62 ieee transactions on systems, man, and …ceit.aut.ac.ir/~sa_hashemi/my teachings/ms-ceit-supply...

62 IEEE TRANSACTIONS ON SYSTEMS, MAN, AND CYBERNETICS—PART A: SYSTEMS AND HUMANS, VOL. 36, NO. 1, JANUARY 2006

Design and Optimization of Integrated E-SupplyChain for Agile and Environmentally

Conscious ManufacturingMariagrazia Dotoli, Member, IEEE, Maria Pia Fanti, Senior Member, IEEE, Carlo Meloni, and

MengChu Zhou, Fellow, IEEE

Abstract—An agile and environmentally conscious manufactur-ing paradigm refers to the ability to reconfigure a flexible systemquickly, economically, and environmentally responsibly. In mod-ern manufacturing enterprises, e-supply chains integrate Internetand web-based electronic market and are promising systems toachieve agility. A key issue in the strategic logistic planning ofintegrated e-supply chains (IESCs) is the configuration of thepartner network. This paper proposes a single- and multiob-jective optimization model to configure the network of IESCs.Considering an Internet-based distributed manufacturing systemcomposed of different stages connected by material and infor-mation links, a procedure is presented to select the appropriatelinks. A set of performance indices is associated with the networklinks. Single-criterion and multicriteria optimization models arepresented under structural constraint definitions. The integer lin-ear programming (ILP) problem solution provides different net-work structures that allow to improve supply chain (SC) flexibility,agility, and environmental performance in the design process.The proposed optimization strategy is applied to two case studiesdescribing two networks for desktop computer production.

Index Terms—Agile manufacturing, computer integrated man-ufacturing, integer linear programming, optimization method,supply chain, system analysis and design.

I. INTRODUCTION

AGILE manufacturing has been introduced as a conceptto satisfy the demand for low-volume and high-variety

products. It requires embedding production flexibility into(re)configuration and control of manufacturing systems [28],[29]. By integrating computer systems, hardware, and infor-mation flows, automated manufacturing systems provide agilemanufacturing with flexibility and reconfigurability. Flexibilityis a manufacturing system’s ability to adjust to customers’preferences and reconfigurability is the ability to meet the

Manuscript received November 30, 2004; revised April 16, 2005. This workwas supported in part by the Italian Ministry for University and Research(MIUR) under project 2003090090 and in part by the National Science Founda-tion of China (NSFC) under Grant 60228004 and Grant 60334020. This paperwas recommended by Guest Editor R. G. Qiu.

M. Dotoli, M. P. Fanti, and C. Meloni are with the Dipartimento diElettrotecnica ed Elettronica, Politecnico di Bari, 70125 Bari, Italy (e-mail:[email protected]; [email protected]; [email protected]).

M. Zhou is with the Department of Electrical and Computer Engineering,New Jersey Institute of Technology, Newark, NJ 07102 USA and also withthe Laboratory of Complex Systems and Intelligence Science, Institute ofAutomation, CAS, Beijing 100080, China (e-mail: [email protected]).

Digital Object Identifier 10.1109/TSMCA.2006.859189

changing demand by reconfiguring the system structure [10],[17], [29]. Customers’ expectation in terms of cost, quality,services, and environmental impact has put industries underpressure to become more agile and environmentally responsive,and provide time- and cost-effective product realization and de-liveries under highly dynamic market and supply conditions. Acommon and accepted way to achieve agility in a manufacturingsystem is manufacturing products in geographically differentsites connected through communication networks [27]. In otherwords, an agile manufacturing system is composed of variousresources that may belong to different companies and canbe viewed as a complex supply chain (SC) network. Moreprecisely, an SC network is defined as a collection of inde-pendent companies, possessing complementary skills and beingintegrated with streamlined material, information and financialflow [24]. Furthermore, such a system has to be dynamically re-configurable according to product and market changes and hasto provide high utilization of resources belonging to differentcompanies.

The core of highly competitive and efficient SC networksis electronic information: Internet and web-based electronicmarket places can provide an inexpensive, secure, and pervasivemedium for information transfer between business units. Ine-commerce, business, partners, and customers connect to-gether through Internet or other electronic communication sys-tems to participate in commercial trading or to interact witheach other [3], [9], [11], [14]. Certainly, e-commerce poses newand hard issues in the company’s logistic systems, involving insome cases new distribution concepts and new design of theSC. Hence, integrated e-supply chain (IESC) is an emergingbusiness strategy that integrates the SC design and power ofe-commerce in order to obtain more flexible and agile manufac-turing processes [3], [8], [9], [11], [16], [19]. In the formationof an effective dynamic IESC network, the selection of partnersin each tier of the SC for fulfillment of each order is extremelyimportant [24]. In previous works [4], [5], the authors proposeda configuration methodology for IESC network design andpartner selection. In order to deal with its complexity, thedecision process is divided into three hierarchical levels: 1) thecandidate selection level, realized by different multicriteria dataanalysis techniques; 2) the network design level, solved by mul-ticriteria optimization problems; and 3) the solution/validationlevel, which considers the impact of higher levels on the oper-ational layer. In [5], the authors focus particularly on first-level

1083-4427/$20.00 © 2006 IEEE

DOTOLI et al.: DESIGN AND OPTIMIZATION OF IESC FOR AGILE AND ENVIRONMENTALLY CONSCIOUS MANUFACTURING 63

decision problems, i.e., the candidate selection module, andpropose several criteria to select the partners of each stage onthe basis of aggregate performance indices. This paper focuseson second-level problems, i.e., network design and optimiza-tion. In each IESC stage, one or more partners (producers,manufacturers, consumers, etc.) have to be selected, makinguse of the information available, for example, on cost, trans-portation mean, distance, and pollution. Significant literaturedeals with the problem of SC component selection and networkdesign [7], [8], [13], [17], [22], [24], [25], [27] to achieveagility and flexibility. Indeed, among the problems encounteredin strategic logistic planning, network configuration and designare considered the most significant issue [1], [13]. In particular,Wu et al. [27] pioneered in formulating partner selection asan integer programming problem to choose one and only onecandidate for each task of the production process. The authorsminimize the sum of the costs for performing all the tasks andthe transportation costs. However, the constraint to select onlyone candidate for each task is restrictive and affects the flexibil-ity of the resulting network structure. Moreover, Jang et al. [13]propose an optimized supply network design module and aplanning scheme of production and distribution activities thatare modeled as three decomposed mathematical formulation.Each mathematical submodel of the SC network is optimizedby a Lagrangian heuristic and an integration methodology.However, the complete procedure appears complex and needsa genetic algorithm procedure to generate the final integratedproduction distribution plan. On the other hand, Viswanadhamand Gaonkar [24] study and analyze a multi-tier SC by usingan optimization technique that takes into account capacitiesand costs. They determine the suboptimal order quantitiesto be allocated to each of the manufacturers, suppliers, andlogistics service providers. A multi-integer programming modelis developed for a dynamic manufacturing network, and itsobjective function maximizes the profit earned by the networksubject to various capacity, production and logistics schedules,and flow balancing constraints. However, information neces-sary to describe and characterize the SC can be so complexand large that the definition of constraints appears critical. Toface such complexity, Talluri and Baker [17] consider a three-phase approach to design an SC: phases I and II design thenetwork and phase III addresses operational issues. However,the network design procedure does not consider the transporta-tion connections among the stages in network design and therouting of material is analyzed in the third phase only. Finally,Luo et al. [14] present a novel approach to describe andoptimize an IESC network incorporating e-commerce andelectronic linkages between actors with the associated envi-ronmental impact. The IESC network is composed of differ-ent consecutive stages connected by material and informationlinks describing costs, distances, pollution, and CO2 emission.Moreover, the structure of the IESC is modeled by a digraphwhere nodes are stage partners and edges are links. Assigningdifferent costs to the material links, Luo et al. [14] obtain theperformance indices and the optimal flow material network bya fuzzy multiobjective optimization approach.

This paper proposes a configuration strategy for IESCnetwork design, considering also the e-business relationship

between operators, network environmental impact, and deman-ufacturing to extract valuable raw materials and componentsfrom products upon their retirement [3], [8], [30], [31]. Inparticular, e-commerce is considered also between differentnonconsecutive stages of the chain, i.e., between suppliers andconsumers, manufacturers and retailers, etc. Indeed, the Inter-net encompasses a wide spectrum of potential commercial ac-tivities and influences the material transport and delivery amongstages. The impact of e-commerce is here taken into account bythe evaluation of particular performance measures assigned tothe links connecting the stage partners [2], [18], [23]. Moreover,single criterion and multicriteria objective problems are formu-lated to optimize the IESC network configuration consideringat the same time material and information connections. Hence,an integer linear programming (ILP) problem provides a set ofpossible alternative solutions. In addition, the IESC network isdescribed by the digraph model proposed in [14], and the objec-tive functions and constraints are characterized and defined onthe basis of the digraph analysis. The presented methodologyallows us to build the IESC network by selecting the edgesof the digraph, for example, to introduce e-commerce linksand recycling links or to impose one or more manufacturers inthe network. Consequently, comparing our methodology withthe optimization technique proposed in [14], our formulationappears more flexible and suited to improve the IESC structure,satisfying the different design requirements such as electroniclinks, material transporters, demanufacturing, etc. The opti-mization model is applied to two case studies producingdesktop computer systems. The first case study is solved in[14], hence, the results obtained by the ILP are compared withthe solutions inferred by the fuzzy-optimization approach.Moreover, in the second case study, the set of obtained networksshows the improvement in flexibility and agility of the IESCstructure.

This paper is organized as follows. In Section II, the IESCstructure and the associated digraph are described. Section IIIsynthesizes the optimization model that is described in de-tail for the two case studies in Sections IV and V. Finally,Section VI summarizes the conclusions.

II. IESC NETWORK MODEL

A. IESC Description

An IESC network can be defined as a hypernetwork ofmaterial flows overlaid with an e-business information network.The considered IESC contains different stages: raw materialsupply, intermediate supply, manufacturing, distribution, retail,customers, and demanufacturing or recycling. For example, weconsider inbound stages in which partners are raw materialsuppliers or plant stages. Moreover, the distribution stages havepartners such as manufacturers, product distributors, and ware-houses. Finally, outbound stages can be composed of retailers,customers, recyclers, and demanufacturers. After the deman-ufacturing stage, recovered material, components, or energyfeedback to suitable SC stages are considered. In the IESCstructure, Internet and e-business extend the benefits of conven-tional proprietary electronic data interchange systems to all the


Fig. 1. Structure of a generic IESC network.

stages of the SC, leading to reduction in procurement cyclesand costs, shortened product development times, and reducedinventories. Moreover, e-business provides direct connectionsbetween final consumers and manufacturers.

We describe an IESC as a set of stages [14], denoted byST = {P1, . . . ,Pk, . . . ,PNs

}, where NS is the number ofstages. In addition, each stage Pk is described as a set of sk

partners representing different actors of the IESC, i.e., Pk ={nik

, nik+1, nik+2, . . . , nik+sk−1}, where ik is the generic in-dex such that ik =

∑k−1h=1 sh with k = 2, 3, . . . , Ns and i1 = 1.

We suppose that there are N partners in the system. In addition,the (k − i)th stage is called the upstream stage of Pk and the(k + i)th stage is called the downstream stage of Pk, withi > 0. Moreover, the bill of material (BOM) of stage Pk is aset of material and components required for processes in thekth stage and produced by upstream stages.

Furthermore, the partners of different IESC stages canbe connected by two types of links: material flow links(m-links) and information links (e-links). More precisely, anm-link represents the physical transportation link betweentwo partners and multiple m-links are allowed between twopartners to model different transportation modes or splitdelivery routes. The second type of link is the e-link, modelinge-business relationship between business entities for streamlin-ing the material flow efficiently and effectively. We assumethat two partners can be connected by m-links and/or bye-links and that an e-link may connect two partners of the SCalso without the presence of an m-link. Hence, the proposedstructure is able to extend the traditional SC into a moresustainable and integrated production system. Note that IESCpartners belonging to the same stage are not connected bylinks since in the considered model material and informationflow through different stages. Formally, the m-links of stagePk are denoted by the set Lmk

= {mij}, where mij is anm-link starting from ni ∈ Pk and ending in nj ∈ Ph, with Pk,Ph ∈ ST and k �= h. The set Lm = ∪Lmk

denotes the overallmaterial flow link set. Analogously, we denote the e-links set ofstage Pk by Lek

= {eij}, where eij is an e-link starting fromni ∈ Pk and ending in nj ∈ Ph, with Pk,Ph ∈ ST and k �=h. The set Le = ∪Lek

denotes the e-link set and L = Lm ∪ Le

is the complete set of links of the IESC. Fig. 1 depicts a genericIESC network.

Moreover, we conclude the description of the IESC net-work by introducing the set of performance indices M ={M1,M2, . . . ,MNM

}, where each element Mq ∈ M corre-

sponds to a performance measure. Typical indices includecost, transportation and process time, product quality, energyconsumption, and environmental impact [2], [14]. A perfor-mance value is assigned to each link, considering m- ande-links: Mq(mij)(Mq(eij)) with q = 1, . . . , NM denoting thevalue of the performance measure Mq associated with the linkmij ∈ Lmk

(eij ∈ Lek). Particularly, an e-link speeds up the

communication process and thus reduces the response time,affecting performance measures such as cost and productivity.Hence, if two partners (for example, ni ∈ Pk and nj ∈ Ph)are connected by both e-link and m-link (i.e., mij and eij),the performance measures are suitably updated and are asso-ciated with the m-link mij only. On the other hand, if just ane-link connects two IESC actors, then it models that fact thatno material flow is possible between the two partners. In such acase, the performance measure that is assigned to the e-link isthe cost of information, such as Internet portals, websites, andelectronic databases.

B. Digraph Definition

To exhibit the interactions among the stages of IESC, wedefine a direct graph D = (N,E). The node set N representsthe complete partner set of the network and each node ni ∈ Nfor i = 1, . . . , N is associated with partner ni ∈ Pk for k ∈{1, . . . , NS} of the SC network. For the sake of simplicity, thesame symbols indicate nodes and partners. Moreover, the edgeset E is such that an arc yij directed from ni to nj is in E ifthere exists an m-link mij ∈ Lm and/or e-link eij ∈ Le. Wedenote with E the number of edges in D.Example 1: Fig. 2 shows an example of e-supply chain com-

posed of NS = 6 stages: four suppliers, one manufacturer, twodistributors, two retailers, one consumer, and four recyclers,for a total of N = 14 partners. The IESC exhibits the m- ande-links m18, e18, m17, e17, m4,10, e4,10, m78, e78, m79, e79,m7,10, e7,10, m9,10, e9,10, m8,10, and e8,10, while the remaininglinks are m-links. Moreover, Fig. 3 depicts the digraph describ-ing the corresponding IESC network. The digraph D = (N,E)has N = 14 nodes and E = 23 edges. Obviously, edges y18,y17, y4,10, y78, y79, y7,10, y9,10, and y8,10 are associated bothwith m- and e-links and the remaining edges of the digraphsare associated with m-links only. Moreover, each edge in Eis labeled by the corresponding variable xh with h = 1, . . . , Eused in the optimization procedure and defined in the nextsection.


Fig. 2. Stages of the IESC network for case study 1.

Fig. 3. Digraph associated with the IESC of case study 1.

III. OPTIMIZATION MODEL

This section develops an optimization model to design theIESC network. The procedure starts from the knowledge of thedigraph D = (N,E) that takes into account all the possibleactors belonging to IESC stages and all the possible links thatcan connect the considered partners. Hence, an optimizationtechnique has to select a subdigraph D∗ = (N∗,E∗) withN∗ ⊂ N and E∗ ⊂ E that corresponds to an IESC subjectto structural constraints and exhibiting optimal or suboptimalperformance indices. To this aim, an ILP problem is establishedconsidering single-criterion or multicriteria optimization undera set of constraints obtained by the structure of the IESC. Moreprecisely, the objective of the model is to minimize a single-criterion or multicriteria cost function subject to the constraintsthat we characterize as BOM, path, mutual exclusion, andstructural constraints.

In order to state the optimization model and to use a clearmathematic notation, we introduce the decision variable vectorx = [x1x2, . . . , xE ]T, where each element xh ∈ {0, 1} withh = 1, . . . , E is associated with an edge yij ∈ E (see theexample in Fig. 3). More precisely, the value of xh indicatesthe presence (xh = 1) or absence (xh = 0) of edge yij ∈ E inthe solution digraph [12].

The objective function is obtained by considering the perfor-mance index functions Mq that assign to each m-link mij ande-link eij the values Mq(mij) and Mq(eij), respectively, andto each m-e-link mij-eij the comprehensive value Mq(mij).Consequently, value Mq(mij) or Mq(eij) is associated withedge yij ∈ E and with the corresponding variable xh. Let us in-

dicate with cq = [cq1c

q2, . . . , c

qE ]T the vector of E entries where

the hth entry is cqh = Mq(mij) or cq

h = Mq(eij) associatedwith yij ∈ E and the decision variable xh labeling yij . Theoptimization problem is as follows

z = min(f(x)) (1)

subject to

Ax ≥ B (2)

xh ∈ {0, 1} for h = 1, . . . ,E (3)

where A is the constraint matrix of dimension v × E andB is a v-entry vector of integers, v representing the numberof constraints. Minimizing the objective function f(x) meanseither to minimize only one performance index (Problem 1)or to minimize a subset of the chosen performance indices(Problem 2), i.e., f(x) represents either a single-criterion ormulticriteria objective function.

A. Objective Function Definition

The Single-Criterion Objective Function of Problem 1: It isdefined as

f(x) = (cq)Tx. (4)

Each solution x∗ of the ILP problem (1)–(4) for a particularvector cq corresponds to a possible IESC structure. Moreprecisely, the optimal solution vector x∗ selects a subdigraphD∗ = (N∗,E∗) of D. If the hth entry of x∗ is x∗

h = 1 and xh


is associated to edge yij outgoing from ni and incoming to nj ,then the solution selects edge yij ∈ E∗ and nodes ni, nj ∈ N∗.In other words, the optimal IESC with respect to index Mq isdescribed by subdigraph D∗ that exhibits the actors (nodes) andlinks (edges) selected in the IESC design.Multicriteria Objective Function of Problem 2: It is

defined as

f(x) = Cx (5)

where

C =

(cq1)T

. . .(cqQ)T

is a qQ × E criteria matrix and cq1, . . . , cqQ are E-entryvectors associated with performance indices Mq1, . . . ,MqQ,respectively.

The set of solutions of the multicriteria ILP problem (1)–(3)and (5) provides the maximal Pareto face of the solution set [6].More precisely, we obtain a subset of solutions X∗ = {x∗

i},where each x∗

i ∈ X∗ is a Pareto optimal solution correspond-ing to a subdigraph D∗

i of D and to an IESC structure.

B. Constraint Definition

BOM Constraints: Each candidate supplier can provide asubset of materials (or components) to each producer. Anal-ogously, the retailer actors provide a subset of products toeach consumer. Generally, the BOM of each partner of themanufacturer stage or consumer stage is described as a list ofspecified materials needed by the stage. We distinguish betweentwo cases: 1) the manufacturer and consumer stages includeonly one actor each (case A); 2) they have more than one partnereach (case B). The BOM constraints for both cases are definedas follows.

Case 1) The actor of the considered stage Pk (for exam-ple, manufacturer or consumer) has to obtain allthe BOM components. For instance, let us supposethat the manufacturer requires all the componentsnecessary to assemble the final product. In addition,suppose that each of these products can be equallyshipped by means of three different edges respec-tively labeled with variables x1, x2, and x3. Thiscondition can be written as an inequality in 0–1 vari-ables: x1 + x2 + x3 ≥ 1. Hence, if we suppose thatthere are v1 BOM constraints, then the followinginequality constraints are formulated, i.e.,

A1x ≥ 1 (6)

where 1 is a v1-dimensional vector with all elementsequal to 1 and A1 is a v1 × E constraint matrixdefined by

A1 =

a11...

av11

(7)

Fig. 4. Fictitious edges and nodes.

with ai1 = [a1(i, 1), a1(i, 2), . . . , a1(i, E)] repre-

senting the ith row of A1 and a1(i, j) ∈ {0, 1} forj = 1, . . . , E and i = 1, . . . , v1.

Case 2) If there are k actors in the considered stage Pk

(manufacturer or consumer), the designer has toselect one or more partners in the stage. We modelthis situation by introducing a fictitious node nfi

and a fictitious edge labeled with variable xfiwith

i = ik, . . . , ik + sk − 1 associated with each nodeni ∈ Pk (see Fig. 4). Hence, vector x is modified,introducing k elements xfi

with i = 1, . . . , k: ifentry xfi

= 1 in the solution vector x∗, then actorni ∈ Pk is chosen. Summing up, the BOM con-straints for each actor ni ∈ Pk are modified as

ai1x ≥ xfi

, with xfi∈ {0, 1}. (8)

Path Constraints: It is necessary to select in the digraph atleast a path starting from a node of the producers and ending atthe nodes of consumers. To impose this condition, we associatewith the digraph the N × E incidence matrix IM where eachelement is IM (i, j) ∈ {−1, 0, 1}. More precisely:

IM (i, j) = 0 if the arc labeled by xj does not belong tonode ni;

IM (i, j) = −1 if the arc labeled by xj starts from node ni;IM (i, j) = 1 if the arc labeled by xj ends at node ni.

Moreover, to define a constraint that imposes the presence ofa path starting from node nh and ending at node nW , we in-troduce the N -vector bh,w = [b1, b2, . . . , bN ]T with bh = −1,bW = 1, and bp = 0 for p �= h, w and p = 1, . . . , N . Theconstraint is written as

IMx ≥ bh,w. (9)

Hence, the constraint submatrix and left side vector are

A2 =

IM...

IM

and B2 =

bh1,w1

...bhi,wi

. (10)

Mutual Exclusion Constraints: In some cases, it is necessaryto choose only one actor in a stage. In other words, the condition“at maximum one edge among those labeled with variables x1,x2, and x3 can be in the solution digraph” is written with thefollowing inequality in 0–1 variables: x1 + x2 + x3 ≤ 1. Onthe other hand, the condition “one and only one edge amongthose labeled with variables x1, x2, and x3 has to be in thesolution digraph” is expressed by x1 + x2 + x3 = 1. Hence, v3

mutual exclusion constraints can be expressed by

A3x ≤ 1 (11)


where 1 is a v3-vector with all elements equal to 1 and matrixA3 is a constraint matrix defined by

A3 =

a13...

av33

(12)

with ai3 = [ai

3(i, 1), ai3(i, 2), . . . , ai

3(i, E)] and ai3(i, j) ∈

{0, 1} for j = 1, . . . , E and i = 1, . . . , v3.Structural Constraints: Some particular structural con-

straints are related to a digraph. For example, the followingcondition can be imposed: “if edge corresponding to x1 belongsto the solution digraph, then edges corresponding to x2 or x3

belong to the solution digraph.” This condition is expressedby the constraint x2 + x3 ≥ x1. A second type of structuralcondition is “edge corresponding to x2 belongs to the solutiondigraph if and only if (iff) edge labeled with variable x1 belongsto the solution digraph,” i.e., x2 = x1. Therefore, v4 structuralconstraints can be expressed by

A4x ≥ 0 (13)

where 0 is a v4-vector with all elements equal to 0 and matrixA4 is a constraint matrix defined by

A4 =

a14...

av44

(14)

with ai4 = [ai

4(i, 1), ai4(i, 2), . . . , ai

4(i, E)] and ai4(i, j) ∈

{−1, 0, 1} for j = 1, . . . , E and i = 1, . . . , v4.Finally, the constraint matrix and specification vector are

A =

A1

A2

−A3

A4

and B =

1B2

−10

. (15)

The optimization models (1) and (4) or (5) subject toconstraints (2) and (3) can be solved by applying standardalgorithmic approaches to ILP problems [26]. However, it isverified that the defined constraints give to the model a par-ticular structure that plays an important role in the effectivesolution of the problem. Indeed, the constraints have a logicalstructure that falls into the case of the shortest path problem[12], [26]. Consequently, the experimental results show that thelinear programming (LP) relaxation of the problem gives aninteger (0,1) solution. Hence, the proposed model is suitableto be straightforwardly solved applying standard approachesto LP problems. It is also verified that the integer solutionis obtained by the standard two-phase simplex method [15].On the other hand, the case studies considered in the followingsection model high level and aggregate real IESCs exhibitingtypical optimization problem dimension. Note that the rigorousproof of these results remains open.

C. Application Methodology

The proposed IESC optimization model is an instrumentto design an SC so that its requirements are satisfied. Moreprecisely, the steps that the decision makers have to accomplishare the following.

1) Identify the SC stages and all the possible partners thatcan compose them.

2) Determine all the possible links that can connect thepartner stages, i.e., single out all the possible material andinformation connections among the IESC partners.

3) Identify the performance indices that can characterizeeach material connection (e.g., cost, environment impact,and type of transporters) and each electronic link (e.g.,cost). If two partners are connected with an m-link ande-link, then establish how the e-business information caninfluence the material connection in terms of the definedperformance indices. This step is very important andis crucial for the good results of the procedure. Theavailability of qualified expert personnel and accuratehistorical database about IESC components can help inaccomplishing this task.

4) Define the optimization model. More precisely, select theperformance indices to optimize and impose the suitableconstraints. By this step, the designers decide IESC struc-tural characteristics e.g., the presence of recycling, one ormore manufacturers, and e-commerce connections.

5) Perform the optimization algorithm so that only onesolution (in the case of Problem 1) or a set of solutions(in the case of Problem 2) is obtained.

IV. CASE STUDY 1

To illustrate the network design optimization procedure, weconsider a case study inspired by an example proposed in[14]. The target product is a typical desktop computer systemconsisting of the computer, hard disk driver, monitor, keyboard,and mouse. The IESC network corresponds to the networkdescribed in Example 1 and depicted in Fig. 2, while its digraphis shown in Fig. 3. The data for the case study are reportedin Table I [14], which shows the values of each performanceindex Mq with q = 1, . . . , 4 associated with the links of theconsidered IESC. More precisely, the adopted performanceindices are total costs (M1), energy (M2), CO2 emission (M3),and cycle time (M4). We indicate generically by cycle timeassociated with an m-link or m-e-link the related time requiredby the transportation and/or production process. The consideredperformance index values are reported in Table I and depend onthe type of link (m- and e-link or m-link), the distance betweenthe connected SC partners, the transportation mode (truck, car,airplane, etc.), and the type of material to be transported. Inparticular, the cost and energy performance indices reportedin the last two rows of Table I, respectively, associated withlinks m11,5 and m14,3 are negative. In fact, in the recycler stageP6, partner n11 is a demanufacturer with an output link m11,5

connecting to manufacturer n5, and partner n14 is a materialrecoverer with an output link m14,3 connecting to supplier n3

(see Fig. 2). Hence, the total cost and energy associated with


TABLE IDATA SHEET FOR THE NETWORK LINKS IN CASE STUDY 1

links m11,5 and m14,3 are negative, i.e., they correspond torecycling material and parts.

Various computational experiments are performed to mini-mize cost, energy consumption, CO2 emission, and total leadtime (TLT), alternatively. In particular, TLT is defined as thetotal time elapsed from the instant at which the raw materialbegins its travel until the instant the finished product is deliv-ered to consumers. Furthermore, a multiobjective function forProblem 2 is chosen. The solutions are obtained by implement-ing the well-known two-phase simplex method in the Matlabframework [20], [21].

A. Constraint Definition

BOM Constraints: The component supplier constraints areobtained by taking into account that the BOMs of the secondstage in Fig. 2, representing the manufacturer, are the computer(C), hard-disk-driver (H), monitor (M), and keyboard/mouse(K). We assume that C is produced by n1 and n2, H is producedby n1, n2, and n3, M is produced by n2, n3, and n4, and K isproduced by n3 and n4 [14]. Hence, with reference to Fig. 3,the constraints imposed on the variables labeling the edgesare as follows:

x3 + x4 ≥ 1

x3 + x4 + x5 ≥ 1

x4 + x5 + x6 ≥ 1

x5 + x6 ≥ 1. (16)

Path Constraints: Case study 1 includes only one manufac-turer and only one consumer (node n5 of stage P2 and n10 of

Fig. 5. Solution digraph of min(f1).

P5, respectively, in Fig. 2). Hence, a path between nodes n5 andn10 is needed. Consequently, we build the N × E incidencematrix IM associated with digraph D. Moreover, we definethe 23-vector b5,10 = [b1, b2, . . . , b23] with b5 = −1, b10 = 1and bp = 0 for p �= 5, 10 and p = 1, . . . , 23. The constraint thatimposes the presence of a path starting from node n5 and endingat node n10 is written as

IMx ≥ b5,10. (17)

Mutual Exclusion Constraints: It is assumed that the designconditions of the SC network include the hypothesis that oneand only one partner is to be included in the recycler stage(stage P6 in Fig. 2). Furthermore, only one type of commerceis present between the second and third stages and one andonly one m-e-link is present among the first stage and theothers. Hence, with reference to Fig. 3, the mutual exclusionconstraints are

x18 + x19 + x20 + x21 ≤ 1

x13 + x14 + x15 ≤ 1

x1 + x2 + x7 = 1. (18)

Structural Constraints: The constraints derived from thedigraph structure in Fig. 3 are

x22 − x18 = 0

x23 − x21 = 0

x5 − x23 ≥ 0

x16 − x1 ≥ 0

x13 + x14 + x15 − x2 ≥ 0

x16 − x13 ≥ 0

x17 − x15 ≥ 0. (19)

For example, the first constraint of (19) means that the edgecorresponding to x22 is selected iff the edge labeled by x18 isselected. In addition, the third constraint of (19) means that ifthe edge labeled by x23 is selected, then the edge correspondingto x5 is selected.





TABLE IIVALUES OF OBJECTIVE FUNCTIONS f1, f2, f3, AND f4 FOR PROBLEM 1

OF CASE STUDY 1

B. Solution of Problem 1

Problem 1 is solved with respect to four objectives alter-natively: costs, energy, CO2 emission, and TLT. The corre-sponding objective functions are denoted by f1, f2, f3, and f4,respectively. The obtained subdigraphs are presented inFigs. 5–8, respectively, and the corresponding objective func-tion values are given in Table II.

Both the solution digraphs respectively minimizing cost andenergy include a recycler (i.e., n14 in Fig. 5 and n11 in Fig. 6),which is missing in the network that minimizes only CO2

emission and TLT (depicted in Figs. 7 and 8), respectively.This result is obvious since including the recycling stage in theIESC network reduces cost and energy (by noticing the negativeentries in the last two rows of Table I) but does not decreaseCO2 emission and TLT.

Fig. 9. Digraph structure of a traditional SC composed of m-links.

Comparing our results with the solutions reported in [14], weremark the following two aspects.

First, while the results of the optimization problems min(f2)and min(f4) provide the same results as [14], the minimizationof the objective functions f1 and f3 does not provide the samedigraphs obtained in [14]. Indeed, the fuzzy optimization canlead to suboptimal solutions: the optimal value of costs andCO2 emission obtained with ILP is f1 = 946.02 US$ and f3 =11.38 KgCE, respectively, while the fuzzy optimization per-formed in [14] determines two solutions with f1 = 951.00 US$and f3 = 14.10 KgCE, respectively. Consequently, the ILPapproach with single criterion objective function guarantees theoptimal solutions.

Second, in [14], the authors use the same structure of BOMconstraints for all the considered performance indices, but sucha structure is not suited to the TLT performance measure.Indeed, the cycle time associated with BOM constraints is notthe sum of the corresponding edge performance indices but themaximum among the performance indices. For example, if wechoose edges y25 and y35 corresponding to variables x4 andx5 as BOM for P2, the corresponding cycle time cannot becomputed as M4(m25) + M4(m35), but as the larger betweenM4(m25) and M4(m35): in such a case, the constraint becomesnonlinear. Hence, to obtain a more rigorous model but withlinear constraints, we modify the constraints (16) to transformthe nonlinear BOM constraints for the TLT in suited linearconstraints [15]. However, since the cycle times assigned tolinks mij ∈ Lm do not differ much, in this particular case theassumption used in [14] is admissible and we obtain the samesolution digraph of problem min(f4) (see Fig. 8).

Finally, the following example shows that the presentedoptimization method can improve the reconfigurability of thenetwork. In particular, let us consider a traditional SC that has anetwork composed of m-links only (see Fig. 9). Moreover, thedesigner has to add the e-links in order to introduce e-commerceand e-business fixtures in the network structure minimizing thecost. Consequently, Problem 1 is solved by selecting cost asthe performance index subject to constraints (16)–(19) and the


Fig. 10. Solution digraph of min(f1) imposing a fixed structure of m-e-links.

TABLE IIIVALUES OF THE PERFORMANCE INDICES FOR PROBLEM 2 OF CASE

STUDY 1: OPTIMAL SOLUTIONS FOR MULTIOBJECTIVE FUNCTION

COST, ENERGY, AND CO2 EMISSION (min(f5))

following mutual exclusion constraints that impose the initialstructure of the SC:

x4 = x5 = x9 = x19 = 1. (20)

The resulting network is depicted in Fig. 10 and exhibits acost of 979.32 US$.

C. Solution of Problem 2

The multiobjective optimization problem is solved consid-ering the following performance indices: cost, energy, andCO2 emission (f5). According to previous remarks, we donot consider the cycle time in the multiobjective optimization,but we compute the TLT values of the problem solutions.Table III reports the ILP solutions and corresponding valuesof the performance indices. The results show the efficiency ofthe proposed method that is able to provide a set of optimalsolutions. For example, solution xA, obtained by minimizingthe objective function f5, is equal to the solution obtainedminimizing f1 (compare the first row of Table III with Fig. 5).Table III shows that solution xA exhibits a large value of CO2

emission. On the other hand, minimizing objective function f5

provides solution xD in Table III, featuring satisfactory valuesof cost, energy, CO2 emission, and TLT. In other words, thebenefits of using a multicriteria optimization approach are in thefact that the method enables us to choose among several near-optimal solutions. The digraph corresponding to solution xD inTable III is depicted in Fig. 11, and the other solution digraphsmay easily be obtained from the last column in Table III.

Comparing the results obtained by solving the ILP problemwith the fuzzy optimization results in [14], we remark that thepresented optimization method provides a set of near-optimal

Fig. 11. Digraph representing solution xD of Problem 2 of case study 1.

solutions instead of only one solution. Hence, the designer canbe guided by priorities and preferences to choose a satisfactoryIESC network, improving system flexibility and agility. Indeed,solution xB of Table III is the solution obtained in [14] by fuzzyoptimization.

V. CASE STUDY 2

To improve the flexibility and agility of the SC, the structureof the IESC of case study 1 is modified by introducing in thesecond stage three manufacturers and in the fifth stage threeconsumers. Indeed, the possibility of choosing from one toseveral manufacturers allows us to obtain a more efficient andreliable IESC. In addition, considering a number of consumersranging from one to three appears a more realistic situationthan the one modeled in case study 1. The digraph describingthe new system is depicted in Fig. 12 and links with the datasheet of the performance indices associated with the digraphedges are reported in Table VI of Appendix 1. As explainedin Section III-B, the IESC model of this case study includessix fictitious nodes, i.e., nf5 , nf6 , and nf7 , associated withmanufacturers n5, n6, and n7, respectively, and nf12 , nf13 , andnf14 , associated with consumers n12, n13, and n14, respec-tively. Hence, the fictitious edges with labels x64, x65, and x66,i.e., arcs connecting nf5 to n5, nf6 to n6, and nf7 to n7, respec-tively, are considered. Analogously, the edges correspondingto variables x67, x68, and x69 are the fictitious arcs connect-ing nf12 to n12, nf13 to n13, and nf14 to n14, respectively.Moreover, the constraints for Problems 1 and 2 are definedby considering two different situations: only one manufacturerand only one consumer are selected (case study 2A) and twomanufacturers and two consumers are selected (case study 2B).

The multicriteria linear program is

z = min(f(x)) = Cx (21)

subject to

Ax ≥ B (22)

xh ∈ {0, 1} for h = 1, . . . , E (23)

with E = 69, but no performance index is assigned to eachfictitious variable xh for h = 64, . . . , 69.

The constraints (22) for case studies 2A and 2B are listed inAppendix 2. Moreover, Tables IV and V report the values ofthe performance indices obtained by minimizing cost, energy,and CO2 emission for case studies 2A and 2B, respectively.An analysis of Table IV shows the benefits of the multicriteriaoptimization approach, which enables us to choose amongseveral near-optimal solutions. In particular, we obtain solution


Fig. 12. Digraph associated with the IESC of case study 2.

TABLE IVVALUES OF THE PERFORMANCE INDICES FOR PROBLEM 2 OF CASE

STUDY 2A: OPTIMAL SOLUTIONS FOR MULTIOBJECTIVE

FUNCTION COST, ENERGY, AND CO2 EMISSION

xF in Table IV representing good trade off among cost, energy,and CO2 emission for case study 2A, and the correspondingnetwork is depicted in Fig. 13. In addition, Table V shows thatsolution xF performs satisfactorily for case study 2B in termsof cost, energy, and CO2 emission: the corresponding digraphis depicted in Fig. 14. Note that the digraph is composed of twoparallel productive chains. In this case, a remarkable advantageis observed: if a transporter is temporarily unavailable and/or acommunication way cannot be momentarily employed, the pro-ductive cycle does not stop. In addition, the considered digraphexhibits edges y15,6 and y18,3 associated to two recycling links.It is evident that such links serve the purpose of facilitatingcomponent recycling and of extracting hazardous componentsfrom products to ensure environmental safety. A consequenceof the presence of such demanufacturing connections is thereduction of the overall costs and CO2 emission performanceindices in the selected IESC configuration.

VI. CONCLUSION

This paper proposes a network design configuration strat-egy to select the stage actors of IESC, which are complexdistributed manufacturing systems incorporating the power of

TABLE VVALUES OF THE PERFORMANCE INDICES FOR PROBLEM 2 OF CASE

STUDY 2B: OPTIMAL SOLUTIONS FOR MULTIOBJECTIVE

FUNCTION COST, ENERGY, AND CO2 EMISSION

e-commerce and electronic information. The network structureof the IESC is modeled by a digraph describing the stagepartners and the material and information links connectingthem. Moreover, a procedure to define the constraints and thesingle- and multiobjective optimization problem is stated on thebasis of the digraph knowledge. The main peculiarity of the pro-posed methodology is its flexibility in building the optimizationconstraints: this characteristic improves agility, environmentalperformance, and reconfigurability in the IESC network design.


Fig. 13. Digraph representing solution xF of Problem 2 of case study 2A.

Fig. 14. Digraph representing solution xF of Problem 2 of case study 2B.

Indeed, it is possible to add or substitute constraints that canrepresent the presence or absence of a transportation mode andof an e-commerce connection. Moreover, the objective functioncan be easily selected and modified.

The proposed optimization strategy is applied to two casestudies inspired by an IESC producing desktop computersdescribed in [14]. The multicriteria optimization problem so-lution proposes different structures for the IESC on the ba-sis of the performance indices associated with the materialand information links. Moreover, to improve agility in theIESC network design, some networks are obtained by selectingfrom the corresponding sets of candidates only one or twomanufacturers and only one or two consumers. The presentedresearch clearly advances state-of-the-art in the area of in-tegrated SC design and shows great promise for industrialapplications.

APPENDIX I

See Table VI on the next page.

APPENDIX II

A. Constraints of Case Study 2A

The constraints for case study 2A may be obtained in the waysimilar to those for case study 1 (see Section III).

BOM Constraints: The BOM constraints of the second stageare defined as for case study 1: C is produced by n1 and n2, H isproduced by n1, n2, and n3, M is produced by n2, n3, and n4,and K is produced by n3 and n4. Then, the obtained constraintsimposed on the edges are

x3 + x4 − x64 ≥ 0

x3 + x4 + x5 − x64 ≥ 0

x4 + x5 + x6 − x64 ≥ 0

x5 + x6 − x64 ≥ 0

x24 + x25 − x65 ≥ 0

x24 + x25 + x26 − x65 ≥ 0

x25 + x26 + x27 − x65 ≥ 0

x26 + x27 − x65 ≥ 0

x33 + x34 − x66 ≥ 0

x33 + x34 + x35 − x66 ≥ 0

x34 + x35 + x36 − x66 ≥ 0

x35 + x36 − x66 ≥ 0. (24)

Path and Structural Constraints: It is necessary to select inthe digraph a path that starts from a node in the manufacturerstage and ends at a node of the consumer stage. For the


TABLE VIDATA SHEET FOR THE NETWORK LINKS IN CASE STUDY 2

sake of simplicity, these constraints are expressed as structuralconstraints derived from the path requirements

x9 + x29 + x40 + x16 + x14 + x17 − x67 ≥ 0

x41 + x42 + x43 + x47 + x51 + x49 − x68 ≥ 0

x44 + x45 + x46 + x48 + x52 + x50 − x69 ≥ 0

x8 + x9 + x41 + x44 + x10 + x11 − x64 = 0

x28 + x29 + x42 + x45 + x30 + x31 − x65 = 0

x12 − x8 − x28 − x38 = 0

x13 + x14 + x51 + x52 + x15 − x10 − x30 − x37 ≥ 0

x16 + x47 + x48 − x12 − x13 ≥ 0

x17 + x49 + x50 − x15 − x11 − x31 − x39 ≥ 0. (25)

In addition, the structural constraints are

x5 + x26 + x35 − x23 ≥ 0

x23 − x21 − x59 − x60 = 0

x64 − x22 ≥ 0

x65 − x32 ≥ 0

x66 − x61 ≥ 0

x67 − x18 ≥ 0

x68 − x53 ≥ 0

x69 − x54 ≥ 0

x7 − x67 ≥ 0

x68 − x62 ≥ 0

x69 − x63 ≥ 0

x48 + x16 + x47 − x1 ≥ 0

x52 + x13 + x15 + x14 + x51 − x2 ≥ 0

x67 − x19 − x20 − x21 − x18 ≥ 0

x68 − x55 − x57 − x59 − x53 ≥ 0

x69 − x56 − x58 − x60 − x54 ≥ 0. (26)

Mutual Exclusion Constraints: We impose that the stages ofrecyclers, consumers, and manufacturers are composed by onlyone actor each. Then the mutual exclusion constraints are

x1 + x2 + x7 + x62 + x63 = 1

x64 + x65 + x66 = 1

x67 + x68 + x69 = 1. (27)

B. Constraints of Case Study 2B

BOM Constraints: The BOM constraints are identical to theBOM constraints expressed by (24) of case study 2A.


Path and Structural Constraints: It is necessary to select inthe digraph a path that starts from a node of the manufacturerstage and ends at a node of the consumer stage. For thesake of simplicity, these constraints are expressed as structuralconstraints derived from the path requirements

x9 + x29 + x40 + x16 + x14 + x17 − x67 ≥ 0

x41 + x42 + x43 + x47 + x51 + x49 − x68 ≥ 0

x44 + x45 + x46 + x48 + x52 + x50 − x69 ≥ 0

x8 + x9 + x41 + x44 + x10 + x11 − x64 = 0

x28 + x29 + x42 + x45 + x30 + x31 − x65 = 0

x38 + x40 + x43 + x46 + x37 + x39 − x66 = 0

x12 − x8 − x28 − x38 = 0

x13 + x14 + x51 + x52 + x15 − x10 − x30 − x37 ≥ 0

x16 + x47 + x48 − x12 − x13 ≥ 0

x17 + x49 + x50 − x15 − x11 − x31 − x39 ≥ 0

x16 + x47 + x48 − x13 ≥ 0

x17 + x49 + x50 − x15 ≥ 0. (28)

In addition, the structural constraints are

x35 + x5 + x26 − x23 ≥ 0

x23 − x21 − x59 − x60 = 0

x64 − x22 ≥ 0

x65 − x32 ≥ 0

x66 − x61 ≥ 0

x67 − x18 ≥ 0

x68 − x53 ≥ 0

x69 − x54 ≥ 0

x22 + x32 + x61 − x18 − x53 − x54 = 0

x7 + x14 + x16 + x17 − x67 ≥ 0

x62 + x47 + x51 + x49 − x68 ≥ 0

x63 + x48 + x52 + x50 − x69 ≥ 0

x67 − x19 − x20 − x21 − x18 ≥ 0

x68 − x55 − x57 − x59 − x53 ≥ 0

x69 − x56 − x58 − x60 − x69 ≥ 0

x48 + x16 + x47 − x1 ≥ 0

x52 + x13 + x15 + x14 + x51 − x2 ≥ 0. (29)

Mutual Exclusion Constraints: We impose that the recyclerstage is composed by at most two actors

x1 + x2 + x7 + x62 + x63 = 2. (30)

Finally, to choose two manufacturers and two consumers, thefollowing constraints are respectively imposed, i.e.,

x64 + x65 + x66 = 2

x67 + x68 + x69 = 2. (31)

REFERENCES

[1] R.-H. Ballou, “Unresolved issues in supply chain network design,”Inf. Syst. Frontiers, vol. 3, no. 4, pp. 417–426, Dec. 2001.

[2] B. M. Beamon, “Measuring supply chain performance,” Int. J. Oper.Prod. Manage., vol. 19, no. 3, pp. 257–292, 1999.

[3] S. M. Disney, M. M. Naim, and A. Potter, “Assessing the impact ofe-business on supply chain dynamics,” Int. J. Prod. Econ., vol. 89, no. 2,pp. 109–118, 2004.

[4] M. Dotoli, M. P. Fanti, C. Meloni, and M. C. Zhou, “A decision supportsystem for the supply chain configuration,” in Proc. IEEE Int. Conf.Systems, Man and Cybernetics, Washington, D.C., 2003, pp. 1867–1872.

[5] ——, “A multi-level approach for network design of integrated supplychain,” Int. J. Prod. Res., vol. 43, no. 20, pp. 4267–4287, 2005.

[6] M. Ehrgott, Multicriteria Optimization. Berlin, Germany: Springer-Verlag, 2000.

[7] R. Gaonkar and N. Viswanadham, “Collaboration and information sharingin global contract manufacturing networks,” IEEE/ASME Trans. Mecha-tronics, vol. 6, no. 4, pp. 366–376, Dec. 2001.

[8] ——, “Strategic sourcing and collaborative planning in Internet-enabledsupply chain networks producing multigeneration products,” IEEE Trans.Autom. Sci. Eng., vol. 2, no. 1, pp. 54–66, Jan. 2005.

[9] G. Graham and G. Hardaker, “Supply chain management across the Inter-net,” Int. J. Phys. Distrib. Logist. Manage., vol. 30, no. 3/4, pp. 286–295,2000.

[10] R. J. Graves, A. Agrawal, and K. Haberle, “Estimating tools to supportmulti-path agility in electronics manufacturing,” IEEE Trans. Compon.,Packag., Manuf. Technol., vol. 19, no. 1, pp. 48–56, Jan. 1996.

[11] M. Grieger, “Electronic marketplaces: A literature review and call forsupply chain management research,” Eur. J. Oper. Res., vol. 144, no. 2,pp. 280–294, 2003.

[12] F. S. Hillier and G. J. Lieberman, Introduction to Operations Research.Singapore: McGraw-Hill, 2005.

[13] Y. J. Jang, S. Y. Jeng, B. M. Chang, and J. Park, “A combined model ofnetwork design and production/distribution planning for a supply net-work,” Comput. Ind. Eng., vol. 43, no. 1/2, pp. 263–281, Jul. 2002.

[14] Y. Luo, M. C. Zhou, and R. J. Caudill, “An integrated e-supplychain model for agile and environmentally conscious manufactur-ing,” IEEE/ASME Trans. Mechatronics, vol. 6, no. 4, pp. 377–386,Dec. 2001.

[15] A. M. Mangini, “Modelli e Algoritmi per la Configurazione di una CatenaLogistico-Produttiva,” M.S. thesis, Dipartimento di Elettrotecnica edEllettronica, Politecnico di Bari, Bari, Italy, 2002–2003.

[16] V. Mantou, M. Vachopou, and D. Folinas, “Virtual e-chain (VeC)model for supply chain collaboration,” Int. J. Prod. Econ., vol. 87, no. 3,pp. 241–250, 2004.

[17] S. Talluri and R. J. Baker, “A multi-phase mathematical programmingapproach for effective supply chain design,” Eur. J. Oper. Res., vol. 141,no. 3, pp. 544–558, 2002.

[18] S. Tayur, R. Ganeshan, and M. Magazine, Quantitative Models forSupply Chain Management. Norwell, MA: Kluwer, 1999.

[19] P. Timmers, “Business models for electronic markets,” EM- Int. J.Electron. Markets, vol. 8, no. 2, pp. 3–8, Jul. 1998.

[20] Matlab Release Notes For Release 13. Natick, MA: MathWorks, 2002.[21] P. Venkataraman, Applied Optimization with MATLAB Programming.

New York: Wiley, 2001.[22] C. J. Vidal and M. Goetschalckx, “Strategic production-distribution

models: A critical review with emphasis on global supply chain models,”Eur. J. Oper. Res, vol. 98, no. 1, pp. 1–8, Apr. 1997.

[23] N. Viswanadham and N. R. S. Raghavan, “Performance analysisand design of supply chains: A Petri net approach,” J. Oper. Res. Soc.,vol. 51, no. 10, pp. 1158–1169, 2000.

[24] N. Viswanadham and R. S. Gaonkar, “Partner selection and syn-chronized planning in dynamic manufacturing networks,” IEEE Trans.Robot. Autom., vol. 19, no. 1, pp. 117–130, Feb. 2003.

[25] C. A. Weber and J. R. Current, “A multiobjective approach to vendorselection,” Eur. J. Oper. Res, vol. 68, no. 2, pp. 173–184, 1993.

[26] L. A. Wolsey, Integer Programming. New York: Wiley, 1998.


[27] N. Wu, N. Mao, and Y. Qian, “An approach to partner selection in agilemanufacturing,” J. Intell. Manuf., vol. 10, no. 6, pp. 519–529, Dec. 1999.

[28] H. Yang, Z. Yu, and T. C. E. Cheng, “A strategic model for supplychain design with logical constraints: Formulation and solution,” Comput.Oper. Res., vol. 30, no. 14, pp. 2135–2155, Dec. 2003.

[29] M. C. Zhou and R. Zurawski, “Introduction to Petri nets in flexibleand agile automation,” in Petri Nets in Flexible and Agile Automation,M. C. Zhou, Ed. Norwell, MA: Kluwer, 1995, pp. 1–42.

[30] E. Zussman and M. C. Zhou, “A methodology for modeling and adaptiveplanning of disassembly processes,” IEEE Trans. Robot. Autom., vol. 15,no. 1, pp. 190–193, Feb. 1999.

[31] M. Gao, M. C. Zhou, and Y. Tang, “Intelligent decision makingin disassembly process based on fuzzy reasoning Petri nets,” IEEETrans. Syst., Man, Cybern. B, Cybern., vol. 34, no. 5, pp. 2029–2084,Oct. 2004.

Mariagrazia Dotoli (S’96–M’99) received theLaurea degree (with honors) in electronic engineer-ing in 1995 and the Ph.D. degree in electrical en-gineering in 1999, both from Politecnico di Bari,Bari, Italy.

She was a Visiting Scholar at the Paris 6 Uni-versity, France, and at the Technical University ofDenmark, Lyngby, Copenhagen. Since 2003, she hasbeen the Expert Evaluator of the European Commis-sion for the Sixth RTD Framework Programme. In1999, she was an Assistant Professor in Systems and

Control Engineering at the Department of Electrical and Electronic Engineer-ing, Politecnico di Bari. Her research interests include process modeling andcontrol via computational intelligence techniques, modeling and control ofdiscrete event systems, Petri nets, flexible production systems, and distributedmanufacturing systems.

Dr. Dotoli was the Co-Chairman of the Training and Education Committee ofERUDIT, the network of excellence for fuzzy logic and uncertainty modeling ininformation technology. She has been key node representative of the EUropeanNetwork of excellence on Intelligent Technologies (EUNITE).

Maria Pia Fanti (M’94–SM’01) received the Laureadegree in electronic engineering from the Universityof Pisa, Pisa, Italy, in 1983.

Since 1983, she has been with the Department ofElectrical and Electronic Engineering, Politecnico diBari, Bari, Italy, where she was an Assistant Profes-sor from 1990 to 1998 and is currently an AssociateProfessor in automation. In 1999, she was a VisitingResearcher at the Rensselaer Polytechnic Institute,Troy, NY. Her research interests include discreteevent systems, Petri nets, control and modeling of

automated manufacturing systems and computer integrated systems, supplychain modeling and management.

Prof. Fanti is the Associate Editor of the IEEE TRANSACTIONS ON

SYSTEMS, MAN, AND CYBERNETICS—PART A.

Carlo Meloni graduated with a degree in electronicengineering and received the Ph.D. degree in op-erations research from the Università degli Studidi Roma La Sapienza, Italy, in 1997 and 2000,respectively.

In 2000, he was a Postdoctoral Fellow in Op-erations Research at Università degli Studi RomaTre. In July 2002, he became the Assistant Pro-fessor of Systems Engineering and Optimization atPolitecnico di Bari, Bari, Italy. His major researchinterests concern combinatorial optimization, graph

theory and algorithms, planning and scheduling, and decision support systems.

MengChu Zhou (S’88–M’90–SM’93–F’03) re-ceived the B.S. degree from Nanjing University ofScience and Technology, Nanjing, China, in 1983,the M.S. degree from the Beijing Institute of Tech-nology, Beijing, China, in 1986, and the Ph.D. degreein computer and systems engineering from Rensse-laer Polytechnic Institute, Troy, NY, in 1990.

In 1990, he joined the New Jersey Institute ofTechnology (NJIT), Newark, and is currently aProfessor of Electrical and Computer Engineeringand the Director of Discrete-Event Systems Labo-

ratory. His research interests are in computer-integrated systems, Petri nets,semiconductor manufacturing, and multilifecycle engineering. He has over 200publications including six books, 80+ journal papers, and 14 book chapters.

Dr. Zhou is a life member of the Chinese Association for Scienceand Technology-USA and served as its President in 1999. He is theManaging Editor of the IEEE TRANSACTIONS ON SYSTEMS, MAN AND

CYBERNETICS—PART C, the Associate Editor of the IEEE TRANSACTIONS

ON AUTOMATION SCIENCE AND ENGINEERING, and the Editor-in-Chiefof the International Journal of Intelligent Control and Systems. He was theGeneral Co-Chair of the 2003 IEEE International Conference on System,Man and Cybernetics, Washington, DC, October 5–8, 2003 and the 2004IEEE International Conference on Networking, Sensors and Control, Taipei,March 21–23, 2004. He has led or participated in 28 research and educationprojects with total budget over $10 M funded by National Science Foundation,Department of Defense, Engineering Foundation, New Jersey Science andTechnology Commission, and industry. He was the recipient of NSF’s ResearchInitiation Award, CIM University-LEAD Award by Society of ManufacturingEngineers, Perlis Research Award by NJIT, Humboldt Research Award for USSenior Scientists, Leadership Award and Academic Achievement Award by theChinese Association for Science and Technology-USA, and Asian AmericanAchievement Award by Asian American Heritage Council of New Jersey.He was founding Chair of Discrete Event Systems Technical Committee ofIEEE Systems, Man and Cybernetics Society and is founding co-chair ofSemiconductor Factory Automation Technical Committee of IEEE Roboticsand Automation Society.

62 ieee transactions on systems, man, and …ceit.aut.ac.ir/~sa_hashemi/my teachings/ms-ceit-supply...

Documents