Alexander Hübl (Auth.) - Stochastic Modelling in Production Planning - Methods For Improvement and Investigations On Production System Behaviour-Gabler Verlag (2018) PDF

Alexander Hübl
Stochastic Modelling
in Production Planning
Methods for Improvement
and Investigations on Production
System Behaviour
Alexander Hübl
Methods for Improvement
and Investigations on Production
System Behaviour
Alexander Hübl
Steyr, Austria
Dissertation University of Vienna, Austria, 2015
ISBN 978-3-658-19119-1 ISBN 978-3-658-19120-7 (eBook)

DOI 10.1007/978-3-658-19120-7
Library of Congress Control Number: 2017949670
Springer Gabler
© Springer Fachmedien Wiesbaden GmbH 2018
This work is subject to copyright. All rights are reserved 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, express 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.
Printed on acid-free paper
This Springer Gabler imprint is published by Springer Nature

The registered company is Springer Fachmedien Wiesbaden GmbH
The registered company address is: Abraham-Lincoln-Str. 46, 65189 Wiesbaden, Germany
Preface
All the work presented henceforth was conducted during my time as Re-
search Assistant at the School of Management at the University of Applied
Sciences Upper Austria. Additionally, the PhD Management Programme
of the faculty of business, economics and statistics of the University of Vi-
enna has been completed. The following funded projects have enabled the
preparation of the thesis:
E-Conwip: Embedded Conwip, Austrian Research Promotion Agency
(FFG), Project number: 814303/13129, Investigates the production plan-
ning method Conwip in an MRPII setting by applying in-depth case studies
in three companies.
HPP: Hierarchical Production Planning, Austrian Science Fund (FWF),
Translational research project, Project number: L534-N13, Investigation of
robust planning strategies in a hierarchical production planning setting.
SimGen: Simulationgenerator, FFG Coin Aufbau, Project number:
826789, Development of a framework for generating discrete event simula-
tion models; decision support for mid-term capacity planning; identifying
the effects of planning strategies and their parameters on real world pro-
duction systems.
ProdNET: Atmende Produktion BAY-AUT, Interreg Bayern – Österre-
ich, Project number: J00317, Identification and promotion of production
potential for small and mid-sized enterprises.
HOPL: Heuristic Optimization in Production and Logistics, FFG
COMET, Project number: 843532, The aim of this project is to develop
innovative optimisation algorithms based on interacting sub-processes to
build the holistic model.
BioBoost: Biomass based energy intermediates boosting biofuel produc-
tion, FP7, Project number: 282873, The overall objective of BioBoost is to
pave the way for de-central conversion of residual biomass to optimised, high
energy density carriers, which can be utilised in large scale applications for
the synthesis of transportation fuel and chemicals or directly in small-scale
combined heat and power plants.
The PhD-thesis has been written under supervision by Prof. Richard
F. Hartl (University of Vienna) and evaluated by Prof. Heinrich Kuhn
(Catholic University of Eichstätt-Ingolstadt) and Prof. Lars Mönch (Uni-
versity of Hagen). The thesis consists of five journal articles, two of which
have already been published and three are working papers:
Chapter 3 : Hübl and Jodlbauer (2013): Optimal Utilization based on

costs and revenue for a hierarchical decision model based on JIT goals
(Working Paper)
VI Preface
Chapter 4 : Altendorfer et al. (2014): Periodical capacity setting methods

for make-to-order multi-machine production systems (published) and
Hübl and Altendorfer (2014b): Queuing model for optimal switching
point for two capacity levels (Working Paper)
Chapter 5 : Hübl and Altendorfer (2014a): Inventory constraint defini-

tion for Conwip in a make-to-order environment – a simulation study
(Working paper)
Chapter 6 : Hübl et al. (2013): Influence of dispatching rules on average

production lead time for multi-stage production systems (published)
The applied simulation model was published in Hübl et al. (2011), Fel-
berbauer et al. (2012) and Altendorfer et al. (2013). Preliminary work
has been published in the following conference papers: Hübl and Gmainer
(2008); Hübl and Jodlbauer (2008); Hübl et al. (2009, 2010); Hübl (2014);
Hübl and Altendorfer (2015).
Alexander Hübl
Wels, March 2015
The PhD thesis has been published by Springer within the FTI (Förde-
rung der wirtschaftlich – technischen Forschung, Technologieentwicklung
und Innovation in Oberösterreich) policy of the federal government of Up-
per Austria for the reserach field ”Supply Chain Planning” of Logistikum,
University of Applied Sciences Upper Austria. I’d like to give special thanks
to my co-authors of the published articles for their inspiration and to Erika
for proofreading this book.
Alexander Hübl
Wels, April 2017
Abstract
In this thesis models for production planning are developed and perfor-
mance indicators are analysed to investigate production system behaviour.
Existing literature is extended by considering uncertainty of customer re-
quired lead time and processing times and by increasing the complexity of
multi-machine multi-items production models.
Results are on the one hand a decision support system for determining
capacity and the further development of the production planning method
Conwip and on the other hand the JIT intensity has been developed and the
effects of dispatching rules on production lead time are analytically proven.
The decision support system for determining the capacity provided is based
on the capacity demanded, whereby process and/or customer uncertainty
can be included. The production planning method Conwip has been ex-
tended by including safety stock and by extending the inventory constraint
to sum of WIP (Work-In-Process) and FGI (Finished Goods Inventory) and
not only WIP. JIT-intensity measures the degree of fulfilment of the ”seven
zeros” in a production system. A high JIT intensity leads to an increase of
excess capacity and to a reduction of utilisation and means that the produc-
tion system follows the customer demand fluctuations without any ”seven
zeros”-relevant losses. Consequently, management has the opportunity to
increase sales by selling more customer required items or to reduce excess
capacity. Finally, the effects of dispatching rules on average production
lead time are investigated. An analytic relationship between ”processing
time weighted average production lead time” and covariance between pro-
cessing time and queuing time has been identified. This relationship has
been proven analytically for single-machine production systems. It has been
found out that the ”processing time weighted average production lead time”
for a multi-machine production system is not invariant with respect to the
applied dispatching rule.
Contents
Preface V
Abstract VII
Contents IX
1 Introduction 1
1.1 Problem Definition . . . . . . . . . . . . . . . . . . . . . . . . 1
1.2 Research Questions . . . . . . . . . . . . . . . . . . . . . . . . 4
1.3 Methodology . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
1.4 Outline . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
2 Literature Review 9
2.1 JIT Goals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
2.1.1 Hierarchical Production Planning Models . . . . . . . 9
2.1.2 Utilisation in Production Planning . . . . . . . . . . . 12
2.1.3 Impact of JIT Activities on Performance . . . . . . . . 12
2.2 Capacity Setting Methods . . . . . . . . . . . . . . . . . . . . 14
2.3 Conwip . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
2.4 Dispatching and Production Lead Time . . . . . . . . . . . . 19
3 Utilisation Concept 23
3.1 Seven Zeros . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
3.2 Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
3.2.1 Utilisation . . . . . . . . . . . . . . . . . . . . . . . . . 27
3.2.2 JIT Intensity . . . . . . . . . . . . . . . . . . . . . . . 30
3.3 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
3.3.1 Long Term Level . . . . . . . . . . . . . . . . . . . . . 31
3.3.2 Medium Term Level . . . . . . . . . . . . . . . . . . . 32
3.3.3 Short Term Level . . . . . . . . . . . . . . . . . . . . . 32
3.4 Concluding Remarks . . . . . . . . . . . . . . . . . . . . . . . 35
4 Capacity Setting Methods 37

4.1 Introductory Remarks . . . . . . . . . . . . . . . . . . . . . . 37
4.2 Model Framework . . . . . . . . . . . . . . . . . . . . . . . . 38
4.3 Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40
4.3.1 Approximation of the Demanded Capacity
(Step 1) . . . . . . . . . . . . . . . . . . . . . . . . . . 42
4.3.2 Setting the Unbounded Provided Capacity (Step 2) . . 47
4.3.3 Capacity Account (Step 3) . . . . . . . . . . . . . . . 49
X Contents
4.4 Multi-Machine Concept . . . . . . . . . . . . . . . . . . . . . 50

4.5 Simulation Study . . . . . . . . . . . . . . . . . . . . . . . . . 52
4.6 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 55
4.6.1 Evaluation of Best Method Combination . . . . . . . . 55
4.6.2 Result comparison . . . . . . . . . . . . . . . . . . . . 56
4.6.3 Scenario Discussion of the Best Methods . . . . . . . . 58
4.7 Extension . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 60
4.7.1 Assumptions . . . . . . . . . . . . . . . . . . . . . . . 62
4.7.2 Model Development . . . . . . . . . . . . . . . . . . . 62
4.7.3 Production Lead Time Approximation . . . . . . . . . 64
4.7.4 FGI and Backorder Calculation . . . . . . . . . . . . . 66
4.7.5 Service Level Calculation . . . . . . . . . . . . . . . . 68
4.7.6 Numerical Examples . . . . . . . . . . . . . . . . . . . 69
4.9 Appendix . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 72
5 Conwip 87
5.1 Model Description . . . . . . . . . . . . . . . . . . . . . . . . 88
5.1.1 Production System . . . . . . . . . . . . . . . . . . . . 88
5.1.2 Production Planning and Control Applying
Conwip . . . . . . . . . . . . . . . . . . . . . . . . . . 88
5.1.3 Simulation Model . . . . . . . . . . . . . . . . . . . . . 90
5.1.4 Simulation Study . . . . . . . . . . . . . . . . . . . . . 91
5.2 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 93
5.2.1 Discussion of WAW and Wipcap . . . . . . . . . . . . 93
5.2.2 Wipcap Method Comparison . . . . . . . . . . . . . . 95
5.4 Appendix Conwip . . . . . . . . . . . . . . . . . . . . . . . . 97
6 Dispatching Rules 109

6.1 Model Framework . . . . . . . . . . . . . . . . . . . . . . . . 109
6.2 Simulation Study . . . . . . . . . . . . . . . . . . . . . . . . . 113
7 Conclusions 121
Bibliography 123
List of Figures
1.1 SCP matrix (Rohde et al., 2000; Stadtler, 2005; Stadtler and
Kilger, 2008) . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
2.1 Manufacturing resource planning MRPII (Wight, 1984; Hopp

and Spearman, 2008) . . . . . . . . . . . . . . . . . . . . . . . 10
4.1 Matrix of stochastic information for demanded capacity –

four methods . . . . . . . . . . . . . . . . . . . . . . . . . . . 39
4.2 Single-machine production system . . . . . . . . . . . . . . . 40
4.3 Basic idea of capacity setting . . . . . . . . . . . . . . . . . . 40
4.4 Timeline . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42
4.5 Cumulated demanded capacity – four methods . . . . . . . . 43
4.6 Operation characteristics . . . . . . . . . . . . . . . . . . . . . 45
4.7 Comparison of the three developed methods calculating the
capacity provided . . . . . . . . . . . . . . . . . . . . . . . . . 47
4.8 Multi-machine production system . . . . . . . . . . . . . . . . 53
4.9 Comparison best capacity setting method with average capacity 58
4.10 Long capacity setting period and capacity flexibility . . . . . 60
4.11 M/M/1 production system with WIP and FGI . . . . . . . . 61
4.12 Approximation error of the PDF of the production lead time 66
4.13 Logistic key figures . . . . . . . . . . . . . . . . . . . . . . . . 71
5.1 Combined BOM and routing structure . . . . . . . . . . . . . 89

5.2 WAW with best Wipcap . . . . . . . . . . . . . . . . . . . . . 94
5.3 Wipcap with best WAW . . . . . . . . . . . . . . . . . . . . . 95
6.1 Multi-stage production system . . . . . . . . . . . . . . . . . 109

List of Tables
3.1 Definitions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
3.2 Decreasing utilisation by realising JIT goals in bottleneck
situations of the equipment considered . . . . . . . . . . . . . 29
3.3 Effects of parameters on total variable costs and revenue by
short term decisions . . . . . . . . . . . . . . . . . . . . . . . 34
4.1 Definition of variables for single-machine model . . . . . . . . 41

4.2 Definition of parameters . . . . . . . . . . . . . . . . . . . . . 42
4.3 Definition of additional variables . . . . . . . . . . . . . . . . 51
4.4 Simulation runs . . . . . . . . . . . . . . . . . . . . . . . . . . 52
4.5 Scenarios . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53
4.6 Service level increase and tardiness decrease potential in basic
scenario . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 56
4.7 Comparison of methods over all scenarios . . . . . . . . . . . 57
4.8 Comparison of scenarios . . . . . . . . . . . . . . . . . . . . . 59
5.1 Processing and set-up times . . . . . . . . . . . . . . . . . . . 92

5.2 Utilisation at each machine group . . . . . . . . . . . . . . . . 93
5.3 Limits for Wipcap and WAW . . . . . . . . . . . . . . . . . . 93
5.4 Introduced safety stock percentage of the periodic demand
for FGI . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 94
5.5 Analysis 1 of Wipcap method . . . . . . . . . . . . . . . . . . 98
5.6 Analysis 2 of Wipcap method . . . . . . . . . . . . . . . . . . 99
5.7 Analysis Wipcap without safety stock 1 . . . . . . . . . . . . 100
5.11 Analysis Wipcap with safety stock 1 . . . . . . . . . . . . . . 104
6.1 Definitions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 118

6.2 Proc. time weighted average production lead time . . . . . . 119
6.3 Correlation coefficients ρ[P, Q] . . . . . . . . . . . . . . . . . 119
6.4 cv [L]for production systems . . . . . . . . . . . . . . . . . . . 120
6.5 E[L]for production systems . . . . . . . . . . . . . . . . . . . 120
List of Theorems
5.2.1 Observation (System load) . . . . . . . . . . . . . . . . . . . . 96

5.2.2 Observation (WIP plus FGI constraint) . . . . . . . . . . . . 96
5.2.3 Observation (Saftey stock) . . . . . . . . . . . . . . . . . . . . 96
5.2.4 Observation (WAW) . . . . . . . . . . . . . . . . . . . . . . . 97
6.1.1 Theorem (Mulit-stage systems) . . . . . . . . . . . . . . . . . 110

6.1.2 Proof (Multi-stage systems) . . . . . . . . . . . . . . . . . . . 110
6.1.3 Theorem (Dispatching rules) . . . . . . . . . . . . . . . . . . 111
6.1.4 Proof (Dispatching rules) . . . . . . . . . . . . . . . . . . . . 112
6.2.1 Observation (”Processing time weighted average production
lead time”) . . . . . . . . . . . . . . . . . . . . . . . . . . . . 115
6.2.2 Observation (Multi-stage production systems) . . . . . . . . . 116
Chapter 1
Introduction
1.1 Problem Definition
Production planning defines production goals, such as on-time delivery, op-
timal utilisation, short production lead times and low inventories. Unfortu-
nately, most of these objectives conflict with each other. Finishing a job on
the customer required due date is much easier if the equipment utilisation
is low because enough excess capacity is available. An efficient utilisation
of the equipment results in a higher return of investment, provided that the
equipment creates products that can be sold. To achieve this, a high utilisa-
tion of the equipment is required. Moreover, the delivery time can easily be
reduced by setting up an enormous Finished Goods Inventory (FGI) leading
to high costs.
However, after setting the objectives the production planning estimates
the resources which will be required to achieve these goals. In manufac-
turing companies various tasks, such as capacity planning, lot sizing, order
release and scheduling or dispatching, have to be fulfilled within the pro-
duction planning process (Schuh, 2006; Jodlbauer, 2008b; Hopp and Spear-
man, 2008). Each of these tasks are associated with a different time horizon
depending among other things on the degree of uncertainty involved. A
hierarchical planning model (Hax and Meal, 1975; Meal, 1984; Schuh, 2006;
Hopp and Spearman, 2008) or the Supply Chain Planning (SCP) matrix
(Rohde et al., 2000; Stadtler, 2005) links various tasks to their different
planning horizons. Detailed reviews on hierarchical production planning
are provided by Fleischmann and Meyr (2003), Meyr (2004) and Missbauer
and Uzsoy (2011).
Many companies have implemented a hierarchical production planning
approach, which is demonstrated by the interests of recent research in the
integrated investigation of two or more planning levels (Jacobs and We-
ston Jr., 2007; Rafiei et al., 2013; Chakrabortty and Hasin, 2013; Jansen
et al., 2013; Olhager, 2013). Jacobs and Weston Jr. (2007) and Olhager
(2013) provide the historical development of production planning systems
such as MRP (Material Requirement Planning), MRPII (Manufacturing
Resource Planning), ERP (Enterprise Resource Planning) and supply chain
planning (SCP). Powell et al. (2012) explore the functionality offered by
ERP for pull production systems. They found out that ERP systems that
were analysed in the case studies lacked support for demand smoothing,
production levelling, order quotation, process planning and Kanban con-
figuration. Olhager (2013) states that the new systems ERP and SCP still

A. Hübl, Stochastic Modelling in Production Planning,
DOI 10.1007/978-3-658-19120-7_1
2 Chapter 1. Introduction
apply a hierarchical planning approach, which stipulates that a strategic de-

cision on capacity investment is made first and subsequently the aggregated
production plans are disaggregated in the medium term level. Finally, on
the short term level, the production control and execution (for instance job
dispatching) is conducted (Hax and Meal, 1975; Meal, 1984; Schneeweiß,
2003; Hopp and Spearman, 2008).
Figure 1.1 shows the supply chain planning matrix (Rohde et al., 2000;
Stadtler, 2005; Stadtler and Kilger, 2008) where the main focus is on sup-
porting the material flow across the supply chain and its related business
functions: (I) purchase (procurement), (II) production, (III) distribution
and (IV) sales (x-axis). On the y-axis the associated hierarchical planning
tasks, from aggregated long term to detailed short term, are considered.
The planning tasks, as shown in the supply chain planning matrix, have
to be split up because not all input data and interrelations can be known
in advance (Steinrücke and Jahr, 2012). Moreover, the SCP matrix covers
cross-company and intra-company tasks.
procurement production distribution sales
Long term Strategic Network Planning
Medium Master Planning

term
Demand
Production Distribution Planning
Purchasing & Planning Planning
Material
Requirements
Planning Transport Demand
Short term Scheduling
Planning Fulfilment & ATP
Figure 1.1: SCP matrix (Rohde et al., 2000; Stadtler, 2005; Stadtler and Kilger,
2008)
Strategic Network Planning is the basis of planning activities. The re-

sult is the configuration of the network consisting of suppliers, production
plants, distribution centres and customers. The task of Demand Planning
is to identify future sales, whereby Demand Fulfilment and Available-To-
Promise (ATP) support the sales force to determine and confirm customer
due dates. Based on future sales the Master Planning synchronises material-
, information- and cash-flow, whereby all resources of the supply chain
should be utilised optimally. Production Planning disaggregates the output
of the Master Planning in terms of time, products and production plants.
1.1. Problem Definition 3
Scheduling defines the sequence of production orders for each resource. Dis-
tribution Planning and Transport Planning optimise the transport costs and
holding costs of finished goods in the supply chain which results in detailed
plans for vehicle routing. The main tasks of Purchasing & Material Re-
quirements Planning are to generate purchase orders and to verify material
availability based on the Demand Fulfilment & ATP module.
Graves (2011) states that three major types of uncertainties affect a
production plan of manufacturing companies:
• Uncertainty in demand forecast: All production plans focus on de-

mand forecast as input. Since forecasts are never perfect, the actual
demand may differ from the forecast and result in a forecast error.
• Uncertainty in external supply processes: A production plan results

in purchase orders which have to be fulfilled. Unfortunately, uncer-
tainties in the delivery date and volume might occur.
• Uncertainty in internal supply processes: A production plan relies on

predictions of internal manufacture, transportation and supply per-
formances. Again, the completion time and volume can deviate from
the production plan.
”Models for production planning which do not recognize the

uncertainty can be expected to generate inferior planning de-
cisions as compared to models that explicitly account for the
uncertainty” (Mula et al., 2006).
Mula et al. (2006) highlight that further research is needed in the area of
production planning under uncertainty: (I) Investigation of new approaches
to modelling of uncertainty, (II) development of new models that consider
additional sources and types of uncertainty, such as Customer Required
Lead Times (CRL), transport times, quality uncertainty, failure of pro-
duction systems and changes to product structure, (III) investigation of
incorporating all types of uncertainty in an integrated manner, (IV) devel-
opment of empirical work that compares the different modelling approaches
with real case studies, (V) development of a comparative evaluation of the
existent models for the different manufacturing systems. Following the hi-
erarchical planning concept, the reviews of Fleischmann and Meyr (2003),
Meyr (2004) and Missbauer and Uzsoy (2011) argue that little research has
been conducted on the interdependence between the planning hierarchies
and therefore further research is needed.
A further complication arises if the customer has the option of revising
the order specifications, which is a typical situation for automotive suppli-
ers. The integration of such stochastic behaviour into the planning process
is even more challenging if the customer changes the specifications directly

via Electronic Data Interchange (EDI) and if these changes have to be in-
tegrated into the next planning run without any human interaction.
In the case of manufacturing companies, a hierarchical production plan-
ning structure is often implemented but there is little help from literature
about how the levels of production planning interact with each other, espe-
cially in a stochastic environment. This thesis focuses on selected topics in
SCP concentrating on intra-company production planning tasks by applying
simulation and analytical methods in order to develop methods for improve-
ment and conducting investigations on production system behaviour.
1.2 Research Questions
In this section the research gap and research questions are identified, where-
by the literature review will be presented in Chapter 2.
The reviewed literature focusing on utilisation in a hierarchical produc-
tion planning structure ignores an investigation of firm’s performance of JIT
activities. However, it is known that opposing goals exist on each hierarchi-
cal level. This research gap will be presented in Section 2.1.
None of the reviewed models for capacity setting assume general distri-
bution of customer required lead time and processing times. Multi-machine
setting is not considered in the reviewed models. Moreover, the investigated
models in literature lack a rolling time horizon. Section 2.2 will cover this
research gap in more detail.
The pull production planning method Conwip will be reviewed in Section
2.3. It will be identified that the implementation of Conwip focusing on
improvements of service level and tardiness by changing Wipcap definitions
itself is rarely discussed. Moreover, it is shown that the implementation of
safety stock in a Conwip system has not been dealt with so far.
After reviewing the literature for dispatching rules influencing average
production lead time, it can be stated that the relevant literature ignores the
analytical link between average production lead time and the ”processing
time weighted production lead time” for multi-stage production systems.
Moreover, the influence of different dispatching rules on average production
lead time has not yet been proven theoretically. Section 2.4 will consider
this research gap thoroughly.
To refine the research field, the following specific research questions are
answered in Chapters 3 to 6 respectively:
1. Which utilisation based on costs and revenue is optimal for a hierar-

chical decision model from a JIT perspective? (Chapter 3)
1.3. Methodology 5
2. How can a decision support system for short and medium capacity set-
ting under a stochastic customer behaviour and stochastic processing
process be configured? (Chapter 4)
3. Does a constrained Wipcap of FGI and WIP for a Conwip system in

a multi-stage and multi-item make-to-order production system lead
to lower total costs in comparisons to classic Conwip where only the
WIP is constrained? Is it beneficial in terms of costs to implement
safety stock for FGI into a Conwip system? (Chapter 5)
4. How do dispatching rules influence average production lead time for

multi-stage production systems? (Chapter 6)
1.3 Methodology
Discrete event simulation and basic concepts of queuing theory are applied
in this thesis to answer the research questions. Mula et al. (2006) have found
out that most of the analytical models address only one type of uncertainty
by assuming a simple structure of the production process. Moreover, for
more complex structures of the production process and more types of un-
certainty the problems are solved by the use of simulation and artificial
intelligence. Since production planning has to deal with a huge amount of
different parameters and variables and a lot of stochastic effects, simulation
is used in recent literature (Kim and Kim, 1994; Huang et al., 1998b; Ku-
tanoglu and Sabuncuoglu, 1999; Gilland, 2002; Abdul-Kader and Gharbi,
2002; Arakawa et al., 2003; Li Sun et al., 2007; Jodlbauer and Huber, 2008;
Hübl et al., 2011; Felberbauer et al., 2012; Altendorfer et al., 2013; Gansterer
et al., 2014) to identify optimal parameters and methods for production
planning.
The most time consuming phase for discrete event simulation studies
is the input data collection and model development (Trybula, 1994; Perera
and Liyanage, 2000; Randell and Bolmsjo, 2001). Traditional simulation
models are created for a certain case only and adapting these models is
both personnel intensive and time consuming (Randell and Bolmsjo, 2001).
Therefore, a scalable simulation model for analysing production systems
with discrete event simulation has been developed (Hübl et al., 2011). The
simulation model is suitable for the three hierarchical planning levels ac-
cording to MRPII (Hopp and Spearman, 2008). The model has also been
applied in some case studies for practitioners (Felberbauer et al., 2012;
Altendorfer et al., 2014). This scalable simulation model is based on a
database, where the structure of the production system is defined in order
to avoid throw away solution (Thompson, 1994). Basically, the structure
of the database distinguishes between master data, such as bills of materi-

als, routings, production planning parameters, calenders, definition of skill
groups of the employees and their capacities, and transaction data defined
by distributions for processing time, set-up time, mean time-to-repair and
mean time-between-failure, customer demand, CRL and replenishment lead
time distribution.
For solving simplified models in the sense of assuming a certain distribu-
tion for random variables or considering only few random variables, analytic
models are developed to discuss production planning issues. Therefore, ba-
sic concepts of queuing theory (Medhi, 1991) are applied to answer some
research questions of this thesis.
The following software packages are used for preparing the dissertation:
• TeXstudio and TeXstudio in combination with MiKTex 2.9
• AnyLogic 6 and 7
• Mathematica 10
• scilab 5.4
• Microsoft Office
• ADOBE Acrobat Professional X
1.4 Outline
Chapter 2 provides an overview of relevant and recent literature focusing
on the defined research field in Chapter 1.
In Chapter 3, which is based on the working paper Hübl and Jodlbauer
(2013), an utilisation concept of the three planning levels (long, medium,
short) of the SCP matrix based on the JIT philosophy is developed and
discussed analytically. The ”seven zeros” as JIT philosophy are introduced
and the optimal utilisation on each hierarchical production level concerning
opposing goals is discussed.
Based on the result in Chapter 3, representing a predefined optimal
utilisation level, a decision support for short and medium term capacity
setting is developed in Chapter 4. This periodical capacity setting method
for MTO multi-machine production systems includes stochastic customer
required lead time and stochastic processing times to improve service level
and tardiness. The basic idea of this chapter is that the cumulated capacity
provided is greater than or equal to the cumulated capacity demanded for
the planning horizon. Based on the information available, processing time
distribution, customer required lead time distribution and a rolling plan-
ning horizon are considered on the periodical capacity setting method for
1.4. Outline 7
a multi-machine setting. Based on the customer orders, the capacity of the

production system is set, which results in an interaction between long term
and medium term planning level of the hierarchical planning approach and
connects demand planning and master planning of the SCP matrix. The
developed methods are evaluated by the use of discrete event simulation.
For a simplified environment the queuing model is developed for switch-
ing capacity. This chapter is based on Altendorfer et al. (2014), Hübl and
Altendorfer (2015) and the working paper Hübl and Altendorfer (2014b).
In Chapter 5, the classical medium term production planning method
Conwip, invented by Spearman et al. (1990) as part of the production plan-
ning, is extended by an FGI and WIP inventory constraint and is compared
to the classical Wipcap setting where only the WIP is limited. Moreover, a
safety stock for finished items is implemented in order to identify the cost
effects in a Conwip environment. The implemented methods are tested and
compared to classical Conwip by a discrete event simulation study. This
chapter is based on the working paper Hübl and Altendorfer (2014a).
Holding customer due dates is an important target for companies and
therefore scheduling, where dispatching rules are applied, affects the pro-
duction lead time. Therefore, Chapter 6, which is published in Hübl et al.
(2013), discusses the influence of dispatching rules on average production
lead time. Two theorems are developed based on the work of Nyhuis and
Wiendahl (2009), Jodlbauer (2005) and Jodlbauer and Stöcher (2006). Ba-
sic statistical relationships concerning the covariance of (in)dependent ran-
dom variables are applied in Theorem 1 to identify the relationship between
average production lead time and ”processing time weighted average pro-
duction lead time” for multi-stage production systems. From the inherent
logic of dispatching rules, the algebraic sign of the covariance between pro-
cessing time and queuing time is identified in this Chapter 6 for a set of
basic dispatching rules and their influence on average production lead time
is analytically discussed in Theorem 2 for single-stage production systems.
A simulation study has been conducted to confirm the developed theorems
and to gain more insight for practitioners.
A summary of the findings is presented in the last chapter of this thesis,
Chapter 7.
Chapter 2
Literature Review
An overview of relevant and recent literature focusing on the defined re-

search field in Chapter 1 is presented in this chapter. Due to the huge
amount of literature only the set of literature which forms the basis for the
developed models is discussed.
2.1 JIT Goals
Section 2.1 of this review covers the literature on key performance indi-
cator utilisation and JIT philosophy in a hierarchical production planning
setting. The reviewed literature lacks of a differentiated investigation on
utilisation for each hierarchical production planning level for a production
system avoiding waste in terms of a JIT perspective.
2.1.1 Hierarchical Production Planning Models
In this section some basic formulations on how production planning de-

cisions are split up into different hierarchical levels are presented. In 1975
Hax and Meal (1975) formalise a hierarchical production planning approach.
Bitran and Hax (1977) suggest an iterative procedure to optimise the sub-
problems defined in Hax and Meal (1975). Wight (1984) develops the hi-
erarchical production planning method MRPII (Manufacturing Resource
Planning), which implements the three planning levels long, medium and
short term (see Figure 2.1). The basis for MRPII is the closed loop MRP ap-
proach invented by Orlicky (1975). The MRP concept has some shortfalls,
such as capacity infeasibility, long planned lead times and system nervous-
ness. The integrated hierarchical planning system MRPII should eliminate
these flaws (Hopp and Spearman, 2008; Jodlbauer, 2008b).

DOI 10.1007/978-3-658-19120-7_2
10 Chapter 2. Literature Review
Demand
Forecast
Resource Aggregate Production

Long term Planning Planning
Rough-cut Capacity Master Production

Planning Scheduling
Bills of
Material Material
Requirements
Medium term Planning
Inventory
Status Capacity
Job Requirements
Pool Planning
Job Routing
Short term Release Data
Job
Dispatching
Figure 2.1: Manufacturing resource planning MRPII (Wight, 1984; Hopp and
Spearman, 2008)
The long term level basically influences capacity investment decisions

based on marketing parameters and results in rough prediction about future
production mix and volume. Long term planning covers three activities:
demand forecast (predicting future demand), resource planning (determines
the required capacities over the long term horizon) and aggregate production
planning (determines the level of production, staffing, inventory on part
families).
The medium term level combines the information from the long term
planning along with information about customer orders. The prediction
about production mix and volume is translated into production orders with
related due dates (Vollmann et al., 1997; Hopp and Spearman, 2008). The
Master Production Schedule (MPS) converts the long-term aggregated for-
casted demand into a detailed forecast while tracking individual customer
orders. The rough-cut capacity planning provides capacity checks of a few
critical resources to ensure the feasibility of the MPS. MRP (Orlicky, 1975)
translates the demands resulting from the MPS into production orders with
related due dates by the use of Bill-Of-Materials (BOM), routing data and
the inventory status. Hopp and Spearman (1996, 2008) implement pull
strategies such as Conwip (Constant Work-In-Process) (Spearman et al.,
1990) or Drum Buffer Rope (DBR) (Goldratt, 1988) into the MRPII con-
cept. Additionally, some methods are used that are not based on production
plans, such as Kanban (Ohno, 1988) and Reorder Policies (Arrow et al.,
1951). The result of medium term level are production plans (Job Pools)
stating the work process that need to be carried out (Hopp and Spearman,
2008). These production plans are based on presumptions and expectations
2.1. JIT Goals 11
concerning a number of factors, such as capacity, production mix, equip-

ment, inventory, lead times, disturbances, scrap and personnel (Wedel and
Lumsden, 1995). Capacity requirements planning provides a more detail
capacity check than the rough-cut capacity planning and generates loading
profiles based on the production plans. The production orders can be based
on customer orders (Make-To-Order, MTO) or on forecasts (Make-To-Stock,
MTS) if the customer required lead time is likely to be shorter than the pro-
duction lead time (Jodlbauer, 2008c; Hübl et al., 2010; Altendorfer et al.,
2014).
According to Hopp and Spearman (2008), the short term level controls
the real-time flow of materials (Job Release) based on the work schedule
developed on the medium term level. Dispatching rules are often used to
decide when which production lot has to be produced on which machine
(Job Dispatching) (Panwalkar and Iskander, 1977; Blackstone et al., 1982).
The hierarchical production planning model has been extended in the
past decades a lot and has become an important method in production plan-
ning. Graves (1982) uses Lagrange multipliers to solve a mixed integer linear
program based on Hax and Meal’s (1975) hierarchical production planning
approach. Kok (1990) discusses the mathematical logic of a hierarchical
planning approach and proposes a procedure to compute the aggregate pro-
duction volume and the allocated quantities in order to achieve a target
service level.
Gfrerer and Zäpfel (1995) propose robust production plans for the ag-
gregated planning level and a disaggregation method for the lower level for
uncertain demand. Zäpfel (1996) develops a hierarchical model for uncer-
tain demand, which can be incorporated into the MRPII concept.
Mula et al. (2006) provide a literature review for production planning
models under uncertainty. Rafiei et al. (2013) investigate the medium term
and short term production planning level in a hierarchical production plan-
ning approach. They develop an MTO/MTS hybrid production planning
strategy, which is modelled as a meta-heuristic algorithm. Chakrabortty
and Hasin (2013) explore the aggregated production planning with fore-
casted demand, related operating costs and capacity for a multi-product,
multi-period model. Their approach minimises total costs including inven-
tory levels, labor levels, overtime, subcontracting and back ordering levels,
as well as labour, machine and warehouse capacity. Jansen et al. (2013)
investigate the non-linear relationship between Work-In-Process (WIP) of a
production unit and lead time in a hierarchical production planning setting.
They present a two-step lead time anticipation procedure where a Linear
Program (LP) is solved regardless of the production capacity available. A
local smoothing heuristic is applied which tracks the stochastic workload
during the planning horizon.
Yang and Fung (2014) use a hierarchical structure to solve an ATP

decision model for a multi-site MTO production supply chain. Ponsignon
and Mönch (2014) combine medium and short term levels in their simulation
study to evaluate the performance of their master planning approach for the
semiconductor industry. Moreover, the authors apply the methodology of
reduced simulation models (Hung and Leachman, 1999) to speed up the run
time of the experiment.
2.1.2 Utilisation in Production Planning

The classical long term objective is to minimise the capacity invested or
equivalently to maximise the utilisation. In the medium term, capacity
adjustments are made resulting from demand fluctuations. In the short
term view, minimising the utilisation means to produce efficiently because
the losses according to the ”seven zeros” philosophy are avoided. In the
following literature, opposing goals on each hierarchical planning level are
reviewed.
According to Hopp and Spearman (2008) utilisation is controversial be-
cause on the one hand high utilisation leads to low costs per unit but on the
other hand low utilisation allows high sales. Bradley and Glynn (2002) show
that higher capacity invested (this means lower utilisation) allows less in-
ventory. Mieghem and Rudi (2002) and Angelus and Porteus (2002) address
the joint (inventory vs. capacity investment) decision problem. Jodlbauer
and Altendorfer (2010) conclude that there is an impact of the utilisation on
the cost for capacity available and inventory needed. Especially in the case
of a small ratio of unit holding cost over unit capacity cost, the costs are
considerably increased if the utilisation is only a little higher or smaller than
optimal utilisation. Carrillo and Gaimon (2000) investigate the manufac-
turing performance through process changes and knowledge creation with
an optimal control model. They argue that process changes may lead to
long term increase in effective capacity but during implementation typically
reduce short term capacity. After reviewing relevant literature it becomes
clear that on each hierarchical level, utilisation is treated differently.
2.1.3 Impact of JIT Activities on Performance

Many industries are facing strong global competition because product life
cycles are shortened, time-to-market decreases and customers require fast
deliveries of a variety of products of an appropriate quality. Therefore,
it is absolutely necessary that a company ensures that the right product
of the right quality is available to the customer in the right quantity at
the right time. Companies applying lean practices such as Just-In-Time
(JIT), Total Productive Maintenance (TPM) or Total Quality Management
2.1. JIT Goals 13
(TQM) contribute substantially to the operating performance (Ahuja and

Khamba, 2008; Emde et al., 2012; Shah and Ward, 2003; Aspinwall and
Elgharib, 2013).
Various empirical studies examine the influence of JIT activities on a
company’s performance. The results of Sim and Killough (1998) provide
empirical evidence that performance gains from synergies of JIT activities
resulting from combining JIT activities with performances goals. In a sur-
vey of US manufacturing companies White et al. (1999) identify ten JIT
activities that are appropriate for implementation. Changes in performance
depend on the degree of JIT implementation and the company size. White
and Pearson (2001) propose how JIT activities can be implemented into the
decision making process of manufacturing companies.
According to their empirical study covering the Canadian automotive
parts manufacturing industry, Callen et al. (2005) show that JIT intensive
plants have more capacity waste than other plants, but they generate more
profit. Inman et al. (2011) observe the influence of JIT activities on a firm’s
manufacturing agility with their structural equation model. They argue that
if JIT activities in the manufacturing processes are already implemented in
a company, then an increased supplier/ customer integration could show a
greater impact on agility than JIT activities alone. Obermaier and Don-
hauser (2012) analyse the financial performance of companies as a function
of inventory holding in their empirical study. They identify a positive re-
lation between inventory holding and financial performance. According to
them, those companies with highest inventory show the best financial per-
formance and vice versa. Their findings indicate that the core principle of
JIT, avoiding inventory, has its limits.
JIT is a philosophy with the primary goal of continuously reducing all
forms of waste (Sugimori et al., 1977; Ohno, 1988; Golhar and Stamm, 1991;
Daugherty et al., 1994). Suzaki (1987) identifies waiting time, transporta-
tion, processing, inventory and motion as possible forms of waste. Accord-
ing to Brown and Mitchell (1991) there are two major forms of waste: high
inventories and unnecessary delays. Daugherty et al. (1994) define six pre-
requisites for continuous improvements: maximum equipment availability,
TQM, minimum changeover, optimum physical layout, multi-trained work-
force and standardised operations.
TPM literature provides different concepts for classifying waste. Naka-
jima (1988), the founder of TPM, describes six waste types (equipment fail-
ure; set-ups and adjustments; idling and minor stoppages; reduced speed;
quality defect and rework; reduced yield). Golhar and Stamm (1991) give an
extensive literature review about JIT and identify eleven critical variables
for eliminating waste. Ljungberg (1998) identifies in his study that many
sources of possible loss are not tackled by companies and he suggests the
use of a comprehensive model of loss management. Muchiri and Pintelon

(2008) give an overview of the performance measurements of TPM.
Already in 1983 Edwards (1983) has introduced the ”seven zeros” – zero
defects, zero lot size, zero set-ups, zero breakdowns, zero handling, zero lead
time, zero surging – as JIT goals, which have to be achieved to eliminate all
forms of waste, especially inventories. The term zero should express that
these figures should be continuously minimised.
2.2 Capacity Setting Methods for Medium Term

Capacity Planning
Section 2.2 is dedicated to literature covering capacity setting problems. In
the beginning, capacity investment problems are reviewed. This is followed
by a discussion about decision problems with capacity expansion and/or
reduction. Finally recent queuing state depended models are introduced.
On the one hand the due dates of the customers can be negotiated to
create a smoother capacity demand (Hopp and Roof Sturgis, 2000; Hegedus
and Hopp, 2001; Keskinocak and Tayur, 2004; Corti et al., 2006). On the
other hand the capacity can be adjusted to the fluctuations of the customer
demand (Kok, 2000; Bradley and Glynn, 2002; Mieghem and Rudi, 2002;
Defregger and Kuhn, 2006; Li et al., 2009; Mincsovics and Dellaert, 2009;
Buyukkaramikli et al., 2013). The methods discussed in this section are
based on capacity adjustment literature whereby a flexible capacity with
upper and lower bounds is assumed. Therefore, this literature review fo-
cuses on the capacity adjustment literature stream. Finally, the reviewed
literature lacks of a periodical decision support for short and medium term
capacity setting for improving service level and tardiness whereby stochastic
customer behaviour and stochastic production process is assumed.
Capacity expansion problems are first studied in capacity investment lit-
erature (Chenery, 1952; Manne, 1961; Luss, 1982; Kok, 2000; Pibernik and
Yadav, 2009). Chenery (1952) and Manne (1961) assume a deterministic
increasing demand and whenever demand exceeds the capacity available,
the capacity is expanded. Manne (1961) includes probabilities instead of a
constant rate of growth in demand and backlogs to the model of Chenery
(1952). Luss (1982) conducts an extensive literature review about capac-
ity expansion problems and classifies capacity investment problems in sev-
eral categories emphasising modelling approaches and algorithmic solutions.
Segerstedt (1996) develops a capacity constrained multi-stage inventory and
production control problem. He minimises the inventory costs and short-
age costs, whereby the cumulated capacity concept is applied as constraint.
The cumulated demanded capacity is not allowed to exceed the cumulated
provided capacity. Kok (2000) compares two capacity allocation strategies.
2.2. Capacity Setting Methods 15
A fixed capacity is assumed and if the demand exceeds the fixed capacity
the orders are delayed. Moreover, an additional capacity is introduced by
hiring additional personnel.
Decision problems with capacity expansion and/or reduction are mod-
elled in most cases as dynamic programs (Mieghem and Rudi, 2002; Bradley
and Glynn, 2002; Li et al., 2009). Bradley and Glynn (2002) develop an an-
alytic model for a single-machine and single product system which describes
the optimal long term balance between capacity and inventory. They show
that optimal inventory policy varies with capacity investment and that
higher capacity invested allows less inventory. Moreover, the authors de-
scribe how inventory should be optimally substituted for capacity in order
to minimise costs when the capacity level varies. Mieghem and Rudi (2002)
and Angelus and Porteus (2002) deal with this issue in a more general situ-
ation and gain similar results. Obviously there is a trade off between capital
invested in capacity and costs of the capital employed in inventories. Defreg-
ger and Kuhn (2006) develop an MTO order acceptance model with limited
inventory capacity, where it is possible to adjust the inventory levels. In Li
et al. (2009), capacity allocation methods with mixed integer programming
methods are compared for supply chain optimisation. The authors iden-
tify that an integrated planning approach achieves better results than an
approach where each subproblem is treated separately.
The MTO ability of production systems is evaluated in Jodlbauer (2008c)
depending on the capacity provided, the customer required lead time dis-
tribution, and the demand fluctuation. The result of this evaluation shows
that applying a capacity adjustment method which enables the reaction to
short term peaks can decrease the capacity provided for MTO environment.
Jodlbauer and Altendorfer (2010) present a concept for optimising the over-
all capacity provided, for which they use the customer required lead time
distribution too. The result of this paper indicates that flexible capacity on
a short and medium term basis can lead to a cost decrease.
The papers of Balakrishnan et al. (1996) and Balakrishnan et al. (1999)
discuss the capacity rationing problem for a two-product production sys-
tem whereby one product class leads to higher profits per unit. An order
rejection policy for the lower profit products to maximise company profit is
described. In Kok (2000), capacity allocation is discussed where capacity
has to be allocated to different product groups while minimising a total cost
function. In this model the production is triggered by an order-up-to policy
for each product group.
Yang and Fung (2014) present an ATP decision support for order accep-
tance/selection, due date assignment and order scheduling in a multi-site
MTO production supply chain system. They propose a two-stage hierarchi-
cal structure.
In recent research queuing state dependent capacity adjustment models

focusing on the transient behaviour of the queuing system when switch-
ing between different capacity levels are studied (Mincsovics and Dellaert,
2009; Buyukkaramikli et al., 2013). In Mincsovics and Dellaert (2009) a
continuous setting is discussed in which an up-switching-point and a down-
switching-point are identified and each switch incurs costs. A periodic
setting with two possible capacity levels has been studied extensively in
Buyukkaramikli et al. (2013).
2.3 Conwip
Section 2.3 summarises literature about the production planning method
Conwip, whereby the literature is distinguished between analytical models
and simulation. The production planning and control method Conwip is
firstly introduced by Spearman et al. (1990) and Spearman and Zazanis
(1992). The basic parameters Wipcap, work-ahead-window and capacity
trigger are already introduced in these papers. The method has since then
been further developed and compared to the most dominant production
planning and control methods MRP and Kanban either analytically or with
simulation. For a detailed review on Conwip and its applications see Frami-
nan et al. (2003). Prakash and Chin (2014) review 15 modified Conwip sys-
tems and classify them according to: (I) the feedback signal from demand
source triggers production, (II) the products are pushed between worksta-
tions and (III) the WIP level (Wipcap) is limited within workstations.
Looking at the literature discussed in the review below shows that Con-
wip performs well in comparison to other traditional production planning
and control methods. However, the implementation of Conwip in an MTO
system is yet rarely discussed as mostly average inventory and throughput
are compared. Especially research on the question of improving service level
and tardiness measures by changing the Wipcap definition itself and the
evaluation of a safety stock implementation in Conwip is still not available.
Analytical models usually describe the behaviour and performance of
Conwip in comparison to Kanban (or other production planning and control
methods) or they deliver approximations for basic logistical key performance
indicators, such as throughput, WIP and production lead time when Conwip
is applied to a more complex production system.
The first paper introducing Conwip, Spearman et al. (1990) model a
multi-stage serial production system as closed loop queuing network. More-
over, the authors compare its performance to an open loop queuing network
representing the push production control method.
Similarly the superior performance of Conwip in comparison to Kanban
and push methods as MRP is proven in Spearman and Zazanis (1992).
2.3. Conwip 17
Hence, Gstettner and Kuhn (1996) demonstrate that Kanban outperforms

Conwip due to the higher number of optimiseable parameters. Geraghty and
Heavey (2004) show that the optimal hybrid push/pull method proposed
in Hodgson and Dingwei Wang (1991a) and Hodgson and Dingwei Wang
(1991b) modelled as a Markov decision process is a Conwip system.
Herer and Masin (1997) develop an optimisation problem formulation
where the right order in which the parts should be produced in a Conwip
system is optimised.
An accurate approximation of the average production rate in a closed
loop system for more complex production system structures than discussed
in the papers mentioned above is developed in Gershwin and Werner (2007)
which can also be applied to a Conwip system. Furthermore, Heragu et al.
(2011) develop an approach to identify the waiting time for semi-open queu-
ing networks, also taking customer waiting time in front of the system into
consideration. Such semi-open queuing networks can be used to model Con-
wip.
An analytical method for determining Wipcap and Work-Ahead-Window
(WAW) of Conwip based on the customer required lead time distribution is
presented in Jodlbauer (2008a).
Helber et al. (2011) optimise the production rate and/or short term
profit of a Conwip system by the use of mixed integer linear program in
combination with a stochastic simulation. The accuracy of their method
proposed depends on the variability of the processing times and the WIP.
Satyam and Krishnamurthy (2013) model the batch size constraints for
a Conwip system as a multi-class closed queuing network with synchronisa-
tion stations. Based on a routing matrix for each station a random variable
for set-up time and processing time is modelled. The authors use a de-
composition approach as approximation because exact methods are hard to
solve. They identify that batch size decisions have an impact on average
production lead time. Practitioners can decide whether they adjust batch
sizes or Wipcap. Park and Lee (2013) study a multi-product Conwip pro-
duction system with a Poisson arrival process with a fixed probability that
an order requires a particular item made of different subcomponents. The
authors also apply a decomposition approach for modelling and solving.
Lagershausen et al. (2013) propose an approximation for the throughput
of a closed loop queuing network with generally distributed processing times.
Gong et al. (2014) quantify and measure information to compare the in-
formation amounts in MRP, Kanban and Conwip to study how information
amount affects the decision-making delay. They argue that MRP requires
the largest amount of information since the WIP in an MRP system is higher
compared to Kanban and Conwip.
Simulation studies either discuss extensions of Conwip and their perfor-

mance, which is also the objective of this section since the integration of
FGI into the Wipcap for MTO production systems is an extension, or they
compare the performance of Conwip to the performance of other production
planning and control methods.
Simulation studies which compare the performance of Conwip to that of
MRP, Kanban and/or DBR are Huang et al. (1998b), Huang et al. (1998a)
and Jodlbauer and Huber (2008), all stating that Conwip outperforms the
other methods. Furthermore, Gilland (2002) and Grosfeld-Nir and Maga-
zine (2002) compare the performance of Conwip to their developed methods
by simulation. They mainly find that Conwip outperforms MRP and DBR,
although their methods outperform Conwip in the respective environments.
Sepehri and Nahavandi (2007) also compare Conwip to other WIP con-
straining work release methods and find that it is partly outperformed by
them. Gstettner and Kuhn (1996) identify that Kanban leads to a lower av-
erage WIP than Conwip for a given production rate. The WIP levels in the
buffers between the working stations of a Kanban system show the ”WIP
bowl phenomenon” while in the Conwip system the WIP has equal values
in all stations except at the bottleneck station where the WIP accumulates.
Duenyas et al. (1993) model a Conwip system with deterministic process-
ing times, exponential failure and repair times as closed queuing network.
Moreover the authors derive mean and variance of the output. By empirical
tests, they show that their approach leads to robust solutions which can be
the basis for selecting production quota and Wipcap for a Conwip line.
Framinan et al. (2000) develop a new Conwip card setting method, which
leads to card counts per product, and discuss the performance of this card
setting method with the application of different dispatching rules. A simula-
tion study shows that this card setting method leads to decreased inventories
and increased service level for an MTS production system.
A dynamic card setting method for the Wipcap is presented in Hopp
and Roof (1998) whereby the number of cards is identified by statistical
process control applied to the throughput. They test the performance of
this method by simulation and find that it is robust against environmental
changes.
Bahaji and Kuhl (2008) conduct a simulation study comparing the influ-
ence of different dispatching rules on logistical performance in a wafer fab-
rication facility. For work release they compare Conwip to a push method
which is a simplified MRP setting. They find that their simplified MRP
method outperforms Conwip when applying some multi-objective compos-
ite dispatching rules developed in their paper. However, they also find
that these rules do not perform significantly better than First-Come-First-
Serve (FCFS) in a Conwip setting which supports the recommendation of
2.4. Dispatching and Production Lead Time 19
Spearman et al. (1990) and Hopp and Spearman (2008) to apply First-In-
System-First-Served (FISFS) within a Conwip production system.
Hübl et al. (2011) develop a simulation model where Conwip is embedded
in a hierarchical production planning structure as proposed by Hopp and
Spearman (2008).
Chong et al. (2013) investigate an integrated Kanban and Conwip pro-
duction system by the use of discrete event simulation. They use total
output, average WIP, average production lead time and average bottleneck
utilisation as performance metrics.
Huang et al. (2015) consider a Conwip assembly production system for
mass production with multi-loop, multi-products, low volume and one-of-a-
kind production environments. The authors develop eight Conwip loop con-
trol policies based on five basic design patterns. The result of the simulation
study shows that the multi-loop design performs better than a single-loop.
Moreover, they have also proposed a WIP upper bound heuristic algorithm
for searching the optimal Wipcap.
2.4 Dispatching and Production Lead Time

In Section 2.4 some literature is reviewed where dispatching rules influence
average production lead time. The goal of scheduling in production plan-
ning is to generate a profitable balance between conflicting objectives in
production planning. Most real-world scheduling applications are NP-hard
(Non-deterministic Polynomial-time hard) problems and therefore in prac-
tice dispatching rules are applied. Dispatching rules consider only local
and current states of the production system and neglect further jobs. How-
ever, dispatching rules effect the production lead time (Hopp and Spearman,
2008). The reviewed literature lacks an analytical relationship between dis-
patching rules and production lead time.
Short production lead times offer several advantages. According to Lit-
tle’s Law (Little, 1961), shorter production lead times also result in lower
WIP levels at the same utilisation and therefore less capital is employed.
Schonberger (1986) uses production lead time reduction to classify com-
panies into different levels of world class manufacturing. According to Neely
et al. (1995) and Wacker (1996) manufacturing performance is defined in
terms of quality, delivery speed, delivery reliability, price (cost) and flexibil-
ity, whereby delivery speed includes the production lead time. Merschmann
and Thonemann (2011), Yang et al. (2011) and Inman et al. (2011) include
the metrics ”reduction on production lead time” for measuring manufactur-
ing flexibility in their structural equation model.
Production lead times for assembly processes can vary because of queu-
ing or transportation delays. Moreover, variable set-up times and variable
processing times result in a distribution of the production lead time (Yano,

1987).
Wedel and Lumsden (1995) show in their case studies that the produc-
tion planning process is affected by reductions in the total manufacturing
lead time because the planning department has a lack of confidence in the
manufacturing department or factory.
Altendorfer and Minner (2011) investigate a two-stage MTO production
system with random demands, processing times and distributed customer
due dates. They minimise total inventory holding and customer order tar-
diness costs to identify the optimal manufacturing capacities and planned
lead times for each stage. As a result they have proven that the distribution
of the customer required lead time has no influence on the optimal planned
lead times whenever capacity is predefined.
Hayya et al. (2011) present a procedure for reducing the mean and vari-
ance for exponentially distributed lead times.
Furthermore, shorter production lead times allow quicker responses to
market demand changes (Altendorfer and Jodlbauer, 2011). Average pro-
duction lead time is influenced by the decisions of production planning and
control. For short term scheduling, especially dispatching rules are often
applied and so their influence on average production lead time is a subject
of research (see the reviews of Blackstone et al. (1982); Waikar et al. (1995);
Rajendran and Holthaus (1999) as well as the book of Pinedo (2008)). Dis-
patching rules are used to select the next order to be processed from waiting
orders in front of the processing station.
Bednowitz et al. (2014) have tested different dispatching and loitering
policies for unmanned aerial vehicle systems. This military application ex-
amines a system with multiple unmanned aerial vehicle systems assigned to
respond to fixed-location,
multiple priority targets. The aim is to ensure rapid service for high priority
targets and effective management of medium priority levels. The authors
identify that dispatching rules with a think-ahead1 and distance-based focus
lead to best results.
Various simulation studies examine the influence of dispatching rules on
average production lead time. Barrett and Kadipasaoglu (1990) show for a
dynamic flow shop that the Shortest Processing Time (SPT) rule performs
best in terms of average production lead time. Waikar et al. (1995) compare
ten different dispatching rules under different shop loads in their simulation
study. Evaluating the average production lead time of jobs, again the SPT
rule performs best and First-In-First-Out (FIFO) is ranked fourth. Hung
and Chen (1998) show that Shortest Remaining Processing Time (SRPT)
1 According to Mantel and Landeweerd (1995) think-ahead implies that known tasks
are combined to routes such that idle (empty) travel time is minimised.
2.4. Dispatching and Production Lead Time 21
and Earliest Due Date (EDD) are good dispatching rules to reduce average
production lead time in semiconductor wafer fabrication. Land et al. (2014)
focus on overcoming the conflict between order release and dispatching. The
authors find out that operating due date dispatching rules are beneficial.
Jayamohan and Rajendran (2000), El-Bouri et al. (2008) and Chen and
Matis (2013) compare new dispatching rules with standard rules with regard
to different performance measures such as average, maximum and variance
of production lead time and tardiness. The results of Jayamohan and Rajen-
dran’s (2000) simulation study show that SPT is the rule with the shortest
average production lead time for flow shops.
Whenever simulation is applied to discuss the effect of dispatching rules
on the average production lead time, the focus is mainly on comparing and
evaluating the performance of dispatching rules and developing new dis-
patching rules. Very little literature is available on the development of ana-
lytical models based on simulation and empirical studies. One such stream is
based on the application of the funnel model for single-stage models (Wien-
dahl et al., 1994; Wiendahl and Breithaupt, 1999) where approximation
functions for production lead time and inventory are identified applying
empirical data for their parametrisation. Based on this funnel model, Ny-
huis and Wiendahl (2009) derive approximations for average production
lead time using SPT and Longest Processing Time (LPT) as dispatching
rules.
From the research work reviewed above, it is obvious that most litera-
ture dealing with the interaction between dispatching rules and production
lead time either applies simulation or empirical data and approximation
equations. Nevertheless, queuing theory can also be applied.
An analytical model based on queuing theory shows that the SPT rule
minimises the expected production lead time in a static job shop with de-
terministic processing times (Buzacott and Shanthikumar, 1993; Hopp and
Spearman, 2008).
The work by Nyhuis and Wiendahl (1999) and Nyhuis and Wiendahl
(2009) introduce the range, which is the ”processing time weighted average
production lead time” in comparison to the arithmetic average production
lead time. They show that the range is independent of the dispatching
rule applied for a single-stage production system. The papers by Jodlbauer
(2005) and Jodlbauer and Stöcher (2006) extend the framework for contin-
uous input and output functions in a single-stage production system.
Chapter 3
Utilisation Concept for Hierarchical
Production Planning
Finding the optimal capacity for the machinery is a standard goal in ac-
counting and finance. Opposing goals on each hierarchical planning level
lead to different utilisation levels. In this chapter, a utilisation concept is
developed in which the JIT goals are taken into account. The concept shows
that JIT capability enables the production system to reduce the excess ca-
pacity by disinvestment or to increase sales without additional investment.
It is shown that low utilisation at the short term level is positively cor-
related to JIT intensive practices. Consequently, low utilisation results in
potential higher profits. Furthermore, there is no reason to maximise the
short term machine utilisation, other than to sell more machine capacity in
the sense of selling more products or producing for external companies.
3.1 Seven Zeros

In the beginning of this chapter, the ”seven zeros” according to Edwards
(1983) are introduced in more detail to derive afterwards the effects on
utilisation while achieving JIT goals.
Zero Defects
No scrap losses and rework are acceptable. One method to approach ”zero
defects” is implementing TQM (Daugherty et al., 1994). The aim of TQM is
to continuously improve and sustain the quality of products and processes.
In addition, TQM has a strong customer as well as process orientation.
Rehder (1989) states that JIT and TQM are mutually supportive. Flynn
et al. (1995) empirically show that TQM practices improve JIT performance
and JIT practices have a positive effect on quality performance. According
to Deming (2000), TQM also reduces fluctuations in all system components.
This is considered to be a prerequisite for a successful implementation of
JIT.
Zero Lot Size

The production lot sizes should be equal to the customer order quantity.
There should be no queuing time because of production lots greater than
the customer orders. Short set-up times (see the JIT goal ”zero set-up”)
are the main prerequisite of small production lots, permitting a good match

DOI 10.1007/978-3-658-19120-7_3
24 Chapter 3. Utilisation Concept
between production rate and customer demand (Daugherty et al., 1994). In

a very highly customised environment this means a production lot size of
one. Furthermore, small lot sizes increase production flexibility. In addition
it may be useful to have smaller transport batches than production lots to
ensure a smooth material flow, such as overlapping production.
Zero Set-Up
No changeovers are necessary or, at least, set-up times are vanishing. Small
lot sizes are only possible if the set-up times are very short. Finch (1986)
addresses a variety of techniques to reduce and simplify set-ups. One key
concept is to distinguish between internal and external set-up. Internal set-
up time causes an interruption of the production process while the external
set-up can be accomplished parallel to the production. Single Minute Ex-
change of Die (SMED) is one of the most widely known methods to reduce
the set-up time (Shingo, 1985). Combining ”zero lot size” and ”zero set-
ups” means that the total time used for changeovers decreases even though
the number of set-ups increases.
Zero Breakdown
No machine failures are acceptable to ensure maximum equipment avail-
ability (Daugherty et al., 1994). ”Zero breakdown” means that there is no
idle time caused by machine or tool failure. TPM is a method to max-
imise equipment effectiveness throughout its entire life (Nakajima, 1988).
Schonberger (1986) emphasises the importance of introducing JIT, TQM
and preventive maintenance practices simultaneously based on his concept
of world class manufacturing. Cua et al. (2001) conducted a survey of TQM,
JIT and TPM practices and their overlaps. In addition their study clearly
demonstrates the importance of joint implementation of JIT, TQM and
TPM.
Zero Handling
Unnecessary movements of the material or unnecessary motions conducted
by the machine operator is not acceptable according to Hines and Rich
(1997). The items go directly from one workstation to the next and double
handling or excessive movements should be avoided, according the pioneer
of TPM (Ohno, 1988). The goal ”zero handling” yields short processing
times as well as short transportation times and no queuing time between
two workstations.
3.2. Model 25
Zero Lead Time

According to ”zero lead time”, no queuing time between two workstations
is acceptable and the processing time should be as short as possible. The
idea is that a downstream workstation requests parts which are provided
immediately without any delay. The pull control system Kanban (Ohno,
1988) is an implementation of this idea. According to Little’s Law (Little,
1961; Jodlbauer and Stöcher, 2006), the inventory equals the lead time
multiplied by the average production rate. Thus, the goal ”zero lead time”
is very close to the main JIT goal ”zero inventory”.
Zero Surging
Only small fluctuations in the production plan are acceptable for fulfilling
”zero surging”. In general a smooth final assembly or production schedule
(nearly constant production rate and production mix) is applied in JIT
systems. Otherwise, a high excess inventory or a high excess capacity is
needed.
Maybe the most challenging task for a JIT environment is to transform
high customer demand fluctuations into a smooth required production rate.
Some authors have developed methods to manage demand uncertainty or
demand peaks (Rees et al., 1987; Bartezzaghi and Verganti, 1995; Verganti,
1997; Co and Sharafali, 1997).
There is a trade-off between excess inventory and excess capacity (Bradley
and Glynn, 2002; Jodlbauer and Altendorfer, 2010). The general trend is
to prefer excess capacity to excess inventory, especially for various finished
goods (Schragenheim and Ronen, 1991). Thus, real JIT systems need some
excess capacity to cope with high demand fluctuations and to ensure contin-
uously high service levels. One way to manage demand peaks is to provide
extra capacity, such as an additional shift or overtime work. The classical
way of Toyota to manage the situation on which the production falls behind
the desired rate is two-shifting, whereby two shifts are scheduled per day
separated by down periods for catching-up the desired production rate, for
preventive maintenance or cross training (Schonberger, 1982).
3.2 Model
In this chapter, a single-machine multi-item production system is assumed.
The concept can be extended easily to a multi-machine setting. The sched-
uled working time T of the production system including preventive mainte-
nance is predefined as the maximum capacity available. Idle time tI repre-
sents machine breakdowns, waiting for an order, tool or worker unavailabil-
ity.
All production orders nO are considered for production and have to be

produced by the production system completely. Production orders consist
of items for customer nC,i (MTO), stock nS,i (MTS) and/or scrap nQ,i .
A set-up of the machine is assumed after each production order and is
described by the set-up time per changeover tS,i . Only the internal set-
up time during which the machine is completely inoperable is considered
whereas external set-up times which occur after techniques such as SMED
are not taken into account since they do not influence the utilisation of an
equipment. The processing time consists of the minimal processing time
per item tP,i and the additional processing time due to waste tAP,i such
as speed (performance) losses, transportation and unnecessary movements.
Table 3.1 presents the notation used in this chapter.
Table 3.1: Definitions
Symbol Description
U Machine utilisation (U < 1)
Umin Minimum machine utilisation (Umin < U )
EC Excess capacity
ECmax Maximum excess capacity (customer enabled excess capac-
ity)
IJIT JIT intensity (see equation 3.6)
tP,i Minimal possible processing time per item of production or-
der i
tS,i Set-up time per changeover of production order i
tAP,i Additional processing time per item of production order i
due to waste
tI Idle time
T Scheduled working time
nC,i Number of items needed for customer orders of production
order i without any production on stock
nS,i Number of items produced on stock of production order i
nQ,i Number of scrapped or obsolete items of production order i
nO Number of production orders
For the cost model, fix as well as variable costs are considered. Fix
costs include personnel costs for permanent workers and depreciation of
the equipment used. The variable costs are divided into variable processing
costs, variable material costs, holding costs and costs for rework or disposal.
The variable processing costs include variable costs for energy, spare parts,
operational supplements, expendables and set-ups and variable costs for
variable capacity (for instance: overtime, leasing workers). Variable mate-
3.2. Model 27
rial costs include the items to produce the parent item. Holding costs occur
for items which are on stock. If rework or disposal is necessary for items,
then costs for rework or disposal will occur. If improvements based on the
”seven zeros” are discussed, then the throughput has to be unchanged.
The total costs for the number of items produced on stock are assumed
to be higher than for the number of items produced for customer orders
because of the holding costs. Assuming enough excess capacity to meet
demand peaks, it is beneficial not to produce on stock because of lower costs
and earlier effective revenue. For the trade-off between excess capacity and
excess inventory, see for example Mieghem and Rudi (2002), Bradley and
Glynn (2002) or Jodlbauer and Altendorfer (2010).
3.2.1 Utilisation
Machine utilisation describes how intensively a machine is being used to
produce items. Thus, machine utilisation equals the ratio time used (run
time plus set-up time) to time available (Kuik and Tielemans, 1997; Cox
and Blackstone, 2002).
The time used is the sum of the run time for all produced items during
one time period (for instance one year) plus the set-up times for all pro-
duction orders during the same time period. All produced items consist of
items for customers, items to be produced on stock (these items go first on
stock and after some queuing time to the customer or are obsolete and are
fault free) and scrapped items.
The time available is the sum of all scheduled working time during the
time period. Summarising all these ideas yields the following relationship
for the utilisation:

n O
(tP,i + tAP,i )(nC,i + nS,i + nQ,i ) + tS,i
i=1
U := (3.1)
T
Idle time (no matter for what reason) of the equipment leads to a de-
crease of utilisation and idle time meets the non-negativity constraint. Since
100 percent utilisation equals to 100 percent minus the percentage of time
lost due to the unavailability of orders, machines and/or workers, the fol-
lowing relationship is introduced (Cox and Blackstone, 2002):

nO
tI := T − [(tP,i + tAP,i )(nC,i + nS,i + nQ,i ) + tS,i ] >= 0 (3.2)
i=1
Based on idle time and utilisation the excess capacity EC can be defined
as the ratio idle time to time available (scheduled working time):

n O
T− (tP,i + tAP,i )(nC,i + nS,i + nQ,i ) + tS,i
tI i=1
EC := = (3.3)
T T
In the following paragraphs, Equations 3.1 to 3.3 are analysed by iden-
tifying the effects on them when the ”seven zeros” methodology is applied.
By achieving ”zero defects”, scrap and rework should be avoided and there-
fore the scrapped items nQ,i are reduced whereby nC,i and nS,i remain
unaffected. This results in a decrease of utilisation.
The two JIT goals ”zero lot size” and ”zero set-ups” are considered to-
gether because changing the lot size influences the amount of set-ups imme-
diately. ”Zero lot size” influences the parameter number of production or-
ders nO because the higher nO the more customer lot sizes are applied in the
production system. ”Zero set-ups” affects the set-up time per changeover
tS,i because it
is assumed that a set-up is necessary after each production
nO
order. Hence, i=1 tS,i is minimised by applying these two ”seven zeros”
and results in a decreased utilisation.
”Zero handling” minimises the additional processing time tAP,i by avoid-
ing unnecessary movements, performance losses and transportations. The
application of ”zero handling” results in a decreased utilisation as well.
According to Little’s Law (Little, 1961), ”zero lead time” influences all
unnecessary inventories such as number of scraped items nQ,i and number
of items produced on stock nS,i . ”Zero lead time” results in a reduction of
unnecessary inventories and reduces the metrics utilisation.
”Zero breakdown” reduces the idle time tI occurred by broken equip-
ment. However, this gain is transferred to idle time due to waiting for order,
since it is assumed that the same numbers of items needed for customer or-
ders are produced. Therefore, this ”zero” has no immediate consequence on
the utilisation metrics. But it allows potential for reducing the scheduled
working time T or the sales department to increase the customer orders
nC,i .
If the JIT goal ”zero surging” is focused, then the idle time tI is reduced
because a production plan with low fluctuation is applied to avoid over and
under load. Again, no immediate consequence on the utilisation metrics
occurs but there is potential for reducing the scheduled working time T or
for the sales department to increase the customer orders nC,i .
All ”seven zeros” themselves result in an increase of idle time tI and
lead to a decreased or unchanged utilisation metrics (see Table 3.2 column
U ). However, all ”seven zeros” cause ”further potential”, which can be
(I) a reduction of the scheduled working time (T ↓) or (II) an increased
3.2. Model 29
production of more customer required items (see Table 3.2 column ”Further
potential”). Even for ”zero breakdown” and ”zero surging” this further
potential occurs since the idle time can be used for producing more customer
required products or reduced after their implementation. The effects of the
”seven zeros” as JIT goal on utilisation are defined in Equation 3.1 and 3.2
and are summarised in Table 3.2. In this table, the ”seven zeros” and the
respective parameters influenced are listed.
Table 3.2: Decreasing utilisation by realising JIT goals in bottleneck situations

of the equipment considered
Seven zeros has an results in U Pot.∗

effect on

n O
Zero defects nQ,i min nQ,i (tP,i + tAP,i ) dec.a yes
nQ,i i=1

n O
Zero lot size tS,i ,nO min tS,i dec.a yes
tS,i ,nO i=1
combined with
zero set-ups
Zero break- tI tI → tI unc.b yes

downs

n O
min tAP,i (nC,i
Zero handling tAP,i tAP,i i=1 dec.a yes
+nS,i + nQ,i )

n O
min (tP,i + tAP,i )
Zero lead nS,i ,nQ,i nS,i ,nQ,i i=1 dec.a yes
time ·(nS,i + nQ,i )
Zero surging tI tI → t I unc.b yes

dec. = decrease
unc. = unchanged
Further potential
If each JIT goal is theoretically achieved, the minimum attainable utilisa-

tion, called minimum utilisation Umin , equals the fraction minimal possible
processing time for produced items for customers divided by the scheduled
working time T . This results in Equation 3.4.

n O
tP,i nC,i
i=1
Umin := (3.4)
T
The minimum utilisation is the customer required utilisation without
capacity wasted for not optimised processing times, scrapped items, over-
production, set-ups or idle times. Moreover, the minimum utilisation (Equa-
tion 3.4) is less than the utilisation U (Equation 3.1). As discussed earlier
in this section, ”zero breakdown” and ”zero surging” have no influence on
utilisation. Therefore, the maximum excess capacity ECmax is introduced
in Equation 3.5 based on Equation 3.1 and 3.2. Excess capacity, which has
been defined as idle time tI , is according to Cox and Blackstone (2002)
the capacity that is not used to either produce or protect the creation of
throughput.

n O
T− tP,i nC,i
i=1
ECmax := (3.5)
T
ECmax can be interpreted as the customer enabled excess capacity for
producing the customer required items nc,i without any waste. If all wastes
based on the ”seven zeros” are avoided in production two scenarios are
possible: (I) divestment of ECmax since it is not used or (II) sales increase
by producing more customer required products so that ECmax equals to
zero.
3.2.2 JIT Intensity

Achieving the JIT goals means (I) minimising the machine utilisation to
the value of the minimum utilisation or (II) maximising excess capacity/idle
time in order to produce only the products requested by the customer. Thus,
the gap between excess capacity and customer enabled excess capacity can
be interpreted as the JIT intensity.
EC tI
IJIT := = (3.6)
ECmax
n O
T− tP,i nC,i
i=1
The JIT intensity expresses the conformance to JIT measured in how
well the ”seven zeros” are achieved. Moreover, JIT intensity is defined
as excess capacity without implementing the ”seven zeros” divided by the
customer enabled excess capacity where all ”seven zeros” are fully applied
(Equation 3.6). JIT intensity can be calculated dividing the idle time by
the idle time of a production system where no waste based on the ”seven
zeros” exists.
3.3. Discussion 31
Literature as discussed in Chapter 2 has shown that achieving JIT goals

leads to an improvement in performance. JIT intensity of 1 means that
every JIT goal is fully reached and JIT intensity of << 1 indicates in a
poor achievement of the ”seven zeros”.
Excess capacity is one way to manage scrapped items, demand fluctu-
ations, demand peaks, breakdowns and unavailability of tools or workers
(the most favoured way especially for many variants of the finished goods)
(Schragenheim and Ronen, 1991). If a high JIT intensity is achieved, it
is possible to reduce the excess capacity by divestment or increasing sales,
whereby the customer requirements are fulfilled without any additional in-
vestment.
The capacity investment enables the system to meet future customer re-
quirements. Especially if the demand is fluctuating, more capacity invested
means higher capability to produce more products, higher excess capacity
and a lower utilisation. But capacity investment costs more money.
Managerial Insight
Excess capacity increases or utilisation decreases by applying all ”seven
zeros” to achieve the customer enabled excess capacity ECmax or minimum
utilisation Umin respectively. The developed metrics JIT intensity provides
the management with a metrics in how well the ”seven zeros” are achieved,
since JIT activities have a positive impact on performance. Management
has two possibilities to deal with the customer required excess capacity: (I)
divestment to reduce the scheduled working time or (II) increase sales by
producing more customer required items.
3.3 Discussion
The influence of the opposing hierarchical goals on utilisation is discussed
for the three planning levels.
3.3.1 Long Term Level

The classical long term objective is to minimise the capacity invested or
equivalently to maximise the utilisation. Bradley and Glynn (2002), Jodl-
bauer and Altendorfer (2010) and Obermaier and Donhauser (2012) pointed
out that there is a trade-off between capacity invested and necessary inven-
tory. Looking at the utilisation Equation 3.1, the capacity invested deter-
mines the scheduled working time T and the minimal possible processing
time per item tP,i . The scheduled working time T depends on the number
of maximum possible working hours and the number of machines invested.
Recapitulating the long term view, the following statements are equivalent:
1. Minimising the fixed costs (depreciation of capacity invested)

2. Maximising the machine utilisation
3. Losing (minimising) the capability (less excess capacity) to meet highly
varying customer demand or increasing inventory levels to manage
demand fluctuations
3.3.2 Medium Term Level

In the medium term, capacity adjustments are made resulting from demand
fluctuations. The number of permanent workers is assumed to be constant
in the medium term. Low customer demand requires shorter working times,
more permanent workers on vacation, less leasing staff or fewer shifts. It
may also result in underutilisation of workers or excess inventory. In con-
trast, high customer demand requires introducing overtime, more leasing
staff, more shifts, selling pre-produced items on stock or failing to meet
customer demand. The main focus for the medium term decision level is on
flexibility and not on utilisation targets. For the medium term, the following
statements are equivalent:
1. Minimising the variable costs for workers (leasing workers and over-
time) by fulfilling customer requirements
2. Inventory increases or more idle time for permanent workers
The best strategy is to introduce cost neutral or nearly cost neutral

flexible working time (Kerkhofs et al., 2008). But there are legal and social
limits to fully applying this flexibility.
3.3.3 Short Term Level

Finally, the effects of short term decisions on the utilisation metrics are
discussed in this section. Fixed costs for capacity invested and costs for
workers are not addressed in the short term level, because these costs cannot
be influenced by the short term decisions. Moreover, only the immediate
effects of the parameters on the utilisation metrics are discussed and no
combinations of parameters are considered. The effects of the parameters
on costs and revenue are summarised in Table 3.3.
By increasing the parameter ”number of items produced for customer
orders” nC,i , the total variable material costs and variable processing costs
will increase. Since these products can be sold due to customer orders, the
revenues will increase immediately.
By increasing the parameter ”number of items produced on stock” nS,i
and ”number of scrapped items” nQ,i , the total variable material costs and
3.3. Discussion 33
variable processing costs are increasing. Moreover, holding costs or costs for
”rework or disposals” are increasing respectively. Items produced on stock
can be sold later as soon as customer orders arrive. Since there is a risk
for obsolescence, the increase of revenue is not guaranteed. By producing
scrap items, the risk for violating due dates is high. Therefore, a risk for a
decrease in revenue exists.
By increasing the total time needed for set-up, the total set-up costs
might increase when additional capacity is needed or when it is not possible
to produce all items needed for customer orders nC,i . By increasing the total
time needed for set-up, the risk for violating due dates is high. Therefore,
a risk for a decrease in revenue exists.
By increasing the ”additional processing time” tAP,i , the total variable
costs might increase when additional capacity is needed or when it is not
possible to produce all items needed for customer orders nC,i . By increasing
the processing time, the risk for violating due dates is high and therefore
carries a risk for a decrease in revenue.
Increasing the idle time ti means that the time for machine breakdowns,
waiting for an order or tool or worker unavailability increases. By increasing
the idle time, the risk for violating due dates is high. This results again in
a risk for a decrease in revenue.
Only the parameter ”numbers of items produced for customer orders”
nC,i leads to an increase of revenues as indicated in Table 3.3. Also the
increase of ”number of items produced on stock” nS,i accounts for potential
revenues only in future. All parameters increase the total variable costs
whereby tI and tAP,i only increase the total variable costs if additional
capacity is required. The expression ”possible decrease of revenue” in Ta-
ble 3.3 means that due to lost capacity caused by quality problems, too
long set-up times, or unavailability and/or ”additional processing times”,
customer due dates may not be met. Missed due dates cause customer dis-
satisfaction and may lead to reduced revenue. Summarising the short term
view, the following statements are equivalent:
1. Maximising profit (maximising revenue with minimal variable material,
processing, holding, rework, disposal and change over costs)
2. Minimising number of items produced on stock nS,i , number of scrapped
items nQ,i , total time needed for change-overs and idle time tI and
additional processing time tAP I,i
3. Minimising the machine utilisation to the minimum utilisation Umin
4. Maximising excess capacity
In the short term view, minimising the utilisation means producing effi-
ciently because losses according to the ”seven zeros” philosophy are avoided.
Table 3.3: Effects of parameters on total variable costs and revenue by short
term decisions
Parameter increase Effect on costs Effect on revenue

Number of items for Increase of total variable Increase of im-
customer orders, nC,i material costs and variable mediate revenue
processing costs
Number of items pro- Increase of total variable Possible increase

duced on stock, nS,i material costs, variable proof future revenue
cessing costs and holding
costs
Number of scrapped Increase of total variable Possible decrease

items, nQ,i material costs, variable proof revenue
cessing costs and costs for
rework or disposal
Total time needed for Possible increase of total Possible decrease

set-up, tS,i nO set-up costs of revenue
Idle time, tI Possible increase of total Possible decrease

variable processing costs of revenue
Additional processing Possible increase of total Possible decrease

time, tAP,i variable processing costs of revenue
Minimising utilisation is equivalent to maximising excess capacity and this

allows high fluctuating customer demands to be met with low inventory lev-
els. In addition, Goldratt’s idea of throughput in the theory of constraints
(Schragenheim and Dettmer, 2000) – that only sold capacity contributes to
the revenue – is incorporated in minimising utilisation.
Obviously, there are some controversial objectives in the long term,
medium term and short term. Fixed costs caused by long term decisions
force the system to focus on high utilisation: this is the classical view of
accounting and finance, although there is a well-known trade-off between
capacity invested and inventory needed. The medium term view tries to ad-
just the workers’ capacity to the customer demand fluctuations with lower
costs and also maximises the utilisation. The short term decisions should
try to minimise utilisation, because losses are avoided due to the ”seven
zeros” philosophy. This ensures the ability to produce directly for customer
3.4. Concluding Remarks 35
orders without any disruption (no quality problems, no machine failures,

fewer changeover times, no production on stock, no additional production
time). Furthermore, total variable costs are minimised by minimising the
short term utilisation.
Hopp and Spearman (2008) state that high utilisation causes low costs
per unit but low utilisation allows high sales. Low costs per unit based on
a high number of items which are produced on stock or are scrapped do not
contribute to profit. In addition, it is known that working at the capacity
limit (Flynn et al., 1995) causes quality problems and a very large inven-
tory (Bradley and Glynn, 2002; Jodlbauer and Altendorfer, 2010). Taking
the general objectives into account, such as maximising profit, higher sales
should be given higher priority than lower costs per unit.
Managerial Insight
Combining the JIT philosophy, utilisation concep,t and costs and revenue,
two interesting strategies arise. If a high JIT intensity is reached, this means
that the production system can follow the fluctuations of the customer de-
mand without any losses according to the ”seven zeros”. Thus it is possible
to reduce the excess machine capacity as well as the flexibility needed for
medium term capacity adjustments. A high JIT intensity allows lower ca-
pacity invested and requires less working time flexibility. Alternatively, it
is possible to increase revenues by increasing sales, whereas no additional
investment is necessary.
3.4 Concluding Remarks

Based on the ”seven zeros”, a measurement for JIT intensity with corre-
sponding managerial insights is introduced. In this chapter, it is shown that
high JIT intensity on the long term level means maximising machine utili-
sation. However, on the medium term planning level, flexibility is the key
for success. Utilisation at the short term decision level has to be minimised
to achieve the seven JIT goals called ”seven zeros”. Minimum utilisation
is reached if production is fully customer order oriented (no production on
stock) and if there are no fluctuations and disruptions in the system (no
set-up, no additional production time, no scrap or rework, no idle time).
If utilisation is very close to the minimum one, excess capacity can be re-
duced. Thus, high JIT intensity allows lower capacity invested and requires
less working time flexibility or allows potential for increasing sales without
any further investments.
Fixed costs based on long term decisions are decreased with high utilisa-
tion but there is a trade-off between capacity invested and inventory levels,
respectively between capacity invested and potential for meeting sales re-
quirements in high fluctuation markets. Total variable costs based on short
term decisions are decreased with low utilisation.
Summarising, the production view in terms of low utilisation or high
excess capacity by fulfilling the seven JIT goals means efficient production.
Chapter 4
Capacity Setting Methods
In this chapter different methods for an MTO environment for a periodi-
cal capacity setting, multi-machine production systems with stochastic cus-
tomer required lead times and stochastic processing times to improve service
level and tardiness are presented. These methods are developed as decision
support when capacity flexibility exists, such as a certain range of possible
working hours a week. The methods differ in the amount of information used
whereby all are based on the cumulated capacity demand at each machine.
In a simulation study the methods’ impact on service level and tardiness is
compared to a constant capacity provided for a single- and a multi-machine
setting. It is shown that the capacity setting methods tested can lead to an
increase in service level and a decrease in average tardiness contrasted to a
constant capacity provided. The methods using information on processing
time and customer required lead time distribution perform best. In the
end of this chapter for a simplified M/M/1 queuing system two different
production rates are introduced. The results found in this chapter can help
practitioners to make efficient use of their flexible capacity.
4.1 Introductory Remarks

High service level and low tardiness are two important objectives of produc-
tion planning and control in MTO manufacturing systems. In a stochastic
environment, practitioners deal with process and customer uncertainties to
set the capacity and to fulfil these objectives.
For the design of capacity setting methods, one key aspect is the in-
formation available about the stochastic environment of a production sys-
tem. Even if information about stochastic customer required lead time and
stochastic processing times is available, practitioners can still decide how
much of this information they are going to use for the capacity setting de-
cision. Such information can be delivered by enterprise resource planning
systems, production data acquisition systems and/or manufacturing execu-
tion systems. However, information processing to ensure its applicability
for decision models often leads to additional efforts or costs.
In order to enable a company to forecast the capacity demanded, the
capacity provided of the production system has to be set based on the ca-
pacity demanded by the customers. Such capacity adjustment is enabled
by flexible working contracts and capacity accounts which become more
and more applied in manufacturing companies. Upper and lower bounds
for the capacity provided are introduced because capacity is not arbitrar-

DOI 10.1007/978-3-658-19120-7_4
38 Chapter 4. Capacity Setting Methods
ily changeable for companies due to, for instance, laws limiting working
times, contract workers and investment/de-investment in machinery and
equipment. These tools allow companies to deal with customer demand
uncertainty by adjusting their capacities. Furthermore, capacity cannot be
adjusted instantaneously, but a periodic capacity adjustment is realistic.
This reflects the situation of a rolling horizon planning in real production
systems where capacity can, for example, be defined for the next week on a
daily basis taking into account the currently known orders.
Based on the assumption that processing time distribution and customer
required lead time distribution are known, these two sources of information
are used to develop different capacity setting methods which are tested in
a discrete event simulation study. Since many manufacturing companies
are dealing with bottlenecks in the production process, a single-machine
framework is considered in this chapter, which is finally extended to a multi-
machine setting.
In this chapter a periodical decision support for short and medium
term capacity setting to improve service level and tardiness is developed.
Stochastic customer behaviour (information uncertainty: not all informa-
tion is available when the decision is taken), the integration of stochastic
production processes (process uncertainty: the processing time of orders is
not deterministic) and the rolling horizon effects of a planning system (re-
evaluation effects: decisions taken in the past influence the current decision
space and not the full length of the decided plan) are considered by the
different methods.
The cumulated capacity concept used by Segerstedt (1996) is applied.
The M/M/1 queuing state dependent capacity adjustment models of Mincso-
vics and Dellaert (2009) and Buyukkaramikli et al. (2013) are extended by
general distributions for the arrival process and processing times. Moreover,
the capacity setting methods are extended to a multi-machine production
system. The proposed framework also covers a distribution of the customer
required lead time. Therefore, the model of Jodlbauer (2008c) is extended
by a rolling horizon and is implemented for calculating the capacity pro-
vided. A limitation of the approaches presented is that no analytical eval-
uation of their performance is possible and therefore simulation has to be
applied. After determining the capacity provided, in a next step a schedul-
ing approach could be considered which is left for further research.
4.2 Model Framework

Since information processing may lead to additional effort and costs, a set of
methods with different information levels is developed for approximating the
capacity demanded as illustrated in Figure 4.1. The matrix indicates the two
4.2. Model Framework 39
dimensions processing time distribution (proc dist) and customer required

lead time distribution (CRL dist). The decision maker can add or neglect
information of the process uncertainty or customer required lead time uncer-
tainty by using the corresponding distribution. Quadrant 1 and 3 apply no
additional information for processing time uncertainties whereby quadrants
2 and 4 consider the processing time distribution. Customer required lead
time uncertainties are implemented in quadrants 3 and 4 whereby quad-
rants 1 and 2 ignore information about the customer required lead time
distribution.
1 3
no proc dist no proc dist
no CRL dist CRL dist
2 4
proc dist proc dist
no CRL dist CRL dist
Figure 4.1: Matrix of stochastic information for demanded capacity – four meth-
ods
Figure 4.2 shows the basic model where order i with the due date Di is
stated by a customer. The customer required lead time represents the time
gap between the point of time when the order is stated and the due date of
the customer order. The production system can either fulfil the customer
order on time or the customer order is delayed. If an order is accomplished
earlier, then it is considered to be on time. This circumstance is measured
by the two metrics service level and tardiness. No lost sales are considered
in this model. If the machine is idle, the order is released into production;
otherwise, the order is waiting in the buffer for processing. Only one order
can be processed at a time. Customer required lead times and processing
times are assumed to be stochastic.
The production system is assumed to work on an MTO basis. This
means production orders are created based on already known demand and
not based on forecast. Only the capacity demand is forecasted in some of
the following methods using the distribution of customer required lead time.
Figure 4.2: Single-machine production system
4.3 Model
In Table 4.1 the variables and in Table 4.2 the parameters for the single-
machine production system are presented.
To fulfil all orders on time, the demanded cumulated capacity ãx (t) has
to be lower than or equal to the cumulated provided capacity cy t at all
points in time as illustrated in Figure 4.3. A combination of (x, y) (see
Table 4.2) is defined as periodical capacity setting method, whereby x and
y indicate the methods used for capacity demanded respectively provided.
Figure 4.3: Basic idea of capacity setting
The capacity setting is performed each Δ periods for an evaluation time

4.3. Model 41
Table 4.1: Definition of variables for single-machine model
Notation Description
Di Random variable for due date of order i
D̂i Random variable for pushed up due date of order i
Ai Random variable for demanded capacity of order i ; A0 is the
current backlog
Ã(t) Random variable for cumulated demanded capacity until
time t
Â(t) Random variable for pushed up cumulated demanded capac-
ity until time t
L Random variable for capacity weighted customer required
lead time of all orders
E[.] Expected value of a random variable
V ar[.] Variance of a random variable
FG (.) CDF (cumulated distribution function) of random variable
G
FG−1 Inverse of the CDF of random variable G
ãx (t) Variable for approximating cumulated demanded capacity at
the machine at time t based on method x
cy Variable of bounded provided capacity after each capacity
setting cycle based on method y
c̃y Variable of unbounded provided capacity after each capacity
setting cycle based on method y
c(l) Variable for capacity provided in past period l
ĉ Variable for capacity account
window of δ periods in the future as shown in Figure 4.4. Compared to

scheduling approaches or medium term production planning like MRP, δ
corresponds to the planning horizon. Hence, the capacity setting is based
on the customer orders already stated. For Δ being one week and δ being
two weeks, for example, the capacity setting would be done weekly taking
into account the known orders for the next 10 working days. Without loss
of generality, the current time of each periodical capacity setting cycle is set
to zero. The periodical capacity setting methods follow three steps:
1. Approximate the cumulated demanded capacity ãx (t) required by the

customers; four methods are presented
2. Calculate unbounded provided capacity based on the demanded ca-

pacity; 3 methods are presented
Table 4.2: Definition of parameters for single-machine model
Notation Description
x Method parameter of cumulated demanded capacity:
x = 1: Deterministic processing times; no CRL distrib.
x = 2: Processing time distribution; no CRL distrib.
x = 3: Deterministic processing times; CRL distribution
x = 4: Processing time and CRL distributions
y Method parameter of provided capacity:
y = 1: Full utilisation
y = 2: Maximum safety
y = 3: Service level target
κmin Parameters for lower bound for provided capacity cy
κmax Parameters for upper bound for provided capacity cy
κavg Parameter for average provided capacity per period
Δ Parameter for period of time for which cy is set
−ψ Parameter for time where the system has (re)started
θ Service level parameter for capacity setting method c3
β Capacity feasibility parameter for capacity setting methods
a2 and a4
OC(t) Operation characteristics of the machine until time t
δ Parameter for evaluation time window
3. Set capacity based on the capacity account
' G
-ʗ -ʗ+1 -ʗ+2 -ʗ+3 ........ 0 1 2 ........ 10
Figure 4.4: Timeline
4.3.1 Approximation of the Demanded Capacity

(Step 1)
In the following section the first step of the periodical capacity setting meth-
ods is detailed. Four methods for approximating the cumulated demanded
capacity are formally defined and illustrated in Figure 4.5.
4.3. Model 43
Figure 4.5: Cumulated demanded capacity – four methods
Deterministic Processing Times, No CRL Distribution: x = 1

Each order i requests a capacity Ai at due date Di . In this case Ai is approx-
imated with its expected value E[Ai ] without any additional information.
This means the decision maker has no information about the process and
demand uncertainty or does not want to invest additional effort into infor-
mation processing. Equation 4.1 cumulates the capacity demanded up to
time t which is illustrated in Figure 4.5.1:

ã1 (t) = E[Ai ] (4.1)
{i|Di ≤t}
Processing Time Distribution, No CRL Distribution: x = 2

Method x = 2 takes the information of the processing time distribution
into account, but uses no information about the customer required lead
time distribution to approximate a cumulated capacity demand. A higher
amount of capacity demanded results from this approximation in compar-
ison to method x = 1 as indicated in Figure 4.5.2 (ã1 (t) ≤ ã2 (t)) due to a
safety capacity for the processing time distribution.
The random variable of the cumulated capacity demanded is defined in
Equation 4.2:

Ã(t) = Ai (4.2)
{i|Di ≤t}
Note that Equation 4.2 includes A0 since D0 = 0. In Equation 4.3, the

capacity demanded to fulfil a set of orders (until time t) with probability β
(FÃ(t) (ã2 (t)) β) is calculated.
−1
ã2 (t) = FÃ(t) (β) (4.3)
Processing times may not be deterministic due to products requiring dif-
ferent amount of machine capacity, stochastic processing speed, set-up time
or down times. In this case, the realisation of the capacity demanded will
exceed the expected value E[Ai ] with a certain probability. If, for example,
β is set to 0.9, then method x = 2 includes safety capacity to fulfil the
stochastic capacity demand with a probability of 90%. This method works
for all distributions of the capacity demanded where a quantile function
exits.
The safety
capacity of method x = 2 in comparison to method x = 1 is
−1
FÃ(t) (β) − {i|Di ≤t} E[Ai ], which is the difference between the dashed and
solid line in Figure 4.5.2.
Deterministic Processing Times, CRL Distribution: x = 3

The customer required lead time distribution is used for evaluating method
x = 3, which results in an earlier allocation of the capacity demanded
compared to method x = 1 to forecast short term orders which are not yet
known. Method x = 3 still assumes deterministic processing times as x = 1.
The customer required lead time is defined as the time between the due
date of an order i and the point in time when this order is stated. At a
certain point in time the capacity demanded is defined by the customer or-
ders already stated. However, according to the customer required lead time
distribution, additional demand – defined as anticipated demand – can be
requested by customers. To include the anticipated demand, the concept
of the operation characteristics (OC) is applied, whereby OC defines the
relationship between the customer required lead time and the capacity de-
manded at a machine (Jodlbauer, 2008c). For a single-stage model, the
OC shows how much of the customer required capacity is known how many
periods in advance. Figure 4.6 shows an example of an OC where 50% of
capacity demanded is known 20 periods in advance.
In Jodlbauer (2008c), the OC concept encompasses the concept of a
constant remaining processing time. This remaining processing time can be
the transportation time to the customer in a single-stage production system
or the remaining processing and handling time in a multi-stage production
4.3. Model 45
system. Since this time is assumed to be constant in Jodlbauer (2008c), but

does not hold especially for job shop production systems, the OC concept
is slightly adapted in this chapter.
Figure 4.6: Operation characteristics
The customer required lead time is weighted by its capacity consump-

tion so that the capacity weighted customer required lead time L results
(Jodlbauer, 2008c; Jodlbauer and Altendorfer, 2010). Finally, the opera-
tion characteristics is calculated in Equation 4.4:
OC(t) = 1 − FL (t) (4.4)

Based on the operation characteristics, it is necessary to produce the

already stated orders earlier in order to have enough safety capacity left
for the anticipated future demand. Therefore, the pushed up due date
D̂i is used instead of the customer required due date Di . The following
Equation 4.5 for approximating the cumulated demanded capacity can be
stated:

ã3 (t) = E[Ai ] (4.5)
{i|D̂i ≤t}
Equation 4.6 shows how the due date of each order based on the OC
is pushed up to ensure some safety capacity for the anticipated demand
(therefore D̂i ≤ Di holds).
Di
D̂i = OC(τ )dτ (4.6)
0
In this method not a typical forecasting approach is conducted where the

customer demand of a certain product or a certain product group is forecast,
but only the capacity demand still to occur on the short and medium time
range is anticipated for a certain machine based on past data.
Processing Time and CRL Distributions: x = 4

In method x = 4, the concepts of methods x = 2 and x = 3 are combined.
Therefore, method x = 4 uses the information of the customer required lead
time distribution and the processing time distribution. Again, a certain
probability β is used to calculate the cumulated demanded capacity based
on the processing time distribution.
−1
ã4 (t) = FÂ(t) (β) (4.7)
However, the random variable of pushed up cumulated demanded ca-

pacity until time t is defined in Equation 4.8, because method x = 4 also
includes the anticipated demand, where the due date of each order is pushed
up according to Equation 4.6.

Â(t) = Ai (4.8)
{i|D̂i ≤t}
Method x = 4 pushes up the capacity demanded (based on the customer

required lead time distribution) and increases it according to the processing
time distribution in comparison to method x = 1 (see Figure 4.5.4).
4.3. Model 47
4.3.2 Setting the Unbounded Provided Capacity (Step

2)
Based on the cumulated demanded capacity, calculated with one of the
methods x = 1...4, the unbounded provided capacity is set. Three methods
for calculating the capacity provided are presented in this section and illus-
trated in Figure 4.7. A0 – the current capacity backlog of the system – is
set to zero in Figure 4.7.
Figure 4.7: Comparison of the three developed methods calculating the capacity
provided
The capacity provided for method y = 1 is set to the cumulated de-

manded capacity at the end of the evaluation time window δ. Method
y = 2 is based on the maximum increase of the cumulated demanded ca-
pacity ãx (t). Finally, method y = 3 uses a kind of service level θ to relax
the constraint of method y = 2, so that not each order has to be ready on
time.
Full Utilisation: y = 1
For this capacity setting method the capacity provided is set to the average
capacity demanded within an evaluation time window δ. Therefore, the
cumulated demanded capacity at the end of the evaluation time window
is divided by the time where the last customer order reaches its due date.
Assuming the approximated demanded capacity occurs, this method ensures

full utilisation of the machine, because the capacity provided is set to the
cumulated demanded capacity for observed evaluation time window δ. Note
that methods x = 2 to x = 4 lead to a safety capacity which will often not
be utilised and therefore the method y = 1 will not lead to a utilisation of
100% in these cases.
ãx (max{Di |Di < δ})

i
c̃1 = (4.9)
max{Di |Di < δ}
i
The maximum over all i is responsible for searching the due date of the
last order within the evaluation time window.
Maximum Safety: y = 2
Method y = 2 works as follows: A straight line starting at zero with mini-
mum slope being greater than or equal to the cumulated demanded capacity
is constructed. The fraction in Equation 4.10 can be interpreted as the slope
of cumulated capacity demand at due date Di . The max operator identifies
the maximum slope over all due dates Di leading to a maximum safety in
capacity provided. As an assumption at least one order has to be in the
system. Note that capacity backlogs are ignored since they would lead to
an infinite increase, however, they are included in ãx (Di ) when the first due
date is evaluated.

ãx (Di )
c̃2 = max (4.10)
{i|Di >0} Di
Service Level Target: y = 3

Since it could be too ambitious to set the capacity provided to the maximum
increase as calculated in Equation 4.10, Equation 4.11 presents a smoothed
version of this approach, where not the whole capacity demand has to be
finished on time. The lowest c̃3 , which still satisfies a service level θ for the
evaluation time window δ, is searched in the following optimisation problem:
c̃3 → min (4.11)

c̃3

f (Di , c̃3 )
{Di |Di ≤δ}
≥θ
1
{Di |Di ≤δ}
4.3. Model 49
The function f (.) defined in Equation 4.12 delivers 1 if a due date (one
or more orders can have the same due date) is on time and 0 if a certain
due date does not hold. c̃δ3 is the slope of the capacity provided.
c̃3
f (Di , c̃3 ) = 1 for ãx (Di ) ≤ Di (on time) (4.12)
δ
c̃3
f (Di , c̃3 ) = 0 for ãx (Di ) > Di (late)
δ
If y = 3 is applied in combination with methods x = 3 and x = 4, Di has
to be substituted by D̂i for Equations 4.11 to 4.12, because the customer
required lead time distribution is included in the approximations.
4.3.3 Capacity Account (Step 3)

The bounded provided capacity cy is introduced for the developed capacity
setting methods. If the unbounded provided capacity c̃y is below κmin then
the provided capacity is set to κmin and if the candidate exceeds κmax then
the provided capacity is set to κmax . Equation 4.13 corresponds to the
flexible capacity range that can be provided in the production system.
κmin ≤ cy ≤ κmax (4.13)

Since most flexible working time contracts have a defined average work-
ing time κavg , a capacity account ĉ is introduced and Equation 4.14 is added
to ensure that the capacity provided is on average lower than or equal to the
allowed average working time. When the capacity account ĉ is positive (e.g.
16 hours), the production system has some capacity which can be provided
above the average working time (e.g. two shifts with eight hours each on
Saturdays).
All periods from (re)start of the system are taken into consideration
which leads to Equation 4.14 with c(l) being the capacity provided in the
past period l. If the production system has collected a positive capacity
account by setting the capacity lower then κavg then the capacity can be
set to a level exceeding κavg if necessary.
cy − κavg ≤ ĉ (4.14)
−1

ĉ = (κavg − c(l))
l=−ψ
The candidate for capacity provided c˜y from methods y = 1...3 is con-
strained by Equation 4.13 and 4.14 in the following Equation 4.15.
−1

κmin ≤ cy ≤ min[κmax , κavg + (κavg − c(l))] (4.15)
l=−ψ
According to the procedure of periodical capacity setting methods, steps

1 to 3 are repeated each Δ period to implement the rolling horizon planning
effect.
4.4 Multi-Machine Concept

In this section the basic single-machine production system is extended to
a multi-machine production system. The methods can be applied to any
arbitrary production system structure according to the respective routing
file. Order i with the due date Di is stated by a customer. On each machine
j a demanded capacity Ai,j is required by order i.
To implement the methods developed for the single-machine case also
in a multi-machine setting, an order due date for each machine Di,j has
to be identified. This order due date is based on the customer order due
date applying a backward scheduling similar to the MRP algorithm (Hopp
and Spearman, 1996). The difference is MRP: For the capacity setting
methods no production plan is generated but only the capacity demand has
to be identified and therefore no or only little safety time is included in this
backward scheduling.
Based on these Di,j values, also the customer required lead times have
to be adapted. Therefore, in the multi-machine concept Di,j , D̂i,j , Lj and
OCj (t) replace Di , D̂i , L and OC(t) respectively in Equations 4.1 to 4.15,
as indicated in Table 4.3. Moreover, the demanded capacity for a machine
within the multi-machine production system needs machine index j to be
added. Therefore, Ai is replaced by Ai,j . Also the lower and upper bound
for the capacity provided κmin and κmax are defined for each machine j.
This backward scheduling is introduced in Equation 4.16:
Di,j = Di − E[Ri,j ] (4.16)

In Equation 4.17 the remaining capacity demand for order i Ri,j is de-
fined, whereby only the capacity demanded at the remaining machines in-
dicated by Ni,j for finishing an order is considered. The parameter Ni,j
represents the set of machines for remaining processing steps of order i af-
ter machine j known from the respective routing file. Assuming for the
multi-machine example presented in Section 4.5, that the current machine
is machine 3, Ni,j consists of machine 4 and 5 since these processing steps
have to be finished after processing at machine 3.
4.4. Multi-Machine Concept 51
Table 4.3: Definition of additional variables for a multi-machine model
Variable Description
Di,j Random variable for machine depended due date of order i at
machine j. Di,j on the last machine equals to Di stated by
the customer
D̂i,j Random variable for pushed up due date of order i at machine
j
Ri,j Random variable of remaining capacity demand for order i
after machine j
Ni,j Set of machines for remaining processing steps of order i after
machine j
Ai,j Random variable for demanded capacity of order i at machine
j
Lj Random variable for capacity weighted customer required
lead time at machine j
OCj (t) Operation characteristics of machine j until time t

Ri,j = Ai,k (4.17)
{k∈Ni,j }
For methods x = 2 and x = 4 two alternatives x = 2b and x = 4b

including the stochastic behaviour of remaining capacity demand are de-
veloped. Hence, the existing methods x = 2 and x = 4 are defined in the
multi-machine production system as x = 2a and x = 4a.
In Equation 4.18 the CDF of Ri,j is used to calculate R̃i,j based on the
distribution of the remaining capacity demand. Thus, E[Ri,j ] is replaced in
Equation 4.17 by R̃i,j .
R̃i,j = FR−1
i,j
(β) (4.18)
Comparing the sum over the means of the capacity provided in Equa-
tion 4.17 with the inverse of the CDF of the processing time distribution
for the remaining processing steps of an order in Equation 4.18 leads to
a difference which is equivalent to the safety time concept also applied in
MRP. This safety time accounts for the possibility that a certain order has
a higher processing time at a certain machine than planned. However, the
approach presented in Equation 4.18 does not consider the queue length
as discussed for example in the Workload Control (Bechte, 1988) literature
where also the queuing between the processing steps is considered. Com-
pared to the safety time concept in MRP, which also includes some safety
time for queuing, this safety time only includes the processing instabilities.
This is assumed since only the workloads and their latest possible date Di,j
for each single machine are needed but no production plan with start dates
is generated.
Lj is calculated in Equation 4.19 based on L. For simplification reasons
in the calculation of Lj , all Ai,j are assumed to have a common stochastic
distribution and therefore E[Rj ] = E[Ri,j ] holds.
Lj = L − E[Rj ] (4.19)
Note that this assumption only influences the OCj (t) applied in the
methods x = 3, x = 4a and x = 4b.
4.5 Simulation Study

The simulation study is conducted for comparing the behaviour of the pre-
sented methods to analyse which information improves the capacity set-
ting methods. In this simulation study 19 500 simulation runs for the
single-machine production system and 28 500 simulation runs for the multi-
machine production system have been conducted as shown in Table 4.4.
Table 4.4: Simulation runs
Description Multi-machine Single-machine

planned utilisation values 15 15
scenarios 10 10
methods for demanded c.∗ 6a 4b
methods for provided c.∗ 3c 3c
replications 10 10
constant capacity 1 1
simulation runs 28 500 19 500
a
x = 1, 2a, 2b, 3, 4a, 4b
b
x = 1...4
c
y = 1...3
*
Capacity
Five sequential machines are investigated for the multi-machine produc-

tion system as illustrated in Figure 4.8. The simulation framework used
is explained in more detail in Hübl et al. (2011). The simulation model
has been validated as proposed in Kleijnen (1995) by checking the model
behaviour of the single-machine case with constant capacity compared to
developed analytic results.
4.5. Simulation Study 53
Figure 4.8: Multi-machine production system
The planned utilisation is set to the same level for all machines. The
following values for planned utilisation are tested: 0.7, 0.75, 0.8, 0.85, 0.875,
0.9, 0.91, 0.92, 0.93, 0.94, 0.95, 0.96, 0.97, 0.98, 0.99.
Ten different scenarios for single- and multi-machine settings are tested
where parameters for processing time, customer required lead time, upper
and lower bound for provided capacity and period for setting the provided
capacity are predefined as shown in Table 4.5. A basic scenario is defined as
the basis for all other scenarios with a variation coefficient for the processing
time distribution of 1, mean customer required lead time of 20 time periods
for the single-machine production system and 100 time units for the multi-
machine production system. The upper and lower bound for the capacity
provided is given by +/- 20% and Δ is set to 10 periods.
Table 4.5: Scenarios
V ar[Ai ]
Description E[Ai ] E[L] κmin , κmax Δ
basic 1 20/100 0.8, 1.2 10
high processing time variation 1.5 20/100 0.8, 1.2 10
low processing time variation 0.5 20/100 0.8, 1.2 10
high customer required lead time 1 50/250 0.8, 1.2 10
low customer required lead time 1 10/50 0.8, 1.2 10
high flexibility 1 20/100 0.5, 1.5 10
low flexibility 1 20/100 0.95, 1.05 10
indefinite flexibility 1 20/100 0, ∞ 10
long capacity setting period 1 20/100 0.8, 1.2 50
short capacity setting period 1 20/100 0.8, 1.2 5
The periodical capacity setting methods are compared to a scenario

where the production system provides constant capacity. The average of

the capacity provided is equal for all tested scenarios, also for the constant
provided capacity scenario. In the single-machine production system, the
four methods x = 1...4 for identifying the capacity demanded are compared.
For the multi-machine production system, the additional two methods for
calculating the capacity demanded due to the machine depended due date
are included (x = 2b and x = 4b). Therefore, six methods are compared in
the multi-machine production system as shown in Table 4.4. Ten replica-
tions are produced for each parameter combination.
The following values are set: κavg = 1, δ = 40 for single-machine pro-

duction system, δ = 200 for multi-machine production system, θ = 0.9 and
β = 0.9. The parameters δ, β and θ have been identified in preliminary
studies to perform well with respect to service level and tardiness for the
basic scenario at 80% utilisation. The simulation runs for 500 000 time
periods, a warm-up period of 100 000 time units is excluded from the re-
sults measured and the system is restarted every 100 000 time units for data
generation reasons.
Processing times are log-normal distributed and the distribution of the

customer required lead time is implemented as exponential distribution for
the simulation experiment. The distributions are chosen for non-negativity.
For simplicity reasons, the distribution of Â(t) is assumed to be log-normal
as well. The mean customer required lead time is differently assigned for the
single-machine and the sequential production system. In the third column
of Table 4.5, the first number is dedicated to the single-machine system
and the second number to the sequential production system. The mean
processing time per order is one period.
The interarrival time of the customer orders, which is used to influence

the planned utilisation, is also exponentially distributed. No set-up times
and no transportation times are included in the simulation.
The capacity setting methods for the capacity provided are implemented
in the simulation model by changing the machine speed. Whenever the
average machine speed (which is equivalent to the capacity account) is below
or equal to 1, the provided capacity can be set – if necessary – higher than
1. The average machine speed is cleared at each system restart. Relating
the simulation periods to one shift of a working day and assuming that two
shifts per day make the working time for five days a week, then the capacity
is set every week (Δ = 10 shifts). The capacity account in the beginning is
0. The simulation is conducted in AnyLogic 6.4.1.
4.6. Results 55
4.6 Results
In the first step, the best capacity setting method (x, y) for the basic sce-
nario is evaluated. Based on this evaluation, the results of the best method
(x, y) are compared to the situation with constant provided capacity for the
two production systems. This is followed by an analysis over all scenarios
defined. For all investigations, the performance measures service level in-
crease and tardiness decrease in comparison to average capacity provided
are used.
4.6.1 Evaluation of Best Method Combination

Table 4.6 compares the best method combination (x, y) to constant provided
capacity for 15 different planned utilisation values. Tardiness and service
level are treated separately for identifying the best method combination.
For the single-machine case the results for the planned utilisation of
94% indicates an increase of 23% in the service level by the use of method
(4, 3). At the same planned utilisation value, a decrease of 91% in average
tardiness by the use of method (3, 3) occurs. The results for the single-
machine production system show that for most of the planned utilisation
values tested, the method (4, 3) leads to the best results. For the multi-
machine case the result from Table 4.6 shows a balance between methods
(4b, 2) and (4b, 3).
As a measure for evaluating which method is the best for the two pro-
duction systems, the percentage of simulation experiments dominated by a
certain method is used, whereby only the average of a replication set is con-
sidered. Therefore, Table 4.7 summarises the percentage of cases where the
tested methods lead to the best results in service level increase and average
tardiness decrease when all ten scenarios are taken into account. In the
multi-machine production system method x = 4b leads to the best result
concerning service level increase in 20% of all parameter combinations.
The first result found in this study is that the same methods perform
well concerning service level increase and average tardiness decrease. In
the single-machine case method x = 4 for demanded capacity, which uses
both customer required lead time and processing time distribution, leads
to better results for service level and tardiness in about 60% of the cases.
Only in about 8% of the cases does the constant provided capacity scenario
lead to better results, which are discussed in the next subsection. The best
method for provided capacity calculation is in the single-machine case y = 3,
which leads to the best result over all scenarios in approximately 50% of
the cases.
For the multi-machine production system, the best result for calculating
the demanded capacity is again found with both versions of method x = 4,
Table 4.6: Service level increase and tardiness decrease potential in basic scenario
U Single-machine case Multi-machine case

Service level Tardiness Service level Tardiness
inc. (x, y) dec. (x, y) inc. (x, y) dec. (x, y)
70% 1% (2, 3) 18% (2, 3) 0% (2a, 3) 9% (4b, 3)
75% 1% (2, 3) 23% (2, 3) 0% (2b, 3) 11% (4b, 2)
80% 1% (2, 3) 28% (4, 3) 1% (2a, 3) 13% (4b, 2)
85% 3% (4, 3) 58% (4, 3) 1% (4b, 2) 15% (4b, 2)
87.5% 6% (4, 3) 72% (4, 3) 2% (3, 2) 16% (3, 2)
90% 10% (4, 3) 82% (4, 3) 3% (4b, 2) 17% (4b, 2)
91% 12% (4, 3) 85% (4, 3) 3% (4b, 2) 17% (3, 3)
92% 15% (4, 3) 88% (4, 3) 4% (4a, 2) 19% (4a, 2)
93% 18% (4, 3) 90% (4, 3) 4% (4a, 3) 20% (4a, 3)
94% 23% (4, 3) 91% (3, 3) 4% (4b, 2) 19% (4b, 2)
95% 27% (4, 3) 94% (4, 2) 5% (3, 3) 19% (4a, 2)
96% 32% (4, 2) 93% (3, 3) 4% (4a, 2) 17% (4a, 2)
97% 37% (3, 3) 95% (4, 1) 5% (3, 3) 20% (3, 3)
98% 43% (4, 1) 94% (4, 1) 2% (1, 3) 14% (1, 3)
99% 44% (4, 1) 85% (3, 1) 0% (1, 3) 16% (1, 3)
x = 1: det proc times, no CRL dist; x = 2: proc time dist, no CRL dist;
x = 3: det proc times, CRL dist; x = 4: proc time dist, CRL dist;
y = 1: Full utilisation; y = 2: Max. safety; y = 3: Service level target;
whereby a slight advantage for alternative x = 4b is identified. Again, this

means the best method uses all the information provided about customer
required lead time distribution and processing time distribution. Due to the
tiny difference in the results outcome between method x = 4a and x = 4b
it is not possible to clearly specify whether it is better to base the machine
dependent due date calculation on expected values for processing times or on
the processing time distribution. In the multi-machine production system
y = 2 is the best method for defining the provided capacity.
4.6.2 Result Comparison Best Method with

Constant Provided Capacity Scenario
Figure 4.9 shows a comparison of the results gained with constant capacity
scenario and method (4, 3) for single-machine production system and (4b, 2)
for multi-machine production system. The basic scenario is compared with
both situations presented in Figure 4.9.
4.6. Results 57
Table 4.7: Comparison of methods over all scenarios
Single-machine case Multi-machine case

Method [%]∗ [%]∗ Method [%]∗ [%]∗
Const. 9.3 8.0 Const. 14.0 9.3
x=1 0.0 2.0 x=1 14.7 17.3
x=2 16.7 14.7 x = 2a 8.7 4.7
x=3 8.0 16.0 x = 2b 12.0 7.3
x=4 66.0 59.3 x=3 14.0 15.3
y=1 24.7 31.3 x = 4a 16.7 20.0
y=2 14.0 14.0 x = 4b 20.0 26.0
y=3 52.0 46.7 y=1 0.7 2.7
y=2 50.7 58.0
y=3 34.7 30.0
*
Percentage of all cases tested
x = 1: det proc times, no CRL dist; x = 2: proc time dist, no CRL dist;
x = 3: det proc times, CRL dist; x = 4: proc time dist, CRL dist;
y = 1: Full utilisation; y = 2: Maximum safety; y = 3: Service level
target;
In the single-machine production system, the service level is on average

14% higher and the tardiness on average 60.1% lower when method (4, 3) is
compared to the constant provided capacity scenario. For the multi-machine
production system a service level increase of 2.2% and a tardiness decrease
of 13.7% has been found in the simulation study. This result shows that
flexible capacities, which are provided for medium term capacity setting,
can lead to significant improvements in service level and tardiness when the
capacity setting methods developed are applied.
For the single- and multi-machine production system, the highest im-
provement potential has been identified between a planned utilisation of
85% and 97%, which is also the range within a lot of production systems
are operating. For low planned utilisation values the developed methods do
not improve the service level and tardiness performance. Nevertheless, in
these situations the methods can still be used to decrease the overall pro-
vided capacity in the production system. The detailed simulation results
for such low utilisation values show that the capacity provided on average
is below 1 with the best methods – (4, 3) and (4b, 2) – applied. In this case
the methods developed may lead to lower capacity costs.
Another finding is that the potential to improve service level and tardi-
ness is lower for the multi-machine production system than for the single-
x = 4: Proc time dist, CRL dist; y = 2: Maximum safety; y = 3: Service

level target
Figure 4.9: Comparison best capacity setting method with average capacity
machine production system. This circumstance may result from the higher
complexity and higher uncertainty of the multi-machine production system
based on the stochastic compensation effect.
4.6.3 Scenario Discussion of the Best Methods

Table 4.8 gives an overview of the average service level increase and tar-
diness decrease in the different scenarios for the best methods identified.
In the single-machine basic scenario, for example, the average service level
increases by 14% when applying method (4, 3) instead of constant provided
capacity scenario.
The best method for estimating the capacity demanded uses both, in-
formation about processing time stochastics as well as customer demand.
This allows the production system to introduce some safety capacity due
to process and demand uncertainties. These methods providing safety ca-
pacity are also preferred for calculating the provided capacity. Therefore,
maximum safety and service level target both perform well.
Another interesting finding is that the best methods found in this study
lead in some of the scenarios to a worse overall performance in comparison to
constant provided capacity. For the ”long capacity setting period” scenario
4.6. Results 59
Table 4.8: Comparison of scenarios
Scenario Single-machine case∗ Multi-machine case∗∗

inc. [%] dec. [%] inc. [%] dec. [%]
basic 14.0 60.1 2.2 13.7
high processing 13.9 56.4 2.5 13.2
time variation
low processing 12.0 50.9 0.5 4.3
time variation
high CRL 8.2 43.4 1.7 12.4
low CRL 7.3 33.6 1.8 10.5
high flexibility 7.8 36.1 0.9 6.3
low flexibility 5.9 36.9 1.2 8.1
indefinite flexi- -7.6 -62.6 -6.9 -57.0
bility
long capacity 3.4 17.8 -0.5 -2.7
setting period
short capacity 16.3 65.8 2.6 16.7
setting period
*
Best method: (4, 3); x = 4: Proc time dist, CRL dist; y = 3: SL target
**
Best method: (4b, 2): x = 4: Proc time dist, CRL dist; y = 2: Max.
safety
(see Table 4.5), the reason for the low performance in the multi-machine
production system is conjectured to be an increasing demand uncertainty
with long periods Δ, for which capacity has to be set. Taking the example
introduced above into account, production systems with Δ1 waste capacity
of area 1 if no additional orders arrive until the next capacity setting. The
production system with Δ2 , however, wastes capacity of area 1 and 2 if no
orders arrive until the next capacity setting. Since it is necessary to fulfil
Equation 4.15 too much wasted capacity in some periods leads to a reduced
speed of the equipment in later periods.
Figure 4.10 illustrates an example of two production systems at a cer-
tain point in time of the timeline (see also Figure 4.4). In both cases the
production system is capable to satisfy cy . Hence, they differ from capacity
setting period Δ, whereby Δ1 assumes a short capacity setting period and
Δ2 a long capacity setting period.
A further result based on Table 4.8 is that for any change in capacity
flexibility the performance on both scenario types is worse than in the basic
scenario. A too high range of capacity flexibility can lead to overreactions
' '
Figure 4.10: Long capacity setting period and capacity flexibility
in many situations (see high and indefinite flexibility scenarios). If a back-

log exists or the customer requests a high amount of capacity, the capacity
provided is set to the upper bound κmax as demonstrated in Figure 4.10.
Again, if in some periods the capacity is set to its maximum amount then it
is only possible to use the lower bound κmin (according to Equation 4.14)
in later periods. However, a too low capacity flexibility allows the meth-
ods developed only a narrow range to react on the customer demand. For
practical application of these methods the capacity flexibility used has to be
evaluated to determine whether it still increases performance or if it already
leads to unnecessary overreactions at single demand peaks.
4.7 Extension: Capacity Setting for

M/M/1 Queue
In this section a similar but simplified problem as previously in this chapter
stated is observed. A simple M/M/1 queue is assumed where two different
production rates are possible. For solving this problem queuing theory – in-
stead of simulation – is applied for a M/M/1 queue with a WIP and FGI (see
Figure 4.11). In literature only a constant production rate μ is used (Medhi,
1991; Buzacott and Shanthikumar, 1993; Altendorfer and Jodlbauer, 2011;
Altendorfer and Minner, 2011). Therefore, literature is extended by apply-
ing two different production rates μL and μH . Altendorfer and Jodlbauer
(2011) demonstrated how a customer required lead time distribution can be
4.7. Extension 61
implemented in an M/M/1 queuing model with constant μ for service level,

FGI, FGI lead time and tardiness.
Figure 4.11 shows the basic model where order i with the due date Di
is stated by a customer with rate λ. The random variable of the customer
required lead time L represents the time gap between the point of time when
the order is stated and the due date of the customer order. Let fL (.) denote
the Probability Distribution Function (PDF) of the customer required lead
time, FL (.) the respective Cumulative Distribution Function (CDF), and
1/β its mean. The customer orders are released to the production system
whenever their remaining time to the due date is smaller or equal to the
planned lead time X.
Figure 4.11: M/M/1 production system with WIP and FGI
The production system can either fulfil the customer order on time or
the customer order is delayed. If an order is accomplished earlier, then it
is considered to be on time and stored in buffer FGI. This circumstance is
measured by the two metrics service level s and tardiness C. If the machine
is idle, the order is released into production; otherwise, the order is waiting
in the buffer for processing. Only one order can be processed at a time. The
production unit consists of a buffer for waiting orders and a processing step
that requires a random processing time. The random variable of production
lead time needed for one order from release until its completion is denoted
by W , and fW (.) denotes the PDF of the production lead time after release
to completion.
Whenever more than k orders are in the WIP of the production system,
it runs at speed μH ; otherwise, μL is used. Let Y denote the random WIP
and let G denote the random finished goods inventory. Backorders B are
charged when orders are late.
4.7.1 Assumptions
The following assumptions are taken into consideration to create the model:
• M/M/1 model is assumed with two levels of μ → {μL , μH }
• Processing times are exponentially distributed and only one order can
be processed at time. Whenever more than k orders are in the WIP,
μH , otherwise μL is used as production rate and no transient be-
haviour at the switching point is assumed
• Customer orders for single items are stated piecewise from many differ-
ent customers which supports the exponentially distributed customer
order interarrival times with rate λ for single items of order i
• Customer required lead time L is exponentially distribut-ed with its
parameter β
• Customer orders are not changed once they are stated and have be
accepted and released into the production system
• Customer required lead time cannot be influenced by the production
system
• Nothing is produced without a customer order (MTO system)
• FIFO is applied as dispatching rule after the customer orders are re-
leased
• Finished items are delivered to the FGI after processing
• Machine capacity cannot be stored
4.7.2 Model Development

As described above, a simplified M/M/1 production system is applied in
this section and whenever the queue length is more than k, the system runs
at speed μH , otherwise at speed μL . For k = 2 the following derivation of
the state probabilities Pn of the Markov chain in Equations 4.20-4.23 holds:
λ
P0 λ = P1 μL ⇒ P1 = P0 (4.20)
μL
P1 λ + P1 μL = P2 μL + P0 λ ⇒ P1 λ = P2 μL
2
λ λ
⇒ P2 = P1 = P0 (4.21)
μL μL
P2 λ + P2 μL = P3 μH + P1 λ ⇒ P2 λ = P3 μH
2
λ λ λ
⇒ P3 = P2 = P0 (4.22)
μH μH μL
2 2
λ λ λ
⇒ P4 = P3 = P0 (4.23)
μH μL μH
4.7. Extension 63
Which can be generalised to:

min(k,n) max(0,n−k)
λ λ
Pn = P 0 (4.24)
μL μH
Based on the development in Equation 4.24 the state probability P0 can
be calculated as (derivation see Appendix Equation 4.40):
∞
∞
λ λ
Pn = 1 ⇔ P0 =1 (4.25)
n=0 n=0
μL μH
μkL (μL − λ) (μH − λ)
⇔ P0 = (4.26)
μk+1
L (μH − λ) − λk+1 (μH − μL )
After determining the state probability P0 , the simplified state proba-

bilities for Pn = P0 μλL λ
μH where n ≤ k and n > k
orders are in the production system, lead to the Equations 4.27 and 4.28
respectively:
for n ≤k
n
λ
⇒ Pn (n ≤ k) =P0
μL
n
μL k (μL − λ) (μH − λ) λ
=
μL k+1 (μH − λ) − λk+1 (μH − μL ) μL
μL k−n λn (μL − λ) (μH − λ)
= (4.27)
μL k+1 (μH − λ) − λk+1 (μH − μL )
for n >k
k n−k
λ λ
⇒ Pn (n > k) =P0
μL μH
μL k (μL − λ) (μH − λ)
=
μL k+1 (μH − λ) − λk+1 (μH − μL )
k n−k
λ λ
μL μH
λn (μL − λ) (μH − λ)
= n−k
μH (μL k+1 (μH − λ) − λk+1 (μH − μL ))
n
λ μH k (μL − λ) (μH − λ)
= (4.28)
μH μL k+1 (μH − λ) − λk+1 (μH − μL )
Equations 4.27-4.28 can be applied to identify PH , the percentage of

time the production system is running at flexible overcapacity μH :

k
k n
λ
PH = 1 − Pn = 1 − P0
n=0 n=0
μL

μL k+1 − λ k+1
= 1 − P0
μL k (μL − λ)
k+1
μL k (μL − λ) (μH − λ) μL − λk+1

=1−
μL k+1 (μH − λ) − λk+1 (μH − μL ) μL k (μL − λ)

(μH − λ) μL k+1 − λk+1

=1−
μL k+1 (μH − λ) − λk+1 (μH − μL )
λk+1 (μL − λ)
= (4.29)
μL k+1 (μH − λ) − λk+1 (μH − μL )
4.7.3 Production Lead Time Approximation

As the PDF and the CDF of the production lead time cannot exactly be eval-
1

∞
uated, the following approximation is applied: fW (t) = 1−P 0
Pn fWn (t).
n=1
This approximation leads to the exact result for k = 0 and k = ∞, however,
it neglects some of the transient behaviour when switching between the
processing rates. Odoni and Roth (1983) give an overview of the transient
behaviour in queuing networks. Missbauer (2002) identifies the hurdles of
the transient behaviour of queuing networks for production units and post-
pones it to further research.
The PDF of the production lead time (waiting and queuing) for each
μn n−1
single queue is Erlang distributed fW(n|n≤k ) (t) = e−μL t (n−1)!
Lt
and
μn tn−1
fW(n|n>k ) (t) = e−μH t (n−1)!
H
for low and high processing rate respectively
(see Appendix Equations 4.41-4.42):
1
fW (t) ≈
1 − P0
k ∞

Pn (n ≤ k) fWn (t) + Pn (n > k) fWn (t) (4.30)
n=1 n=k+1
(μL − λ) (μH − λ)
=
μkL (μH − λ) − λk (μH − μL )

−(μL −λ)t k Γ (k, λt) −(μH −λ)t k Γ (k, λt)
e μL +e μH 1 − (4.31)
Γ (k) Γ (k)
4.7. Extension 65
Where the gamma function Γ(z) satisfies

∞
Γ(z) = t(z−1) e−t dt
0
and the incomplete gamma function satisfies
∞
Γ(a, z) = t(a−1) e−t dt.
z
The expected production lead time W̃ based on this PDF approximation
is calculated in Equation 4.32. The derivation can be found in the Appendix
Equation 4.43.
∞
E W̃ = fW (τ ) τ dτ
0
(μL − λ) (μH − λ)
=
μL k (μH − λ) − λk (μH − μL )

μ L k − λk λk k
−
(μL − λ)
2 μL (μL − λ)

λk λk k
+ + (4.32)
(μH − λ)
2 μH (μH − λ)
The exact value of the expected WIP (where N is a sufficiently high
value to reach the desired precision) can be calculated by summing up all
probability states available based on Medhi (1991):
∞

N
E [Y ] = nPn ≈ nPn (4.33)
n=0 n=0
An evaluation of the approximation error from the approximated PDF
is possible
as shown
in Figure 4.12, applying Little’s Law to Equation 4.32
for E Ỹ = λE W̃ .
The results from Figure 4.12 show that the expected WIP has an error
of up to 7%, however, the average error is rather small and therefore this
approximation is applied in the remainder. μ presents the amount of the
flexible capacity μH = μL + Δμ.
Based on Equation 4.32 also the CDF of the approximated production
lead time can be calculated as:
τ
FW (τ ) = fW (t) dt = νL ϑL (τ ) − νH ϑH (τ ) (4.34)
0
˜
=1, L =0.9 ˜
=1, L =1.1
(EY-EY)/EY (EY-EY)/EY
0.10 0.10
0.05 0.05
0.00 0.00
0.2 0.4 0.6 0.8 1.0 0.2 0.4 0.6 0.8
k=0 k=0
-0.05 k=1 -0.05 k=1
k=5 k=5
k=20 k=20
-0.10 -0.10
(a) Low μL scenario (b) High μL scenario
Figure 4.12: Approximation error of the PDF of the production lead time
The derivation of the CDF of the approximated production lead time

and the auxiliary functions νL , νH , ϑL , ϑH are presented in the Appendix
Equations 4.44-4.46. Moreover, Equation 4.34 is the basis for calculating
the expected FGI and backorders in the next section.
4.7.4 FGI and Backorder Calculation

Altendorfer and Jodlbauer (2011) present a general form of the equation of
FGI lead time and tardiness for a single-stage production system, whereby
a production lead time and customer required lead time distribution and a
WAW work release policy is applied. Altendorfer and Minner (2011) extend
the single-stage production system to a two-stage production system. In
a WAW work release policy, customer orders are released to production
whenever their remaining time to the due date is smaller as or equal to of
the overall planned lead time (work-ahead-window) X.
Based on the findings in Altendorfer and Minner (2011), studying a
single- and a two-stage production system with random due dates and a
planned lead time, which is equivalent to the WAW in this section, the
expected FGI can be stated as:
X
E [G] = λ FW (τ ) (1 − FL (τ )) dτ (4.35)
0
Applying Equation 4.34 and an exponentially distributed customer lead

time provides (derivation see Appendix Equations 4.47-4.9):
4.7. Extension 67

νL λk+1 μk−1 Γ (k, X (μL + β))
E [G] = L
k
1−
(μL + β) Γ (k)

k −(μL +β)X k
νL λμL e (Xλ)
− + e−βX
μL − λ + β Γ (k) X (μL + β)

1 − e−βX (μL − λ) λk Γ (k − 1, X (μL + β))
− − 1 −
β (μL + β)
k Γ (k − 1)

νL λk+1 e−(μL +β)X (XμL )
k−1
Γ (k, Xλ)
−e−(μL −λ+β)X +
Γ (k) β Γ (k − 1) X (μL + β)

Γ (k − 1, XμL )
+e−βX 1 −
Γ (k − 1)

μLk−1
Γ (k − 2, X (μL + β))
− 1−
(μL + β)
k−1 Γ (k − 2)

e−(μH +β)X (Xλ)
k
νH λμkH
+
μH − λ + β Γ (k) X (μH + β)

λk Γ (k − 1, X (μH + β))
− 1−
(μH + β)
k Γ (k − 1)

Γ (k, Xλ)
+e−(μH −λ+β)X 1 −
Γ (k)

k+1 −(μH +β)X k−1
νH λ e (XμH )
−
β Γ (k − 1) X (μH + β)
k−1
βμH Γ (k, X (μH + β))
+ k
1 −
(μH + β) Γ (k)

Γ (k − 1, XμH )
+e−βX 1 −
Γ (k − 1)

μHk−1
Γ (k − 2, X (μH + β))
− 1− (4.36)
(μH + β)
k−1 Γ (k − 2)
Note that based on this solution for E[G] (see last term of Equation 4.36)
only k values greater than 2 can be studied, which means the flexible ca-
pacity cannot be applied when 1, 2 or 3 pieces are in the system.
For the expected backorders, Altendorfer and Minner (2011) provide

the following statement which can also be applied in this section. Using the
exponential customer lead time and the approximated production lead time
it follows:
X
E [B] = E [G] + E Ỹ − λ (1 − FL (τ )) dτ
0
1 − e−βX
= E [G] + E Ỹ − λ (4.37)
β
4.7.5 Service Level Calculation

The general equation for service level from Altendorfer and Minner (2011)
can also be used:
X ∞
s= FW (τ ) fL (τ ) dτ + FW (X) fL (τ ) dτ
0 X
X
= FW (τ ) fL (τ ) dτ + FW (X) (1 − FL (τ )) (4.38)
0
Applying an exponential distributed CRL time and Equation 4.34, Equa-

tion 4.38 leads to (derivation see Appendix Equations 4.51-4.52):

λk μk−1
L β Γ (k, X (μL + β))
s =νL k
1−
(β + μL ) Γ (k)

λk μkL β e−(μL +β)X X k−1 1
− + k
μL − λ + β Γ (k) (μL + β) λ

1 Γ (k − 1, X (μL + β))
− 1− + μkL
(μL + β)
k−1 Γ (k − 1)
−(μL +β)X k−1
k k−1 e X e−(μL +β)X X k−2
+ λ μL +
Γ (k) Γ (k − 1) (μL + β)

1 Γ (k − 2, X (μL + β))
− 1−
(μL + β)
k−1 Γ (k − 2)
4.7. Extension 69

μkL (μL − λ) e−(μL −λ+β)X Γ (k, Xλ)
−
μL − λ + β Γ (k)

k k−1
λ μH β Γ (k, X (μH + β))
− νH k
1 −
(μH + β) Γ (k)

λk μkH β e−(μH +β)X X k−1 e−(μH −λ+β)X
− +
μH − λ + β Γ (k) (μH + β) λk

1 Γ (k − 1, X (μH + β))
− 1−
(μH + β)
k Γ (k − 1)
−(μH +β)X k−2
e X e−(μH +β)X X k−1
+ λk μk−1 +
H
Γ (k − 1) (μH + β) Γ (k)

1 Γ (k − 2, X (μH + β))
− 1−
(μH + β)
k−1 Γ (k − 2)

μkH e−(μH −λ+β)X (μH − λ) Γ (k, Xλ)
+ μkH e−(μH −λ+β)X − (4.39)
μH − λ + β Γ (k)
4.7.6 Numerical Examples

To gain more insight in the equations developed numerical examples are
generated. The influence of the capacity gap between k and Δμ on the
logistic key performance indicators service level, backorders and FGI lead
time are discussed. In these examples a symmetric change of μL and μH
based on the parameter Δμ is assumed so that μL = 1− Δμ 2 and μH = 1+ 2
Δμ
holds true. λ (arrival rate) is set to 1 and β (parameter for customer required
lead time distribution) is set to 0.05 respectively. Backorders are charged a
backorder cost b = 20 currency units per unit of time an order is late. The
holdings costs are set to h = 0.5 for WIP and f = 1 for FGI per order per
time unit.
For the following investigations two scenarios for the switching point
k = {5, 10} are analysed and compared to a constant provided capacity
situation, whereby the optimal WAW X based on Altendorfer and Minner
(2011) is applied. The results are presented in Figure 4.13. Each of these
charts shows the influence on the particular logistic key figure when (I)
flexible capacity is used, (II) maximum capacity μH is always applied and
(III) a constant capacity situation where the average of the flexible capacity
situation is provided.
It is desirable that the results of the flexible capacity setting are close the
maximum capacity provided situation. By increasing k the gap increases
but the flexible capacity situation has in most cases a better performance
than the constant provided capacity scenario.
When the gap between μL and μH is too intense (key performance in-
dicators service level decreases, backorders and FGI increases respectively),
then overreactions occur by providing too much capacity as explained in
Figure 4.10 for capacity flexibility.
In further reserach the developed equations in this section are used to
optimise (minimise costs) the production system parameters μL , μH , k and
X.

In this chapter, capacity setting methods have been developed to improve
service level and tardiness. Information about process uncertainty and/or
customer behaviour including the rolling horizon effects of a planning sys-
tem are implemented in different methods. The results from a simulation
study indicate that the methods (4, 3) and (4b, 2), which use information
about both the processing time distribution and the customer required lead
time distribution, are most efficient. In the single-machine production sys-
tem an average service level increase (over all experiments for the basic
scenario) of 14% and a tardiness reduction of 60.1% have been reached in
comparison to a constant provided capacity. For the multi-machine produc-
tion system the average service level increase is still 2.2% and the average
tardiness decrease is 13.7%. Especially in the utilisation range between 85%
and 97% the methods developed lead to good results. The same methods
perform well concerning the two metrics service level and tardiness. The
scenarios tested show that too much flexibility in capacity provided leads
to overreactions and due to that to an overall lower performance increase
or even to a performance decrease in comparison to the constant provided
capacity. Based on the equations developed in this chapter the best method
can directly be used for practical application to improve logistical perfor-
mance. Decision makers can choose which method to apply based on the
data available, production system complexity and the software available.
Moreover, this chapter provides equations and approximations for the
production lead time and the metrics for expected production lead time,
expected WIP, expected backorders and service level. The numerical study
supports the effect of overreactions when providing too much flexibility.
=1, average =1, k=5, b=20, f=1, h=0.5, =0.05 =1, average =1, k=10, b=20, f=1, h=0.5, =0.05
SL SL
1.0 1.0
0.8 0.8
0.6 Constant capacity 0.6

Flexible capacity
Maximum capacity
0.4 0.4 Constant capacity
Flexible capacity
0.2 0.2 Maximum capacity
0.0 0.0
0.0 0.2 0.4 0.6 0.8 1.0 0.0 0.2 0.4 0.6 0.8 1.0
(a) Service level k = 5 (b) Service level k = 10
E[B] E[B]
50 50
40 40
30 30
20 Constant capacity 20 Constant capacity

Flexible capacity Flexible capacity
10 Maximum capacity 10 Maximum capacity
0 0
0.0 0.2 0.4 0.6 0.8 1.0 0.0 0.2 0.4 0.6 0.8 1.0
(c) Backorders k = 5 (d) Backorders k = 10
E[G] E[G]
10 10
Constant capacity
8 8 Flexible capacity
Maximum capacity
6 6
4 4
Constant capacity
2 Flexible capacity 2
Maximum capacity
0 0
0.0 0.2 0.4 0.6 0.8 1.0 0.0 0.2 0.4 0.6 0.8 1.0
(e) FGI k = 5 (f) FGI k = 10
Figure 4.13: Logistic key figures

4.9 Appendix Capacity Setting Methods

The following equations show the derivation of the state probability P0 :
(4.40)
∞

Pn =1
n=0
∞ min(k,n)
max(0,n−k)
λ λ
⇔ P0 =1
n=0
μL μH
λ
k n k ∞ n−k
λ λ
⇔P0 + P0 =1
n=0
μL μL μH
n=k+1
k n k ∞ n
λ λ λ λ
⇔P0 + P0 =1
n=0
μL μL μH n=0 μH
λ
k n k
λ λ μH
⇔P0 + P0 =1
n=0
μ L μ L μ H μ H −λ
k
λ μλL − μL k
λ λ
⇔P0 + P0 =1
λ − μL μL μH − λ
k k
μL λ λ λ λ
⇔P0 − P0 + P0 =1
μL − λ μL μL − λ μL μH − λ
k
μL λ λ λ
⇔P0 + P0 − =1
μL − λ μL μH − λ μL − λ

μL λk λ (μH − μL )
⇔P0 − P0 k =1
μL − λ μL (μL − λ) (μH − λ)

μL (μH − λ)
k+1
λk+1 (μH − μL )
⇔P0 − =1
μkL (μL − λ) (μH − λ) μkL (μL − λ) (μH − λ)

μk+1 (μH − λ) − λk+1 (μH − μL )
⇔P0 L
=1
⇔ = P0
μk+1
L (μH − λ) − λk+1 (μH − μL )
4.9. Appendix 73
PDF of the production lead time:
∞

1
k
fW (t) ≈ Pn (n ≤ k) fWn (t) + Pn (n > k) fWn (t)
1 − P0 n=1 n=k+1

1
k
μnL tn−1
= Pn (n ≤ k) e−μL t
1 − P0 n=1
(n − 1)!
∞

n n−1
−μH t μH t
+ Pn (n > k) e
(n − 1)!
n=k+1
k
1 λ
n
μn tn−1
= P0 e−μL t L
1 − P0 n=1 μL (n − 1)!
∞
λ
k
λ
n−k n n−1
−μH t μH t
+ P0 e
μL μH (n − 1)!
n=k+1
(μL − λ) (μH − λ)
=
k
λ (μH − λ) − (λ/μL ) (μH − μL )
k
λ n μn tn−1
e−μL t L
n=1
μL (n − 1)!
∞ n−k
λ
k
λ n n−1
−μH t μH t
+ e
μL μH (n − 1)!
n=k+1
(μL − λ) (μH − λ)
=
k
λ (μH − λ) − (λ/μL ) (μH − μL )
k ∞

n n−1 λn−k λk −μH t μnH tn−1
−μL t λ t
e + e
(n − 1)! μn−k μkL (n − 1)!
n=1 n=k+1 H
(μL − λ) (μH − λ)
=
k
λ (μH − λ) − (λ/μL ) (μH − μL )
∞

k
λn tn−1 μkH −μH t λn tn−1
−μL t
e + ke
n=1
(n − 1)! μL (n − 1)!
n=k+1
(μL − λ) (μH − λ)
=
k
λμkL (μH − λ) − (λ/μL ) (μH − μL )
∞

k
λn tn−1 λn tn−1
μkL e−μL t + μkH e−μH t
n=1
(n − 1)! (n − 1)!
n=k+1
(μL − λ) (μH − λ)
= k
μL (μH − λ) − λk (μH − μL )

Γ (k, λt) Γ (k, λt)
e−(μL −λ)t μkL + e−(μH −λ)t μkH 1 − (4.41)
Γ (k) Γ (k)
P0
Auxiliary calculation for 1−P0 is presented as:
μk
L (μL −λ)(μH −λ)
P0 μk+1 (μH −λ)−λk+1 (μH −μL )
= Lk
1 − P0 (μL λ)(μH −λ)−λk+1 (μH −μL )
μk+1
L (μH −λ)−λk+1 (μH −μL )

=
μkL λ (μH − λ) − λk+1 (μH − μL )

=
k
μkL λ (μH − λ) − λ μλk (μH − μL )
L
(μL − λ) (μH − λ)
= (4.42)
k
λ (μH − λ) − (λ/μL ) (μH − μL )
Derivation of the expected production lead time W̃ based on this PDF

approximation is presented in Equation 4.43.
∞
E W̃ = fW (τ ) τ dτ
0
∞
(μL − λ) (μH − λ) Γ (k, λτ )
= e−(μL −λ)τ μL k
μL k (μH − λ) − λk (μH − μL ) Γ (k)
0

Γ (k, λτ )
+e−(μH −λ)τ μH k 1 − τ dτ
Γ (k)
⎛
∞
(μL − λ) (μH − λ) ⎝ Γ (k, λτ )
= k μL k
e−(μL −λ)τ τ dτ
μL (μH − λ) − λk (μH − μL ) Γ (k)
0
⎞
∞ ∞
Γ (k, λτ )
+ μH k e−(μH −λ)τ τ dτ −μH k e−(μH −λ)τ τ dτ ⎠
Γ (k)
0 0
4.9. Appendix 75
(μL − λ) (μH − λ)
=
⎛ k
μL k μL −k λk μL μλLk − 1 − k (μL − λ)
⎝
2
μL (λ − μL )
k ⎞
μH k μH k μH −k λk μH μλHk − 1 − k (μH − λ)
⎠
+ 2 − 2
(λ − μH ) μH (λ − μH )
(μL − λ) (μH − λ)
=

μL k − λk λk k λk λk k
2 − μ (μ − λ) + 2 + μ (μ − λ)
(μL − λ) L L (μH − λ) H H
(4.43)
Based on Equation 4.32 also the CDF of the approximated production

lead time can be calculated as:
τ
FW (τ ) = fW (t) dt
0
τ
(μL − λ) (μH − λ)
=
0

−(μL −λ)t k Γ (k, λt) −(μH −λ)t Γ (k, λt)
e μL +e μH k
1− dt
Γ (k) Γ (k)
(μL − λ) (μH − λ)
= k
τ
Γ (k, λt) Γ (k, λt)
e−(μL −λ)t μL k +e−(μH −λ)t μH k 1 − dt
Γ (k) Γ (k)
0
⎛
τ
(μL − λ) (μH − λ) ⎝ μL k
= k e−(μL −λ)t Γ (k, λt) dt
μL (μH − λ) − λk (μH − μL ) Γ (k)
0
⎞
τ k
τ
μ
+μH k e−(μH −λ)t dt − e−(μH −λ)t Γ (k, λt) dt⎠
H
(4.44)
Γ (k)
0 0
With:
τ
μL k
e−(μL −λ)t Γ (k, λt) dt
Γ (k)
0
μL k e−μL τ (λτ ) μL k λk μL −k Γ (k)

k
μL k Γ (k)
= + −
Γ (k) μL (μL − λ) τ Γ (k) (μL − λ) Γ (k) (μL − λ)
k −1−k
μL k μL τ (λτ ) (μL τ ) Γ (k − 1, μL τ )
−
Γ (k) (μL − λ)
k −1−k
μL k kμL τ (λτ ) (μL τ ) Γ (k − 1, μL τ )
+
Γ (k) (μL − λ)
μL k e−(μL −λ)τ Γ (k, λτ )
−
Γ (k) (μL − λ)
μL k−1 e−μL τ (λτ ) μL k e−(μL −λ)τ Γ (k, λτ )
k
= −
Γ (k) τ (μL − λ) (μL − λ) Γ (k)
μL − λ
k k
λ k
Γ (k − 1, μL τ )
+ +
(μL − λ) (μL − λ) Γ (k − 1)
And:
τ
1 − e−(μH −λ)τ
μH k
e−(μH −λ)t dt = μH k
μH − λ
0
It follows:
(μL − λ) (μH − λ)
FW (τ ) =

μL k−1 e−μL τ (λτ ) μL k e−(μL −λ)τ Γ (k, λτ )
k
−
Γ (k) τ (μL − λ) (μL − λ) Γ (k)

μL k − λk λk Γ (k − 1, μL τ )
+ +
(μL − λ) (μL − λ) Γ (k − 1)
(μL − λ) (μH − λ) 1 − e−(μH −λ)τ
+ μH k
μL (μH − λ) − λ (μH − μL )
k k μH − λ
(μL − λ) (μH − λ)
+

μH k−1 e−μH τ (λτ ) μH k e−(μH −λ)τ Γ (k, λτ )
k
−
Γ (k) τ (μH − λ) (μH − λ) Γ (k)

μH − λ
k k
λ k
Γ (k − 1, μH τ )
+ +
(μH − λ) (μH − λ) Γ (k − 1)
4.9. Appendix 77
(μH − λ)
=
k k−1 k−1 −μL τ
λ μL τ e Γ (k, λτ ) k
− μL k e−(μL −λ)τ + μ L − λk
Γ (k) Γ (k)

Γ (k − 1, μL τ )
+λk
Γ (k − 1)
k k−1 k−1 −μH τ
(μL − λ) λ μH τ e
−
μL k (μH − λ) − λk (μH − μL ) Γ (k)

k −(μH −λ)τ Γ (k, λτ )
k
k Γ (k − 1, μH τ )
−μH e + μH − λ + λ k
Γ (k) Γ (k − 1)
−(μH −λ)τ

(μL − λ) μH 1 − e
k
+
(μH − λ)
= k
λ μL τ e Γ (k, λτ ) k
− μL k e−(μL −λ)τ + μ L − λk
Γ (k) Γ (k)

Γ (k − 1, μ L τ ) (μ L − λ)
+λk −
Γ (k − 1) μL (μH − λ) − λk (μH − μL )
k

μH e 1−
Γ (k)
k−1 k−1 −μH τ

k
λ μH τ e Γ (k − 1, μH τ )
+ −λ 1−
k
Γ (k) Γ (k − 1)
For simplifying the statement the following auxiliary functions are de-
fined:
(μH − λ)
νL =
k
(μL − λ)
νH = k
λk μL k−1 τ k−1 e−μL τ Γ (k, λτ )
ϑL (τ ) = − μL k e−(μL −λ)τ
Γ (k) Γ (k)
k
Γ (k − 1, μ L τ )
+ μ L − λk + λk
Γ (k − 1)

ϑH (τ ) =μH e 1−
Γ (k)
k−1 k−1 −μH τ

k
+ − λk 1 − (4.45)
Γ (k) Γ (k − 1)
Which leads to:
FW (τ ) = νL ϑL (τ ) − νH ϑH (τ ) (4.46)
The following equations provide the derivation of the expected FGI:
X
E [G] =λ FW (τ ) (1 − FL (τ )) dτ
0
X
=λ (νL ϑL (τ ) − νH ϑH (τ )) (1 − FL (τ )) dτ
0
X X
=λνL ϑL (τ ) e−βτ dτ − λνH ϑH (τ ) e−βτ dτ
0 0
X k−1 −μL τ
λk μk−1
L τ e Γ (k, λτ )
=λνL − μkL e−(μL −λ)τ
Γ (k) Γ (k)
0

Γ (k − 1, μL τ ) −βτ
+ μkL − λk + λk (k − 1) e dτ
Γ (k)
X
− λνH μH e 1−
Γ (k)
0
k−1 −μH τ

λk μk−1
H τ e Γ (k − 1, μH τ )
+ − λk 1 − e−βτ dτ
Γ (k) Γ (k − 1)
X
νL λk+1 μk−1
= L
τ k−1 e−(μL +β)τ dτ
Γ (k)
0
X
λνL μkL
− e−(μL −λ+β)τ Γ (k, λτ ) dτ
Γ (k)
0
X X

λk+1 νL
+ λνL μkL − λk e−βτ dτ + Γ (k − 1, μL τ ) e−βτ dτ
Γ (k − 1)
0 0
X X
λνH μkH
− λνH μkH e−(μH −λ+β)τ dτ + e−(μH −λ+β)τ Γ (k, λτ ) dτ
Γ (k)
0 0
4.9. Appendix 79
X
νH λk+1 μk−1
− H
τ k−1 e−(μH +β)τ dτ
Γ (k)
0
X X
λk+1 νH
+ λk+1 νH e−βτ dτ − Γ (k − 1, μH τ ) e−βτ dτ
Γ (k − 1)
0 0
(4.47)
Applying the following integral solutions:
X
−k
τ k−1 e−(μL +β)τ dτ =X k (X (β + μL )) [Γ (k) − Γ (k, X (μL + β))]
0
X
e−(μL +β)X (Xλ)
k
Γ (k)
e−(μL −λ+β)τ Γ (k, λτ ) dτ = +
X (μL + β) (μL − λ + β) μL − λ + β
0
−k
λk (μL + β) Γ (k)
−
μL − λ + β
−k−1
Γ (k − 1, X (μL + β)) (μL + β) λk
+
μL − λ + β μL − λ + β
Γ (k, Xλ) e−(μL −λ+β)X
((k − 1) (μL + β)) −
μL − λ + β
X
1 − e−βX
e−βτ dτ =
β
0
X
1 − e−(μH −λ+β)X
e−(μH −λ+β)τ dτ =
(μH − λ + β)
0
X
e−(μL +β)X (XμL )
k−1
Γ (k − 1)
Γ (k − 1, μL τ ) e−βτ dτ = +
Xβ (μL + β) β
0
−e−βX Γ (k − 1, XμL )
β
1−k
(k − 2) μk−1 (μL + β)
− L
β
Γ (k − 2) − Γ (k − 2, X (μL + β))
(4.48)
β
Leads to:
νL λk+1 μk−1 −k
E [G] = L
X k (X (β + μL )) (Γ (k) − Γ (k, X (β + μL )))
Γ (k)

k
λνL μkL
−
Γ (k) X (μL + β) (μL − λ + β)
−k
Γ (k) λk (μL + β) Γ (k)
+ −
μL − λ + β μL − λ + β
−k−1
Γ (k − 1, X (μL + β)) (μL + β) λk
+ (k − 1) (μL + β)
μL − λ + β

1 − e−βX λk+1 νL
− + λνL μkL − λk +
μL − λ + β β Γ (k − 1)

−(μL +β)X k−1 −βX
e (XμL ) Γ (k − 1) − e Γ (k − 1, XμL )
+
Xβ (μL + β) β
1−k
(k − 2) μk−1 (μL + β) [Γ (k − 2) − Γ (k − 2, X (μL + β))]
− L
β β

1 − e−(μH −λ+β)X e−(μH +β)X (Xλ)
k
λνH μkH
− λνH μkH +
(μH − λ + β) Γ (k) X (μH + β) (μH − λ + β)
−k
Γ (k) λk (μH + β) Γ (k)
+ −
μH − λ + β μH − λ + β
−k−1
Γ (k − 1, X (μH + β)) (μH + β) λk
+ ((k − 1) (μH + β))
μH − λ + β

Γ (k, Xλ) e−(μH −λ+β)X νH λk+1 μk−1 −k
− − H
X k (X (β + μH ))
μH − λ + β Γ (k)
1 − e−βX λk+1 νH
(Γ (k) − Γ (k, X (β + μH ))) + λk+1 νH −
β Γ (k − 1)

e−(μH +β)X (XμH ) Γ (k − 1) − e−βX Γ (k − 1, XμH )
k−1
+
Xβ (μH + β) β
1−k
(k − 2) μk−1
H (μH + β) (Γ (k − 2) − Γ (k − 2, X (μH + β)))
−
β β
(4.49)
4.9. Appendix 81
Which can be restated as:

−k Γ (k, X (β + μL ))
E [G] =νL λk+1 μk−1
L (β + μL ) 1−
Γ (k)

k
− λνL μkL
Γ (k) X (μL + β) (μL − λ + β)
−k
1 − λk (μL + β) e−(μL −λ+β)X Γ (k, Xλ)
+ −
μL − λ + β μL − λ + β Γ (k)

−k k
(μL + β) λ Γ (k − 1, X (μL + β))
+
(μL − λ + β) Γ (k − 1)

1−e −βX
1
+ λνL μkL − λk + λk+1 νL
β β
e−βX Γ (k − 1, XμL ) e−(μL +β)X (XμL )
k−1
− +
β Γ (k − 1) Γ (k − 1) Xβ (μL + β)

μLk−1
Γ (k − 2, X (μL + β))
− 1−
β (μL + β)
k−1 Γ (k − 2)
1 − e−(μH −λ+β)X
− λνH μkH
(μH − λ + β)

−k
k
1 − λk (μH + β)
+ λνH μkH +
Γ (k) X (μH + β) (μH − λ + β) μH − λ + β
−k
(μH + β) λk Γ (k − 1, X (μH + β))
+
μH − λ + β Γ (k − 1)
−(μH −λ+β)X

e Γ (k, Xλ) 1 − e−βX
− + λk+1 νH
μH − λ + β Γ (k) β

−k Γ (k, X (β + μH ))
− νH λk+1 μk−1
H (β + μ H ) 1 −
Γ (k)

−(μH +β)X
e−βX Γ (k − 1, XμH )
k−1
e (XμH ) 1
− λk+1 νH + −
Γ (k − 1) Xβ (μH + β) β β Γ (k − 1)

1−k
μk−1 (μH + β) Γ (k − 2, X (μH + β))
− H 1− (4.50)
β Γ (k − 2)
Derivation for service level:
X ∞
s= FW (τ ) fL (τ ) dτ + FW (X) fL (τ ) dτ
0 X
X
k−1 −μL τ
(μH − λ) λk μk−1
L τ e
=
μkL (μH − λ) − λk (μH − μL ) Γ (k)
0

Γ (k, λτ ) k
Γ (k − 1, μL τ )
−μkL e−(μL −λ)τ + μL − λk +λk
Γ (k) Γ (k − 1)

(μL − λ) k −(μH −λ)τ Γ (k, λτ )
− k μ H e 1 −
μL (μH − λ) − λk (μH − μL ) Γ (k)
k k−1 k−1 −μH τ

+ − λk 1 − βe−βτ dτ
Γ (k) Γ (k − 1)
+ (νL ϑL (X) − νH ϑH (X)) e−βX
⎛X
⎝ λ μL τ e
=νL βe−βτ dτ
Γ (k)
0
X X
Γ (k, λτ ) k
− μkL e−(μL −λ)τ βe−βτ dτ + μL − λk βe−βτ dτ

Γ (k)
0 0
⎞
X
Γ (k − 1, μL τ ) ⎠
+ λk βe−βτ dτ
Γ (k − 1)
0
⎛X

Γ (k, λτ )
− νH ⎝ μkH e−(μH −λ)τ βe−βτ 1 − dτ
Γ (k)
0
X k−1 −μH τ
λk μk−1
H τ e
+ βe−βτ dτ
Γ (k)
0
⎞
X
Γ (k − 1, μH τ )
− λk βe−βτ 1− dτ ⎠
Γ (k − 1)
0
+ (νL ϑL (X) − νH ϑH (X)) e−βX
4.9. Appendix 83
⎛
X
λk μk−1
L β
=νL ⎝ τ k−1 e−(μL +β)τ dτ
Γ (k)
0
X
μkL β

− e−(μL −λ+β)τ Γ (k, λτ ) dτ + μkL − λk 1 − e−βX

Γ (k)
0
⎞
k X
λ β
+ e−βτ Γ (k − 1, μL τ ) dτ ⎠
Γ (k − 1)
0
⎛
X X
Γ (k, λτ )
− νH ⎝μkH β e −(μH −λ+β)τ
dτ − μkH β e−(μH −λ+β)τ dτ
Γ (k)
0 0
X
λk μk−1
H β

+ τ k−1 e−(μH +β)τ dτ − λk 1 − e−βX + λk β

Γ (k)
0
⎞
X
Γ (k − 1, μ τ )
e−βτ dτ ⎠ + (νL ϑL (X) − νH ϑH (X)) e−βX
H
Γ (k − 1)
0

λk μk−1
L β −k
=νL X k (X (β + μL )) (Γ (k) − Γ (k, X (μL + β)))
Γ (k)

−k
μkL β e−(μL +β)X (Xλ)
k Γ (k) − λk (μL + β) Γ (k)
− +
Γ (k) X (μL + β) (μL − λ + β) μL − λ + β
−k−1
Γ (k − 1, X (μL + β)) (μL + β) λk
+ ((k − 1) (μL + β))
μL − λ + β


− + μkL − λk 1 − e−βX
μL − λ + β

e−(μL +β)X (XμL )
k−1
λk β
+
Γ (k − 1) Xβ (μL + β)
Γ (k − 1) − e−βX Γ (k − 1, XμL )
+
β
1−k
(k − 2) μk−1 (μL + β) Γ (k − 2) − Γ (k − 2, X (μL + β))
− L
β β

−(μH −λ+β)X
k
1−e k
μ β
− νH μkH β − H
(μH − λ + β) Γ (k) X (μH + β) (μL − λ + β)
−k
Γ (k) λk (μH + β) Γ (k)
+ −
μH − λ + β μH − λ + β
−k−1
Γ (k − 1, X (μH + β)) (μH + β) λk
+ ((k − 1) (μH + β))
μH − λ + β
−(μH −λ+β)X

Γ (k, Xλ) e
−
μH − λ + β
λk μk−1
H β −k
+ X k (X (β + μH )) (Γ (k) − Γ (k, X (μH + β)))
Γ (k)

− λk 1 − e−βX

e−(μH +β)X (XμH )
k−1
λk β
+
Γ (k − 1) Xβ (μH + β)
Γ (k − 1) − e−βX Γ (k − 1, XμH )
+
β
1−k
(k − 2) μk−1
H (μH + β) (Γ (k − 2) − Γ (k − 2, X (μH + β)))
−
β β
+ (νL ϑL (X) − νH ϑH (X)) e−βX (4.51)
Applying some algebra and the definition of ϑL (τ ) and ϑH (τ ) leads to:

λk μk−1
L β Γ (k, X (μL + β))
s =νL k
1−
(β + μL ) Γ (k)

λk μkL β e−(μL +β)X X k−1 1
− + k
μL − λ + β Γ (k) (μL + β) λ

1 Γ (k − 1, X (μL + β))
− 1−
(μL + β)
k−1 Γ (k − 1)
−(μL +β)X k−1
e X e−(μL +β)X X k−2
+ μkL + λk μk−1 +
L
Γ (k) Γ (k − 1) (μL + β)

1 Γ (k − 2, X (μL + β))
− 1−
(μL + β)
k−1 Γ (k − 2)

μkL (μL − λ) e−(μL −λ+β)X Γ (k, Xλ)
−
μL − λ + β Γ (k)

k k−1
λ μH β Γ (k, X (μH + β))
− νH k
1 −
(μH + β) Γ (k)
4.9. Appendix 85
−(μH +β)X k−1

λk μkH β e X e−(μH −λ+β)X
− +
μH − λ + β Γ (k) (μH + β) λk

1 Γ (k − 1, X (μH + β))
− 1−
(μH + β)
k Γ (k − 1)
−(μH +β)X k−2
e X e−(μH +β)X X k−1
+ λk μk−1 +
H
Γ (k − 1) (μH + β) Γ (k)

1 Γ (k − 2, X (μH + β))
− 1−
(μH + β)
k−1 Γ (k − 2)

μkH e−(μH −λ+β)X (μH − λ) Γ (k, Xλ)
+ μkH e−(μH −λ+β)X − (4.52)
μH − λ + β Γ (k)
Chapter 5
Conwip
The production planning and control method Conwip is firstly mentioned in
Spearman et al. (1990) and has gained since then significant research inter-
est; see (Framinan et al., 2003) for a review. Analytical studies discussing
the behaviour of Conwip are usually linked to MTS systems or they neglect
the distinction between MTO and MTS since only throughput, WIP and
production lead time are focused (Spearman et al., 1990; Duenyas et al.,
1993; Tardif and Maaseidvaag, 2001; Gershwin and Werner, 2007). For
analysing the behaviour of Conwip for more complex production systems, a
set of simulation studies is already available either discussing MTS or also
MTO environments (Hopp and Roof, 1998; Huang et al., 1998b; Framinan
et al., 2000, 2006; Yang et al., 2007; Jodlbauer and Huber, 2008; Hübl et al.,
2011). Most of this literature either focuses on providing optimal policies for
Conwip or it compares the Conwip application to other production planning
and control methods.
None of the available contributions discusses the definition of the in-
ventory constraining parameter in the Conwip method when it is applied
to an MTO environment. This inventory constraining parameter, which is
referred to as Wipcap in this section, can have two different definitions in
an MTO environment. On the one hand, the Wipcap can be related to the
materials being processed on the shop floor or waiting to be processed there.
This implies that the Wipcap actually only constrains the WIP within the
production system. From the first definitions provided concerning Conwip
parameters (Spearman et al., 1990; Hopp and Spearman, 2008), this is the
most widely spread definition which is also applied in most of the litera-
ture. However, on the other hand, the Wipcap can also include the FGI
waiting to be delivered to the customers. Therefore, not only the WIP
but the sum of WIP and FGI can be constrained by the Wipcap. In this
chapter, the influence of these two different Wipcap definitions on the logis-
tical key performance indicators service level and tardiness for a multi-stage
and multi-item production system with stochastic order sizes and stochastic
customer required lead time is studied.
Furthermore, the effect of a safety stock implementation, which is stan-
dard in the widely applied MRP, on performance measures and the Wipcap
definition decision is studied. Simulation is applied since the complexity
of this multi-stage and multi-product model with general distributions for
interarrival time, customer required lead time and customer order sizes also
including set-up times is no more analytically tractable. The main finding
of this simulation study is that for a large set of different production system

DOI 10.1007/978-3-658-19120-7_5
88 Chapter 5. Conwip
parametrisations, constraining WIP by the Wipcap leads to a better service

level and tardiness performance compared to constraining WIP plus FGI.
5.1 Model Description
5.1.1 Production System

The MTO production system modelled in this study is a stylised setting of
a 1st tier supplier in the automotive industry, whereby the customer gets
a certain final item and not the final item which is released to production
based on the customer order. Inter arrival times of customer orders, cus-
tomer order amount, customer required lead time and processing times are
stochastic. The BOM structure is a mixture of sequential and divergent
whereby twelve final products are grouped to three product groups. The
materials are produced on four machine groups all consisting of between
one and four machines. A set of various 1st tier suppliers in the automotive
industry has been studied and based on these findings, the BOM structure
and number of processing stages have been created to include the complexity
needed for deriving reliable results, while keeping the model still tractable.
Figure 5.1 represents the combination of BOM and routing whereby each
time one piece of the respective sub-material is needed to produce a mate-
rial. The materials 1301 and 1303 are raw materials from external suppliers
and the materials with code 62xx are the final products. The product groups
are 5201, 5211 and 5221. Whenever an order is finished at machine M4, the
finished materials are stored in an FGI until the due date of order.
5.1.2 Production Planning and Control Applying

Conwip
The production planning and control for this production system is carried
out with Conwip. According to Spearman et al. (1990), Framinan et al.
(2003) and Hopp and Spearman (2008), the following parameters can be
decided on to implement a Conwip production planning and control method.
Wipcap
The Wipcap constrains the input of materials or orders into the produc-
tion system. In the current implementation, the WIP is measured in final
production units which are released to the shop floor. Whenever an order
is released to the shop floor, the WIP-counter is increased by the number
of finished goods of this order. Each order is assumed to refer to only one
type of finished goods. As stated above, two different types of Wipcap are
implemented. In type 1, which is the definition directly derived from the
5.1. Model Description 89
6201 6202 6203 6211 6212 6213 6214 6221 6222 6223 6224 6225
M4
5201 5211 5221
M3
4101 4102
M1
3101
M2
2101 2102
M1
1301 1303
Figure 5.1: Combined BOM and routing structure
Conwip literature (Framinan et al., 2003; Hopp and Spearman, 2008), the
WIP-counter is reduced by the number of finished goods, whenever an order
is finished at machine group M4. In type 2, which is specifically applica-
ble for MTO production systems, the WIP-counter is reduced whenever an
order leaves the FGI. In both implementations orders are only released to
the production system when the WIP-counter including the next order to
be released is lower or equal the Wipcap.
Work-Ahead-Window (WAW)
In the current implementation the WAW sets a time window into the future,
e.g. 20 periods of time. An order can only be released to the production
system when its due date is within this WAW. This is the typical imple-
mentation of a WAW in MTO production systems (Framinan et al., 2003;
Jodlbauer and Huber, 2008; Hopp and Spearman, 2008).
Order List Sorting

In an MTO environment where the work release to the production system is
constrained, there will be a list of orders waiting to be released. This order
list, which contains only orders with a due date within the WAW, is sorted
according to the EDD dispatching rule (Hopp and Spearman, 2008, for the
EDD application). When performing the order release, only the first order
on the list can be released to the system. Even in case that the first order is
too large, i.e. needs a high number of finished goods, and the second order
on the list is smaller and could therefore be released to the system according
to the Wipcap, this second order would not be released before the first one
has been released.
Production Order Dispatching

Whenever more than one order is waiting in front of a machine, it needs to be
decided which one is the next to be produced. In the current implementation
the FISFS dispatching rule is applied (Hopp and Spearman, 2008, for the
FISFS application). This means that orders are always sorted according
to their release date to the production system. An order which has been
released earlier is always preferred to one being released later.
Capacity Trigger
The capacity trigger is an additional functionality of Conwip which indi-
cates if enough capacity is available or if additional capacity is needed. For
simplification reasons no such functionality is implemented in the current
study and the capacity provided is constant.
Safety Stock
In addition to the five usually applied parameters in a Conwip system, a
safety stock for finished goods is added in this Conwip implementation. For
each finished good a certain percentage of the average periodic demand is
kept as safety stock to increase service level and decrease tardiness.
5.1.3 Simulation Model

The parametrisation as explained below is again based on the 1st tier au-
tomotive supplier assumption. The time frame in the simulation model is
a simplified working year consisting of twelve months with 20 work days
each, which corresponds to a year being 240 time periods long in the simu-
lation. In this section the parametrisation of the basic setting is described,
whereby additional settings are introduced in the simulation study. In the
5.1. Model Description 91
simulation model, the customer orders are generated as compound Poisson

demand with the order amount being log-normal distributed. The products
6201, 6202 and 6203 have a monthly arrival rate of 133.3; 6211, 6212, 6213
and 6214 have a monthly arrival rate of 100; 6221, 6222, 6223, 6224 and 6225
have a monthly arrival rate of 80. The mean order amount is 20 pcs/order
with a coefficient of variation of 0.25 for all finished goods. Each order has
a minimum customer required lead time of 10 periods, which is equivalent
to a frozen zone, plus a log-normal distributed customer required lead time
with mean 40 periods and a coefficient of variation of 0.25. This means
on average the orders are known 50 periods in advance. Each customer
order is released to the production system as single order, so no orders are
merged. The processing times and set-up times, as provided in Table 5.1,
are deterministic whereby a set-up is needed before each production order.
The raw material replenishment is not covered in this study and therefore
raw materials are available whenever an order is released to the production
system.
The simulation framework has been presented in Hübl et al. (2011) and
AnyLogic 6.9 simulation software has been used. Moreover, the simula-
tion model has been validated according to Kleijnen (1995) by checking the
model behaviour of the single-machine case with deterministic parameters
compared to developed analytic results.
5.1.4 Simulation Study

The objective function for evaluating the performance of the two different
Wipcap definitions tested in this study are service level and tardiness related
to the overall inventory (WIP+FGI) needed to reach these values. The
following parameters are varied to generate nine different scenarios:
• Concerning utilisation, three different system loads are tested whereby

the low, medium and high utilisation is implemented by multiplying
the basic arrival rates with 0.875, 1 and 1.125 respectively. This leads
to 72%, 82% and 92% overall utilisation at the bottleneck machine 1
and 4 respectively.
• To reflect customer behaviour, a low, medium and high stochastic part
of the customer required lead time with values of 20, 40 and 80 periods
respectively are tested. The coefficient of variation of the stochastic
customer required lead time remains unchanged.
In combination with the customer order behaviour, the processing and

set-up times lead to the following utilisation levels at each machine group
(see Table 5.2), whereby machine groups 1 and 4 have been designated as
bottleneck for all customer order behaviour scenarios.
Table 5.1: Processing and set-up times
Material Set-up time Processing time

6201 0.033 0.042
6202 0.033 0.021
6203 0.033 0.021
6211 0.033 0.042
6212 0.033 0.021
6213 0.033 0.021
6214 0.033 0.021
6221 0.033 0.042
6222 0.033 0.021
6223 0.033 0.021
6224 0.033 0.021
6225 0.033 0.021
5201 0.001 0.042
5211 0.001 0.042
5221 0.001 0.042
4101 0.050 0.042
4102 0.050 0.042
3101 0.050 0.042
2101 0.050 0.042
2102 0.050 0.042
To identify which Wipcap leads to a better result for each scenario, all
combinations of the following values for Wipcap and WAW are enumerated.
The ranges and values have been identified in a set of preliminary experi-
ments and are presented in Table 5.3. The Wipcap is incremented by steps
of five, and the WAW by steps of one.
Additionally, the influence of implementing a safety stock in both meth-
ods has been identified by including the following safety stock values for each
finished good as percentage of its monthly average demand as presented in
Table 5.4.
For all parameter combinations tested, 40 replications are conducted, for
which 720 time periods are simulated while a warm-up period of 240 time
periods is excluded (2 years, 480 time periods are evaluated). Overall, this
simulation study consists of 3, 321, 720 simulation runs being performed in
AnyLogic 6.9. To hold an item of FGI costs 1 monetary unit per time period
and a WIP item costs 0.75 monetary units per time period. Tardiness costs
are penalised by 19 monetary units per period.
5.2. Results 93
Table 5.2: Utilisation at each machine group
Parameter Machine utilisation [%]

demand Group 1 Group 2 Group 3 Group 4
0.875 72.3 67.9 67.6 72.1
1 82.4 77.1 77.3 82.2
1.125 92.3 86.4 86.4 92.0
Table 5.3: Limits for Wipcap and WAW
Wipcap WAW
Parameter Wipcap Lower Upper Lower Upper
demand method level level level level
0.875 1 400 600 6 21
0.875 2 430 630 6 21
1 1 400 600 6 21
1 2 430 630 6 21
1.125 1 400 600 6 27
1.125 2 430 920 6 27
5.2 Results
In this section the results of the simulation study are presented. Firstly, the
influence of Wipcap and WAW on overall costs for Wipcap method 1 and
Wipcap method 2 respectively is discussed in the basic setting. Secondly,
the performance of the two Wipcap methods without implementing a safety
stock is compared for the nine scenarios (3 demand scenarios, 3 customer
required lead time scenarios). Thirdly, the effect of implementing a safety
stock in Conwip is discussed based on the nine scenarios.
5.2.1 Discussion of WAW and Wipcap

For analysing WAW and Wipcap the medium load (parameter demand = 1)
and medium customer behaviour (mean customer required lead time = 40)
scenario has been selected. The following Figure 5.2 shows the minimum
overall costs, when the WAW changes. To allow a comparison between
Wipcap method 1 and Wipcap method 2, the best Wipcap for each WAW
leading to minimum overall costs is chosen. No significant difference between
Table 5.4: Introduced safety stock percentage of the periodic demand for FGI
Wipcap Param. Saftey stock percentage

method demand
0.875 0 0.1 0.18 0.2 0.3 0.5
1 1 0 0.2 0.3 0.4 0.4
1.125 0 0.05 0.1 0.15 0.2 0.25 0.3
0.875 0 0.1 0.18 0.2 0.3 0.5
2 1 0 0.2 0.3 0.4 0.4
1.125 0 0.05 0.1 0.15 0.2 0.25 0.3
the two methods tested has been observed. Hence, this circumstance leads
to the assumptions that the two Wipcap methods investigated have no
significant influence on overall costs. Additionally, a short WAW leads to
higher costs since the tardiness costs are increasing because the due dates
cannot be hold by the production system. The main purpose of the WAW is
to avoid producing items whose due date is too far in the future (Spearman
et al., 1990; Jodlbauer, 2008a). A WAW between 10 and 21 time unites for
the sets tested leads almost to the same overall costs, because the production
is limited by the Wipcap leading to the minimal overall costs and not by
the WAW.
2500
- - Wipcap Method 2
+
2000 + Wipcap Method 1
1500 +
-
Costs
+
- - + +
-
- - - - - - - + -
+ -
-
+ +
1000 +
- +
+ + + +
+
500
0
5 10 15 20
WAW
Figure 5.2: WAW with best Wipcap
Figure 5.3 illustrates the minimum overall costs, when the Wipcap changes.
Again, to allow a comparison between both Wipcap methods, the best WAW
for each Wipcap leading to minimal overall costs is selected. For the cases
5.2. Results 95
tested, Wipcap method 1, where only the WIP is constrained by the Wip-
cap, leads to lower overall costs then Wipcap method 2, where WIP+FGI
are constrained by the Wipcap parameter. This circumstance holds for a
confidence interval with a probability of 95% except for three values where
no significant difference has been observed.
4000
- - Wipcap Method 2
-- + Wipcap Method 1
3000
-
-
- ---
Costs
2000 --
---
-
---
------
+++ ------- - --
1000 +++++++++ +++ - ----
-++
+ +++++++++++++++++++++++
0
400 450 500 550 600 650
Wipcap
Figure 5.3: Wipcap with best WAW
5.2.2 Wipcap Method Comparison

The comparison of Wipcap method 1 and Wipcap method 2 is based on
the lowest overall costs consisting of WIP+FGI holding costs and costs for
tardiness for each combination of WAW, Wipcap, system loads, customer
required lead time and Wipcap method. Table 5.5 indicates the lowest
overall costs for each parameter combination, whereby all parameters al-
tered (customer required lead time, WAW, Wipcap, bottleneck utilisation)
are indicated with their results for tardiness costs, costs for FGI, costs for
WIP and production lead time for both Wipcap methods in each row.
In Table 5.6 a safety stock for finished items is introduced and again,
Wipcap method 1 and method 2 are compared based on overall costs. In
contrast to the previous table the column safety stock percentage is intro-
duced meaning that e.g. 15% of the monthly demand is kept as safety
stock.
For both investigations (Table 5.5 and Table 5.6) a confidence interval
for 90%, 95%, 99% and 99.9% has been calculated (see Table 5.7 to Table
5.14 in the Appendix of this Chapter). If there is a significant difference
based on the confidence intervals between Wipcap method 1 and method 2

a footnote indicates at which confidence level.
Observation 5.2.1 (System load). Overall costs, FGI costs, WIP costs,
tardiness costs, production lead time and WAW increases by increasing sys-
tem load
As a first intuitive finding, overall costs, FGI costs, WIP costs, tardi-
ness costs, production lead time and WAW increase when the system load
(bottleneck utilisation) of the investigated production systems is increased
since more orders have to be produced at the same simulation time period.
Increasing demand at the same capacity provided results in a higher pro-
duction lead time and WIP according to Little (1961). If more orders are
stated the risk, that more items are produced too early (FGI costs increase)
and also too late (tardiness costs increase), increases.
Observation 5.2.2 (WIP plus FGI constraint). A Conwip systems where

WIP plus FGI is constrained results in worse performance compared to Con-
wip systems where only WIP is constrained
Observation 5.2.2 is the main result of this simulation study and indi-
cates that for the scenarios tested a constrained WIP is beneficial compared
to Wipcap method 2 where WIP+FGI are constrained. For the medium
and high load scenario it has been observed that Wipcap method 2 leads
to more tardiness costs. This observation holds only when no safety stock
is applied. Moreover, there is a tendency that method 2 leads to a longer
production lead time than method 1 for medium and high load scenarios.
An explanation for Observation 5.2.2 could be that classic Conwip tries to
balance the workload independently of the customer behaviour. When FGI
is included in the Wipcap (method 2), then the customer behaviour is part
of the production planning process. It seems that the violation of the basic
Conwip principle leads to a worse performance.
Observation 5.2.3 (Saftey stock). Safety stock is not needed for the in-
vestigated system configurations
For the high demand scenarios, Wipcap method 2 (WIP+ FGI) requests
an introduction of safety stock in terms of overall costs. Nevertheless, Wip-
cap method 1 turns out to be the better one and requests no safety stock.
When safety stock is applied then it could be observed that the Wipcap
decreases.
Observation 5.2.4 (WAW). The work-ahead-window is almost indepen-

dent of the CRL applied for the minimum cost scenarios
For single-machine and two-machine systems, Altendorfer and Minner

(2011) showed analytically that planned lead time is independent of the
applied customer required lead time as long as WIP, FGI and backorder
costs are simultaneously minimised. Observation 5.2.4 supports this ana-
lytic result for a more complex structure in terms of amount of machines and
stochastic parameters since a work-ahead-window and planned lead times
have the same task for a production system.

In this chapter, Conwip is investigated and it compares the implementation
of a Wipcap constraint for WIP+FGI to the classical Conwip where only
the WIP is constrained. The main finding is that for the tested scenarios
both methods perform equally well in terms of overall costs for the low load
scenario. When the load increases then Wipcap method 1, which is the
classical method, leads to lower overall costs.
Moreover, safety stock is not needed in the investigated settings for
the cost optimal Wipcap method. Wipcap method 2 leads to lower FGI
costs but increases tardiness costs and production lead time for medium
and high load scenarios. Another interesting finding is that the WAW is
almost independent of the customer required lead time for the minimum
costs scenarios.
5.4 Appendix Conwip

In the appendix of this chapter, the results of the confidence interval calcu-
lation for Table 5.5 and Table 5.6 are presented.
98
Table 5.5: Analysis of Wipcap method comparison without safety stock [Values for Wipcap method 1
/ Values for Wipcap method 2]
Parameter Results
U* CRL WAW Wipcap Costs Costs Costs Costs Production
[%] FGI WIP tardiness lead time
0.5 9\9 545\590 579\570 156\151b 231\227 192\192 19.3\19
72 1 11\11 565\580 589\634 262\201d 231\232 96\201a 19.3\19.3
2 10\9 460\615 583\540 207\154d 225\227 151\159 18.9\19.1
0.5 12\15 515\625 786\929 306\188d 279\295c 200\446b 20.8\22.1d
82 1 13\9 485\590 771\933a 360\123d 275\282 136\529d 20.4\20.9b
2 13\18 595\620 791\965 369\191d 291\290 131\485c 21.6\21.7
0.5 21\21 585\905 1259\1814b 684\264d 371\445d 204\1104d 24.6\29.8d
92 1 20\18 575\905 1281\1750b 630\257d 366\451d 284\1042c 24.4\30.4d
2 21\21 610\915 1243\1618 704\279d 382\442d 157\897c 25.2\29.8d
a
90% confidence interval
b
c
d
99.9% confidence interval
*
Bottleneck utilisation
Chapter 5. Conwip
Table 5.6: Analysis of Wipcap method comparison with safety stock [Values for Wipcap method 1 / Values
for Wipcap method 2]
Parameter Results
U* CRL WAW Wipcap Safety Costs Costs Costs Costs Prod.
5.4. Appendix Conwip
[%] stock FGI WIP tardiness lead time

0.5 9\9 545\590 0\0 579\570 156\151b 231\227 192\192 19.3\19
72 1 11\11 565\580 0\0 589\634 262\201 d 231\232 96\201a 19.3\19.3
2 10\9 460\615 0\0 583\540 207\154d 225\227 151\159 18.9\19.1
0.5 12\15 515\625 0\0 786\929 306\188d 279\295c 200\446b 20.8\22.1d
82 1 13\9 485\590 0\0 771\933a 360\123 d 275\282 136\529 d 20.4\20.9 b
2 13\18 595\620 0\0 791\965 369\191 d 291\290 131\485c 21.6\21.7
0.5 21\16 585\640 0\0.15 1259\1548b 684\643 371\448d 204\456b 24.6\30.3d
92 1 20\21 575\655 0\0.2 1281\1595c 630\840d 366\473d 284\283 24.4\31.8d
2 21\23 610\550 0\0.2 1243\1617d 704\753 382\458d 157\406b 25.2\30.9d
a
b
c
d
99.9% confidence interval
*
99
100
Table 5.7: Analysis Wipcap without safety stock: confidence interval 0.1
Costs Costs FGI Costs WIP Costs tardiness Prod. lead time
U* Bound CRL Method Method Method Method Method
[%] 1 2 1 2 1 2 1 2 1 2
72 lower 0.5 520 510 154 149 227 224 133 132 19.1 18.8
upper 638 632 158 153 234 231 251 254 19.5 19.2
lower 1 545 577 260 197 228 228 53 144 19.1 19.1
upper 634 689 264 204 235 235 140 257 19.5 19.5
lower 2 530 492 205 152 223 224 97 111 18.8 18.9
upper 637 590 209 156 228 231 205 208 19.1 19.3
82 lower 0.5 720 843 300 181 276 290 131 358 20.7 21.8
upper 856 1021 313 195 283 300 273 539 21.0 22.4
lower 1 719 832 354 119 272 277 82 428 20.3 20.7
upper 822 1040 366 127 278 286 189 636 20.6 21.2
lower 2 732 816 363 185 286 285 71 335 21.3 21.5
upper 848 1113 374 197 296 294 189 632 21.9 22.0
92 lower 0.5 1185 1498 638 244 365 430 118 784 24.3 29.0
upper 1338 2158 730 285 377 461 296 1452 24.8 30.6
lower 1 1199 1448 585 236 361 435 174 737 24.1 29.4
upper 1370 2080 675 279 372 467 402 1376 24.6 31.4
lower 2 1189 1281 663 260 376 428 85 557 24.9 29.0
upper 1299 1987 745 298 389 456 230 1269 25.6 30.5
*
Chapter 5. Conwip
[%] 1 2 1 2 1 2 1 2 1 2
72 lower 0.5 509 498 154 148 227 223 122 121 19.1 18.8
upper 649 644 159 154 234 231 262 266 19.6 19.2
lower 1 537 566 259 196 227 227 44 133 19.1 19.1
upper 642 700 265 205 236 236 148 268 19.6 19.5
lower 2 520 482 205 152 222 223 87 102 18.8 18.9
upper 647 599 209 156 228 231 215 217 19.1 19.3
82 lower 0.5 707 825 299 180 275 289 118 341 20.6 21.8
upper 869 1039 314 196 284 301 287 557 21.0 22.4
lower 1 709 812 353 118 271 276 72 408 20.3 20.7
upper 832 1060 367 127 279 287 200 656 20.6 21.2
lower 2 721 788 362 184 285 284 60 307 21.3 21.4
upper 859 1141 376 199 297 295 201 661 21.9 22.0
92 lower 0.5 1171 1435 629 240 364 427 101 720 24.2 28.8
upper 1353 2221 738 289 378 464 313 1516 24.9 30.8
lower 1 1182 1388 577 232 360 432 152 675 24.0 29.2
upper 1387 2140 683 283 373 470 424 1437 24.7 31.6
lower 2 1179 1213 656 256 374 425 71 488 24.8 28.9
upper 1310 2055 753 302 390 459 244 1338 25.6 30.7
*
101
102
[%] 1 2 1 2 1 2 1 2 1 2
72 lower 0.5 487 475 153 148 226 222 100 98 19.0 18.7
upper 671 667 159 155 236 233 284 288 19.6 19.3
lower 1 521 545 258 195 226 226 28 112 19.0 19.0
upper 659 721 265 207 237 237 165 289 19.6 19.6
lower 2 500 464 204 151 221 222 67 84 18.7 18.8
upper 667 618 210 157 229 232 236 236 19.1 19.4
82 lower 0.5 681 792 297 178 274 287 91 307 20.5 21.7
upper 895 1072 316 199 285 303 313 590 21.1 22.6
lower 1 690 773 350 117 270 275 51 369 20.2 20.6
upper 851 1099 370 129 280 289 220 695 20.7 21.3
lower 2 699 733 360 181 283 282 38 251 21.2 21.3
upper 880 1196 378 201 298 297 223 717 22.0 22.1
92 lower 0.5 1142 1311 612 232 362 421 68 595 24.1 28.5
upper 1381 2345 756 297 380 469 346 1642 25.0 31.1
lower 1 1150 1270 560 224 358 426 109 556 23.9 28.8
upper 1419 2259 700 291 375 476 467 1557 24.8 32.0
lower 2 1158 1081 640 249 372 420 44 355 24.7 28.6
upper 1330 2187 769 309 393 464 271 1471 25.8 31.0
*
Chapter 5. Conwip
[%] 1 2 1 2 1 2 1 2 1 2
72 lower 0.5 461 449 152 147 224 220 74 71 18.9 18.6
upper 697 694 160 156 237 234 310 315 19.7 19.4
lower 1 501 521 257 193 224 225 9 87 18.9 18.9
upper 678 745 266 208 239 239 184 314 19.7 19.7
lower 2 477 443 203 151 220 220 43 63 18.7 18.7
upper 690 639 211 158 231 234 259 257 19.2 19.5
82 lower 0.5 651 753 294 175 272 285 61 268 20.5 21.6
upper 925 1111 319 202 287 305 344 630 21.2 22.7
lower 1 668 727 348 115 268 273 28 323 20.1 20.5
upper 873 1144 372 130 282 291 243 740 20.8 21.4
lower 2 674 668 357 179 281 280 12 186 21.1 21.2
upper 905 1261 380 204 301 299 248 781 22.1 22.2
92 lower 0.5 1109 1168 592 223 359 415 29 450 24.0 28.2
upper 1414 2488 776 306 383 476 385 1787 25.1 31.5
lower 1 1113 1132 541 214 356 419 60 417 23.8 28.4
upper 1456 2396 720 300 377 483 516 1696 24.9 32.4
lower 2 1135 927 622 240 369 414 13 200 24.6 28.2
upper 1354 2340 786 317 396 471 303 1626 25.9 31.3
*
103
104
Table 5.11: Analysis Wipcap with safety stock: confidence interval 0.1
[%] 1 2 1 2 1 2 1 2 1 2
72 lower 0.5 487 475 153 148 226 222 100 98 19.0 18.7
upper 671 667 159 155 236 233 284 288 19.6 19.3
lower 1 521 545 258 195 226 226 28 112 19.0 19.0
upper 659 721 265 207 237 237 165 289 19.6 19.6
lower 2 500 464 204 151 221 222 67 84 18.7 18.8
upper 667 618 210 157 229 232 236 236 19.1 19.4
82 lower 0.5 681 792 297 178 274 287 91 307 20.5 21.7
upper 895 1072 316 199 285 303 313 590 21.1 22.6
lower 1 690 773 350 117 270 275 51 369 20.2 20.6
upper 851 1099 370 129 280 289 220 695 20.7 21.3
lower 2 699 733 360 181 283 282 38 251 21.2 21.3
upper 880 1196 378 201 298 297 223 717 22.0 22.1
92 lower 0.5 1185 1443 638 620 365 431 118 347 24.3 29.3
upper 1338 1664 730 666 377 465 296 577 24.8 31.4
lower 1 1199 1521 585 809 361 450 174 203 24.1 30.4
upper 1370 1678 675 870 372 495 402 372 24.6 33.2
lower 2 1189 1500 663 729 376 440 85 285 24.9 29.9
upper 1299 1738 745 777 389 476 230 530 25.6 31.9
*
Chapter 5. Conwip
[%] 1 2 1 2 1 2 1 2 1 2
72 lower 0.5 487 475 153 148 226 222 100 98 19.0 18.7
upper 671 667 159 155 236 233 284 288 19.6 19.3
lower 1 521 545 258 195 226 226 28 112 19.0 19.0
upper 659 721 265 207 237 237 165 289 19.6 19.6
lower 2 500 464 204 151 221 222 67 84 18.7 18.8
upper 667 618 210 157 229 232 236 236 19.1 19.4
82 lower 0.5 681 792 297 178 274 287 91 307 20.5 21.7
upper 895 1072 316 199 285 303 313 590 21.1 22.6
lower 1 690 773 350 117 270 275 51 369 20.2 20.6
upper 851 1099 370 129 280 289 220 695 20.7 21.3
lower 2 699 733 360 181 283 282 38 251 21.2 21.3
upper 880 1196 378 201 298 297 223 717 22.0 22.1
92 lower 0.5 1171 1422 629 616 364 428 101 325 24.2 29.1
upper 1353 1685 738 671 378 469 313 599 24.9 31.6
lower 1 1182 1506 577 803 360 446 152 187 24.0 30.1
upper 1387 1693 683 876 373 500 424 388 24.7 33.4
lower 2 1179 1477 656 725 374 437 71 261 24.8 29.7
upper 1310 1761 753 782 390 480 244 554 25.6 32.1
*
105
106
[%] 1 2 1 2 1 2 1 2 1 2
72 lower 0.5 487 475 153 148 226 222 100 98 19.0 18.7
upper 671 667 159 155 236 233 284 288 19.6 19.3
lower 1 521 545 258 195 226 226 28 112 19.0 19.0
upper 659 721 265 207 237 237 165 289 19.6 19.6
lower 2 500 464 204 151 221 222 67 84 18.7 18.8
upper 667 618 210 157 229 232 236 236 19.1 19.4
82 lower 0.5 681 792 297 178 274 287 91 307 20.5 21.7
upper 895 1072 316 199 285 303 313 590 21.1 22.6
lower 1 690 773 350 117 270 275 51 369 20.2 20.6
upper 851 1099 370 129 280 289 220 695 20.7 21.3
lower 2 699 733 360 181 283 282 38 251 21.2 21.3
upper 880 1196 378 201 298 297 223 717 22.0 22.1
92 lower 0.5 1142 1380 612 607 362 421 68 282 24.1 28.7
upper 1381 1726 756 679 380 475 346 642 25.0 31.9
lower 1 1150 1477 560 792 358 437 109 155 23.9 29.6
upper 1419 1722 700 887 375 508 467 419 24.8 33.9
lower 2 1158 1433 640 716 372 430 44 215 24.7 29.3
upper 1330 1805 769 791 393 486 271 600 25.8 32.5
*
Chapter 5. Conwip
[%] 1 2 1 2 1 2 1 2 1 2
72 lower 0.5 487 475 153 148 226 222 100 98 19.0 18.7
upper 671 667 159 155 236 233 284 288 19.6 19.3
lower 1 521 545 258 195 226 226 28 112 19.0 19.0
upper 659 721 265 207 237 237 165 289 19.6 19.6
lower 2 500 464 204 151 221 222 67 84 18.7 18.8
upper 667 618 210 157 229 232 236 236 19.1 19.4
82 lower 0.5 681 792 297 178 274 287 91 307 20.5 21.7
upper 895 1072 316 199 285 303 313 590 21.1 22.6
lower 1 690 773 350 117 270 275 51 369 20.2 20.6
upper 851 1099 370 129 280 289 220 695 20.7 21.3
lower 2 699 733 360 181 283 282 38 251 21.2 21.3
upper 880 1196 378 201 298 297 223 717 22.0 22.1
92 lower 0.5 1109 1332 592 597 359 414 29 232 24.0 28.3
upper 1414 1774 776 689 383 482 385 692 25.1 32.4
lower 1 1113 1442 541 778 356 427 60 118 23.8 29.0
upper 1456 1757 720 901 377 518 516 456 24.9 34.5
lower 2 1135 1381 622 705 369 422 13 162 24.6 28.9
upper 1354 1857 786 801 396 494 303 653 25.9 32.9
*
107
Chapter 6
Dispatching Rules
In this chapter the influence of different dispatching rules on the average
production lead time is investigated. Two theorems based on covariance be-
tween processing time and production lead time are formulated and proved
theoretically. Theorem 1 links the average production lead time to the
”processing time weighted production lead time” for multi-stage production
systems analytically. The influence of different dispatching rules on average
production lead time, which is well-known from simulation and empirical
studies, can be proved theoretically in Theorem 2 for a single-stage produc-
tion system. A simulation study has been conducted to gain more insight
into the influence of dispatching rules on average production lead time in a
multi-stage production system. It has been found out that the ”processing
time weighted average production lead time” for a multi-stage production
system is not invariant with the applied dispatching rule and can be used
as an indicator independent of dispatching rules for single-stage production
systems.
6.1 Model Framework

A multi-stage production system consisting of m processing units including
a buffer right before each processing unit, as illustrated in Figure 6.1, is
discussed. Input stream 1 of production orders at the first machine is part
of the production system, whereby changing input streams lead to a different
production system (e.g. dispatching rule). At each buffer the production
orders are sorted according to their dispatching rule.
Figure 6.1: Multi-stage production system
Only one order at a time can be processed at a processing unit. In the

time period observed, n orders leave the production system and for each or-
der the logistical figures processing time, production lead time and queuing
time in front of the processing unit are evaluated for each sub-production
system. Processing time is the time an order spends in the processing unit.

DOI 10.1007/978-3-658-19120-7_6
110 Chapter 6. Dispatching Rules
Production lead time of an order is the time span between entering the
buffer in front of the processing unit and leaving the processing unit. The
queuing time denotes the production lead time minus the processing time
and can be interpreted as the waiting time (while not being processed) in
the buffer stock. Processing time, production lead time and queuing time
of order i at stage α are random variables denoted with Pi,α , Li,α and Qi,α ,
respectively. The terminology used in this chapter is presented in Table 6.1.
There is a lot of literature available examining the proof of Little’s Law
(Little, 1961; Eilon, 1969; Medhi, 1991; Hopp and Spearman, 2008). How-
ever, all of this literature discusses the relationship between expected pro-
duction lead time and expected inventory related to the processing rate, but
none discusses the ”processing time weighted production lead time”. For
the expected production lead time, simulation studies show that relationship
between expected production lead time and expected inventory depends on
the applied dispatching rule. The above mentioned proofs for Little’s Law
are extended by a proof delivered in Jodlbauer and Stöcher (2006) which
shows that, for continuous input and output processes, the ”processing time
weighted average production lead time” is independent of the dispatching
rule in a single-stage production system. They also show that Little’s Law
still holds true for the ”processing time weighted average production lead
time”. Furthermore, their proofs are valid for monotonic step functions
which means that it also holds for a setting with discrete customer orders.
In Theorem 6.1.1, the relationship between average production lead time
and ”processing time weighted average production lead time” is formulated
using the covariance of production lead time and processing time.
Theorem 6.1.1 (Mulit-stage systems). For a multi-stage production sys-

tem, the following identities between average production lead time and pro-
cessing time weighted average production lead time hold true:
Cov[L, P ]
E[L] = l − (6.1)
E[P ]
Cov[P, Q] + V ar[P ]
=l− (6.2)
E[P ]
Proof 6.1.2 (Multi-stage systems).
Cov[L, P ] E[LP ] − E[L]E[P ]

E[L] + =E[L] +
E[P ] E[P ]
E[L]E[P ] + E[LP ] − E[L]E[P ]
=
E[P ]
E[LP ]
= =l (6.3)
E[P ]
6.1. Model Framework 111
Cov[L, P ] =Cov[P + Q, P ]
=Cov[P, P ] + Cov[Q, P ]
=V ar[P ] + Cov[Q, P ] (6.4)
If the input stream of the production system (see Figure 6.1) is changed,
then Cov[P, Q], l and finally E[L] changes subsequently. Therefore, gener-
ally E[Linput1 ]
= E[Linput2 ] holds true. If another dispatching rule is ap-
plied, the Cov[P, Q] changes and therefore
E[Ldispatchingrule1 ]
= E[Ldispatchingrule2 ] holds true. If dispatching rules
are applied, which do not use processing time information for priority cal-
culation, Cov[P, Q] is equal to 0. Therefore, E[Lrandom ] is defined:
V ar[P ]
E[Lrandom ] := lrandom − (6.5)
E[P ]
E[Lrandom ] can be interpreted as the expected production lead time of
dispatching rules which do not use any information about the processing
time. Nyhuis and Wiendahl (1999)(see Nyhuis and Wiendahl (2009) for
the English version) developed the same formula as Equation 6.5 in 1999,
in which the average production lead time equals the mean range minus
the squared coefficient of variation of the processing time multiplied by its
expected value (for details see Nyhuis and Wiendahl (2009) Equation 4.40).
Theorem 6.1.3 introduces dispatching rules to the model framework
based on the property that the ”processing time weighted average produc-
tion lead time” is independent of the dispatching rule applied. Moreover,
Theorem 6.1.3 holds true only for single-stage production systems and is
formulated for (i) random-like, (ii) LPT-like and (iii) SPT-like dispatching
rules.
Theorem 6.1.3 (Dispatching rules). For any dispatching rule D of a single-
stage production system with:
(i) Cov[P, Qrandlike ] = 0 is applied, E[LD ] = E[Lrandom ] holds true.
Especially, FIFO does not use any processing time information for
prioritising and therefore E[LF IF O ] = E[Lrandom ] holds true.
(ii) Cov[P, QLP T like ] < 0 is applied, E[LD ] > E[Lrandom ] holds true.
Especially, LPT prioritises long processing times and therefore
E[LLP T ] > E[Lrandom ] holds true.
(iii) Cov[P, QSP T like ] > 0 is applied, E[LD ] < E[Lrandom ] holds true.
Especially, SPT prioritises short processing times and therefore
E[LSP T ] < E[Lrandom ] holds true.
Proof 6.1.4 (Dispatching rules). Approximating E[LP ] and E[P ] with

their respective sample means (with n orders) results in the following equa-
tion, which is equivalent to the ”processing time weighted average production
lead time” of Jodlbauer and Stöcher (2006) for a single-stage production
system:
1

n
n Pi L i
E[LP ] i=1
l= ≈
n (6.6)
E[P ] 1
n Pi
i=1
According to Jodlbauer and Stöcher (2006) l does not depend on the dis-
patching rule used for single-stage production systems.
(i)
Cov[P, QD ] + V ar[P ]
E[LD ] = l −
(6.2) E[P ]
V ar[P ]
= l− = E[Lrandom ] (6.7)
Cov[P,QD ]=0 E[P ] (6.5)
(ii)
E[LD ] = l −
(6.2) E[P ]
V ar[P ]
> l− = E[Lrandom ] (6.8)
Cov[P,QD ]<0 E[P ] (6.5)
(iii)
E[LD ] = l −
(6.2) E[P ]
V ar[P ]
< l− = E[Lrandom ] (6.9)
Cov[P,QD ]>0 E[P ] (6.5)
The formula proven in Theorem 6.1.3 confirms the relationship presented

in Nyhuis and Wiendahl (2009) for a FIFO system. For other dispatching
rules, Nyhuis and Wiendahl (2009) show by simulation studies that the
average lead time is longer for Longest Processing Time (LPT) rule than
for FIFO. Furthermore, the average lead time is shorter for the SPT policy
than for FIFO. This connection is also well-known from queuing analysis
(Buzacott and Shanthikumar, 1993) and scheduling literature (Blazewicz
et al., 2007; Pinedo, 2008).
6.2 Simulation Study

Due to the fact that Theorem 6.1.3 is only proven for single-stage produc-
tion systems, a simulation study is conducted to analyse Theorem 6.1.3 in
a multi-stage production system. Therefore, twelve different dispatching
rules are compared for a multi-stage production system consisting of four
machines. AnyLogic 6.8.0 is used as discrete event simulator and the model
presented by Hübl et al. (2011) is applied. The model has been validated,
as proposed in Kleijnen (1995), by comparing the analytic results of Al-
tendorfer and Jodlbauer (2011) calculating E[L] of the single-stage FIFO
production system with simulation results.
In this simulation study a Poisson arrival process with an arrival rate
of 25 orders per period is assumed. Simulation length is set to 68 periods
and 20 000 replications are conducted. It is ensured that all orders are
completed (Pinedo, 2008). One order requests one finished item and one
order can be processed at a time in each machine. Customer required lead
time distribution is distributed log-normal
√ with an expected value of 5 time
units and standard deviation of 2 time units. Moreover, processing times
are distributed log-normal
√ with an expected value of 0.4 time units and a
standard deviation of 0.4 time units. The processing times include any
additional random times (e.g. set-up times) but no sequence based times.
For practical reasons long processing times are projected to 4 time units.
Processing times are independent for each production stage.
For each order i, the random variables Pi,α , Li,α and Qi,α are mea-
sured. To gain more insight, the correlation coefficients ρ[Pα , Qα ] for the
four-machine model are presented in Table 6.2. The comparison presented
in this section is linked to the correlation coefficients since they indicate,
in a standardised way, how strong a relationship is. However, the real co-
variances have to be taken into account to calculate the average production
lead time from Equation 6.1 and Equation 6.2. Therefore, the covariance
values are provided in the appendix.
The following twelve dispatching rules are tested and grouped into four
groups according to Theorem 6.1.3:
(i) No processing time information for prioritising orders:
– Random: orders are sorted randomly
– FIFO (First in First out): orders are sorted according to their
entering date
– EDD (Earliest Due Date): orders are sorted based on their due
date
(ii) Jobs with long processing times are prioritised and queuing time is
long:
– LSK/RO (Least Slack per Remaining Operation): orders are

ranked based on the total remaining time to due date minus
the sum of all processing times of the remaining processing steps
divided by the number of remaining processing steps to finish the
order; the order with the least slack time per operation is ranked
first
– LSK (Least Slack): orders are ranked based on the total remain-
ing time to due date minus the sum of processing times of the
remaining processing steps; the order with the least slack time is
ranked first
– CR (Critical Ratio): orders are sorted according to their ratio
processing time divided by remaining time to due date; the order
with the highest ratio is ranked first, whereby already late orders
are also ranked ahead of the queue
– LPT (Longest Processing Time): orders are sorted according to
the longest processing time
– LPT-o: orders are sorted based on the sum of all processing
times in the production system (not only the remaining ones),
the highest sum is ranked first
(iii) Jobs with short processing times are prioritised and queuing time is
short:
– SPT (Shortest Processing Time): orders are sorted according to

the shortest processing time
– SPT-o: orders are sorted based on the sum of all processing
times (not only the remaining ones) in the production system;
the lowest sum is ranked first
– SRPT (Shortest Remaining Processing Time): orders are ranked
based on the sum of the remaining processing time to fulfil the or-
der at all remaining processing steps; the order with the shortest
remaining processing time is ranked first
(iv) Dispatching rule for practical use:
– SPT Slack: orders are sorted according to the shortest process-

ing time and if they have exceed the due date, the orders are
prioritised according to their (negative) slack
Observation 6.2.1 (”Processing time weighted average production lead

time”). ”Processing time weighted average production lead time” at each
stage α depends on the arrival process
Table 6.2 summarises the results for the four-machine case, where l1 to
l4 indicate the ”processing time weighted average production lead time” for
each single stage of the four-machine production system. l12 , l123 and l1234
are the ”processing time weighted average production lead times” for the
stages one to two, one to three and the whole production system respectively.
ρ[P, Q] for stage 1 is the correlation coefficient between the processing time
at stage 1 and the queuing time at stage 1. ρ[P, Q] for the stages 1234 is
the correlation coefficient between the sum of processing times of all stages
and the sum of the waiting time of all stages.
The ”processing time weighted average production lead time” of the first
stage (single-machine production system) is independent of the dispatching
rule as shown in Nyhuis and Wiendahl (2009), Jodlbauer (2005) and Jodl-
bauer and Stöcher (2006). For all other machines in the flow shop (using
the same dispatching rule at each stage), it has been observed that the
”processing time weighted average production lead time” depends on the
dispatching rule used and is increasing over each production state (except
for the dispatching rule SRPT). Especially for LPT an interesting gap be-
tween machine 1 and machine 4 can be observed. The results explicitly
show for all dispatching rules that l1234 is not the sum of l1 to l4 .
For group (i) l is approximately equal at each stage because no processing
time information is used. Moreover, ρ[Pα , Qα ] is 0 for dispatching rules of
group (i) for the single-machine stages. For the multi-stage systems the
covariance term is unequal to 0, which, indeed, makes the analytical proof
of Theorem 6.1.3 for the multi-machine case unfeasible in the current setting.
Category (iv) in Table 6.2 presents the SPT rule in combination with a
slack rule which increases the practical relevance of the SPT dispatching rule
because due date is included in the priority calculation of orders. Based on
this combination it fits into none of the groups discussed in Theorem 6.1.3.
The correlation coefficient for stages 1 and 2 and the multi-stage production
system 12 are positive, which shows a dominant SPT-like behaviour. For
the multi-stage production system 123 and 1234 the slack behaviour gets
dominant in order to fulfil the due dates. The switch between slack and
SPT-like behaviour depends on the switching point between SPT and least
slack rule. In this case the rule switches if orders have exceeded the due
date but any other scenario is possible.
It is conjectured that the input stream of the downstream sub-production
system is influenced by the dispatching of the upstream sub-production sys-
tem. If, for example, LPT is chosen, in the beginning the whole system is
empty and the orders arrive according to a Poisson process at the queue of
the first machine. The order with the longest processing time is released
into the processing at the first machine. However, on the second process-
ing unit no orders are waiting and the finished order at processing unit 1
has no waiting time in front of processing unit 2. Due to the fact that the
processing times are independent, the processing time of the finished order
at processing unit 1 is, with a high probability, higher than the processing
time of processing unit 2. As defined in Figure 6.1, the production system
boundaries include the input stream. If another dispatching rule is applied,
the inputstream changes and a different production system results.
Observation 6.2.2 (Multi-stage production systems). According to the

simulation results Theorem 6.1.3 also holds true for the multi-stage cases
Table 6.5 and Table 6.4 indicate expected value and coefficient of vari-
ation of L for single- and multi-stage production systems. cv [L] for stage 1
for the dispatching rule Random is 2.59. E[L] for the first two stages of the
use of SRPT as dispatching rule is 3.71.
According to Theorem 6.1.3, E[Li ] is equal for all dispatching rules,
where no information about the processing time is used for prioritising jobs
based on a single-stage setting, which is indeed the case for dispatching rules
of group (i). The same behaviour is observed for the multi-stage cases.
For the tested single- and multi-stage production systems where LPT-
like dispatching rules are applied, E[L(ii) ] > E[L(i) ] holds true. Moreover,
if SPT-like dispatching rules are used then E[L(iii) ] < E[L(i) ] holds true for
single- and multi-stage production systems.
cv [L] for a random dispatching rule is higher than for FIFO because the
orders are sorted randomly which has no effect on the expected value but
on the variance and therefore the coefficient of variation is increasing due to
the sorting process. The variance of the customer required lead time effects
the results of cv [L] for the dispatching rule EDD. However, in this case only
little effect on cv [L] compared to FIFO can be observed.
It is not advisable to apply dispatching rules with a high cv [L] (Random,
LPT, LPT-o, SPT, SPT-o, SRTP) because the reliability of the production
schedule is low. Moreover, the more cv [L] is increasing, the more the oper-
ating curves deviate from the ideal operating curves (Nyhuis and Wiendahl,
2009).
Based on Observation 6.2.2 it is assumed that Theorem 6.1.3 holds true
in general for a multi-stage setting and that Theorem 6.1.3 can therefore
be extended. Hence, the analytic proof needs to be investigated in further
research.

In this chapter, a theorem for calculating the average production lead time
based on the covariance of processing and queuing time, and the ”processing
time weighted average production lead time” is formulated for multi-stage
production systems. For single-stage production systems the influence of
different dispatching rules on average production is theoretically proven.
Moreover, a simulation study for a multi-stage production system has
been conducted. One of the main findings is that the ”processing time
weighted average production lead time” for a multi-stage production system
is not invariant with respect to the applied dispatching rule, because the
dispatching rules in multi-stage setting will influence the input stream of
the production orders for the next sub-production system. Nevertheless,
simulation results show that Theorem 6.1.3 holds true for the tested multi-
stage production system and it is assumed that Theorem 6.1.3 holds true
for multi-stage production systems in general.
For practical application, it has been identified that the ”processing
time weighted average production lead time” can be used as a dispatch-
ing rule independent indicator for single-stage production systems. Based
on this measure, new approximation possibilities for production lead time
when applying certain dispatching rules have been developed based on the
covariance of processing and queuing time.
Table 6.1: Definitions
Symbol Description
n Number of orders leaving the system
m Number of machines in the production system
Pα Random variable of processing time with its
respective realisation Pi,α of order i at ma-
chine α including additional random times
(e.g. set-up times) but no sequence based
times
Lα Random variable of production lead time with
its respective realisation Li,α of order i at ma-
chine α
Qα := Lα − Pα Random variable of queuing time with its re-
spective realisation Qi,α of order i at machine
α
E[Lα Pα ]
lα := ”Processing time weighted average production
E[Pα ]
lead time” at machine α according to Jodl-
bauer and Stöcher (2006)

m
P := Pα Random variable of overall processing time
α=1
m
L := Lα Random variable of overall production lead
α=1 time

m
Q := L − P = Qα Random variable of overall queuing time
α=1
E[LP ]
l := ”Processing time weighted average production
E[P ]
lead time” according to Jodlbauer and Stöcher
(2006) extended to a general production sys-
tem (single- and multi-stage production sys-
tem)
E[.] Expected value of a random variable
V ar[.] Variance of a random variable
V ar[.]
cv [.] := Coefficient of variation of a random variable
E[.]
ρ[., .] Correlation coefficient of two random variables
Cov[., .] Covariance of two random variables
Table 6.2: ”Processing time weighted average production lead time”
l for the following stages

Dispatching rule 1 2 3 4 12 123 1234
i Random 3.91 4.35 4.49 4.56 7.40 11.07 14.84
FIFO 3.91 4.36 4.49 4.57 7.32 10.97 14.75
EDD 3.91 4.36 4.50 4.57 7.32 10.97 14.75
ii LSK/RO 3.91 4.41 4.53 4.60 7.46 11.27 15.17
LSK 3.91 4.41 4.55 4.60 7.52 11.33 15.23
CR 3.91 4.49 4.62 4.63 7.84 11.66 15.45
LPT 3.91 8.09 13.45 19.37 16.58 42.53 83.00
LPT-o 3.91 4.54 4.76 4.87 7.63 11.61 15.72
iii SPT 3.91 4.03 4.04 4.04 5.36 6.60 8.25
SPT-o 3.91 4.06 4.08 4.09 7.00 10.23 13.54
SRPT 3.91 4.06 4.05 4.02 6.44 8.88 11.16
iv SPT Slack 3.91 4.39 4.62 4.70 6.46 9.65 13.28
st nd rd th
1 ... 1 stage; 2 ... 2 stage; 3 ... 3 stage; 4 ... 4 stage,
12 ... stages 1 to 2; 123 ... stages 1 to 3; 1234 ... stages 1 to 4
Table 6.3: Correlation coefficients ρ[P, Q]
ρ[Pα , Qα ] for the following stages

i Random 0.00 0.00 0.00 0.00 -0.01 -0.02 -0.03
FIFO 0.00 0.00 0.00 0.00 -0.05 -0.07 -0.09
EDD 0.00 0.00 0.00 0.00 -0.05 -0.07 -0.09
ii LSK/RO -0.06 -0.02 -0.02 -0.01 -0.10 -0.12 -0.13
LSK -0.06 -0.04 -0.02 -0.01 -0.12 -0.14 -0.15
CR -0.15 -0.09 -0.03 -0.01 -0.33 -0.34 -0.33
LPT -0.19 -0.23 -0.24 -0.25 -0.10 -0.20 -0.01
LPT-o -0.12 -0.14 -0.15 -0.15 -0.26 -0.41 -0.57
iii SPT 0.45 0.47 0.47 0.48 0.46 0.45 0.44
SPT-o 0.20 0.23 0.23 0.23 0.38 0.59 0.82
SRPT 0.20 0.26 0.32 0.48 0.42 0.54 0.68
iv SPT Slack 0.22 0.28 0.03 0.00 0.35 -0.39 -0.45
st nd rd th
Table 6.4: cv [L] for single- and multi-stage production systems
cv [L] for the following stages

i Random 2.59 2.55 2.55 2.58 1.86 1.54 1.35
FIFO 1.06 1.03 1.02 1.03 0.72 0.57 0.48
EDD 1.10 1.04 1.03 1.03 0.74 0.58 0.49
ii LSK/RO 1.09 1.50 1.12 1.08 0.94 0.65 0.52
LSK 1.09 1.02 1.02 1.03 0.72 0.56 0.47
CR 1.44 1.19 1.08 1.03 0.88 0.66 0.54
LPT 2.27 1.95 1.87 1.80 1.64 1.36 1.26
LPT-o 2.49 2.20 2.09 2.05 1.21 1.23 1.47
iii SPT 2.28 2.22 2.18 2.16 1.60 1.29 1.11
SPT-o 2.54 2.58 2.60 2.60 2.09 1.91 1.81
SRPT 2.54 2.57 2.56 2.41 2.04 1.81 1.65
iv SPT Slack 1.40 1.49 1.33 1.19 1.06 0.79 0.27
st nd rd th
Table 6.5: E[L] for single- and multi-stage production systems
E[L] for the following stages

i Random 3.20 3.65 3.78 3.86 6.85 10.63 14.49
FIFO 3.20 3.65 3.79 3.86 6.85 10.64 14.50
EDD 3.20 3.65 3.79 3.86 6.85 10.64 14.50
ii LSK/RO 3.53 3.82 3.97 3.96 7.35 11.32 15.28
LSK 3.53 3.90 3.94 3.94 7.43 11.37 15.32
CR 4.55 4.46 4.11 3.97 9.01 13.12 17.09
LPT 7.66 18.76 33.25 48.16 26.46 59.70 107.86
LPT-o 5.42 6.60 6.96 7.11 12.03 18.99 26.10
iii SPT 1.40 1.45 1.46 1.47 2.85 4.31 5.78
SPT-o 1.91 1.90 1.88 1.88 3.81 5.69 7.58
SRPT 1.91 1.81 1.62 1.36 3.71 5.34 6.71
iv SPT Slack 2.29 2.63 3.18 3.58 4.66 7.84 26.09
1 ... 1st stage; 2 ... 2nd stage; 3 ... 3rd stage; 4 ... 4th stage,
Chapter 7
Conclusions
This thesis has investigated production system behaviour such as production
lead time and JIT intensity and has improved capacity setting and the
production planning method Conwip. The topics are focusing on the main
intra-company planning tasks, such as Master Planning, Demand Planning,
Production Planning and Scheduling of the Supply Chain Planning Matrix.
Chapter 3 focuses on the hierarchical planning structure by investigating
the optimal treatment of the metrics utilisation on each hierarchical plan-
ning level. It has been identified that controversial objectives in long term,
medium term and short term planning result in the following advices. For
long term planning tasks high utilisation due to accounting leads to a high
JIT intensity. Medium term planning, however, focuses on the flexibility
of the workforce to compensate customer fluctuations which again results
in a maximisation of utilisation. It is shown that the short term decisions
avoiding waste based on the ”seven zero” philosophy lead to minimising the
metric utilisation. This results in the main finding of this chapter that if a
high JIT intensity is reached, it is possible to reduce excess capacity or sales
can be increased with no additional investments. In further research, an em-
pirical study to confirm the presented relationship between JIT intensity,
utilisation and profit/revenue should be carried out.
In stochastic environments, where practitioners have to deal with cus-
tomer and process uncertainties, information processing is an important
issue to ensure a competitive environment for the manufacturing compa-
nies. In Chapter 4 a decision support system for setting the capacity pro-
vided based on the capacity demanded is developed. Based on the informa-
tion available due to enterprise resource planning systems, production data
acquisition systems and/or manufacturing execution systems, the decision
maker can add or neglect information for processing time uncertainties or
customer uncertainties to determine the capacity demanded. Depending
on capacity demanded some methods supporting the Master Planning for
calculating the capacity provided are developed. The main finding in this
chapter is that methods, which use both processing time distribution and
the customer required lead time distribution, are most efficient. It is shown,
however, that too much capacity flexibility can lead to overreactions and
the performance measured by service level and tardiness decreases. In this
chapter also a simplified production system consisting of an M/M/1 queu-
ing system is investigated. This system is extended so that the system
can switch from a low production rate to a high production rate when the
WIP exceeds a certain level. An approximation for the production lead

DOI 10.1007/978-3-658-19120-7_7
122 Chapter 7. Conclusions
time is provided to calculate expected production lead time, expected WIP,

expected backorders and service level analytically. Since the transient be-
haviour at the switching point is neglected the error of this assumption is
analysed. An extensive parameter optimisation search for the best method
to give some guidelines about optimal parameter sets has been set aside for
further research. Moreover, based on the density function of production lead
time an optimisation problem for identifying the optimal switching point for
the M/M/1 is suggested for further research
Chapter 5 is dedicated to the module Production Planning of the SCP
matrix by focusing on Conwip. The classical Conwip system where the
WIP is constrained by the Wipcap is compared to a Conwip system where
the Wipcap constrains WIP and FGI. Moreover, a safety stock for finished
items is implemented for both scenarios. The main finding is that the tested
Conwip systems where only the WIP is constrained outperforms a system
where WIP and FGI are constrained. The safety stock is not needed in
the current setting of the production system. It is shown that the work-
ahead-window is almost independent of the customer required lead time
distribution for the minimum cost scenarios. Further studies where different
configurations of production systems in terms of complexity (e.g. stochastic
parameters, amount of machines) are suggested for further research since
due to computational power only a limited set of scenarios could have been
tested.
The scheduling module of the SCP matrix is focused in Chapter 6. Since
most real-world applications are NP-hard problems dispatching rules are
applied in practice very often. A theorem for multi-stage production systems
for calculating the average production lead time based on the covariance of
processing and queuing time and the ”processing time weighted average
production lead time” is formulated. For single-stage production systems
the influence of different dispatching rules on average production lead time
is proven analytically. One of the main findings is that the ”processing time
weighted average production lead time” for multi-stage production systems
is not invariant of the applied dispatching rule. The reason for that is that
the dispatching rules reorder the input stream in the next subsystem. A
managerial insight is that the ”processing time weighted average production
lead time” serves as a dispatching independent indicator for single-stage
production systems for identifying the behaviour when applying a certain
dispatching rule. The analytical proof of Theorem 6.1.3 for multi-stage
production systems needs to be postponed to further research.
Bibliography
Abdul-Kader, W. and A. Gharbi, 2002: Capacity estimation of a multi-
product unreliable production line. International Journal of Production
Research, 40 (18), 4815–4834.
Ahuja, I. and J. Khamba, 2008: Total productive maintenance: literature
review and directions. International Journal of Quality & Reliability Man-
agement, 25 (7), 709–756, doi: 10.1108/02656710810890890.
Altendorfer, K., T. Felberbauer, D. Gruber, and A. Hübl, 2013: Applica-
tion of a generic simulation model to optimize production and workforce
planning at an automotive supplier. Proceedings of the 2013 Winter Sim-
ulation Conference, 2689–2697.
Altendorfer, K., A. Hübl, and H. Jodlbauer, 2014: Periodical capacity set-
ting methods for make-to-order multi-machine production systems. In-
ternational Journal of Production Research, doi: 10.1080/00207543.2014.
886822.
Altendorfer, K. and H. Jodlbauer, 2011: An analytical model for ser-
vice level and tardiness in a single machine mto production system.
International Journal of Production Research, 49 (7), 1827–1850, doi:
10.1080/00207541003660176.
Altendorfer, K. and S. Minner, 2011: Simultaneous optimization of capacity
and planned lead time in a two-stage production system with different
customer due dates. European Journal of Operational Research, 213 (1),
134–146, doi: 10.1016/j.ejor.2011.03.006.
Angelus, A. and E. L. Porteus, 2002: Simultaneous capacity and production
management of short-life-cycle, produce-to-stock goods under stochastic
demand. Management Science, 48 (3), 399–413, doi: 10.1287/mnsc.48.3.
399.7726.
Arakawa, M., M. Fuyuki, and I. Inoue, 2003: An optimization-oriented
method for simulation-based job shop scheduling incorporating capac-
ity adjustment function. International Journal of Production Economics,
85 (3), 359–369, doi: 10.1016/S0925-5273(03)00122-1.
Arrow, K. J., T. Harris, and J. Marschak, 1951: Optimal inventory policy.
Econometrica, 19 (3), 250–272, doi: 10.2307/1906813.
Aspinwall, E. and M. Elgharib, 2013: Tpm implementation in large and
medium size organisations. Journal of Manufacturing Technology Man-
agement, 24 (5), 688–710, doi: 10.1108/17410381311327972.

DOI 10.1007/978-3-658-19120-7
124 Bibliography
Bahaji, N. and M. E. Kuhl, 2008: A simulation study of new multi-objective

composite dispatching rules, conwip, and push lot release in semiconduc-
tor fabrication. International Journal of Production Research, 46 (14),
3801–3824, doi: 10.1080/00207540600711879.
Balakrishnan, N., J. W. Patterson, and V. Sridharan, 1996: Rationing ca-

pacity between two product classes. Decision Sciences, 27 (2), 185–214,
doi: 10.1111/j.1540-5915.1996.tb00850.x.
Balakrishnan, N., J. W. Patterson, and V. Sridharan, 1999: Robustness

of capacity rationing policies. European Journal of Operational Research,
115 (2), 328–338, doi: 10.1016/S0377-2217(98)00208-2.
Barrett, R. T. and S. N. Kadipasaoglu, 1990: Dispatching rules for a dy-

namic flow shop. Production & Inventory Management Journal, 31 (1),
54–58.
Bartezzaghi, E. and R. Verganti, 1995: Managing demand uncertainty

through order overplanning. International Journal of Production Eco-
nomics, 40 (2/3), 107–120, doi: 10.1016/0925-5273(95)00047-8.
Bechte, W., 1988: Theory and practice of load-oriented manufacturing con-

trol. International Journal of Production Research, 26 (3), 375–395, doi:
10.1080/00207548808947871.
Bednowitz, N., R. Batta, and R. Nagi, 2014: Dispatching and loitering

policies for unmanned aerial vehicles under dynamically arriving multiple
priority targets. Journal of Simulation, 8 (1), 9–24.
Bitran, G. R. and A. C. Hax, 1977: On the design of hierarchical produc-

tion planning systems. Decision Sciences, 8 (1), 28–55, doi: 10.1111/j.
1540-5915.1977.tb01066.x.
Blackstone, J. H., D. T. Phillips, and G. L. Hogg, 1982: A state-of-the-

art survey of dispatching rules for manufacturing job shop operations.
International Journal of Production Research, 20 (1), 27–45.
Blazewicz, J., K. H. Ecker, E. Pesch, G. Schmidt, and J. Weglarz, 2007:

Handbook on Scheduling: From Theory to Applications. International
Handbook on Information Systems, Springer-Verlag, Berlin Heidelberg,
doi: 10.1007/978-3-540-32220-7.
Bradley, J. R. and P. W. Glynn, 2002: Managing capacity and inventory

jointly in manufacturing systems. Management Science, 48 (2), 273–288,
doi: 10.1287/mnsc.48.2.273.254.
Bibliography 125
Brown, K. A. and T. R. Mitchell, 1991: A comparison of just-in-time and

batch manufacturing: The role of performance obstacles. Academy of
Management Journal, 34 (4), 906–917, doi: 10.2307/256395.
Buyukkaramikli, N. C., J. Bertrand, and H. P. G. v. Ooijen, 2013: Pe-

riodic capacity management under a lead-time performance constraint.
OR Spectrum, 35 (1), 221–249, doi: 10.1007/s00291-011-0261-4.
Buzacott, J. A. and J. G. Shanthikumar, 1993: Stochastic models of man-

ufacturing systems. Prentice-Hall International series in industrial and
systems engineering, Prentice-Hall, Englewood Cliffs NJ.
Callen, J. L., M. Morel, and C. Fader, 2005: Productivity measure-

ment and the relationship between plant performance and jit inten-
sity. Contemporary Accounting Research, 22 (2), 271–309, doi: 10.1506/
GU78-6EDM-1G36-4YBQ.
Carrillo, J. E. and C. Gaimon, 2000: Improving manufacturing performance

through process change and knowledge creation. Management Science,
46 (2), 265–288, doi: 10.1287/mnsc.46.2.265.119.
Chakrabortty, R. K. and M. A. A. Hasin, 2013: Solving an aggregate

production planning problem by using multi-objective genetic algorithm
(moga) approach. International Journal of Industrial Engineering Com-
putations, 4 (1), 1–12, doi: 10.5267/j.ijiec.2012.09.003.
Chen, B. and T. I. Matis, 2013: A flexible dispatching rule for minimiz-

ing tardiness in job shop scheduling. International Journal of Production
Economics, 141 (1), 360–365, doi: 10.1016/j.ijpe.2012.08.019.
Chenery, H. B., 1952: Overcapacity and the acceleration principle. Econo-

metrica, 20 (1), 1–28.
Chong, M. Y., S. A. Mustafa, J. Prakash, and J. F. Chin, 2013: Parallel

kanban-conwip for production system with rework entrance. International
Journal of Manufacturing Research, 8 (4), 357–377, doi: 10.1504/IJMR.
2013.057747.
Co, H. C. and M. Sharafali, 1997: Overplanning factor in toyota’s formula

for computing the number of kanban. IIE Transactions, 29 (5), 409–415,
doi: 10.1080/07408179708966346.
Corti, D., A. Pozzetti, and M. Zorzini, 2006: A capacity-driven approach to

establish reliable due dates in a mto environment. International Journal
of Production Economics, 104 (2), 536–554, doi: 10.1016/j.ijpe.2005.03.
003.
126 Bibliography
Cox, J. F. and J. H. Blackstone, 2002: APICS dictionary. 10th ed., APICS,

Falls Church (Va.).
Cua, K. O., K. E. McKone, and R. G. Schroeder, 2001: Relationships

between implementation of tqm, jit, and tpm and manufacturing per-
formance. Journal of Operations Management, 19 (6), 675–694, doi:
10.1016/S0272-6963(01)00066-3.
Daugherty, P. J., D. S. Rogers, and M. S. Spencer, 1994: Just-in-time

functional model: Empirical test and validation. International Journal
of Physical Distribution & Logistics Management, 24 (6), 20–26, doi:
10.1108/09600039410066150.
Defregger, F. and H. Kuhn, 2006: Revenue management for a make-to-order

company with limited inventory capacity. OR Spectrum, 29 (1), 137–156,
doi: 10.1007/s00291-005-0016-1.
Deming, W. E., 2000: Out of the crisis. 1st ed., Cambridge Univ. Pr, Cam-
bridge Mass.
Duenyas, I., W. J. Hopp, and M. L. Spearman, 1993: Characterizing the

output process of a conwip line with deterministic processing and random
outages. Management Science, 39 (8), 975–988, doi: 10.1287/mnsc.39.8.
975.
Edwards, J., 1983: Mrp and kanban - american style. Twenty-sixth annual
international conference proceedings, 586–603.
Eilon, S., 1969: A simpler proof of L = λW . Operations Research, 17 (5),

915–917.
El-Bouri, A., S. Balakrishnan, and N. Popplewell, 2008: Cooperative dis-

patching for minimizing mean flowtime in a dynamic flowshop: Special
section on advanced modeling and innovative design of supply chain. In-
ternational Journal of Production Economics, 113 (2), 819–833, doi:
10.1016/j.ijpe.2007.11.005.
Emde, S., M. Fliedner, and N. Boysen, 2012: Optimally loading tow trains
for just-in-time supply of mixed-model assembly lines. IIE Transactions,
44 (2), 121–135, doi: 10.1080/0740817X.2011.575442.
Felberbauer, T., K. Altendorfer, and A. Hübl, 2012: Using a scalable simu-

lation model to evaluate the performance of production system segmen-
tation in a combined mrp and kanban system. Proceedings of the 2012
Winter Simulation Conference, 1–12, doi: 10.1109/WSC.2012.6465053.
Bibliography 127
Finch, B., 1986: Japanese management techniques in small manufactur-

ing companies: A strategy for implementation. Production & Inventory
Management Journal, 27 (3), 30–38.
Fleischmann, B. and H. Meyr, 2003: Planning hierarchy, modeling and

advanced planning systems. Supply Chain Management: Design, Co-
ordination and Operation, Elsevier, Handbooks in Operations Research
and Management Science, Vol. 11, 455–523, doi: 10.1016/S0927-0507(03)
11009-2.
Flynn, B. B., S. Sakakibara, and R. G. Schroeder, 1995: Relationship be-

tween jit and tqm: Practices and performance. Academy of Management
Journal, 38 (5), 1325–1360, doi: 10.2307/256860.
Framinan, J. M., P. L. González, and R. Ruiz-Usano, 2003: The conwip pro-

duction control system: review and research issues. Production Planning
& Control, 14 (3), 255–265, doi: 10.1080/0953728031000102595.
Framinan, J. M., P. L. González, and R. Ruiz Usano, 2006: Dynamic card

controlling in a conwip system. International Journal of Production Eco-
nomics, 99 (1-2), 102–116, doi: 10.1016/j.ijpe.2004.12.010.
Framinan, J. M., R. Ruiz-Usano, and R. Leisten, 2000: Input control

and dispatching rules in a dynamic conwip flow-shop. International
Journal of Production Research, 38 (18), 4589–4598, doi: 10.1080/
00207540050205523.
Gansterer, M., C. Almeder, and R. F. Hartl, 2014: Simulation-based opti-

mization methods for setting production planning parameters. Interna-
tional Journal of Production Economics, 151, 206–213, doi: 10.1016/j.
ijpe.2013.10.016.
Geraghty, J. and C. Heavey, 2004: A comparison of hybrid push/pull and

conwip/pull production inventory control policies. International Journal
of Production Economics, 91 (1), 75–90, doi: 10.1016/S0925-5273(03)
00210-X.
Gershwin, S. B. and L. M. Werner, 2007: An approximate analytical method

for evaluating the performance of closed-loop flow systems with unreliable
machines and finite buffers. International Journal of Production Research,
45 (14), 3085–3111, doi: 10.1080/00207540500385980.
Gfrerer, H. and G. Zäpfel, 1995: Hierarchical model for production plan-

ning in the case of uncertain demand. European Journal of Operational
Research, 86 (1), 142–161, doi: 10.1016/0377-2217(95)00066-Y.
128 Bibliography
Gilland, W. G., 2002: A simulation study comparing performance of con-

wip and bottleneck-based release rules. Production Planning & Control,
13 (2), 211–219, doi: 10.1080/09537280110069784.
Goldratt, E. M., 1988: Computerized shop floor scheduling. Interna-

tional Journal of Production Research, 26 (3), 443–455, doi: 10.1080/
00207548808947875.
Golhar, D. Y. and C. L. Stamm, 1991: The just-in-time philosophy: A

literature review. International Journal of Production Research, 29 (4),
657–676, doi: 10.1080/00207549108930094.
Gong, Q., Y. Yang, and S. Wang, 2014: Information and decision-making

delays in mrp, kanban, and conwip. International Journal of Production
Economics, 156 (0), 208–213, doi: 10.1016/j.ijpe.2014.06.010.
Graves, S., 2011: Uncertainty and production planning. Planning Produc-

tion and Inventories in the Extended Enterprise, K. G. Kempf, P. Ke-
skinocak, and R. Uzsoy, Eds., Springer US, International Series in
Operations Research & Management Science, Vol. 151, 83–101, doi:
10.1007/978-1-4419-6485-4\ 5.
Graves, S. C., 1982: Using lagrangean techniques to solve hierarchical pro-

duction planning problems. Management Science, 28 (3), 260–275, doi:
10.1287/mnsc.28.3.260.
Grosfeld-Nir, A. and M. Magazine, 2002: Gated maxwip: A strategy for

controlling multistage production systems. International Journal of Pro-
duction Research, 40 (11), 2557–2567, doi: 10.1080/00207540210128251.
Gstettner, S. and H. Kuhn, 1996: Analysis of production control sys-

tems kanban and conwip. International Journal of Production Research,
34 (11), 3253–3273, doi: 10.1080/00207549608905087.
Hax, A. C. and H. C. Meal, 1975: Hierarchical integration of production

planning and scheduling. Studies in The Management Science, M. K.
Starr, Ed., North Holland American Elsevier, New York, Vol. 1, 53–69.
Hayya, J. C., T. P. Harrison, and X. J. He, 2011: The impact of stochastic

lead time reduction on inventory cost under order crossover. European
Journal of Operational Research, 211 (2), 274–281, doi: 10.1016/j.ejor.
2010.11.025.
Hegedus, M. G. and W. J. Hopp, 2001: Due date setting with supply

constraints in systems using mrp. Computers & Industrial Engineering,
39 (3-4), 293–305, doi: 10.1016/S0360-8352(01)00007-9.
Bibliography 129
Helber, S., K. Schimmelpfeng, and R. Stolletz, 2011: Setting inventory levels

of conwip flow lines via linear programming. BuR - Business Research,
4 (1), 98–115.
Heragu, S. S., X. Cai, A. Krishnamurthy, and C. J. Malmborg, 2011: An-

alytical models for analysis of automated warehouse material handling
systems. International Journal of Production Research, 49 (22), 6833–
6861, doi: 10.1080/00207543.2010.518994.
Herer, Y. T. and M. Masin, 1997: Mathematical programming formula-

tion of conwip based production lines; and relationships to mrp. In-
ternational Journal of Production Research, 35 (4), 1067–1076, doi:
10.1080/002075497195542.
Hines, P. and N. Rich, 1997: The seven value stream mapping tools. In-
ternational Journal of Operations & Production Management, 17 (1),
46–64, doi: 10.1108/01443579710157989.
Hodgson, T. J. and Dingwei Wang, 1991a: Optimal hybrid push/pull control

strategies for a parallel multistage system: Part i. International Journal
of Production Research, 29 (6), 1279.
Hodgson, T. J. and Dingwei Wang, 1991b: Optimal hybrid push/pull con-

trol strategies for parallel multistage system: Part ii. International Jour-
nal of Production Research, 29 (7), 1453.
Hopp, W. J. and M. L. Roof, 1998: Setting wip levels with statistical

throughput control (stc) in conwip production lines. International Journal
of Production Research, 36 (4), 867–882, doi: 10.1080/002075498193435.
Hopp, W. J. and M. L. Roof Sturgis, 2000: Quoting manufacturing due

dates subject to a service level constraint. IIE Transactions, 32 (9), 771–
784.
Hopp, W. J. and M. L. Spearman, 1996: Factory physics: Foundations of

manufacturing management. Irwin, Chicago.
Hopp, W. J. and M. L. Spearman, 2008: Factory physics. 3d ed., The

McGraw-Hill/Irwin series, Irwin/McGraw-Hill, New York.
Huang, G., J. Chen, X. Wang, and Y. Shi, 2015: A simulation study of

conwip assembly with multi-loop in mass production, multi-products and
low volume and okp environments. International Journal of Production
Research, 53 (14), 4160–4175, doi: 10.1080/00207543.2014.980458.
130 Bibliography
Huang, M., D. Wang, and W. H. Ip, 1998a: A simulation and comparative

study of the conwip, kanban and mrp production control systems in a
cold rolling plant. Production Planning & Control, 9 (8), 803–812, doi:
10.1080/095372898233579.
Huang, M., D. Wang, and W. H. Ip, 1998b: Simulation study of conwip

for a cold rolling plant. International Journal of Production Economics,
54 (3), 257–266, doi: 10.1016/S0925-5273(97)00152-7.
Hübl, A., 2014: Selected topics of the supply chain matrix under customer
and process uncertainties. Proceedings of the 2014 Winter Simulation
Conference, 4023–4024.
Hübl, A. and K. Altendorfer, 2014a: Inventory constraint definition for

conwip in a make-to-order environment – a simulation study. Working
Paper.
Hübl, A. and K. Altendorfer, 2014b: Queuing model for optimal switching

point for two capacity levels. Working Paper.
Hübl, A. and K. Altendorfer, 2015: State probabilities for an m/m/1 queu-

ing system with two capacity levels. Proceedings of the 2015 Winter Simu-
lation Conference, IEEE Press, Huntington Beach, California, 2219–2226.
Hübl, A., K. Altendorfer, H. Jodlbauer, M. Gansterer, and R. F. Hartl,

2011: Flexible model for analyzing production system with discrete event
simulation. Proceedings of the 2011 Winter Simulation Conference, 1554–
1565, doi: 10.1109/WSC.2011.6147873.
Hübl, A., K. Altendorfer, H. Jodlbauer, and J. Pilstl, 2010: Customer driven

capacity setting. Advances in Production Management Systems. New
Challenges, New Approaches, IFIP Advances in Information and Commu-
nication Technology, 338, 291–298, doi: 10.1007/978-3-642-16358-6\ 37.
Hübl, A., K. Altendorfer, J. Pilstl, and H. Jodlbauer, 2009: Equipment

performance measurement in production plants based on customer de-
mand. Proceedings of the IEEE 2nd International Symposium on Logis-
tics and Industrial Informatics (Lindi 2009), 1–6, doi: 10.1109/LINDI.
2009.5258750.
Hübl, A. and R. Gmainer, 2008: Critical discussion of oee. Proceedings FH

Science Day 2008, 321–329.
Hübl, A. and H. Jodlbauer, 2008: A combined approach for capacity check-

ing, due date setting and stock availability in an mto environment. Pro-
ceedings FH Science Day 2008, 372–381.
Bibliography 131
Hübl, A. and H. Jodlbauer, 2013: Optimal utilization based on costs and

revenue for a hierarchical decision model based on jit goals. Working
Paper.
Hübl, A., H. Jodlbauer, and K. Altendorfer, 2013: Influence of dispatching

rules on average production lead time for multi-stage production systems.
International Journal of Production Economics, 144 (2), 479–484, doi:
10.1016/j.ijpe.2013.03.020.
Hung, Y.-F. and I.-R. Chen, 1998: A simulation study of dispatch rules
for reducing flow times in semiconductor wafer fabrication. Production
Planning & Control, 9 (7), 714–722, doi: 10.1080/095372898233722.
Hung, Y.-F. and R. C. Leachman, 1999: Reduced simulation models of

wafer fabrication facilities. International Journal of Production Research,
37 (12), 2685–2701, doi: 10.1080/002075499190473.
Inman, R. A., R. S. Sale, K. W. J. Green, and D. Whitten, 2011: Ag-

ile manufacturing: Relation to jit, operational performance and firm
performance. Journal of Operations Management, 29 (4), 343–355, doi:
10.1016/j.jom.2010.06.001.
Jacobs, R. F. and F. C. T. Weston Jr., 2007: Enterprise resource planning

(erp)—a brief history. Special Issue Evolution of the Field of Operations
Management SI/ Special Issue Organisation Theory and Supply Chain
Management, 25 (2), 357–363, doi: 10.1016/j.jom.2006.11.005.
Jansen, M. M., T. G. d. Kok, and J. C. Fransoo, 2013: Lead time anticipa-

tion in supply chain operations planning. OR Spectrum, 35 (1), 251–290,
doi: 10.1007/s00291-011-0267-y.
Jayamohan, M. S. and C. Rajendran, 2000: New dispatching rules for shop

scheduling: A step forward. International Journal of Production Research,
38 (3), 563–586, doi: 10.1080/002075400189301.
Jodlbauer, H., 2005: Definition and properties of the input-weighted average

lead-time. European Journal of Operational Research, 164 (2), 354–357,
doi: 10.1016/j.ejor.2003.04.006.
Jodlbauer, H., 2008a: Customer driven production planning. International

Journal of Production Economics, 111 (2), 793–801, doi: 10.1016/j.ijpe.
2007.03.011.
Jodlbauer, H., 2008b: Produktionsoptimierung: Wertschaffende sowie kun-

denorientierte Planung und Steuerung. 2d ed., Springers Kurzlehrbücher
der Wirtschaftswissenschaften, Springer, Wien.
132 Bibliography
Jodlbauer, H., 2008c: A time-continuous analytic production model for

service level, work in process, lead time and utilization. International
Journal of Production Research, 46 (7), 1723–1744, doi: 10.1080/
00207540601080498.
Jodlbauer, H. and K. Altendorfer, 2010: Trade-off between capacity in-

vested and inventory needed. European Journal of Operational Research,
203 (1), 118–133, doi: 10.1016/j.ejor.2009.07.011.
Jodlbauer, H. and A. Huber, 2008: Service-level performance of mrp, kan-

ban, conwip and dbr due to parameter stability and environmental robust-
ness. International Journal of Production Research, 46 (8), 2179–2195,
doi: 10.1080/00207540600609297.
Jodlbauer, H. and W. Stöcher, 2006: Little’s law in a continuous setting.

International Journal of Production Economics, 103 (1), 10–16, doi: 10.
1016/j.ijpe.2005.04.006.
Kerkhofs, M., H. Chung, and P. Ester, 2008: Working time flexibility across
europe: a typology using firm-level data. Industrial Relations Journal,
39 (6), 569–585, doi: 10.1111/j.1468-2338.2008.00506.x.
Keskinocak, P. and S. Tayur, 2004: Due-date management policies. Hand-

book of quantitative supply chain analysis, D. Simchi-Levi, S. D. Wu, and
Z.-J. M. Shen, Eds., Springer Science + Business Media, New York NY,
International series in operations research & management series, Vol. 74,
485–553.
Kim, M. H. and Y.-D. Kim, 1994: Simulation-based real-time scheduling

in a flexible manufacturing system. Journal of Manufacturing Systems,
13 (2), 85–93, doi: 10.1016/0278-6125(94)90024-8.
Kleijnen, J. P. C., 1995: Verification and validation of simulation models.

European Journal of Operational Research, 82 (1), 145–162, doi: 10.
1016/0377-2217(94)00016-6.
Kok, A. d., 1990: Hierarchical production planning for consumer goods.

European Journal of Operational Research, 45 (1), 55–69, doi: 10.1016/
0377-2217(90)90156-6.
Kok, T. G. d., 2000: Capacity allocation and outsourcing in a process in-

dustry. International Journal of Production Economics, 68 (3), 229–239,
doi: 10.1016/S0925-5273(99)00134-6.
Kuik, R. and P. F. J. Tielemans, 1997: Setup utilization as a performance

indicator in production planning and control. International Journal of
Bibliography 133
Production Economics, 49 (2), 175–182, doi: 10.1016/S0925-5273(97)

00002-9.
Kutanoglu, E. and I. Sabuncuoglu, 1999: An analysis of heuristics in a dy-

namic job shop with weighted tardiness objectives. International Journal
of Production Research, 37 (1), 165–187, doi: 10.1080/002075499191995.
Lagershausen, S., M. Manitz, and H. Tempelmeier, 2013: Performance anal-

ysis of closed-loop assembly lines with general processing times and finite
buffer spaces. IIE Transactions, 45 (5), 502–515, doi: 10.1080/0740817X.
2012.705450.
Land, M., M. Stevenson, and M. Thürer, 2014: Integrating load-based or-

der release and priority dispatching. International Journal of Production
Research, 52 (4), 1059–1073, doi: 10.1080/00207543.2013.836614.
Li, H., L. Hendry, and R. Teunter, 2009: A strategic capacity allocation

model for a complex supply chain: Formulation and solution approach
comparison. International Journal of Production Economics, 121 (2),
505–518, doi: 10.1016/j.ijpe.2007.02.033.
Li Sun, S. S. Heragu, L. Chen, and M. L. Spearman, 2007: Simulation

analysis of a multi-item mrp system based on factorial design. Pro-
ceedings of the 2009 Winter Simulation Conference, 2107–2114, doi:
10.1109/WSC.2009.5429655.
Little, J. D. C., 1961: A proof for the queuing formula L = λW . Operations

Research, 9 (3), 383–387, doi: 10.1287/opre.9.3.383.
Ljungberg, Õ., 1998: Measurement of overall equipment effectiveness as a

basis for tpm activities. International Journal of Operations & Production
Management, 18 (5), 495–507, doi: 10.1108/01443579810206334.
Luss, H., 1982: Operations research and capacity expansion problems: A

survey. Operations Research, 30 (5), 907–947, doi: 10.1287/opre.30.5.907.
Manne, A. S., 1961: Capacity expansion and probabilistic growth. Econo-

metrica, 29 (4), 632–649.
Mantel, R. J. and H. R. Landeweerd, 1995: Design and operational control

of an agv system. International Journal of Production Research, 41 (1–
3), 257–266, doi: 10.1016/0925-5273(95)00018-6.
Meal, H. C., 1984: Putting production decisions where they belong. Harvard
Business Review, 62 (2), 102–111.
134 Bibliography
Medhi, J., 1991: Stochastic models in queueing theory. 1st ed., Academic
Press, Boston u.a.
Merschmann, U. and U. W. Thonemann, 2011: Supply chain flexibility, un-
certainty and firm performance: An empirical analysis of german manu-
facturing firms. International Journal of Production Economics, 130 (1),
43–53, doi: 10.1016/j.ijpe.2010.10.013.
Meyr, H., 2004: Supply chain planning in the german automotive industry.
OR Spectrum, 26 (4), doi: 10.1007/s00291-004-0168-4.
Mieghem, J. A. v. and N. Rudi, 2002: Newsvendor networks: Inven-
tory management and capacity investment with discretionary activities.
Manufacturing & Service Operations Management, 4 (4), 313–335, doi:
10.1287/msom.4.4.313.5728.
Mincsovics, G. Z. and N. P. Dellaert, 2009: Workload-dependent capacity
control in production-to-order systems. IIE Transactions, 41 (10), 853–
865, doi: 10.1080/07408170802369391.
Missbauer, H., 2002: Aggregate order release planning for time-varying
demand. International Journal of Production Research, 40 (3), 699–718,
doi: 10.1080/00207540110090939.
Missbauer, H. and R. Uzsoy, 2011: Optimization models of production plan-
ning problems. Planning Production and Inventories in the Extended En-
terprise, K. G. Kempf, P. Keskinocak, and R. Uzsoy, Eds., Springer US,
Boston and MA, International Series in Operations Research & Manage-
ment Science, Vol. 151, 437–507, doi: 10.1007/978-1-4419-6485-4\ 16.
Muchiri, P. and L. Pintelon, 2008: Performance measurement using overall
equipment effectiveness (oee): literature review and practical application
discussion. International Journal of Production Research, 46 (13), 3517–
3535, doi: 10.1080/00207540601142645.
Mula, J., R. Poler, J. P. Garcı́a-Sabater, and F. C. Lario, 2006: Models for
production planning under uncertainty: A review. International Journal
of Production Economics, 103 (1), 271–285, doi: 10.1016/j.ijpe.2005.09.
001.
Nakajima, S., 1988: Introduction to TPM: Total Productive Maintenance.
Productivity Press, Cambridge Mass. u. a.
Neely, A., M. Gregory, and K. Platts, 1995: Performance measurement
system design: A literature review and research agenda. International
Journal of Operations & Production Management, 15 (4), 80–116, doi:
10.1108/01443579510083622.
Bibliography 135
Nyhuis, P. and H.-P. Wiendahl, 1999: Logistische Kennlinien: Grundlagen,

Werkzeuge und Anwendungen. Springer, Berlin.
Nyhuis, P. and H.-P. Wiendahl, 2009: Fundamentals of production logistics:
Theory, tools and applications. Springer, Berlin.
Obermaier, R. and A. Donhauser, 2012: Zero inventory and firm per-
formance: a management paradigm revisited. International Journal of
Production Research, 50 (16), 4543–4555, doi: 10.1080/00207543.2011.
613869.
Odoni, A. R. and E. Roth, 1983: An empirical investigation of the transient
behavior of stationary queueing systems. Operations research, 31 (3),
432–455, doi: 10.2307/170616.
Ohno, T., 1988: Toyota production system: Beyond large-scale production.
Productivity Press, Cambridge MA.
Olhager, J., 2013: Evolution of operations planning and control: from pro-
duction to supply chains. International Journal of Production Research,
51 (23-24), 6836–6843, doi: 10.1080/00207543.2012.761363.
Orlicky, J., 1975: Material requirements planning; The new way of life in
production and inventory management. McGraw-Hill, New York.
Panwalkar, S. S. and W. Iskander, 1977: A survey of scheduling rules.
Operations Research, 25 (1), 45–61, doi: 10.1287/opre.25.1.45.
Park, C.-W. and H.-S. Lee, 2013: Performance evaluation of a multi-product
conwip assembly system with correlated external demands. International
Journal of Production Economics, 144 (1), 334–344, doi: 10.1016/j.ijpe.
2013.02.021.
Perera, T. and K. Liyanage, 2000: Methodology for rapid identification and
collection of input data in the simulation of manufacturing systems. Simu-
lation Practice and Theory, 7 (7), 645–656, doi: 10.1016/S0928-4869(99)
00020-8.
Pibernik, R. and P. Yadav, 2009: Inventory reservation and real-time order
promising in a make-to-stock system. OR Spectrum, 31 (1), 281–307, doi:
10.1007/s00291-007-0121-4.
Pinedo, M. L., 2008: Scheduling: Theory, algorithms, and systems. 3d ed.,
Springer, New York NY.
Ponsignon, T. and L. Mönch, 2014: Simulation-based performance assess-
ment of master planning approaches in semiconductor manufacturing.
Omega, 46 (0), 21–35, doi: 10.1016/j.omega.2014.01.005.
136 Bibliography
Powell, D., J. Riezebos, and J. O. Strandhagen, 2012: Lean production

and erp systems in small- and medium-sized enterprises: Erp support for
pull production. International Journal of Production Research, 51 (2),
395–409, doi: 10.1080/00207543.2011.645954.
Prakash, J. and J. Chin, 2014: Modified conwip systems: a review and clas-
sification. Production Planning & Control, 1–12, doi: 10.1080/09537287.
2014.898345.
Rafiei, H., M. Rabbani, and M. Alimardani, 2013: Novel bi-level hier-

archical production planning in hybrid mts/mto production contexts.
International Journal of Production Research, 51 (5), 1331–1346, doi:
10.1080/00207543.2012.661089.
Rajendran, C. and O. Holthaus, 1999: A comparative study of dispatching

rules in dynamic flowshops and jobshops. European Journal of Operational
Research, 116 (1), 156–170, doi: 10.1016/S0377-2217(98)00023-X.
Randell, L. and G. Bolmsjo, 2001: Database driven factory simulation: a

proof-of-concept demonstrator. Proceedings of the 2001 Winter Simula-
tion Conference, 2, 977–983, doi: 10.1109/WSC.2001.977402.
Rees, L. P., P. R. Philipoom, B. W. Taylor, and P. Y. Huang, 1987: Dy-

namically adjusting the number of kanbans in a just-in-time production
system using estimated values of leadtime: Iie transactions. IIE Transac-
tions, 19 (2), 199–207, doi: 10.1080/07408178708975387.
Rehder, R. R., 1989: Japanese transplants: In search of a balanced and

broader perspective. Columbia Journal of World Business, 24 (4), 17–
28.
Rohde, J., H. Meyr, and M. Wagner, 2000: Die supply chain planning
matrix. PPS Management, (1), 10–15.
Satyam, K. and A. Krishnamurthy, 2013: Performance analysis of con-

wip systems with batch size constraints. Annals of Operations Research,
209 (1), 85–114, doi: 10.1007/s10479-011-0870-y.
Schneeweiß, C., 2003: Distributed Decision Making. 2d ed., Springer, Berlin

u.a.
Schonberger, R., 1982: Japanese manufacturing techniques: Nine hidden

lessons in simplicity. 23d ed., Free Press, New York NY u.a.
Schonberger, R., 1986: World class manufacturing: The lessons of simplic-

ity applied. The Free Press, New York NY.
Bibliography 137
Schragenheim, E. and H. W. Dettmer, 2000: Manufacturing at warp

speed: Optimizing supply chain financial performance. St. Lucie Press
and APICS, Boca Raton and FL and Alexandria and VA.
Schragenheim, E. and B. Ronen, 1991: Buffer management: A diagnostic

tool for production control. Production & Inventory Management Journal,
32 (2), 74–79.
Schuh, G., 2006: Produktionsplanung und -steuerung: Grundlagen, Gestal-

tung und Konzepte. Springer.
Segerstedt, A., 1996: A capacity-constrained multi-level inventory and pro-

duction control problem. International Journal of Production Economics,
45 (1-3), 449–461, doi: 10.1016/0925-5273(96)00017-5.
Sepehri, M. M. and N. Nahavandi, 2007: Critical wip loops: a mechanism

for material flow control in flow lines. International Journal of Production
Research, 45 (12), 2759–2773, doi: 10.1080/00207540600787077.
Shah, R. and P. T. Ward, 2003: Lean manufacturing: context, practice

bundles, and performance. Journal of Operations Management, 21 (2),
129–149, doi: 10.1016/S0272-6963(02)00108-0.
Shingo, S., 1985: A revolution in manufacturing: The SMED system. Pro-

ductivity Press, Stamford Conn.
Sim, K. L. and L. N. Killough, 1998: The performance effects of comple-

mentarities between manufacturing practices and management account-
ing systems. Journal of Management Accounting Research, 10, 325–346.
Spearman, M. L., D. L. Woodruff, and W. J. Hopp, 1990: Conwip: a

pull alternative to kanban. International Journal of Production Research,
28 (5), 879–894, doi: 10.1080/00207549008942761.
Spearman, M. L. and M. A. Zazanis, 1992: Push and pull production sys-

tems: Issues and comparisons. Operations research, 40 (3), 521–532, doi:
10.1287/opre.40.3.521.
Stadtler, H., 2005: Supply chain management and advanced planning—

basics, overview and challenges. European Journal of Operational Re-
search, 163 (3), 575–588, doi: 10.1016/j.ejor.2004.03.001.
Stadtler, H. and C. Kilger, 2008: Supply chain management and advanced

planning: Concepts, models, software, and case studies. Springer, Berlin
and Heidelberg.
138 Bibliography
Steinrücke, M. and M. Jahr, 2012: Tactical planning in supply chain net-

works with customer oriented single sourcing. International Journal of Lo-
gistics Management, 23 (2), 259–279, doi: 10.1108/09574091211265387.
Sugimori, Y., K. Kusunoki, F. Cho, and S. Uchikawa, 1977: Toyota pro-
duction system and kanban system materialization of just-in-time and
respect-for-human system. International Journal of Production Research,
15 (6), 553–564, doi: 10.1080/00207547708943149.
Suzaki, K., 1987: The new manufacturing challenge: Techniques for con-
tinuous improvement. Free Press, New York.
Tardif, V. and L. Maaseidvaag, 2001: An adaptive approach to control-
ling kanban systems. European Journal of Operational Research, 132 (2),
411–424, doi: 10.1016/S0377-2217(00)00119-3.
Thompson, M. B., 1994: Expanding simulation beyond planning and design.
Industrial Engineering, 26 (10), 64.
Trybula, W. J., 1994: Building simulation models without data. IEEE In-
ternational Conference on Systems, Man, and Cybernetics, 1994. Hu-
mans, Information and Technology, 1, 209–214, doi: 10.1109/ICSMC.
1994.399838.
Verganti, R., 1997: Order overplanning with uncertain lumpy demand: a
simplified theory. International Journal of Production Research, 35 (12),
3229–3248, doi: 10.1080/002075497194057.
Vollmann, T. E., W. L. Berry, and D. C. Whybark, 1997: Manufacturing
planning and control systems. 4th ed., Irwin/McGraw-Hill, New York.
Wacker, J. G., 1996: A theoretical model of manufacturing lead times and
their relationship to a manufacturing goal hierarchy. Decision Sciences,
27 (3), 483–517, doi: 10.1111/j.1540-5915.1996.tb00861.x.
Waikar, A. M., B. R. Sarker, and A. M. Lal, 1995: A comparative study
of some priority dispatching rules under different shop loads. Production
Planning & Control, 6 (4), 301–310, doi: 10.1080/09537289508930284.
Wedel, J. and K. Lumsden, 1995: The influence of lead-time reductions
on decisions and rules in the production planning process. Proceedings
of the 12th International Conference on Production Research, 41 (1–3),
399–404, doi: 10.1016/0925-5273(95)00039-9.
White, R. E. and J. N. Pearson, 2001: Jit, system integration and cus-
tomer service. International Journal of Physical Distribution & Logistics
Management, 31 (5), 313–333, doi: 10.1108/EUM0000000005515.
Bibliography 139
White, R. E., J. N. Pearson, and J. R. Wilson, 1999: Jit manufacturing: A

survey of implementations in small and large u.s. manufacturers. Man-
agement Science, 45 (1), 1–15, doi: 10.1287/mnsc.45.1.1.
Wiendahl, H.-P. and J.-W. Breithaupt, 1999: Modelling and controlling the
dynamics of production systems. Production Planning & Control, 10 (4),
389–401, doi: 10.1080/095372899233136.
Wiendahl, H.-P., E. Ludwig, and W. Ullmann, 1994: Monitoring and di-

agnosis systems: New components of production planning and control -
methodologies, applications and experience. International Transactions
in Operational Research, 1 (1), 95–106, doi: 10.1111/1475-3995.d01-11.
Wight, O. W., 1984: Manufacturing resource planning: MRPII: unlocking

America’s productivity potential. Rev. ed., O. Wight Ltd. Publications
and Van Nostrand Reinhold, Essex Junction and VT and New York.
Yang, M. G., P. Hong, and S. B. Modi, 2011: Impact of lean manufacturing

and environmental management on business performance: An empirical
study of manufacturing firms. International Journal of Production Eco-
nomics, 129 (2), 251–261, doi: 10.1016/j.ijpe.2010.10.017.
Yang, T., H. FU, and K. Yang, 2007: An evolutionary-simulation approach

for the optimization of multi-constant work-in-process strategy—a case
study. International Journal of Production Economics, 107 (1), 104–114,
doi: 10.1016/j.ijpe.2006.02.014.
Yang, W. and R. Y. K. Fung, 2014: An available-to-promise decision

support system for a multi-site make-to-order production system. In-
ternational Journal of Production Research, 52 (14), 4253–4266, doi:
10.1080/00207543.2013.877612.
Yano, C. A., 1987: Stochastic leadtimes in two-level assembly systems. IIE

Transactions, 19 (4), 371–378, doi: 10.1080/07408178708975409.
Zäpfel, G., 1996: Production planning in the case of uncertain individual

demand extension for an mrp ii concept. International Journal of Pro-
duction Economics, 46-47, 153–164, doi: 10.1016/0925-5273(95)00192-1.

Alexander Hübl (Auth.) - Stochastic Modelling in Production Planning - Methods For Improvement and Investigations On Production System Behaviour-Gabler Verlag (2018) PDF

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Alexander Hübl (Auth.) - Stochastic Modelling in Production Planning - Methods For Improvement and Investigations On Production System Behaviour-Gabler Verlag (2018) PDF

Uploaded by

Copyright:

Available Formats

Alexander Hübl

Dissertation University of Vienna, Austria, 2015

ISBN 978-3-658-19119-1 ISBN 978-3-658-19120-7 (eBook)

Printed on acid-free paper

This Springer Gabler imprint is published by Springer Nature

Chapter 3 : Hübl and Jodlbauer (2013): Optimal Utilization based on

Chapter 4 : Altendorfer et al. (2014): Periodical capacity setting methods

Chapter 5 : Hübl and Altendorfer (2014a): Inventory constraint deﬁni-

Chapter 6 : Hübl et al. (2013): Inﬂuence of dispatching rules on average

4 Capacity Setting Methods 37

4.4 Multi-Machine Concept . . . . . . . . . . . . . . . . . . . . . 50

6 Dispatching Rules 109

2.1 Manufacturing resource planning MRPII (Wight, 1984; Hopp

4.1 Matrix of stochastic information for demanded capacity –

5.1 Combined BOM and routing structure . . . . . . . . . . . . . 89

6.1 Multi-stage production system . . . . . . . . . . . . . . . . . 109

4.1 Deﬁnition of variables for single-machine model . . . . . . . . 41

5.1 Processing and set-up times . . . . . . . . . . . . . . . . . . . 92

6.1 Deﬁnitions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 118

5.2.1 Observation (System load) . . . . . . . . . . . . . . . . . . . . 96

6.1.1 Theorem (Mulit-stage systems) . . . . . . . . . . . . . . . . . 110

© Springer Fachmedien Wiesbaden GmbH 2018

apply a hierarchical planning approach, which stipulates that a strategic de-

procurement production distribution sales

Long term Strategic Network Planning

Medium Master Planning

Strategic Network Planning is the basis of planning activities. The re-

• Uncertainty in demand forecast: All production plans focus on de-

• Uncertainty in external supply processes: A production plan results

• Uncertainty in internal supply processes: A production plan relies on

”Models for production planning which do not recognize the

is even more challenging if the customer changes the speciﬁcations directly

1.2 Research Questions

1. Which utilisation based on costs and revenue is optimal for a hierar-

3. Does a constrained Wipcap of FGI and WIP for a Conwip system in

4. How do dispatching rules inﬂuence average production lead time for

of the database distinguishes between master data, such as bills of materi-

a multi-machine setting. Based on the customer orders, the capacity of the

An overview of relevant and recent literature focusing on the deﬁned re-

2.1 JIT Goals

2.1.1 Hierarchical Production Planning Models

In this section some basic formulations on how production planning de-

© Springer Fachmedien Wiesbaden GmbH 2018

Resource Aggregate Production

Rough-cut Capacity Master Production

The long term level basically inﬂuences capacity investment decisions

concerning a number of factors, such as capacity, production mix, equip-

Yang and Fung (2014) use a hierarchical structure to solve an ATP

2.1.2 Utilisation in Production Planning

2.1.3 Impact of JIT Activities on Performance

(TQM) contribute substantially to the operating performance (Ahuja and

use of a comprehensive model of loss management. Muchiri and Pintelon

2.2 Capacity Setting Methods for Medium Term

In recent research queuing state dependent capacity adjustment models

Hence, Gstettner and Kuhn (1996) demonstrate that Kanban outperforms

Simulation studies either discuss extensions of Conwip and their perfor-

2.4 Dispatching and Production Lead Time

processing times result in a distribution of the production lead time (Yano,

3.1 Seven Zeros

Zero Lot Size

© Springer Fachmedien Wiesbaden GmbH 2018

between production rate and customer demand (Daugherty et al., 1994). In

Zero Lead Time