This article was downloaded by: [Texas A&M University-Commerce]

On: 20 July 2013, At: 04:35

Publisher: Taylor & Francis
Informa Ltd Registered in England and Wales Registered Number: 1072954 Registered
office: Mortimer House, 37-41 Mortimer Street, London W1T 3JH, UK

International Journal of Production

Research
Publication details, including instructions for authors and
subscription information:
http://www.tandfonline.com/loi/tprs20

Multi-level supply chain network design

with routing
a b a
Jeong-Hun Lee , Il-Kyeong Moon & Jong-Heung Park
a
Postal Technology Research Center, ETRI, Daejeon 305–700,
Korea
b
Department of Industrial Engineering, Pusan National University,
Busan 609–735, Korea
Published online: 10 Jun 2009.

To cite this article: Jeong-Hun Lee , Il-Kyeong Moon & Jong-Heung Park (2010) Multi-level supply
chain network design with routing, International Journal of Production Research, 48:13, 3957-3976,
DOI: 10.1080/00207540902922851

To link to this article: http://dx.doi.org/10.1080/00207540902922851

PLEASE SCROLL DOWN FOR ARTICLE

Taylor & Francis makes every effort to ensure the accuracy of all the information (the
“Content”) contained in the publications on our platform. However, Taylor & Francis,
our agents, and our licensors make no representations or warranties whatsoever as to
the accuracy, completeness, or suitability for any purpose of the Content. Any opinions
and views expressed in this publication are the opinions and views of the authors,
and are not the views of or endorsed by Taylor & Francis. The accuracy of the Content
should not be relied upon and should be independently verified with primary sources
of information. Taylor and Francis shall not be liable for any losses, actions, claims,
proceedings, demands, costs, expenses, damages, and other liabilities whatsoever or
howsoever caused arising directly or indirectly in connection with, in relation to or arising
out of the use of the Content.

This article may be used for research, teaching, and private study purposes. Any
substantial or systematic reproduction, redistribution, reselling, loan, sub-licensing,
systematic supply, or distribution in any form to anyone is expressly forbidden. Terms &
Conditions of access and use can be found at http://www.tandfonline.com/page/terms-
and-conditions
 International Journal of Production Research
Vol. 48, No. 13, 1 July 2010, 3957–3976

Multi-level supply chain network design with routing

Jeong-Hun Leea, Il-Kyeong Moonb* and Jong-Heung Parka
a
Postal Technology Research Center, ETRI, Daejeon 305–700, Korea; bDepartment of
Industrial Engineering, Pusan National University, Busan 609–735, Korea
(Received 5 September 2008; final version received 24 March 2009)
Downloaded by [Texas A&M University-Commerce] at 04:35 20 July 2013

Recently, the multi-level and multi-facility industrial problem in supply chain

management (SCM) has been widely investigated. One of the key issues, central
to this problem in the current SCM research area is the interdependence
among the location of facilities, the allocation of facilities, and the vehicle
routing for the supply of raw materials and products. This paper studies the
supply chain network design problem, which involves the location of facilities,
allocation of facilities, and routing decisions. The proposed problem has
some practical applications. For example, it is necessary for third party logistics
(3PL) companies to manage the design of the network and to operate vehicle
transportation. The purpose of this study is to determine the optimal location,
allocation, and routing with minimum cost to the supply chain network. The
study proposes two mixed integer programming models, one without routing
and one with routing, and a heuristic algorithm based on LP-relaxation in order
to solve the model with routing. The results show that a developed heuristic
algorithm is able to find a good solution in a reasonable time.
Keywords: supply chain network design; vehicle routing; LP-relaxation; heuristic
algorithm

1. Introduction
The problem with network configuration (design) is the need to specify the structure
through which products flow from their source points to their demand points. This
involves determining the facilities, if any, to be used; how many there should be, where
they should be located, the products and customers assigned to them, the transport
services used between them, the sourcing, inter-facility, and distribution to customers
product flows, and the inventory levels maintained in the facilities (Ballou 2004).
In recent years, many developments in logistics were connected to the need for
information in an efficient supply chain flow. The supply chain is often represented as
a network known as a supply chain network (SCN) that comprises nodes that represent
facilities (suppliers, manufacturers, distribution centres and customers). A multi-level SCN
is a sequence of multiple SCN levels. The flow can only be transferred between two
consecutive stages. The multi-level SCN problem involves the choice of facilities
(manufacturers and distribution centres) to be opened and the distribution network
design that must satisfy the demand with minimum cost. Today’s logistics environments
are characterised by globalisation. Therefore, the competitive strategy of manufacturers

*Corresponding author. Email: ikmoon@pusan.ac.kr

ISSN 0020–7543 print/ISSN 1366–588X online

ß 2010 Taylor & Francis
DOI: 10.1080/00207540902922851
http://www.informaworld.com
 3958 J.-H. Lee et al.
Downloaded by [Texas A&M University-Commerce] at 04:35 20 July 2013

Figure 1. Circumstances of global supply chain network.

and distribution centres has a significant impact on network design decisions within the
supply chain (Chopra and Meindl 2004). Manufacturers and distribution centres tend
to focus on cost leadership and on finding the lower labour cost and closer location
for their manufacturing and distribution facilities (Figure 1). Both academics and
practitioners have recognised the interdependence between the manufacturer, the location
of the distribution centre, and vehicle routing. However, no attempts have been made to
incorporate routing in the location analysis of whole supply chain networks. This study
considers four level supply chains that have multi-suppliers, multi-manufacturers, multi-
distribution centres, and multi-customers. The solution to the supply chain network design
with routing (SCNDR) problem can be defined as one that consists of the location and
vehicle routing problems of manufacturers and distribution centres which have to be
solved simultaneously. A much more general form of the manufacturer location model
needs to be considered in the design of the entire supply chain network, from the supplier
to the customer. The study considers a supply chain in which suppliers send materials to
manufacturers by vehicle and manufacturers send the products to distribution centres by
ship and distribution centres send products to customers by vehicle. Location and capacity
allocation decisions need to be made for manufacturers and distribution centres. Multiple
distribution centres may be used to satisfy customer demand and multiple manufacturers
may be used to transport to distribution centres. It is also assumed that units have been
appropriately adjusted so that one unit of input from a supply source produces one unit
of finished product.
This study presents a mixed integer programming formulation for the SCNDR
problem. This problem combines the location-allocation problem (LAP) and the multi-
depot vehicle routing problem (MDVRP). It is a larger and more complex problem, which
 International Journal of Production Research 3959
Downloaded by [Texas A&M University-Commerce] at 04:35 20 July 2013

Figure 2. Demonstration of the SCNDR problem.

cannot be solved using existing mixed integer programming techniques. The SCNDR
problem proposed here is an NP-hard problem combining the capacitated LAP with
MDVRP. Consequently, this study presents a heuristic algorithm that offers an excellent
solution quality for the SCNDR problem, in a reasonable amount of computational time.
The problems of each supply chain level to be solved by this paper are as demonstrated
in Figure 2.

2. Literature review
The competitive nature of today’s business environment has resulted in an increase in
cooperation among individual companies as members of a supply chain. Accordingly,
third party logistics (3PL) companies must operate supply chains for a number of different
members who wish to improve their logistics operations while decreasing the related costs.
As a result of the dynamic environment in which these supply chains must operate, 3PL
companies must make a sequence of inter-related decisions over time.
All companies that aim to be competitive in the market need to pay attention to their
organisation as it relates to the entire supply chain. In particular, companies need to
analyse the supply chain in order to improve the customer service level without
a concurrent uncontrolled growth in costs. In short, companies need to increase the
efficiency of their logistics operations. It is therefore of fundamental importance to
optimise the flow of goods (and information) among the actors of the supply chain,
 3960 J.-H. Lee et al.

(the suppliers, manufacturers, distribution centres, and customers). Ambrosino and

Scutella (2005) stated that vertical as well as horizontal integrations are required for the
optimisation of the flows in the supply chain, and for the optimisation of all related
activities. This implies that agreements need to be reached between subjects that operate
at different levels of the supply chain (vertical integration), and between actors of the
same level (horizontal integration).
Distribution network design problems involve strategic decisions, which influence
tactical and operational decisions (Crainic and Laporte 1997). In particular, they involve
facility location, transportation and inventory decisions, which affect the cost of the
distribution system and the quality of customer service. Therefore, distribution network
design problems are central for each company. Abdelmaguid and Dessouky (2006)
introduced a new genetic algorithm (GA) approach using a randomised version of
Downloaded by [Texas A&M University-Commerce] at 04:35 20 July 2013

a previously developed construction heuristic to generate the initial random population for
the integrated inventory distribution problem. Ko and Lee (2007) developed a network
design model for an express courier service company using a recursive optimisation and
simulation procedure.
Sha and Che (2006) proposed a mathematical model using the genetic algorithm, the
analytical hierarchy process, and the multi-attribute utility theory in order to solve the
fundamental multi-echelon supply chain network problem that maximises the total overall
utility of fulfilling demands. However, the routing problem is not considered in their
model. Canel and Khumawala (1997) proposed the international facility location problem
(IFLP) model by including multiple periods so that timing of location changes can be more
carefully evaluated. Webb (1968) and, more recently, Salhi and Rand (1989), recognised
the error introduced into location problems when the interdependence between routing
and location decisions is ignored. Since the aforementioned research was published, some
further research has focused on the relationships between facilities and transportation
costs, stressing that the location of distribution facilities and the routing of vehicles from
facilities require decisions to be made interdependently. In particular, in recent years, some
location-routing problems (LRP) arising in the context of distribution network design
problems have been investigated. Fields other than logistics, such as the communication
network design and ship routing also have a tendency to consider problems related to
routing between facilities (Lee et al. 2002, Gunnarsson et al. 2006, Wang et al. 2007,
Lee et al. 2008). Over the past few decades, a great deal of research has been carried out
relating to the location-routing problem (LRP). The LRP consists of two sub-problems;
the facility location problem (FLP) and the vehicle routing problem (VRP). Both of these
are shown to be NP-hard (Karp 1972). Therefore, the SCNDR problem also belongs to the
class of NP-hard problems since the SCNDR problem consists of the capacity location-
allocation problem (CLAP) and MDVRP in the multi-level supply chain. Laporte et al.
(1988) studied how the multi-depot vehicle-routing problem and LRP can be transformed
into a constrained assignment problem through graphical representation, which can solve
up to 80 demand nodes. Another LRP model involves the use of the Clarke-Wright savings
method in a case with stochastically processed demands (Chan et al. 2001). Exact
solutions can only be obtained for not very large problems (from mixed integer
programming or branch-and-bound algorithm with special branching rules). These can
serve as lower bounds for heuristics validation. Breedam (2001) suggested building hybrid
meta-heuristics in order to combine their best features. Wu et al. (2002) presented
a method for solving the multi-depot location-routing problem (MDLRP). They relaxed
some unrealistic assumptions such as homogeneous fleet type and limited number of
 International Journal of Production Research 3961

available vehicles. They solved the problem by dividing it into two sub-problems, i.e., the
location-allocation problem and the general vehicle routing problem. Each sub-problem is
then solved in a sequential and iterative manner by the simulated annealing algorithm.
Meta-heuristics have become prominent approaches in tackling complex, multi-objective
problems (Jones et al. 2002). Lin and Kwok (2006) applied meta-heuristics of a tabu search
and simulated annealing on real data and simulated data, so as to compare their
performances under both simultaneous and sequential routing. They also proposed a new
statistical procedure to compare the two algorithms on the strength of their multi-objective
solutions.
The subsequent information in this paper is organised into the following sections.
In Section 3, the study develops two mixed integer programming models for the supply
chain network design (SCND) problem and the SCNDR problem. In Section 4, we
Downloaded by [Texas A&M University-Commerce] at 04:35 20 July 2013

develop a heuristic algorithm for the developed SCNDR model. Computational

experiments for the developed heuristic algorithm are presented in Section 5. Finally,
conclusions from this study are presented in Section 6.

3. Mathematical models
The study uses an assumed situation for formulating a mixed integer programming
model. In the supply chain network, there are multi-suppliers, multi-manufacturers, multi-
distribution centres (DCs), and multi-customers. We consider a single item and a capacity
limitation for the supplier, the manufacturer, and the DC. The quantity of product
transported to a customer must cover the demand, while a single DC supplies one customer.
The objective of this study is to find a solution that will minimise the sum of the
transportation (routing) cost, variable (operating) cost, and the fixed cost. The trans-
portation (routing) cost is dependent on the distance between the two selected points.
Fixed cost (e.g., for acquiring land, for infrastructure construction or for leasing existing
facilities) will be incurred when these facilities are used. The variable cost of operation for
manufacturers and for DCs is assumed to be linear. The study developed two mixed integer
programming models (one without routing and one with routing) in order to reveal how the
SCNDR model (with routing) can reduce the total cost. Given the previous problem
statement, an approach used for solving the problem is to formulate the problem using
a mixed integer programming model. The model used for solving the problem is explained
below.

3.1 SCND (supply chain network design) model – without routing

The notation is given as follows:
Indices
i index of suppliers;
I set of suppliers (i 2 I );
j index of potential manufacturers;
J set of potential manufacturers ( j 2 J );
k index of potential distribution centres;
K set of potential distribution centres (k 2 K );
l index of customers;
L set of customers (l 2 L).
 3962 J.-H. Lee et al.

Costs
(1) Transportation cost:
TCSM
ij transportation cost per mile from supplier i to manufacturer j, for all
i 2 I, j 2 J;
TCMD
jk transportation cost per mile from manufacturer j to distribution
centre k, for all j 2 J, k 2 K;
TCDC
kl transportation cost per mile from distribution centre k to customer l,
for all k 2 K, l 2 L.
(2) Fixed cost:
Downloaded by [Texas A&M University-Commerce] at 04:35 20 July 2013

FCM
j fixed cost of opening manufacturer j, for all j 2 J;
FCD
k fixed cost of opening distribution centre k, for all k 2 K.
(3) Variable cost:

VCM
j variable cost (operating cost) per unit at manufacturer j, for all j 2 J;
VCD
k variable cost (operating cost) per unit at distribution centre k, for all
k 2 K.
Parameters
(1) Capacity & demand:
CSi supply capacity at supplier i, for all i 2 I;
CM
j production capacity at manufacturer j, for all j 2 J;
CDk capacity at distribution centre k, for all k 2 K;
CCl demand of customer l, for all l 2 L;
CVM
j capacity of vehicle at manufacturer j, for all j 2 J;
CVDk capacity of vehicle at distribution centre k, for all k 2 K.
(2) Distance:
d SM
ij distance from supplier i to manufacturer j (mile), for all i 2 I, j 2 J;
d MD
jk distance from manufacturer j to distribution centre k (mile), for all
j 2 J, k 2 K;
dDC
kl distance from distribution centre k to customer l (mile), for all k 2 K,
l 2 L;
BM a large number.

Decision variables
X SM
ij quantity transported from supplier i to manufacturer j, for all i 2 I, j 2 J;
X MD
jk quantity transported from manufacturer j to distribution centre k,
for all j 2 J, k 2 K;
X DC
kl quantity transported from distribution centre k to customer l, for all
k 2 K, l 2 L;
 International Journal of Production Research 3963

Y SM
ij 1, if supplier i is allocated to manufacturer j, for all i 2 I, j 2 J;
0, otherwise;
Y MD
jk 1, if manufacturer j is allocated to distribution centre k, for all j 2 J,
k 2 K; 0, otherwise;
Y DC
kl 1, if customer l is allocated to distribution centre k, for all k 2 K, l 2 L;
0, otherwise;
ZM
j 1, if manufacturer j is opened, for all j 2 J; 0, otherwise;
ZDk 1, if distribution centre k is opened, for all k 2 K; 0, otherwise.
The SCND model is formulated as the following mixed integer programming model:
XX XX XX
min 2TC SM SM SM
ij d ij Y ij þ TCMD MD MD
jk d jk Y jk þ 2TCDC DC DC
kl d kl Y kl
Downloaded by [Texas A&M University-Commerce] at 04:35 20 July 2013

i2I j2J j2J k2K k2K l2L

! !
X X X X
þ VCM
j CSi Y SM
ij þ VCD
k CC DC
l Y kl
j2J i2I k2K l2L
X X
þ FCM M
j Zj þ FCD D
k Zk ð1Þ
j2J k2K

subject to:
X
X SM S
ij  Ci , all i 2 I ð2Þ
j2J

X
X MD M M
jk  Cj Zj , all j 2 J ð3Þ
k2K

P
X DC D D
kl  Ck Zk , all k 2 K ð4Þ
l2L

X
X DC C
kl  C l , all l 2 L ð5Þ
k2K

X X
X SM
ij  X MD
jk  0, all j 2 J ð6Þ
i2I k2K

X X
X MD
jk  X DC
kl  0, all k 2 K ð7Þ
j2J l2L

X
YDC
kl ¼ 1, all l 2 L ð8Þ
k2K

X
X SM M
ij  CVj , all j 2 J ð9Þ
i2I

X
X DC D
kl  CVk , all k 2 K ð10Þ
l2L

X SM SM
ij  BMY ij , all i 2 I, j 2 J ð11Þ
 3964 J.-H. Lee et al.

X MD MD
jk  BMY jk , all j 2 J, k 2 K ð12Þ

X DC DC
kl  BMY kl , all k 2 K, l 2 L ð13Þ

Y SM MD DC M D
ij , Y jk , Y kl , Z j , Z k 2 f0, 1g, all i 2 I, j 2 J, k 2 K, l 2 L: ð14Þ

The objective function, Equation (1), minimises the sum of the costs required to
transfer the raw materials/products from the source sites (suppliers and DCs) to the
destination sites (manufacturers and customers), the fixed costs for opening and the
variable cost for operating facilities. The transportation costs involve the cost of
transporting the raw materials from the supplier to the manufacturer, the final product
Downloaded by [Texas A&M University-Commerce] at 04:35 20 July 2013

from the manufacturer to the distribution centre, and from the DC to the customer.
We consider the cost of the round trip in the transportation cost to compare it with the
SCNDR model in which the routing cost is considered. Constraint (2) specifies that the
total amount of product transported from a supplier cannot exceed the supplier’s capacity.
Constraint (3) specifies that the amount produced by the manufacturer cannot exceed its
capacity. Constraint (4) specifies that the amount transported through a distribution
centre cannot exceed its capacity. Constraint (5) specifies that the amount transported
to a customer must cover the demand. Constraints (6) and (7) are flow conservation
constrains between suppliers and manufacturers. They state that the amount transported
out of a manufacturer cannot exceed the quantity of raw material received from the
suppliers. Between manufacturers and distribution centres, they specify that the amount
transported from a distribution centre cannot exceed the quantity received from the
manufacturers. Constraint (8) ensures that each customer is supplied by a single
distribution centre. Constraints (9) and (10) are the vehicle capacity of the manufacturer
and the DC. Constraints (11), (12), and (13) are linking constraints for Y and X (if Y is 1, X
has some value, otherwise X is 0). Constraint (14) specifies whether or not there is
transportation between points of each level and that each manufacturer or distribution
centre is either open or closed.

3.2 SCNDR (supply chain network design with routing) model – with routing
The notation is given as follows:
Indices
i index of suppliers;
I set of suppliers (i 2 I );
j index of potential manufacturers;
J set of potential manufacturers ( j 2 J );
k index of potential distribution centres;
K set of potential distribution centres (k 2 K );
l index of customers;
L set of customers (l 2 L);
g, h index of suppliers and manufacturers;
G, H set of suppliers and manufacturers (G, H ¼ I [ J );
s, t index of distribution centres and customers;
S, T set of distribution centres and customers (S, T ¼ K [ L);
 International Journal of Production Research 3965

vM index of vehicles belonging to manufacturers;

VM set of vehicles belonging to manufacturers (vM 2 VM);
vD index of vehicles belonging to distribution centres;
VD set of vehicles belonging to distribution centres (vD 2 VD).
Costs
(1) Transportation cost:
TCSM
gh routing cost per mile from point g to point h, for all g 2 G, h 2 H;
TCMD
jk transportation cost per mile from manufacturer j to distribution centre
k, for all j 2 J, k 2 K;
TCDC
st routing cost per mile from point s to point t, for all s 2 T, t 2 T
Downloaded by [Texas A&M University-Commerce] at 04:35 20 July 2013

(2) Fixed cost:

FCM
j fixed cost of opening manufacturer j, for all j 2 J;
FCD
k fixed cost of opening distribution centre k, for all k 2 K.
(3) Variable cost
VCM
j variable cost (operating cost) per unit at manufacturer j, for all
j 2 J;
VCD
k variable cost (operating cost) per unit at distribution centre k, for all
k 2 K.

Parameters
(1) Capacity & demand:
CSi supply capacity at supplier i, for all i 2 I;
CM
j production capacity at manufacturer j, for all j 2 J;
CD
k capacity at distribution centre k, for all k 2 K;
CC l demand of customer l, for all l 2 L;
CV M
v M capacity of vehicle vM at manufacturer, for all vM 2 VM;
CV D
vD capacity of vehicle vD at distribution centre, for all vD 2 VD.
(2) Distance:
d SM
gh distance from point g to point h, for all g 2 G, h 2 H;
d MD
jk distance from manufacturer j to distribution centre k, for all j 2 J,
k 2 K;
d DC
st distance from point s to point t, for all s 2 T, t 2 T.
(3) Temporary:
UivM, UgvM auxiliary variables for sub-tour elimination constraints in vehicle vM
for suppliers and manufacturers;
WsvD, WlvD auxiliary variables for sub-tour elimination constraints in vehicle vD
for distribution centres and customers;
BM a large number.
 3966 J.-H. Lee et al.

Decision variables
X SM
ghvm 1, if point g immediately precedes point h on vehicle vM for all g 2 G,
h 2 H; 0, otherwise;
X MD
jk 1, if manufacturer j transport to distribution centre k, for all j 2 J,
k 2 K; 0, otherwise;
X DC
stvD 1, if point s immediately precedes point t on vehicle vD, for all s 2 T,
t 2 T; 0, otherwise;
YSM
ij 1, if supplier i is allocated to manufacturer j, for all i 2 I, j 2 J; 0,
otherwise;
YMD
jk transportation quantity from manufacturer j to distribution centre k,
for all j 2 J, k 2 K;
YDC
Downloaded by [Texas A&M University-Commerce] at 04:35 20 July 2013

kl 1, if customer l is allocated to distribution centre k, for all k 2 K, l 2 L;

0, otherwise;
ZM
j 1, if manufacturer j is opened, for all j 2 J; 0, otherwise;
ZD
k 1, if distribution centre k is opened, for all k 2 K; 0, otherwise.
The SCNDR model is formulated as the following mixed integer programming model:
X XX XX X XX
min TC SM SM SM
gh d gh X ghvM þ TCMD MD MD
jk d jk X jk þ TC DC DC DC
st d st X stvD
vM 2VM g2G h2H j2J k2K vD 2VD s2S t2T
! !
X X X X
þ VCM
j CSi Y SM
ij þ VC D
k CC DC
l Y kl
j2J i2I k2K l2L
X X
þ FCM M
j Zj þ FCD D
k Zk ð15Þ
j2J k2K

subject to:
X X
X SM
ihvM ¼ 1, all i 2 I ð16Þ
vM 2VM h2H

X X
CSi X SM M
ihvM  CVvM , all vM 2 VM ð17Þ
i2I h2H

P P P
X SM
ghvM  1 ð18Þ
vM 2VM g2G h2H

X X
X SM
hgvM  X SM
ghvM ¼ 0, all vM 2 VM , h 2 H ð19Þ
g2G g2G

XX
X SM
ijvM  1, all vM 2 VM ð20Þ
j2J i2I

X
CSi YSM M M
ij  Cj Zj  0, all j 2 J ð21Þ
i2I

X X
X SM
ihvM þ X SM SM
jhvM  Yij  1, all i 2 I, j 2 J, vM 2 VM ð22Þ
h2H h2H
 International Journal of Production Research 3967

UivM  UgvM þ ðjIj þ jJjÞX SM

igvM  jIj þ jJj  1, all i 2 I, vM 2 VM , g 2 G ð23Þ
X X
YMD
jk  CSi YSM
ij  0, all j 2 J ð24Þ
k2K i2I

X X
YMD
jk  CC DC
l Ykl  0, all k 2 K ð25Þ
j2J l2L

YMD MD
jk  BM X jk  0, all j 2 J, k 2 K ð26Þ
X X
X DC
slvM ¼ 1, all l 2 L
Downloaded by [Texas A&M University-Commerce] at 04:35 20 July 2013

ð27Þ
vD 2VD s2S

X X
CC
l X DC D
slvD  CVvD , all vD 2 VD ð28Þ
l2L s2S

X XX
X DC
stvD  1 ð29Þ
vD 2VD s2S t2T

X X
X DC
stvD  X DC
tsvD ¼ 0, all vD 2 VD , s 2 S ð30Þ
t2T t2T

XX
X DC
klvD  1, all vD 2 VD ð31Þ
k2K l2L

X
CC DC D D
l Ykl  Ck Zk  0, all k 2 K ð32Þ
l2L

X X
X DC
ktvD þ X DC DC
tlvD  Ykl  1, all k 2 K, l 2 L, vD 2 VD ð33Þ
t2T t2T

WsvD  WlvD þ ðjKj þ jLjÞX DC

slvD  jKj þ jLj  1, all l 2 L, vD 2 VD , s 2 S ð34Þ

X SM MD DC SM DC M D
ghvM , X jk , X stvD , Yij , Ykl , Zj , Zk 2 f0, 1g, all g 2 G, h 2 H, s 2 S,
t 2 T, i 2 I, j 2 J,
ð35Þ
k 2 K, l 2 L,
vM 2 VM , vD 2 VD

UivM , UgvM , WsvD , WlvD  0, all g 2 G, s 2 S, i 2 I,

ð36Þ
l 2 L, vM 2 VM , vD 2 VD :

The objective function, Equation (15), of the SCNDR model is similar to that of the
SCND model. However, in the SCNDR model, we consider the routing cost instead of
the round trip cost. The objective function minimises the sum of the costs required to route
 3968 J.-H. Lee et al.

the raw materials/products from the source sites (suppliers and DCs) to the destination
sites (manufacturers and customers), the fixed costs for opening and the variable cost for
operating facilities. The transportation costs involve transporting the raw materials from
the supplier to the manufacturer (routing), the final product from the manufacturer to the
distribution centre, and from the DC to the customer (routing). Constraints (16)–(23)
guarantee the location/allocation and routing between suppliers and manufacturers.
Constraint (16) requires that each supplier be placed on a single route. This implies that the
requirement of any single supplier is less than the capacity of the vehicle. Constraint (17)
ensures the capacity of the manufacturer’s vehicle. Constraint (18) ensures that every
delivery route should be connected to a manufacturer. Constraint (19) requires that the
vehicle should leave every point that is entered by the vehicle. Constraint (20) specifies that
a route cannot be operated from multiple manufacturers. Constraint (21) limits the
Downloaded by [Texas A&M University-Commerce] at 04:35 20 July 2013

flow through a manufacturer so that it does not exceed the manufacturer’s capacity.
Constraint (22) ensures links between the allocation and routing: so that a supplier can be
allocated to a manufacturer only if there is a route from that manufacturer going via that
supplier. Constraint (23) is a sub-tour elimination constraint in which jIj denotes the number
of elements in set I. Constraints (24)–(26) guarantee the transportation between
manufacturers and DCs. Constraint (24) ensures that the flow out of a manufacturer
should be less than or equal to the flow into a manufacturer. Constraint (25) specifies that
the flow out of a DC should be less than or equal to the flow into a DC. Constraint (26)
specifies that if the manufacturer is allocated to the DC, the product can be transported
from the manufacturer to the DC. Constraints (27)–(34) guarantee the location/allocation
and routing between customers and DCs. Constraint (27) requires that each customer be
placed on a single route. This implies that the requirement of any single customer is less than
the capacity of the vehicle. Constraint (28) ensures the capacity of the DC vehicle.
Constraint (29) ensures that every delivery route should be connected to a DC. Constraint (30)
requires that the vehicle should leave every point that it has entered. Constraint (31)
specifies that a route cannot be operated from multiple DCs. Constraint (32) limits the
flow through a DC to the capacity of the DC. Constraint (33) ensures a link between the
allocation and the routing so that a customer can be allocated to a DC only if there
is a route from that DC going via that customer. Constraint (34) guarantees sub-tour
elimination. Constraint (35) is a binary variable (that specifies whether or not there is
transportation between points of each level and that each manufacturer or DC is either
open or closed). Constraint (36) ensures that the auxiliary variables take positive values.

4. Heuristic algorithm based on LP-relaxation

The study first uses a small problem as a case study. The problem involves four suppliers,
two potential manufacturers, two potential DCs, and four customers. The mixed integer
programming model includes 1024 variables and 149 constraints. An increase in the
number of suppliers (or customers) would cause an exponential increase in the variables
and constraints. As mentioned in Section 1, even for small problems, the model for the
SCNDR problem cannot be solved using the existing optimisation techniques. Therefore,
the study proposes an efficient heuristic algorithm based on LP-relaxation in order to solve
this problem. An increase in the number of decision variables, especially binary variables,
causes an increase in the time it takes to solve an MIP problem. If the number of binary
 International Journal of Production Research 3969

variables is reduced, the problem can be solved in less time. Hence, in advance, the study
determines the binary variables for manufacturers and DCs in the MIP model and
subsequently solves the whole model. Figure 3 shows the flowchart of the proposed
heuristic algorithm based on LP-relaxation. The detailed procedure for this algorithm is
as follows:
Step 1: Determine the lower bounds of required numbers for manufacturers and DCs,
initialise T (number of iterations) and best_ob:
& ’
X n o
LM ¼ CSi =min CM
j ,
i2I
& ’
X  D
Downloaded by [Texas A&M University-Commerce] at 04:35 20 July 2013

LD ¼ CC
l =min Ck :
l2L

Step 2: Determine which manufacturers and DCs are opened, by creating binary values
(0 or 1) satisfying the following conditions for manufacturers and DCs:
X
ZMj ¼ LM,
j2J
X
ZD
k ¼ LD:
k2K

Step 3: If the conditions of Step 2 are satisfied, go to step 4. Otherwise, update the tabu
list and go to Step 2.
Step 4: Solve the suppliers-manufacturers routing problem and DCs-customers routing
problem with ZM D
j , Zk determined in Step 2. Solve the transportation problem between
manufacturers and DCs with routing results.
Step 5: Set temp_ob ¼ calculated objective value. If temp_ob is less than best_ob, set
best_ob ¼ temp_ob. Otherwise, go to Step 6.
Step 6: If T is equal to TMAX, the objective value is best_ob. Otherwise, set T ¼ T þ 1,
update tabu list, and go to Step 2.
In Step 3 and Step 6, the tabu list does not find the binary value that it found
previously. It is a similar function to the tabu list of the tabu search. In Steps 5 and 6,
best_ob is defined as the best objective value and an initial best_ob value is set to a large
number such as BM. The maximum number of iterations (TMAX) is defined as follows: the
number of combinations of manufacturers and DCs is 2jJj  2jKj since their values are
binary. However, at least one manufacturer and one DC should be opened. Therefore, the
number of combinations becomes (2jJj1)  (2jKj1). However, there are also lower bounds
of the required numbers for the manufacturer and DC, and we denote LM and LD,
respectively. Therefore, the total number of combinations is jJjCLM.jKjCLD. Consequently,
the proposed heuristic algorithm based on LP-relaxation can reduce the search area of the
combinations for the binary variables to as much as (2jJj1)  (2jK1)(jJjCLM.jKjCLD).
For example, if jJj ¼ jKj ¼ 10 and LM ¼ LD ¼ 5, then 983,025 ((10241)  (1024–1) 
10C5.10C5 ¼ 1, 046, 52963, 504 ¼ 983,025) iterations can be reduced.
 3970 J.-H. Lee et al.

LM =
⎡∑C ⎡
S
Determine i Min{C Mj }
the lower bound for i∈I
manufactures and DCs,
Set T=1, best_ob
⎡
LD = ∑ C
l ∈L
l
C
Min{C }
⎡
D
k

Create binary values

for Z Mj , Z kD LP-relaxation
Part
Tabu list update

∑ Z Mj ≤ LM
Downloaded by [Texas A&M University-Commerce] at 04:35 20 July 2013

j∈J and
∑ ZkD ≤ LD No
k∈K

Yes

Solve the routing

problem given locations

Solve the transportation

problem between
manufacturers and DCs
given routes

Set
temp_ob = calculated
objective value
T=T+1
Solution
Search Part

temp_ob ≤ best_ob
No

Yes

best_ob = temp_ob

T =T Max
No

Yes

Objective value is
best_ob

Figure 3. Flowchart of solution procedure for the SCNDR.

 International Journal of Production Research 3971

5. Computational experiments
The MIP models were conducted on a 3.0 GHz PC with 1 GB RAM on the Microsoft
Windows XP operating system. They were solved using Xpress-MP release 2007A.
An unacceptable amount of computational time is required when we use this model to find
an optimal solution. We compared the SCND model (MIP) and the SCNDR model (MIP
and heuristic algorithm) in order to test their validity for small problems. The heuristic
algorithm was developed using Xpress-MP release 2007A and Microsoft Visual Basic 6.0.
Table 1 shows the composition of the supply chain network. Table 2 shows the objective
value, computational time, number of constraints, number of variables, and percentage
difference (or savings) between the optimal values of the SCND model and the SCNDR
model. Using these results, the study demonstrated that the SCNDR model could reduce
the total cost. Moreover, the SCNDR model is more realistic. In Table 2, the com-
Downloaded by [Texas A&M University-Commerce] at 04:35 20 July 2013

putational time for finding an optimal solution increases exponentially with an increase in
the number of suppliers, manufacturers, distribution centres, and customers.
Figures 4 and 5 graphically illustrate the objective value and computational time of the
MIP models and the heuristic algorithm for the SCND model and SCNDR model. Due to
the characteristics of the NP-hard problem, the computational time, constraints, and
variables increase exponentially with the number of suppliers, manufacturers, distribution
centres, and customers as shown in Figure 5. Figure 6 shows the number of constraints and
variables for the SCND MIP model and the SCNDR MIP model.
In the application of the real problem, we experimented with the instances of a realistic
size problem (30 suppliers, 10 potential manufacturers, 10 potential DCs, and 30
customers). The supply quantity and customer demand are generated from a uniform
distribution on the ranges [5, 30]. We used the modified data from the Solomon instances
(Solomon 1987) for the suppliers, manufacturers, DCs, and customer locations. The
Solomon instances are divided into six groups, denoted R1, R2, C1, C2, RC1 and RC2.
For all of the instances within a group, the customer locations are the same. In R1 and R2,
the customer locations are randomly generated from a uniform distribution, and in C1
and C2, they are clustered. In RC1 and RC2, the customer locations are a combination
of randomly generated and clustered points. The study created one instance that
corresponded to each group of supplier and customer locations, C1-30, C2-30, R1-30, and
RC1-30, because the Solomon instances yield only four sets of distinct customer locations.
For 10 candidate manufacturers and DC locations, the study randomly generated the
candidate locations from a uniform distribution. In order to calculate the distance matrix,
the study calculated the Euclidean distance between all node pairs and rounded the value
to the nearest integer. Table 3 shows the results of the SCNDR heuristic algorithm. In the

Table 1. Data sets for comparing MIPs and heuristic algorithm.

Problem Number of Number of Number Number of

instance suppliers manufacturers of DCs customers

1 3 2 2 3
2 4 2 2 4
3 5 3 3 5
4 6 3 3 6
5 8 3 3 8
 Downloaded by [Texas A&M University-Commerce] at 04:35 20 July 2013

3972

Table 2. Comparison results of MIPs and heuristic algorithm.

SCND MIP SCNDR MIP Heuristic algorithm

Problem Objective Computational Number of Number of Objective Computational Number of Number of Savings Objective Computational
instance value time (seconds) constraints variables value time (seconds) constraints variables (%) value time (seconds)

1 1538 1.2 32 82 1488 7.7 107 640 3.36 1488 1.5

2 1610 2.5 38 100 1532 80.2 149 1024 5.09 1532 4.6
3 1772 5.8 64 200 1618 1512.7 334 3057 9.52 1618 151.1
4 1928 15.4 72 228 1754 3188.3 439 4068 9.92 1754 256.4
J.-H. Lee et al.

5 3178 74.1 91 308 2890 15,907.3 668 6702 9.97 2890 406.3
 International Journal of Production Research 3973
Downloaded by [Texas A&M University-Commerce] at 04:35 20 July 2013

Figure 4. Comparison of the objective values among the SCND MIP, SCNDR MIP, and heuristic.

Figure 5. Comparison of the computational times among the SCND MIP, SCNDR MIP, and
heuristic.

Figure 6. Comparison of the number of constraints and variables of the MIP models.

results, the study experimented with two kinds of TMAX (¼200 and 400) for the proposed
heuristic algorithm.
In terms of TMAX, the result of the C2-30 clustered instance is partly under the
influence of TMAX and takes a longer amount of time. On the other hand, the RC1-30
 3974 J.-H. Lee et al.

Table 3. Comparison result of SCNDR heuristic algorithm.

TMAX ¼ 200 TMAX ¼ 400

Problem Number of Number Objective Computational Number of Number Objective Computational

instance manufacturers of DCs value* time (seconds)* manufacturers of DCs value* time (seconds)* Remark

C1-30 3 4 7,725 535.7 4 2 7,190 1,385.1 7.44%

C2-30 2 3 6,436 737.4 2 3 6,406 1,567.3 0.47%
R1-30 3 2 6,760 452.4 3 2 6,640 843.5 1.78%
RC1-30 3 5 9,270 951.4 3 4 8,404 1,844.6 9.34%

Note: *Average of 10 evaluations.

Downloaded by [Texas A&M University-Commerce] at 04:35 20 July 2013

Figure 7. The differences according to TMAX for the heuristic algorithm.

instance (combination of randomly generated and clustered points) is to a large extent

under the influence of TMAX, but takes a longer amount of time. Therefore, in the C1-30
instance type it is acceptable to use the TMAX (400), but in the C2 and R1 instance types
it is better to use the TMAX (200) (Figure 7).
In addition, we investigate the efficiency of the proposed heuristic method by
comparing it with the existing heuristic method. Since this problem consists of the location
problem, the allocation problem, and the routing problem in the several layers sim-
ultaneously, it differs from that defined by the past studies and their heuristic methods are
not directly applicable for this problem. To overcome this limitation, we divide this
problem into three sub-problems. The first sub-problem is an LRP between DCs and
customers. We first solve the DC location and allocation problem without routing. Then
we solve the routing problem among one DC and its customers using Wu et al.’s (2002)
algorithm. As a result of solving this sub-problem, we obtain the required quantities of
product in each DC. The second sub-problem is a simple transportation problem between
manufacturers and DCs. It can be solved by a linear program to satisfy the requirement of
each DC. As a result of solving this sub-problem, we obtain the product quantity of each
manufacturer and the relevant transportation scheme between manufacturers and DCs.
The third sub-problem is an LRP between suppliers and manufacturers. We first solve the
manufacturer location and allocation problem without routing. Then we solve the routing
problem among one manufacturer and its suppliers. We call this algorithm the modified
Wu et al.’s algorithm. Table 4 shows the computational results of both the SCNDR
heuristic algorithm and the modified Wu et al.’s algorithm. As expected, the SCNDR
 International Journal of Production Research 3975

Table 4. Comparison result of the SCNDR heuristic algorithm and the modified Wu et al.’s (2002)
algorithm.

SCNDR heuristic algorithm (TMAX ¼ 200) Modified Wu et al.’s algorithm

Problem Number of Number Objective Number of Number Objective Remark

instance manufacturers of DCs value* manufacturers of DCs value*

C1-30 3 4 7,725 3 4 7,952 (3084 þ 760 þ 4108)** 2.94%

C2-30 2 3 6,436 3 3 6,806 (2568 þ 790 þ 3448)** 5.75%
R1-30 3 2 6,760 3 3 7,100 (3654 þ 850 þ 2596)** 5.03%
RC1-30 3 5 9,270 4 5 9,750 (3886 þ 1120 + 4744)** 5.18%

Notes: *Average of 10 evaluations;

Downloaded by [Texas A&M University-Commerce] at 04:35 20 July 2013

**Supplier-manufacturer cost/manufacturer-DC cost/DC-customer cost.

heuristic algorithm outperforms the modified Wu et al.’s algorithm for all problem
instances since the SCNDR heuristic algorithm solves the location-allocation-routing
problem in the four layer supply chain network simultaneously.

6. Conclusions
This paper considered both the supply chain network design model and the supply chain
network design with routing model. We formulated two mixed integer programs for these
models. In order to solve these models, we proposed a heuristic algorithm based on the
relaxed binary variables technique. In order to evaluate the superiority of the proposed
algorithm, an experiment was performed. This experiment compared the results between
the SCND MIP, the SCNDR MIP, and the SCNDR heuristic algorithm. The results
showed that the computational time taken to obtain an optimal solution using Xpress-MP
increases exponentially as the problem size grows, while the proposed algorithm reduced
the influence of the problem size on the computational time. The computational results
demonstrate the efficiency of the developed heuristic algorithm. In terms of real time
application, the size of the problem might be larger than that of the test data. In such
environments, it is not feasible to find an optimal solution in a reasonable amount of time.
Hence, the proposed algorithm can be suitably used in the real situation of designing
a global supply chain network in third party logistics companies. Developing some meta-
heuristics for the SCNDR model and comparing them with our heuristic algorithm might
be an interesting research problem.

Acknowledgements
The authors are grateful to the careful and constructive reviews from the editor-in-chief, two
anonymous referees, and Dr Hongfeng Wang. This work was supported by a grant from the Korean
Ministry of Education, Science and Technology (The Regional Core Research Program/Institute
of Logistics Information Technology).

References

Abdelmaguid, T.F. and Dessouky, M.M., 2006. A genetic algorithm approach to the integrated
inventory-distribution problem. International Journal of Production Research, 44 (21),
4445–4464.
 3976 J.-H. Lee et al.

Ambrosino, D. and Scutella, M., 2005. Distribution network design: new problems and related
models. European Journal of Operational Research, 165 (3), 610–624.
Ballou, R., 2004. Business logistics/supply chain management. 5th ed. Upper Saddle River, NJ:
Prentice Hall.
Breedam, A., 2001. Comparing descent heuristics and metaheuristics for the vehicle routing problem.
Computers and Operations Research, 28 (4), 289–315.
Canel, C. and Khumawala, B.M., 1997. Multi-period international facilities location: an algorithm
and application. International Journal of Production Research, 35 (7), 1891–1910.
Chan, Y., Carter, W., and Burnes, M., 2001. A multiple-depot, multiple-vehicle, location-routing
problem with stochastically processed demands. Computers and Operations Research, 28 (8),
803–826.
Chopra, S. and Meindl, P., 2004. Supply chain management: strategy, planning and operation. 2nd ed.
Englewood Cliffs, NJ: Prentice-Hall.
Downloaded by [Texas A&M University-Commerce] at 04:35 20 July 2013

Crainic, T. and Laporte, G., 1997. Planning models for freight transportation. European Journal of
Operational Research, 97 (3), 409–438.
Gunnarsson, H., Rönnqvist, M., and Carlsson, D., 2006. A combined terminal location and ship
routing problem. Journal of the Operational Research Society, 57 (8), 928–938.
Jones, D., Mirrazavi, S., and Tamiz, M., 2002. Multi-objective meta-heuristics: an overview of the
current state-of-the-art. European Journal of Operational Research, 137 (1), 1–9.
Karp, R., 1972. Complexity of computer computations. New York: Plenum Press.
Ko, C.S. and Lee, H.J., 2007. A recursive optimization/simulation procedure for express courier
service network design: determination of terminal capacity and cut-off time. Journal of the
Korean Institute of Industrial Engineers, 33 (2), 282–289.
Laporte, G., Nobert, Y., and Taillefer, S., 1988. Solving a family of multi-depot vehicle routing and
location-routing problems. Transportation Science, 22 (3), 161–172.
Lee, K.S., et al., 2002. An optimization approach to routing and wavelength assignment in WDM
all-optical mesh networks without wavelength conversion. ETRI Journal, 24 (2), 131–141.
Lee, Y.H., et al., 2008. Tabu search heuristic algorithm for designing broadband convergence
networks. Journal of the Korean Institute of Industrial Engineers, 34 (2), 205–215.
Lin, C. and Kwok, R., 2006. Multi-objective metaheuristics for a location-routing problem with
multiple use of vehicles on real data and simulated data. European Journal of Operational
Research, 175 (3), 1833–1849.
Salhi, S. and Rand, G., 1989. The effect of ignoring routes when locating depots. European Journal
of Operational Research, 39 (2), 150–156.
Sha, D.Y. and Che, Z.H., 2006. Supply chain network design: partner selection and production/
distribution planning using a systematic model. Journal of the Operational Research Society,
57 (1), 52–62.
Solomon, M., 1987. Algorithms for the vehicle routing and scheduling problem with time window
constraints. Operations Research, 35 (2), 254–265.
Wang, H., Xu, D., and Li, L., 2007. Novel globally adaptive load-balanced routing algorithm for
Torus interconnection networks. ETRI Journal, 29 (3), 405–407.
Webb, M., 1968. Cost functions in the location of depots for multiple delivery journeys. Operational
Research Quarterly, 19 (3), 311–328.
Wu, T., Low, C., and Bai, J., 2002. Heuristic solutions to multi-depot location-routing problems.
Computers and Operations Research, 29 (10), 1393–1415.