246 Int. J. Services, Economics and Management, Vol. 2, Nos.

3/4, 2010

A multi-product lot size in make-to-order supply

chain using discrete event simulation and
response surface methodology

Wafik Hachicha*
Unit of Mechanic, Modelling and Production (U2MP),
Engineering School of Sfax,
B.P. 1173, Sfax 3038, Tunisia
and
Department of Industrial Management,
High Institute of Industrial Management of Sfax (ISGI),
B.P. 954, Sfax 3018, Tunisia
Email: wafik_hachicha@yahoo.fr
*Corresponding author

Ahmed Ammeri
Unit of Logistic, Industrial and Quality Management (LOGIQ),
High Institute of Industrial Management of Sfax,
B.P. 954, Sfax 3018, Tunisia
Email: ammariahmed@yahoo.fr

Faouzi Masmoudi
Unit of Mechanic, Modelling and Production (U2MP),
Department of Mechanical Engineering,
Engineering School of Sfax,
B. P.W, Sfax 3038, Tunisia
Fax: + 216-74-274-088
Email: faouzi.masmoudi@enis.rnu.tn

Habib Chabchoub
Unit of Logistic, Industrial and Quality Management (LOGIQ),
University of Economic Sciences and Management,
Sfax, Tunisia
Email: habib.chabchoub@fsegs.rnu.tn

Abstract: This paper develops a simulation optimisation approach for solving

the Lot-Sizing Problem (LSP) in Make-to-Order (MTO) supply chain. For
this purpose, a discrete event simulation model was firstly implemented as a
tool in estimating Order Mean Flow Time (OMFT) performance. Secondly,
a multiple-objective optimisation was achieved by applying Response Surface
Methodology (RSM). A comprehensive case study is detailed which involves
a multi-product, multi-stage, multi-location production planning with capacity-

Copyright © 2010 Inderscience Enterprises Ltd.

 A multi-product lot size in make-to-order supply chain 247

constrained and stochastic parameters such as lot arrivals order, transit time,
set-up time, processing time, etc. The objective of the proposed approach is to
determine the fixed optimal lot size for each manufacturing product type that
will ensure OMFT target value for each finished product type. The study
results illustrate that the LSP in MTO sector is viable and provide a prototype
for further research on simulation optimisation approaches.

Keywords: LSP; lot-sizing problem; supply chain; coordination; make-to-

order; discrete event simulation; response surface methodology; multiple-
objective optimisation; desirability function; simulation optimisation approach;
mean flow time; manufacturing; management; multi-product; multi-location.

Reference to this paper should be made as follows: Hachicha, W., Ammeri, A.,
Masmoudi, F. and Chabchoub, H. (2010) ‘A multi-product lot size in make-to-
order supply chain using discrete event simulation and response surface
methodology’, Int. J. Services, Economics and Management, Vol. 2, Nos. 3/4,
pp.246–266.

Biographical notes: Wafik Hachicha is an Industrial Engineer (since 1999)

in ENIT Tunisia. He obtained his PhD in Manufacturing Management (2009).
He is a Researcher at the Mechanics Modelling and Production Research Unit
(U2MP). His research activities deal with the modelling, analysis, optimisation
and the simulation of manufacturing and supply chain system. He is an
Assistant Professor at the High Institute of Industrial Management of Sfax
in Tunisia (ISGI).

Ahmed Ammeri is a student at Higher Institute of Industrial Management of

Sfax in Tunisia (ISGI). He obtained his Master in Science of Transport and
Logistic in 2009. He is a member of LOGIQ Unit. His research activities deal
with the optimisation and supply chain management.

Faouzi Masmoudi is a Professor at the National School of Engineers of Sfax in

Tunisia and a Researcher at the Mechanics Modelling and Production Research
Unit (U2MP). He received his PhD in Computer-Integrated Manufacturing
(1988) from ENSAM de Paris, France. His research activities are modelling
and simulation of manufacturing systems; design and layout of the cellular
systems of production and simulation of manufacturing cells with unreliable
machines.

Habib Chabchoub is a Professor in Quantitative Methods at the University

of Economic Sciences and Management of Sfax in Tunisia, and the Director
of the Laboratory GIAD (Gestion Industrielle et Aide à la Décision), Sfax,
Tunisia. He is a member of Logistic, Industrial and Quality Management
(LOGIQ) unit.

1 Introduction

A Supply Chain (SC) is a network of facilities and distribution entities such as materials
vendors, manufacturers, distributors, wholesalers and retailers that performs the functions
of procurement of raw materials, transformation of raw materials into intermediate and
finished products and distribution of finished products to customers. A SC is typically
characterised by a forward flow of materials and a backward flow of information. End-
user demand information suffers from delay and distortion as it moves upstream in a SC.
248 W. Hachicha et al.

The coordination between organisations in the SC, through sharing of demand information,
is a possible solution to counter this distortion. Consequently, enterprises have shown
a growing interest for an integrated SC management. An important issue in integrated
logistic network management is to control the inventory at different entities while
meeting end-customer service level requirements, therefore quantifying the trade-off
between inventory investment and end-customer service levels. The dynamic nature of
complex logistic chains causes that this trade-off will change over time.
The coordination between different entities in the SC is not an easy task because each
one is involved in efforts to reduce costs and maximise profits. However, the various
entities may have conflicting objectives and the profits of the entire SC may suffer if the
different stakeholders cannot be brought into agreement and coordination. Traditional
production planning methods, such as Material Requirement Planning (MRP), consider
only the availability of materials when organising demands and totally ignore such
factors as capacity limits and SC configurations. For this reason, MRP cannot provide
feasible production plans since such plans require that capacity limits and multiple
objectives are taken into account. To cope with these challenges, Advance Planning and
Scheduling (APS) methods were developed to integrate the planning activities of the
entire SC, providing powerful planning procedures and methodologies that are able to
react quickly to exceptions and variability.
In addition, each entity of the SC that has the ability to fill customer orders quickly,
as well as offer custom products has the benefit of a competitive advantage. However,
the need to have high product variety and quick response time places conflicting demands
on the production system. For this reason, businesses that compete on response time
focus on producing a limited portfolio of products. Items which are produced ahead of
demand and are kept in stock are ready to be shipped upon receipt of orders. Producing to
stock becomes costly when the number of products is large. It is also risky when demand
is highly variable and/or products have short life cycles. Therefore, an increasing number
of companies have been shifting its production from the Make-to-Stock (MTS) to the
Make-to-Order (MTO) sector such as Dell, BMW, Compaq and Gateway (Gunasekaran
and Ngai, 2009). In addition, the thrust in today’s manufacturing environment is to move
towards Just-in-Time philosophy. However, most of the literature research on MRP and
APS approaches concentrates on MTS systems. As a consequence, many researchers
provide excellent surveys of models and methods for the deterministic and the stochastic
demand cases (Karimi et al., 2003). The MTO area has not received the same degree of
attention. In particular, there are only a handful of research papers that explicitly talk
about the Lot-Sizing Problem (LSP) in MTO sector. In this case, methods based only on
queueing network models have been proposed in the literature. Though, these existing
analytical models are not able to handle all the dynamically changes in SC and they
require many simplification assumptions and constraints. Some authors proposed to
apply stochastic simulation techniques to solve LSP based on queueing model equations.
In fact, this proposed technique has made analytically possible the analysis of further
complex SC although they preserved some disadvantages as mentioned of the analytical
models.
Modelling and analysing SC are not an easy task, as they contain multiple echelons
and are faced with uncertain demand and lead times. Therefore, methodologies based on
Discrete Event Simulation (DES) models are attractive when dealing with SC systems.
They allow both material and information flow to be modelled as well as sophisticated
decision logic for planning and control. DES can also deal with stochastic environments,
 A multi-product lot size in make-to-order supply chain 249

non-stationary demand patterns, multiple performance criteria, capacity constrained

resources, assembly coordination, etc. Opposing to analytical models, any degree of
detail can be modelled, eliminating the need for simplifying assumptions and constraints.
The aim of this paper is to develop a simulation optimisation approach for solving the
LSP in MTO supply chain. A complete case study in MTO sector for which mathematical
model is still extremely complex up to now (multi-stage, multi-product, multi-location,
multi-resource with set-up, capacity constraints and stochastic demand) is provided. Our
objective is to demonstrate that DES method combined with a multi-objective desirability
optimisation tool is able to determine optimal fixed lot sizes in SC which operate in MTO
environment.
The remainder of this paper is organised as follows. Section 2 presents a related
theoretical background which concerns MTS and MTO supply chain management, LSP
and simulation optimisation method. Two relevant literatures are reviewed in Section 3.
The first one concerns the use of different simulation categories modelling in SC. The
second consists of different LSP in MTO supply chain. The details of the proposed
methodology are discussed in Section 4. In Section 5, an illustrative example to explain
how this methodology can be used to determine optimal fixed lot sizes in SC which
operates in MTO environment is described. Conclusions are pointed out in Section 6.

2 Theoretical background

This paper presents a simulation optimisation approach to address LSPs in the MTO
supply chain sector. That is why the current theoretical background subsection is organised
as follows. First, MTO and MTS production strategy are briefly presented. Secondly,
lot-sizing models and methods are provided. Finally, DES methodology is described.

2.1 MTO and MTS production strategy

The manufacturing strategies can be classified summarily in MTS or MTO sector. The
selection of the benefit strategy in the production network is the critical decision point at
which the production system is chosen due to the tight integration of customer-supplier
links along the value chain.
For the MTS items, the strategy is aimed at ready replenishment with on-hand stock
to satisfy customers with common demands, but is burdened with relatively larger amount
of inventory and related costs. The production planning problem involves determining lot
sizes to replenish inventory. Lot sizes models are typically based on some function of
set-up, inventory and production costs.
For the MTO items, the incentive lies in producing customised orders for added
revenue, but must respond with high degree of flexibility. In general, since demand
cannot be predicted with a high degree of confidence, no finished goods inventory is
stocked. As a result, a delivery date is quoted in response to a customer order and the
production planning problem involves determining lot sizes so that delivery promises are
satisfied based on minimised Order Mean Flow Time (OMFT).
250 W. Hachicha et al.

2.2 Lot-sizing problem

The common objective of LSP is to determine the optimal level and/or timing of
production. The research on LSP has a long history. It is stated in various research papers
with different manner which sometimes cause ambiguity. As mentioned in Figure 1, it is
necessary to make out between three adjacent categories: lot-sizing policies, lot-sizing
models and lot-sizing solutions or approaches.

Figure 1 Three levels of lot sizing problem

Inventory and replenishment strategies

Choose Lot sizing policies (or decisions):
(MTS or MTO), capacity constraints,
demand rolling horizon… EOQ, LFL, FOQ, POQ …

The time scale and horizon, the Formulate

demand distribution, the number of Lot sizing models:
stages, the number of products… EOQ, WW, CLSP, ELSP…

Supply chain complexity, accuracy and Solve Lot sizing approaches (dynamic
robustness of the solution… programming, queueing method,
simulation…

The level of production, the lot size, the optimal

timing, the scheduling…

At the beginning, it is necessary to choose a lot-sizing policy based on which inventory

and replenishment strategy is adopted at each entity of supply chain (MTS, MTO and
MTS/MTO). In this category, different lot-sizing policies can be applied. Lot-for-Lot
(LFL) is among the most popular with practitioners since it is simple and produces
the least remnant work-in-process inventory (Ho, 1993). However, set-up costs can be
excessive if too many small lot sizes result. Fixed-Order-Quantity (FOQ) is another policy
used extensively in practice due to its simplicity and lot size consistency. While these
rules may be justified for practical reasons, it has been argued that other lot-sizing
policies are more theoretically sound. Some of these are based on optimisation algorithms
that attempt to achieve the lowest total cost. This usually includes set-up and inventory
holding cost components. Such lot-sizing policies include the famous Economic-Order-
Quantity (EOQ). At this moment, it is necessary to formulate the greatest lot-sizing
model based on which SC hypothesis: time scale and horizon, demand distribution,
number of stages and number of products in SC.
Moreover, EOQ can be viewed also as lot-sizing model which assumes a continuous
time scale, constant demand rate and infinite time horizon. Wagner and Whitin (1958)
solved the deterministic single-item problem for the finite period case. Extensive research
papers generalise the Wagner-Whitin model into various directions thought different
assumptions and constraints. For example, the extension to multiple items and constant
production rates is known as the Economic Lot Scheduling Problem (ELSP) and the
extension to treat the problem with capacity constraints is known as the Capacity Lot-
Sizing Problem (CLSP). These model extensions generally deal with MTS production
strategy. Another group of models is dealing with the stochastic version of LSP. In the
 A multi-product lot size in make-to-order supply chain 251

discrete time case the Markov process model can be used. Only a few papers exist for the
continuous time case. Most of them use a queueing theory formulation. This group of
models deals commonly with MTO production strategy.
After building the suitable lot-sizing model, there is a need to solve the obtained
LSP based on SC complexity, preferred accuracy of the final solution and others. Many
lot-sizing approaches are presented in the literature, such as dynamic programming,
heuristics, meta-heuristics, queueing network method, simulation based on queueing
models and DES. This paper deals with the MTO strategy which a Fixed-Economic-
Order-Quantity is selected.

2.3 Discrete event simulation

Various alternative methods have been proposed for modelling supply chains. They
can be grouped into four categories: deterministic models where all the parameters are
known, stochastic models where at least one parameter is unknown but follows a
probabilistic distribution, economic game-theoretic models and models based on simulation,
which evaluate the performance of various SC strategies. It should be noted that there are
two simulation category applications. The first consists of applying simulation to solve
analytical formulation such as queueing network model and the second one is the DES
which is applied in this paper.
DES is widely used in the world and therefore it is very familiar. The most important
reasons and advantages of simulation methodology for modelling logistic systems are as
follows:
• realistic models are possible, they are a practical approach to represent the important
characteristics of SC management and may incorporate any complex interactions
that exist between different variables;
• options may be considered without direct system experimentation and alternative
designs can be easily evaluated and independently of the real system. It is flexible
and parametric for evaluating different scenarios;
• computer simulation models’ ability to directly address the performance measures
typically used in a real system;
• visual output helps and assists the end user in model development and validation;
• no advanced mathematics is required. It is time efficient even in correspondence of
high number of SC stages and high numbers of items or products;
• analytical methods are perceived to be unhelpful by management or may require
over-simplification.
• repetitive in its architecture for easily changing the number of SC stages.
Basically, the activities of simulation can be depicted as in Figure 2. The computer
simulation is just a model or a function that transforms the inputs into the outputs. The
operational parameters and their variables are described as the inputs and the performances,
which can be derived from simulation, are described as the outputs. The operational
conditions are then tested on this model to achieve the objectives. One objective of the
application of simulation is to search for a set of operational parameters so that system
performance is improved.
252 W. Hachicha et al.

Figure 2 The activities of simulation

Supply chain characteristics

Inputs Outputs
Operational parameters Simulation Performance
and their variables model measures

Presently, there are some studies to advance the activities of simulation and to obtain
optimum performance. Tekin and Ihsen (2004) presented a comprehensive survey on
techniques for simulation optimisation with emphasis given to recent developments.
They classified the existing techniques according to problem characteristics such as
shape of the response surface (global as compared to local optimisation), objective
functions (single or multiple objectives) and parameter spaces (discrete or continuous
parameters). They discussed the major advantages (such as rapid convergence rate, high
search efficiency and robustness, easily applied by well-known statistics tools) and
possible drawbacks (such as local extrema, restriction of the region under investigation,
requirement of prior knowledge about the shape of the response surface) of the different
techniques

3 Literature overview

3.1 The use of DES optimisation method

Literature and practice provide indications that simulation can be used to analyse
complex SC systems (Lee et al., 2002). The use of simulation for SC analysis in recent
years become wealthy in this field: Persson and Olhager (2002) propose a real case study
aiming to evaluate alternative SC scenarios for improving quality and costs and
understanding how these parameters affect each other. Woon et al. (2003) describe the
development of a simulation workbench for modelling and analysing multi-echelon
supply chains. This workbench facilitates study of inventory and forecasting policies
practiced by the echelons, and models different information exchange mechanisms
adopted by them. Reiner (2005) described how process improvements can be dynamically
evaluated under consideration of customer orientation and supported by an integrated
usage of discrete-event simulations models and system dynamics models. Chan and Chan
(2005) used DES for building and testing five different SC models. SC performances are
calculated following a multi-measures based approach. Jammernegg and Reiner (2007)
performed a simulation study of a three-stage supplier network of a company in the
telecommunication and automotive industry. They presented the opportunities and
challenges for improving the performance of SC processes by coordinated application of
inventory management and capacity management. Longo and Mirabelli (2008) presented
an advanced modelling approach and a DES model for supporting SC management
and providing a decision-making tool for SC management. Wang and Ingham (2008)
investigated the customer/manufacturer relationship of the SC and how dynamic
simulation tools can be used to improve and optimise SC performance. Li et al. (2009)
presented a hybrid optimisation approach based on genetic algorithm and DES approach
to optimise the production planning and control policies in remanufacturing system case.
 A multi-product lot size in make-to-order supply chain 253

In DES category, optimisation technique and decision-making tools are strongly

amplified if Design of Experiment and Analysis of Variance are respectively used for
correctly setting parameters levels, number of simulation runs and replications and for
evaluating analytical model to be used for supporting the decision process (Longo and
Mirabelli, 2008). Many recent research articles (e.g. Tsai, 2002; Longo and Mirabelli,
2008; Li et al., 2009) have been written on simulation optimisation method which used
DOE technique.

3.2 Lot-sizing problem in MTO supply chain

To the best of our knowledge, only methods based on queueing theory have been subject
to a mathematical analysis of MTO problems. Williams (1984) noted that classical inventory
control generally ignores capacity constraints and interactions between products. He
considered an M/G/m queuing model that incorporated inventory holding costs, set-up
costs and backorder costs. Lot sizes and reorder points were treated as the decision
variables in cost minimisation. As well, Williams addressed the problem of using
demand patterns to decide whether to choose MTS or MTO. Bertrand (1985) presented
a cost-minimisation model that considered set-up costs and both work-in-process and
finished goods inventory holding costs. Optimal lot sizes were determined through
iteratively estimating lot flow times and then readjusting lot sizes until convergence
occurred. Flow time estimation was based on closed queueing network assumptions and
the lot size optimisation model was based on solving a set of non-linear equations,
derived through partial differentiation of the cost function with respect to product lot
sizes.
Graves et al. (1993) provided a comprehensive review on queueing models for
production systems. Dellaert and Melo (1996) considered a single item capacitated LSP
motivated by a Dutch manufacturer of steel pipes operating in a MTO environment.
The objective is to determine in each period of the planning horizon the optimal size of
production lots so that delivery dates are met as closely as possible with a limited number
of set-ups. To solve this problem, they proposed an exact method based on Markov
Decision Process and dynamic programming for small problems. They also proposed
three heuristic procedures to solve general problems. Ettl et al. (2000) develop a SC
model that took as input the bill-of-materials, the nominal lead times, the demand and
cost data, and the required customer service levels. In return, the model generates the
base-stock level at each store, the stocking location for a component or an end product.
They assumed a distributed inventory control mechanism whereby each site in the
network operates according to a base-stock control policy. This base-stock policy makes
authors avoid the consideration on determining the lot sizes at each store. Srinivasa
Raghavan (2001) presented an analytical method for evaluating the performance of MTO
supply chains using general queueing networks. However, they assumed that demand
arrivals are deterministic.
Missbauer (2002) discussed the need for capacity constrained lot-sizing research
including economic factors. He demonstrated the use of M/G/1 lot-sizing relationships
which minimise the weighted sum of queue times, set-up costs and finished goods holding
costs. Lee et al. (2003) presented a model for computing the parameters of an integrated
inventory replenishment and outbound dispatch scheduling policy under dynamic demand
considerations. Their model can be applied to MTO supply chain attempt to determine
the optimal lot sizes incorporating shipping costs. Choi and Enns (2004) introduced
254 W. Hachicha et al.

relationships to determine lot sizes that minimise costs for the single and multiple product
cases when the production rate is specified. Finally, Dong and Chen (2005) presented an
analytical modelling framework for integrated logistic chains. They developed a network
of inventory-queue models for performance analysis for three cases: base stock policy,
batch-ordering policy and for the LSP. The validity of their analytical model is illustrated
by comparing the results from DES study. However, their framework is used for single
product logistic chains and their performance analysis only. Recently, Liu and Lian
(2009) considered a two-stage distributed manufacturing system under base stock policy.
For modelling, they used a network of inventory-queue model to evaluate the inventory
cost and service level achievable for given inventory control policy. For resolution, they
derived an algorithm to find the optimal inventory control policy that minimises the
overall inventory holding cost and satisfies the given service level requirements.
Some authors allow simulation technique category (not DES techniques) for LSP
solutions. Anupindi and Tayur (1998) used a simulation technique to find good base
stock levels. They combined infinitesimal perturbation analysis with gradient search
which is considered as optimisation tool. Kämpf and Köchel (2006) proposed a
conventional queueing theory models and considered their obtained formulation as too
complex for an analytical solution. Thus, firstly they simplify and restructure their model
by few parameters. Secondly, they applied a simulation technique combined with an
optimisation tool (genetic algorithm) to determine optimal lot sizes solution. As well, the
employ of simulation technique attempt to evaluate each feasible solution. This kind of
simulation technique is different with the DES used in this paper.
The objective of this work is to develop a DES optimisation method for solving the
LSP in MTO supply chain. For this purpose, a DES model was firstly implemented as a
tool in estimating OMFT performance. Secondly, multi-objective desirability optimisation
is achieved by integrating Design of Experiment, Analysis of Variance and desirability
function.

4 The proposed approach

A qualitative description or analysis of the simulation results does not provide a deep
understanding of the SC behaviour and could lead to erroneous conclusions in the
decision-making process. Experiments are natural part of the engineering and scientific
process because they help us in understanding how systems and processes work.
The validity of the decisions taken after an experiment depends strongly on how the
experiment was conducted and how the results were analysed. For these reasons, as
mentioned in Figure 3, the proposed approach suggests to use simulation technique
jointly with the Response Surface Methodology (RSM) and the multi-objective desirability
optimisation. Simulation models are flexible and solve real problems without making too
many restricting assumptions as in most analytical models. The major benefit of using
RSM in post-simulation analysis is the significant reduction in the number of simulation
runs needed (Myers and Montgomery, 1995). The use of RSM allows some
generalisations to be made about the output of the simulation.
 A multi-product lot size in make-to-order supply chain 255

Figure 3 Architecture of the proposed approach

The supply chain

characteristics

(2) Design of (1) Build

Experiment simulation model

(3) Collect simulation

results for each experiment

(4) Response surface A simulation

method confirmation
experiment
(5) Multi-objective
desirability optimisation

Optimise the design and operation of a

supply chain

4.1 Response Surface Methodology (RSM)

RSM is an empirical modelling approach using polynomials as local approximations to
the true input/output relationship. This empirical approach is often adequate for process
improvement in an industrial setting. By careful design of experiments, the objective is to
optimise a response (output variable) that is influenced by several independent variables
(input variables). An experiment is a series of tests, called runs, in which changes are
made in the input variables in order to identify the reasons for changes in the output
response.
The relationship between the response variable of interest (y) and the input variables
(x1, x2... xn) is not usually known. In general, the experimenter approximates the system
function with an empirical model of the form: y = f(x1, x2… xn), where ‘f’ is a first- or
second-order polynomial. This is the empirical or response surface model. The variables
are known as natural variables since they are expressed in physical units of measurement.
In the RSM, the natural variables are transformed into coded variables which are
dimensionless. The successful application of RSM relies on the identification of a
suitable approximation for ‘f’. The ‘f’ function is a low-order polynomial building with
linear regression techniques. The necessary data for building the response models are
generally collected by an experimental design. The most popular of the many classes of
RSM designs is the Central Composite Design (CCD). This is so due to the following
three properties:
1 A CCD can be run sequentially. It can be naturally partitioned into two subsets of
points. The first subset estimates linear and two-factor interaction effects while the
second subset estimates curvature effects. The second subset need not be run when
analysis of the data from the first subset points indicates the absence of significant
curvature effects.
256 W. Hachicha et al.

2 CCDs are very efficient, providing much information on experiment variable effects
and overall experimental error in a minimum number of required runs.
3 CCDs are very flexible. The availability of several varieties of CCDs enables their
use under different experimental regions of interest and operability.
The broad aims of RSM are to investigate the nature of the response surface over a
region of interest and to identify operating conditions associated with maximum or
minimum responses. Several authors have discussed the underlying philosophy and use
of RSM (e.g. Mead and Pike, 1975; Myers et al., 1989; Hood and Welch, 1993; Box and
Liu, 1999; Del Castillo et al., 1999; Myers and Montgomery, 2002; Yang and Tseng,
2002). The methodology involves the use of a variety of statistical techniques, including
experimental design and linear regression, to efficiently discover the unknown relationship
between the factors and the response or to estimate the levels of the treatments (factors)
that optimise the mean system response.
RSM is generally conducted in two phases, as emphasised by Myers and Montgomery
(2002). Phase 1 is concerned with the location of optimum operating conditions by
conducting a sequence of suitable experiments and Phase 2 involves the fitting of an
appropriate empirical model, usually a second-order polynomial model, in order to
examine the nature of the response surface in the vicinity of the optimum.

4.2 Multi-objective desirability optimisation

Desirability appears to have been first proposed as a criterion for response optimisation
by Harrington (1965) and popularised by Derringer and Suich (1980). The first step in
defining a desirability function is to assign values to the responses that reflect their
desirability. The Multi-objective desirability optimisation method involves transformation
of each predicted response, ŷ, to a dimensionless partial desirability function, di, which
includes the researcher’s priorities and desires when building the optimisation procedure.
One or two-sided functions are used, depending on whether each of the n responses has
to be maximised or minimised, or has an allotted target value. If the response i is to be
maximised the quantity di is defined as:
⎧ ⎛ ∧ ⎞
wi

⎪d = ⎜ y − A ⎟ If A ≤ y∧ ≤ B
⎪ i ⎜ B−A ⎟
⎪ ⎝ ⎠
⎪⎪ ∧
⎨d i = 1 If y B (1)
⎪ ∧
⎪d i = 0 If y ≺ A
⎪
⎪
⎪⎩

Likewise, di can be defined when the response is to be minimised or if there is a target

value for the response. In equation (1), A and B are, respectively, the lowest and the
highest values obtained for the response i, and wi is the weight. di ranges between 0, for a
completely undesired response, and 1, for a fully desired response. In both cases, di will
vary non-linearly while approaching the desired value. But with a weight of 1, di varies
linearly. In this work weights are chosen equal to 1 for all responses. The partial
 A multi-product lot size in make-to-order supply chain 257

desirability functions are then combined into a single composite response, the so-called
global desirability function D, defined as the geometric mean of the different di -values as
indicated in equation (2).
1
⎛ n ⎞n
D = ⎜ ∏ d ipi ⎟ (2)
⎝ i =1 ⎠
A value of D different from zero implies that all responses are in a desirable range
simultaneously and, consequently, for a value of D close to 1, the combination of the
different criteria is globally optimum, so the response values are near the target values.
In equation (2), pi is the relative importance assigned to the response i. The relative
importance pi is a comparative scale for weighting each of the resulting di in the overall
desirability product and it varies from the least important (pi = 1) to the most important
(pi = 5). It is noteworthy that the outcome of the overall desirability D depends on the pi
value that offers users flexibility in the definition of desirability functions.

5 Supply chain modelling

The SC configuration investigated in this research is shown in Figure 4. Firstly, Enns and
Suwanruji (2005) provided the same case study in MTS sector. Their problem has been
revised and considerably modified to take into consideration MTO environment.

5.1 Process description

A detailed process description of the SC is a mandatory step for understanding what
is implemented inside and how the simulation model works. The SC under study is
composed of six locations (entities). Two locations, L1 and L2, serve as distribution or
retail function and are exposed to independent customer order. Locations L3, L4, L5 and
L6 serve as production function. These locations may be considered to be constrained
because time delay operations are assumed. The SC under study belongs to multi-stage
category. Each finished product (P1, P2 and P3) has component items (manufacturing
products) as indicated by the Bill-of-Distribution (BOD) and the Bill-of-Material (BOM).
Figure 5 describes BOD and BOM dictating the flow of material. Note that the BOM also
shows the quantity of items required by each finished part. The number in brackets
beside the part number shows the requirements for all finished part order.
Part types P1 and P2 are both derived from P4. P3 is derived from P7. Since P7 is
also a component of P4, the parts going directly to L2 could be considered spare or repair
parts. At L3, P4 is assembled from three unit of P5, one unit of P7 and one P8 parts. At
L5, P7 is produced from two units of P6. At L4, P8 is produced from one unit P6 and P5
is produced from one unit of raw material RM5. At L6, P6 is produced from one unit of
raw material RM6. Supplies of RM5 and RM6 are assumed to be unlimited.

5.2 Model assumption

The current paper considers a situation in which several types of parts are produced at the
same SC. If the production is changed from one type to another, a set-up is needed. For
some reasons, such as a large assortment of parts which is subject to regular changes, a
highly uncertain demand, or unique parts, no safety stocks can be kept and production
scheduling are based on customer’s specifications.
258 W. Hachicha et al.

For specificity, this research makes the following assumptions concerning the
simulation parameters. For all part types, it was assumed that the required lot size order
quantity was not available from the upstream supplier before shipments could be released
(i.e. lot splitting is allowed).
Furthermore, for assembly operations it was assumed that the required lot-size
quantities of all components were not also available before any components were
released for shipment. A common transit time was then applied to all components so
arrival at the assembly location was simultaneous. The transit times are shown along side
the arrows in Figure 4. Moreover, when there are a capacity constraints, lots arriving at
manufacturing locations must undergo a set-up time and a lot processing time. Mean
set-up and part processing times for processed and assembled parts are displayed in
Table 1.

Figure 4 Configuration of the supply chain network

 A multi-product lot size in make-to-order supply chain 259

Figure 5 Bills of distribution and materials

The lot set-up times are stochastic and follow a normal distribution with a coefficient of
variation of 0.3. The lot processing times are deterministic and based on multiplying the
lot size times the fixed part processing time. Processing of all lots in queue is based on
FCFS. Transit times, also shown in Table 1, are defined as the time to move an available
lot of inventory from an upstream location to a downstream location. The transit times
for all part types were assumed to be stochastic and follow a gamma distribution with a
coefficient of variation of 0.1. No capacity constraints were assumed for inventory
transportation.
Furthermore, customer’s order which consists of fixed lot size of same product type
follows a gamma distribution with a mean of 4000 units per week. One week is assumed
to be equal to five working days. Times are given in hours, using the assumption there
are 40 hours per week (or 8 hours per day. Daily order variation is determined on
the basis of having a week order coefficient of variation of 0.1. Based on customer
requirements and specifications, lot sizes for product types P1, P2 and P3 are compulsory
fixed at 200 units, 200 units and 300 units, respectively.
260 W. Hachicha et al.

Table 1 Supply chain data

Mean Setup Part Processing Mean Transit

Part type time (hr) time (hr) Time (hr) Lot size
P1 0.5 0.009 10.0 200
P2 0.5 0.011 13.0 200
P3 0.9 0.023 26.0 300
P4 1.2 0.014 6.5
P5 0.24 0.002 17.0
Unknowns to
P6 0.50 0.003 15.0
be optimised
P7 0.8 0.006 6.5
P8 1.6 0.007 6.5

5.3 Performance measure

Many performance measures (metrics) can be considered for SC analysis such as work-
in-process, mean tardiness, mean flow time and others. The main thing, the coherence
between the criteria of performance guarantees the overall performance of the SC. In
MTO production strategy, the main priority is to minimise mean lot tardiness in order to
avoid associated penalties. On the other hand, it is also important to minimise mean lot
earliness and it is related to extra storage costs. For this reason, the most appropriate
performance measure would be the OMFT. There is a fixed target value (delivery
promise date) proposed for each customer order. This target value must be framed
between the lower and the upper values which are fixed for reduce cost as it is indicated
in Table 2.
Table 2 Order supply chain objectives

Performance measure Lower (hr) Target (hr) Upper (hr)

OMFT P1 57 60 63
OMFT P2 62 65 68
OMFT P3 57 60 63

5.4 Simulation model

The simulation model is developed using Arena 10.0 provided by Rockwell Software
(Law and Kelton, 2000). The global flowchart description is demonstrated in Figure 6.
The main portion of the model’s operation will consist of logic modules to represent
order arrivals, SC locations and order delivery. For instance, the module ‘Create P1
order’ ensures the entry of the product P1 lots within the system. Then, each lot is
assigned by a set of attributes such as part type and sequence routing, via le module
‘Assign P1 attributes’. Additionally, as the part proceeds through the SC, different
attributes record the time delays associated with material handling, set-up and processing
time. All locations can be modelled by a set of machines, which each one is modelled
using the ‘Enter-Process-Leave’ module sequence. Finally, each order leaves the SC
through the module ‘Dispose’.
 A multi-product lot size in make-to-order supply chain 261

Figure 6 Modelling the investigated supply chain by arena

Create P1 order Assign P1 attributes Location 3 Dispose P1 order

Location 1 Location 4
Create P2 order Assign P2 attributes Dispose P2 order
Location 2 Location 5

Create P3 order Assign P3 attributes Location 6 Dispose P3 order

Arrival of the finished product orders Production and distribution Delivery

6 Simulation results analysis and optimisation

Once a simulation model of SC system has been built and verified, it can be used for
analysing the SC performance according to factorial experimental design.

6.1 Design of experiments

Before using DOE notation to divide the parameters into several levels, it is necessary to
assign the variables for the operational parameters. In this study, two levels are chosen
for each manufacturing part: level 1 consists of a lot size of 250 units and level 2 consists
of a lot size 500.
Several simulation runs were made for each SC configuration, each run length is
fixed at 550 000 hours. The results of these simulation runs that are realised with the help
of the simulator Arena 10 were averaged. The result of these runs is shown in Table 3.
Based on the output of the simulation experiments, further analysis is conducted
for the fundamental understanding of SC. Due to multiple decision variables involved,
main effects plots and second-order interaction effects plots are obtained using Minitab
14 software to demonstrate the magnitudes of each factor. For instance, the result of part
P1 is shown in Figure 7. The points in the plot represent the mean of OMFT at the
various levels of each decision variable.

6.2 Multi-objective optimisation

After planning the experiments and identifying the most important factors of the model,
these factors are used as input data for multi-objective desirability optimisation. This
optimisation tool is integrated in many software packages such as Minitab, Statistica,
JMP software, etc.
Applying equation (1) for each response measure, individual desirability for each
product is equal to 1.0. The response optimisation consists in determining how the
solution has satisfied the combined goals for all the responses. Composite desirability has
a range of zero to one. One represents the ideal case; zero indicates that one or more
responses are outside their acceptable limits. Composite desirability is the weighted
geometric mean of the individual desirability for the responses as presented in
262 W. Hachicha et al.

equation (2). The composite desirability for all these three variables is 1.0. To obtain this
desirability, we would set the factor levels at the values shown under global solution in
the Figure 8. That is, Lot size of P4 would be set at 562 units. Lot size of P5 would be set
at 77.5 units, Lot size of P6 would be set at 274 units, Lot size of P7 would be set at
212.5 units and Lot size of P8 would be set at 675 units.

Figure 7 Main effect plots and interaction plots for OMFT: case of product P1
(see online version for colours)
Main Effects Plot (data means) for p1

p8 p7 p6
68

66

64

62
Mean of p1

60

0,5 1,0 1,5 2,0 2,5 0,5 1,0 1,5 2,0 2,5 0,5 1,0 1,5 2,0 2,5

p5 p4
68

66

64

62

60

0,5 1,0 1,5 2,0 2,5 0,5 1,0 1,5 2,0 2,5

Interaction Plot (data means) for p1

0,5 1,0 1,5 2,0 2,5 0,5 1,0 1,5 2,0 2,5
68
p8
0, 5
1, 0
64 1, 5
p8 2, 0
2, 5
60
68
p7
0, 5

64 1, 0
1, 5
p7 2, 0

60 2, 5

68
p6
0, 5

64 1, 0
1, 5
p6 2, 0
60 2, 5

68
p5
0, 5

64 1, 0
1, 5
p5 2, 0
60 2, 5

68
p4
0, 5

64 1, 0
1, 5
p4 2, 0
60 2, 5

0,5 1,0 1,5 2,0 2,5 0,5 1,0 1,5 2,0 2,5 0,5 1,0 1,5 2,0 2,5
 Table 3
Manufacturing part lot levels MFT for each Finished parts (hr) Manufacturing part lot levels MFT for each Finished parts (hr)
Exp. P4 P5 P6 P7 P8 P1 P2 P3 Exp. P4 P5 P6 P7 P8 P1 P2 P3
1 1 1 2 2 1 60.49 64.96 55.06 27 1 1 1 2 2 63.14 63.33 54.84
2 1 1 2 1 1 61.89 63.99 55.15 28 1.5 0.31 1.5 1.5 1.5 63.29 65.04 56.48
3 1 1 1 1 1 60.72 65.65 55.96 29 2 1 1 2 1 60.48 64.92 58.22
4 1.5 1.5 1.5 0.31 1.5 62.38 64.80 63.08 30 2 2 2 1 1 60.70 64.68 58.66
5 1.5 1.5 1.5 1.5 1.5 60.98 63.34 56.55 31 1 2 2 2 1 60.03 64.20 55.21
6 1.5 1.5 1.5 1.5 1.5 60.98 63.34 56.55 32 1.5 1.5 0.31 1.5 1.5 62.40 66.33 59.17
7 1 2 2 1 2 60.23 66.29 55.77 33 2 2 2 2 1 61.50 65.24 55.19
8 2 1 1 1 1 60.75 64.83 59.08 34 1 2 1 1 2 61.38 65.66 54.89
9 1 2 2 1 1 60.23 66.29 55.77 35 2 2 1 1 1 61.57 65.42 59.44
10 1 1 1 2 1 63.14 63.33 54.84 36 1 1 2 1 2 61.89 63.99 55.15
The 25 response surface design

11 1.5 1.5 1.5 1.5 1.5 60.98 63.34 56.55 37 1 2 1 2 2 60.30 65.25 55.25
12 1 2 1 1 1 61.38 65.66 54.89 38 1 1 2 2 2 60.49 64.96 55.06
13 2 1 2 1 1 60.35 64.34 56.18 39 1.5 1.5 1.5 1.5 0.31 60.98 63.34 56.55
14 2 1 1 1 2 60.75 64.83 59.08 40 1.5 1.5 1.5 1.5 1.5 60.98 63.34 56.55
15 2.7 1.5 1.5 1.5 1.5 59.37 64.77 61.06 41 1.5 1.5 1.5 1.5 1.5 60.98 63.34 56.55
16 1.5 1.5 1.5 1.5 1.5 60.98 63.34 56.55 42 1.5 1.5 2.7 1.5 1.5 60.51 66.01 57.35
17 1 2 1 2 1 59.67 64.01 58.08 43 1.5 2.7 1.5 1.5 1.5 61.85 61.42 56.40
18 1.5 1.5 1.5 1.5 1.5 60.98 63.34 56.55 44 1.5 1.5 1.5 1.5 1.5 60.98 63.34 56.55
19 2 2 1 2 1 59.67 64.01 58.08 45 2 1 2 2 2 60.22 64.16 57.23
A multi-product lot size in make-to-order supply chain

20 2 2 1 2 2 59.67 64.01 58.08 46 1.5 1.5 1.5 2.7 1.5 59.69 64.63 56.66
21 2 2 2 1 2 60.70 64.68 58.66 47 2 1 2 1 2 60.35 64.34 56.18
22 1 2 2 2 2 60.03 64.20 55.21 48 1.5 1.5 1.5 1.5 1.5 60.98 63.34 56.55
23 2 2 1 1 2 61.57 65.42 59.44 49 1.5 1.5 1.5 1.5 1.5 60.98 63.34 56.55
24 0.31 1.5 1.5 1.5 1.5 63.54 72.14 56.24 50 2 1 1 2 2 60.48 64.92 58.22
25 2 2 2 2 2 61.50 65.24 55.19 51 1.5 1.5 1.5 1.5 2.7 60.98 63.34 56.55
26 2 1 2 2 1 60.22 64.16 57.23 52 1 1 1 1 2 60.72 65.65 55.96
263
264 W. Hachicha et al.

Figure 8 Multi-objective optimisation based on desirability functions

7 Conclusion

This paper presents a new approach for the LSP in MTO supply chain. The proposed
approach is based on combining discrete event simulation and response surface
methodology for a multi-objective desirability optimisation. The supply chain under
study which operates in MTO environment (no possibility for stock keeping and limited
production capacity) is characterised by multi-product, multi-stage, multi-location
production planning with capacity-constrained and stochastic parameters such as lot
arrivals order, transit time, set-up time, processing time, etc.
The proposed approach is a logical and methodical to solve LSP, which makes it
easily portable into practice. For that reason, design of experiment, analyse of variance
and desirability optimisation tool are available in many commercial software packages
including Minitab, Statistica and others.
Our perspective should give more extensive testing on different problems to support
the proposed approach application in supply chain management. Details and others real
benefits of the application of the metamodel-based simulation optimisation approach into
real-life case study will be presented at our subsequent publications.
 A multi-product lot size in make-to-order supply chain 265

References
Anupindi, R. and Tayur, S. (1998) ‘Managing stochastic multi-product systems: model measures,
and analysis’, Operations Research, Vol. 46, pp.98–111.
Bertrand, J.W.M. (1985) ‘Multi product optimal batch sizes with in-process inventories and multi
work centers’, IIE Transactions, Vol. 17, No. 2, pp.157–163.
Box, G.E.P. and Liu, P.Y.T. (1999) ‘Statistics as a catalyst to learning by scientific method
part I – an example’, Journal of Quality Technology, Vol. 31, No. 1, pp.1–15.
Chan, F.T.S. and Chan, H.K. (2005) ‘Simulation modeling for comparative evaluation of supply
chain management strategies’, International Journal of Advanced Manufacturing and
Technology, Vol. 25, pp.998–1006.
Choi, S. and Enns, S.T. (2004) ‘Multi-product capacity-constrained lot sizing with economic
objectives’, International Journal of Production Economics, Vol. 91, pp.47–62.
Dellaert, N.P. and Melo, M.T. (1996) ’Production strategies for a stochastic lot-sizing problem with
constant capacity’, European Journal of Operational Research, Vol. 92, pp.281–301.
Derringer, G.C. and Suich, R. (1980) ‘Simultaneous optimization of several response variables’,
Journal of Quality Technology, Vol. 12, pp.214–219.
Del Castillo, E., Fan, S.K. and Semple, J. (1999) ‘Optimization of dual response systems:
a comprehensive procedure for degenerate and no degenerate problems’, European Journal of
Operational Research, Vol. 112, No. 1, pp.174–186.
Dong, M. and Chen, F.F. (2005) ‘Performance modeling and analysis of integrated logistic chains:
an analytic framework’, European Journal of Operational Research, Vol. 162, pp.83–98.
Enns, S.T. and Suwanruji, P. (2005) ‘Lot-sizing within capacity constrained manufacturing
systems using time-phased planning’, Proceedings of the 2005 Winter Simulation Conference,
pp.1433–1440.
Ettl, M., Feigin, G.E., Lin, G.Y. and Yao, D.D. (2000) ‘A supply network model with base-stock
control and service requirements’, Operations Research, Vol. 48, pp.216–232.
Gunasekaran, A. and Ngai, E.W.T. (2009) ‘Modeling and analysis of build-to-order supply chains’,
European Journal of Operational Research, Vol. 195, No. 2, pp.319–334.
Graves, S.C., Rinnooy Kan, A.H.G. and Zipkin, P.H. (1993) Logistics of Production and Inventory,
Handbooks in Operations Research and Management Science, Elsevier Science, Holland.
Harrington, J. (1965) ‘The desirability function’, Industrial Quality Control, Vol. 21, pp.494–498.
Ho, C. (1993) ‘Evaluating lot-sizing performance in multilevel MRP systems: a comparative
analysis of multiple performance measures’, International Journal of Operations and
Production Management, Vol. 13, No. 11, pp.52–79.
Hood, S.J. and Welch, P.D. (1993) ‘Response surface methodology and its application in
simulation’, in Evans, G.W., Mollaghasemi, M., Rusell, E.C. and Biles, W.E. (Eds):
Proceedings of the 1993 Winter Simulation Conference, pp.115–122.
Jammernegg, W. and Reiner, G. (2007) ‘Performance improvement of supply chain processes
by coordinated inventory and capacity management’, International Journal of Production
Economics, Vol. 108, pp.183–190.
Karimi, B., Ghomi, S.M.F.T. and Wilson, J.M. (2003) ‘The capacitated lot sizing problem:
a review of models and algorithms’, Omega, Vol. 31, pp.365–378.
Kämpf, M. and Köchel, P. (2006) ‘Simulation-based sequencing and lot size optimization for a
production-and-inventory system with multiple items’, International Journal of Production
Economics, Vol. 104, pp.191–200.
Law, A.M. and Kelton, W.D. (2000) Simulation Modeling and Analysis, McGraw Hill, New York.
Lee, Y.H., Cho, M.K., Kim, S.J. and Kim, Y.B. (2002) ‘Supply chain simulation with discrete
continuous combined modeling’, Computer and Industrial Engineering, Vol. 43, pp.375–392.
266 W. Hachicha et al.

Lee, C.Y., Cetinkaya, S. and Jaruphongsa, W. (2003) ‘A dynamic model for inventory lot-sizing
and outbound shipment scheduling at a third-party warehouse’, Operations Research, Vol. 51,
No. 5, pp.735–747.
Li, J., González, M. and Zhu, Y. (2009) ‘A hybrid simulation optimization method for production
planning of dedicated remanufacturing’, International Journal of Production Economics,
Vol. 117, No. 2, pp.286–301.
Liu, X. and Lian, Z. (2009) ‘Cost-effective inventory control in a value-added manufacturing
system’, European Journal of Operational Research, Vol. 196, pp.534–543.
Longo, F. and Mirabelli, G. (2008) ‘An advanced supply chain management tool based on
modeling and simulation’, Computers and Industrial Engineering, Vol. 54, pp.570–588.
Mead, R. and Pike, D.J. (1975) ‘A review of response surface methodology from a biometric
viewpoint’, International Biometric Society, Vol. 31, pp.803–851.
Missbauer, H. (2002) ‘Lot sizing in workload control systems’, Production Planning and Control,
Vol. 13, No. 7, pp.649–664.
Montgomery, D.C. (2005) Design and Analysis of Experiments, 6th ed., Wiley & Sons, Inc.,
pp.160–335.
Myers, R.H. and Montgomery, D.C. (2002) Response Surface Methodology: Process and Product
Optimization Using Designed Experiments, 2nd ed., Wiley, New York.
Myers, R.H., Khuri, A.I. and Carter Jr., W.H. (1989) ‘Response surface methodology:1966–1988’,
Technometrics, Vol. 31, No. 2, pp.137–157.
Persson, F. and Olhager, J. (2002) ‘Performance simulation of supply chain designs’, International
Journal of Production Economics, Vol. 77, pp.231–245.
Reiner, G. (2005) ‘Customer-oriented improvement and evaluation of supply chain processes
supported by simulation models’, International Journal of Production Economics, Vol. 96,
pp.381–395.
Srinivasa Raghavan, N.R. (2001) ‘Generalized queuing network analysis of integrated supply
chains’, International Journal of Production Research, Vol. 39, No. 2, pp.205–224.
Tekin, E. and Ihsan, S. (2004) ‘Simulation optimization: a comprehensive review on theory and
applications’, IIE Transactions, Vol. 36, pp.1067–1081.
Tsai, C.S. (2002) ‘Evaluation and optimization of integrated manufacturing system operations
using Taguch’s experiment design in computer simulation’, Computers and Industrial
Engineering, Vol. 43, pp.591–604.
Wang, Q. and Ingham, N. (2008) ‘A discrete event modeling approach for supply chain
simulation’, International Journal of Simulation Modeling, Vol. 7, No. 3, pp.124–134.
Williams, T.M. (1984) ‘Special products and uncertainty in production/inventory systems’,
European Journal of Operational Research, Vol. 15, pp.46–54.
Woon, K.N., Piplani, R. and Viswanathan, S. (2003) ‘Simulation workbench for analyzing
multi-echelon supply chains’, Integrated Manufacturing Systems, Vol. 14, No. 5, pp.449–457.
Yang, T. and Tseng, L. (2002) ‘Solving a multi-objective simulation model using a hybrid response
surface method and a lexicographical goal programming approach-a case study on
integrated circuit ink-marking machines’, European Journal of Operational Research,
Vol. 53, pp.211–221.