Professional Documents
Culture Documents
A Supply Chain Distribution Network Design Model A
A Supply Chain Distribution Network Design Model A
net/publication/225714401
CITATIONS READS
221 1,975
2 authors:
All content following this page was uploaded by Irem Ozkarahan on 03 December 2014.
ORIGINAL ARTICLE
Received: 20 April 2006 / Accepted: 12 October 2006 / Published online: 12 December 2006
# Springer-Verlag London Limited 2006
choices for the two decisions, there are various types of least cost while satisfying the desired service level to the
distribution network designs that may be used to move retailers. Since the maximal covering approach [4] has
products from manufacturing plants to customer. The proved to be one of the most useful facility location models
readers may refer to Chopra and Meindl [3] for the detailed from both theoretical and practical points of view, we use it
description of each distribution option and the discussion in statement of the service level. In the proposed model, a
on its strengths and weaknesses. coverage function which may differ among the retailers
In this paper, we deal with a SC distribution network according to their service standard request is defined for
comprised of a set of manufacturing plants, warehouses and each retailer.
retailers. The network is illustrated in Fig. 1. Much of the decision making in the real world takes
In the design option considered, inventory is stored place in an environment in which the goals, the constraints,
locally at retail stores and distributor warehouses. The main and the consequences of the possible actions are not known
advantage of such a network structure is that it can lower precisely. As in the same manner, SCs operate in a
the delivery cost and provide a faster response than other somehow uncertain environment. Uncertainty may be
networks. The major disadvantage is the increased inven- associated with target values of objectives, external supply
tory and facility costs. As emphasized by Chopra and and customer demand etc. Supply chain distribution
Meindl [3], such a network is best suited for fast-moving network design models developed so far either ignored
items where customers value the rapid response. uncertainty or consider it approximately through the use of
Cost or profit-based optimization is the most widely probability concepts. However, when there is lack of
used method for SC distribution network design problems. evidence available or lack of certainty in evidence, the
However, more customer oriented approaches are required standard probabilistic reasoning methods are not appropri-
in order to provide a sustainable competitive advantage in ate. In this case, uncertain parameters can be specified
today’s business environment. Nowadays, there is a trend to based on the experience and managerial subjective judg-
consider customer service level as more critical. Customer ment. Fuzzy set theory (FST) [5] provides the appropriate
service level can be measured by various measures such as framework to describe and treat uncertainty [6]. In decision
customer response time, consistency of order cycle time, sciences, fuzzy sets have had a great impact in preference
accuracy of order fulfillment rate, delivery lead time, modeling and multi-criteria evaluation and have helped
flexibility in order quantity. bringing optimization techniques closer to the users needs.
In this paper, we develop a multiobjective SC distribu- As emphasized previously, SC distribution network
tion network design model. The goal is to select the design problem is a strategic decision problem, the
optimum numbers, locations and capacity levels of plants optimization of which is crucial for the long-term efficient
and warehouses to deliver the products to the retailers at the operation of whole SC. The decisions to be made have
long-lasting effects and are costly, or sometimes impossible location analysis. Pontrandolfo and Okogbaa [13] review
to reverse. Because data are often incomplete and impre- the literature on the configuration as well as the coordina-
cise, SC distribution network design decisions generally use tion of the network of global facilities problems. They
forecasts based on aggregated data. As the environment or define a framework that systematically addresses the global
system changes, such impreciseness also propagates and manufacturing planning problem by identifying and classi-
can exert serious effects on the operation and management fying the variables involved therein.
of the system. Another source of impreciseness may be due The last decades of the twentieth century witnessed a
to vagueness in DMs’ interpretation or intent. This situation considerable expansion of SCs into international locations.
occurs when exactness is not required or is not possible to This growth in globalization, and the additional manage-
specify, or when exactness would unnecessarily limit ment challenges it brings, has motivated both practitioner
options. As stated by Reznik and Pham [7], exact or crisp and academic interest in global SC management [14]. In
models when used in such cases would not be able to this paper, we do not consider the global issues in the
faithfully simulate the richness and subtlety of the real work proposed model.
space. Furthermore, in the worst case, misleading or After mentioning some important reviews on SC
incorrect outcomes may result. Thus, there is a need to network design problem, we present a review of the several
articulate the problems arising from such impreciseness relevant papers in the following. In our review, we focus
with the view to construct appropriate models for their especially on the papers that develop or consider linear
representation, as well as suitable methods for processing deterministic SC network design models.
and manipulating them within the environment. Consider- Geoffrion and Graves [15] presents a new method for the
ing this need, in this study, we use FST in handling SC solution of the problem addresses the optimal location of
distribution network design problem. More specifically, two distribution centers between plants and customers. They
kind of impreciseness that may be faced in the problem, develop an algorithm based on Benders’ decomposition for
i.e., imprecision in retailers’ demand and DMs’ aspiration solving multi-commodity distribution network design prob-
levels for the goals, are treated. Additionally, to provide lem. Brown et al. [16] present a mixed integer model for a
the DMs with a flexible and robust multi-objective multi-commodity production-distribution system. The ob-
decision making technique, we propose a novel and jective of the model is to minimize the variable production
generic interactive fuzzy goal programming (IFGP)-based and shipping cost, fixed cost of equipment assignment and
solution approach. Through the solution approach, DMs plant operations. They apply to the model a primal
determine the preferred compromise solution. decomposition technique similar to Geoffrion and Graves’
The paper is further organized as follows: A review of [15] algorithm. Cohen and Lee [17] present a strategic
the related literature is presented in the next section. model structure and a hierarchical decomposition approach.
Thereafter, basic concepts and the framework of fuzzy goal The scope of their work is to analyze interactions between
programming (FGP), and the proposed IFGP-based solu- functions in a complete SC network. To model these
tion approach are presented in Sects. 3 and 4, respectively. interactions they consider four sub modules where each
Section 5 is devoted to the presentation of the crisp represents a part of the overall SC: (1) material control, (2)
formulation of the proposed SC distribution network design production control, (3) finished goods stockpile, and (4)
model. Computational experiments are presented in Sect. 6. distribution network control.
Finally, conclusions are presented in Sect. 7. Pirkul and Jayaraman [18] consider a tri-echelon, multi-
commodity system concerning production, distribution and
transportation planning. The authors use a Lagrangean
2 Literature review relaxation-based heuristic to provide effective an effective
feasible solution. Jayaraman [19] studies the capacitated
Numerous researchers have extensively studied facility and warehouse location problem that involves locating a given
demand allocation problems. Interested readers may refer to number of warehouses to satisfy customer demands for
detailed survey results, e.g., Francis et al. [8], Aikens [9], different products. Pirkul and Jayaraman [1] extend the
Brandeau and Chiu [10], Beamon [11], Avella et al. [12] previous problem by considering locating also a given
and Pontrandolfo and Okogbaa [13]. number of plants. They present a model for multi-
Beamon [11] provides a focused review of literature in commodity, multi-plant, capacitated facility location prob-
the area of multi-stage SC design and analysis and classify lem, and develop a Lagrangean-based heuristic solution
the models in the area into four categories: deterministic procedure. Dogan and Goetschalckx [20] develop a mixed
analytic models, stochastic analytic models, economic integer linear programming model for the integrated design
models, and simulation models. Avella et al. [12] present of multi-period production-distribution systems. Their
their views on the state of the art and the future trends in paper contributes to the literature by developing an
404 Int J Adv Manuf Technol (2008) 36:401–418
integrated design methodology for strategic production and technique and three local search methods, the pair-wise
distribution systems using primal decomposition theory. It exchange procedure, the insert procedure and the remove
also provides an acceleration methodology for solving the procedure. Eskigun et al. [31] deal with the design of a SC
problem. Tragantalerngsak et al. [21] consider a two- distribution network considering lead time, location of
echelon facility location problem in which the facilities in distribution facilities and choice of transportation mode.
the first echelon are incapacitated and the facilities in the They present a Lagrangian heuristic that gives good
second echelon are capacitated. The goal in their model is solution quality in reasonable computational time. In a
to determine the number and locations of facilities in both recent paper, Amiri [32] addresses the distribution network
echelons in order to satisfy customer demand of the design problem in a SC system. The author develops a
product. They develop a Lagrangean relaxation-based mixed integer programming model and provides a heuristic
branch and bound algorithm to solve the problem. Lee et solution procedure.
al. [22] develop a multi-product mixed integer nonlinear The contribution of our current paper to the literature is
programming model to develop a capacity expansion of an twofold: First, a fuzzy multi-objective model has been
integrated production and distribution system. The system developed for SC distribution network design problem.
comprises the multi-site batch plants and warehouses. Second, a novel and generic IFGP-based solution ap-
Melachrinodis and Min [23] design a multi-objective, proach is proposed to determine the preferred compromise
multi-period mixed integer programming model that deter- solution.
mine the optimal relocation site and phase out schedule of a
combined manufacturing and distribution facility from SC
perspectives. Their research differentiates from the litera-
3 Fuzzy goal programming
ture by considering both dynamic aspects and multi-
echelon network design. Sabri and Beamon [24] develop
Goal programming (GP) is one of the most powerful, multi-
a SC model that considers simultaneous strategic and
objective decision making approaches in practical decision
operational SC planning. The main contribution of the
making. In a standard GP formulation, goals are defined
work is the incorporation of production, delivery, and
precisely. However, application of GP to the real life prob-
demand uncertainty into one model. Pirkul and Jayaraman
lems may be faced with two important difficulties. One of
[25] present an integrated logistic model, and develop an
which is expressing the DMs’ vague goals mathematically
efficient solution procedure for multi-commodity produc-
and the second is the need to optimize all goals simulta-
tion-distribution problem.
neously. In such situations, the use of FST comes in handy.
Tsiakis et al. [26] develop a strategic planning model for
Applying FST into goal programming (GP) has the
SC networks. The paper takes into consideration flexible
advantage of allowing for the vague aspirations of a DM,
production facilities in which a number of products are
which can then be qualified by some natural language
produced making use of shared resources, the economies of
terms. The FST in GP was first considered by Narasimhan
scale in transportation, and uncertainty in product demand.
[33]. Goal programming in fuzzy environment is further
Jang et al. [2] propose a supply network with a global bill
developed by Hannan [34], Ignizio [35], Narasimhan and
of material. They model design and planning problems of a
Rubin [36], Tiwari et al. [37, 38] and others.
supply network in a single combined system. Talluri and
A fuzzy set A can be characterized by a membership
Baker [27] propose a multi-phase mathematical program-
function, usually denoted by μ, which assigns to each
ming approach for effective SC network design. Cakravastia
object of a domain its grade of membership in A. The
et al. [28] develop a mixed integer programming model of
nearer the value of membership function to unity, the higher
the supplier (any manufacturer playing a lower-level
the grade of membership of element or object in a fuzzy set
supporting role) selection process in designing a SC
A. Various types of membership functions can be used to
network. The assumed objective of the SC is to minimize
represent the fuzzy set.
the level of customer dissatisfaction, which is evaluated by
A typical FGP problem formulation can be stated as
two criteria: (i) price and (ii) delivery lead time.
follows:
More recently, Beamon and Fernandes [29] study a
closed-loop SC in which manufacturers produce new 9
products and remanufacture used products. The multi- Find xi ; i ¼ 1; :::; n >
>
>
>
>
period integer programming model uses the present worth Zm ðxi Þ Z m m ¼ 1; :::; M >
=
method to jointly analyze investment and operational costs. ð1Þ
Zk ðxi Þ Z k k ¼ M þ 1; :::; K
Yeh [30] presents a hybrid heuristic algorithm to solve the >
>
gj ð x i Þ bj j ¼ 1; :::; J >
>
multi-stage SC network design problem. The algorithm >
>
;
combines a greedy method, the linear programming xi 0 i ¼ 1; :::; n
Int J Adv Manuf Technol (2008) 36:401–418 405
where, Zm(xi) is the mth goal constraint, Zk (xi) is the kth Table 1 The payoff table
goal constraint, Z m ðxi Þ is the target value of the mth goal,
Z1(X) Z2(X) ... ZM(X)
Z k ðxi Þ is the target value of the kth goal, gj (xi) is the jth
inequality constraint and bj is the available resource of X (1) Z11 Z12 ... Z1M
inequality constraint j. X (2) Z21 Z22 ... Z2M
..
In formulation (1), the symbols “≺ and ≻” denote the . ... ... ... ...
X (M) ZM1 ZM2 ... ZMM
fuzzified versions of “≤ and ≥” and can be read as
approximately less / greater than or equal to. These two
types of linguistic terms have different meanings. Under can be transformed to the following conventional linear
programming problem [40]:
approximately less than or equal to situation, the goal m is 9
allowed to be spread to the right-hand-side of Z m (Z m ¼ lm maximize 1 >
>
>
>
subject to >
>
where lm denote the lower bound for the mth objective) >
>
with a certain range of rm(Z m þ rm ¼ um , where um denote >
=
1 μ Zk k ¼ 1; :::; K
the upper bound for the mth objective). Similarly, with ð4Þ
gj ðxi Þ bj i ¼ 1; :::; n; j ¼ 1; :::; J >
>
>
approximatelygreater than or equal to, pk is the allowed >
>
xi 0 i ¼ 1; :::; n >
>
left side of Z k Z k pk ¼ lk ; and Z k ¼ uk . >
>
;
As can be seen, GP and FGP have some similarities. 1 2 ½0; 1:
Both of them need an aspiration level for each objective.
These aspiration levels are determined by DMs. In addition
to the aspiration levels of the goals, FGP needs max-min 4 The proposed IFGP-based solution approach
limits (uk,lk) for each goal. While the DMs decide the max-
min limits, the linear programming results are starting By use of the interactive paradigm, interactive fuzzy
points and the intervals are covered by these results. multiobjective decision making approaches have been
Generally, the DMs find estimates of the upper (u) and investigated to improve the flexibility and robustness of
lower (l) values for each goal using payoff table (see multiobjective decision making techniques. They provide
Table 1). Therefore, the feasibility of each fuzzy goal is learning process about the system, whereby the DM can
guaranteed. learn to recognize good solutions and relative importance of
Here, Zm(X) denotes the mth objective function, and X(m) factors in the system [41]. The main advantage of
is the optimal solution of the mth single objective problem. interactive approaches is that the DM controls the search
Solving the problem with X(m) (m=1,..., M) for each direction during the solution procedure and, as a result, the
objective, a payoff matrix with entries Zpm ¼ Zm X ðpÞ , m, efficient solution achieves his/her preferences [42]. Litera-
p=1,..., M can be formulated as presented in Table 1. Here, ture in the class of fuzzy interactive programming includes
um ¼ max ðZ1m ; Z2m ; . . . ; ZMm Þ and lm =Zmm, m=1,..., M. Werners [43, 44], Leung [45], Fabian et al. [46], Sasaki et
After constructing fuzzified aspiration levels with respect al. [47] and Baptistella and Ollero [48].
to the linguistic terms of approximately less than or equal Belman and Zadeh’s [39] min operator focuses only on
to, and approximately greater than or equal to, the the maximization of the minimum membership grade. It is
membership functions can be developed for each goal. not a compensatory operator. That is, goals with a high
Using Belman & Zadeh’s [39] min operator approach, degree of membership are not traded off against goals with
one can obtain the feasible fuzzy solution set by the a low degree of membership. Therefore, some computa-
intersection of all membership functions representing the tionally efficient compensatory operators (see [41]) can be
fuzzy goals. This solution set is then characterized by its used in setting the objective function in fuzzy programming
membership μF(x) which is: to investigate better results.
One criterion used to evaluate the performance of
mF ð xÞ ¼ mZ1 ð xÞ \ mZ2 ð xÞ:::: \ mZk ð xÞ compensatory operators in fuzzy optimization is monoto-
nicity. Among the compensatory operators which are well
¼ min mZ1 ð xÞ; mZ2 ð xÞ; ::::; mZk ð xÞ ð2Þ
suited in solving multiobjective programming problems,
Then the optimum decision can be determined to be the Werners’ [49] ‘fuzzy and’ operator has an advantage of
maximum degree of membership for the fuzzy decision: being a strongly monotonically increasing function. That is,
it is positively related with the compensation rate. Further-
max mF ð xÞ ¼ max min mZ1 ð xÞ; mZ2 ð xÞ; :::; mZk ð xÞ ð3Þ more, it is easy to handle, and has generated reasonable
x2F x2F
consistent results in applications. For those reasons, we
By introducing the auxiliary variable λ, which is the employ Werners’s ‘fuzzy and’ operator in the proposed
overall satisfactory level of compromise, formulation (2) IFGP-based solution approach.
406 Int J Adv Manuf Technol (2008) 36:401–418
Compromise solution: A feasible vector X* Z S is called a – Step 2: Obtain efficient extreme solutions (payoff
compromise solution of the problem iff X* Z E and values) used for constructing the membership functions
Z ðX Þ ^X 2S Z ð X Þ where Z(X) is the objective function, S of the objectives. If the DM selects one of them as a
is the feasible region, $ stands for “minimum” and E is the preferred compromise solution go to Step 7. Otherwise
set of efficient solutions. go to Step 3.
This definition imposes two conditions on the solution – Step 3: Define the membership function of each fuzzy
for it to be a compromise solution. First, the solution should objective using upper and lower bounds of the
be efficient. Second, the compromise solution is the closest objectives.
solution to the ideal one that maximizes the underlying – Step 4: Considering the membership functions defined
utility function of the DM. In real-world cases, knowledge in Step 3 and γ (fix the value of γ to 1 in the first
of the set of efficient solutions E is not always necessary. iteration) develop the formulation of the problem using
On the other hand, the DMs’ preferences are to be proposed ‘modified fuzzy and’ operator.
considered in determining the final compromise solution. – Step 5: Obtain a compromise solution and present the
solution to the DM. If the DM accepts it, go to Step 7.
Preferred compromise solution: If the compromise solution Otherwise, go to Step 6.
satisfies the DMs’ preferences, then it is called the preferred – Step 6: Ask the DM if he want to modify the
compromise solution. coefficient of compensation (γ), and membership
functions of the objectives, and go to Step 3. Actually,
definition of a unique rule, e.g., selection of the initial
The proposed IFGP-based solution approach can be
value, change direction or rate of variation in each step,
summarized in the following steps:
by which the value of γ is varied is difficult since it
– Step 1: Develop the conventional (crisp) linear depends on DMs’ preferences. For instance, if the
programming formulation of the problem. problem under concern is so sensitive to γ, the rate of
variation should be sufficiently slow. When the DM
Table 3 Expected demand of the retailers tries to modify membership functions of the objectives
and constraints, only the following variations are
Demand for product I (ai l) Demand for product II (ai2)
acceptable [43]: a) the increase of lower bound (lk)
992 408 556 831 352 229 282 894 976 880 for the maximization objectives, b) the decrease of
423 573 388 683 378 898 968 845 885 988 upper bound (um) for the minimization objective, c) the
659 873 467 653 649 844 960 935 847 629
806 222 501 669 483 306 789 968 334 971
Table 4 Lower and upper bounds for the distance
107 377 777 701 709 845 195 610 776 743
500 439 759 264 449 758 736 976 923 339 Retailers Lower bound (LBi) Upper bound (UBi)
741 783 516 507 455 508 198 496 927 811
257 936 327 174 528 916 630 813 320 466 1–12 500 650
956 317 664 878 484 171 955 424 658 720 13–27 600 750
129 887 929 609 514 403 186 505 700 724 28–50 700 850
408 Int J Adv Manuf Technol (2008) 36:401–418
Table 5 The payoff table Table 7 Compromise solution results with min operator
TCOST 2,720,667 2,663,710 52,046 1 0.5317 2,968,563 2,583,509 56,131 0.5317 0.5317 0.5317
INV 3,250,000 2,229,724 50,000
TSERVL 3,250,000 2,994,668 61,532
decrease of maximum tolerance (dj) is an acceptable – The retailers have demand for a multitude of products,
modification which can guarantee an efficient solution and the warehouses are responsible for right-time
in the recalculated compromise solution step. In order delivery of a right amount of products.
to avoid the possibility of getting into infeasible – Decision makers of the plants, warehouses and retailers
solution sets because of excess increase of lk or excess share information and collaborate with each other to
decrease of um, we should increase lk and decrease um design an effective distribution network.
as few requirements as possible in each iteration. – Decisions are made within a single period.
– Step 7: Stop.
The flow chart of these steps is shown in Fig. 2. 5.1 Notations and definitions
mij coverage parameter that denotes the coverage level houses, the third one maximizes the total service level
of the retailer in zone i by the warehouse in zone j provided to the retailers.
LBi, the parameters that are used in defining the
UBi coverage parameter of the retailer in zone i, LBi 5.3 The constraints
and UBi denote lower and upper bound for the
distance, respectively (see Fig. 3). The constraints of the proposed model and their definitions
are presented in the following.
– Decision variables X
Xijl ¼ ail for all i 2 I and l 2 L; ð11Þ
Yjkl amount of product l transported to the warehouse in j
zone j from the plant in zone k, Constraint set (11) ensures that all demand from retailers
Xijl amount of product l transported to the retailer in zone is satisfied by warehouses.
i from the warehouse in zone j, XX X
Zjr binary variable that indicates whether a warehouse sl Xijl Wjr Zjr for all j 2 J ; ð12Þ
with capacity level r is constructed in zone j, i l r
Pkh binary variable that indicates whether a plant with Constraint set (12) limits the distribution quantities that
capacity level h is constructed in zone k. are shipped from warehouses to retailers to the throughput
limits of warehouses.
X
5.2 The objective functions Zjr 1 for all j 2 J ; ð13Þ
r
Objective functions of the model are formulated as follows: Constraint set (13) ensures that a warehouse can be
As can be seen from Table 2, the first objective function assigned at most one capacity level.
X X
minimizes total cost made of: the transportation costs of Xijl Yjkl for all j 2 J and l 2 L; ð14Þ
products from plants to warehouses and from warehouses to i k
retailers, and the fixed costs associated with the plants and Constraint set (14) guarantees that all demand from
the warehouses. While the second objective function retailer in zone i for product l is balanced by the total units
minimizes the investment in opening plants and ware-
Table 9 Solution results obtained by the proposed solution approach performed using Werners’ (1988) ‘fuzzy and’ operator and
in terms of utilization of the plants and warehouses the proposed solution approach, and the results are com-
γ Utilization of the plants Utilization of the warehouses pared.
A hypothetically constructed SC distribution network
Min Max Average Min Max Average design problem with 50 retailer zones, 20 potential
1 0.87 1 0.97 0.90 1.00 0.97
warehouse sites and 15 potential plant sites is considered
0.9 0.87 1 0.97 0.91 1.00 0.97 in the computational experiments. It is assumed that two
0.8-0.7 0.89 1 0.97 0.82 1.00 0.97 different types of product are demanded by the retailers.
0.6-0.5 0.87 1 0.97 0.90 1.00 0.97 Coordinates of the retailer zones, potential warehouses and
0.4-0.3 0.87 1 0.97 0.86 1.00 0.97 plant sites are generated from a uniform distribution over a
0.2-0 0.88 1 0.96 0.75 1.00 0.93 square with side 3000. Euclidean distances are used in
defining the coverage parameters. CPLEX 9.1 optimization
of product l available at warehouse in zone j that has been software is used at the solution stage.
supplied from open plants. Before presenting the computational experiments, let us
XX X explain the parameter structuring of the hypothetical SC
qkl Yjkl Dkh Pkh for all k 2 K; ð15Þ distribution network design problem under concern.
j l h
Expected demand of the retailers for two different
Constraints in set (15) represent the capacity restrictions products is drawn from a uniform distribution between
of the plants in terms of their total shipments to the
100 and 1000 as given in Table 3.
warehouses. Five capacity levels are used for the capacities available
X
Pkh 1 for all k 2 K; ð16Þ to both the potential plants and warehouses. The opening
h cost of the warehouse in zone j with capacity level 3 (OWj3)
Constraint set (16) ensures that a plant can be assigned are drawn from a uniform distribution between 90,000 and
at most one capacity level. 120,000. The opening costs of the warehouses for the other
Zjr 2 f0; 1g for all j 2 J ; r 2 R; Pkh 2 f0; 1g capacity levels are computed as follows: OWj1 =0.75*OWj3,
ð17Þ OWj2 =0.85*OWj3, OWj4 =1.15*OWj3, OWj5 =1.25*OWj3.
for all k 2 K; h 2 H
Cost coefficients of OPkh are computed in terms of the
Finally, constraint set (17) enforces the binary and non- warehouses costs as OPkh =4*OWkh. Fixed portion of the
negativity restrictions on the decision variables. annual possession and operating costs of the warehouse in
zone j with capacity level 3 (gj3) and the plant in zone k
with capacity level 3 (fk3) are drawn from a uniform
6 Computational experiments distribution between 18,000 and 25,000 and 75,000 and
100,000, respectively. Fixed portion of the annual possession
6.1 The problem and parameter structuring and operating costs of warehouses and plants for the other
capacity levels are computed as follows: gj1 =0.75*gj3, gj2 =
To explore the viability of the proposed model and the 0.85*gj3, gj4 =1.15*gj3, gj5 =1.25*gj3 and fk1 =0.75*fk3, fk2 =
IFGP-based solution approach, computational experiments 0.85*fk3, fk4 =1.15*fk3, fk5 =1.25*fk3.
are presented in this section. The experiments are classified Required throughput capacity of a warehouse for product
into two categories. Imprecision in the DMs’ aspiration l and required production capacity of a plant for product l are
levels for the goals is treated in the first category, while given as follows: s1 =1, s2 =1 and q1 =1, q2 =2. The cost
imprecision in both the retailers’ demand and DMs’ coefficients Cijl and Tjkl are computed as being proportional
aspiration levels for the goals is treated simultaneously in to the Euclidean distance among the locations of warehouses
the second category. Solutions of the proposed model are and retailers, and plants and warehouses, respectively.
Specifically, Cijl and Tjkl are drawn from a uniform 6.2.1 Solution by the proposed approach
distribution between 0.025*dtij and 0.035*dtij and 0.045*dtjk
and 0.055*dtjk, respectively. The parameters that are used in Solution of the SC distribution network design problem by
defining the coverage parameter of retailer in zone i (LBi, the proposed approach is presented step by step in the
UBi), are given in Table 4. Throughput limit of warehouse in following.
zone j with capacity level r (Wjr) and capacity of the plant in
– Step 1:
zone k with capacity level h (Dkh) are taken as follows.
The crisp formulation of the problem has been devel-
Wjr =4000, 6000, 8000, 10000, 12000, Dkh =15000,
oped using Eqs. (8 to 17) in Sect. 5.
20000, 30000, 35000, 40000.
– Step 2:
Efficient extreme solutions of the problem are presented
6.2 Implementation category I: treatment of fuzzy
in Table 5.
aspiration levels for the goals
It is assumed here that the DMs don’t choose any of the
efficient extreme solutions as the preferred compromise
In this category of experiments, the imprecision in goal
solution and proceed to Step 3.
achievement is allowed through the specification of an
– Step 3:
interval of acceptable achievement rather than a crisp
Considering the efficient extreme solutions given in
value.
Table 5, the lower and upper bounds of the objectives can
be determined. In our case, the corresponding minimum
µa il
and maximum values of the efficient extreme solutions are
determined as the lower and upper bounds, respectively, as
presented in Table 6.
8
>
> 0 if INV > 2; 994; 668
>
< 2; 994; 668 INV
μINV ¼ if 2; 229; 724 < INV 2; 994; 668
>
> 2; 994; 668 2; 229; 724
>
:
1 if INV 2; 229; 724:
ð19Þ
8
> 1 if TSERVL > 61; 532
< TSERVL 50; 000
μTSERVL ¼ if 50; 000 < TSERVL 61; 532
>
: 61; 532 50; 000
0 if TSERVL 50; 000:
ð20Þ
– Step 4: 9
maximize γλ þ ð1 γ Þ½0:45λ1 þ 0:20λ2 þ 0:35λ3 > >
Considering the membership functions in Step 3, >
>
subject to >
>
and using the ‘modified fuzzy and’ operator, mathema- =
μTCOST λ þ λ1 ;
tical formulation of the problem can be developed as μINV λ þ λ2 ; >
>
>
follows: μTSERVL λ þ λ3 ; >
>
>
;
μTCOST ; μINV ; μTSERVL ; λ; λk ðk ¼ 1; 2; 3Þ; γ 2 ½0; 1
Table 12 Lower and upper bounds for the objectives
and other system constraints (11 to 17). ð21Þ
Objectives Lower bound Upper bound
As we stated previously, the relative weights for the
TCOST 2,236,907 3,250,000 membership functions of the objectives can be determined
INV 1,480,163 2,994,668 by the DMs using various methods. It is assumed here that,
TSERVL 50,000 73,839 the weights are determined by the DMs as presented in the
objective function of the model.
Table 13 Solution results of the model by the proposed solution approach
Considering the results given in Table 10, let’s DMs don’t select any of the efficient extreme solutions as
suppose that the DMs accept again the results of the model the preferred compromise solution and proceed to the next
with γ=0.4. step.
Considering the efficient extreme solutions given in
6.3 Implementation category II: treatment of fuzzy demand Table 11, the lower and upper bounds of the objectives are
and fuzzy aspiration levels for the goals determined as documented in Table 12.
Using the ‘modified fuzzy and’ operator, mathematical
We assume here that the retailers’ demand pattern can be formulation of the problem can be developed as follows:
affected to some extent using demand management tech-
niques. That is, the retailers’ demand can be altered within a 9
max imize >
range. Under this assumption, we state retailers’ demand as >
>
γλ þ ð1 γ Þ½0:45λ1 þ 0:20λ2 þ 0:35λ3 >
>
fuzzy parameters using triangular membership function as >
>
>
>
illustrated in Fig. 6. The lower and upper bounds for each of >
>
subject to =
the retailer’s demand are assumed 80% and 120% of their μTCOST λ þ λ1 ; ð23Þ
corresponding expected demand (ail) values, respectively. μINV λ þ λ2 ; >
>
>
>
>
>
μTSERVL λ þ λ3 ; >
>
μDEMANDil λ; >
>
6.3.1 Solution by the proposed approach >
;
μTCOST ; μINV ; μTSERVL ; μDEMANDil ; λ; λk ; γ 2 ½0; 1
Efficient extreme solutions of the problem are presented in
and other system constraints (11 to 17).
Table 11. The upper part of the table is constructed by
A set of solutions for the above problem with different
solving the problem considering the individual objective
values of γ are obtained and presented in Table 13.
functions subject to fuzzy constraint set while the crisp
The solution results presented in Table 13 are illustrated
constraint set is considered in the lower part.
graphically in Fig. 7.
As in the same manner of our preceding experiment with
The solution results are also compared in terms of
the proposed solution approach, it is assumed here that the
utilization of the plants and warehouses in Table 14.
Fig. 8 Solution results of the
problem with Werners’ ‘fuzzy
and’ operator
Int J Adv Manuf Technol (2008) 36:401–418 415
Comparing the results given in Table 13 together with 6.3.2 Solution by Werners’ ‘fuzzy and’ operator
consideration of the utilization of the plants and ware-
houses, let’s suppose that the DMs accept the results of the Using Werners’ ‘fuzzy and’ operator, we obtain a set of
model with γ=0.5, and that they consider this solution the solutions corresponding to different values of γ. The results
preferred compromise solution. Then, the procedure is are presented in Table 15 and illustrated graphically in
terminated at Step 7. Fig. 8.
If the results of the model with γ=0.5 are compared to
those of the model with γ=1, it can be realized that 6.4 Comparison of the results
substantial increase (45.9%) can be provided in achieve-
ment level of the total cost objective with a decrease by The solution results obtained by the proposed solution
11.7% in those of the second and third objective function. approach and Werners’ ‘fuzzy and’ operator for the two
In terms of utilization rates of the plants and warehouses, different implementation categories are compared in
the model with γ=0.5 provides better results compared to Table 16.
the other solutions. The results presented in Table 16 are illustrated
graphically in Figs. 9, 10 and 11.
If the results presented in Table 16 are analyzed, it can
be seen that, for the first category of experiments, the
Fig. 9 Comparison of the
results in terms of TCOST and 2,900,000
INV objectives 2,850,000
2,800,000
2,750,000
2,700,000
2,650,000
Proposed approach
2,600,000
2,550,000 Werners' approach
2,500,000
2,450,000
2,400,000
TCOST INV TCOST INV
Category I Category II
416 Int J Adv Manuf Technol (2008) 36:401–418
58,000
TSERVL
56,000 Proposed approach
Werners' approach
54,000
52,000
50,000
Category I Category II
0.8
0.7
0.6
Proposed approach
0.5
Werners' approach
0.4
0.3
min aver. min aver. min aver. min aver.
Category I Category II
Int J Adv Manuf Technol (2008) 36:401–418 417
distribution network design problem can be handled in a 20. Dogan K, Goetschalckx M (1999) A primal decomposition
more flexible, robust and realistic way through the method for the integrated design of multi-period production and
distribution system. IIE Trans 31:1027–1036
proposed model and the IFGP-based solution approach. 21. Tragantalerngsak S, Holt J, Ronnqvist M (2000) An exact method
The proposed model can be used in real life industrial for the two-echelon, single-source, capacitated facility location
application in restructuring, i.e., expanding or narrowing, of problem. Eur J Oper Res 123:473–489
an existing SC distribution network besides the design of a 22. Lee HK, Lee I-B, Reklaitis GV (2000) Capacity expansion
problem of multisite batch plants with production and distribution.
new network. Such a strategic model can be a part of a Comput Chem Eng 24:1597–1602
decision support system developed for the collaborative SC 23. Melachrinodis E, Min H (2000) The dynamic relocation and
management practices. phase-out of a hybrid, two-echelon plant/warehousing facility: a
multiple objective approach. Eur J Oper Res 123(1):1–15
24. Sabri EH, Beamon BM (2000) A multi-objective approach to
simultaneous strategic and operational planning in sc design.
Omega 28:581–598
25. Pirkul H, Jayaraman V (2001) Planning and coordination of
References production and distribution facilities for multiple commodities.
Eur J Oper Res 133:394–408
1. Pirkul H, Jayaraman V (1998) A multi-commodity, multi-plant, 26. Tsiakis P, Shah N, Pantelides CC (2001) Design of multi-echelon
capacitated facility location problem: formulation and efficient supply chain networks under demand uncertainty. Ind Eng Chem
heuristic solution. Comput Oper Res 25(10):869–878 Res 40:3585–3604
2. Jang Y-J, Jang S-Y, Chang B-Y, Park J (2002) A combined model 27. Talluri S, Baker RC (2002) A multi-phase mathematical program-
of network design and production/distribution planning for a ming approach for effective supply chain design. Eur J Oper Res
supply network. Comput Ind Eng 43:263–281 141:544–558
3. Chopra S, Meindl P (2001) Supply chain management. Prentice 28. Cakravastia A, Toha IS, Nakamura N (2002) A two-stage model for
Hall, Upper Saddle River, NJ the design of supply chain networks. Int J Prod Econ 80:231–248
4. Church RL, Revelle C (1974) The maximal covering location 29. Beamon BM, Fernandes C (2004) Supply chain network config-
problem. Pap Reg Sci Assoc 32:101–118 uration for product recovery. Prod Plan Control 15(3):270–281
5. Zadeh LA (1965) Fuzzy sets. Inform Control 8:338–353 30. Yeh W-C (2005) A hybrid heuristic algorithm for the multistage
6. Petrovic D, Roy R, Petrovic R (1999) Modeling and simulation of supply chain network problem. Int J Adv Manuf Techol 26:675–685
a supply chain in an uncertain environment. Eur J Oper Res 31. Eskigun E, Uzsoy R, Preckel PV, Beaujon G, Krishnan S, Tew JD
109:299–309 (2005) Outbound supply chain network design with mode
7. Reznik L, Pham B (2001) Fuzzy models in evaluation of selection, lead times and capacitated vehicle distribution centers.
information uncertainty in engineering and technology applica- Eur J Oper Res 165:182–206
tions. Proceedings of the 10th IEEE International Conference on 32. Amiri A (2006) Designing a distribution network in a supply
Fuzzy Systems, Melbourne, Austraila, pp 972–975 chain: formulation and efficient solution procedure. Eur J Oper
8. Francis RL, Mcginnis LF, White JA (1983) Locational analysis. Res 171:567–576
Eur J Oper Res 12:220–252 33. Narasimhan R (1980) Goal programming in a fuzzy environment.
9. Aikens CH (1985) Facility location models for distribution Decision Sci 11:325–336
planning. Eur J Oper Res 22:263–279 34. Hannan EL (1981) Some further comments on fuzzy priorities.
10. Brandeau ML, Chiu SS (1989) An overview of representative Decision Sci 13:337–339
problems in location research. Manage Sci 35:645–674 35. Ignizio JP (1982) On the rediscovery of fuzzy goal programming.
11. Beamon BM (1998) Supply chain design and analysis: models Decision Sci 13:331–336
and methods. Int J Prod Econ 55:281–294 36. Narasimhan R, Rubin PA (1984) Fuzzy goal programming with
12. Avella P et al (1998) Some personal views on the current state and nested priorities. Fuzzy Set Syst 14:115–129
the future of locational analysis. Eur J Oper Res 104:269–287 37. Tiwari RN, Dharmar S, Rao JR (1986) Priority structure in fuzzy
13. Pontrandolfo P, Okogbaa OG (1999) Global manufacturing: a goal programming. Fuzzy Set Syst 19:251–259
review and a framework for planning in a global corporation. Int J 38. Tiwari RN, Dharmar S, Rao JR (1987) Fuzzy goal programming -
Prod Res 37(1):1–19 an additive method. Fuzzy Set Syst 24:27–34
14. Meixell MJ, Gargeya VB (2005) Global supply chain design: a 39. Bellman RE, Zadeh LA (1970) Decision making in a fuzzy
literature review and critique. Transport Res E-Log 41:531–550 environment. Manage Sci 17:141–164
15. Geoffrion AM, Graves GW (1974) Multicommodity distribution 40. Zimmermann H-J (1978) Fuzzy programming and linear program-
system design by benders decomposition. Manage Sci 20(5):822–844 ming with several objective functions. Fuzzy Sets Syst 1:45–55
16. Brown GG, Graves GW, Honczarenko MD (1987) Design and 41. Lai Y-J, Hwang C-L (1994) Fuzzy multiple objective decision
operation of a multicommodity production/distribution system making. Springer, Berlin Heidelberg New York
using primal goal decomposition. Manage Sci 33(11):1469– 42. Abd El-Wahed WF, Lee SM (2006) Interactive fuzzy goal
1480 programming for multi- objective transportation problems. Omega
17. Cohen MA, Lee HL (1988) Strategic analysis of integrated 34:158–166
production-distribution systems: models and methods. Oper Res 43. Werners B (1987) Interactive multiple objective programming
36:216–228 subject to flexible constraints. Eur J Oper Res 31:342–349
18. Pirkul H, Jayaraman V (1996) Production, transportation, and 44. Werners B (1987) An interactive fuzzy programming system.
distribution planning in a multi-commodity tri-echelon system. Fuzzy Set Syst 23:131–147
Transport Sci 30:291–302 45. Leung Y (1987) Hierarchical programming with fuzzy objectives
19. Jayaraman V (1998) An efficient heuristic procedure for practical- and constraints. In: Kacprzyk J, Orlovski SA (eds) Optimization
sized capacitated warehouse design and management. Decision models using fuzzy sets and possibility theory. D. Reidel,
Sci 29:729–745 Dordrecht, pp 245–257
418 Int J Adv Manuf Technol (2008) 36:401–418
46. Fabian Cs, Ciobanu Gh, Stoica M (1987) Interactive polyopti- 48. Babtistella LFB, Ollero A (1980) Fuzzy methodologies for
mization for fuzzy mathematical programming. In: Kacprzyk J, interactive multicriteria optimization. IEEE T Syst Man Cyb
Orlovski SA (eds) Optimization models using fuzzy sets and 10:355–365
possibility theory. D. Reidel, Dordrecht, pp 272–291 49. Werners B (1988) Aggregation models in mathematical program-
47. Sasaki MY, Nakahara Y, Gen M, Ida K (1991) An efficient ming. In: Mitra G, Greenberg HJ, Lootsma FA, Rijckaert MJ,
algorithm for solving fuzzy multiobjective 0-1 linear program- Zimmermann H-J (eds) Mathematical models for decision
ming problem. Comput Ind Eng 21:647–651 support. Springer, Berlin Heidelberg New York, pp 295–305