Capacity and Inventory Planning For Make-to-Order Production Systems The Impact of A Customer Required Lead Time Distribution

Lecture Notes in Economics and Mathematical Systems 671
Klaus Altendorfer
Capacity and
Inventory Planning
for Make-to-Order
Production Systems
The Impact of a Customer Required
Lead Time Distribution
Lecture Notes in Economics
and Mathematical Systems 671
Founding Editors:
M. Beckmann
H.P. Künzi
Managing Editors:
Prof. Dr. G. Fandel
Fachbereich Wirtschaftswissenschaften
Fernuniversität Hagen
Hagen, Germany
Prof. Dr. W. Trockel

Murat Sertel Institute for Advanced Economic Research
Istanbul Bilgi University
Istanbul, Turkey
Institut für Mathematische Wirtschaftsforschung (IMW)

Universität Bielefeld
Bielefeld, Germany
Editorial Board:
H. Dawid, D. Dimitrov, A. Gerber, C-J. Haake, C. Hofmann, T. Pfeiffer,

R. Slowiński, W.H.M. Zijm
For further volumes:

http://www.springer.com/series/300
ThiS is a FM Blank Page
Klaus Altendorfer
Capacity and Inventory

Planning for Make-to-Order
Production Systems
The Impact of a Customer Required Lead
Time Distribution
Klaus Altendorfer
School of Management
University of Applied Sciences
Upper Austria
Steyr, Austria
Published with the support of Austrian Science Fund (FWF)
ISSN 0075-8442
ISBN 978-3-319-00842-4 ISBN 978-3-319-00843-1 (eBook)
DOI 10.1007/978-3-319-00843-1
Springer Cham Heidelberg New York Dordrecht London
Library of Congress Control Number: 2013944963
© Springer International Publishing Switzerland 2014

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. Exempted from this legal reservation are brief excerpts
in connection with reviews or scholarly analysis or material supplied specifically for the purpose of being
entered and executed on a computer system, for exclusive use by the purchaser of the work. Duplication
of this publication or parts thereof is permitted only under the provisions of the Copyright Law of the
Publisher’s location, in its current version, and permission for use must always be obtained from
Springer. Permissions for use may be obtained through RightsLink at the Copyright Clearance Center.
Violations are liable to prosecution under the respective Copyright Law.
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.
While the advice and information in this book are believed to be true and accurate at the date of
publication, neither the authors nor the editors nor the publisher can accept any legal responsibility for
any errors or omissions that may be made. The publisher makes no warranty, express or implied, with
respect to the material contained herein.
Printed on acid-free paper
Springer is part of Springer Science+Business Media (www.springer.com)

Preface
Queuing models for the design and analysis of manufacturing and service systems
have a long tradition in operations management. There exists a considerable
knowledge of stochastic models and methods. Nevertheless, there is still demand
for further extensions, in particular when it comes to combining an operations view
with other business perspectives like a stronger customer focus.
This book comprises a collection of four research papers the author has prepared
during his time as a Ph.D. student in the “Logistics and Operations Management”
program at the University of Vienna. The work’s innovation is the simultaneous
consideration of classical performance measures and decision variables of
manufacturing systems and the impact of customer-required lead times which
vary among different customers.
The contributions in detail analyze inventory, waiting time performance, and
service-level measures under distributed customer-required lead times, both for
single- and multistage manufacturing systems. The basic results are utilized to
simultaneously optimize strategic and tactical means of a manufacturing system,
i.e., capacity on the one hand and work-release rules as tactical parameters on the
other hand.
Using simple queuing models, Klaus Altendorfer’s thesis aims at deriving
analytical expressions for performance measures and insights derived from those.
The obtained results will inspire future work on production control and
manufacturing systems design with a particular emphasis on customer
requirements.
Munich, March 2013 Stefan Minner
v
Acknowledgments
The Ph.D. thesis, which is published with minor improvements in this book, was
greatly inspired by one of Prof. Jodlbauer’s papers. The thesis was written during
my time as a Research Assistant and later on as an Assistant Professor in his
department at the Upper Austria University of Applied Sciences in Steyr. I would
like to thank Prof. Jodlbauer for his support and for the working conditions enabling
the generation of a Ph.D. thesis in his department. Furthermore, I would like to
thank Prof. Hartl for the possibility to start my Ph.D. studies at the University of
Vienna and for his further support.
I am especially thankful to my Supervisor Prof. Minner for his critical comments
on my manuscripts, leading to a considerable improvement of the results and their
presentation. This critical review has enabled a much deeper understanding of the
concepts used and will guide my future research work as well.
Lastly, I would like to thank my wife Gerda for her ongoing support during all
the challenges of my Ph.D. studies. She was always an island of rest and love for me
and provided me the strength to go on whenever the path seemed too rocky.
Steyr, March 2013 Klaus Altendorfer
vii
Contents
1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
2 Literature Review . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
2.1 Relationships Between Logistical Key Performance Indicators . . . 6
2.1.1 Internal View on Production Logistics . . . . . . . . . . . . . . . 6
2.1.2 External View on Production Logistics . . . . . . . . . . . . . . . 8
2.2 Work Release Rules . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
2.2.1 Planned Lead Time Setting . . . . . . . . . . . . . . . . . . . . . . . 12
2.2.2 Advance Demand Information . . . . . . . . . . . . . . . . . . . . . 13
2.3 Capacity Investment . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
2.3.1 Queuing System Design . . . . . . . . . . . . . . . . . . . . . . . . . . 15
2.3.2 Backorder Cost or Service Level Models . . . . . . . . . . . . . 16
3 Single-Stage Service Level and Tardiness Model . . . . . . . . . . . . . . . 19
3.1 Introductory Remarks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
3.2 Model Development . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
3.2.1 Model Assumptions and Notation . . . . . . . . . . . . . . . . . . . 20
3.2.2 Queuing Model with Deterministic Customer Required
Lead Time . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
3.2.3 Distribution of Customer Required Lead Time . . . . . . . . . 23
3.2.4 Integration of a WAW Work Release Policy . . . . . . . . . . . 25
3.2.5 Application for G/G/1 Production System . . . . . . . . . . . . . 28
3.3 Numerical Study . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
3.3.1 Relationship Between Logistical Key Figures . . . . . . . . . . 29
3.3.2 FGI Reduction Potential with WAW Policy . . . . . . . . . . . 31
3.4 Concluding Remarks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
4 Simultaneous Capacity and Planned Lead Time Optimization . . . . . 43
4.2 The General Optimization Model . . . . . . . . . . . . . . . . . . . . . . . . 44
4.2.1 Assumptions and Notation . . . . . . . . . . . . . . . . . . . . . . . . 44
4.2.2 Model Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45
4.2.3 Multi-Stage Extension . . . . . . . . . . . . . . . . . . . . . . . . . . . 48
ix
x Contents
4.3 Simultaneous Optimization of a Single-Stage Production

System . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49
4.3.1 Optimal Planned Lead Time when Capacity
is Predefined . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49
4.3.2 Simultaneous Capacity and Planned Lead Time
Setting . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51
4.4 Simultaneous Optimization of a Two-Stage Production
System . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53
is Predefined . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53
4.4.2 Simultaneous Capacity and Planned Lead Time Setting . . . 55
4.5 Numerical Study . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 59
5 Optimal Composition of Number and Size of Machines . . . . . . . . . . 73
5.2 General n-Stage Optimization Problem . . . . . . . . . . . . . . . . . . . . 74
5.2.1 Model Description and Assumptions . . . . . . . . . . . . . . . . 74
5.2.2 Model Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 76
5.3 Optimization Problem for a Two-Stage M/M/s Production
System . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 78
5.3.1 M/M/s Queue . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 78
5.3.2 Problem Statement . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 79
5.3.3 WIP, FGI, and Backorders Calculation . . . . . . . . . . . . . . . 80
5.3.4 Solution Heuristic for the Two-Stage M/M/s Problem . . . . 83
5.4 Numerical Study . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 84
5.4.1 Influence of Machine Size on Optimal Cost . . . . . . . . . . . 84
5.4.2 Influence of Machine Number on Optimal Costs . . . . . . . . 87
5.4.3 Performance Evaluation of Solution Heuristic . . . . . . . . . . 88
5.4.4 Influence of Uncertain Input Rates . . . . . . . . . . . . . . . . . . 89
6 Service Level Constraint Models . . . . . . . . . . . . . . . . . . . . . . . . . . . 99
6.2 Single-Stage M/M/1 Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 100
6.2.1 Model Development . . . . . . . . . . . . . . . . . . . . . . . . . . . . 100
6.2.2 Numerical Example . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 102
6.3 Single-Stage, Multi-Item, Normally Distributed Demand
Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 104
6.3.1 Model Assumptions and Notation . . . . . . . . . . . . . . . . . . . 105
6.3.2 Capacity Demand Smoothing . . . . . . . . . . . . . . . . . . . . . . 107
6.3.3 WAW Calculation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 109
6.3.4 Determination of the Required Inventory . . . . . . . . . . . . . 111
6.3.5 Optimal Capacity Invested . . . . . . . . . . . . . . . . . . . . . . . . 111
Contents xi

6.3.7 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 115
6.4 Two-Stage M/M/s Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 116
6.4.1 Optimization Problem with Service Level Constraint . . . . . 116
7 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 125
Appendix . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 129
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 133
List of Figures
Fig. 1.1 Basic model framework . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3

Fig. 3.1 Single-stage production system . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
Fig. 3.2 Trajectory utilization loss versus FGI reduction in M/M/1
model . . . . .. . . . . . .. . . . . .. . . . . . .. . . . . . .. . . . . .. . . . . . .. . . . . . .. . . . . .. . . . . . .. . . . 28
Fig. 3.3 Service level, FGI lead time, and tardiness in M/M/1 model . . . . . . . 30
Fig. 3.4 On-time probability in joint probability space of W and L . . . . . . . . . . 35
Fig. 4.1 Two-stage production system with planned lead times . . .. . .. . .. . .. . 44
Fig. 4.2 Planned lead time costs in single-stage M/M/1 model with
predefined capacity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51
Fig. 4.3 Optimal parameters for single-stage M/M/1 model with
backorder costs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 52
Fig. 4.4 Influence of X1 and X2 on the expected FGI lead time and tardiness
in two-stage model . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . 56
Fig. 4.5 Cost with respect to planned lead time and utilization in two-stage
M/M/1 model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 57
Fig. 4.6 Optimal planned lead time with respect to utilization in two-stage
M/M/1 model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 58
Fig. 4.7 Optimal parameters for two-stage M/M/1 model with
backorder costs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 58
Fig. 4.8 Cost reduction in two-stage model with simultaneous optimization
problem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61
Fig. 5.1 Multi-stage production system with parallel machines . . . . . . . . . . . . . . 75
Fig. 5.2 Influence of processing rate and number of machines on logistical
key figures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 82
Fig. 5.3 Optimal overall costs dependent on machine size in two-stage
M/M/s system . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 85
Fig. 5.4 Optimal number of machines with predefined machine
sizes in two-stage M/M/s model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 87
Fig. 5.5 Cost increase for more than one machine at each stage in two-stage
M/M/s model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 88
xiii
xiv List of Figures
Fig. 5.6 Influence of uncertain input rate on optimal costs in two-stage

M/M/1 model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 90
Fig. 5.7 Influence of range and uncertainty of input rate in two-stage
M/M/1 model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 90
Fig. 6.1 Backorder to FGI cost ratio for equal WAW in single-stage
M/M/1 model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 102
Fig. 6.2 Costs and WAW in single-stage M/M/1 model with service level
constraint . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 103
Fig. 6.3 Sensitivity analysis of single-stage M/M/1 model with service
level constraint . . . . . . . . . . . .. . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . .. . . . . . . . . 104
Fig. 6.4 Multi-item, normally distributed demand, single-stage production
system . .. . .. . .. . .. . .. . .. . .. . .. . .. . .. . .. . .. . .. . .. . .. . .. . .. . .. . .. . .. . .. . .. . 105
Fig. 6.5 Overall lead time in normally distributed demand, single-stage
model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 107
Fig. 6.6 WAW calculation based on capacity oriented customer required
lead time distribution . . . . .. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . 110
Fig. 6.7 Costs and WAW in normally distributed demand model with
service level constraint . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 114
Fig. 6.8 Sensitivity analysis of normally distributed demand model with
service level constraint . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 115
Fig. 6.9 Optimal costs dependent on machine size in two-stage service
level constraint model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 118
List of Tables
Table 3.1 Numerical examples for logistical relationships in

M/M/1 model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30
Table 3.2 Results from numerical study of FGI reduction in
M/M/1 model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32
Table 5.1 Numerical examples for overall cost evaluation in
two-stage M/M/s model . . . . . . .. . . . . . .. . . . . . .. . . . . . .. . . . . . .. . . . . . .. . . 85
Table 5.2 Two-stage M/M/s optimization heuristic performance
comparison . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 89
Table 5.3 Two-stage M/M/s range uncertainty experiment
parameters . . . . . .. . . . . . . .. . . . . . . .. . . . . . . . .. . . . . . . .. . . . . . . .. . . . . . . . .. . . . 90
Table 5.4 Results from numerical study on input rate uncertainty in
two-stage M/M/s model . . . . . . .. . . . . . .. . . . . . .. . . . . . .. . . . . . .. . . . . . .. . . 92
Table 6.1 Numerical examples for M/M/1 model with service level
constraint . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 103
Table 6.2 Numerical examples for normally distributed demand
model with service level constraint . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 114
Table 6.3 Numerical examples for overall costs in two-stage
M/M/s service level constraint model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 118
Table A.1 Definition of variables used in the book .. . .. . .. .. . .. . .. .. . .. . .. . .. 129
Table A.2 Definition of units used in the book . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 131
xv
Chapter 1
Introduction
Decisions on manufacturing capacity and production planning parameters are

continuously faced by manufacturing companies to improve their performance.
To determine this performance, internal and external performance indicators can
be applied. Important external performance indicators for make-to-order (MTO)
production systems are service level and average tardiness. Service level denotes
the percentage of orders or pieces delivered on time and average tardiness states
how much later than the due date the orders are delivered. Important internal
performance indicators are capacity invested and inventory held. To identify an
optimum between these conflicting internal and external objectives of a
manufacturing company, inventory and capacity invested are weighted by costs
which are minimized whereby either tardiness costs are added or a service level
constraint is considered. These four conflicting key performance indicators can be
influenced by capacity investment and work release to the production system.
In real production systems the increasing number of different products and
variants is forcing more and more companies to apply an MTO production policy
to keep the inventory at an acceptable level. In such MTO production systems the
due dates observed are often random. This can arise from different products ordered
by different customers, it can be the result of customer diversification through
marketing, or it occurs due to the planning uncertainty the customers of a
manufacturing company face. Defining the time between ordering and the due
date as random customer required lead time, a distribution of this customer required
lead time can be stated (see Jodlbauer 2008b). The influence of this customer
required lead time on the conflicting key performance indicators capacity invested,
inventory held, average tardiness, and service level is studied in this work. There-
fore, the main research question discusses the optimal capacity to invest and the
optimal work release rule parameterization to minimize capacity, inventory, and
backorder costs or to fulfill a certain service level constraint under minimum
capacity and inventory costs for an MTO production system with random customer
required lead time. To further refine the research field, the following specific topics
are covered in Chaps. 3, 4, 5 and 6 respectively:
K. Altendorfer, Capacity and Inventory Planning for Make-to-Order Production 1

Systems, Lecture Notes in Economics and Mathematical Systems 671,
DOI 10.1007/978-3-319-00843-1_1, © Springer International Publishing Switzerland 2014
2 1 Introduction
• Influence of a random customer required lead time and a work-ahead-window

(WAW) work release rule on the service level, the expected FGI lead time, and
the expected tardiness of a single-stage, MTO production system.
• Influence of a random customer required lead time on the optimal capacity to
invest and the optimal planned lead times in a single-stage and a two-stage MTO
production system when capacity, inventory, and backorder costs are
minimized.
• Influence of predefined machine sizes on the optimal number and size of
machines in a two-stage, MTO production system with random customer
required lead time and parallel machines at each stage when capacity, inventory,
and backorder costs are minimized.
• Optimal capacity to invest and the optimal WAW in a single-stage or a two-stage
production system with random customer required lead time when capacity and
inventory costs are minimized with respect to a service level constraint.
The literature review in Chap. 2 shows that the specific research topics treated in
this book have not yet been covered but there has recently been increased research
interest in simultaneously optimizing inventory and capacity investment consider-
ing the timing of customer demands in addition to the actual distribution of demand.
The work release rules discussed in this book are the planned lead time rule and
the WAW rule. The planned lead time rule, which is applied in material-
requirements-planning (MRP) states that for each production stage a planned lead
time is defined and the orders are released to the next stage whenever the remaining
time to due date is shorter than the sum of the remaining planned lead time values
(see Hopp and Spearman 2008). The WAW rule states that an order is released to
the production system whenever its remaining time to due date is less than the
WAW. This WAW rule is applied by the production planning methods constant-
work-in-process (CONWIP) (see Hopp and Spearman 2008) or drum-buffer-rope
(DBR) (see Schragenheim and Dettmer 2001).
Insights are derived by setting up and solving simultaneous optimization
problems for capacity investment and work release rule parameterization for
different MTO production systems all including a distribution of customer required
lead time in addition to the actual demand distribution. The common framework for
all the models discussed in this book is based on the assumption that the customers
specify different due dates when ordering an item from which the distribution of
customer required lead time can be derived. Orders which are finished earlier than
the due date are stored in the finished-goods-inventory (FGI). Therefore, the
production lead time defines the time between work release to the production
system and its completion at the last work station. The FGI lead time is the time
between arrival at the FGI buffer and the due date. Tardiness is defined as the time
between due date and order delivery if an order is late. Additionally, two different
service level definitions are used depending on the actual models. Service level is
either calculated as probability that an order is delivered on time or as percentage of
capacity delivered on time. The work-in-process (WIP) is defined as number of
items in the buffers in front of the machines and in the machines, FGI is the number
1 Introduction 3
Fig. 1.1 Basic model Customer required Tardiness

framework lead time = L C = max(0, W–L)
Customer
Service level
h = P(W <= L)
Buffer Buffer
Processing
stages
FGI = G
WIP = Y
FGI lead time

Production lead time = W I = max(0, L–W )
of items in the FGI buffer, and backorder is the number of items currently late,
i.e. the due date is in the past and the item has not been delivered yet. Utilization
denotes the percentage of time the machines are busy at the respective processing
stage.
Figure 1.1 shows the common model framework for the following chapters.
As stated above, only MTO production systems are studied which means that
nothing is produced without a customer order. All customer orders are accepted and
the random due date values, which do not change after they have been stated,
cannot be influenced by the production system. All buffers in the production system
are unbounded.
In Chap. 3 an analytical model to calculate service level, FGI, and tardiness for
an MTO production system based on the production lead time, utilization, and WIP
when the customer required lead time follows a stochastic distribution, is presented.
Explicit expressions for service level, FGI, FGI lead time, and tardiness are derived
for an M/M/1 (for this notation see Sect. 2.1.1) production system within an MTO
environment based on the known equations for production lead time, utilization,
and WIP from queuing literature. Moreover, the integration of a WAW work
release policy is discussed. A high potential to decrease FGI when applying such
a WAW policy is found based on a numerical study. This is an interesting manage-
rial insight since the competition between manufacturing companies is more and
more forced into the field of logistical performance and it is clearly shown that
implementing such a WAW policy helps to keep costs down for high service levels.
In Chap. 4 a two-stage, MTO production system where the work is released
to each stage based on a planned lead time is considered. A general approach to
minimize total inventory holding and customer order tardiness cost is presented
to find the optimal manufacturing capacities and planned lead times for each
manufacturing stage. Expressions are derived for WIP, FGI, and expected
backorders under the assumption of a series of M/M/1 queuing systems and
exponentially distributed customer required lead times. It is proven that the distri-
bution of customer required lead time has no influence on the optimal planned lead
4 1 Introduction
times whenever capacity is predefined but it influences the optimal capacity invest-
ment. For the simultaneous optimization of capacity and planned lead times a
numerical study is presented which shows that only marginal cost decreases can
be gained by setting a planned lead time for the upstream stage and that a
considerable cost penalty is incurred if capacity and planned lead time optimization
are performed sequentially.
In Chap. 5 an optimization model to minimize costs for WIP, FGI, backorders,
and capacity for a multi-stage production system is developed. The parameters to
optimize are the capacity invested at each stage, consisting of the number of
machines and the processing rate (defined by a set of possible processing rates),
as well as the WAW. An optimality condition is provided showing in which
production systems it is optimal to invest in a single machine at each stage if
processing rates are continuous decision variables. The optimality of increasing
capacity towards the customer end of the production line is proven under certain
conditions. For a special case consisting of two stages with M/M/s (for this notation
see Sect. 2.1.1) queues with exponentially distributed customer required lead time,
explicit expressions for WIP, FGI, and backorders are developed. A set of numeri-
cal examples illustrates the influence of predefined processing rates on the optimal
number and selection of machines. A solution heuristic for the problem with a
predefined set of processing rates is proposed and its performance is shown in a
numerical study. Additionally, the influence of uncertain input rates into the
production system is discussed.
Chapter 6 tackles the topic of service level constraint models. In Sects. 6.2 and
6.4, the models developed in Chaps. 4 and 5 are extended to an optimization
problem with a service level constraint. The managerial insight from the differences
between the service level constraint models and the backorder costs models is that
for simultaneous capacity investment and work release rule parameterization,
management has to decide on one model since they are not equivalent. In
Sect. 6.3 the relationship between the available capacity and the inventory needed
to meet customer requirements with a predefined service level is modeled for a
single-stage, multi-item production system with normally distributed demand. In
this model service level is defined as percentage of capacity delivered on time.
Explicit expressions for optimal capacity and inventory are provided showing that
both increase if demand variation increases.
A conclusion of the results is presented in Chap. 7.
Chapter 2
Literature Review
In this literature review, a broad overview of the relevant and recent publications in
the research fields studied in this book is provided. Based on the mass of literature
available this review can only cover the relevant set of literature which builds the
basis for the developed models.
Section 2.1 of this review is dedicated to literature covering the relationship
between the logistical key performance indicators utilization, WIP, production lead
time, FGI, FGI lead time, tardiness, and backorders. The literature reviewed in this
section shows that there is already extensive research in this field available.
Nevertheless, none of the models explicitly addresses the link between the internal
logistical key figures, utilization and production lead time, and the external logisti-
cal key figures, service level and tardiness, for an MTO production system includ-
ing random due dates. This research gap, as stated in the first research topic of the
introduction, is covered in Chap. 3 by developing an analytical model linking these
internal and external key performance indicators.
Section 2.2 covers the literature on planned lead time setting and advance
demand information. The literature discussed in Sect. 2.2 shows that most of the
models in both fields of research have neglected the possibility of influencing the
distribution of production lead time by capacity investment decisions. Additionally,
most models on planned lead time setting define the customer required lead time to
be constant. It is, often implicitly, assumed that the customer required lead time is
always higher than the sum of the planned lead times. Also in most of the advance
demand information models it is assumed that information is available a certain
constant time period in advance which neglects the stochastic behavior of customer
required lead time. Furthermore, all reviewed advance demand information models
are make-to-stock (MTS). Therefore, the investigation of a simultaneous treatment
of capacity investment and planned lead time setting decision and the evaluation of
the influence of a random customer required lead time on the optimal result, as
stated in the second research topic of the introduction, are possible extensions of the
currently available literature. Such a model is developed in Chap. 4.
The topic of capacity investment either minimizing capacity and WIP costs or
also including backorder costs or a service level constraint is reviewed in Sect. 2.3.

6 2 Literature Review
The literature on queuing system design, discussing the optimal number of servers
and the optimal service rate in queuing systems, indicates that there is a lack of
models including the customer required lead time and the FGI into this quite strong
framework of queuing systems. This influence of a customer required lead time
distribution on the optimal number and size of machines in a queuing theoretical
model including FGI and backorder costs, as stated in the third research topic of the
introduction, is examined in Chap. 5.
The review of Sect. 2.3 shows that in addition to models including backorder
costs, literature is available on service level constraint models jointly optimizing
capacity and FGI costs. However, these models also mostly lack the integration of a
customer required lead time distribution and most models are MTS. Therefore, in
addition to backorder cost models for capacity and inventory cost minimization
when customer required lead time is distributed, the influence of a service level
constraint has not been identified yet. This research gap, as addressed in the fourth
research topic of the introduction, is treated in Chap. 6.
2.1 Relationships Between Logistical Key Performance

Indicators
In this section the literature concerning the logistical relationships between utiliza-
tion, WIP, production lead time, FGI, FGI lead time, tardiness, and backorders is
reviewed. The literature can be divided into literature covering the relationship
between utilization, WIP, and production lead time, which is called the internal
view on the production logistics in this book (see Sect. 2.1.1) and literature that
covers the topics of service level, FGI, tardiness, and backorders based on capacity
or inventory. This is called the external view on the production logistics in this book
(see Sect. 2.1.2).
2.1.1 Internal View on Production Logistics
There are two different approaches for studying the internal view on production
logistics. On the one hand there is the analytical approach being able to deliver
exact proofs for different problems and to identify the exact influence of different
parameters on the key measures utilization, WIP, and production lead time. The
limitation of this approach is the low level of complexity of production systems
treated. On the other hand there exists the simulation approach, where based on
simulation models the influence of different parameters on the logistical key figures
is investigated. This approach enables the discussion of more complex production
system structures but is limited by the fact that the findings are always linked to
2.1 Relationships Between Logistical Key Performance Indicators 7
single simulation models and no analytical proofs can be delivered. Both

approaches are covered in this section.
2.1.1.1 Analytical Approaches
One approach to address utilization, WIP, and production lead time is the queuing
theory approach. Queuing systems are often described in a structured way applying
the Kendall-Notation. In the notation A/S/s/K, A describes the distribution of the
arrival process, S the distribution of the service process, s the number of parallel
servers, and K, which is optional, is the maximum number of items in the queue.
For A and S, the letters M, Ph, and G describe an exponential, phase type, and a
general distribution respectively (see Tijms (2003)). Little (1961) describes the
relationship between WIP, production lead time, and throughput with the following
formula: L ¼ λW, whereby L is the mean number of units in the system, W is the
mean time spent by a unit in the system and λ is the arrival rate. Using queuing
theory the relationship between WIP, utilization, and production lead time is
described by Medhi (1991), Chen and Yao (2001), Tijms (2003), and Hopp and
Spearman (2008) who deliver equations for the relationship between expected
production lead time and utilization as well as between expected WIP and utiliza-
tion in an M/M/1 production system based on the arrival and processing rates.
Additionally, the books of Medhi (1991), Chen and Yao (2001), Tijms (2003),
Beichelt (2006), and Ross (2010) provide a broad overview about queuing system
basics and include exact as well as approximate equations for the relationship
between production lead time, WIP, and utilization for different and more complex
production system structures than M/M/1. In all these structures a linear increase in
utilization leads to a strictly convex increase in WIP. Furthermore, Medhi (1991)
proves that in the M/M/1 case the production lead time is exponentially distributed
and Chen and Yao (2001) give an approximation for the production lead time of
G/G/1 production systems. They approximate the production lead time with an
exponential distribution with its mean depending on the coefficients of variation of
interarrival time and service time. Karmarkar (1987) states that the actual produc-
tion lead time is highly dependent on actual workloads and lot sizes of a single-
machine production system.
Clearing functions also describe the relationship between WIP and throughput or
flow time. Graves (1986) and Karmarkar (1993) where the first to mention the idea
of clearing which relates the workload of the shop floor with the flow time of the
next job released. Since then such clearing functions have been widely applied to
anticipate (planned) lead times for planning decisions, recently, for example, by
Missbauer (2002), Spitter et al. (2005), Selcuk et al. (2006, 2008).
Another approach is covered by Jodlbauer (2008a) who shows this relationship
between utilization, WIP, and production lead time for a single-stage, multi-item
production system by means of a continuous time mathematical model based on the
average and the coefficient of variation of inventory.
2.1.1.2 Simulation and Empirical Approaches
Further approaches to discuss logistical relationships are empirical studies based on

real world data and simulation studies. The empirical study approach is conducted
by, for example, Nyhuis and Wiendahl (2002), Wiendahl and Breithaupt (1999),
and Wiendahl et al. (2005) who describe the relationship between inventory,
utilization, and production lead time based on empirical data. These papers derive
mathematical approximation functions for these relationships which have to be
fitted to the production system studied by certain scaling parameters using some
empirical data. Also Souza et al. (2002), studying the influence of focusing factories
on products, find in their empirical study that an increase in utilization leads to a
disproportionate increase in production lead time.
Applying simulation, Jones (1973), Hopp and Roof-Sturgis (2000), Altendorfer
et al. (2007a, b) and Jodlbauer and Huber (2008) find similar relationships based on
simulation results. These simulation studies show that the same shape of the curves,
as proven in analytical models for single- or multi-machine setting, also hold for
more complex production systems. A review of the application possibilities of
simulation for research is provided by Bertrand and Fransoo (2008).
2.1.2 External View on Production Logistics
The external view on production logistics expressed through FGI, tardiness,

backorders, and service level, is either linked to a certain stock available (mostly
MTS systems) or to a certain capacity available (mostly MTO systems). Further-
more, the literature can be divided into literature covering the relationship between
two of these key figures directly for a certain system configuration and literature
discussing cost (or profit) optimal settings. Another possible distinction is that most
relevant literature covers either service level or tardiness. This review is structured
according to the last differentiation between service level and tardiness. Since
especially the development of a link between the internal view, which is already
quite well analyzed in literature, and this external view is developed in this book,
the literature review is focused on the external view.
2.1.2.1 Service Level Calculation
For MTO production systems, often the inventory needed is discussed to reach a
predefined service level when capacity available is constant. For example Jodlbauer
(2008b) presents a model to evaluate the minimum inventory needed for a single-
stage, multi-item production system based on a constant capacity available to reach
a certain service level in an MTO production system. Similar qualitative influences
of changes in capacity and demand uncertainty on the inventory needed to reach a
predefined service level are also studied in Hartl (1990). Mapes (1993) shows in his
simulation study that a decrease in available capacity, in comparison to the average
capacity needed, leads to an increase of safety stock to ensure a certain service
level. An MTS production system with a certain permanent capacity and contingent
capacity is considered in Mincsovics et al. (2009), who develop a model to
economically evaluate the balance between inventory, permanent capacity and
contingent capacity for non-stationary stochastic demand. A framework for the
modeling and analysis of MTO, MTS, and delayed product differentiation (DD) is
developed by Gupta and Benjaafar (2004). They present a model for minimization
of inventory costs subject to a service level constraint in a multi-item production
system. Based on a certain accepted customer waiting time, the inventory holding
costs are evaluated for MTS, MTO, and DD production systems. No FGI is held in
this model based on the assumption that a customer does not require a certain lead
time but is satisfied whenever the realized customer waiting time is shorter than the
accepted customer waiting time. The setting of safety stocks for a predefined
service level is also shown in Dar-El and Malmborg (1991). For a supply chain
setting, del Vecchio and Paschalidis (2006) use approximations to set up a nonlin-
ear optimization problem where optimal stock levels for maintaining a certain
service level constraint are developed.
Another question often discussed is how much capacity is needed to reach a
certain service level. Paraskevopoulos et al. (1991) present a model for planning
capacity under uncertainty. It is shown that in situations with high uncertainty (is
equal to high fluctuations in demand and changes in the mean demand) a higher
excess capacity is needed which leads to a lower utilization to reach the same
service level. An assembly line fed by two processing lines is discussed in Liu and
Yuan (2001) whereby the throughput is maximized while a certain service level
should be maintained which also creates a link between capacity available and
service level. A good survey of capacity expansion and capacity management
literature is given in Luss (1982) and more recently in van Mieghem (2003).
In addition to the question of how much inventory is required to reach a
predefined service level in a certain production system, also the replenishment
strategy is often discussed for stock replenishment systems under a service level
constraint. For example Schneider (1981) identifies when to order in a stock
replenishment system with different replenishment strategies in order to ensure
that a certain service level is reached under normally distributed demand. In Brauer
(1985) the relationship between safety stock and replenishment lead time duration
with a known demand distribution is studied under a service level constraint. He
discusses the trade-off between service level and system costs linked to replenish-
ment lead time concluding that also purchase alternatives should be considered to
optimize system performance. A replenishment strategy is developed in Segerstedt
(1994) which leads to a replenishment order if the probability of shortage of an
order is higher than a specified service level. In Bertsimas and Paschalidis (2001)
production policies are derived for an MTS production system with a constant
production rate and multiple items to guarantee predefined service levels. In Zipkin
(2000) a broad overview concerning stock replenishment systems and different
problem structures is given. Most of the problems (and solution procedures)

presented in Zipkin (2000) consider either service level or customer waiting time
or both as part of the cost function or as constraint, whereby the parameters for the
replenishment strategy are searched for. Such systems usually correspond to MTS
production systems with production lead time values which are distributed inde-
pendently from the actual workload of the system. Additionally, a good review of
inventory management is delivered in Silver et al. (1998). For replenishment and
production systems, Greis (1994) develop in their article a reliability curve describ-
ing the relationship between service level and FGI. The developed model takes into
account the current system state and the demand distribution. In Lejeune and
Ruszczynski (2007) an inventory-production-distribution plan for a supply chain
is developed to minimize the costs while ensuring to reach a predefined service
level.
There are also approaches focusing on the calculation of the service level
reached for certain production system specifications. One of these papers is Lutz
et al. (2003), which calculates the relationship between mean inventory and service
level or mean tardiness for a single-machine production system. Their calculation is
based on mathematical approximations which have to be parameterized by
analyzing empirical data. In van Nieuwenhuyse et al. (2007) a model to determine
the service level for an MTS production system is presented by applying queuing
theory to calculate the distribution of demand within the production lead time and to
compare it with FGI. A model to estimate the service level for a production system
applying an MRP method with a frozen zone is developed in Sridharan and LaForge
(1994). Chopra et al. (2004) compares cycle service level and service level based on
demand uncertainty, replenishment lead time uncertainty for different distributions
of replenishment lead time, and demand, for a stock replenishment system. Differ-
ent definitions of service level for safety stock determination in stock replenishment
systems are addressed in Coleman (2000). Especially different problems of service
level definitions are mentioned and their influence on the optimal solution for safety
stock is identified. In Blackstone et al. (1982) a review of dispatching rules
considering different objectives is provided. Among these objectives the influence
of dispatching rules on service level is discussed. It is shown that for predefined due
dates, the application of certain dispatching rules can lead to service level
improvements.
2.1.2.2 Tardiness Calculation
The literature on tardiness can be divided into literature concerning MTO produc-
tion systems and MTS production systems. For MTO production systems only few
analytical approaches can be found since the topic of tardiness is usually addressed
by simulation studies. For MTS production systems, the term used is often customer
waiting time instead of tardiness. Nevertheless, looking at the definition of cus-
tomer waiting time, which is the time a customer has to wait for his ordered
products in an MTS system, this term is equivalent to the tardiness definition as

used in this book (see also Hopp and Spearman (2008)).
Concerning MTO production systems, Nyhuis and Wiendahl (2002) as well as
Yu (2001) present approximation equations to calculate average tardiness with
respect to average WIP based on the relationship between planned WIP to actual
WIP. Bertrand and van Ooijen (2008) also discuss the topic of tardiness in MTO
production systems. Optimal work release rules are developed for job shop produc-
tion systems working under MTO. The customer order lead time costs, tardiness
costs, and WIP costs are considered. Simulation studies also covering the topic of
tardiness evaluation are, for example, Vepsalainen and Morton (1987) and
Kutanoglu and Sabuncuoglu (1999) which study optimal dispatching strategies to
reduce tardiness whereby the relationship between utilization and tardiness is
evaluated.
For stock replenishment systems, Kruse (1980) discusses the customer waiting
time when applying the (S-1, S) replenishment policy (which means that inventory
is continuously reviewed and whenever one piece is taken from stock a new one is
ordered) and Altiok (1989) studies backorders for an (R, r) continuous review
system (whenever inventory falls below r production is started and it is produced
until an inventory of R is reached) and compound Poisson demand. Van der Heijden
and de Kok (1992) show an approximation for the customer waiting times in an
(r, S) inventory system (every r periods the inventory is reviewed and an order to fill
the inventory position up to level S is issued) with compound Poisson demand, and
Tempelmeier (2000) calculates the customer order waiting times in an (r, S)
inventory system emphasizing the importance of customer waiting times to evalu-
ate the logistical performance of a company. For an (r, s, Q) replenishment policy
(every r periods the inventory is reviewed and whenever the inventory position is
below the level s enough orders of size Q are issued to raise the inventory position
up to a level between s and s + Q) an approximation of the customer waiting time
distribution is provided by Kiesmüller and de Kok (2006). Furthermore, in Nyhuis
and Wiendahl (2002) and Inderfurth and Schulz (2007, 2008), characteristic curves
to discuss the average inventory held on hand in comparison to the customer
waiting time are considered. Stability conditions of capacitated supply chain
systems are analyzed in Glasserman and Tayur (1994). It is shown that for a
multi-echelon production and distribution system with limited capacity operating
a base-stock policy, the inventory and backorder values stay stable if the average
demand is lower than the average production capacity at each node.
Closely linked to the topic of tardiness is also the question of how many pieces
are late, which is covered, for example, in Adenso-Diaz (1996) calculating the
average number of items for a stockout based on proportion of items delivered on
time and the demand distribution.
2.2 Work Release Rules
Most of the MTO production planning methods presented in the literature such as
MRP (see Hopp and Spearman (2008)); DBR (see Schragenheim and Dettmer
(2001)); or CONWIP (see Hopp and Spearman (2008)) need a parameter to define
when the work has to be released to the production system. This parameter can have
different scopes of functionality, as for example in CONWIP only the work release
to the production system is defined with the parameter WAW, while in MRP
usually for each stage of production a start and end date is planned, which
constrains the material flow. In the case of DBR, the buffer before the bottleneck
fulfills this function. Literature on the optimization of this parameter, be it a single
parameter or one for each production stage, is reviewed in this section. The
literature stream discussed in Sect. 2.2.1 usually defines this parameter as planned
lead time, which stems from the definition as used in MRP. The recently growing
research interest on advance demand information shows that in this field a similar
parameter to the planned lead time is established, which also rations the work
release to an MTS production system based on the due date the customer states.
This stream of literature is considered in Sect. 2.2.2.
2.2.1 Planned Lead Time Setting
Early papers that address the topic of planned lead time setting are Weeks (1981),
for the single-machine case, and Yano (1987) for the two-stage case. Yano (1987)
delivers a model for planned lead time setting based on the production lead time
distribution for each production step. The optimal planned lead time is found by
minimizing WIP, FGI, and backorder costs. The stochastic production lead time is
treated as an exogenous variable in her model. An extension of how to calculate the
planned lead time for a multi-stage production system is also sketched in
Yano (1987).
One extension of Yano (1987) is the finding of Gong et al. (1994), who show that
the optimal solution of the production lead time problem in an n-stage serial
production system is equivalent to an n-stage serial inventory model as studied by
Clark and Scarf (1960). They argue that this problem has a unique solution based on
the finding that such a problem can be tackled by solving a sequence of newsvendor
problems and based on the findings in Federgruen and Zipkin (1984). Another
extension provided by Elhafsi (2002) is the integration of penalty costs at each
production stage into the model. This extends Yano (1987) because there only a
backorder or penalty cost is paid when an order is late towards the customer. In
Elhafsi (2002) a recursive solution procedure for their more complex model is
developed. The model developed in Song et al. (2000) describes the optimal
planned lead time calculation for an assembly station again minimizing holding
and backorder costs. Additionally, in their model the optimal order quantity is
2.2 Work Release Rules 13
calculated based on overproduction costs. Inputs are the single product replenish-
ment lead time distributions. A heuristic to find the optimal planned lead time in
more complex assembly networks is then developed in Song et al. (2001). They
consider WIP, earliness as well as tardiness costs and show how to include a service
level constraint into their model. An approximate decomposition technique for a
multi-stage assembly network is further developed in Axsäter (2005) to identify the
planned lead times for minimizing WIP and tardiness costs.
In addition to the cost perspective on planned lead times, Weng (1996, 1999)
discuss profit maximization problems. Firstly Weng (1996) develop an optimiza-
tion problem with two different types of customers. In this model, type 1 customers
are sensitive to production lead time and the optimization problem involves the
determination of a constant customer required lead time quoted to these type
1 customers. The longer the values for customer required lead time, the lower the
profit with these type 1 customers. Type 2 customers are lead time insensitive.
Costs for WIP and FGI are balanced against the revenue of both customer groups.
For an M/M/1 queuing system, explicit expressions are derived and the optimal
order acceptance rates and optimal lead time quoted are discussed. In Weng (1999)
an optimization problem for a phase type service distribution is then set up. In this
model with similar assumptions to Weng (1996) it is found that marketing should
not only focus on customer order rate but also on quoted lead times and that a too
high utilization is much worse than a too low utilization. Another model not directly
focusing on profit maximization, but using the same idea that longer quoted lead
times lead to higher costs is Bertrand and van Ooijen (2008). In their model a
CONWIP policy is assumed and the optimal value for quoted lead time and WIP
level are searched for. A large cost reduction potential is found in their study which
compares their restricted WIP policy with an immediate order release to the
production system.
2.2.2 Advance Demand Information
One of the first articles to consider the possibility that some information about the
customer demand in advance of the respective demand can be used in MTS
production systems is Buzacott and Shanthikumar (1994). The difference between
safety stock and safety time within an MTS inventory replenishment system is
discussed. It is found that safety time is preferable to safety stock if there is a good
forecast, i.e., the customer required orders and their due dates are known for a long
period in advance. The safety time considered in this paper is defined as planned
lead time minus average production lead time. This concept presented in Buzacott
and Shanthikumar (1994) for safety time is equivalent to applying a planned lead
time in an MTO system but additionally a safety stock is held. Hariharan and Zipkin
(1995) discuss the influence of customer required lead time, which is equivalent to
advance demand information, for a single-stage and multi-stage MTS production
system. They test different reorder policies and find that in their model a constant
customer required lead time leads to the same result as a respective reduction of
supply lead time.
This concept of knowing the demand in advance even in stock replenishment
systems working under MTS is the basis for the literature stream on advance
demand information. In this stream of literature either the optimal replenishment
strategy or the optimal parameters of a predefined replenishment strategy are
optimized usually assuming a constant customer required lead time. Karaesmen
et al. (2002) for example analyze a discrete time single-stage MTS production
system with customer demand known a constant period in advance. For a base stock
policy the parameters order release time, which is equivalent to planned lead time,
and base stock level are optimized based on inventory and backorder costs. In
Karaesmen et al. (2004) a continuous time M/M/1 model with constant customer
required lead time and a base stock reorder policy, is discussed whereby the value
of advance demand information in such a system is shown. Additionally, in this
paper the simple MTO case with constant customer required lead time is explicitly
addressed, and for the M/M/1 case similar results to the ones found in Buzacott and
Shanthikumar (1994) are derived. The functional relationship between the base-
stock level and the length of the constant customer required lead time is evaluated
in Liberopoulos (2008). It is found that the base stock level can even drop to zero
under certain conditions if the customer required lead time is sufficiently long. This
case corresponds to an MTO production system without safety stock. The topic of
constructing an optimal reorder policy applying advance demand information is
addressed in Wijngaard (2004), Wijngaard and Karaesmen (2007), Tan et al.
(2007), and Gayon et al. (2009) for different production system assumptions
whereby all of these articles find a kind of base-stock policy to be optimal.
Gayon et al. (2009) is the first to also include a distribution of customer required
lead time values.
2.3 Capacity Investment
To generate some managerial insights from the results concerning the logistical
relationships as presented in Sect. 2.1, one method is to evaluate the economic
influence of these logistical key figures. In this section, some literature covering the
topic of optimizing operations costs or profit taking into account capacity costs,
inventory costs, backorder costs, and service level constraints, is discussed. The
literature on capacity investment can be divided into the queuing system design
stream (see Sect. 2.3.1), considering mainly capacity costs compared to customer
waiting time, and models also including backorder costs or service level constraints
(see Sect. 2.3.2).
2.3 Capacity Investment 15
2.3.1 Queuing System Design
The questions of how much capacity should be invested and whether a single or
multiple servers are optimal have a long research history. A good overview on the
topic of queuing system design is provided by Crabill et al. (1977) and Buzacott and
Shanthikumar (1992) or more recently in Rao et al. (1998) and Stidham (2002).
The roots of literature on this topic stem from Brigham (1955), Hillier (1963),
and Stidham (1970) all stating some conditions or queuing system specifications
where single servers lead to the optimal capacity and waiting time costs. This
already shows the typical optimization problem discussed in the queuing system
design literature stream where usually waiting time related costs and capacity
related costs are compared.
In Brigham (1955) the overall costs for WIP and number of servers are
minimized for an M/M/s queue by searching for the optimal number of servers.
In this paper the processing rate of servers is predefined. Hillier (1963) later on gave
a summary of different kinds of models all with parallel servers on a single
processing stage minimizing waiting and capacity costs. For a set of different
service disciplines and nonlinear waiting cost functions Stidham (1970) proves
that the single server queue leads to the minimum capacity and waiting time costs.
Brumelle (1971) states some conditions under which a single server system always
leads to lower WIP costs at the same capacity costs than a system of parallel servers.
Recently there have been some publications on conditions where the single server
setting is not optimal. Mandelbaum and Reiman (1998) discuss the optimality of
single or multiple servers in a queuing system with phase type service times. While
minimizing waiting times it is shown that multiple servers are optimal at high
coefficients of variation and at high utilization values. MacGregor-Smith (2007)
formulates a set of optimization problems for an M/G/s/K queue with different
objectives and different variables to optimize also including number of servers.
Other objective functions in queuing system design models can be found in
Driscoll and Weiss (1976) where the processing rate for reaching a predefined
waiting time with a certain probability is developed in the M/M/1 case. This
objective function can directly be transformed to a manufacturing context where
the required processing rate to reach a predefined service level is calculated when
the customer has a constant required lead time. De Cani (1962) discusses a profit
maximization problem where not only waiting time costs and capacity costs are
considered, but also the profit loss of customers not joining the queue because of the
queue length. This paper also includes the behavior mentioned in Chao and Scott
(2000) that for some customers, waiting before the service may be worse than
waiting when serviced. Chao and Scott (2000) show that for a G/G/s queuing
system the average and stochastic number in queue (without the items in service)
is a decreasing function of s. This means that the waiting time before service
decreases when more servers are available if the sum of all processing rates is
kept constant. Another influence on the customer arrival rate can be the price
charged to the customer as discussed in Stidham (1992), where for a G/G/1 queue
the price and the processing rate are optimized. A system with two priority classes
of customers is considered in Wierman et al. (2006) for an M/Ph/s queue and it is
shown that the optimal number of servers increases when job size variability
increases.
This subset of queuing system design literature shows that the basic optimization
problem balances capacity costs against the waiting time costs, which are equiva-
lent to WIP costs, neglecting other effects of MTO or MTS production systems.
This becomes very clear in the review by Buzacott and Shanthikumar (1992), who
examine the application of queuing models for the design of manufacturing
systems. However, in any manufacturing context, be it MTO or MTS, backorder
costs, lost sales or service level constraints can be important topics.
2.3.2 Backorder Cost or Service Level Models
For production systems with stochastic customer demand, the capacity invested and
the inventory held are economic substitutes. All of the following models either
discuss the FGI, mostly in MTS systems, or they discuss a joint FGI and WIP, but
the inventory mentioned is never only the WIP, as was the case in Sect. 2.3.1. This
relationship of inventory as an economic substitute for capacity invested is implic-
itly included in many of the following papers setting up optimization problems.
Explicitly is this relationship mentioned, for example, in Angelus and Porteus
(2002), who derive an optimal simultaneous capacity and inventory plan for an
MTS production system under stochastic demand. Van Mieghem and Rudi (2002)
also address the joint (inventory against capacity investment) decision problem.
The model is based on periodic production and demand occurrences. Furthermore,
Bradley and Glynn (2002) consider a joint optimization of capacity and inventory
decisions in a single-item, single-machine, MTS production system with the objec-
tive of minimizing the long-run average operating cost due to penalty, holding, and
capacity costs. A profit maximization problem including capacity costs and inven-
tory costs based on a known production plan for aggregated product groups is set up
in Zäpfel (1998). One study ignoring the influence of capacity utilization on service
level and lost sales is Coelli et al. (2002), which studies the profit lost due to unused
capacity, taking into account the maximum possible capacity depending on fixed
and variable input factors. They apply their model to analyze 28 firms (airlines) and
find a high potential for more profit by increasing capacity utilization. Rajagopalan
and Swaminathan (2001) explore the relationship between production planning and
capacity acquisition in a multi-item and multi-period environment with known,
varying and long-term growing demand. A balance between capacity leading (for
definition see Slack and Lewis (2002)), resulting in low average inventory with high
average capacity, and capacity lagging (for definition see Slack and Lewis (2002)),
leading to high inventory with low average capacity, is discussed. The capacity
lagging leads to high inventory because of the preproduction for the increasing
demand.
2.3 Capacity Investment 17
The differentiation between constant capacity available and flexible capacity

available is also considered in recent literature. For example, Raman and Kim
(2002) study the influence of inventory holding costs and flexible capacity at an
apparel manufacturer. They conclude that an increase in capacity can reduce WIP
and FGI as well as costs for markdown and stockouts. The trade-off between
permanent and contingent capacity taking inventory holding cost into account for
an MTS production system is discussed in Mincsovics et al. (2009) for a
non-stationary stochastic demand situation. They show that the value of flexibility
decreases with an increasing capacity acquisition lead time. A mathematical model
for optimizing the capacity extension (based on a currently available capacity) in a
multi-machine, multi-item production system with lost sales is developed by Zhang
et al. (2004). Capacity investment costs are balanced against lost sales for a
stochastic demand and a capacity plan minimizing total cost is delivered. Another
paper covering the topic of flexible capacity is van Mieghem (1998), which
discusses optimal capacity investments balancing capacity costs against revenue.
In this model, flexible capacity has again a positive influence on revenue.
Chapter 3
Single-Stage Service Level and Tardiness
Model
3.1 Introductory Remarks
As already discussed in the introduction and in the literature review, the two
performance indicators service level and average tardiness are important measures
of the logistical performance a company provides.
For production systems, the relationship between utilization, WIP, and produc-
tion lead time is analytically well defined in the literature. Nonetheless, this
relationship lacks a link to the customer’s perspective as shown in Chap. 2. This
lack of analytical models to identify the link between the internal production system
behavior and the external customer related measures of service level and tardiness
for MTO production systems leads to the model developed in this chapter. The
distribution of customer required lead time is combined with the distribution of
production lead time which leads to a set of equations defining the service level, the
expected FGI lead time, the expected FGI, and the expected tardiness in a general
form for a single-stage production system working under MTO. Assuming expo-
nentially distributed customer required lead time, explicit expressions are provided
for an M/M/1 queue. Additionally, a set of equations for the aforementioned
measures is developed for the application of a WAW policy (see Hopp and
Spearman (2008) for WAW). It is proven, that such a WAW policy keeps the
Poisson input stream in the production system for an arbitrary customer required
lead time distribution. For the G/G/1 case the influence of interarrival time as well
as processing time variation is discussed based on the production lead time approx-
imation of Chen and Yao (2001).
For capacity investment decisions this model provides equations to balance the
service level, tardiness, and FGI against capacity invested and WIP held. Such
problems are addressed in Chaps. 4, 5 and 6. Additionally, the influence of the
WAW work release policy on FGI, service level, and tardiness is discussed. In a
numerical example the application of such a WAW policy is found to significantly
reduce FGI, especially for high service level targets. It can therefore be applied as a

20 3 Single-Stage Service Level and Tardiness Model
managerial tool to stay competitive in the more and more service level oriented
environment which manufacturing companies face.
The remainder of this chapter based on Altendorfer and Jodlbauer (2011) is
structured as follows: Sect. 3.2 provides the analytical development of the model in
a general form as well as for the M/M/1 production system. In Sect. 3.3 a numerical
study is provided to show the application of the developed model and examine the
FGI reduction potential of the WAW policy. In Sect. 3.4 some concluding remarks
are stated. All proofs are added in the Appendix.
3.2 Model Development
The model developed in this chapter provides a deeper understanding of the

relationship between production lead time, utilization, and WIP in a single-stage
production system and the service level as well as tardiness of such a system. An
extension of the model is presented to include the possibility of reducing FGI. This
model can be used by decision makers in cases of capacity investment decisions to
define the capacity needed for a certain target service level or a certain target
average tardiness and to define an optimal WAW policy.
3.2.1 Model Assumptions and Notation
Figure 3.1 shows the single-stage production system with a WIP and an FGI which
is discussed. Customer orders arrive at the manufacturing system according to a
Poisson process with mean arrival rate λ. Each customer requests an individual due
date value that determines the non-negative customer required lead time. This
^ or an exponential random
customer required lead time can either be constant, i.e. L,
variable, i.e. L with fL ðÞ denoting the probability density function (pdf), FL ðÞ the
respective cumulated distribution function (cdf), and 1=β its mean. The customer
required lead time cannot be influenced by the production system and all customer
orders are accepted. Customer orders are processed by the single machine with
exponential processing rate μ and utilization ρ following a first-in-first-served (FIFS)
discipline. W denotes the random variable and 1=k the mean of production lead time
needed for one order from release to the buffer in front of the machine until its
completion and delivery to the FGI buffer. fW ðÞ denotes the pdf of the production
lead time and FW ðÞ the respective cdf. I denotes the random FGI lead time, which is
the time between arrival at the FGI buffer and the due date. C is the random
tardiness which is the time between due date and delivery date whenever an order
is late. Y and G are the random WIP and FGI respectively. η denotes the service
level, which is the probability that an order is delivered on time.
3.2 Model Development 21
Customer required Tardiness

lead time = L C = max (0, W–L)
Customer
Service level
h = P(W <= L)
Buffer Buffer
Machine
FGI = G
WIP = Y
FGI lead time

Production lead time = W I = max (0, L–W )
Fig. 3.1 Single-stage production system
The customer orders are released to the production system, i.e. to the buffer in front
of the machine, whenever their remaining time to the due date is smaller or equal to
the WAW X (see also Jodlbauer (2008b) or Hopp and Spearman (2008)). The WAW
policy leads to the situation that all the orders with a customer required lead time being
greater than the WAW are transferred to an order list (or WAW buffer) when they
arrive at the production system. Their release to the production system is triggered
when their remaining time to due date becomes smaller than the WAW. To ensure the
M/M/1 assumption, the WAW policy must not disturb the Poisson input stream into
the production system. The necessary property is stated in Proposition 3.1.
Proposition 3.1. For every Poisson arrival process the application of the WAW
policy leads to a Poisson input stream into the production system for any customer
required lead time distribution.
Proof see Appendix.
3.2.2 Queuing Model with Deterministic Customer Required

Lead Time
In this part some available queuing theory findings are applied to define the basic
equations for the developed approach. In this section the WAW policy is neglected,
i.e. X ¼ 1, and customer required lead time is constant.
3.2.2.1 M/M/1 Queue
The M/M/1 queue, which consists of a Poisson (Markovian) input and a Poisson
(Markovian) service process with one server, is the most basic queuing model.
Nevertheless, it is often applied to prove some basic relationships in operations
research (e.g. Stidham (1970), Buzacott and Shanthikumar (1992), and Karaesmen
et al. (2004)). Transformed to the manufacturing setting this model has exponen-
tially distributed interarrival times between two consecutive orders and exponen-
tially distributed processing times of the orders which are processed at one
machine. The following equations for WIP, production lead time, and utilization
can be stated (see Tijms (2003) and Medhi (1991)):
fW ðtÞ ¼ ðμ λÞeðμλÞt
λ
ρ¼
μ
1 1 (3.1)
E½W ¼ ¼
μλ k
λ
E½Y ¼
μλ
This model can be applied to show the nonlinear increase of WIP and production
lead time with respect to the utilization. Furthermore, the M/M/1 model provides an
exponentially distributed production lead time and a Poisson output stream. For a
multi-stage model this Poisson output stream means that several processing stages
can be linked, and as long as the first input stream is Poisson and all the processing
time distributions are exponential, this production system can be modeled as a
series of M/M/1 queues.
3.2.2.2 Service Level, FGI Lead Time, and Tardiness
When an order has a certain customer required lead time, the on-time probability of
this order is equal to the probability that the production lead time is shorter than this
value for customer required lead time (see Duenyas and Hopp (1995) and Liu and
Yuan (2001)):

η ¼ P W L^ ¼ FW L^ ¼ 1 eðμð1ρÞLÞ
^
(3.2)
Each order in the production system which has a production lead time being
smaller than the deterministic customer required lead time has the FGI lead time
þ
I ¼ L^ W and 0 otherwise, so I ¼ L^ W holds. Based on the customer required
lead time and on the exponentially distributed production lead time the following
equation holds for the expected value of the FGI lead time:
ðL^ ðL^

E½I ¼ fW ðτÞ L τ dτ ¼ μð1 ρÞeðμð1ρÞτÞ L^ τ dτ
^
0 0 (3.3)
^ ð1 ρÞ þ e ðμð1ρÞL^Þ
Lμ 1
¼
μ ð 1 ρÞ
Based on Little’s law (Little (1961)) the expected value of FGI can then be
calculated as:
E½G ¼ E½Iλ (3.4)
The same structure as for FGI lead time also applies for tardiness, so
þ
C ¼ W L^ and (see Duenyas and Hopp (1995) for the integral form):
ð
1 ð
1

E½C ¼ fW ðτÞ τ L^ dτ ¼ μð1 ρÞ eðμð1ρÞτÞ τ L^ dτ
L^ L^ (3.5)
ðμð1ρÞL^Þ
e
¼
μð1 ρÞ
These Eqs. (3.1, 3.2, 3.3, 3.4 and 3.5) from queuing theory will be linked to the
distribution of customer required lead time in the next section.
3.2.3 Distribution of Customer Required Lead Time
A distribution of customer required lead time is implemented in this section

whereby the WAW policy is still neglected. If the customer required lead time is
not a deterministic number, but follows a certain distribution, for each possible
value the customer required lead time can take, the service level for that customer
required lead time value can be calculated according to Eq. (3.2). By weighting
each of these resulting service levels according to the probability of occurrence of
the respective customer required lead time value the following expected service
level results:
ð
1 ð
1
β
η¼ FW ðτÞfL ðτÞdτ ¼ 1 eðkτÞ βeðβτÞ dτ ¼ 1 (3.6)
k¼μð1ρÞ kþβ
0 0
Derivation see Appendix.

Equation (3.6) is, as well as all the equations developed in the remainder of
Sect. 3.2, in its integral form valid for any production lead time and customer
required lead time distribution. Calculating the limit ρ ! 1 shows that the service
level approaches 0 independently of the distribution parameter β of customer
required lead time. For ρ ! 0 (i.e. λ ! 0) the service level reaches a value below
100 % ð1 β=ðμ þ βÞÞ depending on the relation between processing rate μ and the
parameter of the customer required lead time distribution β. The result for ρ ! 0 can
intuitively be argued by the fact that each order needs a certain time to be processed
which depends on μ and its on-time probability is at ρ ! 0 exactly equal to the
probability that the processing time is shorter than the customer required lead time.
Note that in the case of ρ ! 0 with μ ! 1 it clearly follows that η ! 1.
Weighting the expected FGI lead time values from Eq. (3.3) with their probabil-
ity of occurrence leads to:
ð ðθ
1 ð
1
θk þ eðkθÞ 1
E½I ¼ fW ðτÞðθ τÞdτfL ðθÞdθ ¼ fL ðθÞdθ
k
0 0 0 (3.7)
1 1
¼
β kþβ

For the limit ρ ! 1 the expected value of FGI lead time approaches 0. From
service level equal to 0 at ρ ! 1 it is intuitively clear that nothing can be in the FGI
when the utilization reaches 100 %. For ρ ! 0 (i.e. λ ! 0) a certain maximum
expected value of FGI lead time can be calculated as μ=ðβðμ þ βÞÞ. This value
decreases in β which means the maximum expected FGI lead time value increases
when the expected customer required lead time increases. The maximum expected
FGI lead time increases in μ which means that a higher processing rate leads to a
higher maximum expected value for the FGI lead time.
The expected tardiness from Eq. (3.8) is calculated analogous to the expected
FGI lead time in Eq. (3.7):
ð1
1 ð ð
1 ð
1
ekθ ekθ βθ
E½C ¼ fW ðτÞðτ θÞdτfL ðθÞdθ ¼ fL ðθÞdθ ¼ βe dθ
k k
0 θ 0 0 (3.8)
β
¼
k ðk þ β Þ
For ρ ! 1 the expected value of tardiness approaches 1. The reason is that the
production lead time approaches 1 and so the tardiness has to have the same
behavior. In the case of ρ ! 0 (i.e. λ ! 0) a certain minimum value exists for the
expected tardiness reached ðβ=ðμðμ þ βÞÞÞ similarly to the expected FGI lead time.
This value decreases in μ which means the higher the processing rate is, the lower
the minimum expected tardiness. Furthermore, it increases in β which means the

higher the expected customer required lead time, the lower the minimum expected
tardiness is.
For the calculation of the FGI, Eq. (3.4) still holds in the case of distributed
customer required lead time values.
3.2.4 Integration of a WAW Work Release Policy
When a WAW work release policy is applied, no order released to the production
queue has a remaining customer required lead time value longer than the WAW X.
This means that the distribution of remaining customer required lead time as
observed by the production system at order release is a truncated version of the
original one. At order release, all orders with a customer required lead time L > X
have a remaining customer required lead time of X. Based on the remaining
customer required lead time at order release, the integral from Eq. (3.6) for the
service level can be split up into two parts, whereby the second part of Eq. (3.9)
covers all values L > X:
ðX ð
1
keðkþβÞX þ β
η ¼ FW ðτÞfL ðτÞdτ þ FW ðXÞfL ðτÞdτ ¼ 1 (3.9)
kþβ
0 X

For limit ρ ! 1 the service level approaches 0, which is the same behavior as in
limit ρ ! 0 (i.e.
Eq. (3.6). When calculating λ ! 0) the following maximum service
ðμþβÞX
level can be calculated 1 μe μþβ
þβ
which is smaller than the service level
without WAW.
The two equations for the expected FGI lead time and the expected tardiness are
created by using Eqs. (3.7) and (3.8) for two different cases. The first case is that the
customer required lead time is smaller than the WAW and the second case is that
customer required lead time is larger than the WAW. Whenever L X holds true,
Eq. (3.7) also holds in the case of conducting a WAW policy. For the situation
where L > X, the realized customer required lead time in the production system is X.
The first integral of Eq. (3.7) integrates over all possible customer required lead
time values, so this integral has to be split up into the two parts. Adding up these
two parts leads to the following equation:
ðX ðθ ð ðX
1
E½I ¼ fW ðτÞðθ τÞdτfL ðθÞdθ þ fW ðτÞðX τÞdτfL ðθÞdθ

0 0 X 0 (3.10)
ðkþβÞX βX
βe ðk þ βÞe þk
¼
β ðk þ β Þ

Calculating the limit ρ ! 1 shows that the FGI lead time approaches 0 (see
Appendix), which is the same case as for Eq. (3.7). For ρ ! 0 (i.e. λ ! 0) the FGI
ðμþβÞX βX
lead time leads to the following maximum value βe ðμþβÞe þμ
βðμþβÞ which is
smaller than μ=ðβðμ þ βÞÞ (see Appendix) for the case without WAW policy.
The expected tardiness can be stated analogous to the expected FGI lead time, so
the first integral of Eq. (3.8) also has to be split up into two parts and then added to
the expected tardiness. This leads to the following equation:
ðX 1
ð ð1
1 ð
E½C ¼ fW ðτÞðτ θÞdτfL ðθÞdθ þ fW ðτÞðτ XÞdτfL ðθÞdθ
0 θ X X
(3.11)
ðX ð
1
ekθ βθ ekX ðβθÞ keðkþβÞX þ β
¼ βe dθ þ βe dθ ¼
k k k ðk þ β Þ
0 X
Calculating the limit ρ ! 1 shows that the tardiness approaches

ðμþβÞX1.For ρ ! 0
(i.e. λ ! 0 ) a minimum tardiness level can be calculated μe μðμþβÞþβ which is
higher than in the case without WAW.
Equation (3.4) still holds for the calculation of the FGI in the case of distributed
customer required lead time values and the application of a WAW policy.
Proposition 3.2. Whenever the WAW policy is applied to a production system, the
expected FGI lead time and the service level with the WAW policy are lower than
without the WAW policy and the expected tardiness with the WAW policy is higher
than without the WAW policy if the pdf of production lead time is not affected by the
WAW policy.
Proof see Appendix.
Proposition 3.3. Increasing the mean of the exponential distribution for produc-
tion lead time in an M/M/1 production system facing exponentially distributed
customer required lead time without changing the distribution of customer required
lead time and without changing the input rate leads to a reduction of service level,
expected FGI, and expected FGI lead time as well as to an increase in expected
tardiness.
Proof see Appendix.
Based on Proposition 3.1, the Proposition 3.2 holds for an M/M/1 production
system with arbitrary customer required lead time distribution. It shows that in a
system with a WAW policy no improvement of FGI can be achieved without
reducing the service level (or equivalently the utilization as shown later on).
Nevertheless, the balance between service level and FGI or utilization and FGI
can be discussed. Based on Eq. (3.4), the following equation can be stated for the
expected value of FGI:

ρμkeðkþβÞX ρμβ ρμk eβX 1 ρμβ
E½G ¼ λE½I ¼ þ
λ¼ρμ kðk þ β Þ k ðk þ β Þ kβ kβ
ðkþβÞX βX (3.12)
ρμ e 1 ρμ e 1
¼
kþβ β
Calculating the limit ρ ! 1 and ρ ! 0 (i.e. λ ! 0) for Eq. (3.12) delivers the
expected FGI value of 0. This means in both extreme cases for the utilization there
is no FGI in pieces available. The comparison to the limit ρ ! 0 of the FGI lead
time shows, that even with the maximum expected FGI lead time at utilization
0 there is no expected FGI in pieces available. The reason for that is that at
utilization zero, the input rate has to be zero and for that reason no products are
produced.
The utilization at which the expected FGI reaches its maximum can implicitly be
stated by:

μeðμð1ρÞþβÞX ð1 þ XρμÞ μ ρμ2 eðμð1ρÞþβÞX 1 μ eβX 1
þ ¼ (3.13)
μ ð 1 ρÞ þ β ð μ ð 1 ρÞ þ β Þ 2 β

Equation (3.13) can only be solved numerically. The utilization value found with
this equation gives the production manager the information whether an increase in
utilization still leads to an increase in average FGI or if it already leads to a
decrease, which is a valuable information concerning FGI costs and FGI storage.
The application of a WAW policy within an M/M/1 production system facing
exponentially distributed customer required lead time leads to the following utili-
zation and FGI reduction in comparison to an M/M/1 production system without
WAW policy for equal service level values between the two systems:
1 ρX ðμð1ρX ÞþβÞX
Δρ ¼ e (3.14)
1η

ρX μ eβX 1 ðρX þ ΔρÞμk ρX μ eðkX þβÞX 1
ΔE½G ¼ þ (3.15)
β ðk þ βÞβ kX þ β
5 140
120
4
100
ΔE[G]
ΔE[G]
3 80
2 60
40
1
20
0 0
0% 10% 20% 30% 40% 0% 2% 4% 6% 8% 10%
Δρ Δρ
μ = 0.5; β = 0.05; ρX = 0.6 μ = 2; β = 0.0125; ρX = 0.9
Fig. 3.2 Trajectory utilization loss versus FGI reduction in M/M/1 model
whereby Δρ is the utilization loss and ΔE½G is the FGI reduction when in the
production system applying the WAW policy, the utilization ρX and mean produc-
tion lead time 1=kX are reached. Derivation see Appendix.
Based on Eqs. (3.14) and (3.15), management can discuss the balance between
utilization loss and FGI cost reduction and define an optimal strategy based on
utilization loss costs and FGI costs. The trajectory shown in Fig. 3.2 can be defined
for each combination of μ, β, and ρX .
A detailed numerical study discussing the FGI reduction potential is presented in
Sect. 3.3.2.
The influence of increasing the mean of the exponential distribution for produc-
tion lead time on service level, FGI, FGI lead time, and tardiness is identified in
Proposition 3.3.
3.2.5 Application for G/G/1 Production System
A single machine production system with a general input and a general service
process is called a G/G/1 queue. Since for the general processes it is not possible to
state exact equations for pdf and mean of production lead time, the following
approximations are developed by Chen and Yao (2001):
2ð1 ρÞ λð2αð21ρ Þ
t
fW ðtÞ ¼ 2 e a þα2s Þ
λ αa þ αs 2
λ
ρ¼ (3.16)
μ

λ α2a þ α2s
E½ W ¼
2ð 1 ρÞ
3.3 Numerical Study 29
Whereby αa and αs are the coefficients of variation of interarrival and processing

time distribution respectively. Chen and Yao (2001) argue that the production lead
time in a G/G/1 queue can be approximated with an exponential distribution.
Applying this approximation from Chen and Yao (2001), the mean of the
production lead time increases whenever the coefficient of variation of the
interarrival time distribution or the processing time distribution increases. In com-
bination with Proposition 3.3, this shows that the results for service level, expected
FGI, expected FGI lead time, and expected tardiness depend, for a more general
system, on the second moment of the processing and interarrival time distribution.
Especially, the link to Proposition 3.3 illustrates that the lower the variation of
processing time and interarrival time, the better the results for service level,
expected FGI, expected FGI lead time, and expected tardiness will be.
3.3 Numerical Study
In this section, firstly a comparison between three examples shows the typical
characteristic of the relationships between utilization, expected WIP, expected
FGI, expected FGI lead time, expected tardiness, and service level. The advantage
of implementing a WAW is discussed in this section based on the calculated curves
and some general insights from the shape of the curves are given. The second part of
this section presents the result of an extensive experiment set to evaluate the FGI
reduction potential based on the implementation of a WAW work release policy.
3.3.1 Relationship Between Logistical Key Figures
In this section a numerical example is conducted with three different sets of

parameters. The results for these three different sets of parameters for the M/M/1
queuing model are compared according to the logistic relationship between utiliza-
tion, WIP, production lead time, service level, FGI, FGI lead time, and tardiness for
an exponentially distributed customer required lead time. Table 3.1 shows the
parameter set for the numerical example.
In the numerical example the production rate μ is held constant and the input rate
λ is varied to find different points of the curves. All resulting figures in this section
are shown with the expected WIP on the x-Axis. This is the same format as used in
Wiendahl and Breithaupt (1999); Wiendahl et al. (2005); Jodlbauer (2008a) or also
partly in Hopp and Spearman (2008).
Figure 3.3 shows the relationships between logistical key figures for the three
examples. The service level calculated without an additional WAW in Example A
and B is the maximum reachable service level if each order is released to the
production system without any delay and the FIFS dispatching rule is applied (see
Proposition 3.2). The reduction of service level with increasing WIP, as shown in
Table 3.1 Numerical Example X β μ

examples for logistical
relationships in M/M/1 model A 1 0.05 0.5
B 1 0.2 0.5
C 20 0.05 0.5
a b
1 20
0.9 18
0.8 16
FGI lead time

14
Service level
0.7
0.6 12
0.5 10
0.4 8
0.3 6
0.2 4
0.1 2
0 0
0 5 10 15 20 25 30 0 5 10 15 20 25 30
WIP WIP
Example A Example B Example C Example A Example B Example C
c d
60
6
50
5
40
Tardiness
4
FGI
3 30
2 20
1 10
0 0
0 5 10 15 20 25 30 0 5 10 15 20 25 30
WIP WIP
Example A Example B Example C Example A Example B Example C
Fig. 3.3 Service level, FGI lead time, and tardiness in M/M/1 model
Fig. 3.3a, is a result of the increasing expected production lead time (increasing λ).
The maximum service level, maximum FGI lead time, and minimum tardiness
values for utilization 0 (i.e. WIP is zero from Eq. (3.1)) as well as the asymptotical
behavior of the curves for utilization approaching 100 % (i.e. WIP is 1 from
Eq. (3.1)) is shown in Fig. 3.3. The restated equations for service level, FGI, FGI
lead time, and tardiness with respect to the expected WIP when μ is predefined and
the first derivatives showing the monotonicity behavior of service level, FGI lead
time, and tardiness can be found in the Appendix. For expected FGI, Eq. (3.13)
provides the utilization at which the expected FGI has its maximum. Applying
Eq. (3.1) shows that for this utilization only one solution for the expected WIP
(denoted as E½YG ) exists. Therefore, it follows that expected FGI increases in

expected WIP for E½Y < E½YG and it decreases if E½Y > E½YG .
The result that with a smaller average value for the customer required lead time
the service level decreases faster, the FGI lead time decreases faster and the
tardiness increases faster when WIP increases (see comparison of Example A
with B) is consistent with intuition. Example C shows that the integration of a
WAW can reduce the FGI and for this reason the costs.
The comparison of Example A and C in Fig. 3.3a and d shows the influence of a
policy reducing the longest customer required lead time values on service level and
tardiness. Such a WAW policy leads to a reduction of service level and an increase
of expected tardiness. Figure 3.3b, c show the expected FGI lead time and the
expected FGI in items. The comparison with a WAW of 20 periods, which is
exactly the expected value of the exponential distribution of customer required
lead time, shows that approximately half of the FGI in pieces can be reduced with
this policy. The service level as well as the expected tardiness at low WIP values are
only influenced marginally by this policy. This leads to the managerial insight, that
especially the orders with long customer required lead time values should not be
released immediately. A detailed numerical study about the FGI reduction potential
of such a WAW policy is given in the next section.
3.3.2 FGI Reduction Potential with WAW Policy
To identify the potential of reducing expected FGI when a WAW work release
policy is applied, a broad spectrum of test instances has been generated. All
possible combinations of the following parameter settings have been tested:
• η 2 f0:5; 0:8; 0:9; 0:95; 0:98; 0:99g;
• λ 2 f0:5; 1; 2; 4; 8; 16; 32g;
• β 2 f0:0125; 0:025; 0:05; 0:1; 0:2; 0:4; 0:8; 1:2g;
• ρ 2 f0:5; 0:7; 0:8; 0:9; 0:95; 0:98; 0:99g:
For each test instance the FGI reduction potential is calculated based on the FGI
with X ¼ 1 and the FGI with the minimum required WAW that exactly meets the
service level constraint η (for a derivation based on Eq. (3.9) see Appendix –
Chap. 6). The values in Table 3.2 correspond to the average FGI reduction potential
with respect to the order arrival rates for which it is possible to reach this service
level constraint. This means that for each combination of ρ and β the value in the
respective η tables consists of a different number of λ instances for which the service
level constraint η can be reached. The results of the numerical study show that the
higher η is, the lower is the number of λ instances being able to reach it (“—” in
Table 3.2 means that the service level target cannot be reached for any λ value
concerning the respective combination of ρ and β). This is the reason why the FGI
reduction potential in Table 3.2 is not strictly increasing in certain parameters, but
32
Table 3.2 Results from numerical study of FGI reduction in M/M/1 model
η ρ η ρ
50 % 0.5 0.7 0.8 0.9 0.95 0.98 0.99 Avg 95 % 0.5 0.7 0.8 0.9 0.95 0.98 0.99 Avg
β 0.0125 100 100 99 98 96 93 92 97 β 0.0125 98 98 95 97 96 92 88 96
0.025 100 99 99 96 86 92 90 95 0.025 95 97 94 96 94 89 — 95
0.05 99 98 97 88 84 90 87 93 0.05 95 97 92 95 92 — — 94
0.1 99 96 91 86 81 87 83 90 0.1 94 96 90 93 — — — 93
0.2 97 87 89 83 76 83 74 87 0.2 92 95 86 — — — — 91
0.4 91 85 87 79 69 75 — 84 0.4 90 92 79 — — — — 88
0.8 89 82 84 73 57 — — 81 0.8 86 — — — — — — 86
1.2 92 88 88 80 66 — — 87 1.2 89 — — — — — — 89
Avg 96 93 93 88 82 90 88 91 Avg 94 97 92 96 95 91 88 94
η ρ η ρ
80 % 0.5 0.7 0.8 0.9 0.95 0.98 0.99 Avg 98 % 0.5 0.7 0.8 0.9 0.95 0.98 0.99 Avg
β 0.0125 99 99 97 91 88 92 90 94 β 0.0125 99 98 98 97 96 90 – 97
0.025 99 97 93 90 86 90 87 93 0.025 98 98 97 96 94 — — 97
0.05 97 90 92 88 82 87 81 90 0.05 98 97 97 94 — — — 97
0.1 93 89 90 85 77 81 — 88 0.1 97 96 95 — — — — 97
0.2 92 86 88 81 69 — — 86 0.2 97 94 — — — — — 96
0.4 90 83 85 74 51 — — 82 0.4 95 — — — — — — 95
0.8 88 79 79 59 — — — 81 0.8 — — — — — — —
1.2 90 84 82 — — — — 87 1.2 — — — — — — —
Avg 94 90 91 86 81 89 87 90 Avg 98 97 97 96 95 90 97
η ρ η ρ
90 % 0.5 0.7 0.8 0.9 0.95 0.98 0.99 Avg 99 % 0.5 0.7 0.8 0.9 0.95 0.98 0.99 Avg
β 0.0125 99 97 94 96 95 92 89 96 β 0.0125 99 99 98 98 96 — — 98
0.025 98 97 93 96 94 89 84 95 0.025 99 98 98 97 — — — 98
0.05 94 96 91 94 92 84 — 93 0.05 98 98 97 — — — — 98
0.1 93 95 89 93 89 — — 92 0.1 98 97 — — — — — 97
0.2 91 94 87 90 — — — 91 0.2 97 — — — — — — 97
0.4 89 92 82 — — — — 88 0.4 — — — — — — —
0.8 87 89 72 — — — — 84 0.8 — — — — — — —
1.2 89 80 — — — — — 86 1.2 — — — — — — —
Avg 94 95 90 95 94 89 87 93 Avg 99 98 98 97 96 98
3 Single-Stage Service Level and Tardiness Model
3.4 Concluding Remarks 33
only correlations can be stated. Note that whenever the service level constraint can
be met, there is also an FGI reduction potential.
On average, the numerical study shows that higher service level targets correlate
with a higher FGI reduction potential and lower utilization values correlate with a
higher FGI reduction potential. Table 3.2 shows that the application of such a
WAW leads to a considerable FGI reduction. The lowest FGI reduction potential
found concerning the average values is still 50 %. The original data including a
single value for each input rate λ indicates that in all the test instances created the
lowest FGI reduction is 30 %. This means whenever the targeted service level can
be reached, there is a high potential to reduce FGI with such a WAW policy.
The result of this numerical study leads to an interesting managerial insight since
the competition between production companies is more and more forced into the
field of logistical performance and it is clearly shown that implementing such a
WAW policy helps to keep costs down for high service levels.
3.4 Concluding Remarks
This chapter introduced an analytical model for relating the external logistical key
figures service level, FGI, FGI lead time, and tardiness to the internal logistical key
figures WIP, utilization, and production lead time for a single-stage production
system. Exact expressions for service level, expected FGI lead time, expected FGI,
and expected tardiness are presented based on an exponentially distributed
customer required lead time for an M/M/1, MTO production system. Furthermore,
the effect of the WAW work release policy on these logistical figures can be
determined with the model. Especially in the case of capacity investment decisions,
the service level and tardiness values reached can be balanced against the capacity
invested and the costs for holding FGI.
The implementation of the WAW work release policy enables the discussion of
reducing FGI. The numerical study leads to the managerial insight that a company
using the WAW policy for order release to the production system can significantly
reduce the FGI in comparison to releasing all orders directly to the production
system. Especially for high service level targets and moderate utilization values, the
FGI reduction potential of such a WAW policy is quite high.
Appendix
Proof of Proposition 3.1. The comparison of a stochastic customer required lead

time value L to a deterministic WAW value X leads to a random assignment of
orders to type 1 and type 2 orders. Type 1 orders have L X with probability p and
type 2 orders have L > X with probability ð1 pÞ. Based on the splitting property of
the Poisson input stream with rate λ, the two resulting streams of events are again
Poisson streams with rates φ ¼ pλ and ψ ¼ ð1 pÞλ respectively (see Tijms
(2003)). The Poisson stream 1 (L X) with rate φ ¼ pλ directly feeds the M/M/1
production system. The Poisson stream 2 with rate ψ ¼ ð1 pÞλ feeds the buffer of
units waiting to be released to the system. This WAW buffer can be transformed to
an M/G/1 queuing system since the service time has an arbitrary distribution,
which is based on the arbitrary customer required lead time distribution, and each
order has an own server since the waiting time is L X . For such an M/G/1
queuing system, the output process is a Poisson process with the same rate as the
input process, which is ψ ¼ ð1 pÞλ (see Mirasol (1963) and Newell (1966)). This
output process of the WAW buffer feeds the M/M/1 production system. Based on
the merging property of two Poisson processes, the input process for the M/M/1
production system is Poisson with rate λ ¼ φ þ ψ (see Tijms (2003)).
Derivation of Equation (3.6). Figure 3.4 visualizes The probability of an order
being delivered on time in the production lead time W and customer required lead
time L space. The shaded area indicates when an order is delivered on time.
Assuming that the distribution of W and L are independent of each other (which
holds true for FIFS dispatching discipline) the following equation for the service
level holds:
ð1
1 ð ð
1 ð
1 ð
1
η¼ fW ðτÞfL ðθÞdθdτ ¼ f W ðτ Þ fL ðθÞdθdτ ¼ fW ðτÞð1 FL ðτÞÞdτ

0 τ 0 τ 0
ð
1 ð
1
¼1 fW ðτÞFL ðτÞdτ ¼ FW ðτÞFL ðτÞj1

0 fW ðτÞFL ðτÞdτ
(3.17)
0 0
ð
1
¼ FW ðτÞfL ðτÞdτ
0
with FL ð0Þ ¼ 0; FL ð1Þ ¼ 1; FW ð0Þ ¼ 0; FW ð1Þ ¼ 1;
Appendix 35
Fig. 3.4 On-time probability 30

in joint probability space of
W and L
25
Production lead time W

20
15
10
0
0 5 10 15 20 25 30
Customer required lead time L
Derivation of Equation (3.7).

ð ðθ
1 ð
1
θk þ ekθ 1 βθ
E½I ¼ fW ðτÞðθ τÞdτfL ðθÞdθ ¼ βe dθ
k
0 0 0
01 1
ð 1ð ð
1
β@
¼ θkeβθ dθ þ eðkþβÞθ dθ eβθ dθA
k
0 0 0 (3.18)
01 1
ð
β 1 1
¼ @ θkeβθ dθ þ A
k kþβ β
0

β k 1 1 1 1
¼ þ ¼
k β2 k þ β β β kþβ

ðX ð
1
η ¼ FW ðτÞfL ðτÞdτ þ FW ðXÞfL ðτÞdτ

0 X
ðX ð
1

¼ 1 ekτ βeβτ dτ þ 1 ekX βeβτ dτ
0 X
(3.19)
X X !
eβτ eðkþβÞτ βτ 1
kX e
¼β þ 1e β
β ðk þ βÞ 0 β 0 X
ðkþβÞX ðkþβÞX
βe β ðk þ βÞe keðkþβÞX þ β
¼1þ ¼1
kþβ kþβ kþβ
ðX ðθ ð ðX
1
E½I ¼ fW ðτÞðθ τÞdτfL ðθÞdθ þ fW ðτÞðX τÞdτfL ðθÞdθ

0 0 X 0
ðX ð
1
θk þ ekθ 1 ðβθÞ Xk þ ekX 1 βθ
¼ βe dθ þ βe dθ
k k
0 X
0X 1
ð ðX ðX ð
1
β@ βθ ðkþβÞθ βθ A Xk þ ekX 1
¼ θke dθ þ e dθ e dθ þ β eβθ dθ
k k
0 0 0 X
βX
βX k e 1 ð kþβ ÞX βX
β e e 1 e 1
¼ Xk þ
k β β2 ðk þ β Þ β
XkeβX þ eðkþβÞX eβX
þ
k
keðkþβÞX þ β k eβX 1 þ β βeðkþβÞX ðk þ βÞeβX þ k
¼ ¼
k ðk þ β Þ kβ β ðk þ β Þ
(3.20)
Limes calculation ρ ! 1:

βeðμð1ρÞþβÞX ðμð1 ρÞ þ βÞeβX þ μð1 ρÞ
lim
ρ!1 β ð μ ð 1 ρÞ þ β Þ
βeβX βeβX
¼ ¼0 (3.21)
β2
Limes calculation ρ ! 0:

βeðμð1ρÞþβÞX ðμð1 ρÞ þ βÞeβX þ μð1 ρÞ
lim
ρ!0 β ð μ ð 1 ρÞ þ β Þ
βeðμþβÞX ðμ þ βÞeβX þ μ βeðμþβÞX ðμ þ βÞeβX μ
¼ ¼ þ
β ðμ þ β Þ β ðμ þ β Þ β ðμ þ β Þ (3.22)
ðμþβÞX βX ðμþβÞX βX
with e <e ) βe βe <0
ðμþβÞX βX βX
βe βe μe μ μ
) þ <
β ðμ þ β Þ βðμ þ βÞ βðμ þ βÞ
Proof of Proposition 3.2. For the pdf of production lead time being independent of
the WAW X, the service level with WAW policy is lower than without WAW
policy:
Appendix 37
ðX ð
1 ð
1
!
FW ðτÞfL ðτÞdτ þ FW ðXÞfL ðτÞdτ < FW ðτÞfL ðτÞdτ
0 X 0
(3.23)
ð
1 ð
1
!
, FW ðXÞfL ðτÞdτ < FW ðτÞfL ðτÞdτ
X X
which holds since any cdf is an increasing function.

For the pdf of production lead time being independent of the WAW X, the
expected FGI lead time with WAW policy is lower than without WAW policy:
ðX ðθ ð ðX
1
fW ðτÞðθ τÞdτfL ðθÞdθ þ fW ðτÞðX τÞdτfL ðθÞdθ

0 0 X 0
ð ðθ
1
!
< fW ðτÞðθ τÞdτfL ðθÞdθ (3.24)
0 0
ð ðX
1 ð ðθ
1
!
, fW ðτÞðX τÞdτfL ðθÞdθ < fW ðτÞðθ τÞdτfL ðθÞdθ
X 0 X 0
which holds for any production lead time and customer required lead time distribu-
tion since θ X on the right hand side of the last line in inequality (3.24).
For the pdf of production lead time being independent of the WAW X, the
expected tardiness with WAW policy is higher than without WAW policy:
ðX 1
ð ð1
1 ð
fW ðτÞðτ θÞdτfL ðθÞdθ þ fW ðτÞðτ XÞdτfL ðθÞdθ
0 θ X X
ð1
1 ð
!
> fW ðτÞðτ θÞdτfL ðθÞdθ (3.25)
0 θ
ð1
1 ð ð1
1 ð
!
, fW ðτÞðτ XÞdτfL ðθÞdθ > fW ðτÞðτ θÞdτfL ðθÞdθ
X X X θ
which holds for any production lead time and customer required lead time distribu-
tion since θ X on the right hand side of the last line in inequality (3.25).

ρμ eðμð1ρÞþβÞX 1 ρμ eβX 1
E½G ¼
μ ð 1 ρÞ þ β β
ðμð1ρÞþβÞX
dE½G μ e 1 Xρμ e 2 ðμð1ρÞþβÞX ρμ2 eðμð1ρÞþβÞX 1
¼ þ þ
dρ μ ð 1 ρÞ þ β μð1 ρÞ þ β ðμð1 ρÞ þ βÞ2
βX
μ e 1
¼0
β

μeðμð1ρÞþβÞX ð1 þ XρμÞ μ ρμ2 eðμð1ρÞþβÞX 1 μ eβX 1
, þ ¼
μð1 ρÞ þ β ðμð1 ρÞ þ βÞ2 β
βX
ðμð1ρÞþβÞX 2 ðμð1ρÞþβÞX μ e 1
μe ð1 þ XρμÞ ρμ e
, þ ¼
μð1 ρÞ þ β ð μ ð 1 ρÞ þ β Þ 2 β
μ ρμ2
þ þ
μð1 ρÞ þ β ðμð1 ρÞ þ βÞ2
(3.26)
With defining
μeðμð1ρÞþβÞX ð1 þ XρμÞ ρμ2 eðμð1ρÞþβÞX

lhsðρÞ ¼ þ
μð1 ρÞ þ β ðμð1 ρÞ þ βÞ2
βX (3.27)
μ e 1 μ ρμ2
rhsðρÞ ¼ þ þ
β μð1 ρÞ þ β ðμð1 ρÞ þ βÞ2
it follows that Eq. (3.26) has one unique solution which is a maximum since
d d d d
dρ lhsðρÞ > 0, dρ rhsðρÞ > 0, d 2 ρ lhsðρÞ > 0, d2 ρ rhsðρÞ > 0 and lhsðρÞ > rhsðρÞ for
ρ ¼ 0 as well as lhsðρÞ < rhsðρÞ for ρ ¼ 1. For ρ ¼ 0 it follows
ðμþβÞX βX
e 1 e 1
lhsð0Þ > rhsð0Þ , > (3.28)
ðμ þ βÞX βX
which holds since

d ea 1 ea ea 1
>0, > 0 , ea 1 > a (3.29)
da a a a2
and for ρ ¼ 1 it holds:

μeβX ð1 þ XμÞ μ2 eβX 1 μ eβX 1 μ
lhsð1Þ < rhsð1Þ , þ < þ
β β2 β β (3.30)
βX
, e 1 > βX
Appendix 39
which is fulfilled with ex 1 > x.

Derivation of Equations (3.14) and (3.15). Based on Eqs. (3.6) and (3.9) for the
service level in a system without and with WAW policy the following can be stated
provided the service level in both systems is equal:
β
η¼1 , kð1 ηÞ βη ¼ 0
ðk þ β Þ
and
kX eðkX þβÞX þ β
ηX ¼ 1 , kX ð1 ηX Þ kX eðkX þβÞX βηX ¼ 0
kX þ β
with η ¼ ηX (3.31)
) kð1 ηÞ βη ¼ kX ð1 ηÞ kX eðkX þβÞX βη
, ðk kX Þð1 ηÞ ¼ kX eðkX þβÞX
, ρ ¼ ρX þ e
k¼μð1ρÞ 1η
k X ¼ μ ð 1 ρX Þ
The index X indicates the system applying the WAW policy. The utilization loss
can be calculated as:
Δρ ¼ ρ ρX ¼ e (3.32)
1η
Based on Eqs. (3.12) and (3.4), the following can be stated for the FGI reduction:

kμρ ρX μ eðkX þβÞX 1 ρX μ eβX 1
ΔE½G ¼ E½G E½GX ¼ þ
kþβ kX þ β β
βX ðk þβÞX (3.33)
ρX μ e 1 ðρX þ ΔρÞμk ρX μ e X
1
¼ þ
β β ðk þ β Þ kX þ β
Proof of Proposition 3.3. Service level increases with increasing k (decreases with
increasing mean production lead time 1=k):
keðkþβÞX þ β
η¼1
kþβ
ðkþβÞX
dη kXe eðkþβÞX keðkþβÞX þ β ! (3.34)
¼ þ >0
dk kþβ ðk þ β Þ2
!
) k2 XeðkþβÞX þ βkXeðkþβÞX þ β > βeðkþβÞX
which is fulfilled with β > βeðkþβÞX and k2 XeðkþβÞX þ βkXeðkþβÞX > 0.

The expected FGI lead time increases with increasing k (decreases with increas-
ing mean production lead time 1=k):
βeðkþβÞX ðk þ βÞeβX þ k
E½ I ¼
β ðk þ β Þ
dE½ I 1 eβX XβeðkþβÞX k þ βeðkþβÞX ðk þ βÞeβX ! (3.35)
¼ >0
dk β ðk þ β Þ β ðk þ β Þ2
! !
) 1 ðk þ βÞXeðkþβÞX eðkþβÞX > 0 ) eðkþβÞX > 1 þ ðk þ βÞX

The expected tardiness decreases with increasing k (increases with increasing
mean production lead time 1=k):
keðkþβÞX þ β
E½ C ¼
k ðk þ β Þ
dE½C eðkþβÞX kXeðkþβÞX keðkþβÞX þ β keðkþβÞX þ β !
¼ 2 <0
dk k ðk þ β Þ k ðk þ β Þ2 k ðk þ β Þ
ðk þ βÞkeðkþβÞX ðk þ βÞk2 XeðkþβÞX k2 eðkþβÞX þ kβ (3.36)
) 2

k 2 ðk þ β Þ k2 ðk þ βÞ2
ðk þ βÞkeðkþβÞX þ ðk þ βÞβ !
2
<0
k 2 ðk þ βÞ
!
) k3 XeðkþβÞX βk2 XeðkþβÞX k2 eðkþβÞX 2kβ β2 < 0
which is fulfilled with k; X; β > 0.

The expected FGI increases with increasing k (decreases with increasing mean
production lead time 1=k):

λ eðkþβÞX 1 λ eβX 1
E½G ¼
kþβ β
ðkþβÞX
dE½G λ e 1 XλeðkþβÞX !
¼ >0
dk ðk þ β Þ 2 kþβ (3.37)
!
) λ eðkþβÞX 1 ðk þ βÞXλeðkþβÞX > 0
!
) eðkþβÞX > 1 þ ðk þ βÞX

Appendix 41
Service Level, FGI, FGI Lead Time, and Tardiness

with Respect to Expected WIP
E½Y
With k ¼ 1þEμ ½Y and ρ ¼ 1þE ½Y from Eq. (3.1), the Eqs. (3.9, 3.10, 3.11 and 3.12) can
be restated with respect to expected WIP when μ is predefined:

μ
μ 1þE½Y
þβ X
keðkþβÞX þ β 1þE½Y e þβ
η¼1 ¼1 μ
kþβ 1þE½Y þ β
(3.38)
μ
1þE½Y
þβ X
μe þ βð1 þ E½Y Þ
¼1
μ þ βð1 þ E½Y Þ
E½I ¼
βðk þ βÞ

μ
(3.39)
þβ X
βð1 þ E½Y Þe 1þE½Y
ðμ þ βð1 þ E½Y ÞÞeβX þ μ
¼
β ð μ þ β ð 1 þ E½ Y Þ Þ

μ
þβ X
ðkþβÞX 1þE½Y
ke þ β μð1 þ E½Y Þe þ βð1 þ E½Y Þ2
E½C ¼ ¼ (3.40)
k ðk þ β Þ μðμ þ βð1 þ E½Y ÞÞ

ρμ eðkþβÞX 1 ρμ eβX 1
E½G ¼
kþβ β
!
μ (3.41)
1þE½Y
þβ X
E½Y μ e 1
βX
E½Y μ e 1
¼
μ þ βð1 þ E½Y Þ βð1 þ E½Y Þ
Taking the first derivative of service level with respect to expected WIP leads to:
dη
¼
dE½Y
! !!
μ
þβ X
μ X þ ð1 þ E½Y Þβ ð1 þ E½Y Þ e
2 1þE½Y
1 þ μX
μ
1þE½Y
þβ X
2 2
μe
ð1 þ E½Y Þ ðβð1 þ E½Y Þ þ μÞ
(3.42)

μ
1þE½Y
þβ X
which is negative because e > 1.
The first derivative of expected FGI lead time with respect to expected WIP is
negative as well:
dE½I
¼
dE½Y
! !
μ
þβ X
ð 1 þ E½ Y Þ 1 e 1þE½Y
þ Xβ þ E½Y Xβ þ Xμ
μ
1þE½Y
þβ X !
2
μe <0
ð1 þ E½Y Þðμ þ βð1 þ E½Y ÞÞ (3.43)
!
Xμ μ
þβ X !
, ð1 þ E½Y Þ Xβ þ þ 1 e 1þE½Y <0
1 þ E½ Y

μ
1þE½Y
þβ X ! μ
,e 1> þβ X
1 þ E½ Y
which holds since ex 1 > x.

The first derivative of the expected tardiness with respect to expected WIP is
positive:

μ
1þE½Y
þβ X
dE½C ð1 þ E½Y Þ3 β2 þ e μ2 ð1 þ E½Y þ XμÞ
¼
dE½Y ð1 þ E½Y Þμðμ þ βð1 þ E½Y ÞÞ2
(3.44)
μ
1þE½Y
þβ X
2βð1 þ E½Y Þ þ Xμβe
þ >0
ðμ þ βð1 þ E½Y ÞÞ2
Chapter 4
Simultaneous Capacity and Planned
Lead Time Optimization
The literature on simultaneous capacity and inventory investment problems

(see Chap. 2) shows that, especially for the MTS case, there are several contri-
butions available, but for the MTO case, there is still a research gap about the
influence of capacity investment on optimal planned lead time. A simultaneous
approach to capacity and planned lead time setting is presented in this chapter based
on costs for capacity, WIP at each stage, FGI, and backorders. This extends existing
planned lead time approaches by introducing a distributed customer required lead
time rather than applying a single value and a variable production lead time that can
be optimized by investments. Furthermore, it extends existing capacity investment
approaches by including a distributed customer required lead time and by rationing
the work release, and therefore the inventory, with the application of planned lead
times. First the simultaneous optimization problem is stated for a general setting
before detailed results for a single- and a two-stage production system, assuming
M/M/1 queues for the processing steps, are developed. For the two-stage production
system consisting of M/M/1 queues and an exponentially distributed customer
required lead time it is shown that only very little cost improvement can be gained
by including a planned lead time for the upstream processing stage. This extends the
findings of Yano (1987), Buzacott and Shanthikumar (1994), and Karaesmen et al.
(2004) who analyze an MTO system with a constant customer required lead time.
Furthermore, a large cost improvement potential is found when capacity investment
and planned lead times are simultaneously optimized.
The remainder of this chapter based on Altendorfer and Minner (2011) is
structured as follows. The general optimization problem is stated in Sect. 4.2 and
a special case for a single-machine model is discussed in Sect. 4.3. The two-stage
M/M/1 model is analyzed in Sects. 4.4 and 4.5 reports on the results of a numerical
study. Conclusions are provided in Sect. 4.6. Some additional proofs are added in
the Appendix.

44 4 Simultaneous Capacity and Planned Lead Time Optimization

lead time = L C = max(0, W–L)
Customer
Buffer
Buffer Proces - Buffer Proces -
2 sing step 1 sing step
2 1 FGI = G
WIP 2 = Y 2 WIP 1 = Y 1
Production lead time FGI lead time

Production lead time = W 2 = W1 I = max(0, L–W )
Waiting time until release to stage 1

= max(0, X 2 –W 2)
Fig. 4.1 Two-stage production system with planned lead times
4.2 The General Optimization Model
4.2.1 Assumptions and Notation
A two-stage production system as shown in Fig. 4.1 is considered.

Customer orders arrive to the manufacturing system according to a stochastic
process with mean arrival rate λ. Each customer order requests an individual due
date value that determines the non-negative random customer required lead
time L. Let fL ðÞ denote the pdf of the customer required lead time, FL ðÞ the
respective cdf, and 1=β its mean. The customer required lead time cannot be
influenced by the production system and all customer orders are accepted. Cus-
tomer orders are processed consecutively by two stages of the manufacturing
system following a FIFS discipline. Each stage consists of a buffer for waiting
orders and a processing step that requires a random processing time. Wi denotes the
random variable of production lead time at stage i needed for one order from release
at stage i until its completion. fWi ðÞ denotes the pdf of the production lead time after
release to stage i. μi denotes the average number of processed orders per unit of time
at each stage i. In the following it is assumed that this capacity can be determined by
investment decisions and therefore is a decision variable of the model, i.e. the
processing rate μi increases (linearly) with investment. cμ;i denotes the capacity cost
per unit of installed capacity at stage i.
The orders are released into the manufacturing system and to consecutive stages
as follows. The customer orders are released to stage 2 of the production system
whenever their remaining time to the due date is smaller or equal to the overall
planned lead time X ¼ X1 þ X2 . Whenever an order is finished at stage 2 it is
forwarded to the buffer of stage 1. At stage 1 the order is released to be processed
(when machine is available) whenever the remaining lead time is smaller than X1 .
After finishing an order at stage 1 it is stored in the FGI until the due date.
4.2 The General Optimization Model 45
Let Yi denote the random WIP at stage i and cy,i the respective holding cost per
order per unit of time stored at stage i. WIP holding costs are assumed to increase
after each processing step, cy;1 > cy;2. The accounting of WIP at stage 2 starts when
the order is released to buffer 2 and ends when the order is delivered to buffer 1. The
evaluation of WIP at stage 1 starts when an order is delivered to buffer 1 and ends
when it is delivered to the FGI. Let G denote the random FGI and cf the respective
inventory holding cost per order per unit of time. Backorders B are charged a
backorder cost cc per unit of time an order is late.
The minimization of expected costs for WIP at each stage, FGI, backorders, and
capacity leads to the following optimization problem.
E½Y1 cy;1 þ E½Y2 cy;2 þ E½Gcf þ E½ Bcc þ μ1 cμ;1 þ μ2 cμ;2 ! min (4.1)
X1 ;X2 ;μ1 ;μ2
4.2.2 Model Analysis
For the analysis of Eq. (4.1), expressions for WIP, FGI, and backorders are required.
4.2.2.1 Relationship Between FGI Lead Time and Tardiness
In Chap. 3 a general form of the equations of FGI lead time and tardiness is
developed based on the production lead time distribution and the customer required
lead time distribution for a production system applying a WAW work release
policy. Let I denote the expected FGI lead time and C denote the tardiness random
variable of an order. Then, E½G ¼ λE½ I and E½ B ¼ λE½C holds. From the expected
FGI lead time (see Eq. (3.10))
ð ðX
1 ðX ðθ
E½ I ¼ fW ðτÞðX τÞ dτfL ðθÞ dθ þ fW ðτÞðθ τÞ dτfL ðθÞ dθ
X 0 0 0
ðX
¼ FW ðτÞð1 FL ðτÞÞdτ ð4:2Þ
0
and the expected tardiness (see Eq. (3.11))
ð1
1 ð ð
ðX 1
E½C ¼ fW ðτÞðτ XÞ dτfL ðθÞ dθ þ fW ðτÞðτ θÞ dτfL ðθÞdθ
X X 0 θ
ðX
¼ E½W ð1 FW ðτÞÞð1 FL ðτÞÞ dτ ð4:3Þ
0
the following relationship can be derived:
ðX ðX
E½C E½ I ¼ E½W ð1 FW ðτÞÞð1 FL ðτÞÞ dτ FW ðτÞð1 FL ðτÞÞ dτ
0 0
ðX
, E½C ¼ E½ I þ E½W ð1 FL ðτÞÞ dτ ð4:4Þ
0
In a serial production system with planned lead times at each stage, the work
release to the first stage is triggered by the sum of all planned lead times. This sum
is equivalent to the WAW applied in Chap. 3. Based on this transformation,
Eqs. (4.2, 4.3 and 4.4) hold for any serial production system with stage-dependent
planned lead times being independent of order size.
4.2.2.2 WIP, FGI, and Backorders Calculation
The expected WIP at stage 2 is independent of the customer required lead time.
ð
1
E ½ Y2 ¼ λ fW2 ðτ; μ2 Þ τdτ ¼ λE½W2 (4.5)

0
The expected WIP at stage 1 depends on the customer required lead time
distribution since the orders are released for production at the stage 1 whenever
their time to due date is lower than X1 .
ð
1 ð Xð2
1
E ½ Y1 ¼ λ fW1 ðτ; μ1 Þ τdτ þ λ fW2 ðτ; μ2 ÞðX2 τÞfL ðθÞ dτdθ

0 X 0
ð1
ðX θX
þλ fW2 ðτ; μ2 Þðθ X1 τÞfL ðθÞ dτdθ
X1 0
ð2
X
¼ λE½W1 þ λð1 FL ðXÞÞ FW2 ðτ; μ2 Þ dτ

0
ðX θX
ð1
þλ FW2 ðτ; μ2 ÞfL ðθÞdτdθ
X1 0
ð2
X
¼ λE½W1 þ λ FW2 ðτ; μ2 Þð1 FL ðτ þ X1 ÞÞ dτ ð4:6Þ

X¼X1 þX2
0
4.2 The General Optimization Model 47
The expected inventory at stage 1 includes the waiting time until the release to
processing and the production lead time at step 1.
Based on the assumptions above, the expected FGI lead time for the random
variable I ¼ ðL W Þþ can be calculated as follows:
ð
1 ð1
X
E½ I ¼ FW2 ðX2 ; μ2 Þ fW1 ðτ1 ; μ1 ÞðX1 τ1 Þdτ1 fL ðθÞdθ

X 0
ð ðX Xτ
1 ð2
þ fW2 ðτ2 ; μ2 Þ fW1 ðτ1 ; μ1 ÞðX τ2 τ1 Þ dτ2 dτ1 fL ðθÞdθ
X X2 0
ðX ð1
X
þ FW 2 ð θ X1 ; μ 2 Þ fW1 ðτ1 ; μ1 ÞðX1 τ1 Þ dτ1 fL ðθÞdθ

X1 0
ðX ðθ ð2
θτ
þ fW2 ðτ2 ; μ2 Þ fW1 ðτ1 ; μ1 Þðθ τ2 τ1 Þ dτ2 dτ1 fL ðθÞ dθ

X1 θX1 0
ð1 ðθ θτ
X ð2
þ fW2 ðτ2 ; μ2 Þ fW1 ðτ1 ; μ1 Þðθ τ2 τ1 Þ dτ2 dτ1 fL ðθÞ dθ ð4:7Þ
0 0 0
The first two lines of Eq. (4.7) cover the case where the customer required lead
time is longer than the overall planned lead time. In the first line, stage 2 finishes on
time and in the second line, stage 2 finishes late but the production lead time at stage
2 is still below the overall planned lead time. Lines three and four cover the case
where X1 L X. Again, the third line covers the case when stage 2 is on time
(with the remaining planned lead time at this stage) and the fourth line covers the
case where stage 2 is late concerning the remaining planned lead time at this stage.
Finally, the last line covers the case where the customer required lead time is
already smaller than the planned lead time at stage 1 which implies that the order
is released to both stages immediately.
From Eq. (4.7) the expected tardiness for the random variable C ¼ ðW LÞþ
can be expressed as:
ðX
E½C ¼ E½ I þ E½W ð1 FL ðτÞÞdτ
0
ð1
X
¼ ð1 FL ðXÞÞ FW2 ðτ þ X2 ; μ2 ÞFW1 ðX1 τ; μ1 Þdτ

λE½W ¼E½Y1 þE½Y2
0
ð1
X ð2
X
þ FW1 ðτ; μ1 Þdτ fW2 ðθ; μ2 Þ fL ðθ þ X1 Þ dθ

0 0
ðX ðθ ð2
θτ
þ fW2 ðτ2 ; μ2 Þ fW1 ðτ1 ; μ1 Þðθ τ2 τ1 Þ dτ2 dτ1 fL ðθÞ dθ

X1 θX1 0
ð1 ðθ θτ
X ð2
þ fW2 ðτ2 ; μ2 Þ fW1 ðτ1 ; μ1 Þðθ τ2 τ1 Þ dτ2 dτ1 fL ðθÞdθ
0 0 0
ð2
X ðX
þ E½W1 þ E½W2 þ FW2 ðτ; μ2 Þð1 FL ðτ þ X1 ÞÞ dτ ð1 FL ðτÞÞ dτ
0 0
(4.8)
Due to the additional complexity added by the integration of a distribution for

customer required lead time and the pdf of production lead time at both stages being
dependent on capacity invested, this problem in its general form cannot be solved
analytically. Nevertheless, by applying numerical optimization methods, the prob-
lem statement can be used to identify the optimal parameters for a production
system.
4.2.3 Multi-Stage Extension
The two-stage model presented above can be extended to a multi-stage model with
n stages by defining:
þ þ
þ þ
Δi ¼ ðXLÞ þ Wn Xn þWn1 Xn1 þ ...þWi1 Xi1 þWi (4.9)
4.3 Simultaneous Optimization of a Single-Stage Production System 49
The WIP per stage, FGI lead time, and tardiness are defined by:
E½Yi ¼ λE½Xiþ1 Δiþ1 þ þ λE½Wi

E½ I ¼ E½ X1 Δ 1 þ
E½C ¼ E½Δ1 X1 þ ð4:10Þ
Using Eq. (4.10) the optimization problem as defined in Eq. (4.1) can be
extended to:
X
n X
n
E½Yi cy;i þ E½Gcf þ E½ Bcc þ μi cμi ! min (4.11)
fXi g;fμi g
i¼1 i¼1
4.3 Simultaneous Optimization of a Single-Stage

Production System
In this section the simplified case of a single-stage production system with an

upfront buffer and a planned lead time is analyzed. This system is studied for a
general distribution of customer required lead time and an M/M/1 queue. In this
case, Eq. (4.1) reduces to
E½Y cy þ E½Gcf þ E½ Bcc þ μcμ ! min (4.12)

X;μ

is Predefined
For a single-stage production system with given capacity, the only parameter
to optimize is the planned lead time X. The relevant costs in this system are
the backorder and FGI costs as the production lead time cannot be influenced
(see also Buzacott and Shanthikumar (1994) that WIP costs have no influ-
ence on the optimal solution). The following optimization problem can be
stated:
CðXÞ ¼ cf E½G þ cc E½ B ! min (4.13)

X
Proposition 4.1. The optimal planned lead time X for minimizing the FGI and
backorder costs for a single-stage production system with distributed customer
required lead time and predefined processing rate is independent of the distribution
of customer required lead time and given by

cc
X ¼ F1
W (4.14)
cf þ cc
Proof. The first-order optimality condition for planned lead time X is:
dCðXÞ
¼ λcf FW ðXÞð1 FL ðXÞÞ λcc ð1 FW ðXÞÞð1 FL ðXÞÞ ¼ 0 (4.15)
dX
The second derivative is positive for any X satisfying Eq. (4.15) and FW ðÞ is a
cdf such that Eq. (4.14) yields a unique minimum:
CðXÞ
¼ λcf ðFW ðXÞfL ðXÞ þ fW ðXÞð1 FL ðXÞÞÞ
d2 X
!
λcc ðð1 FW ðXÞÞð fL ðXÞÞ ð fW ðXÞÞð1 FL ðXÞÞÞ > 0
cc
, f W ð X Þ ð 1 FL ð X Þ Þ þ c þ c f L ð X Þ
1 cc f c
X ¼FW cf þcc

cc !
FW F1
W fL ðX Þ > 0
cf þ cc
, fW ðX Þð1 FL ðX ÞÞ > 0 ð4:16Þ
An important managerial implication of Proposition 4.1 is that the optimal

planned lead time for a production system without the opportunity to invest in
new capacity can be determined based on backorder and FGI costs without knowing
the distribution of customer required lead time. It further follows that X ¼ 0 if
cc ! 0 or cf ! 1 and that X ! 1 if cf ! 0 or cc ! 1.
Following Proposition 3.1, the implementation of a planned lead time value (WAW)
in front of the production system modeled as M/M/1 queue still leads to a Poisson input
stream into the production stage such that all properties of an M/M/1 queue remain
valid. Figure 4.2 illustrates the costs for an M/M/1 production system with exponential
customer required lead time for three different mean customer required lead time
values. This comparison shows that different customer required lead time values result
in different curve shapes, but the optimal value for X is always the same.
4.3 Simultaneous Optimization of a Single-Stage Production System 51
Fig. 4.2 Planned lead time 350 μ = 0.9

costs in single-stage M/M/1
300 λ = 0.8
model with predefined
capacity 250 cf = 10
200 cc = 30
Costs
150
100
50
0
0 20 40 60 80 100
Planned lead time
β=0.005 β=0.025 β=0.05 Xopt
4.3.2 Simultaneous Capacity and Planned Lead Time Setting
In addition to FGI and backorder costs, here WIP and capacity costs can be
influenced by investing into capacity. This further extends Buzacott and
Shanthikumar (1994) to the capacitated case. The optimization problem is:
CðX; μÞ ¼ cy E½Y þ cμ μ þ cf E½G þ cc E½ B ! min (4.17)

X;μ
Proposition 4.2. The optimal processing rate μ and the optimal planned lead time
X for minimizing the capacity, WIP, FGI, and backorder costs for an M/M/1
production system with distributed customer required lead time have to fulfill the
first-order conditions (4.18).
cc
FW ðXÞ ¼
cf þ cc
ðX (4.18)
cμ ðμλÞτ cy þ cc
þ cf þ cc τe ð1 FL ðτÞÞdτ ¼
λ ðμ λÞ2
0
Proof. The first derivative with respect to X is provided in Eq. (4.15) and the first
derivative with respect to μ is:
ðX
λ
CðX; μÞ ¼ cy þ cc þ cμ μ þ cf þ cc λ 1 eðμλÞτ ð1 FL ðτÞÞdτ
μλ
0
ðX
λcc ð1 FL ðτÞÞdτ
0
ðX
dCðX; μÞ cy þ cc λ
¼ cμ þ cf þ cc λ τeðμλÞτ ð1 FL ðτÞÞdτ ¼ 0 ð4:19Þ
dμ ðμ λÞ2
0
12 1
Optimal planned lead time
10
0.9
Optimal utilization
8
0.8
6
0.7
4
0.6
2
0 0.5
0 10 20 30 40 50 0 10 20 30 40 50
Mean customer required lead time Mean customer required lead time
Example 1 Example 2 Example 1 Example 2
λ = 0.8; c f = 10; c c = 30; c y = 5.5; cμ ( Example 1 ) = 100; cμ ( Example 2 ) = 1000;
Fig. 4.3 Optimal parameters for single-stage M/M/1 model with backorder costs
Contrary to the result of Proposition 4.1, the customer required lead time
distribution is relevant for the optimal decision if the capacity can be optimized, too.
Proposition 4.3. The optimal processing rate μ and the optimal planned lead time
X for minimizing the capacity, WIP, FGI, and backorder costs for an M/M/1
production system with exponentially distributed customer required lead time are
defined by Eq. (4.20). The optimal processing rate decreases in mean customer
required lead time and the optimal planned lead time increases in mean customer
required lead time.

cμ cf þ cc 1 eðμ þβλÞX ðμ þβλÞX cy þ cc
þ e X ¼
λ μ þ β λ ðμ þ β λÞ ðμ λÞ2

1 cf þ cc
X ¼ ln ð4:20Þ
μ λ cf
Proof see Appendix.

Figure 4.3 illustrates the optimal utilization and the optimal planned lead time
versus the mean customer required lead time for two examples with different
capacity costs.
The comparison of these two examples in Fig. 4.3 shows that the capacity costs
have a high influence on the optimal utilization, which is an intuitive result, but also
on the optimal planned lead time. This high influence of capacity costs and capacity
investment itself on the optimal planned lead time already shows that the simulta-
neous treatment of these two optimization problems, which are usually treated
separately, leads to a significant difference in optimal parameters. Furthermore,
Fig. 4.3 shows that the influence of the mean customer required lead time cannot be
neglected.
4.4 Simultaneous Optimization of a Two-Stage Production System 53
4.4 Simultaneous Optimization of a Two-Stage Production

System
In this section the general model stated in Sect. 4.2 is solved for two processing
steps with exponentially distributed processing times and interarrival times as well
as exponentially distributed customer required lead times.
4.4.1 Optimal Planned Lead Time when Capacity is

Predefined
In addition to the backorder and FGI costs from the single-stage model, the WIP
holding costs after stage 2 have to be added, which occur when orders are finished
with a production lead time at stage 2 being smaller than the planned lead time X2
(see also Yano (1987) for this formulation). Equation (4.1) simplifies to:
CðX1 ; X2 Þ ¼ E½Y1 cy;1 þ E½Gcf þ E½ Bcc ! min

X1 ;X2
ðX

, cy;1 þ cc E½Y1 þ λ cf þ cc E½ I λcc ð1 FL ðτÞÞdτ ! min ð4:21Þ
ð4:4Þ X1 ;X2
0
Using a Lagrangian approach ( χ 1 and χ 2 are the Lagrange multipliers) and taking
the first derivative of the cost function leads to the first-order optimality conditions:
ð2
X
dCðX1 ; X2 Þ
¼ 0 , cf þ cc FW1 ðX1 Þ cy;1 þ cc FW2 ðτÞfL ðτ þ X1 Þdτ
dX1
0
X1 ð
þX2

þ c f þ c c ð 1 FL ð X1 þ X 2 Þ Þ FW2 ðτÞfW1 ðX1 þ X2 τÞdτ
X2
cc ð1 FL ðX1 þ X2 ÞÞ χ 1 =λ ¼ 0 ð4:22Þ
dCðX1 ; X2 Þ
¼0
dX2

, cy;1 þ cc cf þ cc FW1 ðX1 Þ ð1 FL ðX1 þ X2 ÞÞFW2 ðX2 Þ
X1 ð
þX2

þ c f þ c c ð 1 FL ð X1 þ X2 Þ Þ FW2 ðτÞfW1 ðX1 þ X2 τÞdτ
X2
cc ð1 FL ðX1 þ X2 ÞÞ χ 2 =λ ¼ 0 ð4:23Þ

An analysis of the Lagrange multipliers shows that X1 > 0 and that a boundary
solution X2 ¼ 0 can exist. The optimality conditions (4.22) and (4.23) have the
same second and third term (for X1 > 0 and X2 > 0 and for this reason χ 1 ¼ 0 and
χ 2 ¼ 0) which means the first two terms have to be equal in an optimal solution.
This leads to the following relative optimality condition:
0X
ð2

cy;1 þ cc cf þ cc FW1 ðX1 Þ @ FW2 ðτÞfL ðτ þ X1 Þdτ
0
!
þð1 FL ðXÞÞFW2 ðX2 Þ ¼0 ð4:24Þ
Based on the results found in the single-stage optimization problem as well as

the results of Yano (1987), it is conjectured that the first-order optimality conditions
are sufficient to identify the optimum of problem (4.21).
Proposition 4.4. The optimal planned lead times X1 and X2 minimizing WIP, FGI,
and backorder costs for a two-stage production system with predefined capacity
and distributed customer required lead time are independent of the distribution of
customer required lead time and implicitly defined by Eqs. (4.25) and (4.26).
Case X2 ¼ 0 :
ð1
X
cc
X1 : FW2 ðτÞfW1 ðX1 τÞdτ ¼ ð4:25Þ
cf þ cc
0
Case X2 > 0 :

cy;1 þ cc cy;1 þ cc
X1 : ¼ FW1 ðX1 Þ , X1 ¼ F1W1
cf þ cc cf þ cc
X1 ð
þX2
cc
X2 : FW2 ðτÞfW1 ðX1 þ X2 τÞdτ ¼ ð4:26Þ
cf þ cc
X2
Even though the optimality conditions found in Eqs. (4.22) and (4.23) are
different from the ones found in Yano (1987) for the case without distributed
customer required lead times, the result presented in Proposition 4.4 is equal to the
result found in Yano (1987). One managerial implication of Proposition 4.4 is that
the distribution of customer required lead time has in the case of predefined capacity
no influence on the optimal parameterization of the planned lead times. Comparing
the solution where X2 ¼ 0 with the single-stage production system shows that
Eq. (4.25) is equivalent to Eq. (4.14) since the left hand side of Eq. (4.25) is only
the cdf of the convolution of the two production lead times at stage 1 and stage 2.
4.4.2 Simultaneous Capacity and Planned Lead Time Setting
The special case of exponentially distributed customer required lead time values
and a series of two M/M/1 queues is treated in this section. That the Poisson input
into stage 2 remains valid is directly shown in Proposition 3.1. Since the output of
stage 2 is a Poisson stream and the planned lead time application between the stages
is a WAW work release policy with arbitrary distribution, Proposition 3.1 holds and
the input into stage 1 is Poisson as well. The following expected values for WIP at
stage 1 and 2 can be calculated:
λ
E ½ Y2 ¼
μ2 λ
βX1
λ e eβX eβX1 eðμ2 λÞX2 βX
E½Y1 ¼ þλ ð4:27Þ
μ1 λ β ðμ 2 λ þ β Þ
The FGI lead time is:
eðμ2 λþβÞX2 ðμ2 λþβÞX1 ðμ1 λÞ eðμ2 λþβÞX2 ðμ1 λþβÞX1 ðμ2 λÞ
E½ I ¼
ðμ1 μ2 Þðμ2 λ þ βÞ ðμ1 μ2 Þðμ2 λ þ βÞ
eðμ2 λþβÞX2 βX1 eðμ1 λþβÞX1 ðμ2 λÞ
þ þ
ðμ2 λ þ βÞ ðμ 1 λ þ β Þðμ 2 λ þ β Þ
eβX1 ðμ2 λÞ ðμ1 λÞðμ2 λÞ
þ ð4:28Þ
β ðμ 2 λ þ β Þ β ðμ 1 λ þ β Þðμ 2 λ þ β Þ
Derivation see Appendix. Applying Eq. (4.4) leads to the following expected
tardiness:
ðX
0
¼
ðμ 1 μ 2 Þðμ 2 λ þ β Þ ðμ1 μ2 Þðμ2 λ þ βÞ
eðμ1 λþβÞX1 ðμ2 λÞ eβX1 ðμ2 λÞ ðμ1 λÞðμ2 λÞ
þ þ
ðμ1 λ þ βÞðμ2 λ þ βÞ βðμ2 λ þ βÞ βðμ1 λ þ βÞðμ2 λ þ βÞ
eðμ2 λþβÞX2 βX1 1 1 eβX1 eðμ2 λÞX2 βX 1 eβX1
þ þ þ ð4:29Þ
ðμ2 λ þ βÞ μ2 λ μ1 λ ðμ2 λ þ βÞ β β
Figure 4.4 shows the influence of the parameters X1 and X2 on the expected FGI
lead time and tardiness (applying Eqs. (4.28) and (4.29)). A stronger influence of X1
a b
14 14
FGI lead time / tardiness
FGI lead time / tardiness

12 12
10 10
8 8
6 6
4 4
2 2
0 0
0 5 10 15 20 25 0 5 10 15 20 25
Planned lead time X1 Planned lead time X2
Tardiness FGI lead time Tardiness FGI lead time
λ = 0.8; μ 1 = 1.1; μ 2 = 1.0 ; X1 (Figure 4.4b) = 5; X2 (Figure 4.4a) = 5;
Fig. 4.4 Influence of X1 and X2 on the expected FGI lead time and tardiness in two-stage model
on expected FGI lead time and tardiness is indicated in Fig. 4.4 and the intuitive
result that tardiness decreases and FGI lead time increases with increasing planned
lead time is depicted.
The influence of X1 and X2 on the optimal solution will be discussed more in
detail in Sect. 4.5. Applying Eqs. (4.28) and (4.29), the optimization problem from
Eq. (4.1) for non-identical processing rates becomes:

Cðμ1 ; μ2 ; X1 ; X2 Þ ¼ cy;1 þ cc E½Y1 þ cy;2 þ cc E½Y2 þ cf þ cc λE½ I
ðX
cc λ ð1 FL ðτÞÞdτ þ cμ1 μ1 þ cμ2 μ2 ! min
X1 ;X2 ;μ1 ;μ2
0
βX1
λ e eβX eβX1 eðμ2 λÞX2 βX
, cy;1 þ cc þλ
μ1 λ β ðμ 2 λ þ β Þ

þ cf þ cc λ
eðμ2 λþβÞX2 βX1 eðμ1 λþβÞX1 ðμ2 λÞ eβX1 ðμ2 λÞ
þ
ðμ2 λ þ βÞ ðμ1 λ þ βÞðμ2 λ þ βÞ βðμ2 λ þ βÞ
ðμ1 λÞðμ2 λÞ λ 1 eβX
þ þ cy;2 þ cc cc λ
βðμ1 λ þ βÞðμ2 λ þ βÞ μ2 λ β
þ c μ1 μ 1 þ c μ2 μ 2 ! min ð4:30Þ
X1 ;X2 ;μ1 ;μ2
The explicit optimality conditions for X1 and X2 for this problem are provided in
the Appendix.
a 500
b 500
400 400
300 300
Costs
Costs
200 200
100 100
0 0
0 20 40 60 80 100 0.4 0.5 0.6 0.7 0.8 0.9 1
Planned lead time X1 Utilization
β=0.005 β=0.025 β=0.05 β=0.005 β=0.025 β=0.05
λ = 0.8; cf = 10; cc = 30; cy,1 = 6; cy,2 = 5; cμ 1 = cμ 2 = 50 ;
Fig. 4.5 Cost with respect to planned lead time and utilization in two-stage M/M/1 model
For equal processing rates at both stages it holds:

βX1
λ e eβX eβX1 eðμλÞX2 βX
Cðμ; X1 ; X2 Þ ¼ cy;1 þ cc þλ
μλ β ðμ λ þ β Þ

eðμλþβÞX þ eðμλþβÞX ðμ λÞX1 eðμλþβÞX2 βX1
þ cf þ cc λ
ðμ λ þ β Þ ðμ λ þ β Þ
#
e ðμλþβÞX1
ðμ λÞ e βX1
ðμ λ Þ ðμ λ Þ 2 λ
þ þ þ cy;2 þ cc
ðμ λ þ β Þ 2 β ðμ λ þ β Þ β ðμ λ þ β Þ 2 μλ
1 eβX
cc λ þ 2cμ μ ! min ð4:31Þ
β X1 ;X2 ;μ
Figure 4.5a shows an example of the cost curve as a function of the planned lead
time X1 when processing rates and X2 are optimized. It indicates that the mean
customer required lead time has a strong influence on the costs when planned lead
time X1 is predefined. The shape of the cost curves in Fig. 4.5a is similar to the
single stage setting depicted in Fig. 4.2. Figure 4.5b shows the costs with respect to
the utilization when X1 and X2 are optimized. In this example μ1 ¼ μ2 ¼ μ is
assumed and therefore problem (4.31) is numerically solved. The example depicted
in Fig. 4.5b implies that the optimal utilization strongly depends on the mean
customer required lead time whereas its influence on the optimal costs is lower.
Figure 4.6a, b show the optimal X and X2 values as functions of utilization for
equal processing rates at both stages (problem (4.31) is numerically solved). The
global optimum for a low and a high capacity cost example are shown, too. Based
on the optimality conditions stated in Eqs. (4.25) and (4.26) it is clear that the
optimal parameters for X1 and X2 do not depend on the capacity costs and therefore
a b
20
Optimal overall planned lead time when
140
Optimal planned lead time X2 when

18
120
16
capacity is predefined
capacity is predefined
100 14
12
80
10
60
8
40 6
4
20
2
0 0
0.4 0.5 0.6 0.7 0.8 0.9 1 0.4 0.5 0.6 0.7 0.8 0.9 1
Utilization Utilization
X Opt Ex 1 Opt Ex 2 X2 Opt Ex 1 Opt Ex 2
λ = 0.8; β = 0.005; cf =10; cc =30; cy,1=6; cy,2=5;

cμ (Example 1) = 50; cμ(Example 2) = 500;
Fig. 4.6 Optimal planned lead time with respect to utilization in two-stage M/M/1 model
18 1
16
Optimal Planned lead time X
14 0.9
Optimal utilization
12
0.8
10
8
0.7
6
4 0.6
2
0 0.5
0 10 20 30 40 50 0 10 20 30 40 50
Mean customer required lead time Mean customer required lead time
Example 1 Example 2 Example 1 Example 2
λ = 0.8; cf =10; cc =30; cy,1=6; cy,2=5; cμ (Example 1) = 50; cμ (Example 2) = 500;
Fig. 4.7 Optimal parameters for two-stage M/M/1 model with backorder costs
Example 1 and Example 2 lead to the same curves with respect to the utilization.
Figure 4.6 shows that the optimal values for X1 and X2 are increasing in the
utilization and that in this case, the functions of both optimal values have the same
shape. Nevertheless, there exists a large set of optimal solutions where X2 ¼ 0 as
shown in the next section. This shape of the X curves shows that, whenever
capacity is assumed to be constant, the deviation of X ðρÞ to Xglobal

can be very high.
Figure 4.7 illustrates the optimal parameters for overall planned lead time and
average utilization ρ over both stages for the two examples also discussed in
Sect. 4.3. In this figure, problem (4.30) is numerically solved. Comparing Fig. 4.7 to
Fig. 4.3 shows that the curve shapes for optimal planned lead time and optimal
utilization are similar for the single and the two stage setting.
This section shows, that the distribution of customer required lead time, which is
in past publications implicitly assumed to have only values larger than the sum of
the planned lead times, has no influence on the optimal solution for the planned lead
time setting if capacity is predefined. Furthermore, it is shown that the customer
required lead time has an influence on the optimal capacity to invest. Since the
optimal capacity invested also influences the optimal planned lead time setting, it is
shown that the influence of the customer required lead time distribution on optimal
planned lead time is mediated through the capacity invested. This finding
emphasizes that the distribution of customer required lead time is an important
factor which should be considered when optimizing production planning
parameters including the capacity invested.
4.5 Numerical Study
This numerical study has the following objectives:

• Determine the cost effect of neglecting the planned lead time on stage 2 (X2 ¼ 0).
• Identify the cost reduction potential of simultaneously solving the capacity
investment and the planned lead time decision in comparison to firstly decide
for the capacity and secondly decide for the planned lead times.
• Evaluate the influence of mean customer required lead time and WIP, FGI, as
well as backorder costs on this cost reduction potential when simultaneously
optimizing capacity investment and planned lead time.
Following the literature, the queuing design question on how much capacity to
invest can be analyzed either with or without including backorder and FGI costs.
Therefore, the optimization problem sequentially solving the capacity and planned
lead time decision is evaluated for both situations with and without including
backorder and FGI costs in the first step. Based on the objectives of this numerical
study, the following four optimization problems (all based on problem (4.30)) are
solved for each test instance:
• For the simultaneous optimization of capacity and planned lead times, problem
(4.30) is solved. This is the basic scenario with the optimal solution.
• To identify the influence of X2 ¼ 0 on the optimal result, problem (4.30) is
solved for X2 ¼ 0.
• In the case of sequential optimization of capacity and planned lead time includ-
ing FGI and backorder costs in the capacity optimization step, problem (4.30)
is in the first step solved for μ1 and μ2 with X2 ¼ 0 and X1 ¼ 1. In the second
step, problem (4.30) is then solved for X1 and X2 with applying μ1 and μ2 from step
one. In the first step this corresponds to a situation with unconstrained work
release to the production system and the results of this optimization problem are
therefore referenced as unconstrained work release sequential optimization in
the remainder.
• In the case of sequential optimization of capacity and planned lead time exclud-
ing FGI and backorder costs in the capacity optimization step, problem (4.30) is
in the first step reduced to a capacity and WIP cost model (cf ¼ 0 and cc ¼ 0) and
solved for μ1 and μ2. In the second step, problem (4.30) is then again solved for X1
and X2 with applying μ1 and μ2 from step one. In the first step this corresponds to
the typical queuing design model only including WIP costs for the capacity
investment decision and the results of this optimization problem are therefore
referenced as WIP cost based sequential optimization in the remainder.
The test instances are defined as all combinations of the following parameter
sets, which leads to more than 70,000 test instances:
• 1=β 2 f7:5; 15; 30; 45; 60; 75; 90; 105; 120; 135; 150g;
• λ ¼ 0:8;
• cμ;1 ¼ cμ;2 ¼ cμ 2 f100 þ 90k; k ¼ 0; . . . ; 10g;
• cc 2 f10 þ 2k; k ¼ 0; . . . ; 10g;
• cy;1 2 f1 þ k; k ¼ 0; . . . ; 9g;
• cy;2 2 f1 þ k; k ¼ 0; . . . ; 9g cy;1 :
The cost parameters are chosen to reflect a broad range of cost structures and the
β values are based on the unconstrained expected production lead time of a
two-stage M/M/1 system with 75 % utilization E½W jX2 ¼ 0; X1 ¼ 1; ρ ¼ 0:75
¼ 7:5. They account for a mean customer required lead time between 1 and 20 times
that value, which seems to be a feasible region for possible MTO situations.
The results concerning the influence of X2 on the optimal solution show, that
neglecting X2 leads on average to a cost penalty of 0.08 % and X2 ¼ 0 is optimal for
51 % of all problem instances. Within 0.1 % of the optimal solution are 83 % of the
instances with X2 ¼ 0 and 97 % are within 0.5 %. The maximum cost increase
between the optimal costs with X2 > 0 and X2 ¼ 0 is 4 % which occurs in an instance
where a high value increase is generated at production stage 1. The result supports the
common practice in production control systems like CONWIP or DBR which use
only single planned lead time values for the work release into the production system.
The simultaneous optimization leads to an average cost reduction of 6.7 % in
comparison to the unconstrained work release sequential optimization and of 7.7 %
in comparison to the WIP cost based sequential optimization. A detailed analysis of
the numerical results shows that the cost reduction potential due to simultaneously
optimizing the capacity and planned lead time decision is very dependent on the
parameter settings. Therefore, the following Fig. 4.8 shows the influence of the cost
parameters and the mean customer required lead time on the cost reduction poten-
tial. All values shown in Fig. 4.8 are the average of all test instances with the
respective parameter. Looking at Fig. 4.8b, for example, shows that simultaneously
optimizing capacity and planned lead times leads for capacity costs of 100 to an
average cost reduction potential of 10 % in comparison to the WIP cost based
sequential optimization and 23 % in comparison to the unconstrained work release
sequential optimization.
a b
25 25
simultaneuosly optimized
% Cost reduction when
20 20
15 15
10 10
5 5
0 0
0 50 100 150 100 250 400 550 700 850 1000
Mean customer required lead time 1/b Capacity costs cµ
Unconstrained work release sequential optimization Unconstrained work release sequential optimization
WIP cost based sequential optimization WIP cost based sequential optimization
c 25
d 30

20 25
20
15
15
10
10
5 5
0 0
10 15 20 25 30 0 2 4 6 8 10
Backorder costs cc WIP costs cy1
Unconstrained work release sequential optimization Unconstrained work release sequential optimization
WIP cost based sequential optimization WIP cost based sequential optimization
Fig. 4.8 Cost reduction in two-stage model with simultaneous optimization problem
An interesting finding from Fig. 4.8 is that the influence the parameters have on
the cost reduction potential strongly depends on the sequential optimization model
applied. Compared to the unconstrained work release sequential optimization, the
cost reduction potential of simultaneous optimization increases with respect to the
mean customer required lead time, while it decreases compared to the WIP cost
based sequential optimization. The contrary is true for backorder costs. Increasing
WIP and capacity costs both lead to a decreasing cost reduction potential for the
simultaneous approach (the influence of cy;2 is similar to the one of cy;1 ).
The results presented in Fig. 4.8 show that independent of the respective cost
parameters, the simultaneous optimization always leads to a considerable cost reduc-
tion in comparison to the sequential optimization approaches. The low cost reduction
potential in comparison to the unconstrained work release sequential optimization for
low mean customer required lead time values comes from the very high optimal
planned lead time values in this situation which lead to similar results as X1 ¼ 1.
For practical application, this rather high cost reduction potential means that the
planned lead time decision should be included into the optimization problem to
decrease overall costs. This even holds when the result for planned lead times is not
of interest at the capacity investment point in time.
The simultaneous optimization of planned lead time and capacity invested is

discussed for a single-stage and a two-stage manufacturing system with random
demands and random due dates. It is shown that the optimal planned lead time
values are independent of the distribution of customer required lead time when
capacity is predefined. This extends the findings of Buzacott and Shanthikumar
(1994) and Yano (1987) discussing the planned lead time setting for single and
two-stage systems respectively without the consideration of distributed due dates.
Nevertheless, it is shown that the distribution of customer required lead time has a
strong influence on the optimal capacity invested.
Additionally, a numerical study shows that in the simultaneous optimization
problem the planned lead time at stage 2 can be neglected in a large number of
cases. This finding supports the implementation of production planning strategies
like CONWIP or DBR which only have one parameter for work release. In the
numerical examples an average cost reduction potential of 6.7 % and 7.7 % is found
when simultaneously optimizing capacity investment and planned lead times in
comparison to sequentially optimizing these values with and without considering
FGI and backorder costs in the first step respectively. A managerial implication of
this finding is that the simultaneous treatment of both problems can lead to signifi-
cant cost reductions.
Appendix
Proof of Proposition 4.3. Based on Eq. (4.18) the following first-order optimality
conditions can be stated:

dCðX; μÞ cc 1 cf þ cc
¼ 0 , FW ðXÞ ¼ ,X ¼ ln
dX cf þ cc μλ cf
ð
X
dCðX; μÞ cμ ðμλÞτ cy þ cc
¼ 0 , þ cf þ cc τe ð1 FL ðτÞÞdτ ¼
dμ λ ðμ λÞ2
0
ðX
cμ cy þ cc
, þ cf þ cc τeðμλÞτ eβτ dτ ¼
λ ðμ λ Þ2
0

cμ cf þ cc 1 eðμþβλÞX cy þ cc
, þ eðμþβλÞX X ¼ ð4:32Þ
λ ðμ þ β λÞ ðμ þ β λÞ ðμ λÞ2
Appendix 63
The second derivative of the cost function with respect to the processing rate μ is
positive. Therefore, the cost function is convex and the first-order condition defines
a unique minimum:

dCðX; μÞ 2λ cy þ cc
¼
d2 μ ðμ λ Þ3

ðμλþβÞX
2 þ e 2 þ 2Xðμ λ þ βÞ þ X2 ðμ λ þ βÞ2 !
cc þ cf λ 3
>0
ðμ λ þ β Þ

2 cy þ cc
,
cc þ cf
ðμ λÞ3 ðμλþβÞX

2
!
2 þ e 2 þ 2X ð μ λ þ β Þ þ X 2
ðμ λ þ β Þ >0
ðμ λ þ β Þ3

2 cy þ cc
with > 1 from cc > cf
cc þ cf
ðμ λÞ3 !
, 3
2 þ eðμλþβÞX 2 þ 2Xðμ λ þ βÞ þ X2 ðμ λ þ βÞ2 < 1
ðμ λ þ β Þ
ðμ λ þ β Þ3
with >1
ðμ λÞ3

ðμλþβÞX 1 2 2 ! 3
,e 1 þ X ðμ λ þ β Þ þ X ðμ λ þ β Þ <
2 2
1 ! 1
, eðμλþβÞX þ eðμλþβÞX > 1 þ Xðμ λ þ βÞ þ X2 ðμ λ þ βÞ2 ð4:33Þ
2 2
which is fulfilled with:
ea 1 > a ) eðμλþβÞX > 1 þ Xðμ λ þ βÞ

and ea > a2 ) eðμλþβÞX > ðXðμ λ þ βÞÞ2 ð4:34Þ
The first derivative of X with respect to β is obviously negative, so X increases

in mean customer required lead time 1=β.
Implicit differentiation of the optimality condition (4.20) leads to:
Xð
cy þ cc λ
Φ ðμ ; μ ðβ ÞÞ ¼ c μ 2
þ cf þ cc λ τeðμ λþβÞτ
dτ ¼ 0
ðμ λÞ
0

dΦðμ ðβÞÞ 2 cy þ cc λ dX
¼ þ cf þ cc λ X eðμ λþβÞX
dμ ðβÞ ðμ λ Þ3 dμ ðβÞ
ð
X

cf þ cc λ τ2 eðμ λþβÞτ
dτ
0
ð
X
dΦðμ ðβÞÞ
¼ cf þ cc λ τ 2 e ð μ λþβÞτ
dτ
dβ
0
dΦðμ ðβÞÞ dμ ðβÞ dΦðμ ðβÞÞ

with þ ¼0
dμ ðβÞ dβ dβ

2 cy þ cc dμ ðβÞ dX ðμ λþβÞX dμ ðβÞ
) þ c f þ c c X e
ðμ λÞ3 dβ dμ ðβÞ dβ
ð
X
2 ðμ λþβÞτ dμ ðβÞ
cf þ cc τ e dτ 1 þ ¼0 (4.35)
dβ
0

dX 1 cf þ cc dX X
with ¼ ln , ¼
dμ ðβÞ ðμ λÞ2 cc dμ ðβÞ ðμ λÞ
XÐ 2 ðμ λþβÞτ
c f þ c c τ e dτ
dμ ðβÞ 0
) ¼ X (4.36)
dβ 2ðcy þcc Þ Ð λþβÞτ ðX Þ2 eðμ λþβÞX
cf þ cc τ e
2 ð μ dτ þ
ðμ λÞ3 ðμ λÞ
0
which is positive if the denominator is positive:
0X 1
ð
2 cy þ cc !
2
ðX Þ
> cf þ cc @ τ2 eðμ λþβÞτ dτ þ eðμ λþβÞX A

(4.37)
ðμ λÞ3 ðμ λ Þ
0
Appendix 65
From the optimality condition (4.20):

2cμ 2 cf þ cc 1 eðμ þβλÞX ðμ þβλÞX !
þ e X >
λðμ λÞ ðμ λÞ μ þ β λ ðμ þ β λÞ
0 X 1
ð
2
ðX Þ ðμ λþβÞX A
cf þ cc @ τ2 eðμ λþβÞτ dτ þ

e
ðμ λÞ
0

!
ðμ þ β λÞ2 2eðμ þβλÞX X þ ðμ þ β λÞðX Þ2 eðμ λþβÞX

,
ðμ λÞ ðμ þ β λÞ3

! Xð
ðμ þ β λÞ 2 2eðμ þβλÞX 2cμ !
þ þ > τ2 eðμ λþβÞτ dτ (4.38)
ðμ λÞ ðμ þ β λÞ3 λðμ λÞ c f þ c c
0
with
Xð

τ 2 e ð μ λþβÞτ
dτ ¼
0

λþβÞX
2 e ð μ 2 þ 2ðμ λ þ βÞX þ ðμ λ þ βÞ2 ðX Þ2
¼ (4.39)
ðμ λ þ β Þ3
it follows that:
Xð
2cμ ðμ λ þ β Þ !
þ
1 τ2 eðμ λþβÞτ dτ > 0 (4.40)

λðμ λÞ cf þ cc ðμ λÞ
0
which obviously holds for any cost structure.

Derivation of Equations (4.22) and (4.23). Restating the cost function from
Eq. (4.21) leads to:
0 1
ð2
X

CðX1 ; X2 Þ ¼ λ cy;1 þ cc @E½W1 þ FW2 ðτÞð1 FL ðτ þ X1 ÞÞdτAþ
0
ð
1 ð1
X

þ λ cf þ cc FW 2 ð X 2 Þ fW1 ðτ1 ÞðX1 τ1 Þdτ1 fL ðθÞdθ
X 0
ð ðX Xτ
1 ð2

þ λ cf þ cc fW2 ðτ2 ÞfW1 ðτ1 ÞðX τ2 τ1 Þdτ2 dτ1 fL ðθÞdθ
X X2 0
ðX ð1
X

þ λ cf þ cc FW 2 ð θ X1 Þ fW1 ðτ1 ÞðX1 τ1 Þdτ1 fL ðθÞdθ
X1 0
ðX ðθ ð2
θτ

þ λ cf þ cc fW2 ðτ2 ÞfW1 ðτ1 Þðθ τ2 τ1 Þdτ2 dτ1 fL ðθÞdθ
X1 θX1 0
ð1 ðθ θτ
X ð2

þ λ cf þ cc fW2 ðτ2 ÞfW1 ðτ1 Þðθ τ2 τ1 Þdτ2 dτ1 fL ðθÞdθ
0 0 0
ðX
λcc ð1 FL ðτÞÞdτ χ 1 X1 χ 2 X2 ð4:41Þ
0
Taking the first derivative of the restated cost function with respect to X1 shows:
ð2
X
dCðX1 ;X2 Þ
¼ λðcy;1 þ cc Þ FW2 ðτÞfL ðτ þ X1 Þdτ λcc ð1 FL ðXÞÞ
dX1
0
ð
1 ð1
X ð1
X !
þ λðcf þ cc ÞFW2 ðX2 Þ fL ðθÞ fW1 ðτ1 Þdτ1 dθ fL ðXÞ fW1 ðτ1 ÞðX1 τ1 Þdτ1
X 0 0
ð
1 ðX Xτ
ð2 !
þ λðcf þ cc Þ fL ðθÞdθ fW2 ðτ2 ÞfW1 ðτ1 Þdτ1 dτ2
X X2 0
ð2
ðX Xτ
λðcf þ cc ÞfL ðXÞ fW2 ðτ2 ÞfW1 ðτ1 ÞðX τ2 τ1 Þdτ1 dτ2
X2 0
Appendix 67
ðX ð1
X
þ λðcf þ cc Þ FW2 ðθ X1 ÞfL ðθÞdθ fW1 ðτ1 Þdτ1 þ λðcf þ cc ÞFW2 ðX X1 ÞfL ðXÞ
X1 0
ð1
X ðX
fW1 ðτ1 ÞðX1 τ1 Þdτ1 λðcf þ cc Þ fW2 ðθ X1 ÞfL ðθÞdθ
0 X1
ð1
X ðX ð2
Xτ
fW1 ðτ1 ÞðX1 τ1 Þdτ1 þ λðcf þ cc ÞfL ðXÞ fW2 ðτ2 ÞfW1 ðτ1 ÞðX τ2 τ1 Þdτ2 dτ1
0 XX1 0
ð1 X1ðτ2
X
λðcf þ cc ÞfL ðX1 Þ fW2 ðτ2 ÞfW1 ðτ1 ÞðX1 τ2 τ1 Þdτ2 dτ1 þ λðcf þ cc Þ
0 0
ðX ð1
X
fL ðθÞ fW2 ðθ X1 ÞfW1 ðτ1 ÞðX1 τ1 Þdτ1 dθ þ λðcf þ cc ÞfL ðX1 Þ

X1 0
ð1 X1ðτ2
X
fW2 ðτ2 ÞfW1 ðτ1 ÞðX1 τ2 τ1 Þdτ2 dτ1 χ 1 ¼ 0 , ðcf þ cc ÞFW1 ðX1 Þ
0 0
X1 ð
þX2
FW2 ðτ X1 ÞfL ðτÞdτ þ ðcf þ cc Þð1 FL ðXÞÞ

X1
X1 ð
þX2
FW2 ðτÞfW1 ðX1 þ X2 τÞdτ ðcy;1 þ cc Þ

X2
ð2
X
χ1
FW2 ðτÞfL ðτ þ X1 Þdτ cc ð1 FL ðXÞÞ ¼0
λ
0
ðX !
, ðcf þ cc ÞFW1 ðX1 Þ FW2 ðX2 ÞFL ðXÞ fW2 ðτ X1 ÞFL ðτÞdτ þ ðcf þ cc Þ
X1
X1 ð
þX2
χ1
ð1 FL ðXÞÞ FW2 ðτÞfW1 ðX1 þ X2 τÞdτ cc ð1 FL ðXÞÞ
λ
X2
ð2
X !!
¼ ðcy;1 þ cc Þ FW2 ðX2 ÞFL ðXÞ FW2 ðX2 ÞFL ðXÞ FW2 ðτÞfL ðτ þ X1 Þdτ (4.42)
0
Rearranging terms:
ð2
X

, cf þ cc FW1 ðX1 Þ cy;1 þ cc FW2 ðτÞfL ðτ þ X1 Þdτ
0
X1 ð
þX2

þ cf þ cc ð1 FL ðX1 þ X2 ÞÞ FW2 ðτÞfW1 ðX1 þ X2 τÞdτ
X2
χ
c c ð 1 FL ð X1 þ X2 Þ Þ 1 ¼ 0 (4.43)
λ
Taking the first derivative of the restated cost function with respect to X2 leads to:
dCðX1 ;X2 Þ
¼ λ cy;1 þ cc FW2 ðX2 Þð1 FL ðXÞÞ
dX2
ð1
X

λ cf þ cc FW2 ðX2 ÞfL ðXÞ fW1 ðτ1 ÞðX1 τ1 Þdτ1
0
ð
1 ð1
X ð
1

þλ cf þ cc fW2 ðX2 Þ fL ðθÞdθ fW1 ðτ1 ÞðX1 τ1 Þdτ1 λ cf þ cc fL ðθÞdθ
X 0 X
ð2
XX ð
1

fW2 ðX2 ÞfW1 ðτ1 ÞðX1 τ1 Þdτ1 þ λ cf þ cc fL ðθÞdθ
0 X
ð2
ðX Xτ

fW2 ðτ2 ÞfW1 ðτ1 Þdτ1 dτ2 λ cf þ cc fL ðXÞ
X2 0
ðX Xτ
ð2

fW2 ðτ2 ÞfW1 ðτ1 ÞðX τ2 τ1 Þdτ1 dτ2 þ λ cf þ cc FW2 ðX X1 ÞfL ðXÞ
X2 0
ð1
X

fW1 ðτ1 ÞðX1 τ1 Þdτ1 þ λ cf þ cc fL ðXÞ
0
ðX ð2
Xτ
fW2 ðτ2 ÞfW1 ðτ1 ÞðX τ2 τ1 Þdτ2 dτ1 λcc ð1 FL ðXÞÞ χ 2

XX1 0

, cf þ cc ð1 FL ðXÞÞ
ðX
χ
fW2 ðτÞFW1 ðX τÞdτ þ cy;1 þ cc FW2 ðX2 Þð1 FL ðXÞÞ cc ð1 FL ðXÞÞ 2 ¼ 0
λ
X2
Appendix 69
, ððcy;1 þ cc Þ ðcf þ cc ÞFW1 ðX1 ÞÞð1 FL ðX1 þ X2 ÞÞFW2 ðX2 Þ þ ðcf þ cc Þ

X1 ð
þX2
ð1 FL ðX1 þ X2 ÞÞ FW2 ðτÞfW1 ðX1 þ X2 τÞdτ

X2
χ2
cc ð1 FL ðX1 þ X2 ÞÞ ¼0 (4.44)
λ
ð1
X

E½ I ¼ ð1 FL ðXÞÞ 1 eðμ2 λÞðτþX2 Þ 1 eðμ1 λÞðX1 τÞ dτ
0
ð1
X
Xð2
þ 1 eðμ1 λÞτ dτ ðμ2 λÞeðμ2 λÞθ fL ðθ þ X1 Þdθ
0 0
ðX ðθ ð2
θτ
þ ðμ2 λÞeðμ2 λÞτ2 ðμ1 λÞeðμ1 λÞτ1 ðθ τ2 τ1 Þdτ2 dτ1 fL ðθÞdθ

X1 θX1 0
ð1 ðθ θτ
X ð2
þ ðμ2 λÞeðμ2 λÞτ2 ðμ1 λÞeðμ1 λÞτ1 ðθ τ2 τ1 Þdτ2 dτ1 fL ðθÞdθ
0 0 0

1 eðμ1 λÞX1 eðμ2 λÞX eðμ2 λÞX eðμ2 λÞX2
¼ ð1 FL ðXÞÞ X1 þ þ
μ1 λ μ1 μ2 μ2 λ μ2 λ
ð
X 2
eðμ2 λÞX2 ðμ1 λÞX1 1 eðμ1 λÞX1
þ X1 ðμ2 λÞ eðμ2 λÞθ fL ðθ þ X1 Þdθ
μ2 λ μ1 λ
0
ðX ðθ
1 þ eðμ1 λÞðθτ2 Þ þ ðμ1 λÞðθ τ2 Þ
þ fL ðθÞ ðμ2 λÞeðμ2 λÞτ2 dτ2dθ
ðμ1 λÞ
X1 θX1
ð1
X ðθ
1 þ eðμ1 λÞðθτ2 Þ þ ðμ1 λÞðθ τ2 Þ
þ fL ðθÞ ðμ2 λÞeðμ2 λÞτ2 dτ2 dθ
ðμ1 λÞ
0 0

1 eðμ1 λÞX1 eðμ2 λÞX eðμ2 λÞX eðμ2 λÞX2
¼ ð1 FL ðXÞÞ X1 þ þ
ð2
X
eðμ2 λÞX2 ðμ1 λÞX1 1 eðμ1 λÞX1
þ X1 ðμ2 λÞ eðμ2 λÞθ fL ðθ þ X1 Þdθ
μ2 λ μ1 λ
0
ðX
eðμ2 λÞθ eðμ2 λÞðθX1 Þ ðμ2 λÞ eðμ2 λÞθ eðμ2 μ1 ÞðθX1 Þðμ1 λÞθ
þ f L ðθ Þ
ðμ1 λÞ ðμ1 λÞ ðμ 2 μ 1 Þ
X1

eðμ2 λÞθ eðμ2 λÞðθX1 Þ
þ þ eðμ2 λÞðθX1 Þ θ eðμ2 λÞðθX1 Þ ðθ X1 Þ dθ
ðμ 2 λ Þ
Xð1 ðμ λÞθ
e 2 1 ðμ2 λÞ eðμ2 λÞθ eðμ1 λÞθ eðμ2 λÞθ 1
þ fL ðθÞ þ þ θ dθ
ðμ1 λÞ ðμ1 λÞ ðμ2 μ1 Þ ðμ2 λÞ
0
(4.45)
With exponential distribution for customer required lead time:

βX 1 eðμ1 λÞX1 eðμ2 λÞX eðμ2 λÞX eðμ2 λÞX2
E½ I ¼ e X1 þ þ

eðμ2 λÞX2 ðμ λÞX1
1

μ2 λ

1 eðμ1 λÞX1 eβX1 eðμ2 λþβÞX2 βX1 βðμ2 λÞ
þ X1
μ1 λ ðμ2 λ þ βÞ
ðμ λþβÞX
ðμ2 λÞβ e 1 1
eðμ1 λþβÞX1 ðμ2 λþβÞX2
þ
ðμ2 μ1 Þðμ2 λ þ βÞðμ1 λÞ
eðμ2 λþβÞX2 βX1 βðμ1 þ μ2 2λ X1 ðμ1 λÞðμ2 λÞÞ
þ
ðμ2 λ þ βÞðμ1 λÞðμ2 λÞ
eðμ2 λþβÞX βððμ1 λÞμ2 þ λðμ2 λÞ μ1 ðμ1 þ μ2 2λÞÞ
þ
ðμ2 μ1 Þðμ2 λ þ βÞðμ1 λÞðμ2 λÞ
eðμ2 λþβÞX1 βðμ1 ðμ1 þ μ2 2λÞ ðμ1 λÞμ2 λðμ2 λÞÞ
þ
ðμ2 μ1 Þðμ2 λ þ βÞðμ1 λÞðμ2 λÞ
eβX1 βððμ2 λÞ þ ðμ1 λÞð1 þ X1 ðμ2 λÞÞÞ 1 eβX1
þ
ðμ2 λ þ βÞðμ1 λÞðμ2 λÞ ðμ1 λÞ
ðμ2 λþβÞX1

β 1e βðμ2 λÞ
þ þ
ðμ1 λÞðμ2 λ þ βÞ ðμ1 λÞðμ2 μ1 Þ

1 eðμ1 λþβÞX1 1 eðμ2 λþβÞX1 1 eβX1

ðμ 1 λ þ β Þ ðμ2 λ þ βÞ ðμ 2 λ Þ
1 eðμ2 λþβÞX1 1 eβX1 ð1 þ βX1 Þ
þβ þ (4.46)
ðμ2 λÞðμ2 λ þ βÞ β
Appendix 71
Simplifying and rearranging terms leads to:
E½ I ¼
eðμ2 λþβÞX2 βX1 eðμ1 λþβÞX1 ðμ2 λÞ
þ (4.47)
ðμ2 λ þ βÞ ðμ 1 λ þ β Þðμ 2 λ þ β Þ
eβX1 ðμ2 λÞ ðμ1 λÞðμ2 λÞ
þ
β ðμ 2 λ þ β Þ β ðμ 1 λ þ β Þðμ 2 λ þ β Þ
Optimality Conditions in Two-Stage M/M/1 System
The following optimality conditions for X1 and X2 can be stated with Eqs. (4.25) and
(4.26):
X2 ¼ 0 :
ð1
X
cc
X1 : 1 eðμ2 λÞτ ðμ1 λÞeðμ1 λÞðX1 τÞ dτ ¼
cf þ cc (4.48)
0
cf ðμ λÞeðμ2 λÞX1 ðμ2 λÞeðμ1 λÞX1
, ¼ 1
cf þ cc ðμ 1 μ 2 Þ
X2 > 0 :

cy;1 þ cc 1 cf þ cc
X1 :
¼ FW 1 ð X1 Þ , X1 ¼ ln
cf þ cc ðμ1 λÞ cf cy;1
X1 ð
þX2
cc
X2 : 1 eðμ2 λÞτ ðμ1 λÞeðμ1 λÞðX1 þX2 τÞ dτ ¼
cf þ cc
X2
μ2 λ ! !
1 cf cy;1 cf cy;1 μ1 λ cy;1 μ2 μ1
, X2 ¼ ln ln
μ2 λ cf þ cc cf þ cc cf þ cc μ 1 λ
(4.49)
Chapter 5
Optimal Composition of Number and Size of
Machines
An important trade-off in manufacturing capacity planning is between the number

and size of machines, i.e. a few large versus a larger number of small machines. The
complexity of such a tactical decision further increases in a multi-stage system and
costs for investments need to be further traded-off with inventory and backorder
costs whereby investment decisions need to correctly anticipate operational replen-
ishment and workload control decisions.
For MTS production systems, the optimal replenishment strategy minimizing
inventory and backorder costs, both with an independent (exogenous) and depen-
dent (endogenous) distribution of replenishment lead time is a broadly discussed
topic (see Silver et al. (1998) and Zipkin (2000)). The literature on determining
optimal number and size of machines in MTO production systems focuses mainly
on balancing costs of waiting with costs of capacity. The majority derives that a
single server queue leads to the optimal result when linear capacity costs and a
continuous range of processing rate is assumed (see Chap. 2). These queuing design
models are extended by including customer due dates, FGI as well as backorder
costs, and workload is released to the system by a WAW policy.
Most contributions on capacity investment analyze single-machine or single-
stage production systems with continuous capacity investment possibilities where
inventory and/or backorder costs (or instead, a service level constraint) are balanced
against capacity investment. In the majority of contributions, processing rates can
be set as continuous variables. In reality, however, there often exists a finite set of
possible technology options. The finiteness of manufacturing rates can be motivated
either by available technologies, e.g. a machine can only run at certain speeds,
requires certain batch sizes, etc. or by restrictions imposed by the workforce.
Regulations might require that a worker has to be compensated for a minimum
number of hours and that working time is organized in full shifts (either 1, 2, or
3 per day) or a certain number of full working days. Furthermore, different
workforce qualifications and the technology of automatic handling systems can

74 5 Optimal Composition of Number and Size of Machines
also lead to a predefined set of processing rates. Therefore the influence of a

predefined set of processing rates on the optimal capacity investment is investigated
in an MTO production system with stochastic due dates, a WAW work release
policy, as well as FGI and backorder costs. The question of which technology
option to take and how much machines to invest for minimizing overall costs in
such a system is one continuously faced by production managers and therefore of
practical relevance.
It is evaluated under which conditions a single-machine policy at each stage is
optimal in a serial MTO production system with stochastic due dates. The cost
penalty of investing in more than one machine at each stage is calculated in a
two-stage M/M/s production system with exponentially distributed customer
required lead time when processing rates have a continuous range. Additionally,
the impact of predefined processing rates on overall costs in a two-stage M/M/s
production system with exponentially distributed customer required lead time is
investigated. Building on the results of Chap. 4, the trade-off between multiple
machines and machine sizes and here especially on a predefined set of technologies
rather than only allowing for a single machine with continuous processing rate at
each stage, is addressed.
The chapter based on Altendorfer and Minner (2012) is structured as follows. In
Sect. 5.2, the general multi-stage model is introduced and structural properties of
optimal solutions are derived. Expressions for WIP, FGI, and backorder costs for
the two-stage M/M/s production system are developed in Sect. 5.3. A numerical
example in Sect. 5.4 illustrates the impact of predefined processing rates and
uncertain customer order input rates. All proofs are added in the Appendix.
5.2 General n-Stage Optimization Problem
5.2.1 Model Description and Assumptions
A manufacturing system with i ¼ 1,2,. . .,n stages numbered in reverse order from n
(most upstream) to 1 (most downstream) as shown in Fig. 5.1 is considered. Each
stage i has si (identical) parallel machines (si being integer).
Customer orders arrive to the manufacturing system according to a stochastic
process with mean arrival rate λ. Each customer order requests an individual due
date that determines the non-negative random customer required lead time L. Let
fL ðÞ denote the pdf of the customer required lead time, FL ðÞ the respective cdf, and
1=β its mean. The customer required lead time cannot be influenced by the
production system and all customer orders are accepted. Customer orders are
processed consecutively by the n stages of the manufacturing system following a
FIFS discipline. Each stage i consists of one shared buffer for waiting orders and si
identical parallel machines with random processing rate with mean μi per unit of
time of one machine at stage i. In the general case, μi is assumed to be a continuous
5.2 General n-Stage Optimization Problem 75

lead time = L C = max(0, W–L)
Customer
Proces- Proces-
sing step n sing step 1
msn m s1
...
...
m2 m2
Buffer
Buffer Buffer
n ...
m1 1
m1
FGI = G
WIP = Yn WIP = Y1
FGI lead time

Production lead time = Wn Production lead time = W1 I = max(0, L–W )
Production lead time W = Wn +...+W1
Fig. 5.1 Multi-stage production system with parallel machines
decision variable. However, it might be further restricted to a set of feasible values

Ω. The capacity at each stage is therefore given by si μi which is assumed to be
determined by an investment decision either including both, the processing rate and
the machine number, or any of the two single variables. cμ;i denotes the capacity cost
per unit of installed capacity at stage i. Wi denotes the random variable of
production lead time at stage i needed for one order from arrival at stage i until
Pi
~i ¼
its completion. W Wj denotes the remaining production lead time from arrival
j¼1
at stage i until its delivery to the FGI buffer. Orders are released to stage n of the
production system whenever their remaining time to due date is lower than a
predefined WAW X.
Let Yi denote the random WIP at stage i and cy,i the respective holding cost per
order per unit of time stored at stage i. A value added WIP holding cost structure is
assumed, i.e. an increase after each processing step, cy;i cy;iþ1 . Let cy;nþ1 ¼ 0;
i.e. holding costs for orders not yet released to the production system and therefore
waiting before stage n are neglected. The accounting of WIP at stage i starts when
the order is released to the buffer i and ends when the order is delivered to buffer
i1. Let G denote the random FGI and cf the respective inventory holding cost per
order per unit of time. Backorders B are charged a backorder cost cc per unit of time
an order is late. C denotes the tardiness random variable of an order and I the
random FGI lead time.
Minimizing expected WIP, FGI, backorder, and capacity costs leads to optimi-
zation problem (5.1).
X
n X
n
~ i λ cy;i cy;iþ1 þ E½Gcf þ E½ Bcc þ
E W si μi cμ;i ! min
X;fsi g;fμi g
i¼1 i¼1
si μi > λ8i and si 2 N þ (5.1)
The WIP cost calculation in Eq. (5.1) is based on the relative WIP costs added
per stage. The second line of Eq. (5.1) is the stability condition stating that
utilization at each stage has to be lower than 100 %.
5.2.2 Model Analysis
For a single-stage as well as a serial production system, expressions for expected

FGI and backorders without stochastic due dates are provided by Buzacott and
Shanthikumar (1994) and Yano (1987). Chapters 3 and 4 extend these equations to
include a random due date as follows (see Eqs. (4.3) and (4.4)):
ðX
E½ I ¼ FW ðτÞð1 FL ðτÞÞdτ (5.2)
0
ðX
E½C ¼ E½ I þ E½W ð1 FL ðτÞÞdτ (5.3)
0
P
i
~i ¼
From the definition of the remaining production lead time (W Wj) it follows
j¼1
that the pdf of W~ i is a convolution of the pdfs of all remaining production lead time
values Wj with 1 j i. Denoting the convolution between the pdf of stochastic
Ðt
variable W ~ i1 with the pdf of Wi with f ~ ðτÞfWi ðt τÞdτ ¼ f ~ ðtÞ fWi ðtÞ
W i1 W i1
0
leads to:
fW~ i ðtÞ ¼ ðððfW1 ðtÞ fW2 ðtÞÞ fW3 ðtÞÞ . . . fWi ðtÞÞ (5.4)
For the general optimization problem, the following proposition concerning the
single-machine optimality can be stated.
Proposition 5.1. For any manufacturing system consisting of a series of
processing stages with parallel machinesdefined
by ðfsi g; fμi gÞ and continuous
processing rates, a single-machine policy f1g; μ1i is optimal to minimize WIP,
FGI, backorder, and capacity costs as defined in Eq. (5.1) when the remaining
production lead time of the parallel machine system at each stage second order
5.2 General n-Stage Optimization Problem 77
stochastic dominates the remaining production lead time of the single-machine

stage (see condition (5.5)).
ðθ ðθ
FW~ 1 ðτÞFW~ i ðτÞdτ 0 8θ > 0 and 9θ : FW~ 1 ðτÞFW~ i ðτÞdτ > 0 8i (5.5)
i i
0 0
Proof see Appendix. The superscript 1 defines the system with a single machine
per stage.
For a single-stage M/M/s queuing system without considering a distribution of
customer required lead time, Stidham, Jr. (1970) shows that condition (5.5) is
fulfilled and therefore the optimality of a single server. However, this does not
hold in general. Brumelle (1971) provides an example for a squared coefficient of
variation of the processing time greater than 1 where condition (5.5) is not fulfilled.
Nevertheless, the definition in Eq. (5.4) together with the finding of Stidham,
Jr. (1970) shows that the condition is fulfilled for a series of M/M/s queues as
discussed in the next sections.
Proposition 5.1 is intuitive for a series of M/M/s queues since the single machine
at each stage is exactly tailored to the customer order rate.
processing
stages with single machines and capacity costs increasing in processing
Pn
rate Cðfμi gÞ ¼ ξðμi Þcμ;i and dξdμðμi Þ > 0 and non-increasing in stage
i
i¼1
c cμ;i1 , the optimal processing rate is non-decreasing towards the FGI
μ;i
μi μiþ1 when WIP, FGI, backorder, and capacity costs are minimized.
Proof see Appendix.
A managerial implication of Proposition 5.2 is that (if not overruled by suffi-
ciently increasing capacity costs) it is always better to have a higher downstream
capacity and that the integration of a customer required lead time distribution has no
influence on that property.
processing stages with parallel machines, equal processing rates, and linear
capacity
costs with respect to the number of machines at each stage
Pn
C ð fs i gÞ ¼ si μcμ;i which are non-increasing in stage cμ;i cμ;i1 , the
i¼1
optimal allocation of machines is non-decreasing towards the FGI si siþ1
when WIP, FGI, backorder, and capacity costs are minimized.
The proof is similar to the one for Proposition 5.2 and therefore omitted.
Proposition 5.3 shows that higher capacities should be kept towards the customer
end of the production line not only for continuous processing rates but also for cases
with a predefined machine size.
Proposition 5.4. For a manufacturing system consisting of a series of processing
stages with parallel machines defined by ðfsi g; fμi gÞ the cost for inventory, backorder,
and capacity as defined in Eq. (5.1) with cy;k ¼ cf ¼ cy are unimodal in si .
Proof see Appendix.

Assuming that the condition cy;k ¼ cf ¼ cy has only minor influence on the
unimodality and is rather a technical necessity for the proof, Proposition 5.4 can be
applied to find a solution heuristic for the general optimization problem restricted to
a predefined set of processing rates where cy;k 6¼ cf .
The general form of the optimization problem as stated in this section can be
applied to find optimal parameters for any kind of n-stage production system as
long as the production lead time distributions for the n stages as well as the
customer required lead time distribution are known. In the next section a solution
heuristic for such a problem in a two-stage M/M/s production system is developed.
5.3 Optimization Problem for a Two-Stage M/M/s Production

System
In this section a special case with two stages, M/M/s queues, exponentially
distributed customer required lead time, and a predefined set of processing rates Ω
is discussed. This two-stage setting, chosen for simplicity of exposition, allows for
closed form expressions. However, by convolution of production lead time
distributions, this could be extended to general serial systems.
5.3.1 M/M/s Queue
The M/M/s queue has an exponential interarrival time and an exponential

processing time whereby the orders are processed at s parallel servers fed by one
common queue. This processing of orders at parallel servers leads in this system to
the following equations for WIP, production lead time, and utilization (see Tijms
(2003) and Medhi (1991)):
X
s1
ðρsÞi ðρsÞs sμ2
fWM=M=s ðtÞ ¼ p0 μeμt þ p0 eμt eð1ρÞsμt
i¼0
i! s! ð s 1 Þμ λ
" #1
Xs1
ðρsÞi ðρsÞs
p0 ¼ þ ;
i¼0
i! s!ð1 ρÞ
ðρsÞs sμ2 X
s1
ðρsÞi
with o ¼ p0 ; m¼ p0 μ
s! ðs 1Þμ λ i¼0
i!
) fWM=M=s ðtÞ ¼ ðm þ oÞeμt oeð1ρÞsμt (5.6)
5.3 Optimization Problem for a Two-Stage M/M/s Production System 79
λ
ρ¼
sμ
1 1 ðsρÞs
E½ W ¼ þ p0
μ sμ λ s!ð1 ρÞ
λ ðsρÞs
E½Y ¼ sρ þ p0 (5.7)
sμ λ s!ð1 ρÞ
The utilization of machines is denoted by ρ and p0 is the probability that no order

is in the production system. m and o are defined as auxiliary variables for simplifi-
cation of statements.
Equation (5.6) shows that the production lead time is Erlang distributed in the
M/M/s setting. Nevertheless, the M/M/s queue still provides a Poisson output
stream (see Tijms (2003)). This means that a serial production system with parallel
machines can be modeled as a series of M/M/s queues.
5.3.2 Problem Statement
Following Proposition 3.1, the implementation of a WAW in front of the production

system still leads to a Poisson input stream into the production stage. Since
the M/M/s queue still maintains a Poisson output stream (see Tijms (2003)) all
equations developed in this section are exact.
The optimization problem (5.1) simplifies to:

E½W1 λ cy;1 þ cc þ E½W2 λ cy;2 þ cc þ E½ I λ cf þ cc
ðX
ð1 FL ðτÞÞdτλcc þ s1 μ1 cμ;1 þ s2 μ2 cμ;2 ! min
X;s1 ;μ1 ;s2 ;μ2
0
s1 μ1 > λ and s2 μ2 > λ;
μ1 ; μ2 2 Ω and s1 ; s2 2 N þ (5.8)
To solve the optimization problem, the distributions fW~ i ðtÞ have to be calculated
first. Based on Eq. (5.6) it follows:
fW~ 1 ðtÞ ¼ fW1 ðtÞ ¼ ðm1 þ o1 Þeμt o1 eð1ρÞsμt

ðt
fW ðtÞ ¼ fW~ 2 ðtÞ ¼ fW1 ðtÞ fW2 ðtÞ ¼ fW1 ðτÞfW2 ðt τÞdτ
0
ðt
¼ ðm1 þ o1 Þeμ1 τ o1 eð1ρ1 Þs1 μ1 τ ðm2 þ o2 Þeμ2 ðtτÞ o2 eð1ρ2 Þs2 μ2 ðtτÞ dτ
0
eμ1 t eμ2 t eðs2 μ2 λÞt eμ1 t
¼ ðm1 þ o1 Þðm2 þ o2 Þ ðm1 þ o1 Þo2
ðμ 2 μ 1 Þ ðμ 1 s2 μ 2 þ λ Þ
eðs1 μ1 λÞt eμ2 t eðs1 μ1 λÞt eðs2 μ2 λÞt
ðm2 þ o2 Þo1 þ o1 o2
ðμ 2 s1 μ 1 þ λ Þ ðs2 μ2 s1 μ1 Þ
(5.9)
For Eqs. (5.9, 5.10, 5.11, 5.12, 5.13, and 5.14) non-identical processing rates are
assumed. The equations for identical processing rates and identical number of
machines (referenced as balanced production system) are delivered in the
Appendix.
5.3.3 WIP, FGI, and Backorders Calculation
Using Eq. (5.9) leads to the following expected production lead time values:
ðmi þ oi Þ oi
E½Wi ¼ (5.10)
μ2i ðð1 ρi Þsi μi Þ2
Applying Eq. (5.9) to Eq. (5.2) leads to the expected value for FGI lead time:

ðt1 þ t2 Þ ðt3 t1 Þ ðt 2 þ t 4 Þ ðt4 t3 Þ 1 eβX
E½I ¼ þ þ
μ1 μ2 ðs2 μ2 λÞ ðs1 μ1 λÞ β
ðt1 þ t2 Þ 1 eðμ1 þβÞX ðt3 t1 Þ 1 eðμ2 þβÞX

μ1 ðμ 1 þ β Þ μ2 ðμ 2 þ β Þ
ðt2 þ t4 Þ 1 eðs2 μ2 λþβÞX ðt4 t3 Þ 1 eðs1 μ1 λþβÞX
þ (5.11)
ðs2 μ 2 λ Þ ðs2 μ 2 λ þ β Þ ðs1 μ1 λÞ ðs1 μ1 λ þ βÞ
Derivation see Appendix. t1 ; t2 ; t3 and t4 are defined as auxiliary variables for

simplification of statements (see Appendix).
The expected value for tardiness can be calculated based on Eqs. (5.3) and (5.11)
as follows:
ðX
0

ðt1 þ t2 Þ ðt3 t1 Þ ðt 2 þ t 4 Þ ðt4 t3 Þ 1 eβX
¼ þ þ 1

μ1 ðμ1 þ βÞ μ2 ðμ 2 þ β Þ
þ
ðs2 μ2 λÞ ðs2 μ2 λ þ βÞ ðs1 μ1 λÞ ðs1 μ1 λ þ βÞ
ðm1 þ o1 Þ o1 ðm2 þ o2 Þ o2
þ þ (5.12)
μ21 ðð1 ρ1 Þs1 μ1 Þ 2 μ 2
2 ðð1 ρ2 Þs2 μ2 Þ2
In Fig. 5.2 the WIP, FGI lead time, and tardiness are shown to illustrate their
dependency on utilization, number of machines, and processing rates of machines.
Figure 5.2a, c, e show the influence of a different number of machines at stages
1 and 2. The number of machines at each stage is predefined and the processing rate
is varied. The influence of predefined processing rates is depicted in Fig. 5.2b, d, f.
The processing rate is kept equal at both stages and the number of machines is varied.
The results from Fig. 5.2a, c, e show that an imbalance between the processing
rates at the two stages which creates an imbalance in number of machines, leads to
an increase in WIP, a decrease in FGI lead time, and an increase in tardiness at the
same utilization. Figure 5.2b, d, f show that whenever the processing rate is
predefined, contrary to Fig. 5.2a, c, e there are only certain utilization rates possible.
In the examples shown in Fig. 5.2b, d, f, especially high utilization values cannot be
reached because of the relation between processing rate and order arrival rate.
Nevertheless, Fig. 5.2b, d, f show that higher processing rates lead to lower WIP,
higher FGI lead time, and lower tardiness values at the same utilization.
For the optimization problem (5.8) minimizing the FGI, backorder, WIP, and
capacity costs when a set of processing rates are predefined, only the problem
statement can be provided in explicit form; the solution has to be found numeri-
cally. For given capacity investment, in Eq. (4.18) the following solution for
optimal WAW is provided:
cc
X : FW~ n ðXÞ ¼ (5.13)
cf þ cc
Applying Eq. (5.32) from Appendix for the cdf of overall production lead time
leads to the following optimality condition:
a 20 b 20
18 18
16 16
14 14
12 12
WIP
WIP
10 10
8 8
6 6
4 4
2 2
0 0
0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1
Ex A.1 Ex A.2 Ex A.3 Ex A.4 Ex B.1 Ex B.2 Ex B.3 Ex B.4
c 50
d 50
40 40
FGI lead time
FGI lead time
30 30
20 20
10 10
0 0
0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1
e 3 f 3
2 2
Tardiness
Tardiness
1 1
0 0
0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1
l = 1; X = 20; b = 0.005; s1 = 2; l = 1; X = 20; b = 0.005;

s2 ( Ex A.1) = 2; s2 ( Ex A.2 ) = 4; m ( Ex B.1) = 0.5; m ( Ex B.2 ) = 0.3;
s2 ( Ex A.3 ) = 6; s2 ( Ex A.4 ) = 8; m ( Ex B.3 ) = 0.25; m ( Ex B.4 ) = 0.16
m i = l / ( si r ) ;
Fig. 5.2 Influence of processing rate and number of machines on logistical key figures
cc 1 eμ1 X 1 eμ2 X
¼ ðt1 þ t2 Þ þ ðt3 t1 Þ
cf þ cc μ1 μ2
1 eðs2 μ2 λÞX 1 eðs1 μ1 λÞX
ðt2 þ t4 Þ þ ðt4 t3 Þ (5.14)
ðs2 μ2 λÞ ðs1 μ 1 λ Þ
5.3.4 Solution Heuristic for the Two-Stage M/M/s Problem
In this section a simple solution algorithm for problem (5.8) is proposed based on
the two-stage M/M/1 problem and on Proposition 5.4. In the numerical example
provided in the following section it is shown that this simple algorithm leads to
good results when being compared to an enumeration scheme for finding the
optimal solution.
Algorithm.
Step 1: Calculate μ1 and μ2 of the two-stage M/M/1 problem with continuous
processing rates (see Eqs. (4.30) and (4.48)).
Step 2: Calculate the number of machines si;j for each predefined processing rate
l m j k
μ
μj 2 Ω and for both stages i 2 f1; 2g by si;j ¼ max μλ ; μi .
j j
Step 3: For each processing rate μj 2 Ω:

Calculate the cost term C s1; j ; μj ; s2 ; μ2 ; X s1; j ; μj ; s2 ; μ2 , whereby s2 ¼ 1 from
μ2 being calculated in the single machine case. If

C s1; j þ 1; μj ; s2 ; μ2 ; X s1; j þ 1; μj ; s2 ; μ2

< C s1; j ; μj ; s2 ; μ2 ; X s1; j ; μj ; s2 ; μ2 (5.15)
then increase s1; j until

C s1; j þ 1; μj ; s2 ; μ2 ; X s1; j þ 1; μj ; s2 ; μ2 >

C s1; j ; μj ; s2 ; μ2 ; X s1; j ; μj ; s2 ; μ2 (5.16)
is fulfilled and otherwise decrease s1; j until

C s1; j 1; μj ; s2 ; μ2 ; X s1; j 1; μj ; s2 ; μ2 >

C s1; j ; μj ; s2 ; μ2 ; X s1; j ; μj ; s2 ; μ2 (5.17)
is fulfilled and set s1;j ¼ s1; j .

n o
Select the values s1;j ; μ1 ðΩÞ minimizing the cost function. X is implicitly
defined in Eq. (5.14).
Step 4: Redo the same procedure as in Step 3 for
C s1;j ; μ1 ðΩÞ; s2;j ; μj ; X s1;j ; μ1 ðΩÞ; s2;j ; μj (5.18)
n o
whereby s2;j is increased/decreased and s2;j ; μ2 ðΩÞ is selected.
Two limiting assumptions are made in this solution heuristic. The first assump-
tion is that Proposition 5.4 holds for any cost structure. The second assumption is
that it is sufficient to optimize the number of machines and processing rates stage-
wise whereby the other stage is set to the overall optimal capacity. When
optimizing stage 1, the parameters for stage 2 are set to one machine with μ2
whereas when stage 2 is optimized, the optimal parameters for stage 1 from Step
3 are applied.
Additionally, an efficient starting point for the optimal capacity search is
included in the algorithm. Note that also all combinations of optimality candidates
on both stages could be tested but this would lead to a considerable additional
computational effort.
5.4 Numerical Study
The first effect studied is the influence of predefined processing rates on the optimal
cost and optimal number of machines. The second effect studied is the cost penalty
for investing multiple machines when the processing rates can be optimized from a
continuous set. In the third subsection the performance of the developed solution
heuristic is evaluated. Additionally, the costs of different processing rate sets are
shown. The fourth effect studied is the influence of uncertainty about the customer
order rate at the point in time of capacity investment.
5.4.1 Influence of Machine Size on Optimal Cost
In this section the influence of different processing rates on the optimal costs for a
production system with balanced capacities (μ1 ¼ μ2) is investigated (equations for
this case are provided in the Appendix). The balanced case is chosen since the main
observations are not influenced by that assumption but it reduces complexity.
Table 5.1 shows the parameters for the four examples compared in this section.
Example A is the basic scenario. Example B is a situation where capacity costs are
low. In Example C the mean customer required lead time is very short and in
Example D the tardiness penalties are very high.
For these four examples the optimal costs and the optimal number of machines
are numerically determined by solving the optimization problem (5.8) for
processing rate values between 0.001 and 1.2. The calculation has been performed
Table 5.1 Numerical Example cc cf cy;2 cy;1 cμ;1 ¼ cμ;2 λ β

examples for overall cost
evaluation in two-stage A 30 10 5 7.5 1,000 1 0.005
M/M/s model B 30 10 5 7.5 250 1 0.005
C 30 10 5 7.5 1,000 1 0.05
D 100 10 5 7.5 1,000 1 0.005
Ex A) Ex B)
20 20
2 machines are optimal


1 machine is optimal
Optimal number of machines

18 1800 18
4500
16 1600 16
14 1400 14
Overall costs
Overall costs
4000
12 1200 12
3500 10 1000 10
8 800 8
3000
6 600 6
4 400 4
2500
2 200 2
2000 0 0 0
0 0.2 0.4 0.6 0.8 1 1.2 0 0.2 0.4 0.6 0.8 1 1.2
Processing rate Processing rate
Overall costs Optimal number of machines Overall costs Optimal number of machines
Ex C) Ex D)
20 20



18 18
4500 4500
16 16
14 14
Overall costs
Overall costs
4000 4000
12 12
3500 10 3500 10
8 8
3000 3000
6 6
4 4
2500 2500
2 2
2000 0 2000 0
0 0.2 0.4 0.6 0.8 1 1.2 0 0.2 0.4 0.6 0.8 1 1.2
Fig. 5.3 Optimal overall costs dependent on machine size in two-stage M/M/s system
in Wolfram Mathematica 6.0. Based on the results shown in Fig. 5.3 the following
observations are stated.
Observation 5.1. Whenever different machine sizes with different processing rates
are predefined and only the WAW and the number of machines can be optimized, the
optimal number of machines is highly dependent on the predetermined processing
rates.
This observation shows that for practical applications it is not always optimal to
invest in only a single machine capable of handling all the customer demands. A lot
of capacity can be wasted by this one big machine instead of applying several
smaller machines.
Observation 5.2. The range of optimal costs is a convex function with respect to
the predefined processing rates. Very big and very small machines induce a high
risk of incurring too high costs.
This statement is based on the difference between the highest and the lowest
optimal costs when a certain range of processing rates is focused. In Example A,
the highest and lowest costs for a processing rate range between 1.0 and 1.2
is 1,650 whereas it is only 715 for a processing rate range of 0.4–0.6 and it
increases again to 1 for a processing rate range of 0–0.2. Assuming a
predetermined set of available processing rates, Observation 5.2 states that also
small machines can fit with a high probability. This observation is supported by
the results from Sect. 5.4.3 where 1,875 test instances are compared applying
4 processing rate sets. In 38 % of the cases more than one machine is optimal at
least at one of the two stages.
The result for Example B shows that the range of optimal costs as a function of
the predefined processing rate reduces when capacity costs are lower. The
Examples C and D confirm that the observations are robust to changes in mean
customer required lead time and backorder costs.
Figure 5.4 shows the optimal number of machines for two different machine size
policies. Parameters of Example A are used and the vertical black lines show the
available processing rates.
In Fig. 5.4a, there are only three different machine sizes available defined by
three different processing rates. In this case the smallest machine (lowest
processing rate) leads to the lowest overall costs, which is consistent with Observa-
tion 5.2. In Fig. 5.4b, there are 6 processing rates available. In this case the optimal
processing rate is neither the smallest nor the largest one. As the example from
Fig. 5.4 shows, the optimal number of machines highly depends on the machine
sizes available. This is true although the overall minimum is always found with one
machine per stage when the processing rate can be continuously scaled as stated in
Proposition 5.1.
Comparing Observations 5.1 and 5.2 with the TPS (Toyota Production System)
philosophy to invest into small and simple machines, which in TPS is mainly based
on maintenance, flexibility, and machine failure arguments, shows that this philos-
ophy can, depending on the actual processing rates available, lead to a cost
premium paid. However, this cost premium is smaller than expected based on
Proposition 5.1. Especially the result of Observation 5.2 reinforces that smaller
machines lead to low cost penalties or are even optimal. This finding, which cannot
be derived in a continuous processing rate model, supports the application of TPS
from an investment cost perspective.
a 20 b 20
optimal point: 18
machine size µ =0,25 18

4500 machine number s =5 4500
16 16
14 14
4000 4000
Overall costs
Overall costs
12 12
3500 10 3500 10
8 8
3000 3000
6 6
4 4
2500 2500
2 2
2000 0 2000 0
0 0.2 0.4 0.6 0.8 1 1.2 0 0,2 0,4 0,6 0,8 1 1,2
optimal point:
Processing rate machine size µ = 0.6 Processing rate
machine number s =2
Fig. 5.4 Optimal number of machines with predefined machine sizes in two-stage M/M/s model
5.4.2 Influence of Machine Number on Optimal Costs
For a set of test instances the optimal processing rate is determined for a predefined
machine number. Therefore, problem (5.8) is solved under condition (5.14) with
continuous processing rates for the balanced case. The test instances consist of all
combinations of the following parameter sets:
• 1=β 2 f12; 36; 60; 84; 108g;
• λ ¼ 1;
• cμ;1 ¼ cμ;2 ¼ cμ 2 f100 þ 225k; k ¼ 0; . . . ; 4g;
• cc 2 f10 þ 5k; k ¼ 0; . . . ; 4g;
• cy;1 2 f2k; k ¼ 1; . . . ; 5g;
• cy;2 2 f2k; k ¼ 1; . . . ; 5g cy;1 :
The cost parameters are chosen to reflect a broad range of relative cost structures
and the β values are based on the unconstrained expected production lead time of a
two-stage M/M/1 system with 75 % utilization E½W jX ¼ 1; ρ ¼ 0:75 ¼ 6. They
account for a mean customer required lead time between 2 and 18 times that value
which seems to be a feasible region for possible MTO situations.
In Fig. 5.5, the average cost increase over all test instances for a predefined
number of machines is calculated. The results show that a considerable cost penalty
occurs when more than one machine is implemented at both stages and the
processing rate is a continuous optimization variable.
Fig. 5.5 Cost increase for 20%

more than one machine at 18%
each stage in two-stage 16%
Cost increase in %
M/M/s model 14%
12%
10%
8%
6%
4%
2%
0%
2 3 4 5 6 7 8 9 10
Number of machines
5.4.3 Performance Evaluation of Solution Heuristic
In this section the same problem instances as described in Sect. 5.4.2 are used and
problem (5.8) is solved in the general case. Four different processing rate sets are
tested:

• μi ðSet 1Þ 2 1=10 2k ; k ¼ 0; . . . ; 4 ;

• μi ðSet 2Þ 2 1=7:5 2k ; k ¼ 0; . . . ; 4 ;
• μi ðSet 3Þ 2 f1=6 k; k ¼ 1; . . . ; 9g;
• μi ðSet 4Þ 2 f1=4 k; k ¼ 1; . . . ; 6g:
From Fig. 5.4, it is conjectured that not only the number of values in a set of
possible processing rates, but also their basic value and the construction scheme,
have a significant influence on costs. Therefore, two different μ-value schemes each
with two different basic values are tested.
For the overall 1,875 test instances the optimal solution with continuous range of
processing rates, the optimal solution applying the four processing rate sets, and the
heuristic solution from Sect. 5.3.4 have been calculated. The optimal solution is
found by enumeration of processing rates and number of machines under condition
(5.14).
The results in Table 5.2 indicate that the solution heuristic leads to good
results. Only in a few cases the solution found deviates from the optimal one,
but the average cost increase is negligible. Additionally, the results in Table 5.2
also show the average cost increase caused by the respective processing rate
set.
Table 5.2 Two-stage M/M/s optimization heuristic performance comparison

Processing Avg. cost increase by proc. % of test instances heuristic Avg. cost increase by
rate set rate set is optimal heuristic
Set 1 3.2 % 100 % 0%
Set 2 8.4 % 95.9 % 0.006 %
Set 3 0.6 % 99.1 % 0.001 %
Set 4 2.6 % 99.7 % 0.0003 %
5.4.4 Influence of Uncertain Input Rates
In real investment problems the customer order arrival rate, i.e. the parameter λ, can
be uncertain at the time of investment. In this section it is assumed that prior to the
capacity investment decision only uncertain information about the order arrival rate
is available. In particular, the order arrival rate follows a three point distribution
with a low, medium, and high expected value λl , λm , and λh occurring with
probabilities Pðλl Þ, Pðλm Þ, and Pðλh Þ respectively. After the investment, the real
order arrival rate is observed and the optimal WAW is determined. For the
two-stage case with equal processing rates at both stages and a continuous range
of processing rates, the optimization problem is:
Pðλl ÞCðXl ; μ; s; λl Þ þ Pðλm ÞCðXm ; μ; s; λm Þ

þPðλh ÞCðXh ; μ; s; λh Þ ! min (5.19)
Xl ;Xm ;Xh ;μ;s
CðÞ is the cost function from Eq. (5.8) with a continuous range of processing
rates. All cost factors are taken from Example A and the range parameters are
shown in Table 5.3.
The numerical results in Figs. 5.6 and 5.7 illustrate the influence of the range
between λl and λh as well as the influence of the order arrival rate probabilities Pðλl Þ
and Pðλh Þ in comparison to Pðλm Þ. In Fig. 5.6 the number of machines is predefined
whereas in Fig. 5.7 a single machine (as result of the numerical solution of problem
(5.19)) is invested at each stage (i.e. s ¼ 1).
Based on the findings of this numerical example two further observations can be
stated.
Observation 5.3. The uncertainty in order arrival rates has no influence on the
single-machine optimality property stated in Proposition 5.1. Uncertain order
arrival rates do not favor the investment of more than one machine if processing
rates can be chosen continuously.
In this case, the cost functions concerning the processing rate look similar to the
ones presented in Figs. 5.3 and 5.4.
Observation 5.4. Order arrival rate parameter uncertainty leads to a consider-
able increase in optimal costs and the optimal capacity invested increases.
Table 5.3 Two-stage M/M/s Example λl λm λh

range uncertainty experiment
parameters Low range 0.95 1 1.05
Medium range 0.90 1 1.10
High range 0.80 1 1.20
a b
3000 3000
2900 2900
Expected costs
Expected costs
2800 2800
2700 2700
2600 2600
2500 2500
0 2 4 6 8 10 0 2 4 6 8 10
Number of machines Number of machines
constant low range medium range high range constant low range medium range high range
P ( λl ) = 0.25; P ( λm ) = 0.50; P ( λh ) = 0.25 P ( λl ) = 0.1; P ( λm) =0.8; P ( λh ) = 0.1
Fig. 5.6 Influence of uncertain input rate on optimal costs in two-stage M/M/1 model
1.4 2800
2750
1.3
2700
Copt
µopt
1.2 2650
2600
1.1
2550
1 2500
0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
P (λ m ) P (λ m )
constant low range medium range high range constant low range medium range high range
1− P (λ m )
P(λl ) = P (λ h ) =
2
Fig. 5.7 Influence of range and uncertainty of input rate in two-stage M/M/1 model
Observation 5.4 is intuitive as variability increases when uncertainty of order

arrival rate increases and this variability increase has to be covered with additional
capacity and additional costs for inventory and backorders. Even though this
finding is linked in the current example to a symmetric structure in uncertainty it
is conjectured that an asymmetric uncertainty has no influence on that principal
5.5 Concluding Remarks 91
result. Furthermore, no possibility to reduce the order arrival rate, if the high order
arrival rate occurs, is included in this setting which could be a field of further
research.
A second experiment with a limited capacity decision postponement option is
conducted for a predefined set of processing rates. When λ is uncertain, the
technology, i.e. the machine size has to be decided. After observing the true order
arrival rate, additional machines with the same processing rate can be added. This
leads to the following optimization problem:
Pðλl ÞCðXl ; μ; sl ; λl Þ þ Pðλm ÞCðXm ; μ; sm ; λm Þ

þPðλh ÞCðXh ; μ; sh ; λh Þ ! min (5.20)
Xl ;Xm ;Xh ;μ;sl ;sm ;sh
The optimization problem is solved for the test instances from Sect. 5.4.3 with the
simplification that a balanced production system is considered and that cy;1 ¼ cy;2 .
This leads to 625 test instances which are compared for the four processing rate sets
and Pðλl Þ ¼ 0:25; Pðλm Þ ¼ 0:50; Pðλh Þ ¼ 0:25. The percentages of test instances
where more than one machine is optimal are shown in Table 5.4.
Observation 5.5. An increase in range uncertainty about the order arrival rate
leads to a higher probability that more than one machine (with smaller processing
rates) are optimal in comparison to a known order arrival rate when only the
machine type (processing rate) has to be decided under uncertainty.
For most of the processing rate sets, order arrival rate uncertainty leads to an
increase of instances where more than one machine, however with a lower
processing rate, is optimal. As a managerial insight, under demand uncertainty it
is beneficial to invest into smaller machines if investment into additional machines
of the same type is possible after observing the real order arrival rate.
In this chapter the influence of machine size on the optimal costs for WIP, FGI,
backorders, and capacity is analyzed. A general optimization problem for an n-
stage production system is stated and explicit expressions for expected WIP, FGI,
and backorders are developed for a two-stage M/M/s production system with
exponentially distributed customer required lead time and a WAW order release
policy.
For the n-stage production system a general condition for single-machine opti-
mality at each stage is provided and it is shown that in a system with value added
WIP cost structure the distribution of customer required lead time has no influence
on the property that processing rates should increase towards the customer end of
the line.
Table 5.4 Results from numerical study on input rate uncertainty in two-stage M/M/s model
% of instances where more than Average processing rate of
1 machine is optimal machines invested
Set 1 Set 2 Set 3 Set 4 Set 1 Set 2 Set 3 Set 4
Known λ 81.9 % 65.9 % 0.0 % 0.0 % 0.61 0.57 1.22 1.29
Low range of λ 81.8 % 100 % 0.0 % 2.6 % 0.62 0.31 1.23 1.26
Medium range of λ 82.2 % 100 % 1.6 % 23.5 % 0.58 0.32 1.24 1.06
High range of λ 82.6 % 100 % 69.4 % 81.3 % 0.49 0.32 0.61 0.50
Based on numerical examples it is shown that predefined processing rates have a

considerable impact on optimal costs and the penalty for investing in more than one
machine is analyzed when the processing rate is a continuous variable. A simple
solution heuristic for the mixed integer problem of capacity investment and WAW
optimization with a predefined set of processing rates is developed. For an experi-
mental design including 7,000 test instances the heuristic shows close to optimal
results with negligible extra costs. Additionally, it is found that uncertainty about
order arrival rates prior to the investment decision leads to significant additional
costs due to higher capacity investment. If additional machines of the same type can
be added after observing the real order arrival rate, smaller machines become more
attractive as the range of uncertainty increases.
Appendix
Proof of Proposition 5.1. For any multi-machine production system with

parameters ðfsi g; fμigÞ, a single
1 machine production system with the same capacity
costs and parameters f1g; μi and μi ¼ si μi exists. The superscript 1 indicates the
1
system with a single machine at each stage. The WIP, FGI, and backorder costs in the
single-machine system have to be smaller than in the parallel-machine system:
X
n
~ i λ cy;i cy;iþ1 þ E W
E W ~ n λcc þ E½ I λ cf þ cc
i¼1
X
n
! X
n 1
Xð1 FL ðXÞÞλcc þ ci μi cμ;i > ~ λ cy;i cy;iþ1
E W i
i¼1 i¼1

X
n
~ 1 λcc þ E I 1 λ cf þ cc Xð1 FL ðXÞÞλcc þ
þE W μ1i cμ;i
i
i¼1
X
n !
, E W ~ n λcc þ E½ I λ cf þ cc >
~ i λ cy;i cy;iþ1 þ E W
i¼1
X
n 1 1
~ λ cy;i cy;iþ1 þ E W
E W ~ λcc þ E I 1 λ cf þ cc
i i
i¼1
! 1 !
~i > E W
,E W ~ and E½ I > E I 1 (5.21)
i
Appendix 93

Condition E W ~ i > E½W
~ 1 is fulfilled with condition (5.5). It remains to show that
i
E½ I > E½I 1 holds. Applying Eq. (5.2) leads to:
!
E½ I > E I 1
ðX ðθ ð ðX
1
!
, fW~ n ðτÞðθ τÞdτfL ðθÞdθ þ fW~ n ðτÞðX τÞdτfL ðθÞdθ >
0 0 X 0
ðX ðθ ð ðX
1
fW~n1 ðτÞðθ τÞdτfL ðθÞdθ þ fW~n1 ðτÞðX τÞdτfL ðθÞdθ

0 0 X 0
0θ 1
ð ðX ðθ
, @ fW~ n ðτÞðθ τÞdτ fW~n1 ðτÞðθ τÞdτAfL ðθÞdθ
0 0 0
0X 1
ð
1 ð ðX
!
þ @ f ~ ðτÞðX τÞdτ f ~ 1 ðτÞðX τÞdτAfL ðθÞdθ > 0 (5.22)
Wn Wn
X 0 0
by partial integration it follows:
ðθ θ
ðθ
θ !
, FW~ n ðτÞðθ τÞ 0
FW~ n ðτÞdτ > FW~n1 ðτÞðθ τÞ FW~n1 ðτÞdτ 8θ > 0
0
0 0
ðθ ðθ
, FW~n1 ðτÞFW~ n ðτÞdτ 0 8θ > 0 and 9θ : FW~n1 ðτÞFW~ n ðτÞdτ > 0 (5.23)
0 0
which is the definition of second order stochastic dominance.

Proof of Proposition 5.2. Assume a set of different processing rates in a
manufacturing system fμð1Þ μð2Þ . . . μðnÞg. From the structure of
Eqs. (5.2, 5.3, and 5.4) it follows that backorder and FGI costs are independent of
the sequence in which these processing rates are allocated to the stages. Therefore,
these costs can be omitted for the proof. Comparing any unordered system, e.g.
μ1 ¼ μð2Þ < μ2 ¼ μð1Þ and fμ3 ¼ μð3Þ; μ4 ¼ μð4Þ; . . . ; μn ¼ μðnÞg to the
corresponding ordered system fμ1 ¼ μð1Þ μ2 ¼ μð2Þ . . . μn ¼ μðnÞg,
leads to the following cost inequality:
!
Cunordered > Cordered
with
Cunordered : μ1 ¼ μð2Þ < μ2 ¼ μð1Þ and fμ3 ¼ μð3Þ; μ4 ¼ μð4Þ; . . . ; μn ¼ μðnÞg
Cordered : fμ1 ¼ μð1Þ; μ2 ¼ μð2Þ; . . . ; μn ¼ μðnÞg
, E½W1 ðμð2ÞÞλcy;1 þ E½W2 ðμð1ÞÞλcy;2 þ ξðμð2ÞÞcμ;1 þ ξðμð1ÞÞcμ;2
!
> E½W1 ðμð1ÞÞλcy;1 þ E½W2 ðμð2ÞÞλcy;2 þ ξðμð1ÞÞcμ;1 þ ξðμð2ÞÞcμ;2
, (5.24)
ξðμð1ÞÞ !
1Þ cμ;2 cμ;1 > cμ;2 cμ;1 (5.25)
ξðμð2ÞÞ
which is fulfilled with μð1Þ > μð2Þ and cμ;2 cμ;1 . And
cy;1 !
2Þ ðE½W1 ðμð2ÞÞ E½W1 ðμð1ÞÞÞ > E½W2 ðμð2ÞÞ E½W2 ðμð1ÞÞ (5.26)
cy;2
which is fulfilled with
E½W1 ðμð2ÞÞ E½W1 ðμð1ÞÞ ¼ E½W2 ðμð2ÞÞ E½W2 ðμð1ÞÞ

(5.27)
and cy;1 cy;2
Proof of Proposition 5.4. Restating the costs from Eq. (5.1) and applying Eq. (5.3)
leads to:
X
n
CðX; fsi g; fμi gÞ ¼ E½Wi ðsi ; μi Þλ cy;i þ cc
i¼1

þ E½I ðX; fsi g; fμi gÞλ cf þ cc
ðX X
n
ð1 FL ðτÞÞdτλcc þ si μi cμ;i (5.28)
i¼1
0
For any sk ¼ s~ from fsi g the following cost delta when increasing s~ by
1 follows:
Appendix 95
ΔCðs~; Þ ¼ CðX; fsi 6¼ sk g; s~ þ 1; fμi gÞ CðX; fsi 6¼ sk g; s~; fμi gÞ

¼ ðE½Wk ðs~ þ 1; μk Þ E½Wk ðs~; μk ÞÞλ cy;k þ cc

þ ðE½I ðX; fsi 6¼ sk g; s~ þ 1; fμi gÞ E½I ðX; fsi 6¼ sk g; s~; fμi gÞÞλ cf þ cc
þ μk cμ;k
(5.29)
which can be rewritten as:

ΔCðs~; Þ ¼ ðE½Wk ðs~ þ 1Þ E½Wk ðs~ÞÞλ cy;k þ cc

þ ðE½I ðfsi 6¼ sk g; s~ þ 1Þ E½I ðfsi 6¼ sk g; ~sÞÞλ cf þ cc þ μk cμ;k
01 1
ð ð
1

¼ @ τfW ðτ; s~ þ 1; Þdτ τfW ðτ; s~; ÞdτAλ cy;k þ cc
0 0
0X
ð
þ @ FW ðτ; ðs~ þ 1; ÞÞð1 FL ðτÞÞdτ
0
1
ðX

FW ðτ; ðs~; ÞÞð1 FL ðτÞÞdτAλ cf þ cc þ μk cμ;k
0
ð
1

¼ ðFW ðτ; s~ þ 1; Þ FW ðτ; s~; ÞÞdτλ cy;k þ cc
0
ðX

þ FW ðτ; ðs~ þ 1; ÞÞ FW ðτ; ðs~; ÞÞð1 FL ðτÞÞdτλ cf þ cc þ μk cμ;k
0
(5.30)
with cy;k ¼ cf ¼ cy
ð
1

¼ ðFW ðτ; s~ þ 1; Þ FW ðτ; s~; ÞÞdτλ cy þ cc
0
ðX

þ FW ðτ; ðs~ þ 1; ÞÞ FW ðτ; ðs~; ÞÞð1 FL ðτÞÞdτλ cy þ cc þ μk cμ;k (5.31)
0
which increases in s~ since E½Wk ðsk ; μk Þ decreases in sk. For an M/M/s queue
this property is shown by Grassmann (1983) who proves a convex decrease of
E½Wk ðsk ; μk Þ with respect to sk. However, it holds for any production system, since
an additional machine can never increase production lead time and whenever the
additional machine processes a job while others are waiting, the waiting time of
the other jobs decreases. The serial setting of more than one processing stage does
not change this property.
Derivation of Equation (5.11). The cdf of the overall production lead time is:
ðτ
FW ðτÞ ¼ fW ðθÞdθ
0
ðτ

¼ t1 eμ1 θ eμ2 θ t2 eðs2 μ2 λÞθ eμ1 θ
0
t3 eðs1 μ1 λÞθ eμ2 θ þ t4 eðs1 μ1 λÞθ eðs2 μ2 λÞθ dθ

1 eμ1 τ 1 eμ2 τ 1 eðs2 μ2 λÞτ 1 eμ1 τ 1 eðs1 μ1 λÞτ
¼ t1 t1 t2 þ t2 t3
μ1 μ2 ðs2 μ 2 λ Þ μ1 ðs1 μ 1 λ Þ
1 eμ2 τ 1 eðs1 μ1 λÞτ 1 eðs2 μ2 λÞτ
þ t3 þ t4 t4
μ2 ðs1 μ1 λÞ ðs2 μ 2 λ Þ
1 eμ1 τ 1 eμ2 τ 1 eðs2 μ2 λÞτ
¼ ðt1 þ t2 Þ þ ðt3 t1 Þ ðt2 þ t4 Þ
μ1 μ2 ðs2 μ2 λÞ
1 eðs1 μ1 λÞτ
þ ðt4 t3 Þ
ðs1 μ 1 λ Þ
ðm1 þ o1 Þðm2 þ o2 Þ ðm1 þ o1 Þo2
with t1 ¼ ; t2 ¼ ;
ðμ2 μ1 Þ ðμ1 s2 μ2 þ λÞ
ðm2 þ o2 Þo1 o1 o2
t3 ¼ ; t4 ¼
ðμ2 s1 μ1 þ λÞ ðs2 μ2 s1 μ1 Þ
(5.32)
From Eq. (5.32) the expected FGI lead time can be calculated as:
ðX
E½ I ¼ FW ðτÞð1 FL ðτÞÞdτ
0
ðX
1 eμ1 τ 1 eμ2 τ 1 eðs2 μ2 λÞτ
¼ ðt1 þ t2 Þ þ ðt3 t1 Þ ðt2 þ t4 Þ
μ1 μ2 ðs2 μ 2 λ Þ
0
1 eðs1 μ1 λÞτ βτ
þðt4 t3 Þ e dτ
ðs1 μ 1 λ Þ
ðX
eβτ eðμ1 þβÞτ eβτ eðμ2 þβÞτ
¼ ðt1 þ t2 Þ þ ðt3 t1 Þ
μ1 μ2
0
eβτ eðs2 μ2 λþβÞτ eβτ eðs1 μ1 λþβÞτ
ðt2 þ t4 Þ þ ðt 4 t 3 Þ dτ (5.33)
ðs2 μ 2 λ Þ ðs1 μ1 λÞ
Appendix 97

ðt1 þ t2 Þ ðt3 t1 Þ ðt2 þ t4 Þ ðt4 t3 Þ 1 eβX
¼ þ þ

μ1 ðμ 1 þ β Þ μ2 ðμ2 þ βÞ
þ (5.34)
ðs2 μ2 λÞ ðs2 μ2 λ þ βÞ ðs1 μ1 λÞ ðs1 μ1 λ þ βÞ
Derivation of Expressions for Two-Stage M/M/s Production System with Identical

Processing Rates and Identical Number of Machines. For the calculations in
Sect. 5.4, the expressions for expected WIP, FGI, and backorders are derived for
a two-stage M/M/s production system with identical processing rates and identical
number of machines at both stages.
With μ1 ¼ μ2 ¼ μ; s1 ¼ s2 ¼ s; ρ1 ¼ ρ2 ¼ ρ it follows m1 ¼ m2 ¼ m; o1 ¼ o2 ¼ o
and the pdf of the overall production lead time is:
ðt
fW ðtÞ ¼ ðm1 þ o1 Þðm2 þ o2 Þeðμ2 μ1 Þτμ2 t ðm1 þ o1 Þo2 eðs2 μ2 þλÞtþðs2 μ2 λμ1 Þτ
0
ðm2 þ o2 Þo1 eμ2 tþðμ2 s1 μ1 þλÞτ þ o1 o2 eðs2 μ2 c1 μ1 Þτþðλs2 μ2 Þt dτ

with m1 ¼ m2 ¼ m; o1 ¼ o2 ¼ o; μ1 ¼ μ2 ¼ μ; s1 ¼ s2 ¼ s; ρ1 ¼ ρ2 ¼ ρ
ðt
) fWbalanced ðtÞ ¼ ðm þ oÞ2 eμt ðm þ oÞoeðsμþλÞtþðsμλμÞτ
0
μtþðμsμþλÞτ
ðm þ oÞoe þ o2 eðsμsμþλλÞτþðλsμÞt dτ
ðt
¼ ðm þ oÞ2 eμt ðm þ oÞoeðsμþλÞtþðsμλμÞτ
0
ðm þ oÞoeμtþðμsμþλÞτ : þ o2 eðλsμÞt dτ
eðλsμÞt eμt
¼ ðm þ oÞ2 teμt þ o2 teðλsμÞt ðm þ oÞo
ðμ sμ þ λÞ
eðλsμÞt eμt
ðm þ oÞo
ðμ sμ þ λÞ
eðsμλÞt eμt
¼ ðm þ oÞ2 teμt þ o2 teðsμλÞt 2ðm þ oÞo (5.35)
ðμ sμ þ λÞ
Based on the pdf of the overall production lead time its cdf is:
ðτ
FW ðτÞ ¼ fW ðθÞdθ
0
ðτ
eðsμλÞθ eμθ
¼ ðm þ oÞ2 θeμθ þ o2 θeðsμλÞθ 2ðm þ oÞo dθ
ðμ sμ þ λÞ
0
1 eμτ ð1 þ μτÞ 1 eðsμλÞτ ð1 þ ðsμ λÞτÞ
¼ ð m þ oÞ 2 þ o2
μ 2
ðsμ λÞ2
1 eðsμλÞτ 1 eμτ
2ðm þ oÞo þ 2ðm þ oÞo (5.36)
ðμ sμ þ λÞðsμ λÞ ðμ sμ þ λÞμ
From Eqs. (5.5) and (5.2) the expected FGI lead time is:
ðX
E½Ibalanced ¼ FW ðτÞð1 FL ðτÞÞdτ
0
ðX
1 eμτ ð1 þ μτÞ 21e
ðsμλÞτ
ð1 þ ðsμ λÞτÞ
¼ ð m þ oÞ 2 þ o
μ2 ðsμ λÞ2
0
1 eðsμλÞτ 1 eμτ
2ðm þ oÞo þ 2ðm þ oÞo eβτ dτ
!
ð m þ oÞ 2 o2 2ðm þ oÞo 2ðm þ oÞo 1 eβX
¼ þ þ
μ 2
ðsμ λÞ 2 ðμ sμ þ λÞðsμ λÞ ðμ sμ þ λÞμ β
!
ð m þ oÞ 2 2ðm þ oÞo 1 eðμþβÞX ðm þ oÞ2 1 eðμþβÞX
þ
μ2 ðμ sμ þ λÞμ ðμ þ β Þ μ ðμ þ βÞ2
ðm þ oÞ2 eðμþβÞX X o2 1 eðsμλþβÞX ð1 þ ðsμ λ þ βÞXÞ
þ
μ ðμ þ β Þ ðsμ λÞ ðsμ λ þ βÞ2
!
2ðm þ oÞo o2 1 eðsμλþβÞX
þ
ðμ sμ þ λÞðsμ λÞ ðsμ λÞ2 ðsμ λ þ βÞ
(5.37)
and the optimality condition (5.14) becomes:
cc 1 eμX ð1 þ μXÞ 21e

ðsμλÞX
ð1 þ ðsμ λÞXÞ
X : ¼ ð m þ oÞ 2 þ o
cf þ cc μ 2
ðsμ λÞ2
1 eðsμλÞX 1 eμX
2ðm þ oÞo þ 2ðm þ oÞo
(5.38)
Chapter 6
Service Level Constraint Models
In this chapter the problem of capacity investment and WAW setting is discussed in
a service level constraint context. Since in production systems it is often difficult to
identify a cost factor for backorders, a service level target can be defined by
management as a strategic decision. To be able to use the models developed in
Chaps. 4 and 5 in such an environment, an extension of these models to include a
service level constraint is developed. For the single-stage M/M/1 model, an explicit
expression for the optimal WAW and the optimality condition for the optimal
processing rate are derived. For the two-stage M/M/s model a service level equation
is provided to enable its application in such settings. Some additional numerical
examples are provided which show that the results of the service level constraint
models are similar to the backorder cost models. Additionally, a capacity invest-
ment optimization model for a multi-item, single-stage production system with
normally distributed demand is developed in this chapter. The capacity investment
is optimized under a service level constraint including inventory holding costs.
The remainder of this chapter is structured as follows. The single-stage M/M/1
model is extended to the service level setting in Sect. 6.2. In Sect. 6.3 the multi-
item, normally distributed demand, single-stage production system is modeled.
This model is the simplified case from Jodlbauer and Altendorfer (2010) without
setup times. The calculation of the service level and the optimization problem setup
for the two-stage model from Chap. 5 is presented in Sect. 6.4. Some concluding
remarks are stated in Sect. 6.5 and all proofs are added in the Appendix.

100 6 Service Level Constraint Models
6.2 Single-Stage M/M/1 Model
The single-stage model as presented in Chap. 4 is designed to minimize the overall

costs for capacity, WIP, FGI, and backorders. The parameters to optimize are
capacity available and WAW. In this section the model is transformed to fulfill a
service level constraint instead of including the backorder costs.
6.2.1 Model Development
The optimization problem stated in Chap. 4 is (see Eq. (4.12)):
CðX; μÞ ¼ cy E½Y þ cμ μ þ cf E½G þ cc E½ B ! min (6.1)

X;μ
Whereby cy, cf , cc, and cμ are WIP, FGI, backorder, as well as capacity costs and
Y, G, B, as well as μ are WIP, FGI, backorders, and processing rate (i.e. capacity
invested) respectively. Transforming the backorder costs to a service level con-
straint leads to the following optimization problem (applying Eq. (3.9) for service
level η):
CðX; μÞ ¼ cy E½Y þ cμ μ þ cf E½G ! min

μ;X
ðμ λÞeðμλþβÞX
þβ (6.2)
w:r:t: η ¼ 1 ~η
ðμ λ þ β Þ
With λ, 1=β, and X being the order arrival rate, mean customer required lead time,
and WAW respectively. E½Y and E½G depend on the WAW (see Eqs. (3.1, 3.4, and
3.10)) and ~η is the service level constraint. Based on Eq. (3.9) the following optimal
WAW can be stated with respect to the binding constraint on ~η:

1 β
X ¼ ln 1 ~η 1 þ (6.3)
k¼μλ kþβ k

β
Equation (6.3) only holds true for ~η 1 kþβ which is the maximum possible
service level for the single-stage M/M/1 production system (corresponds to X ¼ 1)
6.2 Single-Stage M/M/1 Model 101
from Eq. (3.6). One finding from Eq. (6.3) is that in the service level constraint
model, the optimal WAW depends on the customer required lead time distribution
which is not the case in the backorder cost model. To identify the MTO ability of a
production system, Eq. (3.6) can be restated as:
β~η
μλþ (6.4)
1 ~η
Condition (6.4) is a measure of the MTO ability of a production system. For any
predefined service level and mean customer required lead time, (6.4) states (for the
assumptions as provided in Chap. 3) how much capacity is at least needed to fulfill
the service level constraint and for this reason to be able to work under the MTO
regime.
The service level constraint optimization problem can be compared to the
backorder cost problem applying Eqs. (6.3) and (4.14) for optimal WAW. When
both models lead to the same optimal WAW for an equal available capacity, the
relationship between backorder costs and FGI costs follows as:

Xservice ¼ Xbackorder

1 β 1 cf þ cc
, ln 1 ~η 1 þ ¼ ln
μλþβ μλ μλ cf

μλ β cf þ cc (6.5)
, ln 1 ~η 1 þ ¼ ln
μλþβ μλ cf
μλþβ
μλ
β cc
, 1 ~η 1 þ 1¼ :
μλ cf
From Eq. (6.5) it follows that a transformation between the backorder cost and
the service level constraint model solely based on the ratio cc =cf independent of
the processing rate μ is not possible. However, for any predefined μ, a backorder
cost ratio cc =cf can be determined which is consistent with inventory management
theory; see Axsäter (2006) who provides some examples where backorder costs
can be transformed to a service level constraint for exogenous replenishment lead
time. Figure 6.1 shows the range of the cost ratio cc =cf to reach the same optimal
WAW in both models. This ratio is even in this single-machine model highly
dependent on the utilization, i.e. it is not possible to state a ratio which leads to
approximately the same results for both models over a considerable utilization
range. The ratio is unbounded for both service level and utilization increase.
100 100
90 90
80 80
70 70
Cost ratio cc/cf

Cost ratio cc/cf
60 60
50 50
40 40
30 30
20 20
10 10
0 0
0.5 0.6 0.7 0.8 0.9 0.8 0.85 0.9 0.95 1
Utilization Service level
β = 0.005; λ = 0.5; η = 0.95 β = 0.005; λ = 0.5; ρ = 0.8
Fig. 6.1 Backorder to FGI cost ratio for equal WAW in single-stage M/M/1 model
Proposition 6.1. To minimize the capacity, WIP, and FGI costs under service level
constraint ~η (optimization problem from Eq. (6.2)) for an M/M/1 production system
with exponentially distributed customer required lead time, the first-order condi-
tion for optimal processing rate μ is given by:
cμ ~ηβð2μ 2λ þ βÞ þ ~ηðμ λÞ2 cy

þ cf
λ 2
ðμ λ þ β Þ ðμ λ Þ 2
ðμ λÞ2
2
μλþβ
β
β 4 ~ηβ
¼ cf 1 ~η 1 þ (6.6)
ðμ λ Þ ðμ λ þ βÞ ð1 ~ηÞðμ λÞ2 ~ηβðμ λÞ
#
ln½1 ~ηð1 þ β=ðμ λÞÞ

ðμ λ þ β Þ2
Proof see Appendix.

This equation can be solved numerically for μ. Even though it was not possible to
proof the existence and uniqueness of its solution, it existed and was unique in all
the numerical examples conducted in this section (see also Sect. 6.2.2).
6.2.2 Numerical Example
The following numerical example shows the results of this single-stage M/M/1
model. The parameters for the test examples as defined in Table 6.1 are similar to
the ones in Chap. 5.
The following Fig. 6.2a, b show the optimal costs and the optimal WAW from
Eq. (6.3) when the utilization, and for this reason the capacity invested μ, is predefined.
6.2 Single-Stage M/M/1 Model 103
Table 6.1 Numerical Example ~

η cμ β
examples for M/M/1 model
with service level constraint A 0.95 1,000 0.005
B 0.95 500 0.005
C 0.95 1,000 0.0075
D 0.97 1,000 0.005
cf ¼ 10; cy ¼ 5; λ ¼ 1;
a b
2000 60
1800 50
Optimal WAW
1600
Overall costs
40
1400
30
1200
20
1000
800 10
600 0
0.5 0.6 0.7 0.8 0.9 0.5 0.6 0.7 0.8 0.9
Example A Example B Example C Example D Example A Example B Example C Example D
Fig. 6.2 Costs and WAW in single-stage M/M/1 model with service level constraint
For Example C and D, Fig. 6.2a shows that the maximum utilization to reach the
targeted service level is 87.5 % and 86 % respectively which is indicated by the ends
of the lines (see also condition (6.4) for minimum processing rate needed). In Example
A and B the maximum utilization to reach the targeted service level is above 90 %.
Comparing Example B with Example A shows the intuitive result that a lower
capacity cost cμ leads to lower overall costs and a lower optimal utilization (84 %
in Example A and 80 % in Example B). When the mean customer required lead time
decreases, Example C shows that optimal costs as well as optimal WAW increases
and that optimal utilization decreases. The strong influence of the service level
constraint on the optimal costs, optimal utilization, and optimal WAW is illustrated
by Example D. It indicates that capacity invested and WAW, i.e. FGI, increase when
the service level target is tighter. The comparison of Example A and B in Fig. 6.2b
shows that the optimal WAW is independent of the capacity costs (see also Eq. (6.3)).
A sensitivity analysis is conducted and Fig. 6.3 visualizes the influence of the
parameters mean customer required lead time, capacity costs, FGI holding costs,
and service level constraint on the optimal capacity invested and the optimal WAW
(whereby all the other parameters are set to the values of Example A).
An increase in the mean customer required lead time leads to an increase
in optimal WAW and a decrease in optimal capacity invested (see Fig. 6.3a).
Figure 6.3b illustrates the influence of a capacity cost increase and shows the
intuitive result that in this case the optimal capacity invested decreases and the
optimal WAW increases. The influence of FGI holding costs is visualized in
Fig. 6.3c. Whenever FGI holding costs increase in comparison to WIP and capacity
a b
1.6 25 1.6 25
Optimal capacity invested

1.4 1.4
20 20
1.2 1.2
Optimal WAW
Optimal WAW
1 15 1 15
0.8 0.8
0.6 10 0.6 10
0.4 0.4
5 5
0.2 0.2
0 0 0 0
0 200 400 600 800 1000 0 200 400 600 800 1000
Customer required lead time Capacity costs
Optimal capacity invested Optimal WAW Optimal capacity invested Optimal WAW
c d
1.6 25 1.6 25

1.4 1.4
20 20
1.2 1.2
Optimal WAW
Optimal WAW
1 15 1 15
0.8 0.8
0.6 10 0.6 10
0.4 0.4
5 5
0.2 0.2
0 0 0 0
0 10 20 30 40 50 0.6 0.7 0.8 0.9 1
FGI holding costs Service level constraint
Fig. 6.3 Sensitivity analysis of single-stage M/M/1 model with service level constraint
costs, the WAW and for this reason the average FGI decreases which is
compensated by an increased capacity investment to still fulfill the service level
constraint. An increase in the service level constraint leads, according to Fig. 6.3d,
to an increase in capacity invested. However, the WAW reaches a maximum at a
service level constraint of 97 % (in this example) and decreases thereafter. This
means that for very high service level targets it is better to invest more money in
capacity and less in FGI than for moderate service level targets.
6.3 Single-Stage, Multi-Item, Normally Distributed

Demand Model
In this section which is based on Jodlbauer and Altendorfer (2010), a service level
constraint, single-stage, multi-item capacity investment model considering inven-
tory costs is developed for normally distributed demand applying an earliest due
date (EDD) dispatching rule.
6.3 Single-Stage, Multi-Item, Normally Distributed Demand Model 105
Fig. 6.4 Multi-item, Item dependent customer

normally distributed demand, required lead time = Li
single-stage production Customer
Service level
system h = P(D(n) <= µ)
Buffer Buffer
EDD
Machine
Orders FGI
Inventory = Y
6.3.1 Model Assumptions and Notation
The production system as shown in Fig. 6.4 is discussed.

For each item i 2 1; . . . ; N, the random customer demand in pieces Ai ðnÞ, which
is independent of other item demands, follows a Gauss Process (see Ross (2010) or
Beichelt (2006) for details on Gauss Processes) with E½Ai ðnÞ ¼ λi n and Var ½Ai ðnÞ
¼ σ 2i n which states that the demand for each item in any arbitrary time period n is
normally distributed. λi and σ 2i denote the mean demand rate and the demand
variance of item i respectively. The applicability of the normal distribution for item
demands is discussed in Bish et al. (2005) showing that the probability of negative
values is negligible. Each piece produced needs an item dependent deterministic
capacity qi (in time units) from the single processing stage leading to the random
PN
capacity demand DðnÞ ¼ Ai ðnÞqi which is again normally distributed for any
i¼1
P
N P
N
arbitrary time period n with E½DðnÞ ¼ λi qi n and Var ½DðnÞ ¼ ðσ i qi Þ2 n. Let
i¼1 i¼1
P
N P
N
λ ¼ λi qi denote the capacity demand rate and σ ¼ 2
ðσ i qi Þ2 the capacity
i¼1 i¼1
demand variance, then E½DðnÞ ¼ λn and Var ½DðnÞ ¼ σ n holds. fDðnÞ ðÞ and 2
FDðnÞ ðÞ denote the pdf and cdf of the random variable DðnÞ respectively.
Nothing is produced without an order and each piece ordered has a random
non-negative, item dependent customer required lead time Li (in time units)
which is the difference between order time and due date. This definition
implies that each order consists of one piece of any item i. Li is assumed
to be lognormally distributed (see also Jodlbauer (2008b) for this assumption).
L denotes the random customer required lead time of one capacity unit (one
time unit needed from the machine) and is the capacity demand weighted
PN
sum of all Li values. It has a mean of E½L ¼ E½Li qi λi =λ and a variance of
i¼1
P
N P
N
Var ½L ¼ 2
Var ½Li ðqi λi Þ = 2
ðqi λi Þ and follows a lognormal distribution as
i¼1 i¼1
well. fLðÞ and FLðÞ denote the pdf and cdf of the random variable L
respectively. All orders are processed according to the EDD dispatching rule.
The service level η denotes the proportion of capacity which is delivered on time.
Such a capacity oriented service level is useful whenever the customer perceived
utility depends on the capacity needed for an order, which can, for example, be the
case in machine intensive industries. This capacity oriented service level is equal to
the on-time probability of orders if the capacity needed qi of all items i is identical.
The orders are released to the production system whenever their remaining time to
the due date is smaller or equal to the WAW X. W denotes the random overall lead
time of an order which is the time it takes from release to the production system
until shipment to customer, i.e. leaving the FGI buffer. The processing stage is
assumed to have a constant minimum required lead time P (in time units), which is
the sum of required processing and transportation time needed to finalize an order
and to deliver it to the customer (see Jodlbauer (2008b) for details on P). Figure 6.5
shows the overall lead time assumption. According to Fig. 6.5, capacity units
ordered with capacity oriented customer required lead times L > X are released
to production exactly at X time before due date. Therefore, the overall lead time is
equal to X for these capacity units ordered. Capacity units ordered with capacity
oriented customer required lead times L < P have at least an overall lead time of P
and cannot be fulfilled on time which leads to a basic service level loss ηb. All other
capacity units ordered with capacity oriented customer required lead times P L
X are released to production at the time when the orders are received and
therefore the overall lead time equals the capacity oriented customer required
lead time.
Therefore, W follows as:
¼ maxðP; minðX; LÞÞ

W (6.7)
Since the effect that orders not delivered on time have longer overall lead times
is neglected here, Eq. (6.7) only holds for production systems with high service
levels. Since the system studied is an MTO system it is assumed that mostly high
service levels are requested by the customers and therefore Eq. (6.7) holds. The
random inventory, which is measured in capacity units in this model, is denoted by

Y.
It is assumed that the capacity processing rate μ, which defines the capacity units
(in time units) to be processed per time unit, is a decision variable in the model and
can be determined by investment with cost parameter cμ . The following optimiza-
tion problem minimizing capacity investment and inventory holding cost under a
service level constraint ~η is solved:
; XÞ ¼ E½Ych þ μ
Cðμ cμ ! min
μ
;X
(6.8)
w:r:t: η ~η
Fig. 6.5 Overall lead time in Pdf of L

normally distributed demand,
single-stage model
()
fL t
=P =X
=t
P X Capacity oriented customer

required lead time t
Overall lead time W
P X Capacity oriented customer

required lead time t
ch denotes the inventory holdings costs for one unit of capacity stored one
time unit.
6.3.2 Capacity Demand Smoothing
From the definition of the random capacity demand DðnÞ it is obvious that this DðnÞ
will for some time periods n be greater than the available capacity μ
n in this time
period. Additionally, whenever a time period n is discussed, this n can be
interpreted as an averaging period for smoothing capacity demand, since the
coefficient of variation of DðnÞ decreases with increasing n:
pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
Var ½DðnÞ σ dαDðnÞ 1σ 3
αDðnÞ ¼ ¼ pffiffiffi ) ¼ n 2 (6.9)
E½DðnÞ λ n dn 2λ
Applying this capacity smoothing period n leads to the following service level η,
which is by definition the percentage of capacity delivered on time:
FDðnÞ ðμ
nÞ ¼ η (6.10)
Note that Eq. (6.10) holds because the dispatching is EDD, so all orders are
processed according to their respective due dates. Since the production system is
studied in a stationary setting, the capacity processing rate has to be greater than the
capacity demand rate, so μ > λ has to hold. Equation (6.10) is for a predefined
service level explicitly solvable for the time-averaging period n or the capacity
processing rate μ .
Proposition 6.2. Considering normally distributed capacity demand and assuming
> λ, the time averaging period n to ensure a capacity oriented service-level η with
μ
capacity processing rate μ is given by:
!2
F1
Nð0;1Þ ðηÞ
n¼ σ2 (6.11)
λ
μ
whereby F1Nð0;1Þ ðÞ denotes the inverse of the standard normal distribution
function. Furthermore, the capacity needed to ensure a service level η with a
predefined time-averaging period n can be expressed by:
1
FNð0;1Þ ðηÞσ
¼ λ þ
μ pffiffiffi (6.12)
n
Proof see Appendix.

From Eq. (6.12) the capacity processing rate μ can be split up into one part
independent of η, which is denoted by basic capacity μb , and one part depending
on η, which is denoted by excess capacity μexcess. The basic capacity μb is equal to
the capacity demand rate and the excess capacity μexcess is needed to handle the
demand peaks:
μ
¼ μb þ μexcess
whereby
1
(6.13)
FNð0;1Þ ðηÞσ
μb ¼ λ and μexcess ¼ pffiffiffi
n
Proposition 6.2 shows that the time averaging period needed to manage tempo-
rary demand peaks, i.e. the variation in the demand process, is an increasing
function with respect to the capacity demand variance, the capacity demand rate
as well as to the service level and a decreasing function with respect to the capacity
processing rate (i.e. capacity invested).
6.3.3 WAW Calculation
The current time in the production system is set to t ¼ 0 without loss of generality.
From the definition of the capacity oriented customer required lead time, the
expected capacity demand known with due date between time P and time X can
be calculated as:
ðX 1
ð ðX
fLðτÞλdτdθ ¼ ð1 FLðθÞÞdθλ (6.14)
P θ P
The first integral in Eq. (6.14) (from P to X) corresponds to the due date of the
orders and the second integral corresponds to the orders known at time 0 with due
date θ, i.e. all orders with customer required lead time longer than θ. From Eq. (6.14)
it follows that 1 FLðθÞ is the expected value of the proportion of capacity needed
at due date θ that is already specified (with fixed orders) at time 0. Furthermore, the
expected capacity oriented customer required lead time can be expressed as:
ð
1
1 FLðτÞdτ ¼ E½L (6.15)

0

Based on the result for the averaging period n from Eq. (6.11), the WAW work
release rule can be parameterized. If all orders were known an infinite time in
advance without any capacity oriented customer required lead time distribution, the
WAW X would be the sum of the averaging period n and the minimum required
lead time P. This holds since any order needs at least P time to run through the
production system and to satisfy the service level constraint an additional time
period n has to be taken into account for demand smoothing. Figure 6.6 illustrates
the problem that not all orders within the period P to P þ n are known and for this
reason the WAW X has to be greater than P þ n.
Since the expected value for capacity demand within the demand smoothing
period is λn and a certain proportion of capacity demand with due date within period
P to P þ n is not yet known, see area A in Fig. 6.6, this unknown capacity demand
has to be compensated by orders with due dates after P þ n which is depicted as area
B in Fig. 6.6. Applying Eq. (6.14) and setting the expected value of known demand
within period P to WAW X to the value λn , which corresponds to the demand
needed for capacity smoothing, shows:
ðX ðX

ð1 FLðθÞÞdθλ ¼ λn , ð1 FLðθÞÞdθ ¼ n (6.16)
P P
Fig. 6.6 WAW calculation Proportion of

based on capacity oriented booked capacity
Capacity for unknown but expected customer orders
customer required lead time
distribution A
Capacity for known orders
1 − FL (⋅)
Capacity oriented customer required lead time

P n
An implicit equation for the WAW X is provided in Eq. (6.16) which is

equivalent to setting area A equal to area B in Fig. 6.6. Equation (6.16) has a
Ð
1
solution if and only if the time period n is shorter than 1 FLðτÞdτ. If no finite
P
WAW X is found to fulfill Eq. (6.16), it is not possible to install a pure MTO system
because not enough customer orders are known early enough.
ÐP
Approximating 1 FLðτÞdτ with P and applying Eq. (6.15) leads to a
0
simplified inequality which is assumed to hold since only MTO production systems
are discussed with this model. Note that this inequality (6.17) is slightly stricter than
ÐP
needed based on the applied approximation (since P > 1 FLðτÞdτ):
0
E½L P > nðμ

; ηÞ (6.17)
In condition (6.17) as well as in the remainder of this section, the time averaging
period n is formulated as a function with respect to the capacity processing rate μ
and the service level η, to illustrate the dependency of n on μ and η according to
Eq. (6.11).
Figure 6.6 additionally shows that for any capacity oriented customer required
lead time distribution, there is a proportion of capacity that cannot be delivered on
time because it has a customer required lead time shorter than the minimum
required lead time. This basic service level loss is:
ηb ¼ FLðPÞ (6.18)
Applying Eqs. (6.11) and (6.16) to identify the WAW X for reaching a
predefined service level η will lead to a realized service level η ηb . For the
optimization problem this means that the service level constraint ~η has to include
this effect.
6.3.4 Determination of the Required Inventory
Based on the overall lead time definition in Eq. (6.7), an equation for the expected
inventory with respect to the available capacity is yielded by applying Little’s Law
to the overall lead time.
> λ as well as E½L P > n, the mean inventory needed to ensure a service-level η
μ
with available capacity μ is given by
; ηÞÞλ
E½Y ¼ ðP þ nðμ (6.19)
whereby nðμ
; ηÞ is defined by Eq. (6.11).
Proof see Appendix.
The expected inventory needed from Eq. (6.19) can be split up (similar to
the capacity processing rate needed) into a basic inventory Yb, which is independent
of η, and a surplus inventory Ysurplus needed for handling the demand peaks:
E½Y ¼ E½Yb þ E Ysurplus

whereby
(6.20)
E½Yb ¼ Pλ
; ηÞλ
E Ysurplus ¼ nðμ
6.3.5 Optimal Capacity Invested
Applying Eqs. (6.19) and (6.13), optimization problem (6.8) can be solved. The
WAW X which is also a result of the optimization problem is calculated from
Eqs. (6.11) and (6.16) with respect to the capacity invested. As input for the
optimization problem, the management has to decide on a certain service level
target η. The service level constraint for the optimization problem is then calculated
as the sum of the service level target defined by management plus the basic service
level loss which leads to:
~η ¼ η þ ηb (6.21)
In the second stage the capacity investment is optimized based on the manage-
ment decision.
E½L P > n , the solution of the non-linear cost minimization problem from
Eq. (6.8) under a service level constraint ~η is defined by:
0 2 113
B F1
Nð0;1Þ ð~
ηÞ σ2C
¼ @2λch
μ A þ λ (6.22)
cμ
0 !23 1
F1
Nð0;1Þ ð~
ηÞσcμ
E Y ¼ @P þ
Aλ (6.23)
2λch
!23
F1
Nð0;1Þ ð~
ηÞσcμ
; ~ηÞ ¼
nð μ (6.24)
2λch
2 13
1
Þ ¼ λcμ þ 2λch FNð0;1Þ ð~ηÞcμ σ
Cðμ þ Pλch
0 !2 113
F1
Nð0;1Þ ð~
ηÞcμ σ
þ @λch A (6.25)
2
Proof see Appendix.

From Eqs. (6.22) and (6.23) the optimal excess capacity and optimal surplus
inventory can be identified by applying Eqs. (6.13) and (6.20):
0 2 113
1
F ð~ηÞ σ2 C
B Nð0;1Þ
μexcess ¼ @2λch A
cμ
(6.26)
!23
h i F1
Nð0;1Þ ð~
ηÞσcμ
λ
E Ysurplus ¼
2λch
Proposition 6.5 shows the cost minimum trade-off between excess capacity and
surplus inventory, which is an interesting managerial insight.
Proposition 6.5. Under the condition of Proposition 6.4 for the optimal choice
of capacity invested and inventory needed, the following identity holds true: The
double of the surplus inventory cost is equal to the cost of the excess capacity (see
Eq. (6.27)).
h i
μexcess cμ ¼ 2E Ysurplus

ch (6.27)
Proof see Appendix.

This very simple ratio, which may depend on the measurement of inventory in
capacity units and service level as the proportion of capacity units delivered on
time, nevertheless, provides a guideline for evaluating the current status of produc-
tion systems. It enables a quick check of optimality for real production facilities but
has always to be viewed in the light of the model assumptions stated above.
The optimal utilization, which is the ratio of capacity demand rate to capacity
invested (which is capacity processing rate), follows from Eq. (6.22) as:
λ λ 1
ρ ¼ ¼ 2 113 ¼ 13 (6.28)
0
μ 2
σ2ch
F1 ð~ηÞ σ2 2 F1 ð~η Þ þ1
@2λch Nð0;1Þ
A þ λ Nð0;1Þ λ cμ
2
cμ
The optimal utilization is a decreasing function in the service level constraint,

the variance of capacity demand, and the inventory holding cost. This means that
more excess capacity is needed for a higher targeted service level, a higher capacity
demand variance or for greater inventory holding cost to ensure cost optimality.
Furthermore, the optimal utilization is an increasing function with respect to
capacity cost which shows that in the case of greater capacity cost, it is advanta-
geous to reduce excess capacity.
The following numerical example shows the results of this multi-item, single-stage
model with normally distributed demand. The parameters for the test examples
defined in Table 6.2 are similar to the ones in Chap. 5 and in Sect. 6.2.2.
The following Fig. 6.7a shows the optimal costs when the utilization, and for this
reason the capacity processing rate μ, is predefined and the optimal time-averaging
period from Eq. (6.11) is applied. Figure 6.7b illustrates which WAW is optimal for
the respective predefined utilization values according to Eq. (6.16).
Figure 6.7a shows that the optimal costs with respect to the utilization are
identical in Example A and Example C. This comes from Eq. (6.8) for costs and
Eq. (6.19) for expected inventory which only include the time-averaging period and
a constraint for the expected customer required lead time but not the distribution of
customer required lead time. Nevertheless, the line for Example C ends at a

η cμ
E½ L
examples for normally
distributed demand model A 0.95
1,000 200
with service level constraint B 0.95 500 200
C 0.95
1,000 133
D 0.97
1,000 200
ch ¼ 10; P ¼ 1; λ ¼ 1; σ 2 ¼ λ ; Var ½L ¼ E½L
2 2
a b
3000 200
2500 175
150
Optimal WAW
Overall costs
2000
125
1500 100
75
1000
50
500
25
0 0
0.5 0.6 0.7 0.8 0.9 0.5 0.6 0.7 0.8 0.9
Example A Example B Example C Example D Example A Example B Example C Example D
Fig. 6.7 Costs and WAW in normally distributed demand model with service level constraint
utilization of 87.5 % which shows that for higher utilization values the constraint
(6.17) is no longer fulfilled and therefore higher utilization values cannot be
reached in this setting. The comparison of optimal WAW values in Fig. 6.7b
indicates that the optimal WAW for Example A and Example B is the same. The
reason is that the optimal WAW only depends on the utilization, the service level
constraint, and on the distribution of customer required lead time (see Eqs. (6.11)
and (6.16)).
Example B visualizes the intuitive result that higher capacity costs lead to a
higher optimal utilization. Example C shows that a lower mean customer required
lead time leads to a higher optimal WAW and Example D indicates that a higher
service level constraint leads to higher costs and requires a higher WAW.
A sensitivity analysis is conducted and Fig. 6.8 visualizes the influence of the
parameters mean customer required lead time, capacity costs, FGI holding costs,
and service level constraint on the optimal capacity processing rate and the optimal
WAW whereby all the other parameters are set to the values of Example A.
Figure 6.8a shows that even though the distribution of customer required lead
time is not part of the optimization problem and has, for this reason, no influence on
the optimal capacity invested, it still influences the optimal WAW since the time
averaging period, which is a result of the optimization problem, has to be
transformed to the planning parameter WAW applying its distribution (see
Eqs. (6.16) and (6.24)).
The intuitive result that optimal capacity invested decreases and optimal WAW
increases when capacity costs increase is illustrated in Fig. 6.8b. The contrary holds
a b
2 25 2 25

1.8 1.8
1.6 20 1.6 20
Optimal WAW
Optimal WAW
1.4 1.4
1.2 15 1.2 15
1 1
0.8 10 0.8 10
0.6 0.6
0.4 5 0.4 5
0.2 0.2
0 0 0 0
0 50 100 150 200 0 200 400 600 800 1000
Capacity oriented customer required lead time Capacity costs
c d
2 25 2 25

1.8 1.8
1.6 20 1.6 20
Optimal WAW
Optimal WAW
1.4 1.4
1.2 15 1.2 15
1 1
0.8 10 0.8 10
0.6 0.6
0.4 5 0.4 5
0.2 0.2
0 0 0 0
0 10 20 30 40 50 0.85 0.9 0.95 1
Inventory holding costs Service level constraint
Fig. 6.8 Sensitivity analysis of normally distributed demand model with service level constraint
for increasing inventory holding costs shown in Fig. 6.8c. However, the balance
between capacity and inventory cost is kept the same as proven in Proposition 6.5.
Figure 6.8d shows that an increase in the service level constraint leads to
increased capacity invested and an increased WAW. Since the model is only
valid for high service level targets the evaluated region starts at a service level
constraint of 85 %.
6.3.7 Summary
In this section a multi-item production system with normally distributed demand

and random due dates is considered. An optimization problem identifying the
optimal capacity investment under a capacity service level constraint minimizing
capacity and inventory holding costs is solved. It is shown that the inventory needed
to satisfy a certain service level is a decreasing and convex function with respect to
the invested capacity. Furthermore, an increase in targeted service level and mean
capacity demand leads to a higher inventory needed for predefined capacity. The
cost function with respect to the capacity invested is convex and for the optimal
capacity invested the following characteristic is gained: Double the surplus inven-
tory cost equals the excess capacity cost.
The characteristic of the optimal solution can be used to analyze the relationship
between capacity and inventory. If the excess capacity cost is much greater than
double the surplus inventory cost, divesting the capacity will improve the cost
structure, and if the excess capacity cost is much smaller than double the surplus
inventory cost, an increase in available capacity will improve the cost structure.
6.4 Two-Stage M/M/s Model
In this section the service level equation for a two-stage M/M/s production system
applying a WAW work release rule with exponentially distributed customer
required lead time is developed. Based on this service level equation the same
optimization problem as stated in Sect. 6.2 can be stated for this two-stage setting.
A numerical example is provided, showing the influence of predefined processing
rates on the optimal cost and optimal number of machines similar to the numerical
example from Sect. 5.4.1. It is shown that the general findings from Sect. 5.4.1 do
not change when the backorder costs are replaced by a service level constraint. All
variables used in Sect. 6.4 are defined in Chap. 5.
6.4.1 Optimization Problem with Service Level Constraint
Since in this section the two-stage problem is evaluated similarly to Chap. 5 with a
WAW work release policy, the service level Eq. (3.9) can be combined with the cdf
of the production lead time from Chap. 5 (see Eqs. (5.32) and (5.36)). The following
service level equation can be calculated for non-identical processing rates:
ðX ð
1
ðβτÞ
η ¼ FW ðτÞβe dτ þ FW ðXÞβeðβτÞ dτ
0 X
ðt1 þ t2 Þ 1 eðμ1 þβÞX ðt3 t1 Þ 1 eðμ2 þβÞX ðt2 þ t4 Þ 1 eðs2 μ2 λþβÞX

¼ þ
ðμ1 þ βÞ ðμ2 þ βÞ ðs2 μ 2 λ þ β Þ
ðt4 t3 Þ 1 eðs1 μ1 λþβÞX
þ
ðs1 μ 1 λ þ β Þ
(6.29)
For identical processing rates and equal number of machines at both stages the
following holds:
6.4 Two-Stage M/M/s Model 117
ðX ð
1
ðβτÞ
ηbalanced ¼ FW ðτÞβe dτ þ FW ðXÞβeðβτÞ dτ
0 X
ðmðsμ λ þ βÞ oðμ sμ þ λÞÞ2 ðm þ oÞ2 eðμþβÞX ðm þ oÞ2 XeðμþβÞX

¼
2
ðμ þ βÞ ðsμ λ þ βÞ 2
ðμ þ β Þ 2 ðμ þ βÞ
ðsμλþβÞX 2 ðsμλþβÞX
o Xe
2
o e 2ðm þ oÞoeðμþβÞX

ðsμ λ þ βÞ ðsμ λ þ βÞ2 ðμ þ βÞðμ sμ þ λÞ
2ðm þ oÞoeðsμλþβÞX
þ
ðsμ λ þ βÞðμ sμ þ λÞ
(6.30)

Restating the optimization problem from Chap. 5 (see Eq. (5.1)) in the same way
as in Sect. 6.2 leads to the following service level constraint optimization problem
minimizing costs for capacity, WIP, and FGI by searching for the optimal number
of machines for predefined processing rate with μ1 ¼ μ2 ¼ μ and s1 ¼ s2 ¼ s:
Cðs; XÞ ¼ cy E½Y þ cμ sμ þ cf E½G ! min

s;X
(6.31)
w:r:t: η ~η
Whereby E½Y and E½G can be calculated according to Eqs. (5.10) and (5.34)
respectively. Based on the findings of Sect. 6.2 it is conjectured that the service
level constraint is binding and for this reason Eq. (6.30) implicitly defines the WAW
X. The optimization problem can be solved for s numerically and a similar problem
can easily be stated for the case with non-identical processing rates, a continuous
range of processing rates, and a predefined set of possible processing rates.
In this section, a numerical example, similar to the one presented in Sect. 5.4.1, shows
the influence of predefined processing rates in the service level constraint model. The
different examples are defined in Table 6.3. Example A is the basic scenario and
Example B shows a situation with low capacity costs. The influence of a lower mean
customer required lead time is included in Example C and Example D shows the
effect of a higher targeted service level. Example A and Example B directly corre-
spond to the examples discussed in Sect. 5.4.1 (only cy;1 ¼ cy;2 is assumed here). In
Example C a lower decrease in mean customer required lead time than in Sect. 5.4.1
is tested. Example D in Sect. 5.4.1 corresponds to an increase in backorder costs,
which is transformed to an increase in the service level target in this section.
Figure 6.9 illustrates optimal costs and the optimal number of machines.

η cf cy cμ λ β
examples for overall costs in
two-stage M/M/s service level A 95 % 10 5 1,000 1 0.005
constraint model B 95 % 10 5 250 1 0.005
C 95 % 10 5 1,000 1 0.075
D 97 % 10 5 1,000 1 0.005
Ex A) Ex B)


20 20

18 1800 18
4500
16 1600 16
14 1400 14
Overall costs
Overall costs
4000
12 1200 12
3500 10 1000 10
8 800 8
3000
6 600 6
4 400 4
2500
2 200 2
2000 0 0 0
0.1 0.3 0.5 0.7 0.9 1.1 1.3 0.1 0.3 0.5 0.7 0.9 1.1 1.3
Ex C) Ex D)
20


20

5500 18 5500 18
16 16
5000 5000
14 14
Overall costs
Overall costs
4500 12 4500 12
4000 10 4000 10
3500 8 3500 8
6 6
3000 3000
4 4
2500 2 2500 2
2000 0 2000 0
0.1 0.3 0.5 0.7 0.9 1.1 1.3 0.1 0.3 0.5 0.7 0.9 1.1 1.3
Fig. 6.9 Optimal costs dependent on machine size in two-stage service level constraint model
The shape of the curves created with this service level constraint model is
exactly the same as in the backorder costs model from Sect. 5.4.1. Example B
shows, similar to the backorder cost model, that lower capacity costs lead to a lower
range of optimal costs over the predefined processing rate. Example C illustrates
that for a tighter mean customer required lead time the optimal number of machines
is greater than or equal to the basic scenario. An increase in the service level
constraint has qualitatively the same influence on the optimal costs and the optimal
number of machines as the reduction of mean customer required lead time (see
Example C compared to Example D).
The only difference between the results of the service level constraint model and
the backorder cost model is that in the former this constraint cannot be satisfied for
low processing rates whereas in the latter low processing rates just lead to an
unlimited increase in costs. In Fig. 6.9 the start of the lines on the left hand side
Appendix 119
of the chart shows the minimum possible processing rate which still fulfills the
service level constraint.
The results presented in this chapter are twofold. Firstly, the models as presented in
Chaps. 4 and 5 are extended to an environment with a service level constraint
instead of the backorder costs. Secondly, a multi-item, normally distributed
demand, single-stage model for balancing inventory and capacity costs is
introduced in this chapter. Furthermore, the effect of replacing backorder costs
with a service level constraint is studied.
For the single-stage M/M/1 model the first order optimality conditions for
minimizing capacity, WIP, and FGI costs under a service level constraint are
provided and for the two-stage M/M/s model a numerically solvable optimization
problem is stated. For both models numerical examples show that the service level
constraint models and the backorder cost models lead to similar results. The main
difference is that in the service level constraint models too low processing rates fail
to reach the targeted service level. Furthermore, it is found that the transformation
of a backorder cost model to a service level constraint model cannot be conducted
by setting a constant backorder cost factor with respect to the targeted service level.
This leads to the finding that for simultaneous capacity and planned lead time
setting it has firstly to be decided if a service level model or a backorder cost model
is more suitable and then the optimization can be performed.
The multi-item, normally distributed demand, single-stage model is solved and
explicit expression for the optimal capacity investment and optimal inventory are
delivered.
Appendix
keðkþβÞX þ β
η¼1 , keðkþβÞX ¼ ð1 ηÞðk þ βÞ β
kþβ
(6.32)
β β
, eðkþβÞX ¼ 1 η 1 þ , ðk þ βÞX ¼ ln 1 η 1 þ
k k
Since the function for X is increasing in η and the cost function (6.34) increases in X,
the service level constraint is binding and it holds:

1 β
X ¼ ln 1 ~η 1 þ (6.33)
kþβ k
Proof of Proposition 6.1. Restating the cost function and applying Eqs. (3.1 and
(3.10) leads to:
λ
CðX; μ ¼cy E½Y þ cμ μ þ cf E½G ¼ cy þ cμ μ
k
(6.34)
þ cf λ
β ðk þ β Þ
Including the binding service level constraint by replacing X with the optimal
WAW X from Eq. (6.3) provides:
βeð½μλþβðμλþβ ln½1~ηð1þμλÞÞÞ
1 β
λ
CðμÞ ¼ cy þ cμ μ þ cf λ
μλ β ðμ λ þ β Þ
ðβðμλþβ
1
ln½1~ηð1þμλβ
ÞÞÞ
ðμ λ þ βÞe μλ
cf λ þ cf λ
β ðμ λ þ β Þ β ðμ λ þ β Þ
ð½μλþβðμλþβ
1
ln½1~ηð1þμλ
β
Þ ÞÞ
λ e
¼ cy þ cμ μ þ cf λ
μλ ðμ λ þ βÞ
(6.35)
eðβðμλþβ ln½1~ηð1þμλÞÞÞ
1 β
μλ
cf λ þ cf λ
β β ðμ λ þ β Þ

1 ~η 1 þ μλ β μλþβ
β
λ cf λ β
¼ cy þ cμ μ þ cf λ 1 ~η 1 þ
μλ ðμ λ þ β Þ β μλ
μλ
þ cf λ
β ðμ λ þ β Þ
Taking the first derivative with respect to μ and setting it zero leads to:

β
dCðμÞ cy λ ~ηβ 1 ~
η 1 þ ðμλÞ
¼ þ cμ þ cf λ cf λ
dμ ðμ λÞ 2
ðμ λ þ β Þðμ λ Þ 2
ðμ λ þ βÞ2
μλþβ
β
cf λ β ~ηβ2
1 ~η 1 þ
β ðμ λÞ ðμ λ þ βÞðμ λÞ2 ð1 ~ηð1 þ β=ðμ λÞÞÞ
!
β ln½1 ~ηð1 þ β=ðμ λÞÞ ðμ λÞ 1
cf λ þ cf λ ¼0
ðμ λ þ βÞ2 β ðμ λ þ β Þ2 β ðμ λ þ β Þ
(6.36)
cμ ~ηβð2μ 2λ þ βÞ þ ~ηðμ λÞ2 cy
, þ cf
λ ðμ λ þ βÞ2 ðμ λÞ2 ðμ λÞ2
0
μλþβ
β
β @ ~ηβ
¼ cf 1 ~η 1 þ
ðμ λÞ ðμ λ þ βÞ ð1 ~ηÞðμ λÞ2 ~ηβðμ λÞ
!
ln½1 ~ηð1 þ β=ðμ λÞÞ

ðμ λ þ β Þ2
Appendix 121
Proof of Proposition 6.2.

FNðλn;σ2 nÞ ðμ
nÞ ¼ η
pffiffiffi
n λnÞ
ðμ λÞ n
ðμ 1
, FNð0;1Þ pffiffiffi ¼η, ¼ FNð0;1Þ ðηÞ
σ n σ
pffiffiffi 1
, nðμ λÞ nFNð0;1Þ ðηÞσ ¼ 0 (6.37)
!2
F1
Nð0;1Þ ðηÞ
,n¼ σ2
n>0 λ
μ
Rearranging terms leads to μ

with respect to the service level η:
1
pffiffiffi 1 FN ð0;1Þ ðηÞσ
λÞ
nð μ ¼ λ þ
nFNð0;1Þ ðηÞσ ¼ 0 , μ pffiffiffi (6.38)
n

0 1
ð
1 ða
E½L ¼ τfLðτÞdτ ¼ lim @τFLðτÞja0 FLðτÞdτA
a!1
0 0
0 0a 1 1 1
ða ð ð
¼ lim @aFLðaÞ FLðτÞdτA ¼ lim @ FLðaÞ FLðτÞdτA ¼ 1 FLðτÞdτ
a!1 a!1
0 0 0
(6.39)
Proof of Proposition 6.3.

ð
1
¼
E½W maxðP; minðX; τÞÞfLðτÞdτ
0
ðP ðX ð
1
¼ P fLðτÞdτ þ τfLðτÞdτ þ X fLðτÞdτ

0 P X
(6.40)
ðX
¼ PFLðPÞ þ XFLðXÞ PFLðPÞ FLðτÞdτ þ X XFLðXÞ
P
ðX ðX
¼ X FLðτÞdτ ¼ P þ 1 FLðτÞdτ
P P
Applying Eq. (6.16) and Little’ Law leads to:
E½Y ¼ ðP þ nÞλ (6.41)

Proof of Proposition 6.4. From applying Eq. (6.19) to the cost function from
Eq. (6.8) a cost function with respect to capacity processing rate follows:
Þ ¼ E½Ych þ μ
Cðμ ; ~ηÞÞλch þ μ
cμ ¼ ðP þ nðμ cμ
(6.42)

w:r:t: η ~η; E½L P > n
The condition E½L P > n is assumed to hold in the Proposition 6.4 so it is

neglected in the remainder. Since from Eq. (6.11) it is obvious that nðμ ; ~ηÞ is an
increasing function in ~η and Cðμ Þ from Eq. (6.42) increases in nðμ ; ~ηÞ the service
level constraint ~η is binding. Therefore, the binding service level constraint can be
replaced by Eq. (6.11) which leads to:
0 !2 1
F1
Nð0;1Þ ðηÞ
Þ ¼ @P þ
C ðμ σ 2 Aλch þ μ
cμ (6.43)
λ
μ
The first order optimality condition follows as:

2
dCðμ
Þ F1
Nð0;1Þ ð~ηÞ
¼ 2λch σ 2 þ cμ ¼ 0
d
μ ðμ λÞ 3
0 2 113 (6.44)
1
F ð~ηÞ σ2 C
B Nð0;1Þ
)μ
¼ @2λch A þ λ
cμ
and the second derivative shows that this is a unique optimum:

2
dCðμÞ F1
Nð0;1Þ ð~ηÞ
¼ 6λch σ2 > 0 (6.45)
d2 μ
λÞ
4
ðμ
The optimal inventory held is then:

0 0 0 2 113 12 1
1
B B 1 B FNð0;1Þ ð~ηÞ σ 2 C C C
2 C
E Y ¼B
@ P þ BF
@ Nð0;1Þ ð~ηÞ @ 2 λc h A CA σ Aλ
cμ
(6.46)
0 !2 1
F1
Nð0;1Þ ð~
ηÞσcμ 3
¼ @P þ Aλ
2λch
μ ; ~ηÞ follows from Eq. (6.19) as:

Whereby nð
Appendix 123
!23

F1
Nð0;1Þ ð~
ηÞσcμ
μ ; ~ηÞ ¼
nð (6.47)
2λch
Furthermore, the optimal costs can be calculated as:

Þ ¼ E Y ch þ μ
C ðμ cμ
0 !2 113
2 13 1
F Nð0;1Þ ð~η Þcμ σ
¼ λcμ þ 2λch F1
Nð0;1Þ ð~
ηÞcμ σ þ Pλch þ @λch A
2
(6.48)
Proof of Proposition 6.5. Applying Eq. (6.26) leads to:
0 2 113
F1 ð~ηÞ σ2
@2λch Nð0;1Þ
A cμ
cμ
μ c
h excess μi ¼ 2 123
0
E Ysurplus ch 2 F1 ð~ηÞ σ2
F1 @2λch Nð0;1Þ
A σ 2λch
Nð0;1Þ ðηÞ cμ (6.49)
23 2 1 1 2
23 F1 σ 3λ3 c3h c3μ
1
Nð0;1Þ ð~ηÞ
¼ 23 1 1 2 ¼ 2
23 F1 ~ σ 3λ3 c3h c3μ
2
ð η Þ
2
Nð0;1Þ
Derivation of Equations (6.29) and (6.30). Applying the cdf of production lead
time for the two-stage production system with non-identical processing rates from
Eq. (5.32) to Eq. (3.9) leads to:
ðX ð
1
βτ
η ¼ FW ðτÞβe dτ þ FW ðXÞβeβτ dτ
0 X
ðX
ðt1 þ t2 Þðeμ1 τ 1Þ ðt3 t1 Þðeμ2 τ 1Þ ðt2 þ t4 Þ eðs2 μ2 λÞτ 1
¼ þ
μ1 μ2 ðs2 μ 2 λ Þ
0
!
ðt4 t3 Þ eðs1 μ1 λÞτ 1 βτ ðt1 þ t2 Þðeμ1 X 1Þ
βe dτ þ
ðs1 μ1 λÞ μ1
!
ðt3 t1 Þðeμ2 X 1Þ ðt2 þ t4 Þ eðs2 μ2 λÞX 1 ðt4 t3 Þ eðs1 μ1 λÞX 1
þ eβX
μ2 ðs2 μ 2 λ Þ ðs1 μ1 λÞ
(6.50)
Rearranging terms:
ðt1 þ t2 Þ 1 eðμ1 þβÞX ðt3 t1 Þ 1 eðμ2 þβÞX ðt2 þ t4 Þ 1 eðs2 μ2 λþβÞX

η¼ þ
ðμ 1 þ β Þ ðμ2 þ βÞ ðs2 μ2 λ þ βÞ
ðt4 t3 Þ 1 eðs1 μ1 λþβÞX
þ
ðs1 μ1 λ þ βÞ
(6.51)
For identical processing rates and equal number of machines at both stages
Eq. (5.36) applied to Eq. (3.9) shows:
ðX ð
1
βτ
ηbalanced ¼ FW ðτÞβe dτ þ FW ðXÞβeβτ dτ
0 X
ðX
ðm þ oÞ2 eðμþβÞτ ðm þ oÞ2 τeðμþβÞτ 2ðm þ oÞoeðμþβÞτ
¼β
μ2 μ μðμ sμ þ λÞ
0
!
o2 eðsμλþβÞτ o2 τeðsμλþβÞτ 2ðm þ oÞoeðsμλþβÞτ
þ dτ
ðsμ λÞ2 ðsμ λÞ ðsμ λÞðμ sμ þ λÞ
! ðX
ð m þ oÞ 2 o2 2ðm þ oÞo 2ðm þ oÞo
þ þ þ β eβτ dτ
μ2 ðsμ λÞ2 ðsμ λÞðμ sμ þ λÞ μðμ sμ þ λÞ
0
2 μX 2 μX μX 2 ðsμλÞX
ð m þ oÞ e ðm þ oÞ Xe 2ðm þ oÞoe oe
þ
μ2 μ μðμ sμ þ λÞ ðsμ λÞ2
o2 XeðsμλÞX 2ðm þ oÞoeðsμλÞX ð m þ oÞ 2 o2
þ þ þ
ðsμ λÞ ðsμ λÞðμ sμ þ λÞ μ2 ðsμ λÞ2

2ðm þ oÞo 2ðm þ oÞo
þ eβτ
ðsμ λÞðμ sμ þ λÞ μðμ sμ þ λÞ
(6.52)
Rearranging terms:
ðmðsμ λ þ βÞ oðμ sμ þ λÞÞ2 ðm þ oÞ2 eðμþβÞX
ηbalanced ¼
ðμ þ βÞ2 ðsμ λ þ βÞ2 ðμ þ β Þ2
ðm þ oÞ2 XeðμþβÞX o2 XeðsμλþβÞX o2 eðsμλþβÞX
(6.53)
ðμ þ β Þ ðsμ λ þ βÞ ðsμ λ þ βÞ2
2ðm þ oÞoeðμþβÞX 2ðm þ oÞoeðsμλþβÞX
þ
ðμ þ βÞðμ sμ þ λÞ ðsμ λ þ βÞðμ sμ þ λÞ
Chapter 7
Conclusion
In this book the influence of a customer required lead time distribution on the
optimal capacity investment and the optimal work release rule parameterization is
evaluated either including backorder costs or considering a service level constraint.
A set of equations for the logistical key figures FGI, FGI lead time, backorders,
tardiness, and service level has been developed in integral form based on the
production lead time distribution when a planned lead time or WAW work release
rule is applied and customers request random due dates. These equations can be
used in real production systems to identify the influence of the customer required
lead time distribution on the above mentioned logistical key figures. Additionally,
explicit equations for these logistical key figures have been derived for a set of
different production systems. The following main findings can be stated.
The implementation of a WAW work release rule to constrain the order release
into the production system in combination with a random customer required lead
time leads to significant FGI reduction potentials. In a numerical study of a
two-stage production system, only minor additional cost savings are found if
work release is also constrained within the production system. This supports
production planning methods only constraining work release at the beginning of
the production system like CONWIP or DBR.
For a two-stage production system with a planned lead time at each stage the
customer required lead time distribution is found to have no influence on the
optimal planned lead time values as long as WIP, FGI, and backorder costs are
simultaneously minimized. The implication of this finding is that if capacity is
predefined in a backorder cost model, it is not necessary to know the customer
required lead time distribution to optimize the planned lead times. However, if
capacity invested can also be optimized, the customer required lead time distribu-
tion has a strong influence on the optimal result.
A significant cost reduction potential has been identified when simultaneously
optimizing the capacity invested and the work release rule parameterization in
comparison to sequentially making these decisions. For practical application, this
implies that although the optimal work release rule might not be of interest in the

126 7 Conclusion
capacity investment phase, it still leads to a better overall performance if this

decision variable is included in the optimization problem.
For service level constraint models it is shown that it is not possible to transform
a service level constraint model into a backorder cost model with a constant
backorder cost factor if capacity invested and work release rule parameterization
are simultaneously optimized. This implies that in the first step management has to
decide either to apply a backorder cost model or to apply a service level constraint
model before the simultaneous optimization can be performed.
It is proven that capacity in a serial production system should increase towards
the customer end of the production line for continuous and predefined machine
sizes, under certain conditions. A managerial insight is therefore that it is usually
better to have slightly more overcapacity at the end of the production line than at the
beginning.
Predefined machine sizes are found to have a considerable impact on optimal
costs. Furthermore, uncertainty about customer order arrival rates prior to the
investment decision leads to significant additional costs and capacity investments.
It is shown that with uncertain customer order arrival rates and the option to reinvest
in machines, there is a higher probability that smaller machines lead to the optimal
result than without the reinvestment option. A managerial insight is therefore, that
whenever the machine type invested has to be decided under uncertain customer
demand information, the possibility to reinvest the same machine type later on often
leads to smaller machines being optimal.
For a multi-item, single-stage production system with normally distributed
demand under a capacity oriented service level constraint, the optimal costs are
reached when overcapacity costs are equal to the double surplus inventory costs.
This is an interesting managerial insight since it can be applied to check the current
cost situation of real manufacturing systems and derive measures to reach the
optimum.
For further research, explicit expressions for the external logistical key figures
service level, FGI lead time, and tardiness could be derived for more general
production systems and other distributions of the customer required lead time.
The equations developed in this book show that for the calculation of these figures,
the distribution of production lead time is important. Although there is already a lot
of literature available to approximate the mean production lead time for complex
production networks, further research could be conducted to also approximate the
variance of production lead time for such complex networks. These could be used to
identify approximations for the key figures developed in this work which only need
mean and variance of production lead time and customer required lead time as
input data.
Based on the results of this book, further research could be conducted
concerning the optimality of MTO or MTS production strategies when customers
have random due dates. Additionally, the already available literature on advance
demand information could be used as a basis to identify new safety stock strategies
including the customer required lead time distribution. The focus here was on
stationary production systems facing long-term decisions for capacity invested
7 Conclusion 127
and work release rule parameterization. Another direction of further research could
be to develop models for identifying the short term capacity needed using the
current system status information and the expected customer orders based on the
customer required lead time distribution.
Appendix
List of Variables
The following Table A.1 provides a summary of all variables used in this book.
Table A.1 Definition of variables used in the book

Symbol Description Unit
W Random variable of overall production lead time time
FW ðÞ cdf of the random variable W for overall production lead time 1
fW ðÞ pdf of the random variable W for overall production lead time 1
W Random overall lead time of an order in normally distributed time
demand model
Wi ; FWi ðÞ; fWi ðÞ Random variable, cdf, and pdf of production lead time at stage i time; 1
respectively
W~ i ; F ~ ðÞ; f ~ ðÞ Random variable, cdf, and pdf of remaining production lead time time; 1
Wi Wi
from arrival at stage i until its delivery to the FGI buffer
respectively
Y; FY ðÞ; fY ðÞ Random variable, cdf, and pdf of WIP in the production system pcs; 1
respectively
Y Random variable of WIP in the production system measured in time
capacity units in normally distributed demand model
Yi Random variable of WIP at stage i pcs
Yb Minimum required inventory to cover P in the normally time
distributed demand model
Ysurplus Surplus inventory in the normally distributed demand model time
I Random variable of FGI lead time time
G Random variable of FGI pcs
C Random variable of tardiness time
B Random variable of backorders pcs
L; FL ðÞ; fL ðÞ Random variable, cdf, and pdf of customer required lead time time; 1
respectively
(continued)

DOI 10.1007/978-3-319-00843-1, © Springer International Publishing Switzerland 2014
130 Appendix
Table A.1 (continued)

FLðÞ; fLðÞ
L; Random variable, cdf, and pdf of capacity oriented customer time; 1
required lead time respectively in normally distributed demand
model
Li ; FLi ðÞ; fLi ðÞ Random variable, cdf, and pdf of customer required lead time time; 1
respectively of the i-th product type in the normally distributed
demand model
β Parameter of the customer required lead time distribution (1=β is 1/time
the mean customer required lead time)
L^ Deterministic value of customer required lead time time
μ Processing rate of the machine pcs/time
μ
Capacity processing rate of the machine in normally distributed 1
demand model
μi Processing rate at stage i pcs/time
Ω Set of predefined processing rates in the two-stage M/M/s model pcs/time
μexcess Excess capacity in the normally distributed demand model 1
λ Order arrival rate to the system pcs/time
λ Capacity demand rate in normally distributed demand model 1
σ2 Capacity demand variance in normally distributed demand model 12
λi Demand rate of item i pcs/time
σ 2i Variance of item i demand pcs2/time
λl Low value for expected customer order arrival rate when demand pcs/time
is uncertain
λm Medium value for expected customer order arrival rate when pcs/time
demand is uncertain
λh High value for expected customer order arrival rate when demand pcs/time
is uncertain
k Unutilized processing rate (1=k is the mean production lead time pcs/time
in M/M/1)
ρ Utilization of the machine 1
αa Coefficient of variation of interarrival time 1
αc Coefficient of variation of processing time 1
X WAW for work release to the production system or overall time
planned lead time
n Time-averaging operator for demand smoothing in normally time
Xi Planned lead time at stage i time
Δi Random variable for waiting time until order release at stage i time
because of planned lead time policy
η Service level reached in the production system 1
~η Service level constraint for the production system 1
ηb Basic service level loss in normally distributed demand model 1
cy WIP holding costs for one order being stored one period of time curr/time
cy;i WIP holding costs at stage i for one order being stored one period curr/time
of time
cf FGI holding costs for 1 order being stored 1 period of time curr/time
ch Inventory holding costs for one capacity unit being stored one curr/time2
period of time
(continued)
Appendix 131
Table A.1 (continued)

cc Backorder costs for one order being late one period of time curr/time
cμ Capacity costs for one unit of capacity invested curr
cμ;i Capacity costs for one unit of capacity invested at stage i curr
s Number of machines in the balanced two-stage model 1
si Number of machines at stage i 1
p0 Probability that no order is in the production system for M/M/s 1
queue
CðÞ Costs to be minimized curr
Ai ðnÞ The random customer demand in pieces of item i in an arbitrary pcs
time period n in normally distributed demand model
qi Processing time for the i-th product type in normally distributed time
demand model
DðnÞ; fDðnÞ ðÞ; Random variable, cdf, and pdf of capacity demand which is time; 1
normally distributed for any arbitrary time period n
FDðnÞ ðÞ
αDðnÞ Coefficient of variation of capacity demand in time period n in time; 1
normally distributed demand model
N Number of different product types considered in the normally 1
P Minimum required lead time in the normally distributed demand time
model
m; o; t1 ; t2 ; t3 ; t4 Auxiliary variables in the two-stage M/M/s model for simplifica- 1
tion of statements
FNð0;1Þ ðÞ cdf of the standard normal distribution 1
E½ Expected value of a random variable
Var ½ Variance of a random variable
The following units are used for the variables.

Table A.2 Definition of Unit Description
units used in the book
time Continuous time
pcs Pieces or orders processed
curr Currency units; Money
References
Adenso-Diaz, B. (1996). How many units will be short when stockout occurs? International
Journal of Operations & Production Management, 16(4), 112–118.
Altendorfer, K., & Jodlbauer, H. (2011). An analytical model for service level and tardiness in a
single machine MTO production system. International Journal of Production Research, 49(7),
1827–1850.
Altendorfer, K., & Minner, S. (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.
Altendorfer, K., & Minner, S. (2012). Optimal composition of number and size of machines in a
multi-stage make-to-order system with due dates. International Journal of Production
Research, 50(3), 603–621.
Altendorfer, K., Jodlbauer, H., & Huber, A. (2007a). Behaviour of MRP, Kanban, CONWIP and
DBR under dynamic environmental variability. In C. Engelhardt-Nowizki, O. Nowitzky, &
B. Krenn (Eds.), Management komplexer Materialflüsse mittels Simulation – State-of-the-Art
innovative Konzepte (pp. 113–128). Wiesbaden: Deutscher Universitätsverlag.
Altendorfer, K., Kabelka, B., & Stöcher, W. (2007b). A new dispatching rule for optimising
machine utilisation at a semiconductor test field. ASMC Proceedings 2007, Stresa, Italy,
188–193.
Altiok, T. (1989). (R, r) Production/inventory systems. Operations Research, 37(2), 266–276.
Angelus, A., & Porteus, E. L. (2002). Simultaneous capacity and production management of short-
life-cycle, produce-to-stock goods under stochastic demand. Management Science, 48(3),
399–413.
Axsäter, S. (2005). Planning order releases for an assembly system with random operation times.
OR Spectrum, 27(2), 459–470.
Axsäter, S. (2006). Inventory control (2nd ed.). Boston a.o.: Kluwer Academic.
Beichelt, F. (2006). Stochastic processes in science, engineering, and finance. Boca Raton:
Chapman & Hall/CRC.
Bertrand, J. W. M., & Fransoo, J. C. (2008). Modelling and simulation. In C. Karlsson (Ed.),
Researching operations management (pp. 265–306). New York: Routledge.
Bertrand, J. W. M., & van Ooijen, H. P. G. (2008). Optimal work order release for make-to-order
job shops with customer order lead-time costs, tardiness costs and work-in-process costs.
International Journal of Production Economics, 116(2), 233–241.
Bertsimas, D., & Paschalidis, I. C. (2001). Probabilistic service level guarantees in make-to-stock
manufacturing systems. Operations Research, 49(1), 119–133.
Bish, E. K., Muriel, A., & Biller, S. (2005). Managing flexible capacity in a make-to-order
environment. Management Science, 51(2), 167–180.

DOI 10.1007/978-3-319-00843-1, © Springer International Publishing Switzerland 2014
134 References
Blackstone, J. H., Jr., Phillips, D. T., & Hogg, G. L. (1982). A state-of-the-art survey of
dispatching rules for manufacturing job shop operations. International Journal of Production
Research, 20(1), 27–45.
Bradley, J. R., & Glynn, P. W. (2002). Managing capacity and inventory jointly in manufacturing
systems. Management Science, 48(2), 273–288.
Brauer, D. C. (1985). Effect of lead time and service level on safety stock for a continuous review
inventory system with independent demand. Production and Inventory Management, 26(2),
146–151.
Brigham, G. (1955). On a congestion problem in an aircraft factory. Journal of the Operations
Research Society of America, 3(4), 412–428.
Brumelle, S. L. (1971). Some inequalities for parallel-server queues. Operations Research, 19(2),
402–413.
Buzacott, J. A., & Shanthikumar, J. G. (1992). Design of manufacturing systems using queueing
models. Queueing Systems, 12(1), 135–213.
Buzacott, J. A., & Shanthikumar, J. G. (1994). Safety stock versus safety time in MRP controlled
production systems. Management Science, 40(12), 1678–1689.
Chao, X., & Scott, C. (2000). Several results on the design of queuing systems. Operations
Research, 48(6), 965–970.
Chen, H., & Yao, D. D. (2001). Fundamentals of queueing networks: Performance, asymptotics,
and optimization. New York: Springer.
Chopra, S., Reinhardt, G., & Dada, M. (2004). The effect of lead time uncertainty on safety stocks.
Decision Sciences, 35(1), 1–24.
Clark, A. J., & Scarf, H. (1960). Optimal policies for a multi-echelon inventory problem.
Management Science, 6(4), 475–490.
Coelli, T., Grifell-Tatje, E., & Perelman, S. (2002). Capacity utilisation and profitability: A
decomposition of short-run profit efficiency. International Journal of Production Economics,
79(3), 261–278.
Coleman, B. J. (2000). Determining the correct service level target. Production and Inventory
Management Journal, 41(1), 19–23.
Crabill, T. B., Gross, D., & Magazine, M. J. (1977). A classified bibliography of research on
optimal design and control of queues. Operations Research, 25(2), 219–232.
Dar-El, E. M., & Malmborg, C. J. (1991). Improved strategy for service level-based safety stock
determination. Production Planning & Control, 2(2), 116–121.
de Cani, J. S. (1962). A queueing model for business students. Management Science, 8(3),
278–286.
del Vecchio, C., & Paschalidis, L. C. (2006). Enforcing service-level constraints in supply chains
with assembly operations. IEEE Transactions on Automatic Control, 51(12), 2000–2005.
Driscoll, M. F., & Weiss, N. A. (1976). An application of queueing theory to reservation networks.
Duenyas, I., & Hopp, W. J. (1995). Quoting customer lead times. Management Science, 41(1),
43–57.
Elhafsi, M. (2002). Optimal leadtimes planning in serial production systems with earliness and
tardiness costs. IIE Transactions, 34(3), 233–243.
Federgruen, A., & Zipkin, P. (1984). Computational issues in an infinite-horizon, multiechelon
inventory model. Operations Research, 32(4), 818–836.
Gallego, G., & Özer, Ö. (2001). Integrating replenishment decisions with advance demand
information. Management Science, 47(10), 1344–1360.
Gayon, J.-P., Benjaafar, S., & de Vericourt, F. (2009). Using imperfect advance demand informa-
tion in production-inventory systems with multiple customer classes. Manufacturing & Service
Operations Management, 11(1), 128–143.
Glasserman, P., & Tayur, S. (1994). The stability of a capacitated, multi-echelon production-
inventory system under a base-stock policy. Operations Research, 42(5), 913–924.
References 135
Gong, L., de Kok, T., & Ding, J. (1994). Optimal leadtimes planning in a serial production system.
Grassmann, W. (1983). The convexity of the mean queue size of the M/M/c queue with respect to
the traffic intensity. Journal of Applied Probability, 20(4), 916–919.
Graves, S. C. (1986). A tactical model for a job shop. Operations Research, 34(4), 522–533.
Greis, N. P. (1994). Assessing service level targets in production and inventory planning. Decision
Sciences, 25(1), 15–40.
Gupta, D., & Benjaafar, S. (2004). Make-to-order, make-to-stock, or delay product differentiation?
A common framework for modeling and analysis. IIE Transactions, 36(6), 529–546.
Hariharan, R., & Zipkin, P. (1995). Customer-order information, leadtimes, and inventories.
Hartl, R. J. (1990). Inventory: An overlooked issue in financial management. The Journal of
Education for Business, 66(1), 57–62.
Hillier, F. S. (1963). Economic models for industrial waiting line problems. Management Science,
10(1), 119–130.
Hopp, W. J., & Roof, M. L. (1998). Setting WIP levels with statistical throughput control (STC) in
CONWIP production lines. International Journal of Production Research, 36(4), 867–882.
Hopp, W. J., & Roof-Sturgis, M. L. (2000). Quoting manufacturing due dates subject to a service
level constraint. IIE Transactions, 32(9), 771–784.
Hopp, W. J., & Spearman, M. L. (2008). Factory physics: Foundation of manufacturing manage-
ment (3rd ed.). Chicago: McGraw-Hill/Irwin.
Inderfurth, K., & Schulz, T. (2007). Zur Exaktheit der Lagerkennlinie nach Nyhuis und Wiendahl.
In A. Otto & R. Obermaier (Eds.), Logistikmanagement. Analyse, Bewertung und Gestaltung
logistischer Systeme (pp. 23–49). Wiesbaden: Deutscher Universitätsverlag.
Inderfurth, K., & Schulz, T. (2008). Zur Analyse von Lagerkennlinien. In K. Inderfurth,
G. Neumann, M. Schenk, G. Wäscher, & D. Ziems (Eds.), Netzwerklogistik (pp. 295–310).
Magdeburg: Logisch GmbH.
Jodlbauer, H. (2008a). 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.
Jodlbauer, H. (2008b). Customer driven production planning. International Journal of Production
Economics, 111(2), 793–801.
Jodlbauer, H., & Altendorfer, K. (2010). Trade-off between capacity invested and inventory
needed. European Journal of Operational Research, 203(1), 118–133.
Jodlbauer, H., & Huber, A. (2008). Service-level performance of MRP, Kanban, CONWIP and
DBR due to parameter stability and environmental robustness. International Journal of
Production Research, 46(8), 2179–2195.
Jones, C. H. (1973). An economic evaluation of dispatching rules. Management Science, 20(3),
293–307.
Karaesmen, F., Buzacott, J. A., & Dallery, Y. (2002). Integrating advance order information in
make-to-stock production systems. IIE Transactions, 34(8), 649–662.
Karaesmen, F., Liberopoulos, G., & Dallery, Y. (2004). The value of advance demand information
in production/inventory systems. Annals of Operations Research, 126(1), 135–157.
Karmarkar, U. S. (1987). Lot sizes, lead times and in-process inventories. Management Science, 33
(3), 409–418.
Karmarkar, U. S. (1993). Manufacturing lead times, order release and capacity loading. In S. C.
Graves, H. G. Rinnooy Kan, & P. H. Zipkin (Eds.), Logistics of production and inventory
(pp. 287–329). Amsterdam a. o.: Elsevier.
Kiesmüller, G. P., & de Kok, A. G. (2006). The customer waiting time in an (R, s, Q) inventory
system. International Journal of Production Economics, 104(2), 354–364.
Kruse, W. K. (1980). Waiting time in an S-1, S inventory system with arbitrarily distributed lead
times. Operations Research, 28(2), 348–352.
136 References
Kutanoglu, E., & Sabuncuoglu, I. (1999). An analysis of heuristics in a dynamic job shop with
weighted tardiness objectives. International Journal of Production Research, 37(1), 165–187.
Lejeune, M. A., & Ruszczynski, A. (2007). An efficient trajectory method for probabilistic
production-inventory-distribution problems. Operations Research, 55(2), 378–394.
Liberopoulos, G. (2008). On the tradeoff between optimal order-base-stock levels and demand
lead-times. European Journal of Operational Research, 190(1), 136–155.
Little, J. D. C. (1961). A proof for the queuing formula L ¼ λW. Operations Research, 9(3),
383–387.
Liu, L., & Yuan, X.-M. (2001). Throughput, flow times, and service level in an unreliable
assembly system. European Journal of Operational Research, 135(3), 602–615.
Luss, H. (1982). Operations research and capacity expansion problems: A survey. Operations
Research, 30(5), 907–947.
Lutz, S., Löedding, H., & Wiendahl, H.-P. (2003). Logistics-oriented inventory analysis. Interna-
tional Journal of Production Economics, 85(2), 217–231.
MacGregor-Smith, J. (2007). Multi-server, finite waiting room, M/G/c/K optimization models.
Infor, 45(4), 257–274.
Mandelbaum, A., & Reiman, M. I. (1998). On pooling in queueing networks. Management
Science, 44(7), 971–981.
Mapes, J. (1993). The effect of capacity limitations on safety stock. International Journal of
Operations & Production Management, 13(10), 26–33.
Medhi, J. (1991). Stochastic models in queuing theory. Boston: Academic.
Mincsovics, G., Tan, T., & Alp, O. (2009). Integrated capacity and inventory management with
capacity acquisition lead times. European Journal of Operational Research, 196(3), 949–958.
Mirasol, N. M. (1963). The output of an M/G/1 queuing system is poisson. Operations Research,
11(2), 282–284.
Missbauer, H. (2002). Aggregate order release planning for time-varying demand. International
Journal of Production Research, 40(3), 699–718.
Newell, G. F. (1966). The M/G/1 queue. SIAM Journal on Applied Mathematics, 14(1), 86–88.
Nyhuis, P., & Wiendahl, H.-P. (2002). Logistische Kennlinien. Grundlagen, Werkzeuge und
Anwendungen (2nd ed.). Berlin u. a.: Springer.
Paraskevopoulos, D., Karakitsos, E., & Rustem, B. (1991). Robst capacity planning under uncer-
tainty. Management Science, 37(7), 787–800.
Rajagopalan, S., & Swaminathan, J. M. (2001). A coordinated -production planning model with
capacity expansion and inventory management. Management Science, 47(11), 1562–1580.
Raman, A., & Kim, B. (2002). Quantifying the impact of inventory holding cost and reactive
capacity on an apparel manufacturer’s profitability. Production and Operations Management,
11(3), 358–373.
Rao, S. S., Gunasekaran, A., Goyal, S. K., & Martikainen, T. (1998). Waiting line model
applications in manufacturing. International Journal of Production Economics, 54(1), 1–28.
Ross, S. M. (2010). Introduction to probability models (10th ed.). Amsterdam, a. o.: Elsevier.
Scheller-Wolf, A. (2003). Necessary and sufficient conditions for delay moments in FIFO
multiserver queues with an application comparing s slow servers with one fast one. Operations
Research, 51(5), 748–758.
Schneider, H. (1981). Effect of service-levels on order-points or order-levels in inventory models.
International Journal of Production Research, 19(6), 615–631.
Schragenheim, E., & Dettmer, W. (2001). Manufacturing at warp speed. Optimizing supply chain
financial performance. Boca Raton: CRC Press.
Segerstedt, A. (1994). Inventory control with variation in lead times, especially when demand is
intermittent. International Journal of Production Economics, 35(1), 365–372.
Selcuk, B., Fransoo, J. C., & de Kok, A. G. (2006). The effect of updating lead times on the
performance of hierarchical planning systems. International Journal of Production Economics,
104(2), 427–440.
References 137
Selcuk, B., Fransoo, J. C., & de Kok, A. G. (2008). Work-in-process clearing in supply chain
operations planning. IIE Transactions, 40(3), 206–220.
Silver, E. A., Pyke, D. F., & Peterson, R. (1998). Inventory management and production planning
and scheduling (3rd ed.). New York: Wiley.
Slack, N., & Lewis, M. (2002). Operations strategy. Harlow, England: Prentice-Hall.
Song, J.-S., Yano, C. A., & Lerssrisuriya, P. (2000). Contract assembly: Dealing with combined
supply lead time and demand quantity uncertainty. Manufacturing & Service Operations
Management, 2(3), 287–296.
Song, D., Hicks, C., & Earl, C. F. (2001). Setting planned job release times in stochastic assembly
systems with resource constraints. International Journal of Production Research, 39(6),
1289–1301.
Souza, G. C., Ketzenberg, M. E., & Guide, V. D. R., Jr. (2002). Capacitated remanufacturing with
service level constraints. Production and Operations Management, 11(2), 231–248.
Spearman, M. L., Woodruff, D. L., & Hopp, W. J. (1990). CONWIP: A pull alternative to Kanban.
International Journal of Production Research, 28(5), 879–894.
Spitter, J. M., Hurkens, C. A. J., de Kok, A. G., Lenstra, J. K., & Negenman, E. G. (2005). Linear
programming models with planned lead times for supply chain operations planning. European
Journal of Operational Research, 163(3), 706–720.
Sridharan, V., & LaForge, L. R. (1994). A model to estimate service levels when a portion of the
master production schedule is frozen. Computers and Operations Research, 21(5), 477–486.
Stidham, S., Jr. (1970). On the optimality of single server queuing systems. Operations Research,
18(4), 708–732.
Stidham, S., Jr. (1992). Pricing and capacity decisions for service facility: Stability and multiple
local optima. Management Science, 38(8), 1121–1139.
Stidham, S., Jr. (2002). Analysis, design, and control of queuing systems. Operations Research, 50
(1), 197–216.
Tan, T., Güllü, R., & Erkip, N. (2007). Modelling imperfect advance demand information and
analysis of optimal inventory policies. European Journal of Operational Research, 177(2),
897–923.
Tempelmeier, H. (2000). Inventory service-levels in the customer supply chain. OR-Spektrum, 22
(3), 361–380.
Tijms, H. C. (2003). A first course in stochastic models. Amsterdam: Wiley.
van der Heijden, M. C., & de Kok, A. G. (1992). Customer waiting times in an (R, S) inventory
system with compound Poisson demand. Zeitschrift für Operations Research, 36(4), 315–332.
van Mieghem, J. A. (1998). Investment strategies for flexible resources. Management Science, 44
(8), 1071–1078.
van Mieghem, J. A. (2003). Capacity management, investment, and hedging: Review and recent
developments. Manufacturing & Service Operations Management, 5(4), 269–302.
van Mieghem, J. A., & Rudi, N. (2002). Newsvendor networks: Inventory management and
capacity investment with discretionary activities. Manufacturing & Service Operations Man-
agement, 4(4), 313–335.
van Nieuwenhuyse, I., Vandaele, N., Rajaram, K., & Karmarkar, U. S. (2007). Buffer sizing in
multi-product multi-reactor batch processes: Impact of allocation and campaign sizing
policies. European Journal of Operational Research, 179(2), 424–443.
Vepsalainen, A. P. J., & Morton, T. E. (1987). Priority rules for job shops with weighted tardiness
costs. Management Science, 33(8), 1035–1047.
Weeks, J. K. (1981). Optimizing planned lead times and delivery dates. 21st Annual Conference
Proceeding, American Production and Inventory Society, pp.177–188.
Weng, K. Z. (1996). Manufacturing lead times, system utilization rates and lead-time-related
demand. European Journal of Operational Research, 89(2), 259–268.
Weng, K. Z. (1999). Strategies for integrating lead time and customer-order decisions. IIE
Transactions, 31(2), 161–171.
138 References
Wiendahl, H.-P., & Breithaupt, J.-W. (1999). Modelling and controlling the dynamics of produc-
tion systems. Production Planning & Control, 10(4), 389–401.
Wiendahl, H.-H., von Cieminski, G., & Wiendahl, H.-P. (2005). Stumbling blocks of PPC:
Towards the holistic configuration of PPC systems. Production Planning & Control, 16(7),
634–651.
Wierman, A., Osogami, T., Harchol-Balter, M., & Scheller-Wolf, A. (2006). How many servers
are best in a dual-priority M/PH/k system? Performance Evaluation, 63(12), 1253–1272.
Wijngaard, J. (2004). The effect of foreknowledge of demand in case of a restricted capacity: The
single-stage, single-product case. European Journal of Operational Research, 159(1), 95–109.
Wijngaard, J., & Karaesmen, F. (2007). Advance demand information and a restricted production
capacity: On the optimality of order base-stock policies. OR Spectrum, 29(4), 643–660.
Yano, C. A. (1987). Setting planned leadtimes in serial production systems with tardiness costs.
Yu, K.-W. (2001). Terminkennnlinie. Eine Beschreibungsmethodik für die Terminabweichung im
Produktionsbereich. Düsseldorf: VDI Verlag.
Zäpfel, G. (1998). Customer-order-driven production: An economical concept for responding to
demand uncertainty? International Journal of Production Economics, 56(57), 699–709.
Zhang, F., Roundy, R., Çanyildirim, M., & Huh, W. T. (2004). Optimal capacity expansion for
multi-product, multi-machine manufacturing systems with stochastic demand. IIE
Transactions, 36(1), 23–36.
Zipkin, P. H. (2000). Foundations of inventory management. Boston, a. o.: McGraw-Hill Higher
Education.

Capacity and Inventory Planning For Make-to-Order Production Systems The Impact of A Customer Required Lead Time Distribution

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

Capacity and Inventory Planning For Make-to-Order Production Systems The Impact of A Customer Required Lead Time Distribution

Uploaded by

Copyright:

Available Formats

Lecture Notes in Economics and Mathematical Systems 671

Prof. Dr. W. Trockel

Institut für Mathematische Wirtschaftsforschung (IMW)

H. Dawid, D. Dimitrov, A. Gerber, C-J. Haake, C. Hofmann, T. Pfeiffer,

For further volumes:

Capacity and Inventory

Published with the support of Austrian Science Fund (FWF)

© Springer International Publishing Switzerland 2014

Printed on acid-free paper

Springer is part of Springer Science+Business Media (www.springer.com)

Steyr, March 2013 Klaus Altendorfer

4.3 Simultaneous Optimization of a Single-Stage Production

6.3.6 Numerical Example . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 113

Fig. 1.1 Basic model framework . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3

Fig. 5.6 Influence of uncertain input rate on optimal costs in two-stage

Table 3.1 Numerical examples for logistical relationships in

Decisions on manufacturing capacity and production planning parameters are

K. Altendorfer, Capacity and Inventory Planning for Make-to-Order Production 1

• Influence of a random customer required lead time and a work-ahead-window

Fig. 1.1 Basic model Customer required Tardiness

FGI lead time

K. Altendorfer, Capacity and Inventory Planning for Make-to-Order Production 5

2.1 Relationships Between Logistical Key Performance

2.1.1 Internal View on Production Logistics

single simulation models and no analytical proofs can be delivered. Both

2.1.1.1 Analytical Approaches

2.1.1.2 Simulation and Empirical Approaches

Further approaches to discuss logistical relationships are empirical studies based on

2.1.2 External View on Production Logistics

The external view on production logistics expressed through FGI, tardiness,

2.1.2.1 Service Level Calculation

problem structures is given. Most of the problems (and solution procedures)

2.1.2.2 Tardiness Calculation

products in an MTS system, this term is equivalent to the tardiness definition as

2.2 Work Release Rules

2.2.1 Planned Lead Time Setting

2.2.2 Advance Demand Information

2.3 Capacity Investment

2.3.1 Queuing System Design

2.3.2 Backorder Cost or Service Level Models

The differentiation between constant capacity available and flexible capacity

3.1 Introductory Remarks

K. Altendorfer, Capacity and Inventory Planning for Make-to-Order Production 19

3.2 Model Development

The model developed in this chapter provides a deeper understanding of the

3.2.1 Model Assumptions and Notation

Customer required Tardiness

FGI lead time

Fig. 3.1 Single-stage production system

3.2.2 Queuing Model with Deterministic Customer Required

3.2.2.1 M/M/1 Queue

3.2.2.2 Service Level, FGI Lead Time, and Tardiness

E½G ¼ E½Iλ (3.4)

3.2.3 Distribution of Customer Required Lead Time

A distribution of customer required lead time is implemented in this section

Derivation see Appendix.

Derivation see Appendix.

the minimum expected tardiness. Furthermore, it increases in β which means the

3.2.4 Integration of a WAW Work Release Policy

Derivation see Appendix.

E½I ¼ fW ðτÞðθ τÞdτfL ðθÞdθ þ fW ðτÞðX τÞdτfL ðθÞdθ

Derivation see Appendix.

Calculating the limit ρ ! 1 shows that the tardiness approaches

(denoted as E½YG ) exists. Therefore, it follows that expected FGI increases in