance times need to be considered.

5.6 Prototypical Implementation

147

Figure 5.42 Results of the peak consumption optimization

The total energy consumption using the optimized parameters for the simulation model sums up to 2.097.912,3 kWh while the maximum peak is found at 158,9 kW. Comparing these values with the values from the reference scenario reveals a significant reduction for the total energy consumption value of 14,2% and a reduction of the maximum peak by 1,3% (Table 5.3). While the share of energy consumed by non-value adding machine times adds up to 19,7% of the total energy consumption in the reference scenario, the nonvalue adding consumption share is reduced to only 6,3% through the idle and standby optimizer parameters. The determined optimal offset not only reduces the highest peak but also leads to a general reduction of the number of peaks. While there were 27 consumption peaks measured with 160 kW or more in the reference scenario, no peak above 160 kW occurs using the machine offset parameter12 . The comparison of the reference scenario and the optimized production scenario clearly shows that by deliberately avoiding non-value-adding machine states, a reduction of the total energy consumption and, by using offset times a decrease of peak loads can be achieved. Thus, the simulation-based optimization method proves to be a helpful tool for the energy-efficient design of production processes. The presented methodology requires the use of combined simulation, which entails increased data generation and modeling efforts. For this reason, it should be clarified at this point whether the use of the combined simulation is actually 12 A full list of the 50 highest measured consumption peaks for the reference and the optimization scenario can be found in the additional material of this book.

148

5

Development of a Simulation-based Methodology …

Table 5.3 Comparison of scenario key figures Reference Scenario

Energy Efficiency Optimization

Deviation

Output Quantity

2028 pieces

2028 pieces

0

Total Energy Consumption

2.395.082,3 kWh

2.097.912,3 kWh

14,2%

Share of the non-value energy consumption

471.866,8 kWh (19,7% of total energy consumption)

132.411,2 kWh (6,3% of total energy consumption)

/

Maximum Peak Consumption

161,0 kW

158,9 kW

1,3%

required, or if potential savings in production can be recognized even without the use of simulation paradigms with different time progress. To answer this question, the presented methodology is reduced by the continuous model part on the internal machine level. The real energy consumptions are no longer simulated in the energy component of the simulation module but are reduced to a mean value and stored as parameters in the machine logic at the corresponding machine states. This reduces the model complexity from three to only two simulation paradigms (DES and ABS), both following a discrete time progress. The simulation methodology easily allows the usage of mean values, which significantly reduces the data collection efforts. The depiction of the exact energy consumption values with its maximum load peaks cannot be done anymore. On the macro level, it is never the less possible to determine the energy consumption of the production line at any time. The optimization scenario is run using mean values and afterwards being compared to the run with the exact energetic load profiles. The results are compared in Table 5.4. Using mean values instead of exact load profiles leads to less accurate results. The total energy consumption of the production line is reported 0,2% lower than it actually is13 . The same applies to the depiction of the non-value-adding energy share with a deviation of 0,4% of the exact value. Despite the less exact depiction of energy values, the optimization methodology can be used very well with mean values, since it is not relevant for detection and elimination of non-value-adding machine times, whether mean values or exact loading profiles have been used. 13 At

this point, the presentation of the results of mean values in the reference model is omitted since the deviations were identical to those of the mean values in the optimization model compared to the use of exact load profiles.

5.6 Prototypical Implementation

149

Table 5.4 Comparison of the use of energetic load profiles and mean values for the energy consumption Mean Values

Energetic Load Profiles

Deviation

Output Quantity

2028 pieces

2028 pieces

0

Total Energy Consumption

2.093.223,5 kWh

2.097.912,3 kWh

0,2%

Share of the non-value energy consumption

131.864,7 kWh

132.411,2 kWh

0,4%

Maximum Peak Consumption

122,2 kW

158,9 kW

23,1%

In both cases, optimal parameters to rise the energy efficiency are found. While the total consumption experiment was successful using mean values, the peak consumption experiment did not yield any useful results (Figure 5.43). The maximum occurring peak in the mean value scenario is found at 122,198 kWh, a deviation of over 23% compared to the peak that can be determined by using the exact energy load profiles. Due to practical restrictions, the peak optimizer is only testing off set values between zero and three minutes. As the machine states have durations between four and 90 minutes, an offset of three minutes is insufficient to achieve a shift that avoids a maximum value having the length of a machine state. The peak optimizer therefore does not reach any optimal value for the offset parameter. The maximum consumption value remains at 122,198 kWh with every parameter variation possible. The use of the average values for the duration of an entire machine state produces a blur regarding peak loads in the power consumption profile. Occurring extreme values are no longer recognizable in the data and thus not shown in the simulation. It can therefore be concluded, that the use of the peak optimizer together with mean values is not effective. As described, two objective functions are pursued to optimize energy efficiency in production. The associated optimization parameters were determined successively in two separate optimization experiments. At this point, the question arises whether both objective functions can be summarized in one single optimization experiment. By separating the experiments, the first ideal parameter configuration is determined, which promises the lowest possible total energy consumption. Assuming, that this optimal scenario is followed, the peak loads occurring in this scenario are determined in a second step and the required offset parameter for a peak reduction in this scenario is proposed. The offset of one machine

150

5

Development of a Simulation-based Methodology …

Figure 5.43 Peak optimization experiment in combination with the use of mean values for the energy consumption

has no influence on the total energy consumption, which still corresponds to the determined possible consumption minimum. If both objective functions are pursued in parallel in one single optimization experiment, both, the idle and standby optimizer parameters as well as the offset parameters of the machines are varied in parallel in various iterations. Due to the increased number of parameters that are optimized in parallel, the number of possible parameter configurations increases massively. For this reason, the number of iterations to be performed must be adjusted in order to obtain reliable optimization results. A raise in the number of iterations drastically increases the computational effort of the entire optimization experiment. Test runs in AnyLogic found that the proposed parameter configurations resulted in significantly higher total energy consumption values but were rated as the optimal solution because the effective reduction of power peaks occurring in this constellation was very high. However, the reduced power peak values were not lower than those found in the separate experiments. The execution of only one optimization experiment, which simultaneously pursues both objective functions, thus does not provide a total optimum and additionally went along with higher computation times for the optimization experiment runs. The optimization of both objective functions in one experiment requires the definition of a cost function.

5.6 Prototypical Implementation

151

In summary, the application of the optimization methodology on the fictional use case can be described as successful. The use of different simulation techniques was proven to be possible. The material flow of the production line was built following a process-oriented DES logic, while the internal logic of the machines was created using ABS and then combined with the energy simulation in SD. Thus, the dynamic depiction of the energy consumption behavior became possible, being able to get an exact consumption value for any point in time. The reference scenario reflects the current behavior in production, which is basically to make no efforts to bring machines into a less energy-intensive state during production interruption periods. By introducing the idle and standby optimization parameters, production employees receive precise information as of which interruption length a change of the machine states makes sense and saves energy. The determination of exact starting times of machines that are used later in the production process opens up the possibility of making production more energy efficient. In addition, the offset parameter gives an exact recommendation for action defining which offset for an energy-intensive machine is required in order to prevent energy consumption peaks. The use of the optimization methodology has shown that significant savings in total energy consumption have been achieved compared to the reference scenario. For the consumption peaks, a slight improvement could be achieved by using the offset parameter. It is believed that the peak optimization provides better results when a larger area of production is considered in the simulation model. The optimization methodology was also tested for the use with mean values instead of exact energy load profiles. While the optimization of the total energy consumption has delivered results in similar high quality as for the exact load profiles, the peak optimization was not possible under these circumstances. In addition to the settings and setups tested in this section, the fictional case model was also tested with varied delivery schedules, machine settings, and production parameters in further simulation experiments. The results have been evaluated by comparing the reference and the optimization models as discussed before. Since the results were similar to the extensively described setup, further explanations are omitted here. The next section summarizes gained findings to consider peculiarities for the practical application of the methodology.

152

5.6.6

5

Development of a Simulation-based Methodology …

Conclusions for the Practical Application of the Methodology

Through different simulation experiments with a manually varied machine schedule, batch size variations for individual machines of the production line and the deliberate avoidance of non-value-added machine states, it was possible to test how the energy consumption of the production can be reduced without negatively affecting the total daily output. The simulated energy profiles of the machines in the fictional case study depict the dynamic behavior of the energy consumption and even the consideration of single machines using mean values instead of exact load profiles was successfully tested. By treating the peak optimization objective as a restriction in the total consumption optimizer, it was tried to summarize all objective functions in only one optimization experiment. The solutions achieved by this were not better, as neither the total consumption nor the peak reduction was more successful. In terms of occurring peak loads, the individual execution of the optimization experiments led to the same results. However, the equal consideration of the consumption peaks has led to experiments with poorer total energy consumption and their optimization parameters being presented as ideal solutions. As the reduction of total energy consumption is the number one objective, this optimizer should be run first. Afterwards, the peaks for this ideal solution should be optimized again to get not only the solution with the lowest total energy consumption but to optimize the occurring peaks for this low consumption parameter set. Therefore, the experiments have to be separated and are run in the above defined order. This is referred to as lexicographical ordering14 of objective functions [Br2008, p. 19; NM2005, p. 14]. While the fictional example describes a production line in which the individual machines are rather loosely dependent on one another and are decoupled from each other by smaller buffers, the practical case in chapter 6 depicts two production lines with strongly connected production machines. The failure of one machine in the production line leads to a downtime of the entire line with a minimum offset of a few seconds between the single machines. For this reason, 14 “In lexicographic ordering […], [the decision maker] must arrange the objective functions according to their absolute importance. This means that a more important objective is infinitely more important than a less important objective. After the ordering, the most important objective function is minimized subject to the original constraints. If this problem has a unique solution, it is the final one and the solution process stops. Otherwise, the second most important objective function is minimized. Now, a new constraint is introduced to guarantee that the most important objective function preserves its optimal value” [Br2008, p. 19].

5.6 Prototypical Implementation

153

neither the calculation of the start dates, nor the optimization of the idle and standby parameters are required for each individual machine. Therefore, the simulation model logic is slightly adapted compared to the fictional scenario. Each machine retains its own machine logic as it is needed to correctly depict the energy consumption behavior, but the idle and the standby optimization parameters are used per line and not at the machine level anymore. Details on the changes are described in section 6.3.1.

6

Experimental Validation of the Methodology

“Scientific research is a search after truth, and it is only after discovery that the question of applicability can be usefully considered.” Henri Moissan (Prof. Henri Moissan (28 Sep 1852—20 Feb 1907) was a French chemist and Nobel Laureate (1906). The quote comes from a lecture held by Prof. Henri Moissan at the Royal Institution of Great Britain, May 28,1897.)

This chapter summarizes the results of the practical test of the simulation-based optimization methodology. Section 6.1 presents the use case in detail while section 6.2 comprises the creation of the simulation model. Section 6.3 contains the process description and evaluation of the optimization experiments using different data resolutions as well as the application of the methodology using mean values. The evaluation of the results is followed by a financial evaluation of the potential energy savings in section 6.3.6 as well as a list of recommendations for action for the practical partner in section 6.4. Chapter 6 concludes with a critical statement on the general suitability of the methodology for practical use and an evaluation of the methodology with regard to the concept objectives and requirements defined in section 5.1.

Electronic supplementary material The online version of this chapter (https://doi.org/10.1007/978-3-658-32971-6_6) contains supplementary material, which is available to authorized users.

© The Author(s), under exclusive license to Springer Fachmedien Wiesbaden GmbH, part of Springer Nature 2021 A. C. Römer, Simulation-based Optimization of Energy Efficiency in Production, Forschung zur Digitalisierung der Wirtschaft | Advanced Studies in Business Digitization, https://doi.org/10.1007/978-3-658-32971-6_6

155

156

6.1

6

Experimental Validation of the Methodology

Presentation of the Practical Area of Application

The validation of the methodology is based on practical data1 from Bosch, a global company operating in various fields. The pilot area, which supplies the data for validating the simulation-based optimization methodology, is part of a Bosch plant in Germany that manufactures engine components for the automotive industry. With more than 4200 employees, approximately 16 million parts are produced per year for passenger cars, commercial vehicles, as well as off-highway applications. The plant has a yearly energy consumption of 54 GWh and therefore energy costs of more than 10 million Euro2 . About 95% of the electricity demand is covered by a framework contract with an external energy supplier, the remainder is provided plant-internally. The type of power supply does not play an important role for this thesis, since it is purely about the total amount of energy savings. Whether the energy requirement is satisfied by a company-owned energy source or the energy grid of an external supplier, is not initially important at this point. To assess the results and optimizations, a uniform cost rate, namely the price of the energy supply contract, is used for calculations. The electricity tariff, as described in detail in section 3.3, provides a distinction between a high and low tariff working price3 and includes a peak-dependent power price for grid usage.

6.1.1

Presentation of the Data Basis

Bosch has a very detailed energy monitoring system in use. The energy data acquisition in the considered plant is available on machine level factory-wide. The collected energy consumption data is managed in a purpose-developed software solution named energy platform. Energy efficiency measures regarding the leak management of pipes, the compressed air supply, the control of the heating, cooling, ventilation, and illumination of production halls, as well as a shut-down management of unneeded ventilation to avoid energy peaks are monitored and evaluated using the energy platform data. So far, the energy monitoring system is 1 The

Bosch Company provides current product, production, and energy data in full for this work. For data protection and privacy reasons, the production data (produced quantities, production cycle times, delivery schedules, etc.) is anonymized and adapted by the author in the text, the illustrations, and the simulation model. The energy consumption data of the machines is not changed. 2 The 10 million Euro include 19% value-added tax. 3 The high tariff is used from 6:00 am to 10:00 pm and the low tariff from 10:00 pm to 6:00 am.

6.1 Presentation of the Practical Area of Application

157

not connected to the enterprise resource planning (ERP) system. In the context of this work, the linkage between Bosch energy data and production data has been made, to use the energy data for energy efficiency optimizations of non-valueadding production times of machines and the prevention of energy peaks during production.

Figure 6.1 Bosch production lines 1 and 2

Two production lines have been selected as an example for the validation. The selected production lines are referred to as “line 1” and “line 2”. The lines share the raw material warehouse as well as the finished goods stock. There is no other logical link between the lines. Both production lines produce similar finished goods, which basically differ in their length, and are thus constructed equally (Figure 6.1). The finished parts produced on production line 1 can come in five different product variants, on production line 2 eight product variants are produced. Since the different product variants are very similar and no noticeable changes in the energy consumption of the production equipment during the production processes of the different product types can be determined (Figure 6.2), no differentiation according to individual product types in the energy consumption analysis of the production processes is made. Both production lines 1 are operated in a two-shift model with 10 shifts per week. The machines of a line are linked by conveyor belts. The machine capacity and the number of workpiece carriers in the production lines limit the number of parts that can be processed in the production line at the same time (Table 6.1). The loading station can be filled with a raw material container containing 192 (4 × 48) parts, which will supply the production line for about two hours with raw material before a worker has to refill it as the cycle time of the production

158

6

Experimental Validation of the Methodology

Figure 6.2 Comparison of load profiles of three different product variants on machine 1 (of line 1) for a batch of 80 pieces/ 50 minutes per variant

lines is 33 seconds per part. Additionally, the loading station can handle four parts at the same time with its grippers loading the parts from the raw material container into M2 and taking them from M2 into a workpiece carrier for the conveyor belt transport to M3. The pressing machine M3 has four process steps in which one part at a time is processed. The finish machine performs five process steps, meaning five parts can be in the machine at the same time. The cleaning machine is subdivided into eight partitions, in which six parts at a time can go through the cleaning process. The unloading portal is able to handle three parts at a time, first putting the parts from the workpiece carrier into the cleaning machine and after the cleaning process back into containers with a capacity of 48 for the finished goods warehouse. The machines M2, M3, and M4 are linked through automated conveyor belts. The parts are transported on the conveyor belts in workpiece carriers with a capacity of two. Ideally, there should not be more than ten workpiece carriers in a production line which basically means in practice, that six parts are on the conveyor belt between M2 and M3 and six parts are on the conveyor belt between M3 and M4. The remaining four workpiece carriers are located on the

6.1 Presentation of the Practical Area of Application

159

returning conveyor belt to the loading portal in front of M2 after unloading by M5. Table 6.1 Overview of parts and carriers in the production line Machine state Number of parts in machine M1_load

4 plus 192 parts supply

M2_press

4

M3_finish

5

M4_clean

48

M5_unload

3 plus 47 in container for finished parts

6.1.2

Capacity of Conveyer Belt

3×2

4 empty carriers 3×2

Data Acquisition, Processing, and Validation

The processing of relevant data is done using the Bosch Energy Platform, where the measured values are recorded and checked, so that processed data is then stored in the database as error-free as possible. The data check includes the control of complete data transmission, the value calculation of missing data records as well as checks on measuring ranges and data consistency. In addition, the necessary conversions and aggregation of the raw data is done. The data of the energy platform is made available to the individual company divisions. To protect workers against productivity measurements, the energy data is measured with a 1-minute-resolution but then aggregated to a 15-minute interval. With process times of half a minute, such grossly aggregated data cannot be used for energy efficiency optimizations. Therefore, different data resolutions4 are subsequently tested to determine an appropriate aggregation level. The energy data acquisition of the high-resolution data was done by hardware controllers installed at the machines specially for this purpose. Over a period of

4 The

data measured at a resolution of less than 15 minutes has been collected for a short period of up to three weeks and is used for scientific purposes only without any linkage to employee data.

160

6

Experimental Validation of the Methodology

four weeks, the energy consumption behavior of the ten machines was documented. Thus, three different resolutions of the energy consumption data are now available for the simulation and optimization experiments: • 15-minute resolution The 15-minute resolution data is exported from the energy data portal. The data in the energy portal is based on the machine controller data. • 1-minute resolution The 1-minute resolution is exported from the original machine controller. • 1-second resolution The 1-second resolution is exported from the additionally installed controller. Since the energy data collection is done in a separate system that has no interface to the ERP or MES systems, the combination of energy and production data proved to be difficult. While the energy consumption data is collected automatically, the documentation of production quantities, production rejects, interruptions, and maintenance times is a purely manual process, usually done at the end of a shift. Thus, these data do not have a usable timestamp, which, in turn, makes automatic assignments to the data from the energy measurements impossible. To solve this data issue, all machine states in manufacturing were manually triggered and documented with exact timestamps and associated production quantities. All machines went through a controlled state simulation5 to be able to precisely allocate the energy consumptions to the respective machine states. This method is very time consuming, as it causes the shutdown of a production line for the duration of the tests of all states. Each of the ten machines was operated in each machine state for a total of 30 minutes in order to generate the energy database required to set up the simulation. In this way, a clean, error-free database was generated that can be used for the optimization of the total energy consumption and occurring power peak loads. Besides the simulation of machine states using the real production machines, the load profile clustering (LPC) algorithm described in section 3.2.3 has been tested, using only electrical load profiles and processing times to identify different load level. The automated extraction of state-based machine information from available load profile data in combination with a manual processing cluster assignment saves time compared to the manual machine state simulation and leads 5 The controlled state simulation has been carried out by a Bosch employee for both production

lines. Every machine was manually put into every possible machine state and has been run for about 30 minutes in every state. The energy load profiles generated in this time can thus be assigned exactly to the individual states.

6.1 Presentation of the Practical Area of Application

161

according to Teiwes et al. [Te+2018] to reliable results. The LPC has also been tested with different machines data for this work to simplify the extraction of machine-state-based load profiles from large datasets. It has become evident that the results of the clustering were only clearly separated according to the machine states, if the power values of the machine states significantly differed in the height of their values.

Figure 6.3 Load profile extract from the finish machine of line 2

The example of the finish machine illustrates why the clustering could not be used successfully. As shown in Figure 6.3, the productive and the idle state can be distinguished from each other very clearly. The finish machine switches at 2 pm from productive into idle state (for the shift handover from the early to the late shift), from which the machine then returns into productive state at around 2:39 pm in the first hour of the late shift. The visualization of the data makes it very easy to recognize the different machine states, but it also becomes apparent that the maximum values of the idle state are in a similar value range as the minimum values of the productive state (Figure 6.4). At this point, the clustering method fails. Using the k-means algorithm, the data clusters are formed according to fixed value limits. If the value ranges of the machine states overlap, it is no longer possible to automatically build up the clusters matching the machine states only using value limits. Individual records from the productive state that have, for example, a particularly low value that is characteristic for the idle state cluster are thus assigned to the same cluster

162

6

Experimental Validation of the Methodology

as the values of the idle state. On the other hand, the data sets of the idle state with rather high values are assigned to the cluster of the productive state. The use of a clustering algorithm requires a complex post-processing of the clustered data, to resolve blurring caused by overlapping values between machine states. Since the elaborate post-processing of the cluster data thus makes the argument of the time savings obsolete, the machine state simulation in the real production is the method of choice for the verification of the state-based energy consumption profiles in this work.

Figure 6.4 Detailed load profile view (L2_finish)

Since a clean and complete database forms the basis for the simulation and optimization experiments, it is essential for the application of the described simulation-based methodology to determine the machine-state-dependent load profiles and production quantities either in the manner described above or from

6.2 Simulation Model Details

163

information technology systems that are based on a common time stamp and can thus be merged automatically.

6.2

Simulation Model Details

The simulation model as well as the optimization experiments are created in AnyLogic PLE 8.5.1. As already for the fictional case study, the simulation model is built using the three different simulation paradigms. The data from the real production is used for the model. Quantities, processing times, maintenance plans, buffer sizes, and delivery schedules are set according to the real production scenario. The machine logic types have already been modeled for the fictious case study and can be used via drag and drop for every model including all optimization parameters and restrictions described in sections 5.6.2 and 5.6.3. The modeling of the Bosch simulation model as well as the setup of the optimization experiments is—in large parts—similar to the procedure in the fictional case study. Deviations caused by the special production situation (strongly linked production machines) are described in the following sections.

Figure 6.5 Machine logic and switching options of the Bosch production machines

164

6

Experimental Validation of the Methodology

The two considered production lines consist of five machines each, with three machines actually performing machining processes on the product and two machines per line being installed for loading and unloading processes. The machines per line are strongly linked to each other without any buffers between the single machines. A machine stoppage for one production machine causes a production stop for the entire line. In terms of machine logic, all machines have a similar complexity that corresponds to the high complex machine logic type. All machines have a main switch that can either be on or off, a Programmable Logic Controller (PLC) that can either be turned on or off, and an operating mode which can either be switched to manual or automatic (Figure 6.5). The combination of the different switching options results in different machine states. Table 6.2 contains an overview of the assigned machine states. Table 6.2 Assignment of switching options to machine states Machine state

Main switch

PLC

Operating mode

Energy consumption

Off

Off

Off

Both possible

No

Standby

On

Off

Both possible

Yes

Manual mode

On

On

Manual

Yes

Warmup

On

On

Automatic

Yes

Idle

On

On

Automatic

Yes

Producing

On

On

Automatic

Yes

Failure

On

On

Both possible

Yes

Fast warmup

On

On

Automatic

Yes

Since the production machines of the two production lines are very similar in their states and control options, the complex machine type is assigned to all machines in the simulation model. The Bosch machines use the fast warmup state, for example, to bring spindles back to working temperature after longer lingering in non-productive modes like standby, idle, or manual mode. This procedure prevents from fast abrasion of the spindles, reduces the number of machine downtimes, and prevents machining errors and inaccuracies that can occur when producing with cold spindles. The washing machines in both lines are not switched off (main switch off) completely, since the timer which times the heating of the cleaning tanks is integrated in the control and does not constitute a separate module. A solution to turn off the main switch for the cleaning machines might be to separate the

6.2 Simulation Model Details

165

timer unit from the controller and install it in front of the main switch. Thus, the machine does not have to remain in standby but can be switched off. Additionally, after completion of the production, it is necessary to leave the machine in idle or manual state for about 60 minutes. A water separation operation is required to separate the lubricants and oil of previous production steps that have been washed of the parts from the cleaning liquid. Setup processes for all machines, required to produce different product variants, are generally performed in manual mode. All restrictions are taken over into the simulation model, to be able to depict a realistic production scenario. The final simulation model is shown in Figure 6.6. To generate a lucid model, the energy flows are first separated per line in flow 1 and flow 2, before they run together in the total energy consumption stock element.

Figure 6.6 Simulation model of the Bosch production lines 1 and 2

166

6

Experimental Validation of the Methodology

For the model validation, the model logic has been crosschecked several times before it was used to reproduce a number of production days of historical data6 . After repeated fine-tuning of the modeled material flow parameters as well as multiple revisions of the entered machine state transition rules, the hybrid simulation reproduced the production data with a deviation of less than 3.2 percent. The occurring deviation of 3.2 percent per shift can be explained considering machine failures in the real production. As the machine failures can hardly be predicted correctly, they have not been considered in the hybrid model. As failures occurred in practice, but not in the model, the part numbers produced in the simulation runs have been slightly higher than in practice. For future simulation runs, it is possible to consider a failure probability for all machines in the machine logic if the focus is on getting the output numbers of the model closer to the ones in the real production. Including failure probabilities adds a stochastic to the simulation model and thus requires the setup of replications in the optimization experiments. To be able to simulate and compare different energy data resolutions as well as the use of mean values7 , the simulation model is duplicated, using different energy datasets. An overview of tested scenarios, the energy data type, and the data resolution is provided in Table 6.3. In section 6.3, the optimization experiment setup is depicted using the 1-minute-resolution data. The results of the simulations and optimizations with the other data resolutions are partially presented afterwards. All results are then discussed in section 6.3.5. Table 6.3 Overview of created simulation models Resolution

Energy data type

Optimization

Model name

1-minute

real load profiles

no

Reference Scenario (1-min)

real load profiles

yes

Optimization Scenario (1-min)

mean values

yes

Mean Value Scenario (1-min)

real load profiles

no

Reference Scenario (1-sec)

real load profiles

yes

Optimization Scenario (1-sec)

1-second

6 At this point, reference is made again to the alienation and anonymization of real production

data in order to eliminate any possible conclusions about sensitive or confidential company data. The changes in the data basis were considered for the model validation. 7 The simulation of mean values is only tested for the data resolution of 1 minute, as the data in a 15-minute resolution is highly aggregated already. The mean value creation for the 1-second resolution data will lead to the same mean values as for the 1-minute resolution, as they go back on the same data basis.

6.2 Simulation Model Details

167

The parameter setups, as well as delivery schedules, process times, etc., are identical in all simulation model setups. The data at the 15-minute resolution is available as a data export from the Bosch energy portal but cannot be used for the simulation-based optimization methodology. Due to the rather short cycle and machine state times in production, it is not possible to cleanly assign the data records to the machine states. The example of the resolution comparison of the energy consumption profiles of machine L1_load clearly shows that the number of measuring points for depicting the realistic energy consumption behavior separated according to the single machine states is not possible (Figure 6.7). For the duration of the machine warmup of 19 minutes, e.g., only two data records are available when using such a coarse data resolution; in the productive state of the machines, the period between two measuring points covers the production of more than 40 produced parts. With such strong data aggregation, the boundaries between the machine states become blurred.

Figure 6.7 Comparison of data resolutions of L1_load over eight hours/one shift

Therefore, the data in a 15-minute resolution was not used to build a model for the simulation-based optimization of energy efficiency in the Bosch production scenario. The following section describes the optimization experiment runs using the data with the resolutions of one data record per second and one data record per minute.

168

6.3

6

Experimental Validation of the Methodology

Optimization Experiments

To increase the energy efficiency in the Bosch production scenario, the previously defined optimization experiments were used. As in the fictional case study, the reduction of the total energy consumption of the production lines has the highest priority. For this reason, two separate optimization experiments were performed. Firstly, the optimizer to minimize the total energy consumption of the two production lines was run, followed by the consumption peak optimization experiment.

6.3.1

Optimization Parameter Adaptations

As already briefly indicated, there is a very strong dependence of the machines among one another in the production lines considered. No safety stocks are provided between the machines to decouple the processing steps, so failures of individual machines in the line cannot be intercepted by material buffers but lead to a shutdown of the entire line with a delay of only one to two machine cycles. Due to the strong coupling of the machines, it does not make sense to optimize and evaluate the idle and standby optimizer parameters per machine. For a holistic optimization approach, the standby and idle optimizer parameters are introduced per line, with one exception: for a number of reasons, the washing machines are technically unable to change machine states as quickly as the remaining four machines of the respective line. In addition, the parts are buffered in front of the washing machines until the required lot size is reached and the cleaning process can be started. For this reason, L1_wash and L2_wash have their own idle optimizer parameters to take into account the technical requirements of the real washing machines (Figure 6.8). In total, the number of optimization parameters in the optimization experiment is reduced from 30 to eight, which has a positive effect on the experiment runtime but does not alleviate the quality of results. Six of the eight optimization parameters were used for the total consumption optimization and the optimal configuration of the other two parameters is determined in the peak optimization runs.

6.3 Optimization Experiments

169

Figure 6.8 Optimization parameter setup of the production machines

6.3.2

Simulation-based Optimization using the 1-Min Resolution Data

The data for the machine states in the 1-minute resolution is taken from the machine controllers. The data records from the monitored machine state simulation in the production are assigned to the machine states. Before, the exported files have been checked for completeness and correctness. Duplicates, empty or incorrect data records were not found. The data records were imported into the simulation model and linked to the corresponding machine states. The simulation model time unit was set to “minutes”, as well as the timeout unit of the machine state transitions. To be able to compare the achieved optimizations against reference values, one week of production8 is simulated without using the optimization parameters in the Reference Scenario. The reference scenario was then duplicated to have exactly the same starting point and stored as Optimization Scenario. For the optimization scenario, the total consumption optimizer experiment and the peak consumption experiment were created and executed.

8 At this point, the week from May, 20th to 26th was chosen, as the energy data for this period

is available and the model can therefore be validated well.

170

6

Experimental Validation of the Methodology

Figure 6.9 Results of the total consumption optimizer experiment run (data resolution 1min)

During the total consumption optimization experiment runtime of 38 minutes, 500 iterations and thus 500 parameter configurations were simulated and evaluated with regard to the objective function (Figure 6.9). The optimization parameters were varied within the applicable minimum and maximum values for the individual parameters, which result from the specifications of the production management. Figure 6.10 shows very clearly that the first parameter configurations had a significantly greater variance than the parameter sets after a larger number of iterations, such as in the graph presented in Figure 6.9. At this point, the use of tabu search in AnyLogic’s optimization algorithm becomes noticeable. With each iteration, promising configurations are followed, while configurations that provide inferior values for the performance of the objective function are not further refined. After determining the optimal idle and standby optimizer parameters, they were taken over as fixed values in the peak consumption optimizer to eliminate peak loads in the scenario with the lowest total energy consumption. The advantages of separating the two objective functions in two optimizer experiments are

6.3 Optimization Experiments

171

Figure 6.10 First iterations of the total consumption optimizer experiment run (1-minute resolution of the energy data)

the endorsement of the prioritization of the objective functions as well as the simplification of the experiments. If all optimization parameters for the total energy consumption and the peak load optimization were varied in parallel in one experiment, it would require a much higher number of iterations to obtain a similarly good solution as in the two separate experiments. As in the first optimization experiment, the parameter configurations for the offset parameter are varied within the limits set by the production manager. Since the total output of the production should not be adversely affected, the offset parameter may cause a delay of one production line by a maximum of five minutes (or ten finished parts). The results of the peak consumption optimizer are presented in Figure 6.11. The offset causes a slightly delayed start of the machines of line 1, therefore other values of the energy consumption profiles of the single machines meet and add up to lower power load peaks as shown in Figure 6.12. At this point it is important to understand that the total energy consumption of the production lines is not affected in forms of energy savings, the offset parameter only causes a shift of the consumption point in time into the future (Figure 6.13). The level of total energy consumption for the two production lines determined in the first optimization experiment remains the same.

172

6

Experimental Validation of the Methodology

Figure 6.11 Peak consumption optimizer experiment run (1-min resolution)

Figure 6.12 Effects of the offset parameter on the power consumption peaks

After the completion of both optimization experiments, the optimal parameter configurations were tested in the simulation model9 and, when compared with the reference scenario, led to the deviations shown in Table 6.4. Having the same production output quantity, the setup of the optimization scenario results in a reduction of the total energy consumption of 6,2% by cutting 9 The screenshot of the simulation experiment run with the optimized parameter can be found

in Attachment D.

6.3 Optimization Experiments

173

Figure 6.13 Depiction of the offset parameter effect on the machine starting time

Table 6.4 Comparison of the reference and the optimization scenario (1-min resolution) Reference Scenario

Optimization Scenario

Deviation

Output Quantity

13.728 pieces

13.728 pieces

0%

Total Energy Consumption

193.195,6 kWh

181.422,8 kWh

6,2%

Maximum Peak Consumption

58,6 kW

57,1 kW

2,6%

the non-value-adding machine times occurring as far as possible. Additionally, the occurring peaks in this energy-efficient scenario are reduced by 2,6% causing a delay in production of 1.2 minutes without affecting the production output.

174

6.3.3

6

Experimental Validation of the Methodology

Simulation-based Optimization using the 1-sec Resolution

The machine controllers of the production machines under consideration were technically not suitable for recording energy data with a resolution of one dataset per second. For this reason, external voltage quality measuring devices (Janitza UMG 604 EP) were temporarily installed to document four weeks of energy data for each machine. The raw data of the external measuring devices was compared with the data measured by the machine controllers during the same period. Using the example of L1_load, the deviations found in the data are shown.

Figure 6.14 Energy consumption profile of L1_load with one data record per second

Figure 6.14 shows a four-hour extract10 of the energy consumption profile of the L1_load with a resolution of one data set per second, Figure 6.15 shows the same energy consumption time period of the same machine with a resolution of one data set per minute. In the direct comparison, the different representation of consumption peaks is particularly noticeable. While the maximum values in the 1-second resolution are 2.5 kW, no values above 1.57 kW are displayed in the 1-minute resolution graph. When evaluating the data tables, it was found that the different accuracies lead to different total energy consumptions of the machines per shift. In the example 10 The energy consumption profiles documenting an eight-hour shift of L1_load and L2_wash can be found in Attachment C of this book.

6.3 Optimization Experiments

175

Figure 6.15 Energy consumption profile of L1_load with one data record per minute

of L1_load, the consumption per shift on the day evaluated adds up to 514.2 kWh, if the 1-min resolution is used as a basis. If the sum of a shift is calculated on the basis of the 1-sec resolution, the machine L1_load consumes 525.8 kWh. The table data of the L1_load thus differs by 2,2%. The deviations occur in varying degrees of severity in the raw data of the other nine machines of production lines 1 and 2 as well and range from 0,1% to 2,8%. This deviation caused by the raw data might be visible when comparing the reference scenario results with the two data resolutions after the application of the simulation-based optimization methodology. To test the methodology with the 1-sec resolution data, the data records from the monitored machine state simulation in the production were assigned to the machine states after the exported records have been checked for completeness and duplicates, empty or incorrect data records. The data was imported into the simulation model and linked to the corresponding machine states. The simulation model time unit was set to “seconds”, as well as the timeout unit of the machine state transitions. To be able to compare the achieved optimizations against reference values, one week of production was simulated without using the optimization parameters in the Reference Scenario. Again, the reference scenario was then duplicated to have exactly the same starting point and stored as Optimization Scenario. For the optimization scenario, the total consumption optimizer experiment and the peak consumption experiment were created and executed. Figure 6.16 and

176

6

Experimental Validation of the Methodology

Figure 6.16 Results of the total consumption optimization experiment run (1-second resolution)

Figure 6.17 show the results of the total consumption optimization and the peak consumption optimization experiment runs. The number of iterations for the consumption optimization was set to 500, the simulated time period was exactly the same week as in the 1-min resolution experiments. While the total consumption optimization experiment runtime at the 1-min resolution was 39,3 minutes in total, the experiment runtime at the 1-sec resolution was 8,75 hours, more than 13 times as long. The representation of the offset in the simulation model can be seen more clearly with the 1-sec resolution (Figure 6.18). The determined optimal parameter configurations were transferred to the simulation model and tested. For the Optimization Scenario11 , the optimization experiments resulted in lower overall consumption and lower consumption peaks. Table 6.5 summarizes the results of the comparison from the Reference and the Optimization Scenario. By reducing the non-value-adding machine time through the determination of the optimal parameter configuration to leave the

11 The screenshots of the optimization scenario experiment run can be found in Attachment D.

6.3 Optimization Experiments

177

Figure 6.17 Result of the peak consumption optimizer experiment run (1-second resolution)

Figure 6.18 Depiction of the determined offset in the visualization of the energy flow at the example of six machines

178

6

Experimental Validation of the Methodology

idle and standby states, the total energy consumption was reduced by 6,6%. The introduction of the offset parameter led to a power peak reduction of 5,7%. Table 6.5 Comparison of the reference and the optimization scenario (1-second resolution) Reference Scenario

Optimization Scenario

Deviation

Output Quantity

13.728 pieces

13.728 pieces

0%

Total Energy Consumption

192.775,5 kWh

180.079,9 kWh

6,6%

Maximum Peak Consumption

112,9 kW

106,5 kW

5,7%

When comparing the two reference scenarios with the 1-min and the 1-sec resolution, it is noticeable that the deviation of the total energy consumption values is 0,2%. This deviation can be explained by the difference in the raw data of the two resolutions mentioned at the beginning of this section. The massive deviation of 48,1% between the consumption peaks is also understandable, as the high-resolution data already shows significantly more volatile curves with significantly higher values in the raw data as well (Table 6.6). Table 6.6 Comparison of the reference scenarios at 1-min and 1-sec resolution Reference Scenario (1-min)

Reference Scenario (1-sec)

Deviation

Output Quantity

13.728 pieces

13.728 pieces

0%

Total Energy Consumption

193.195,6 kWh

192.775,5 kWh

0,2%

Maximum Peak Consumption

58,6 kW

112,9 kW

48,1%

The comparison of the reference scenarios shows that with both resolutions the data of the real production can be reproduced.

6.3.4

Examination of the Necessity of Combined Simulation in Practice

The fact that hybrid simulation brings advantages for the representation of the processes in production by combining different simulation paradigms is beyond question. By using the ABS together with the process-oriented material flow

6.3 Optimization Experiments

179

simulation (DES), the machine states and machine state changes can be exactly reproduced. In combination with the time-continuous presentation of the energy consumption behavior of the machines (SD), the hybrid simulation model becomes a combined one, linking the discrete-event and the continuous time advance in one model. The extraction and evaluation of the exact energy consumption profiles is complex and time-consuming. In addition, administrative regulations often prohibit the collection and use of high-resolution data that could be used to determine employee productivity. The reality in production companies often looks like that energy data is selectively collected for single machines or production lines and the use of this collected data is restricted. For this reason, the possible application of the developed methodology for the case of unavailable energy consumption load profiles is described below.

Figure 6.19 Required machine setup to use energy consumption profiles

To depict the energy consumption behavior of the machines, the reproduction of the exact energy consumption behavior will be dispensed with and a representation of the energy consumption via calculated mean values is used instead. This results in the utilization of one energy consumption value per machine state

180

6

Experimental Validation of the Methodology

which is used as the energy value over the entire duration of the state. The realization in the simulation model requires the definition of a parameter in the machine logic and thus eliminates the complex representation of the power consumption via table functions. The Bosch simulation model requires one table function per machine state (Figure 6.19), summing up to 50 table functions in total. While the number of data records used for the 1-minute resolution remains manageable, the total number of datasets for the energy consumption per second sums up to over 35.000 data records that have to be accessed in a one-secondfrequency during the simulation runs. The replacement of these amounts of data with a total of 50 mean values reduces the access time and at the end the simulation and experimentation runtimes. The mean values are inserted in the setup of the single machine logics (Figure 6.20).

Figure 6.20 Required machine setup to use calculated mean values

The calculation of the mean values for the Bosch machines was done using the 1-minute-resolution data for the comparison of the simulation run and the optimization experiment results to the Bosch optimization scenario (1-min resolution). The decision for the 1-min resolution data is based on the fact that the energy consumption data was recorded by the machine controllers and not by external measuring equipment. Additionally, the resolution comparison showed that the

6.3 Optimization Experiments

181

results of the optimizations were very close and that the use of the high-resolution data proved to be rather disadvantageous, due to significantly longer simulation runtimes times. For the calculation of the mean values, all energy values per state are summed up and divided by the total number of energy values. Although the energy consumption visualization in the simulation shows significantly less volatile curves due to averaging, the machine state changes can be seen clearly in the visualization of the simulation (Figure 6.21). Extreme values are no longer displayed. The execution of the total consumption optimizer experiment revealed nearly identical parameter configurations (compared to the experiment run with load profile data) for the idle and standby optimizer parameters as optimal (Figure 6.22). The minimum deviations of 0.15% for the standby optimizer parameter and 0.33% for the idle optimizer parameter can be justified by rounding errors in the averaging calculations. Following the determination of the optimal parameter set to fulfill the objective function for the total energy consumption, the peak optimization experiment was started. Within the scope of the optimization, however, it was not possible to determine an optimal offset parameter that achieves a reduction of the occurring peak loads in the mean value scenario within the permitted limits for the parameter. Looking at the data of the first few minutes of the simulation, it quickly becomes clear why the load peaks cannot be reduced. The offset parameter can cause delays of up to five minutes. During the warmup phase, the maximum peak of the mean value scenario is reached for the first time. The entire warmup has a duration of 20 minutes and the peak occurs over the entire duration of the warmup phase. An offset of five minutes cannot counteract here. A reduction of the power peak load through the parameter variation in the peak optimization experiment run is therefore not meaningful (Figure 6.23). Comparing the mean value scenario with the optimization scenario (1-min resolution), it becomes clear that only very minor deviations of 0,4% in the depiction of the total energy consumption for the same production output quantity can be noted (Table 6.7). The depiction of the exact consumption peaks is not possible in the mean value scenario. While it is possible to optimize the total energy consumption by using the total consumption optimizer experiment, the peak load optimization with averages does not lead to useful results. This is solely due to the fact that extreme values are no longer recognizable in the consumption profiles due to averaging and is neither caused by failure of the methodology nor the simulator.

182

6

Experimental Validation of the Methodology

Figure 6.21 Visualization of the energetic profiles of two example machines and the total energy consumption of both lines using table functions and mean values

As the comparison of the real energetic power profiles and the mean values has shown, the ideal choice of the data basis is highly dependent on the formulated objectives. If the focus is only on reducing the total energy consumption of a production, the use of mean values leads to satisfying results. The use of the realistic load profiles is necessary as soon as peak optimization experiments are requested to improve the energy efficiency.

6.3 Optimization Experiments

Figure 6.22 Optimization experiment run results using mean values

Figure 6.23 Peak optimizer experiment run for the mean value scenario

183

184

6

Experimental Validation of the Methodology

Table 6.7 Comparison of the mean value and the optimization scenario Optimization Scenario

Mean Value Scenario

Deviation

Output Quantity

13.728 pieces

13.728 pieces

0%

Total Energy Consumption

181.422,8 kWh

180.597,4 kWh

0,4%

Maximum Peak Consumption

57.1 kW

55.4 kW

3,0%

6.3.5

Summary and Evaluation of Simulation-based Optimization Results

After successfully testing the simulation-based optimization methodology in section 5.6 on a fictional production example, the suitability for practical use was examined in this chapter. Using the example of two production lines, each with five strongly linked machines, it was examined to what extent the lines could be operated more energy-efficiently by reducing non-value-adding machine times and decreasing power consumption peaks without negatively affecting the output quantity of the production lines. Since 2015, the practice partner Bosch has been carrying out a comprehensive energy data documentation for all machines in production. The energy data is made available via an energy portal with a resolution of one data record per 15 min time interval. All relevant production planning and control information is available, but processed and stored in different, noncoupled systems. With manually maintained data such as output quantities per shift, missing parts and machine downtimes, the problem arises that there is no time stamp for the data records synchronized with the production times. For this reason, it is not possible to automatically assign these data records to the energy data recorded for the production machines. A targeted machine state simulation on the real production machines made it possible to assign energy consumption profiles and exact production numbers and thus generate a database that could be used to build the simulation model for the method. Since the energy consumption data from the energy portal could not be assigned to the machine states with a satisfying accuracy using a resolution of 15 minutes, the machine controllers and external measuring devices recorded the energy consumption data with a resolution of one data record per minute and one data record per second. This data was assigned it to the machine states. For each of the two resolutions, a simulation model for a reference scenario and an optimization scenario were created. In addition, the energy load profiles were

6.3 Optimization Experiments

185

replaced by machine condition-related mean values in an extra model (mean value scenario) for the 1-min resolution datasets. The reference scenarios with the different data resolutions of one data record per minute and one data record per second were tested, both with the same production task (13.728 pieces/week) using the same time frame for the simulation experiment (May 20th starting at 06:00 am until May 24th finishing at 11:59 pm). Comparing both simulation runs, the deviation of the total energy consumption was only 0,2%, even though the energy profiles of the different machine states in the two data resolutions differed between 0,1 and 2,8% depending on the machine looked at. The comparison of the total energy consumption of the optimization scenarios with the 1-min and the 1-sec resolution showed a deviation of 0,7%. At this point it becomes clear that the deviations in the energy consumption profiles of the machine states caused by the resolution affect the results. If one compares the ideal parameter configurations determined for the idle and standby optimizer parameters in the two optimization scenarios, one also notices deviations here (Table 6.8). Table 6.8 Comparison of the idle and standby optimizer parameter configurations in the optimization scenarios at different data resolutions Optimization Scenario (1-second resolution)

Optimization Scenario (1-minute resolution)

IdleOptimizer_Line1

1.872 sec (31,2 min)

44,8 min

IdleOptimizer_Line2

1.874 sec (31,2 min)

19 min

StandbyOptimizer_Line1

1.140 sec (19 min)

19 min

StandbyOptimizer_Line2

1.140 sec (19 min)

19 min

IdleOptimizer_L1wash

3.600 sec (120 min)

120 min

IdleOptimizer_L2wash

3.600 sec (120 min)

120 min

The deviation of the single machine state energy consumption profiles in the two data resolutions resulted in different optimal parameter configurations to reach the minimal total energy consumption value. The data-resolution-dependent optimal parameter configurations led to the fact that the single machine states have different shares of the total simulation time varying in a range of 0,1% to 3,3% (Figure 6.24). The data resolution comparison shows that impurities and deviations in the database affect the entire simulation and its different experiment runs and have a significant impact on the results. The use of different measuring instruments to

186

6

Experimental Validation of the Methodology

Figure 6.24 Comparison of the machine state time shares of the optimization scenarios in different data resolutions

record the different data resolutions has proven to be a potential source of error. Nevertheless, it can be stated at this point, especially for the practical implementation, that both data resolutions deliver realizable results with a similarly high total energy consumption. The solutions differ noticeable regarding the retention time in the idle state, which at the higher resolution is to leave both lines just over 30 minutes in the idle state, while at the 1-min resolution the suggestion is to leave line 1 over 40 minutes and line 2 under 20 minutes in the idle state. It was assumed that the limited number of allowed iterations in AnyLogic per experiment might be a reason for the differing optimization parameter values. The test described below confirmed this assumption. To cross-check the effect of the optimizer parameters on the total energy consumption value, the optimal parameter configuration set of the 1-min resolution optimization experiment was tested in the 1-sec resolution scenario simulation (Figure 6.25) and vice versa. The total consumption value in the 1-sec resolution simulation is thus about 0,35% lower than with the parameter configuration determined in the 1-sec resolution experiment run. This result lead to the check, if the optimization experiments test the optimal parameter configuration of the other resolution during the experiment run at all. While the values close to the optimal parameter set of the 1-sec resolution experiment were tested during the

6.3 Optimization Experiments

187

Figure 6.25 Results of the simulation run (1-sec resolution optimization scenario) with the optimal parameter set of the 1-min resolution optimization experiment

500 iterations of the total consumption optimization experiment run with the 1min resolution, the optimal parameter set of the 1-min resolution is not tested during the 1-sec resolution experiment run at all. It therefore seems reasonable to assume that the limitation of iterations per optimization experiment in the AnyLogic software to a maximum of 500 is not enough. The higher time resolution in the 1-sec scenario leads to the fact that the values of the optimization parameters in smaller steps, so that the 500 experiments are not sufficient to determine an overall optimum. The examination of the necessity of the combined simulation in practice was done using mean energy consumption values instead of exact energy consumption profiles to eliminate the SD component of the methodology on the internal machine level. The mean values were calculated on the basis of the machine state-dependent energy load profiles and used as parameters in the machine logic. The total consumption optimizer experiment and the peak consumption optimizer experiment were executed. While the peak consumption optimizer was not able to define an offset leading to a lower power peak level, the total consumption optimizer calculated nearly the same optimal parameter configurations as in the optimization scenario with the load profile data (Table 6.9). The use of mean values in the simulation-based optimization methodology proved that the total consumption optimization for a production scenario led to reliable results that were comparable in quality to those calculated using the real energy consumption profiles. If the objective in production is limited to the reduction of total energy consumption only, this can be done using the methodology

188

6

Experimental Validation of the Methodology

Table 6.9 Comparison of the parameter configurations in the mean value and the optimization scenario (1-min resolution)

IdleOptimizer_Line1

Optimization Scenario (1-minute resolution)

Mean Value Scenario (1-minute resolution)

44,8 min

45,9 min

IdleOptimizer_Line2

19 min

19 min

StandbyOptimizer_Line1

19 min

19 min

StandbyOptimizer_Line2

19 min

19,03 min

IdleOptimizer_L1wash

120 min

120 min

IdleOptimizer_L2wash

120 min

120 min

with mean values. It is also conceivable to integrate individual machines for which there are no exact energy profiles but only mean values into an optimization scenario with exact energy consumption profiles.

6.3.6

Financial Evaluation of the Increase in Energy Efficiency

This section assesses the financial evaluation of the energy savings that can be achieved by the methodology. The data with a resolution of one data record per minute is used as the basis for this. This is since the data produced results of the same quality as the higher resolution data but was recorded by the machine controller and thus no additional costs were incurred in this scenario for the purchase and installation of external measuring equipment. As already described at the beginning of this chapter, Bosch purchases 95% of the total energy needed externally. The details of the framework agreement for the electricity supply are available for the evaluation of the optimization results. Table 6.10 lists all relevant consumption-dependent energy price components. The power price (electricity costspower ), which is essential to financially evaluate the peak reduction, is known but only for the currently required grid capacity. Since there is no information on price differentiation depending on the height of the grid capacity, it is not possible to financially assess the savings achieved by the reducing the energy consumption peaks. As the grid capacity costs account for about 12% of the total energy costs per months, the missing option to financially evaluate this data is quite disturbing but assumptions about possible price differentiations do not contribute to a realistic assessment. For this reason, the financial evaluation of the peak reduction is omitted.

6.3 Optimization Experiments

189

Table 6.10 Price components of the electricity supply contract

Electricity costswork (day tariff)

Unit

Price

MWh

41,15 e

EEG apportionment

MWh

67,92 e

Sum of electricity taxes and other charges

MWh

32,95 e 142,02 e

Total Costs

The total energy consumption savings that result from using the optimization parameters proposed in the total consumption experiment are calculated as follows: E save/week = E total Re f Scene − E total O pt Scene

(6.1)

E save/week = 193.195,6 kW h − 181.422,8 kW h = 11.772,8 kW h E save/week = 11.772,8 kW h = 11,8 M W h E save/month = 11,8 M W h ∗

52 12

= 51,0 M W h

csave/month = 51,0 M W h ∗ 142,02 e = 7.243,02 e

with

E save/week E save/month csave/month

calculated weekly energy savings calculated monthly energy savings calculated monthly cost savings

By optimizing the operation mode of the production lines under consideration, the energy costs for these two lines can be reduced by 7.200 e/month (87.000 e/year). The next step is to extend the scope of the simulation-based optimization to the consideration of entire production areas in the model. The extension of the model allows a statement about the scalability of the results. If the optimization potential for all production lines and machines is as high as in the area under consideration, it can be assumed that consumption-dependent energy costs at Bosch can be reduced by 6%.

190

6.4

6

Experimental Validation of the Methodology

Recommendations for Action for Bosch Derived from the Simulation-based Optimization Methodology

In the previous sections, it was demonstrated that the simulation-based optimization methodology is suitable for practical application. Basically, the creation of a simulation model requires the existence of the corresponding data. In addition to the generally needed data for production planning and control, the energy consumption data of the machines must be available. It has been shown that the resolution of the underlying data base for the energy consumption profiles should be chosen in a way that it is accurate enough to represent the consumption behavior and the load peaks of the machine states. In the case of Bosch, it has been shown that the resolution of the energy data provided on the energy platform is not suitable for use in the simulation model to conduct optimization studies. Due to the selected 15-minute interval, the energy states cannot be clearly recognized and separated from each other in the energy data. Thus, a realistic reproduction of the energy consumption behavior of machines in the simulation model is not possible. It is therefore recommended to use energy data with a resolution of one data record per minute. Should data protection regulations prevent such a high resolution from being made available (temporarily or permanently), machine state-based mean values should be calculated on the basis of the 1-min resolution data and used in the simulation model instead of the exact energy profiles, since good results can be achieved with correctly calculated mean values, at least with regard to the total consumption optimization. The use of different information technology (IT) systems to document different production and machine data has proven to be disadvantageous, as non-synchronized time stamps and different data acquisition methods require high manual efforts for the merging of production and energy data. The documentation of all data in one single system having the big advantage of one consistent time used for all data records, would considerably simplify and accelerate the connection of relevant information. It is recommendable to integrate the documentation of production quantities, machine failures and maintenance plans in the ERP/MES system instead of maintaining them in several different applications. The parameter configurations determined as optimal within the framework of the methodology can only be used as recommendations for action to switch off machines during production shutdowns if they are based on a correct and reliable data basis.

6.5 Evaluation of the Practical Applicability …

191

With a complete database, the methodology can be used to simulate production forecasts of products that have already been produced on the production lines before. New product start-ups are hard to simulate in advance, since the methodology can only be used if the energy consumption of the individual machines during the productive machine states is known in the form of value tables, mean values or mathematical equations. If those data do not exist for new products, the simulation-based optimization methodology cannot be used. For a complete database, the simulation-based optimization methodology can be used to optimize individual production lines or entire production areas. It can thus be utilized in machine or production line related projects to increase energy efficiency, but also for the holistic optimization of the energy efficiency of the total production area. In order to become familiar with the methodology and the simulation software, it is advisable to start with small projects and later combine the models to a holistic production model. The financial valuation of the achievable energy savings was based on the energy prices of the current valid framework agreement from 2013. It is striking that the tariff for consumption-dependent energy costs is relatively high compared to electricity bills from other companies that were used as a reference. The negotiation of new framework agreements for the electricity supply might be a further starting point for reducing energy costs.

6.5

Evaluation of the Practical Applicability of the Simulation-based Optimization Methodology

Following the evaluation of the optimization results and the summary of the resulting recommendations for action for the practical partner, this section assesses the applicability of the simulation-based optimization methodology in practical use. For this purpose, the defined concept objectives and requirements listed in section 5.1 are evaluated with regard to their implementation in the methodology (Table 6.11). Table 6.11 shows that the methodology meets the requirements in many respects. By applying the methodology, the interactions between production processes and energy use in production can be presented with high transparency. Additionally, processes can be modeled very accurately using the various required simulation paradigms. The modular, clear structure and the many parametrization possibilities support the simulation of small production areas, but also the extension to entire productions. The systematic approach to model creation and the execution of optimization experiments provides the user with instructions for the

192

6

Experimental Validation of the Methodology

Table 6.11 Evaluation of the proposed simulation-based optimization methodology Requirements

Evaluation of the methodology

Transparency

The methodology contains all relevant material and energy flows as well as their dynamic interdependencies in one single software solution to increase transparency in production towards improvements. The optimization of energy parameters is possible in two separate optimization experiments clearly having the focus to reduce operating costs of processes.

Accuracy

The methodology follows a realistic depiction of the machine processes and the linked energy consumption behavior of all machines. The introduced machine logic in combination with the energy component depict the machine state and machine state transitions in detail to support the event-based material flow and time-continuous energy flow simulation in different, customer-chosen levels of detail.

Simplicity

The implementation in one software tool (for example in AnyLogic) to provide a holistic view without the complexity of interface management, complex data exchanges, and synchronization requirements is possible.

Expandability

The methodology is applicable to different production scenarios and it can be used to model and optimize single production sub-systems, production lines or the entire production.

Parametrization The methodology can easily be configured for different production scenarios. Case-specific adaptations (e.g. the optimization of entire production lines instead of single machines) are supported by the approach. The use without a simulation expert is possible. Modularity

The methodology is built up modular to allow the flexible coupling and fast adaptation of single system components. All relevant functions and systems of a production are represented in fast configurable components. Interactions in the production are depicted using defined interfaces.

Optimization

The total energy consumption and peak load optimization experiments are included in the methodology supporting the optimization of production machines, production lines or entire shop floors. The optimization of the energy efficiency is done following an integrative approach, considering the energy consumption together with traditional production planning aspects as time, quantity and quality. So far only parameter optimizations but no scheduling optimization approaches have been considered in the approach.

Predictability

The methodology supports the simulation of forecasts for known products that have already been produced in the shop floor. A forecasting of energy requirements for unanalyzed combinations of operating machines, process scheduling tasks and process parameters is not possible yet. (continued)

6.5 Evaluation of the Practical Applicability …

193

Table 6.11 (continued) Evaluation

The optimization parameter configurations are portrayed comprehensibly to function as an active decision-making tool. An evaluation tool to provide Energy and Key Performance Indicators (EnPIs and KPIs) and their development over time are not proposed as the practical application of the methodology has shown, that KPI’s and visualization tools are company specific and target group dependent.

Visualization

The visualization of the energy consumption in the simulation was implemented, a visualization of the optimization results for the use in the company was omitted, since the presentation forms of results are company-specific and strongly target group dependent.

Application

The description of the methodology in this book allows a systematic implementation in production. The guidelines intend to avoid unnecessary repetitions of implementation steps, support the individual implementation phases by means of proven procedures, and describe required database setups and data validation procedures.

Usability

The developed methodology can be used to simulate and optimize different production scenarios. Two examples for the implementation in AnyLogic are given in this book. The use of other software systems supporting multi-method modeling should be possible as well but has not been tested so far. For the implementation in AnyLogic, software licenses need to be bought. The implementation is possible without simulation specialists but requires the presence of employees with basic simulation skills and knowledge of the java programming language.

systematic application of proven procedures. For reasons of convenience the use of a single, multi-method-capable simulation software is recommended for the implementation of the methodology. However, depending on the chosen granularity of data, certain limitations might occur, as it is the case in the optimization experiment of the practical case study using a 1-sec resolution in AnyLogic. So far, the simulation-based optimization methodology does not support any forecast functionalities, neither does it include a full set of evaluation or visualization tools. These three points offer starting points for improving the methodology. In summary, however, it can be said that the developed methodology is well suited for practical use in manufacturing companies in order to increase energy efficiency, as it provides significant support in uncovering non-value-adding but energy-consuming machine times. In addition, recommendations for action to reduce the total energy consumption by depicting ideal machine state changes and timed offsets for the avoidance of energy load peaks.

7

Summary and Outlook

“As the saying goes, the stone age did not end because we ran out of stones; we transi-tioned to better solutions. The same opportunity lies before us with energy efficiency...” Steven Chu (Former United States Secretary of Energy and Nobel Laureate (1997) Dr. Steven Chu in his letter to the employees of the U.S. Energy Department [Ch2013].)

The final chapter briefly summarizes the contents and the research results of this book. In addition, an evaluation of the developed methodology is conducted and an outlook on possible future work is given.

7.1

Summary

The aim of this book was to develop a simulation-based optimization methodology for improving energy efficiency in production without requiring large financial investments in new, efficient production technologies. The problem, the motivation, and the research questions were described in chapter 1. In chapter 2 the relevant theoretical background on the state of the art of simulation and optimization technologies, as well as on the combination of the two was outlined. Chapter 3 comprised the basic principles of energy, necessary definitions of energy consumption, energy efficiency and energy saving potentials in manufacturing processes and production systems. It also described the composition of energy costs. The detailed literature review in chapter 4 showed that the approaches described in literature do not satisfactorily solve the problem of integrating © The Author(s), under exclusive license to Springer Fachmedien Wiesbaden GmbH, part of Springer Nature 2021 A. C. Römer, Simulation-based Optimization of Energy Efficiency in Production, Forschung zur Digitalisierung der Wirtschaft | Advanced Studies in Business Digitization, https://doi.org/10.1007/978-3-658-32971-6_7

195

196

7

Summary and Outlook

energy aspects in production simulations. Chapter 4 therefore concluded with a derivation of the current research demand. The research gaps identified were the lack of an easy applicability of approaches, the missing depiction of dynamic interdependencies of the discrete material flow and the continuous energy consumption, lacking support of integrated optimization, and missing approaches solving the combined depiction of production details and energy aspects in one single software solution. The identified research gaps were concretized into requirements and objectives in chapter 5, followed by the development of a conceptual framework for the simulation-based optimization of energy efficiency in production which was implemented in a fictional case scenario at the end of the chapter. In chapter 6, the developed methodology has been tested on a practical example. In addition to details on data acquisition, processing and validation, the simulation model and the optimization experiments using differently resolved energy data have been described. Additionally, the necessity of combined simulation approaches in practice has been examined. Chapter 6 closed with a financial evaluation of the optimization potential in the practical use case and an analysis of the practical applicability of the simulation-based optimization methodology. The methodology consists of two modules, a simulation and an optimization module. The simulation module is divided into three components in order to represent the production processes at all levels, from the internal process level to the macro level. Following the industry standard in production simulation, a material flow component is defined in which the processes in production are modeled discretely in a process-oriented manner. In order to exactly reproduce the machine processes with their individual machine states in production, an agentbased machine logic was developed to represent the processes at the micro level. Thus, the machine states, state changes, and state durations can be modeled realistically. The link to the energy consumption behavior is established via the energy component of the simulation module. The energy component comprises energy load profiles which are assigned to the individual machine states and are used accordingly during the simulation experiment runs. In order to not only model the energy aspects in production but also to use them for optimization scenarios, lexicographic objective functions have been derived that propose ideal parameter configurations for the energy-efficient operation of the production line using simulation-based optimization experiments. The focus of the optimization is on the reduction of the total energy consumption by avoiding non-value adding machine states. In addition, the methodology includes the possibility to determine an offset parameter to change the machine starts within an allowed time frame to reduce occurring load peaks. The developed methodology was first implemented in a fictitious case study using the software AnyLogic (PLE

7.2 Critical Appraisal of the Methodology

197

8.5.1), before it was tested in industrial practice using the example of two production lines at an automotive supplier in chapter 6. The implementation shows that the developed methodology is working for practical applications, but the quality of the results is highly reliant on the quality of the database and the completeness of information from the real production used for the construction of the simulation model.

7.2

Critical Appraisal of the Methodology

The objective of the presented approach is the depiction of combined material and energy flow simulation in production to create a realistic model showing the dynamic behavior of the energy consumption in production process. The use of a multi-method simulation software simplifies the hybrid modeling process. The complexity of interface management, complex data exchanges, and synchronization requirements are reduced compared to the usage of different software packages for the single simulation paradigms. Occurring peak demands and timecontinuous energy consumption can be shown closer to reality compared to DES models based on measured operating states, which are considered to be constant over a defined period of time. The practical implementation has shown that it is possible to build a hybrid simulation model, using the three modeling paradigms SD, DES, and ABS for representing the energy consumption behavior in production on the basis of historical data. It was also pointed out that the simulation model, in combination with forecast figures regarding quantities and planned schedules, can be used to depict the future consumption very precisely with its upcoming peak loads and non-value-adding production phases. From the optimization experiments carried out, alternative courses of action for the energy-efficient control of machine states emerge which lead to energy consumption reductions of 10% and more in production situations similar to the fictional example and savings of still 6% in strongly connected production lines. The level of optimization potential is strongly dependent on the starting point. If measures to increase efficiency have already been implemented in production, e.g., rough specifications have been installed that machines must be switched off during downtimes exceeding 60 or 90 minutes, the optimization potential is correspondingly lower. If machines are not decoupled from one another by buffers, it makes sense to take a holistic view of the production line. In this case again, the optimization potential will be lower than for machine state optimization, in which the machines can be individually set to their optimum.

198

7

Summary and Outlook

In this work failures have not been considered, as the timing of failures in the model will always be different from their actual future occurrence in production. Thus, a small deviation in production quantities between the real production and the simulation model has been noticed. As the consideration of failure probabilities would have a negative effect on the ability to detect consumption peaks, they might stay undetected due to a machine failure occurrence in the model but not in the real production, machine failures are not part of the simulation model. One prerequisite for the proposed method is the existence of measured energy consumption data at adequate levels of resolution. While energy monitoring systems are gradually being recognized and installed as efficiency tools in many companies, such an accurate tracking of the energy consumption on machine level in a very detailed resolution is rather the exception. To grant flexibility in case load data is not available, the simulation approach has been tested with mean values instead of energy load profiles. For each machine state a mean value has been added as a parameter, which is taken as the consumption value for the entire duration of the state. While this did not have any influence on the detection and avoidance of unproductive states and only a small deviation regarding the total energy consumption of the line through the use of mean values was noticed, the existing peak loads in the consumption profiles disappeared completely. The model accuracy should therefore be chosen depending on the objective of the investigation. A combination of the use of energy data tables and mean values is also conceivable if not all machines have yet been connected to the energy data acquisition in a production but should already be taken into account. There are only limited advantages of the use of real energy profiles, when peak reduction is not an issue in production. As the use of the exact energetic load profiles requires intensive data work, it might be advantageous to use the mean value version which reduces modeling efforts and the experiment run in the simulation and optimization experiments drastically. Both, the fictional and the real example have shown that a lexicographic optimization provides better results than the optimization of both objectives in one single experiment. Since the reduction of the total energy consumption is the primary goal, it makes sense to define two consecutive optimization experiments. This procedure also has the advantage that not both objectives need to be pursued in every project, but only the total energy consumption or only the peak loads can be optimized. An optimization of both objectives in a single optimization experiment would technically be possible by using a cost model. In practice, however, this approach failed because the energy companies were not willing to reveal consumption-based grid fees. Thus, it was impossible to assess the peak consumption reductions monetarily.

7.2 Critical Appraisal of the Methodology

199

In section 6.5 the simulation-based methodology is evaluated with regards to the practical applicability. While several requirements are included and fully met by the simulation-based optimization methodology, for example, the transparent presentation of interactions of single production aspects, the modular structures with a high number of parametrization options and the included optimization experiments that can be executed without requiring complex interface management to link different software solutions, some concept objectives and requirements remain unsolved. On the one hand, the commitment to AnyLogic meets the requirement of having one single software solution supporting both, multi-method simulation and optimization experiments. On the other hand, limitations regarding the number of supported database sizes, the number of possible optimization parameters, and allowed iterations in the optimization experiments arise. The coupling of a multi-method simulation software with an optimization software could eliminate those limitations but results in higher efforts for interface management and data-exchange and thus, probably demands for a simulation expert. The simulation-based optimization methodology proposes optimized parameter configurations for energy-efficient machine control. The idle and standby optimizer parameters define from what production interruption lengths on it is more efficient to turn the machine off rather than having them run in idle or standby state. However, to bring these parameter configurations in the real production, the employees need to know at the beginning of each interruption how long it will be. Depending on the causes for the production break down, the interruption length cannot be estimated in advance. While the point in time when a restocking of the buffers will end a material shortage might be known from supplier information or planned material deliveries in the ERP system, machine break downs add an unknown factor to the system. If production interruption durations can only be roughly estimated, the worker has to decide on his own, if he follows the proposed parameter configurations and switches the machines off or if he leaves them in idle mode in case the interruption might be too short to invest the energy and time required for the warmup of the production lines. It can be summarized, that the total consumption optimization requires the knowledge of interruption lengths to make use of the optimization results. The optimal value for the offset parameter on the other hand is determined within a range of a few seconds to a few minutes to avoid negative effects on the production output. The applicability to manually switch machines on and off that precise highly depends on the level of automation in production. To realize the recommendations for the machine offset, it requires an automated mechanism,

200

7

Summary and Outlook

e.g., a direct signal to the machine controller. Depending on the level of digitalization in production, this will only be feasible if Industry 4.0 machine and system networking standards are already installed. Another problematic aspect regarding the offset parameter is the consideration of machine failures in the simulation model. The consideration of machine failures would add a stochastic to the model, which requires the setup of replications for the single iterations of the optimization experiment, which can be done easily. But the occurrence of the machine failures in the simulation run and the real production would never be the same point in time. Therefore, in this work, failures are not considered, to be able to detect all possible consumption peaks that might occur in practice and would be hidden in the simulation model due to simulated failures. The disadvantage of this procedure is, that once a failure actually occurs in the real production line, the calculated offset parameter is not valid anymore, because in practice, the energy consumption profiles that are added up to get the total consumption of the production are different from what is simulated in the model. Despite the limitations mentioned above, the advantages of using the simulation-based optimization methodology to reduce the energy consumption in production predominate. For researchers, there is a contribution to knowledge in the fields of hybrid simulation and hybrid system modeling. At this point it is emphasized that particularly the approach for creating the conceptual model for hybrid simulations is scientifically relevant, since only a very few examples but no established methods can be found in this area. The contribution of knowledge for practical application lies in the generation of a step-by-step approach that can be implemented in a multi-method simulation software without the need for a simulation expert or programmer. The methodology can easily be applied for single machines, production lines or whole production areas and is enlargeable at any time in a project. In principle, only energy consumptions can be simulated that have previously been measured. Relationships between technological production parameters (feed rate, processing speed, etc.) and the resulting energy consumption are not modeled. The presented research results demonstrate that the developed method is suitable for considering energy efficiency goals in traditional production simulation. Thus, the research questions formulated at the beginning of the thesis can be answered as follows: Q1.

How can the energy consumption of a production system be depicted in a simulation model that can as well be used for optimization scenarios?

7.2 Critical Appraisal of the Methodology

201

The combination of different simulation paradigms allows the exact depiction of different production aspects to create a holistic simulation model. Following the current state of the art for production simulation, all relevant material flow details are modeled using the process-oriented discrete event simulation. To be able to model the energy consumption of machines, it is required to model the different possible machine states as well as state transitions and allowed state changes accordingly, as the energy requirements of the individual operating states can vary greatly. The agent-based simulation provides state graphs, which are ideal to model the operational inside of machines. While the agent-based simulation is also following a discrete time progress, the energy consumption behavior of machines is modeled continuously using stock and flow elements of the system dynamics library. The individual energy profile sections of the corresponding machine states are extracted from the measured total energy consumption profiles of the production machines and can be assigned to the machine states. Continuous variables for the energy state are created by considering the ratio of the passed time in a state and the remaining time. Thus, it is possible, to depict the energy consumption at any point in time, even in between events of the discretely modeled material flow. The single machine and energy states of the machines can be differentiated according to time and optimization aspects and can thus be split up into three groups, technically required operations, value-adding, and non-value-adding operations. While the technically required and the value-adding operations do not provide any optimization potential, the non-value adding but energy consuming machine times are in focus of energy efficiency optimizations. In preparation for the use of the simulation model for those optimizations, the simulation module needs to be parameterized as far as possible. Applicable restrictions imposed by the production scenario as well as required parameters, allowed deviations from a given schedule, and mandatory retention times for single machine states need to be introduced. To have machines in a production ready state just-in-time, an automated calculation to determine the remaining time until a machine has to be ready for production can be added to the model in preparation to define the most efficient production line setup in the optimization scenarios. Building the hybrid simulation model using the three sub-models based on different simulation paradigms together with the described model adaptations allows a realistic reproduction of production and energy consumption data and thus forms the basis for energy efficiency optimizations such as the optimization of the total energy consumption or peak-load avoidance.

202

Q2.

7

Summary and Outlook

How can an energy efficiency optimization of a production system be executed without causing any restrictions on production flexibility, without influencing the quality or the output of the production?

Firstly, all existing production requirements need to be included in the simulation model. Relevant production data, e.g., shift and maintenance plans, setup times, and production quantities, as well as known technical restrictions, for example minimum retention times in a machine state, batch sizes and machine capacities, have to be modeled. Secondly, the production flow functions as a pace maker process. The number of produced goods, the production machine states or state durations are no subjects to optimization. It is assumed, that production times, quality and process setups are ideal already. Thus, the optimization experiments are only focusing on non-value-adding production times and therefore optimization proposals to increase the energy efficiency are not at the expense of the output quantity. The same applies for the peak load optimization. As the peaks can only be reduced by using machine offsets of a length that does not cause a change of the output, delays are permitted only within a time frame of a few minutes per production shift. Can energy be integrated as a control parameter for production optimization? The machines with the medium and the high complexity machine logic types contain non-productive but energy-consuming states (idle and standby state). The non-productive states are generally used to bridge production interruptions. Depending on the length of a production stop, it is more energy efficient to shut the machines down and restart them later than having them remain in idle or standby for the whole duration of the interruption. It is possible to determine the period of time that the non-value-adding states should be used for bridging production interruptions and as of when a shutdown of the machines is the more energy-efficient variant. To be able to perform the calculation, optimization parameters (idle and standby optimizer) were included in the machine logic. The simulation-based optimization methodology contains two optimization experiments, each comprising an objective function and different optimization parameters. The optimal parameter configurations for the objective functions are determined during the optimization experiment runs. The total energy consumption optimization aims at finding the parameter configuration for the

7.2 Critical Appraisal of the Methodology

203

lowest possible total energy consumption value for the given production scenario. The optimal values determined for the idle and the standby optimization parameter can be brought back to the setup of the simulation model and are thus used to simulate the ideal switching behavior of machines in cases of production interruptions. Besides the optimization parameters for the non-productive states, an offset parameter was added to the model to reduce occurring power peak loads. The offset parameter causes a defined delay in the machine start of production machines and this leads to the fact that the single energy consumption profiles add up differently in specific points in time. The height of power peaks may thus be reduced. Those examples show that it is possible to integrate energy as a control parameter for different scenarios of production optimization. Q3.

How can the profitability of the energy optimization methodology be rated considering the various fields of application?

As the prototypical implementation as well as the practical validation have shown, the simulation-based optimization methodology offers great potentials for the energy-efficient design of production processes. The optimization of the total energy consumption through a reduction of non-value adding but energy consuming production times can be implemented immediately in every production without requiring high financial investments in advance. The approach is not limited to a certain production scenario but can be applied to any production situation, starting from single machine optimizations to the energy efficient design of the total production. For the calculation of the profitability, only the costs for the simulation model generation have to be opposed to the cost saving potentials. The savings are calculated using the prices of the current energy contract. The height of the optimization potentials that can be realized vary depending on the production and its optimization state. Have general rules and measures to save energy already been applied, the achievable potential of the simulation-based optimization methodology will be evaluated to be lower than in a production were no steps to improve the energy efficiency have been taken so far. To what extent should the periphery of a production system be included in the simulation-based optimization in order to obtain a quantifiable statement about the energetic behavior of the entire system?

204

7

Summary and Outlook

The more details a simulation model contains, the more realistic is the reproduction of the actual production scenario. The efforts to create the model and especially to collect and validate the data need to be adapted to the model purpose and the followed objectives. It is not target-oriented, if the efforts to include energy consumption data of the entire production periphery require high investments in measuring equipment when the share of the production periphery is very low compared to the share of production machines. At the end of the day, the decision is the sole responsibility of the model maker. Does the production hall contain an energy-intensive cooling system to temper the production, it makes sense to include it in the simulation model as the energy share of this equipment is high compared to a vent that might be placed at a worker’s desk. Generally, all main energy consumers should be considered in the model creation process to obtain a realistic picture of the production and to be able to use the simulation model for energy efficiency optimizations. Nevertheless, optimization potentials on machine or production line level are a first step towards an energy efficient production.

7.3

Outlook and Future Work

An essential requirement of the method for simulation-based optimization of energy efficiency in production was the realistic and exact representation of energy consumption in production. Particularly during the practical validation of the methodology, it has been proved that the energy consumption of production machines and production lines can be depicted very precisely by means of hybrid simulation and that it is possible to use the hybrid simulation models for optimization experiments to reduce the total energy consumption as well as peak consumptions in production. So far, the methodology has only been tested with single production lines. Following the requirement of an integrated view of production systems, the methodology has to be tested considering the whole production area of a company. Due to the simple possibility to extend the existing model and to easily integrate the next production lines into the same simulation model. With increasing model size the performance of the simulator might be influenced negatively. Subsequent to the extension of the simulation model on the production side, the consideration of transport and logistic processes that have been excluded in the production flow component could be a future enhancement. In a next step, the

7.3 Outlook and Future Work

205

extension of the methodology to depict energy consumption of the technical building equipment could be considered as an option. Cooling and heating systems, air conditioning and other energy consuming equipment of the building infrastructure could be modeled to be able to assess and to optimize the share of energy consumption they cause. The simulation-based optimization methodology grants flexibility which overcomes the need for predefined system borders. Besides the production processes, logistics processes, as well as technical building equipment can be modeled with the simulation paradigms used. A limitation to the production area is therefore not necessary. It should be examined whether the creation of further simulation modules for uniform modeling of the building infrastructure would simplify the creation of the simulation model, as the defined machine logic types of the machine component might have to be changed to represent heating and cooling processes or transportation systems realistically. The cooperation with the practice partner repeatedly led to the demand for the use of the methodology to model compressed air consumptions as well as cooling and lubricant circuits. Initial research and experiments have shown that the hybrid approach could also be suitable for depicting such media, as they also exhibit continuous behavior which is highly dependent on machine behavior. So far, all required data has been collected and matched manually. The automated data integration would bear potential for improvement. An interface to the required data sources such as the ERP system for production and process data as well as a coupling to the energy data portal could provide the latest data and thus improve the data quality used for the simulation model. In order to provide a better and faster decision support, the automated transfer of optimization parameter values without manual interference to the production machines should be tested. In addition to the different approaches for extending and improving the simulation model, the optimization component of the methodology also offers starting points for future work. Whereas in this approach only a parameter optimization was carried out in two separate optimization experiments, a next step could be an optimization based on a cost function. The definition of a cost function would allow the overall optimization purely under consideration of the energy costs focusing the total consumption and the peak optimization costs in parallel. Due to the lack of a reliable database for the grid costs, the creation of a cost function has so far been dispensed. The introduction of a cost function would additionally allow the consideration of time-dependent energy prices in the course of the day as proposed in Johannes, Wichmann, and Spengler [JWS2019]. One option to overcome the limitation of the methodology to simulate and optimize only known and measured machine consumptions might be the deep

206

7

Summary and Outlook

learning-based prediction of energy consumption for hybrid simulation. The combination of hybrid simulation and machine learning is well suited for problems in which behavioral aspects are difficult to model, but enough data of the real system exists [Be+2019, p. 1357]. Wörrlein et al. propose the use of a hybrid system model with a Recurrent Neural Network (RNN) encoder-decoder architecture that returns a discrete time series when a behavior sequence has been inserted into a neural network model of the machine. This allows the energy consumption of the machine to be displayed for each job executed. When weighted and called, the neural network emits the length of a job as well as a corresponding time series indicating the quasi-continuous time consumption of the order [Wö+2019, p. 121]. While being work-in-progress, this approach could have the potential for a better depiction of causal relationships between parameters of the simulation model and the predicted energy consumption.

References

[AGA2016] Ait El Cadi, A.; Gharbi, A.; Artiba, A.: Matlab/Simulink -vs- Arena/ OptQuest: Optimal Production Control of Unreliable Manufacturing Systems. In: 11th International Conference on Modeling, Optimization and Simulation; MOSIM’16. 2016. pp. 1–10. [AK2014] Ahrends, J.; Kabelac, S.: Das Ingenieurwissen: Technische Thermodynamik. Springer-Verlag, Berlin Heidelberg, 2014. [Am+2016] Amaran, S.; Sahinidis, N. V.; Sharda, B.; Bury, S. J.: Simulation optimization. In: Annals of Operations Research. 240(1), 2016, pp. 351–380. [An1998] Andradottir, S.: Simulation Optimization. In (Banks, J., eds.): Handbook of Simulation. Wiley; Co-published by Engineering & Management Press, New York, 1998, pp. 307–333. [An+2017] Angermann, A.; Beuschel, M.; Rau, M.; Wohlfarth, U.: MATLAB – Simulink – Stateflow: Grundlagen, Toolboxen, Beispiele. De Gruyter Oldenbourg, Berlin, Boston, 2017. [An2019] AnyLogic Company: Simulation Software for Every Business Challenge: Features. website accessed on October 19th , 2019. URL: https://www.anylogic. com/features/. [Ap+2004] April, J.; Better, M.; Glover, F.; Kelly, J.: New Advances and Applications for Marrying Simulation and Optimization. In (Ingalls, R. G.; Rossetti, M. D.; Smith, J. S.; Peters, B. A. eds.): Proceedings of the 2004 Winter Simulation Conference. IEEE, Piscataway, NJ, 2004, pp. 76–82. [Au2009] Augenstein, E. M. G.: Rechnergestützte Analyse und Konzeption industrieller Energiesysteme, Aachen, 2009, dissertation. [Au2019a] AutoSimulations, Inc.: AutoStat. website accessed on July 15th , 2019. URL: https://www.autosim.com. [Au2019b] AutoSimulations, Inc.: Evolutionary Optimizer (Extend). website accessed on July 15th , 2019. URL: https://www.imaginethatinc.com. [Ba1998] Banks, J.: Principles of Simulation. In (Banks, J., eds.): Handbook of Simulation. Wiley; Co-published by Engineering & Management Press, New York, 1998, pp. 3–30.

© The Editor(s) (if applicable) and The Author(s), under exclusive license to Springer Fachmedien Wiesbaden GmbH, part of Springer Nature 2021 A. C. Römer, Simulation-based Optimization of Energy Efficiency in Production, Forschung zur Digitalisierung der Wirtschaft | Advanced Studies in Business Digitization, https://doi.org/10.1007/978-3-658-32971-6

207

208

References

[Ba+2005] Banks, J.; Carson, J.; Nelson, B.; Nicol, D.: Discrete-event system simulation. Pearson Prentice Hall, Upper Saddle River, New Jersey, 2005. [Ba2012a] Balci, O.: A life cycle for modeling and simulation. In: SIMULATION. 88(7), 2012, pp. 870–883. [Ba2012b] Bangert, P.: Optimization for industrial problems. Springer, Berlin, 2012. [Ba2015] Bangsow, S.: Tecnomatix plant simulation: Modeling and programming by means of examples. Springer, Chem, 2015. [Ba+2015] Bader, N.; Bluecher, F. v.; Boeve, S.; Bourgault, C.; Hazrat, M.; Grave, K.: Stromkosten der energieintensiven Industrie. Website assessed on 18.01.16, https://www.ecofys.com/files/files/ecofys-fraunhoferisi-2015-str omkosten-der-energieintensiven-industrie.pdf. [Ba+2016] Baumann, F.; Wilson, H.; Seidel, S.; Franke, M.; Gromnitza, U.: OptPlanEnergy – An Optimization and Scheduling Platform for the Energy-Efficient Production of Tempered Glas. In (Wiedemann, T., eds.): Tagungsband ASIM 2016. 7. bis 9. September 2016, Dresden, pp. 121–124. [Ba+2017] Baumann, F.; Wilson, H.; Seidel, S.; Franke, M.; Gromnitza, U.: OptPlanEnergie – Der Einsatz von Simulation und Optimierung zur Verringerung des Energiebedarfs bei der Produktion von Sicherheitsglas. In (Wenzel, S.; Peter, T., eds.): Simulation in Produktion und Logistik 2017. 20. – 22. September 2017, Kassel, pp. 59–68. [Be2010] Benker, H.: Ingenieurmathematik kompakt – Problemlösungen mit MATLAB: Einstieg und Nachschlagewerk für Ingenieure und Naturwissenschaftler. Springer, Berlin, 2010. [Be2017] Beier, J.: Simulation Approach Towards Energy Flexible Manufacturing Systems. Springer International Publishing, Cham, 2017. [Be+2011] Berglund, J.; Michaloski, J.; Leong, S.; Shao, G.; Riddick, F.; Arinez, J.; Biller, S.: Energy efficiency analysis for a casting production system. In (S. Jain, R.R. Creasey, J. Himmelspach, K.P. White, and M. Fu, eds.): Proceedings of the 2011 Winter Simulation Conference. IEEE, Piscataway, NJ, 2011, pp. 1060–1071. [Be+2019] Bell, D.; Groen, D.; Mustafee, N.; Osik, J.; Strassburger, S.: Hybrid Simulation Development – Is it just Analytics? In (Mustafee, N.; Bae, K.-H.G.; LazarovaMolnar, M.; Szabo, C.; Haas, P.; Son, Y.-J., eds.): Proceedings of the 2019 Winter Simulation Conference. IEEE, Piscataway, NJ, 2019, pp. 1352–1365. [BF2004] Borshchev, A.; Filippov, A.: From System Dynamics and Discrete Event to Practical Agent Based Modeling: Reasons, Techniques, Tools. In (Kennedy, M.; Winch, G. W.; Langer, Robin, S.; Rowe, J. I.; Yanni, J. M., eds.): Proceedings of the 22nd International Conference of the System Dynamics Society. 2004, pp. 1–23. [BFG2015] Bursi, F.; Ferrara, A.; Grassi, A.; Ronzoni, C.: Simulation Continuous Time Production Flows in Food Industry by Means of Discrete Event Simulation. In (Chen, X. D., eds.): International Journal of Food Engineering. Band 11, Heft 1, De Gruyter, Berlin/Boston, 2015, pp. 139–150. [Bh2000] Bhatti, M. A.: Practical Optimization Methods: With Mathematica Applications. Springer New York, New York, NY, 2000. [Bi+2004] Biethahn, J.; Lackner, A.; Range, M.; Brodersen, O. B.: Optimierung und Simulation. Oldenbourg, München, 2004.

References

209

[BKN2009] Becker, J.; Krcmar, H.; Niehaves, B.: Wissenschaftstheorie und gestaltungsorientierte Wirtschaftsinformatik. Physica-Verlag Heidelberg, Heidelberg, 2009. [Bo2013] Borshchev, A.: The big book of simulation modeling: Multimethod modeling with AnyLogic 6. AnyLogic North America, Chicago, 2013. [Bö2013] Böge, W.: Vieweg Handbuch Elektrotechnik. Vieweg+Teubner Verlag, Wiesbaden, 2013. [BP2017] BP p.l.c.: BP Energy Outlook- 2017 edition. website accessed on January 12th , 2018, https://www.bp.com/content/dam/bp/pdf/energy-economics/energy-out look-2017/bp-energy-outlook-2017.pdf. [Br2008] Branke, J.: Multiobjective Optimization: Interactive and Evolutionary Approaches. Springer, Berlin, 2008. [Br2014] Brailsford, S.: Modeling Human Behavior – An (ID)Entity Crisis? In (Tolk, A.; Diallo, D.; Ryzhov, I. O.; Yilmaz, L.; Buckley, S.; Miller, J. A. eds.): Proceedings of the 2014 Winter Simulation Conference. IEEE, Piscataway, NJ, 2014, pp. 1539–1548. [Br+2019] Brailsford, S. C.; Eldabi, T.; Kunc, M.; Mustafee, N.; Osorio, A. F.: Hybrid Simulation Modelling in Operational research: A State-of-the-Art Review. In: European Journal of Operational Research. Volume 278, Issue 3, 2019, pp. 721– 737. [BSP2011] Bertsche, B.; Schauz, A.; Pickard, K.: Reliability in automotive and mechanical engineering: Determination of component and system reliability. Springer, Berlin, London, 2011. [Ca2013] Cavazzuti, M.: Optimization Methods: From Theory to Design Scientific and Technological Aspects in Mechanics. Springer, Berlin, Heidelberg, 2013. [Ch2009] Chahal, K.: A Generic Framework for Hybrid Simulation in Healthcare, London, 2009, dissertation. [Ch2013] Chu, S.: Letter from Secretary Steven Chu to Energy Department Employees. U.S. Department of Energy, Washington, 2013. website accessed on October 17th , 2019, https://www.energy.gov/articles/letter-secretary-stevenchu-energy-department-employees. [Ch+2013] Chen, G.; Zhang, L.; Arinez, J.; Biller, S.: Energy-Efficient Production Systems Through Schedule-Based Operations. In: IEEE Transactions on Automation Science and Engineering. 10(1), 2013, pp. 27–37. [CHR2002] Cassidy, D.; Holton, G.; Rutherford, J.: Understanding Physics. Springer, New York, NY, 2002. [CT2018] Chopard, B.; Tomassini, M.: An Introduction to Metaheuristics for Optimization. Springer International Publishing, Cham, 2018. [CTS2013] Cataldo, A.; Taisch, M.; Stahl, B.: Modelling, simulation and evaluation of energy consumptions for a manufacturing production line. In (Zucker, G.; Meléndez Augusto Nogueiras, A.; eds.): Proceedings IECON 2013 – 39th annual conference of the IEEE Industrial Electronics Society; 10 – 13 Nov. 2013, Vienna, Austria, pp. 7537–7542. [Da+2014] Djanatliev A., Kolominsky-Rabas P., Hofmann B.M., Aisenbrey A., German R.: System Dynamics and Agent-Based Simulation for Prospective Health Technology Assessments. In (Obaidat, M.; Filipe, J.; Kacprzyk, J.; Pina, N.;

210

References

[De+2007]

[De+2008]

[DG2013]

[DG2015]

[Di2014] [DI2017]

[DLR2004] [Do+2015] [DS2008] [DS2010] [DT2005]

[Du+2012]

[Du2018]

eds.): Simulation and Modeling Methodologies, Technologies and Applications. Advances in Intelligent Systems and Computing, vol 256. Springer, 2014, pp. 85–96. Devoldere, T.; Dewulf, W.; Deprez, W.; Duflou, J.: Improvement Potential for Energy Consumption in Discrete Part Production Machines. In (Takata, S.; Umeda, Y., eds.): Advances in Life Cycle Engineering for Sustainable Manufacturing Businesses. Springer-Verlag London Ltd, London, 2007, pp. 311–316. Devoldere, T.; Dewulf, W.; Deprez, W.; Eberspächer, P.: Energy Related Life Cycle Impact and Cost Reduction Opportunities in Machine Design: The Laser Cutting Case. In (Kaebernick, H., eds.): Applying life cycle knowledge to engineering solutions, Sydney, 2008, pp. 412–419. Djanatliev, A.; German, R.: Prospective Healthcare Decision-Making by Combined System Dynamics, Discrete-Event and Agent-Based Simulation. In (Pasupathy, R.; Kim, S.-H.; Tolk, A.; Hill, R. R.; Kuhl, M. E., eds.): Proceedings of the 2013 Winter Simulation Conference. IEEE, Piscataway, NJ, 2013, pp. 270–281. Djanatliev, A.; German, R.: Towards a Guide to domain-specific Hybrid Simulation. In (Yilmaz, L.; Chan, W. K.; Moon, I.; Roeder, T. M.; Macal, C. M.; Rossetti, M. D., eds.): Proceedings of the 2015 Winter Simulation Conference. IEEE, Piscataway, NJ, 2015, pp. 1609–1620. DiStefano, J. J.: Dynamic systems biology modeling and simulation. Academic Press, London, UK, 2014. DIHK Deutscher Industrie- und Handelskammertag Berlin – Brüssel: Faktenpapier Strompreise in Deutschland 2017. Website assessed on 01.02.2018, https://www.rostock.ihk24.de/blob/hroihk24/innovation_und_umwelt/downlo ads/3301616/3c7673170127d2b27c99962a6852b846/Faktenpapier--Stromp reise-in-Deutschland--data.pdf. Dyckhoff, H.; Lackes, R.; Reese, J.: Supply chain management and reverse logistics. Springer, Berlin, 2004. Domschke, W.; Drexl, A.; Klein, R.; Scholl, A.: Einführung in Operations Research. Springer Gabler, Berlin, Heidelberg, 2015. Dyckhoff, H.; Souren, R.: Nachhaltige Unternehmensführung: Grundzüge industriellen Umweltmanagements. Springer, Berlin, 2008. Dyckhoff, H.; Spengler, T. S.: Produktionswirtschaft: Eine Einführung. Springer, Berlin, Heidelberg, 2010. Dubiel, B.; Tsimhoni, O.: Integrating Agent Based Modeling into a Discrete Event Simulation. In (Kuhl, M. E.; Steiger, N. M.; Armstrong, F. B.; Joines, J. A., eds.): Proceedings of the 2005 Winter Simulation Conference. pp. 1029– 1037. Duflou, J. R.; Sutherland, J. W.; Dornfeld, D.; Herrmann, C.; Jeswiet, J.; Kara, S.; Hauschild, M.; Kellens, K.: Towards energy and resource efficient manufacturing. In: CIRP Annals. 61(2), 2012, pp. 587–609. Durán, J. M.: Computer Simulations in Science and Engineering: Concepts – Practices – Perspectives. Springer International Publishing, Cham, 2018.

References

211

[DV2009] Dietmair, A.; Verl, A.: Energy Consumption Forecasting and Optimisation for Tool Machines. In: MM Science Journal. 2009(01), 2009, pp. 63–67. [Eb+2014] Eberspächer, P.; Schraml, P.; Schlechtendahl, J.; Verl, A.; Abele, E.: A Modeland Signal-based Power Consumption Monitoring Concept for Energetic Optimization of Machine Tools. In: Procedia CIRP. 152014, pp. 44–49. [El+2016] Eldabi, T.; Balaban, M.; Brailsford, S.; Mustafee, N.; Nance, R.; Onggo, B.: Hybrid Simulation: Historical Lessons, Present Challenges and Futures. In (Roeder, T. M.; Frazier, P. I.; Szechtman, R.; Zhou, E., eds.): Proceedings of the 2016 Winter Simulation Conference. IEEE, Piscataway, NJ, 2016, pp. 1388– 1403. [El+2018] Eldabi, T.; Brailsford, S.; Djanatliev, A.; Kunc, M.; Mustafee, N.; Osorio, A. F.: Hybrid Simulation Challenges and Opportunities: A Life-Cycle Approach. In (Rabe, M.; Juan, A. A.; Mustafee, N.; Skoogh, A.; Jain, S.; Johansson, B., eds.): Proceedings of the 2018 Winter Simulation Conference. IEEE, Piscataway, NJ, 2018, pp. 1500–1514. [Er2014] Erlach, K.: Energiewertstrom – Steigerung der Energieeffizienz in der Produktion. In (Neugebauer, R., eds.): Handbuch Ressourcenorientierte Produktion. Carl Hanser Verlag, München, 2014, pp. 41–63. [Es+2011] Eskandari, H.; Mahmoodi, E.; Fallah, H.; Geiger, C. D.: Performance Analysis of Comercial Simulation-Based Optimization Packages: OptQuest and Witness Optimizer. In (S. Jain, R.R. Creasey, J. Himmelspach, K.P. White, and M. Fu, eds.): Proceedings of the 2011 Winter Simulation Conference. IEEE, Piscataway, NJ, 2011, pp. 2363–2373. [EV2014] Eberspächer, P.; Verl, A.: Realizing Energy Reduction of Machine Tools Through a Control-integrated Consumption Graph-based Optimization Method. In (Cunha, Pedro Filipe do Carmo, eds.): Economic development and wealth through globally competitive manufacturing systems. Curran, Red Hook, NY, 2014, pp. 640–645. [Fa+2014] Fakhimi, M.; Anagnostou, A.; Stergioulas, L.; Taylor, S.: A Hybrid AgendBased and Discrete Event Simulation Approach for Sustainable Strategic Plabbubg and Simulation Analytics. In (Tolk, A.; Diallo, D.; Ryzhov, I. O.; Yilmaz, L.; Buckley, S.; Miller, J. A. eds.): Proceedings of the 2014 Winter Simulation Conference. IEEE, Piscataway, NJ, 2014, pp. 1573–1584. [FGA2005] Fu, M. C.; Glover, F. W.; April, J.: Simulation Optimization: A Review, New Developements, and Applications. In (Kuhl, M. E.; Steiger, N. M.; Armstrong, F. B.; Joines, J. A., eds.): Proceedings of the 2005 Winter Simulation Conference. IEEE, Piscataway, NJ, 2005, pp. 83–95. [Fo1961] Forrester, J. W.: Industrial Dynamics. Productivity Press, Wiley, Cambridge, 1961. [Fo1968] Forrester, J. W.: Industrial Dynamics. After the First Decade. In (INFORMS – The Institute for Operations Research and the Management Sciences Hrsg), Maryland, USA, 1968, pp. 398–415. [Fu2013] Fu, M. C.: Simulation Optimization. In (Gass, S. I.; Fu, M., eds.): Encyclopedia of Operations Research and Management Science. Springer Science + Business Media, New York, NY, 2013, pp. 1418–1423.

212

References

[Fu2015] Fu, M. C.: Handbook of simulation optimization. Springer New York, New York, NY, 2015. [Fu+2014] Fu, M. C.; Bayraksan, G.; Henderson, S. G.; Nelson, B. L.; Powell, Warren, B.; Ryzhov, I. O.: Simulation Optimization: A Panel on the State of the Art in Research and Practice. In (Tolk, A.; Diallo, D.; Ryzhov, I. O.; Yilmaz, L.; Buckley, S.; Miller, J. A., eds.): Proceedings of the 2014 Winter Simulation Conference. IEEE, Piscataway, NJ, 2014, pp. 3696–3706. [Fy+2013] Fysikopoulos, A.; Papacharalampopoulos, A.; Pastras, G.; Stavropoulos, P.; Chryssolouris, G.: Energy Efficiency of Manufacturing Processes: A Critical Review. In (Cunha, P. F., eds.): Forty Sixth CIRP Conference on Manufacturing Systems 2013. Elsevier B.V.2013, pp. 628–633. [Ga+2018] Garwood, T. L.; Hughes, B. R.; Oates, M. R.; O’Connor, D.; Hughes, R.: A review of energy simulation tools for the manufacturing sector. In: Renewable and Sustainable Energy Reviews. 812018, pp. 895–911. [GB2018] Grimme, C.; Bossek, J.: Einführung in die Optimierung: Konzepte, Methoden und Anwendungen. Springer Vieweg, Wiesbaden, 2018. [GDT2006] Gutowski, T.; Dahmus, J.; Thiriez, A.: Electrical Energy Requirements for Manufacturing Processes. In (Duflou, J. R., eds.): Towards a closed loop economy. Katholieke Univ. Leuven, Leuven, 2006, pp. 623–627 [Ge+2015] Geilhausen, M.; Bränzel, J.; Engelmann, D.; Schulze, O.: Energiemanagement: Für Fachkräfte, Beauftrage und Manager. Springer Fachmedien Wiesbaden, Wiesbaden, 2015. [GLM2000] Glover, F.; Laguna, M.; Martí, R.: Fundamentals of Scatter Search and Path Relinking. In: Control and Cybernetics Journal (Vol. 29, No. 3), 2000, pp. 653– 684. [Go2015] Gosavi, A.: Simulation-based optimization: parametric optimization techniques and reinforcement learning. Springer US, Boston, MA, 2015. [Gr+2015] Grave, K.; Hazrat, M.; Boeve, S.; Blücher, F. von; Bourgault, C.; Breitschopf, B.; Friedrichsen, N.; Arens, M.; Aydemir, A.; Pudlik, M.; Duscha, V.; Ordonez, J.; Lutz, C.; Großmann, A.; Flaute, M.: Electrical Costs of Energy Intensive Industries, https://www.isi.fraunhofer.de/isi-wAssets/docs/x/de/publs-mit arbeiterseiten/Electricity-Costs-of-Energy-Intensive-Industries.pdf. [Ha2013] Haag, H.: Eine Methodik zur modellbasierten Planung und Bewertung der Energieeffizienz in der Produktion. Fraunhofer Verlag, Stuttgart, 2013. [HC2010] Hevner, A.; Chatterjee, S.: Design Research in Information Systems: Theory and Practice. Springer Science+Business Media LLC, Boston, MA, 2010. [He2006] Henriksen, J. O.: Taming the Complexity Dragon. In (Perrone, L. F.; Wieland, F. P.; Liu, B.; Lawson, B. G.; Nicol, D. M.; Fujimoto, R. M., eds.): Proceedings of the 2006 Winter Simulation Conference. IEEE, Piscataway, NJ, 2006, pp. 1–20. [He2008] Helal, M.: A Hybrid System Dynamics-discrete Event Simulation approach To Simulating The Manufacturing Enterprise, Florida, 2008, dissertation. [He2012] Hesselbach, J.: Energie- und klimaeffiziente Produktion: Grundlagen, Leitlinien und Praxisbeispiele. Vieweg+Teubner (GWV), 2012. [He+2008] Hesselbach, J.; Herrmann, C.; Detzer, R.; Martin, L.; Thiede, S.; Lüdemann, B.: Energy Efficiency through optimized coordination of production and technical building services. In (Kaebernick, H., eds.): Applying life cycle knowledge

References

[He+2010]

[He+2011]

[He+2013]

[He+2018] [HGM2013] [HMW2012]

[HS2008]

[HV2000] [In2017a]

[In2017b] [JH2015]

[Jo+2012]

[JWS2019]

213 to engineering solutions: Conference Proceedings: 15th CIRP International Conference on Life Cycle Engineering: Applying Life Cycle Knowledge to Engineering Solutions, Sydney, 2008, pp. 624–628. He, Y.; Liu, F.; Wu, T.; Zhong, F.-P.; Peng, B.: Analysis and estimation of energy consumption for numerical control machining. In: Proceedings of the Institution of Mechanical Engineers, Part B: Journal of Engineering Manufacture. 226(2), 2010, pp. 255–266. Heath, S. K.; Brailsford, S. C.; Buss, A.; Macal, C. M.: Cross-Paradigm Simulation Modeling: Challenges and Success. In (S. Jain, R.R. Creasey, J. Himmelspach, K.P. White, and M. Fu, eds.): Proceedings of the 2011 Winter Simulation Conference. IEEE, Piscataway, NJ, 2011, pp. 2788–2802. Heinzl, B.; Rossler, M.; Popper, N.; Leobner, I.; Ponweiser, K.; Kastner, W.; Dur, F.; Bleicher, F.; Breitenecker, F.: Interdisciplinary Strategies for Simulation-Based Optimization of Energy Efficiency in Production Facilities. In (Al-Dabass, D.; Orsoni, A. Y. J.; Cand, R.; Ibrahim, Z., eds.): 2013 UKSim 15th International Conference on Computer Modelling and Simulation (UKSim 2013). Cambridge, pp. 304–309. Hering, E.; Martin, R.; Gutekunst, J.; Kempkes, J.: Elektrotechnik und Elektronik für Maschinenbauer. Springer Vieweg, Berlin, 2018. Hering, E.; Gutekunst, J.; Martin, R.: Elektrotechnik für Maschinenbauer: Grundlagen. Springer Berlin Heidelberg, Berlin, Heidelberg, 2013. Hoffmann, K.; Mathis, W.; Wiesemann, K.: Elektronik. In (Czichos, H.; Hennecke, M.; eds.): Hütte – Das Ingenieurwissen. Springer, Berlin, Heidelberg, 2012, pp. G1–G188. Herzwurm, G.; Stelzer, D.: Wirtschaftsinformatik versus Information Systems – eine Gegenüberstellung. In (Bankhofer, U.; Nissen, V.; Stelzer, D.; and Straßburger, S.; eds): Ilmenauer Beiträge zur Wirtschaftsinformatik, Ilmenau, 2008. Hindesberger, M.; Valqui Vidal, R. V.: Tabu Search – A Guided Tour. In: Control and Cybernetics Journal. 2000 (Vol. 29, No. 3), 2000, pp. 633–651. International Energy Agency: World Energy Outlook 2017 – Executive Summary, website accessed on March 22nd , 2018, https://www.iea.org/Textbase/ npsum/weo2017SUM.pdf. International Energy Agency: World Energy Outlook – Homepage Article. website accessed on March 22nd , 2018, https://www.iea.org/weo2017/. Jian, N.; Henderson, S. G.: An Introduction to Simulation Optimization. In (Yilmaz, L.; Chan, W. K.; Moon, I.; Roeder, T. M.; Macal, C. M.; Rossetti, M. D., eds.): Proceedings of the 2015 Winter Simulation Conference. IEEE, Piscataway, NJ, 2015, pp. 1780–1794. Johansson, T. B.; Naki´cenovi´c, N.; Gomez-Echeverri, L.; Patwardhan, A. P.: Global Energy Assessment – Toward a Sustainable Future. Cambridge University Press, Cambridge, 2012. Johannes, C.; Wichmann, M. G.; Spengler, T. S.: Energy-oriented production planning with time-dependent energy prices. In: Procedia CIRP. 802019, pp. 245–250.

214

References

[Ka1993] Karayanakis, N. M.: Computer-assisted simulation of dynamic systems with block diagram languages. CRC Press, Boca Raton, 1993. [Kh+2015] Khedri Liraviasl, K.; ElMaraghy, H.; Hanafy, M.; Samy, S. N.: A Framework for Modelling Reconfigurable Manufacturing Systems Using Hybridized Discrete-Event and Agent-based Simulation. In: IFAC-PapersOnLine. 48(3), 2015, pp. 1490–1495. [KS2011] Kletti, J.; Schumacher, J.: Die perfekte Produktion: Manufacturing Excellence durch Short Interval Technology (SIT). Springer-Verlag Berlin Heidelberg, Berlin, Heidelberg, 2011. [KS2015] Krischer, K.; Schönleber, K.: Physics of Energy Conversion. De Gruyter, Berlin/Boston, 2015. [KSZ2002] Kelton, W. D.; Sadowski, R. P.; Zupick, N. B.: Simulation with Arena. McGrawHill Education, New York, NY, 2002. [KW2000] Klinger, A.; Wenzel, S.: Referenzmodelle – Begriffsbestimmung und Klassifikation. In (Wenzel, S., eds.): Referenzmodelle für die Simulation in Produktion und Logistik. Society for Computer Simulation International, Ghent, 2000, pp. 13–29. [La2000] Lane, D. C.: You just don’t understand me: modes of failure and success in the discourse between system dynamics and discrete event simulation. LSE OR Working Paper 00.34, London, 2000. [La2003] Law, A. M.: How to conduct a simulation study. In (Chick, S.; Sanchez, P.; Ferrin, D.; Morrice, D., eds.): Proceedings of the 2003 Winter Simulation Conference. Association for Computing Machinery, New York, N.Y, Piscataway, N.J, 2003, pp. 66–70. [La2007] Law, A. M.: Simulation modeling and analysis. McGraw-Hill, Boston, Mass., 2007. [La2011] Laguna, M.: OptQuest: Optimization of Complex Systems, website accessed on July 12th, 2019. URL: https://www.opttek.com/sites/default/files/pdfs/Opt Quest-Optimization%20of%20Complex%20Systems.pdf. [La2019] Lanner Group, Inc.: WITNESS. website accessed on July 15th, 2019. URL: https://www.lanner.com. [LHF2012] Lorenz, S.; Hesse, M.; Fischer, A.: Simulation and optimization of robot driven production systems for peak-load reduction. In (C. Laroque, J. Himmelspach, R. Pasupathy, O. Rose, and A. M. Uhrmacher, eds.): Proceedings of the 2012 Winter Simulation Conference. IEEE, Piscataway, NJ, 2012, pp. 2875–2886. [Li1992] Littger, K.: Optimierung: Eine Einführung in rechnergestützte Methoden. Springer Berlin Heidelberg, Berlin, Heidelberg, 1992. [Li+2015] Liraviasl, K.; ElMaraghy, H.; Hanafy, M.; Samy, S.: A Framework for Modelling Reconfigurable Manufacturing Systems Using Hybridized Discrete-Event and Agent-based Simulation. In Proceedings of the 15th IFAC Symposium on Information Control Problems in Manufacturing. 2015. [LK2000] Law, A. M.; Kelton, W. D.: Simulation modeling and analysis. McGraw-Hill, Boston, 2000. [LM2013] Laguna, M.; Marti, R.: Heuristics. In (Gass, S. I.; Fu, M., eds.): Encyclopedia of Operations Research and Management Science. Springer Science + Business Media, New York, NY, 2013, pp. 695–703.

References

215

[Lu2002] Lunze, J.: What is a Hybrid System? In (Engell, S.; Frehse, G.; Schnieder, E., eds.): Modelling, Analysis, and Design of Hybrid Systems. Springer, Berlin, 2002, pp. 3–14. [Ma2019] MathWorks: Stateflow – Overview. Website accessed November 19th , 2019. URL: https://mathworks.com/products/stateflow.html. [MA2019] Mykoniatis, K.; Angelopoulou, A.: A modeling framework for the application of multi-paradigm simulation methods. In: SIMULATION. 2019. [MB1997] Mosterman, P. J.; Biswas, G.: Formal Specifications for Hybrid Dynamical Systems. In (Pollack, M. E., ed.): Proceedings of the 15th International Joint Conference on Artificial Intelligence (IJCAI’97), 1997, pp. 568–577. [MC2007] Miller, J.A., He, C.; Couto, J. I.: Impact of the Semantic Web on Modeling and Simulation. In (Fishwick, P. A., eds.): Handbook of dynamic system modeling. Chapman & Hall/CRC, Boca Raton, 2007, pp. 3.1–3.22. [MCP2018] Martí, R.; Corberán, Á.; Peiró, J.: Scatter Search. In (Martí, R.; Pardalos, P. M.; Resende, M. G. C. Eds.): Handbook of Heuristics. Springer International Publishing, Cham, 2018, pp. 717–740. [MF2004] Michalewicz, Z.; Fogel, D. B.: How to solve it: Modern heuristics. Springer, Berlin, 2004. [Mi2002] Milling, P.: Understanding and Managing Innovation Processes. In: System Dynamics Review, 2003, Vol. 18, No. 1, pp. 73–86. [MK2011] März, L.; Krug, W.: Kopplung von Simulation und Optimierung. In (März, L.; Krug, W.; Rose, O.; Weigert, G., eds.): Simulation und Optimierung in Produktion und Logistik. Springer Berlin Heidelberg, 2011, pp. 41–45. [MN2014] Macal, C. M.; North, M. J.: Tutorial on agent-based modeling and simulation. In (Taylor, S., eds.): Agent-based Modeling and Simulation. Palgrave Macmillan, Basingstoke, 2014, pp. 11–31. [Mo1999] Mosterman P.J.: An Overview of Hybrid Simulation Phenomena and Their Support by Simulation Packages. In (Vaandrager, F.W.; van Schuppen, J.H., eds.): Hybrid Systems: Computation and Control. HSCC 1999. Lecture Notes in Computer Science, vol 1569. Springer, Berlin, Heidelberg, 1999, pp. 165– 177. [MP2018] Mustafee, N.; Powell, J. H.: From Hybrid Simulation to Hybrid System Modelling. In (Rabe, M.; Juan, A. A.; Mustafee, N.; Skoogh, A.; Jain, S.; Johansson, B., eds.): Proceedings of the 2018 Winter Simulation Conference. IEEE, Piscataway, NJ, 2017, pp. 1430–1439. [MPR2018] Martí, R.; Pardalos, P. M.; Resende, M. G. C.: Handbook of Heuristics. Springer International Publishing, Cham, 2018. [Mu+2017] Mustafee, N.; Brailsford, S.; Djanatliev, A.; Eldabi, T.; Kunc, M.; Tolk, A.: Purpose and Benefits of Hybrid Simulation: Contributing to the Convergence of its Definitions. In (Chan, W. K.; D’Ambrogio, A.; Zacharewicz, G.; Mustafee, N.; Wainer, G.; Page, E. H., eds.): Proceedings of the 2017 Winter Simulation Conference. IEEE, Piscataway, NJ, 2017, pp. 1631–1645. [Mü+2009] Müller, E.; Engelmann, J.; Löffler, T.; Strauch, J.: Energieeffiziente Fabriken planen und betreiben. Springer Berlin Heidelberg, 2009. [MVG2010] Meschede, D.; Vogel, H.; Gerthsen, C.: Gerthsen Physik. Springer, Berlin, 2010.

216

References

[MW2011] März, L.; Weigert, G.: Simulationsgestützte Optimierung. In (März, L.; Krug, W.; Rose, O.; Weigert, G., eds.): Simulation und Optimierung in Produktion und Logistik. Springer Berlin Heidelberg, 2011, pp. 3–12. [My2006] Myers, R. L.: The basics of physics. Greenwood Press, Westport, Conn, 2006. [Ne+2008] Neugebauer, R.; Westkämper, E.; Klocke, F.; Spath, D.; Schenk, M.; Michaelis, A.; Hompel, M. t.; Weidner, E.: Energieeffizienz in der Produktion: Untersuchungen zum Handlungs- und Forschungsbedarf. Fraunhofer Gesellschaft, München, 2008. [Ng+2017] Nguyen, V.; Novak, A.; Shokr, M.; Pash, K.: Aircrew Manpower Supply Modeling under Change: An Agent-Based Discrete Event Simulation Approach. In (Chan, W. K.; D’Ambrogio, A.; Zacharewicz, G.; Mustafee, N.; Wainer, G.; Page, E. H., eds.): Proceedings of the 2017 Winter Simulation Conference. IEEE, Piscataway, NJ, 2017, pp. 4070–4080. [NM2005] Nedjah, N.; Macedo Mourelle, L. d.: Evolutionary Multi-Objective Optimisation: A Review. In (Nedjah, N.; Macedo Mourelle, L. d., eds.): Real-world multi-objective system engineering. Nova Science Publ, New York, 2005, pp. 1–28. [NM2007] North, M. J.; Macal, C. M.: Managing business complexity: Discovering strategic solutions with agent-based modeling and simulation. Oxford University Press, Oxford, New York, 2007. [NS2012] Niedrig, H.; Sternberg, M.: Physik. In (Czichos, H.; Hennecke, M., eds.): Hütte – Das Ingenieurwissen. Springer, Berlin, Heidelberg, 2012, S. B1–B289. [NW2012] Nyhuis, P.; Wiendahl, H.-P.: Logistische Kennlinien: Grundlagen, Werkzeuge und Anwendungen. Springer, Berlin, Heidelberg, 2012. [Op2013] Optimization. In (Gass, S. I.; Fu, M., eds.): Encyclopedia of Operations Research and Management Science. Springer Science + Business Media, New York, NY, 2013, p. 1092. [Op2019] OptTek Systems, Inc.: OptQuest. website accessed on July 15th, 2019. URL: https://www.opttek.com. [Ör2009] Ören, T. I.: Modeling and Simulation: A Comprehensive and Integrative View. In (Yilmaz, L.; Ören, T. I., eds.): Agent-directed simulation and systems engineering. Wiley-VCH, Weinheim, 2009, pp. 3–36. [Pa1991] Page, B.: Diskrete Simulation. Eine Einführung mit Modula – 2. SpringerVerlag, Berlin/Heidelberg, 1991. [Pa2019] Palisade Corp.: RISKOptimizer. website accessed on July 15th, 2019. URL: https://www.palisade.com. [Pe2010a] Pedgen, D. C.: Advanced Tutorial: Overview of Simulation World Views. In (Johansson, B.; Jain, S.; Montoya-Torres, J.; Hugan, J.; Yücesan, E. Hrsg.): Proceedings of the 2010 Winter Simulation Conference. IEEE, Piscataway, NJ, 2017, pp. 210–215. [Pe2010b] Pehnt, M.: Energieeffizienz – Definitionen, Indikatoren, Wirkungen. In (Pehnt, M., eds.): Energieeffizienz. Springer, Berlin, 2010, pp. 1–34. [Pe2017] Pedgen, D. C.: The Evolution of Simulation Languages. In (Tolk, A.; Fowler, J.; Shao, G.; Yucesan, E., eds.): Advances in modeling and simulation. Springer, Cham, 2017, pp. 81–96.

References

217

[Pe+2017] Peter, T.; Wenzel, S.; Reiche, L.; Fehlbier, M.: Coupled Simulation of Energy and Material Flow – A Use Case in an Aluminium Foundry. In (Chan, W. K.; D’Ambrogio, A.; Zacharewicz, G.; Mustafee, N.; Wainer, G.; Page, E. H., eds.): Proceedings of the 2017 Winter Simulation Conference. pp. 3792–3803. [PH2010] Paulun, T.; Haubrich H.-J.: Long-term and Expansion Planning for Electrical Networks Considering Uncertainties. In (Rebennack, S., eds.): Handbook of power systems I. Springer-Verlag Berlin Heidelberg, Berlin, Heidelberg, 2010, pp. 391–408. [PO1999] Pritsker, A. A. B.; O’Reilly, J. J.: Simulation with Visual SLAM and AweSim. Wiley, New York, 1999. [Po2011] Posch, W.: Ganzheitliches Energiemanagement für Industriebetriebe. Gabler Verlag /Springer Fachmedien Wiesbaden GmbH Wiesbaden, Wiesbaden, 2011. [Pr1995] Pritsker, A. A. B.: Introduction to simulation and SLAM II. Halsted Pr, New York, 1995. [Pr1998] Pritsker, A. B.: Principles of Simulation Modeling. In (Banks, J., eds.): Handbook of Simulation. Wiley; Co-published by Engineering & Management Press, New York, 1998, pp. 31–51. [PR2002] Pardalos, P. M.; Resende, M. G. C.: Handbook of applied optimization. Oxford Univ. Press, Oxford, 2002. [Pr2019] ProModel Corp.: ProModel. website accessed on July 15th, 2019. URL: https:// www.promodel.com. [PSJ2017] Pawletta, T.; Schmidt, A.; Junglas, P.: A Multimodeling Approach for the Simulation of Energy Consumption in Manufacturing. In: SNE Simulation Notes Europe. 27(2), 2017, pp. 115–124. [PW2015] Peter, T.; Wenzel, S.: Simulationsgestützte Planung und Bewertung der Energieeffizienz für Produktionssysteme in der Automobilindustrie. In (Rabe, M.; Clausen, U., eds.): Simulation in production und logistics 2015. Fraunhofer Verl., Stuttgart, 2015, pp. 535–544. [Ra2008] Rager, M.: Energieorientierte Produktionsplanung: Analyse, Konzeption und Umsetzung. Betriebswirtschaftlicher Verlag Dr. Th. Gabler | GWV Fachverlage GmbH, Wiesbaden, 2008. [Ro1998] Rohrer, M. W.: Simulation of Manufacturing and Material Handling Systems. In (Banks, J., eds.): Handbook of Simulation. Wiley; Co-published by Engineering & Management Press, New York, S.l., 1998, pp. 519–545. [Ro2004] Robinson, S.: Simulation: The practice of model development and use. John Wiley & Sons Ltd, Chichester, West Sussex, England, Hoboken, NJ, 2004. [Ro2008] Robinson, S.: Conceptual modelling for simulation Part I. In: Journal of the Operational Research Society. 59(3), 2008, pp. 278–290. [Ro2010] Robinson, S.: Conceptual Modeling for Simulation: Definition and Requirements. In (Robinson, S.; Brooks, R.; Kotiadis, K.; van der Zee, D., eds.): Conceptual modeling for discrete-event simulation. CRC Press, Boca Raton, Fla., 2010, pp. 3–30. [RRS2018] Römer, A. C.; Rückbrod, M.; Straßburger, S.: Eignung kombinierter Simulation zur Darstellung energetischer Aspekte in der Produktionssimulation. In (Deatcu, C.; Schramm, T.; Zobel, K., eds.): ASIM 2018 – 24. Symposium Simulationstechnik: Tagungsband. ARGESIM/ASIM, Wien, 2018, pp. 73–80.

218

References

[RS2011] Reggelin, T.; Schenk, M.: Mesoskopische Modellierung und Simulation logistischer Flusssysteme. docupoint, Magdeburg, 2011. [RS2016] Roemer, A. C.; Strassburger, S.: A review of literature on simulation-based optimization of energy efficiency in production. In (Roeder, T. M.; Frazier, P. I.; Szechtman, R.; Zhou, E., eds.): Proceedings of the 2016 Winter Simulation Conference. IEEE, Piscataway, NJ, 2016, pp. 1416–1427. [RS2019] Roemer, A. C.; Strassburger, S.: Hybrid System Modeling Approach for the Depiction of the Energy Consumption in Production Simulation. In (Mustafee, N.; Bae, K.-H. G.; Lazarova-Molnar, S.; Rabe, M.; Szabo, C.; Haas, P.; Son, Y.-J., eds.): Proceedings of the 2019 Winter Simulation Conference. IEEE, Piscataway, NJ, 2019, pp. 1366–1377. [Ru2018] Rueckbrod, M.: Eignung kombinierter Simulation für die Abbildung energetischer Aspekte in der Simulation von Produktionssystemen, Ilmenau, 2018. [RW2008] Rudolph, M.; Wagner, U.: Energieanwendungstechnik: Wege und Techniken zur effizienteren Energienutzung. Springer, Berlin, Heidelberg, 2008. [Sa2000] Saint-Exupéry, A. d.: Wind, sand and stars. Penguin, London, 2000. [Sc2004] Schöneborn, F.: Strategisches Controlling mit System Dynamics. SpringerVerlag GmbH, Berlin, Heidelberg, 2004. [Sc2006] Schieferdecker, B.: Energiemanagement-Tools: Anwendung im Industrieunternehmen. Springer-Verlag, 2006. [Sc2016] Schellong, W.: Analyse und Optimierung von Energieverbundsystemen. Springer Vieweg, Berlin, Heidelberg, 2016. [Sc+2017] Schlüter, W.; Henninger, M.; Buswell, A.; Schmidt, J.: Schwachstellenanalyse und Prozessverbesserung in Nichteisen-Schmelz- und Druckgussbetrieben durch bidirektionale Kopplung eines Materialflussmodells mit einem Energiemodell. In (Wenzel, S.; Peter, T., eds.): Simulation in Produktion und Logistik 2017. 20. – 22. September 2017, Kassel, pp. 19–28. [SG2003] Schieritz, N.; Größler, A.: Emergent Structures in Supply Chains – A Study Integrating Agent-Based and System Dynamics Modeling. In (Sprague, R. H, eds.): Proceedings of the 36th Annual Hawaii International Conference on System Sciences. IEEE Computer Society Press, Los Alamitos, California, 2003, pp. 94–1 ff. [SG2013] Soerensen, K.; Glover, F. W.: Metaheuristics. In (Gass, S. I.; Fu, M., eds.): Encyclopedia of Operations Research and Management Science. Springer Science + Business Media, New York, NY, 2013, pp. 960–968. [Sh1975] Shannon, R. E.: Systems simulation: The art and science. Prentice-Hall, Englewood Cliffs, NJ, 1975. [Sh1998] Shannon, E.: Introduction to the Art and Science of Simulation. In (1998 Winter Simulation Conference. IEEE, Piscataway, N.J, New York, N.Y, San Diego, Calif, 1998, pp. 7–14. [Sh+2018] Shaheen, A. M.; Spea, S. R.; Farrag, S. M.; Abido, M. A.: A review of metaheuristic algorithms for reactive power planning problem. In: Ain Shams Engineering Journal. 9(2), 2018, pp. 215–231. [Si2016] Siarry, P.: Metaheuristics. Springer, Cham, 2016. [Si+2018] Siegel, A.; Turek, K.; Michelini, E.; Schmidt, T.: Hybrid modeling approach for prediction of energy demand and power peaks in intralogistic systems. In

References

[SKS2017]

[SM2003]

[SM2009] [So2018]

[SP2005]

[SP2014]

[SSP2013]

[St2000] [St2014]

[St2017]

[St+2006]

[Sw2015]

219 (Deatcu, C.; Schramm, T.; Zobel, K., eds.): ASIM 2018 – 24. Symposium Simulationstechnik: Tagungsband. ARGESIM/ASIM, Wien, 2018, pp. 81–88. Sobottka, T.; Kamhuber, F.; Sihn, W.: Increasing Energy Efficiency in Production Environments through an Optimized, Hybrid Simulation-based Planning of Production and its Periphery. In: Procedia CIRP. (61)2017, pp. 440–445. Schieritz, N.; Milling, M.: Modeling the Forest or Modeling the Threes – A Comparison of System Dynamics and Agent-Based Simulation. In (Eberlein, R. L. eds): Proceedings of the 21st International Conference of the System Dynamics Society. System Dynamics Society, Albany, NY, 2003, pp. 1–15. Suhl, L.; Mellouli, T.: Optimierungssysteme: Modelle, Verfahren, Software, Anwendungen. Springer, Dordrecht, 2009. Sobottka, T.: Eine anwendungsorientierte simulationsbasierte Methode, unter Berücksichtigung von Energieeffizienz in der optimierenden Planung von Produktion und Logistik, Wien, 2018, dissertation. Solding, P.; Petku, D.: Applying Energy Aspects on Simulation of Energy Intensive Production Systems. In (M. E. Kuhl, N. M. Steiger, F. B. Armstrong, and J. A. Joines, eds.): Proceedings of the 2005 Winter Simulation Conference. IEEE, Piscataway, NJ, 2005, pp. 1428–1432. Schmidt, A.; Pawletta, T.: Hybride Modellierung fertigungstechnischer Prozessketten mit Energieaspekten in einer ereignisorientierten Simulationsumgebung. In: Proc. ASIM 2014 – 22. Symposium Simulationstechnik, Berlin, 03./05.09.2014, ARGESIM Report 43, ASIM Mitteilung AM 151, ARGESIM/ASIM Pub. Vienna, Austria, 2014, pp. 109–116. Schlegel, A.; Stoldt, J.; Putz, M.: Erweiterte Integration energetischer Betrachtungen in der Materialflusssimulation. In (Dangelmaier, W.; Laroque, C.; Klaas, A., eds.): Simulation in Produktion und Logistik 2013: Advanced integration of energetic consideration in discrete event simulation. HeinzNixdorf-Institut Universität Paderborn, Paderborn, 2013, pp. 187–196. Sterman, J. D.: Business dynamics: Systems thinking and modeling for a complex world. Irwin/McGraw-Hill, Boston, 2000. Statistisches Bundesamt (Destatis): Erzeugerpreise für Strom seit Januar 2000. Press release commenting on the development of energy prices from 2000 to 2014, https://www.destatis.de/DE/PresseService, press release of October 9th , 2014 – 354/14. Statistisches Bundesamt (Destatis): Daten zur Energiepreisentwicklung. Website assessed on 10.01.2018, https://www.destatis.de/DE/Publikationen/The matisch/Preise/Energiepreise/EnergyPriceTrendsPDF_5619002.pdf Straßburger, S.; Seidel, H.; Schady, R.; Masik, S.: Werkzeuge und Trends der Fabrikplanung – Analyse der Ergebnisse einer Onlinebefragung. In (Wenzel, S. eds.): Simulation in Produktion und Logistik 2006. SCS Publishing House, Erlangen, 2006, pp. 391–402. Swat, M.: Methode zur Planung und Gestaltung energieeffizienter Prozessketten für die Serienfertigung am Beispiel ausgewählter Feinbearbeitungsverfahren. Saarbrücken, 2015, dissertation.

220

References

[SWM2014] Schenk, M.; Wirth, S.; Müller, E.: Fabrikplanung und Fabrikbetrieb: Methoden für die wandlungsfähige, vernetzte und ressourceneffiziente Fabrik. Springer Vieweg, Berlin, 2014. [Ta2014] Taylor, S.: Introducing agent-based modeling and simulation. In (Taylor, S., eds.): Agent-based Modeling and Simulation. Palgrave Macmillan, Basingstoke, 2014, pp. 1–10. [TB2017] Trivedi, K. S.; Bobbio, A.: Reliability and availability engineering: Modeling, analysis, and applications. Cambridge University Press, Cambridge, 2017. [Te+2018] Teiwes, H.; Blume, S.; Herrmann, C.; Rössinger, M.; Thiede, S.: Energy Load Profile Analysis on Machine Level. In: Procedia CIRP 69, 2018, pp. 271–276. [Th2019a] The MathWorks, Inc.: Optimization Toolbox. website accessed on July 16th, 2019. URL: https://de.mathworks.com/help/optim/index.html [Th2019b] The MathWorks, Inc.: Getting Started with Global Optimization Toolbox. website accessed on July 16th, 2019. URL: https://de.mathworks.com/help/gads/ getting-started-with-global-optimization-toolbox.html?s_tid=CRUX_lftnav. [Th2012] Thiede, S.: Energy Efficiency in Manufacturing Systems. Springer-Verlag, Berlin Heidelberg, 2012, dissertation. [Ti2001] Tiller, M.: Introduction to Physical Modeling with Modelica. Springer US, Boston, MA, 2001. [Tu2019] Turk, I.: Practical MATLAB: With Modeling, Simulation, and Processing Projects. Springer Science+Business Media, New York, 2019. [TRP2013] Tuy, H.; Rebennack, S.; Pardalos, P. M.: Global Optimization. In (Gass, S. I.; Fu, M., eds.): Encyclopedia of Operations Research and Management Science. Springer Science + Business Media, New York, NY, 2013, pp. 650–658. [Vö+2003] Völker, S.; Gmilkowsky, P.; Biethahn, J.; Böselt, M.: Reduktion von Simulationsmodellen zur simulationsbasierten Optimierung in der Termin- und Kapazitätsplanung. Lang, Frankfurt am Main, 2003. [VS2010] Völker, S.; Schmidt, P.-M.: Simulationsbasierte Optimierung von Produktionsund Logistiksystemen mit Tecnomatix Plant Simulation. In (Zülch, G.; Stock, P., eds.): Integrationsaspekte der Simulation: Technik, Organisation und Personal. KIT Scientific Publ, Karlsruhe, 2010, pp. 93–100. [Wa2009] Wainer, G. A.: Discrete-event modeling and simulation: A practitioner’s approach. CRC Press, Boca Raton, 2009. [Wa+2012] Wang, Z.; Gao, F.; Zhai, Q.; Guan, X.; Liu, K.; Zhou, D.: An Integrated Optimization Model for Generation and Batch Production Load Scheduling in Energy Intensive Enterprise. IEEE Power and Energy Society general meeting, 2012. [We2000] Wenzel, S.: Referenzmodelle für die Simulation in Produktion und Logistik. Society for Computer Simulation International, Ghent, 2000. [We2010] Weinert, N.: Vorgehensweise für Planung und Betrieb energieeffizienter Produktionssysteme. Fraunhofer-Verlag, Stuttgart, 2010, dissertation. [WH2006] Wilde, T.; Hess, T.: Methodenspektrum der Wirtschaftsinformatik, München, 2006. [WH2007] Wilde, T.; Hess, T.: Forschungsmethoden der Wirtschaftsinformatik. In: WIRTSCHAFTSINFORMATIK. 49(4), 2007, pp. 280–287. [Wi1998] Wieting, R.: Modellbildung und Simulation mit hybriden höheren Netzen. Shaker, Aachen, 1998.

References

221

[WKD2012] Wolff, D.; Kulus, D.; Dreher, S.: Simulating Energy Consumption in Automotive Industries. In (Bangsow, S., eds.): Use cases of discrete event simulation. Springer, Berlin, 2012, pp. 59–86. [Wo2009] Wooldridge, M. J.: An introduction to multiagent systems. Wiley, Chichester, 2009. [Wö+2019] Wörrlein, B.; Bergmann, S.; Feldkamp, N.; Straßburger, S.: Deep-Learningbasierte Prognose von Stromverbrauch für die hybride Simulation. In (Putz, M.; Schlegel, A., eds.): ASIM 2019: Deep Learning based Prediction of Energy Consumption for Hybrid Simulation. Wissenschaftliche Scripten, Auerbach /Vogtl., 2019, pp. 121–131. [WW2002] Webster, J.; Watson, R. T.: Analyzing the Past to Prepare for the Future: Writing A Literature Review. In: MIS Quarterly. 2002, pp. xiii–xxiii. [Ya2018] Yang, X.-S.: Engineering optimization: An introduction with metaheuristic applications. Wiley, Hoboken, NJ, 2018. [YK2011] Yang, X.-S.; Koziel, S.: Computational Optimization: An Overview. In (Koziel, S.; Yang, X.-S., eds.): Computational Optimization, Methods and Algorithms. Springer Berlin Heidelberg, Berlin, Heidelberg, 2011. [Zä2001] Zäpfel, G.: Grundzüge des Produktions- und Logistikmanagement. Oldenbourg, München, 2001. [ZO1986] Zeigler, B. P., Oeren, T. I.: 1986. Multifacetted, Multiparadigm Modelling Perspectives: Tools for the 90’s. In (Wilson, J.; Henriksen, J.; Roberts, S., eds.): Proceedings of the 1986 Winter Simulation Conference. IEEE, Piscataway, NJ, 1986, pp. 708–712. [ZPK2010] Zeigler, B. P.; Praehofer, H.; Kim, T. G.: Theory of modeling and simulation: Integrating discrete event and continuous complex dynamic systems. Academic Press, Amsterdam, 2010.

List of Standards [DIN2002] DIN 40110-2:2002-11: Quantities used in alternating current theory – Part 2: Multi-line circuits. Beuth Verlag, Berlin, 2002. [VD2007] VDI Verein Deutscher Ingenieure e. V.: VDI 4602:2007-10: Energy management. Terms and definitions. VDI-Richtlinie 4602. Beuth Verlag, Berlin, 2007. [VD2014a] VDI Verein Deutscher Ingenieure e. V.: VDI 3633:2014-12: Simulation of systems in materials handling, logistics and production. Fundamentals. VDI-Richtlinie 3633. Beuth Verlag, Berlin, 2014. [VD2014b] VDI Verein Deutscher Ingenieure e. V.: VDI 4661:2014-08: Energetic characteristics: Fundamentals – methodology. VDI-Richtlinie 4661. Beuth Verlag, Berlin, 2014. [VD2016a] VDI Verein Deutscher Ingenieure e. V.: VDI 3633 Blatt 12: Simulation von Logistik-, Materialfluss- und Produktionssystemen. Simulation und Optimierung. VDI-Richtlinie 3633 (Entwurf). Beuth Verlag, Berlin, 2016.