ESWA 8700 21 July 2013

Expert Systems with Applications xxx (2013) xxxxxx 1

No. of Pages 11, Model 5G

Contents lists available at SciVerse ScienceDirect

Expert Systems with Applications

journal homepage: www.elsevier.com/locate/eswa
5 6 3 4 7 8 9

A model of resilient supply chain network design: A two-stage programming with fuzzy shortest path
Q1

Yohanes Kristianto a,, Angappa Gunasekaran b, Petri Helo a, Yuqiuqe Hao a

a b

Department of Production, University of Vaasa, FI-65101 Vaasa, Finland Department of Decision and Information Sciences, University of Massachusetts Dartmouth, North Dartmouth, MA 02747-2300, USA

10 1 2 2 2 13 14 15 16 17 18 19 20 21

a r t i c l e

i n f o

a b s t r a c t
A supply chain network design needs to consider the future probability of reconguration due to some problems of disaster or price changes. The objective of this article is to design a recongurable supply chain network by optimizing inventory allocation and transportation routing. A two-stage programming is composed according to Benders decomposition by allocating inventory in advance and anticipating the changes of transportation routings; thus the transportation routing is stochastic in nature. In addition, the fuzzy shortest path is developed to solve the problem complexity in terms of the multi-criteria of lead time and capacity with an efcient computational method. The results and analysis indicate that the proposed two-stage programming with fuzzy shortest path surpasses the performance of shortest path problem with time windows and capacity constraint (SPPTWCC) in terms of less computational time and CPU memory consumption. Finally, management decision-making is discussed among other concluding remarks. 2013 Elsevier Ltd. All rights reserved.
23 24 25 26 27 28 29 30 31 32 33 34 35

Keywords: Benders decomposition Fuzzy set Shortest path problem Resilient supply chain Supply chain design Two-stage programming

36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59

1. Introduction Supply chain (SC) network design requires not only robustness to cope with errors during execution, but also resilience as the ability of a system to return to its original state or move to a new, more desirable state after being disturbed, which demands a combination of exibility and adaptability (Christopher & Peck, 2004). Thus, the term risk is introduced to signify the probability that the supply chain could be damaged. The risk comes from internal process, external demand, supply and control. There are abundant topics on internal process optimization and control: for instance, the introduction of radio frequency and identication (RFID) and vendor managed inventory (VMI) to increase SC visibility (Kristianto, Helo, Jiao, & Sandhu, 2012a; Lin, Chang, Hung, & Pai, 2010). On the other hand, while disruptions in the SC are realistic possibility that companies must address, there is lack of capability to capture them efciently. Therefore a recovery logic by managing time and inventory as two buffers in the SC is the key to making a supply chain resilient (Schmitt & Singh, 2012). The nature of resilient SC networks compared to other SC networks (i.e., agile and lean) presents its own set of challenges, which must be addressed effectively in order for a company to remain competitive. First, SC network design comprises several congurations of transportation routings to reduce the impact of a disrup Corresponding author.
E-mail address: ykristiantonugroho@gmail.com (Y. Kristianto). 0957-4174/$ - see front matter 2013 Elsevier Ltd. All rights reserved. http://dx.doi.org/10.1016/j.eswa.2013.07.009

tion if and when it occurs. Second, the whole scenario of SC network congurations must trade off efciency and redundancy so as to accommodate the strategic disposition of additional capacity and/or inventory at potential pinch points (Christopher & Peck, 2004). The additional surplus capacity provides more exibility than inventory investment if the resilience is taken into consideration. The additional surplus capacity enables new location delivery for a new SC member whenever a disruption occurs. In that case production capacity changes are necessary to meet a new demand level. Furthermore, there is an opportunity to improve SC velocity and acceleration by reducing inbound lead times and non-value added time. A study of dynamic facility layout problems (Dong, Wu, & Hou, 2009) provides a useful insight on using a multi-criteria decision of inventory, excess capacity allocation and transportation routings to allow potential pinch points for improving supply chain velocity (Christopher & Peck, 2004). The multi-criteria decision provides supply chain robustness by generating capabilities of (1) reducing cost and/or improving customer satisfaction under normal circumstances, and (2) sustaining SC operations during and after a major disruption (Tang, 2006). Considering the above mentioned important properties of SC resilient network design, previous contributions are available from operational to strategic level. On the strategic level, jointly capacitated facility location and transportation planning are optimized by using, for instance, a stochastic solution (Santoso, Ahmed, Goetschalckx, & Shapiro, 2005), path diversity design and ow restora-

60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86

Please cite this article in press as: Kristianto, Y., et al. A model of resilient supply chain network design: A two-stage programming with fuzzy shortest path. Expert Systems with Applications (2013), http://dx.doi.org/10.1016/j.eswa.2013.07.009

ESWA 8700 21 July 2013

2 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 Y. Kristianto et al. / Expert Systems with Applications xxx (2013) xxxxxx

No. of Pages 11, Model 5G

tion (Tomaszewski, Pioro, & Zotkiewicz, 2010). In general, these methods provide a backup plan in providing several solution scenarios if the primary networks are disrupted. On the tactical level, strategic inventory allocation under demand uncertainty (Kristianto, Gunasekaran, Helo, & Sandhu, 2012b) reduces safety stock placement (Christopher & Peck, 2004) by anticipating demand and supply uncertainties. Concurrently, multi-commodity distribution systems (Geoffrion & Graves, 1974) link the tactical and strategic planning. While the operational level planning could affect SC velocity and capacity, however, little attention paid to linking the operational to the strategic level. Among the few existing studies, for instance Kristianto, Helo, and Jiao (2012c) have designed recongurable production systems to increase the level of path diversity and ow restoration and therefore attain surplus capacity (Christopher & Peck, 2004). The solutions in the latest literature are not available for enterprise wide application so that other sources of SC network disruption, for instance non-reliable transportation lead times and suppliers are not considered in the solution. Therefore, the research gap has to be lled by designing an integrated multi-level resilient SC network for reducing the impact of a disruption by allocating additional capacity and/or inventory at potential pinch points efciently, with or without reliable delivery and supply. In considering the objective of the article, the following research questions (RQ) are raised: 1. (RQ1) How does resilient SC design reduce the impact of a disruption if and when it occurs? 2. (RQ2) How does resilient SC design trade off efciency and redundancy so as to accommodate the strategic disposition of additional capacity and/or inventory at potential pinch points? 3. (RQ3) In terms of allowing for potential pinch points, what is the key factor that inuences supply chain velocity? To address the above questions, the following sections are composed as follows: Section 2 identies a solution strategy. Section 3 focuses on the shortest path problem with time windows and capacity constraint. Section 4 introduces two-stage programming. Section 5 examines the Benders decomposition for solving twostage programming. Section 6 validates the models according to the problem example. Section 7 discusses the optimization result and provides managerial implications. Finally, Section 8 concludes the paper by giving future research directions based on research limitations. 2. Solution strategy A comprehensive study of strategic network planning and the ow of goods in the networks need to be conducted also on a space and time basis (Goetschalcks & Fleischmann, 2002). On a spatial basis, the capability to expand or shrink the number of potential pinch points is necessary to adapt the required level of capacity and or/inventory due to future potential disruptions. For example, during a situation where the supply location is relocated or the product specications are changed due to a severe disaster or government regulation changes, the SC may choose a different geographical area of supply for up to the next 3 years. But if, starting in year four, another scenario of raw material scarcity and price changes occurs, the most economic production-distribution strategy may be to manufacture the products closer to the area of supplies (Goetschalcks & Fleischmann, 2002). The expanding or shrinking of the number of potential pinch points has to be holistic, and it potentially becomes redundant without global optimization and considering the stochastic nature of resilient SC networks

(Goetschalcks & Fleischmann, 2002). Thus, a two-stage optimization of structural and master planning is then used to maximize prot, customer service and exibility at minimum risk. In this regard, this paper proposes a two-stage approach to resilient SC network design. At the rst stage, inventory allocation is scrutinized for meeting the different scenarios of excess capacity and total costs of the second stage. At the second stage, SC network design is constrained by total costs and time windows, which is formulated as a shortest path problem with time windows and capacity constraint (SPPTWCC). The Benders decomposition algorithm is applied to solve this optimal conguration problem owing to its high computational efciency (Saharidis, Minoux, & Ierapetritou, 2010). Unlike the ordinary shortest path problem, which can be solved in polynomial time in graphs without negative cycles, the SPPTWCC is NP-hard due to the nonlinear relationship between time windows (TW) and capacity constraint (CC), and, as such, is believed not to be efciently solvable. In solving the NP-hard problem, soft computing (i.e., fuzzy logic, neural networks and genetic algorithm GA) has been widely successfully implemented for optimizing SC design. Some applications, for instance manufacturing ow management, order fulllment, demand management, supplier relationship management, product development and commercialization, return management, customer service management and customer relationship management are most improvement areas of SC design. GA and fuzzy logic are the two most used tools and manufacturing ow management is the most popular application area (Ko, Tiwari, & Mehnen, 2010 and references therein). While GA most probably cannot guarantee the global optimum solution, fuzzy logic can potentially be implemented within a deterministic global optimization tool for achieving the global optimum solution. The current article embeds a fuzzy shortest path into two-stage programming in order to nd the global optimum solution at polynomial time by converting the second stage formulation from SPPTWCC to ordinary SPP. 3. Shortest path problem with time windows and capacity constraint (SPPTWCC) for inventory and excess capacity allocations The SPPTWCC is built over generic networks and routings {GNR} which are presented as a directed graph Gk = (Vk, Ak) for feasible vehicle routing with k serving a set of generic operations (GOs) Vk and a set of transportation arcs Ak. The Ak is a set of binary network ows, variable xk i;j that has a unity value if there is a pairing from GOs i to j to congure a route k, and zero for otherwise. The {GNR} has restricted routes from the lowest to the highest hierarchy of station. Each feasible route of {GNR} qk is indexed by p and is included in the set of extreme points of {GNR} structures, "p 2 qk. Each k is composed by linking one GO to another GO in such a way that the combination of some feasible routes forms a single shortest path. For each path k, arc xk i;j is congured according to a set of feasible routing "p 2 qk in such a way that i, j 2 V, "k 2 K, "p 2 qk and k k therefore the arc xk i;j is rewritten as xi;j;p : Each xi;j requires a transl portation time tij to link GOs i and j within time windows te i ; t i . Consequently, the transportation cost for path k within a set of exP P treme points p is reformulated as C k c xk . A path p p2qk i;j2Ak i;j i;j;p k parameter up 1 holds if the path p is used by {GNR} k and zero for otherwise. Denition 1. A path xij from i to j dominates another path xij from i to j if the objective vector of path xij, pij is higher than the objective vector of path xij ; pij for at least one k.

148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182

183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208

130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147

Please cite this article in press as: Kristianto, Y., et al. A model of resilient supply chain network design: A two-stage programming with fuzzy shortest path. Expert Systems with Applications (2013), http://dx.doi.org/10.1016/j.eswa.2013.07.009

ESWA 8700 21 July 2013

Y. Kristianto et al. / Expert Systems with Applications xxx (2013) xxxxxx 209 210 211

No. of Pages 11, Model 5G

Denition 2. A path xij from i to j is a pareto optimal path from i to j if there is no other objective vector of path xij ; or pij has higher value than the objective vector of path xij, pij for all k.

Indices i, j

q 2 (1, . . . , p)

Set of all predecessor and successor variables of the SC networks, Set of all SC network patterns,

Fig. 1. An example of the quickest path problem with time windows and capacity constraint.

Set denitions G = (V, A) A set of activities V and a set of location networks in the . . .. . .. . .? of arcs A that connects two modules i and j within a module set V sj The slack variable of stage j, Equal to one if the direct connection from nij p activity i to j is performed by worker in a feasible conguration of centralized SC, p, xijp Equal to one if arcs i ? j are connected and are used by path p, up The up = xijp if arcs i ? j are connected and are used by path p, Deterministic parameters tij The transportation time from i to j, Cp The transportation cost of path p, Iij The allocated inventory level for delivering products from i to j, cij The production cost incurred if activities i and j are directly connected, l Holding cost, the earliest and the latest hi ; te i ; ti departure time at location i, Arc label of transportation time, lowest, highest ; q t ij ; Q li ; Q u ij i and actual vehicle loads from i to j, aij Fuzzy membership value of (cij, tij) for connecting location i and j, h Benders cut, Stochastic parameters Sp Stochastic scenario of SC networks design p, before, during or after disruption, a = (a, l, b) The lowest, most acceptable and highest value of departure time or vehicle loads, f(a) The fuzzy membership grade Decision variables wp The path variable of the Benders master problem for p 2 {1, 2, . . . , P} hp The path variable of the Benders sub-problem for p 2 {1, 2, . . . , P}, ai, bij, c The dual multipliers of the Benders master problem
278

ations subsequently. For simultaneous pickup and delivery operations, the actual delivery capacity qij from i to j represents a vehicle load for arc i j. The quickest path problem considers the delay of transportation time t li t e i is within a certain range of time windows. The transporter must wait for a GOs to open the warehouse at te j before continuing to the next sequence. However, a transporter must leave a GOs at most tlj . Therefore, a feasible arc i j is formulated as follows:

292 293 294 295 296 297 298 299

300 302
303 304 305 306 307 308 309 310 311 312 313 314 315

tlj P tli tij

Considering pickup and delivery operations, the capacity of delivery is described in Fig. 1 as the demand of destination GO and the pickup quantity from the previous GO. While the changes in SC networks topology are not anticipatory, future optimum network design is reformulated as sub-problems with time windows and capacity constraint labeled as tij ; Q li ; Q u the i ; qij . At the rst stage (master problem), current decil sion is the inventory allocation labeled as hi ; t e i ; t i . Fig. 2 shows that prior to the transportation routings, the inventory distribution is optimized to obtain the rst stage decision of inventory placement. The inventory due to a connection between GOi and GOj is used as a parameter of the networks arc and a holding cost hij is incurred. 4. Two-stage programming (TSP) To integrate inventory allocation and transportation routings within a predetermined delivery schedule, Fig. 2 shows a twostage solution by optimizing {GNR} instantiations at the rst stage and transportation routings at the second. While the rst stage deals with safety stock allocation, the decision is xed in earliest period of planning. The second stage decisions of transportation routings are about the optimal paths of transporting the products, so as to provide a backup for SC networks to tailor their transportation routings in anticipating future disruptions. Since there is more than one state of transportation routing scenarios, the second stage decision is also stochastic in nature. In order to nd the global optimum of inventory allocation and transportation routings, Benders decomposition is an implemented solution algorithm where the master problem solution (inventory allocation) must be feasible to the sub-problem feasibility constraint (transportation routings with production scheduling). In addition to the previous contributions on TSP applications in SC networks design (i.e., Beasley & Cao, 1998; Cordeau, Laporte, & Mercier 2001; Mercier, Cordeau, & Soumis 2005; Mingozzi, Boschetti, Ricciardelli, & Bianco 1999; Stojkovic & Soumis, 2001; Yen and Birge, 2006), production Q2 scheduling (i.e., Faneyte, Spieksma, & Woeginger, 2002; Saharidis et al., 2010) and inventory allocation-vehicle routing problem (Federgruen & Zipkin, 1984), the current TSP considers simultaneously pickup and delivery transportation routings with production schedule constraints with variations in delivery lead times and batch size. The fuzzy shortest path algorithm is used to convert the SPPTWCC to the original SPP, and thus the problem is easier to solve.

316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344

279 280 281 282 283 284 285 286 287 288 289 290 291

3.1. The quickest path problem with labeling method for SPPTWCC The method of labeling is used in this paper based on the principle of Pareto optimality and dominance principle of Denitions 1 and 2. At each iteration, a set of new labels are obtained after selecting a set of outgoing arcs that link node i to j, and Denitions 1 and 2 apply for selecting a dominance label and discarding the rest of the dominated paths. When there are no more labels left to be scanned, the algorithm terminates (Park & Storch, 2002). A quickest path of Fig. 1 is used to handle the complexity of SPPTWCC for combing lead times and vehicle capacity constraints in the form of transmission time. The time windows constraint stretches the transmission time into a certain range. The capacities of outgoing and incoming arcs stand for pickup and delivery oper-

Please cite this article in press as: Kristianto, Y., et al. A model of resilient supply chain network design: A two-stage programming with fuzzy shortest path. Expert Systems with Applications (2013), http://dx.doi.org/10.1016/j.eswa.2013.07.009

ESWA 8700 21 July 2013

4 Y. Kristianto et al. / Expert Systems with Applications xxx (2013) xxxxxx

No. of Pages 11, Model 5G

Fig. 2. {GNR} and their instantiation. 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368

An illustration of two stage programming is exhibited in Fig. 3. The gure shows that the rst stage deals with the optional decisions of inventory allocation and the second stage decision of vehicle routings. The stochastic scenario of vehicle loads and transportation time is predened. The second stage has to take several inputs, for instance, time windows, delivery cost and time, delivery quantity at each location and vehicle capacity. The optimization of rst and second stage objective functions should result in nding locations of inventory placement and excess capacity. Synthesis of the rst and second stages is motivated by the requirement of optimizing not only the transportation routings efciency in reducing the impact of a disruption if and when it occurs, but also the investment decision of inventory and excess capacity. The synthesis is formulated as multi-commodity network ows (i.e., Cordeau et al., 2001; Federgruen & Zipkin, 1984; Geoffrion & Graves, 1974; Mercier et al., 2005) consisting of a set of delivery sequences dened by the producer, and the future delivery sequences or pairings are called {GNR} instantiation. Every pairing begins and ends at a specic location. These commodities are formulated as a SPPTWCC, in which one commodity (inventory and excess capacity allocations) is formulated as a master problem and another commodity (a {GNR} instance) as a sub-problem. A linking constraint of maximum allocated inventory in each location is added to connect a sub-problem to the master problem.

4.1. Optimization of inventory and excess capacity allocation (stage 1) Let Ik ip be any specic inventory location i (GOi) that is serviced by a vehicle k to the path p at a certain delivery period k. The inventory allocation at GOi demands an unit inventory cost hij for satisfying the delivery demand qij from i to j. Therefore, for each run of a pattern p, we get a number of inventory levels for each GOi which we denote as Ik ip . Consequently, the decisions in our model are the inventory path according to the pattern p in a given delivery period k. Since the inventory allocation is constrained by the amount of delivery demands, the problem is formulated as a cutting stock problem. Thus the master problem is formulated as follows:

369 370 371 372 373 374 375 376 377 378 379

X XX k k z1 min hi Iip wp
p2qk k2K i2V

380 382 383

8i 2 V ;

8k 2 K ;

8p 2 qk

1a

s:t:

XX
p2qk kK

k uk 8k 2 K ; p wp 1

8p 2 qk

1b
385 386

XX
p2qk kK

wk 8k 2 K ; p 1

8p 2 q k 8p 2 qk

1c
388 389

wk 8k 2 K ; p f0; 1g

1d

391

Fig. 3. System model for a resilient SC network design.

Please cite this article in press as: Kristianto, Y., et al. A model of resilient supply chain network design: A two-stage programming with fuzzy shortest path. Expert Systems with Applications (2013), http://dx.doi.org/10.1016/j.eswa.2013.07.009

ESWA 8700 21 July 2013

Y. Kristianto et al. / Expert Systems with Applications xxx (2013) xxxxxx

No. of Pages 11, Model 5G

5 450 451 452 453 454 455 456 457 458 459 460 461

392 394
395 396 397 398 399 400 401 402 403

X XX k k Iip wp P qij
p2qk kK i;j2V

8i ; j 2 V ;

8k 2 K ;

8p 2 q k

1e

The objective function of Eq. (1a) holds the optimality still with binary path variable wk p , and the ow on inventory path p by using a vehicle k. The number of deliveries from GOi cannot be less than the demand qij (1e). uk p 1 if an inventory location i is used by a path p and a vehicle k (1c). Since constraints (1b, 1d) are the integrality constraints, they are relaxed for the rst phase and will be re-introduced for obtaining the IP optimum. Proposition 1. The formulation (1a)(1e) possesses an optimal k solution with wk p ; 8k 2 K ; 8p 2 q binary. Proof. The wk p is a binary variable and equal to one if there is a ow which passes a path p of operation k. While the ow for one path is allowed only once (Eq. (1b)), there must be only one fractional path that minimizes the total duration time (i 1) ? i ? (i + 1), and thus the optimal path wk h p 1 exists. Since the path is capacitated (Eq. (1e)), a column generation is then implemented to solve stage 1. The rationale of implementing column generation is multifold. First, though the existing vehicle capacity constraints, column generation lends itself as a exible tool for solving complex {GNR} structures for individual instantiations of standard product and process platforms, as well as a wide variety of rules that dene the feasibility of a route. Second, the absence of integrality properties helps to cancel the positive gap between relaxed and restricted solutions (Desrosiers, Soumis, & Desrochers, 1984). Third, efcient algorithms exist at least for some important variants of the SPPRC. 4.2. Optimization of transportation routings (stage 2) Each of the transportation routings at stage 2 is constrained by the number of inventory allocation patterns of the master problem and sequentially generated subject to the master problem feasible inventory allocation path constraint (Eq. (3d)). Since one feasible path of the master problem has the opportunity to be satised more than once (Eq. (1e)), each path of the feasible transportation routings has a certain level of delivery capacity and the problem is formulated as a shortest path with capacity constraint (Eq. (2e)) within a time window. The formulation of sub-problems is presented as follows:

The objective function of Eq. (2a) holds the optimality still with the binary path variable of transportation routings hk p . Constraint (2e) adds a vehicle capacity constraint to the solution of Eq. (2a). Consequently, the transportation time for path k within a set of extreme P P k points p is reformulated as T k c t . uk p xi;j;p if paths p p2qk i;j2Ak i;j ij;p i ? j are connected and are used by path p and transportation routing k. Since constraints (2b, 2f) are the integrality constraints, for the sake of avoiding the greedy solution, they are relaxed for the rst phase and will be re-introduced for obtaining the IP solution. Eq. (2d) links the master and sub-problem in such a way that the allocated inventory satises the required demands within path p, from i to j, nij p. 5. Benders decomposition for two stage programming of a resilient SC network design Benders decomposition solves the two-stage programming by considering that each scenario is independent and therefore the problem structure can be decomposed into a number of smaller problem structures to speed up the solution time. The second stage is set to zero at the rst iteration and will be updated each time the new Benders cut constraint is introduced to the sub-problems. The decisions of sub-problems are linked to the master problem through an approximation of the second stage cost h as a function of decision variables of the master problem since the delivery time of the second stage cannot exceed the delivery time of the rst stage. However, since we have not dealt with the second stage yet, we have no clue about the value of h, and therefore we optimistically set its initial value to 1 for the rst iteration. 5.1. Benders second-stage A resilient supply chain considers variations of vehicle loads and transportation times as two stochastic parameters. It is assumed that the variation sources are only in the supply side and the demand side is deterministic (master problem). Since the variations of vehicle loads and transportation times are normally distributed, the transformation of the stochastic parameters into their deterministic equivalent on each arc simplies the solution shortest path problem with time windows and capacity constraint (SPPTWCC) by reducing the networks into a shortest path problem (SPP). Considering the stochastic nature of parameter tij and qij, the fuzzy shortest path is employed to do the analysis. The appeal of fuzzy shortest path into the solution of SPPTWCC is its capability to nd an optimum solution from a large number scenario of parameter tij and qij. The scenario of each arc is represented as a triangular fuzzy number which directly clusters the parameters variations into the lowest expectation, most acceptable and highest expectation. 5.2. Fuzzy shortest path for the primal of second stage Benders The fuzzy parameters are the deterministic equivalent of the stochastic parameters. Fuzzy shortest path problem (FSPP) is discussed extensively in the literature (i.e., Chanas & Kamburowski, 1981; Chanas & Zielinski, 2001) by focusing on the degree of possibility that a certain path is chosen as a shortest path and the solution follows a dynamic programming with recursion (Klein, 1991). However, none of the past literature considers the resource constraint and time windows in the discussion of fuzzy shortest path. Thus, the problem appears as a multi-criteria decision making in routings (Niaraki & Kim, 2009). Multi-criteria shortest path covers the rest of the constraints, which are not studies using an ordinary

462 463 464 465 466 467 468 469 470 471 472 473 474 475 476

404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431

477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494

432 434 435

z2 min

XX k k T p hp
p2qk kK

8k 2 K ;

8p 2 qk Dual multipliers 8p 2 qk a al jl 2 L

2a

s:t:
437 438 440 441

XX k uk 8k 2 K ; p hp 1
p2qk kK

2b

XX k hp 1 8k 2 K ;
p2qk kK

8p 2 q k b 6 0

2c

495 496 497 498 499 500 501 502 503 504 505 506

X XX X XX k k k k nij nij p qp hp 6 p I ip wp
p2qk k2K i;j2V p2qk k2K i;j2V

8i; j 2 A;


8k 2 K ;
2d 2e

443 444 446 447 449

8p 2 qk cij P 0li ; lj 2 Ak
XX k qk p hp 6 Q
p2qk kK

8k 2 K ;

8p 2 qk l 6 0

hk 8k 2 K ; p f0; 1g

8p 2 qk

2f

Please cite this article in press as: Kristianto, Y., et al. A model of resilient supply chain network design: A two-stage programming with fuzzy shortest path. Expert Systems with Applications (2013), http://dx.doi.org/10.1016/j.eswa.2013.07.009

ESWA 8700 21 July 2013

6 507 508 509 510 511 512 513 514 515 Y. Kristianto et al. / Expert Systems with Applications xxx (2013) xxxxxx

No. of Pages 11, Model 5G

shortest path. When disregarding a constraint, an acceptable outcome would be unlikely (Park & Storch, 2002). While considering the latest issue, a fuzzy shortest path is used s possibility index as a constraint. The by utilizing Dubois and Prade membership grade is obtained from the time windows of each arc as follows: Denition 3. A triangular fuzzy number is represented by a = (a, l, b) with a membership function f(a) = (a, l, b), dened by an expression:

min

X
l2 L

al b

XX k k nij p Iip wp cij

p2qk kK

8i; j 2 V ;

8k 2 K ;

8p
8a
564 565

2 qk s:t: uk p al b X k k ij nij p qp cij P fp pp

ij2A

8i; j 2 V ;

8k 2 K ;

8p 2 q k
8b
567 568

516

cij P 0
b60

518
519 520 521 522 523 524 525 526 527 528 529 530 531 532 533 534

8 0 > > > > > > ala > > > a > > < f a 1 > > > > lba > > > > b > > > : 0

k cij P 0  li ; lj 2 A

8c 8d

570 571 573

574 575 576 577 578 579

if if if if if

a6la

la<a<a
al 5

l<a<lb
aPlb

Notice that only the objective function (8a) contains inventory allocation information. This implies that the feasible region of the dual second-stage is independent of the inventory allocation. Introducing the additional free variable z0, the MIP relaxation of model (1)(8) can thus be reformulated as the following Benders master problem:

535 537
538 539

e e l e where a, a, l, b are t li t ij ; t e j t i ; t j ; t j t j respectively. Each scenario of vehicle routings results in a different value of f(a) due to the actual arrival time being different. Using the quickest path of Fig. 2, each load of a vehicle for each route from a source node i to a destination node j is pre-calculated in advance. Any route which has loads over a one vehicle capacity is unfeasible and can be omitted from the calculation. Since the transporter would like to minimize the capacity loss, any feasible route which has lower capacity utilization is also not recommended. Therefore, the decision is formulated as a fuzzy number a, a, l, b standing for qij, Ql, Q, Qu, the actual load from source node i to destination node j, lower allowance of capacity loss, normal capacity and upper allowance of overloaded capacity. The nal fuzzy number of the SPPTWCC is the arithmetical summation of the time windows and capacity constraint fuzzy numbers, as follows:

z3 max z0

X XX
p2qk k2K i2V

580 582 583

k hi Ik ip wp

8i 2 V ;

8k 2 K ;

8p 2 qk

9a

s:t: z0

XX k k X nij al b 8i; j 2 V ; p Iip wp cij 6

p2qk kK l2 L

8k
9b 9c
588 589 585 586

2 K; XX
p2qk kK k k nij p I ip wp cij P

8p 2 qk

X
l2 L

al b 8i; j 2 V ; 8k 2 K ; 8p 2 qk
8p 2 qk 8p 2 q k 8p 2 qk

XX
p2qk kK

1 8k 2 K ;

9d
591 592

XX
p2qk kK

wk 8k 2 K ; p 1

9e
594 595

k TW fp fa ^ faCC

wk 8k 2 K ; p f0; 1g XX
p2qk kK k k uk p wp P qij wp

9f

597 598

541 543 544 546 547 549 550

540

The existence of two fuzzy numbers (time and capacity) avoids any negative cycle in the networks, and thus the SPPTWCC can be approached as a SPP and Eq. (2) is reformulated as follows:

8i; j 2 V ;

8k 2 K ;

8p 2 q k

9g
600
601 602 603 604 605 606 607 608 609 610 611

XX k k z2 max fp hp
p2qk k2K

8k 2 K ;

8p 2 qk Dual multipliers
k

7a

XX k s:t: uk 8k 2 K ; p hp 1
p2qk kK

8p 2 q
k

a al jl 2 L

7b

XX k hp 1 8k 2 K ;
p2qk kK

8p 2 q

b0

7c

X XX X XX k k k k nij nij p qp hp 6 p I ip wp
p2qk k 2 K i; j 2 V p2qk k2K i;j2V

8i; j 2 A;
 7d 7e

Eqs. (9b) and (9c) are the optimality and feasibility constraints to constraint the value of wk p by its dual and to minimize the duality gap and therefore the solution to be optimum (Eq. (9b)). The relaxed master problem is iterated together with the second-stage Benders to set up constraints (2d) in the second-stage Benders. If the primal second-stage Benders is feasible, the dual variables of the primal determine an extreme point of the primal, otherwise a Benders cut is added after identifying an extreme ray that violates one of the constraints (9c). The iterations are continued until its optimal solution yields a feasible primal second-stage Benders whose cost is equal to the value of z0. 6. Computational study Considering the running example of HP printer supply chains (Fig. 4), there are 11 locations to manufacture the desk-jet printer. The case was written by Kopczak and Lee (2004). The case was chosen since the HP deskjet printer SC is concerned with zero inventory and has three major problems of long delivery lead times, internal process and demand variations. At that time, HP had to solve the following problems: 1. Should they concentrate on warehousing or integration (transportation)?

552 553 555

556 557 558 559 560

8k 2 K ;
hk 8k 2 K ; p f0; 1g

   8p 2 qk cij P 0li ; lj 2 Ak

612 613 614 615 616 617 618 619 620 621

8p 2 qk

The objective function of Eq. (7a) holds the optimality still with binary path variable of transportation routings hk p ; the prot of having a fuzzy number at a certain level. 5.3. The dual of Benders second-stage and Benders master problem The dual of the Benders second-stage is used to generate a Benders cut as per the following:

562

561

Please cite this article in press as: Kristianto, Y., et al. A model of resilient supply chain network design: A two-stage programming with fuzzy shortest path. Expert Systems with Applications (2013), http://dx.doi.org/10.1016/j.eswa.2013.07.009

ESWA 8700 21 July 2013

Y. Kristianto et al. / Expert Systems with Applications xxx (2013) xxxxxx

No. of Pages 11, Model 5G

Fig. 4. Printer supply chains network design (modied from Kopczak & Lee, 2004).

622 623 624 625 626 627 628 629 630 631 632 633 634 635 636 637 638 639 640 641 642 643 644 645 646 647 648 649 650 651 652 653 654 655 656 657 658

2. At which location does HP need to allocate the inventory? From these two problems, HP considered the two areas of inventory management and logistics critical to their operations. While HP concentrated on nding the most accurate forecast method, the warehousing and integration strategies did not consider the optimization of their network design to handle both the sectors of warehousing and integration. Therefore, the focus of their study was the minimization of inventory level by choosing quicker transportation means. The current article contributes to the HP case study by showing that reconguring the SC network design could potentially reduce the level of inventory without adding excessive investment in transportation means. Furthermore, the extension can help HP to mitigate the risk of future SC disruptions. The importance of a recongurable SC network comes from the case of a Phillips facility in Albuquerque, New Mexico, that went up in ames. The plant made radio frequency (RF) chips, key components for mobile telephones for both Scandinavian companies. The re damaged the plant, but Nokias manager quickly carried out design changes and contacted back-up sources from two suppliers in Japan and the United States by making a modular product platform. Conversely, Ericsson was unable to hedge against this emergency because it did not prepare for a dual sourcing strategy, resulting in lost sales (Lee, 2004). In analyzing the SC resilience for the HP case, the original network comprising suppliers, manufacturing sites, distribution centers (DCs), dealers and customers for the deskjet product is studied and we generate a scenario of having future situations where one or more manufacturing sites cannot meet the expected transportation time and delivery quantity. The manufacturing sites have two locations for printer circuit assembly and test (PCAT) for components such as application specic integrated circuit (ASIC), read only memory (ROM) and raw printed circuit board (PCB), and nal assembly test (FAT) where the components are outsourced to worldwide suppliers. We modify the HP case by enriching the task of IC manufacturer by adding

slack capacity of PCAT if in future disruptions the original location of PCAT fails to deliver the components. Similarly, suppliers 2 and 3 could provide surplus capacity in providing components for IC manufacturer, PCAT and print mechanism. Note that suppliers 2 and 3 might comprise several providers from the same geographical area and therefore have different sizes of costs and lead times. The overall computations are delivered by implementing LINGO 13.0 software into Intel Centrino 2 processor. Different to the original problem of HP deskjet printer SCs, the transportation times and vehicle loads are assumed to be stochastic to cover the demand variations of the three DCs with a range of 2730%. Even the departure time is allowed to be delayed within the time windows; however, a longer departure time at a location is not advisable. The reason is that a longer delay of the upstream will be propagated to the downstream, nally reducing the ability of the SC to absorb the demand shock. The SC robustness can be observed by benchmarking the total lead times and operation costs among two methods, SPPTWCC and FSPP. The shorter lead times and lower operation costs provide a higher level of SC robustness. Each circle in Fig. 4 stands for one location and the directed arc signies the allowable sequence from upstream to downstream locations. A combination of location and arc is presented as a graph G = (V, A) for each node "i 2 V corresponding to the location i. In addition, arc "i,j 2 A connects two locations i and j within an arc A and therefore the total costs of transportation, inventory and transportation cij are incurred that require tij, an operation lead time of connecting two locations i and j. There are 36 possible routings from starting to selling point that exploit supply chain resilience. Each arc that connects locations signies the feasible connection between two locations, and thus the connections store predened information of transportation time and vehicle capacity (lower, upper and actual loads), tij ; Q li ; Q u i ; qij . Each node stores information about holding cost and l time windows, hi ; te i ; t i . The networks show that the maximum vehicle load is 25 units in connecting node 7 to node 9. Thus, vehicle loads of 25 units are the required vehicle capacity. The underutilization of vehicle capacity must be avoided. In addition to vehi-

659 660 661 662 663 664 665 666 667 668 669 670 671 672 673 674 675 676 677 678 679 680 681 682 683 684 685 686 687 688 689 690 691 692 693 694 695

Please cite this article in press as: Kristianto, Y., et al. A model of resilient supply chain network design: A two-stage programming with fuzzy shortest path. Expert Systems with Applications (2013), http://dx.doi.org/10.1016/j.eswa.2013.07.009

ESWA 8700 21 July 2013

8 Table 1 Degree of possibility for each arc in Fig. 2. arc tj + tij Y. Kristianto et al. / Expert Systems with Applications xxx (2013) xxxxxx

No. of Pages 11, Model 5G

a
e te j ti

l
te j 7 7 8 7 8 8 8 8 8 8 7 7 7 11 7 11 11 11 11 11 11

b
e ti j tj

TW fa

qij

a
QI 3 7 9 3 12 8 12 2 15 14 9 8 9 11 7 8 9 13 4 5 4

l
Q 4 12 10 5 14 9 18 3 20 16 10 11 12 24 30 15 20 17 8 10 7

b Qu 5 14 31 25 15 11 20 4 21 16 12 12 14 25 32 17 21 19 9 11 8

CC fa

TW CC fa ^ fa

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21

2 2 2 18 14 17 18 35 21 19 28 48 66 68 80 72 72 71 100 100 100

7 7 8 1 1 1 1 1 0 0 1 1 1 1 4 4 4 2 0 0 0

7 7 8 7 10 10 29 48 29 48 54 54 54 67 54 80 80 80 113 113 113

0.29 0.29 0.25 0.00 0.40 0.10 0.66 0.44 0.55 0.77 0.61 0.24 0.00 0.15 0.00 0.24 0.24 0.25 0.21 0.21 0.21

4 12 22 16 13 9 14 3 16 16 11 11 12 25 23 15 19 16 6 7 7

1.00 1.00 2.33 4.67 0.92 1.00 0.67 1.00 0.73 1.00 1.11 1.00 1.00 1.09 0.00 1.00 0.89 0.92 0.50 0.40 1.00

0.29 0.29 0.25 0.00 0.40 0.10 0.66 0.44 0.55 0.77 0.61 0.24 0.00 0.15 0.00 0.24 0.24 0.25 0.21 0.21 0.21

696 697 698 699 700 701 702 703 704 705 706 707 708 709 710 711 712 713 714 715 716 717 718 719 720 721 722 723 724 725 726 727 728 729 730 731 732 733 734 735 736 737

cle capacity, some of those routings consume too much time. Therefore, the sub-problem is categorized as a multi-objective network ow and thus an NP-Hard problem. In responding to RQs 1 and 2, the following calculation results of the degree of possibility for each transportation routings can be summarized as follows: The higher degree of possibility implies that the transportation route has a higher possibility to deliver the order on time. Therefore the need for placing safety stock at the potential pinch point (destination node) is at a minimum. On the contrary, a higher level of safety stock is required by a destination node which is supplied by a transportation arc having a lower degree of possibility. The same meaning is employed with the transportation batch size, where the higher degree of possibility implies a lower chance of making the delivery overloaded or under-utilized. Thus, the disruption due to transportation could be avoided. A good choice of transportation path enables the transporter not to lose capacity and at the same time minimize the chance of lacking the required number of vehicles. The combination of transportation and vehicle load possibility degree results in the ranking for the intended arc being chosen by the FSPP. A validation of the proposed FSPP is done by benchmarking the FSPP against non-stochastic shortest path problem with time windows and capacity constraint (SPPTWCC). The non-stochastic SPPTWCC considers only one scenario of demands. The same levels l of cij ; te i ; t i are employed in the two instances for rening meaningful results. The SPPTWCC and FSPP solve a situation where the inventory allocation and the transportation routing are optimized as two-stage programming and the departure time at each location is restricted. An instance of problems 1 and 2 is found in an integrated inventory allocation and vehicle routings (Federgruen & Zipkin, 1984; Geoffrion & Graves, 1974), where a set of inventory locations supply a set of destinations according to minimum transportation cost of sub-problems. Cutting stock problem is used to solve the Benders master problem (inventory allocation) and FSPP nds the solution of transportation routings by following the procedure of Section 4. Two suppliers can have a link whenever they Q3 have similarities of, for instance, process technology (see Table 1). Table 2 shows that the SPPTWCC requires a higher level of inventory costs than FSPP. The value can be validated by considering that SPPTWCC focuses on the highest demand to achieve the solution feasibility (i.e., if the highest demand is met, the lower le-

Table 2 Characteristics and solutions of the three problems. SPPTWCC Number of inventory at all nodes Number of connection arcs Number of node/location with excess capacity Number of Benders cut Total lead times Number of transporters Number of stocking points Number of variables Number of constraints Number of iterations CPU time (s) Memory used 250 21 5 30 52 7 15 290 36 3 1 117 Kbyte FSPP 115 21 7 48 40 6 10 836 104 6 2 278 KByte

vel of demand will be met). However, this is not an efcient solution. Therefore, it requires more vehicles than FSPP (7 vehicles) to transport components from upstream to downstream, and more inventory level to satisfy the demands. FSPP, on the other hand, requires 6 vehicles. The FSPP has a robust solution in a more efcient way than SPPTWCC. Table 2 shows that FSPP surpasses SPPTWCC in terms of the number of excess capacity locations. An excess capacity location delivers products to the next location based on the demand scenario. Fig. 5 shows that routes 36 and 37 are categorized as excess capacity locations. Conversely, a stockholding point always delivers product to the next location regardless of the demand scenario. Therefore, a higher number of excess capacity locations increases the SC efciency without ignoring the demand changes. A higher number of pinch points improves supply chain velocity (Christopher & Peck, 2004) and sustains SC operations during and after a major disruption (Tang, 2006) by expanding or shrinking the number of potential pinch points to adapt the required level of capacity and or/inventory due to future potential disruptions (Goetschalcks & Fleischmann, 2002). Thus, FSPP can discriminate between excess capacity location and stockholding points for generating an agile SC. In terms of algorithm efciency, the proposed FSPP has at least two feasible benets. First, the computational time per iteration is more efcient than SPPTWCC. The high efciency can be viewed on rows 811 of Table 3, where FSPP has twice the number of itera- Q4

738 739 740 741 742 743 744 745 746 747 748 749 750 751 752 753 754 755 756 757 758 759 760 761 762 763

Please cite this article in press as: Kristianto, Y., et al. A model of resilient supply chain network design: A two-stage programming with fuzzy shortest path. Expert Systems with Applications (2013), http://dx.doi.org/10.1016/j.eswa.2013.07.009

ESWA 8700 21 July 2013

Y. Kristianto et al. / Expert Systems with Applications xxx (2013) xxxxxx

No. of Pages 11, Model 5G

Fig. 5. The inventory allocation and transportation routings of the three scenarios.

764 765 766

tions than SPPTWCC for a much higher number of variables and constraints than SPPTWCC. The results imply that the memory used is lower because the algorithm per iteration is easier to solve

than SPPTWCC. The CPU time and memory consumption of the current model look very small due to the problem size. The higher amount of memory used by FSPP than SPPTWCC comes from the

767 768 769

Please cite this article in press as: Kristianto, Y., et al. A model of resilient supply chain network design: A two-stage programming with fuzzy shortest path. Expert Systems with Applications (2013), http://dx.doi.org/10.1016/j.eswa.2013.07.009

ESWA 8700 21 July 2013

10 770 771 772 773 774 775 776 777 778 779 780 781 782 783 784 785 786 787 788 789 790 791 792 793 794 795 796 797 798 799 800 801 802 803 804 805 806 807 808 809 810 811 812 813 814 815 816 817 Y. Kristianto et al. / Expert Systems with Applications xxx (2013) xxxxxx

No. of Pages 11, Model 5G

implementation of two-stage stochastic programming. Considering the benet of FSPP to solve the stochastic problems, the higher consumption of memory is reasonable. In more complex SC networks, the number of variables involved adds not only computational time but also the size of computer memory. The future solution has to able to reduce the number of variables in order to reduce the number of constraints. If the number of constraints is reduced and the problem formulation simplied, then the optimal and feasible solution can be achieved in a reasonable time. The FSPP surpasses the other two of VRPTW and SPPRC in terms of its capability to reduce the number of constraints and variables, and therefore it has a simpler form of optimization model. It is summarized in Table 3 that FSPP has the highest level of supply chain robustness by giving the lowest cost and lead times. From the inventory allocation viewpoint, the three scenarios reveal quiet similar results when the number of vehicles is six to serve pickup and delivery routings of (12; 14681213; 1 35811; 36; 479; 37). The black arcs signify that the inventory allocations on these routings are unfeasible and thus another option of excess production capacity is allowed to cover the demands. The solution results show that the number of location (nodes) which are assigned as stockholding points is less than the number of transportation routings. This implies that one stockholding point can act as a depot and serves more than one vehicle and route. Considering the SC resilience, the situation prevents a stockholding point getting into a chaotic situation whenever one of those destinations is wildly uctuated, and therefore it increases path diversity and enables ow restoration by providing a backup plan of giving several solution scenarios if the primary network is disrupted (Tomaszewski et al., 2010). Conversely, non-stockholding points are unfeasible for inventory allocation due to demand uctuations and excessive inventory level, and thus an excess capacity is more efcient than providing inventory to cover the variations. Even though the problem example is small, the solution algorithm is scalable to a considerable large number of nodes (facility location) and vertices (transportation arcs). The reason is that the rst and the second stage are optimized by implementing column generation. Column generation has been widely implemented to optimize SPPRC with considerably large size of pilot and ight scheduling of airlines (i.e., Cordeau et al. 2001; Mercier et al. 2005; Stojkovic & Soumis, 2001; Yen and Birge, 2006). Furthermore, the developed fuzzy shortest path algorithm sub-problem transforms the SPPTWCC (which is NP hard) into an ordinary SPP. As a consequence of the transformation, the solution time is at polynomial time. Therefore, an extension of the problem size would not be a problem to the proposed algorithm. 7. Discussion and management decision-making This article proposes SC network design from the supply chain resilience viewpoint. The proposed model has a number of additions and extensions to the current system which are of both theoretical and practical importance. This article proposes SC network design solution from the SC resilience viewpoint. The proposed model was used to answer the research questions as follows: 1. How does resilient SC design reduce the impact of a disruption if and when it occurs? SC design considers several scenarios of future disruptions and therefore the size of SC networks can be shrunk or expanded to adapt the required level of capacity and or/ inventory due to future potential disruptions (Goetschalcks

818 819 820 821 822 823 824 825 826 827 828 829 830 831

& Fleischmann, 2002). The adaptation is, however, limited by constraining the operations schedule and transportation capacity by still considering the prior decision of inventory allocations. 2. How does resilient SC design trade off efciency and redundancy so as to accommodate the strategic disposition of additional capacity and/or inventory at potential pinch points? The capability of hedging a future scenario of supply and demand disruption is the main objective of SC resilience subject to time windows and transportation capacity constraints. Non-scheduled deliveries from those excess capacity locations are used to buffer the future possibility of SC disruption. Conversely, scheduled deliveries from those stockholding points are used to buffer the normal level of demands. Therefore, lower levels of production costs are necessary for non-scheduled deliveries and lower levels of holding costs are necessary for scheduled deliveries. If an arc is neither scheduled nor non-scheduled delivery, outsourcing is an attractive solution. 1. In terms of allowing potential pinch points, what is the key factor that inuences supply chain velocity? The output of Table 3 summarized that supply chain velocity does not depend at all on the number of vehicles (row 8). In fact, it depends on the number of stockholding points (row 9) and excess capacity locations. A higher number of excess capacity locations increases SC velocity and generates SC agility. An agile SC requires less inventory and compensates the required delivery speed by adding more production rates. Therefore, it does not generate an excessive amount of operation cost at a higher number of stock holding points. The key factors for determining the inuence on the SC velocity are the structure of the SC network and the number of stockholding points (Geoffrion & Graves, 1974). In order to generalize the ndings toward different contexts and time periods, the proposed SPPTWCC is potentially used for solving ow restoration and path diversity design problems (Tomaszewski et al., 2010). Thus, the applications area could be transport telecommunication networks for digital voice and data transfer. In addition to the original problem of transport telecommunication networks, variations of data transfer capacity and physical distribution can be managed simultaneously. For instance, it is predicted that Wal-Mart will generate over 7 terabytes of operational RFID data a day. Wall Mart Traditional technology architectures are not prepared to handle this volume. Supply chain traceability, auditing, and real-time tracking will drive the need to store much of this operational data, yet databases designed to store traditional transactional data will never be able to handle the load. Since the data is precious, losing the data transfer capability will not only disrupt business communication networks. Indeed, the situation will cause the disruption of physical distribution networks since the delivery depends on RFID technology. The data communication network is even more uid and susceptible to future disruptions. Therefore the solution time for disruptions becomes the most important issue. For instance, the disruptions of ight scheduling need a quicker resolution time or otherwise there will be a lot of ight delay and re-scheduling of the ights, which are cost and time consuming. In conclusion, the application areas of the proposed two-stage programming with fuzzy shortest path span from manufacturing to service industries with a considerable complex SC network. Time and exibility are components of competitive strategies beside quality and cost in todays business competition. SC decision makers have to consider the adverse effects of disruptions to SC competitiveness or otherwise the business operations could break down sometimes and nally result in losing business opportunities and customer loyalty.

832 833 834 835 836 837 838 839 840 841 842 843 844 845 846 847 848 849 850 851 852 853 854 855 856 857 858 859 860 861 862 863 864 865 867 866 868 869 870 871 872 873 874 875 876 877 878 879 880 881 882 883 884 885 886 887 888 889 890 891 892 893 894 895 896 897 898 899 900

Please cite this article in press as: Kristianto, Y., et al. A model of resilient supply chain network design: A two-stage programming with fuzzy shortest path. Expert Systems with Applications (2013), http://dx.doi.org/10.1016/j.eswa.2013.07.009

ESWA 8700 21 July 2013

Y. Kristianto et al. / Expert Systems with Applications xxx (2013) xxxxxx 901 902 903 904 905 906 907 908 909 910 911 912 913 914 915 916 917 918 919 920 921 922 923 924 925 926 927 928 929 930 931 932

No. of Pages 11, Model 5G

11 949 950 951 952 953 954 955 956 957 958 959 960 961 962 963 964 965 966 967 968 969 970 971 972 973 974 975 976 977 978 979 980 981 982 983 984 985 986 987 988 989 990 991 992 993 994 995 996 997 998 999 1000 1001 1002 1003 1004 1005 1006 1007 1008 1009 1010 1011 1012 1013 1014 1015

8. Concluding remarks An integrated inventory allocation and transportation planning is represented as a two stage stochastic programming where the inventory allocation acts as the rst stage decision and transportation planning as the two stage decision at different sets of future stochastic scenarios. The model takes into account both the lead time and capacity as two nonlinear parameters which need to be simultaneously considered during the optimization process. An example that incorporates a modied HP desk-jet printer supply chain is used with a variety of locations worldwide. This is an initial step toward the goal of developing a global supply chain strategy. For both academics and practitioners, the proposed algorithm gives at least two benets. First, the fuzzy shortest path could simplify and transform the NP hard SPPRC to an ordinary SPP. The implication of the transformation is that the solution time is polynomial and there are is lot of cheap and easily operated software for the later SPP problem. Subsequently, easy to operate software is benecial for a wide spectrum of computing skills, and therefore practitioners can benet from it. More importantly, shorter computational time is now the main criterion for practitioners to choose and use the algorithm. Therefore, the proposed two-stage programming with fuzzy shortest path has good potential to be sustained in the face of competition. However, the proposed model has some drawbacks. First, the FSPP implements a fuzzy inference system without optimizing the triangular fuzzy number and therefore the optimality fuzzy system response is still questionable. As a consequence, the second drawback is that there is no sensitivity analysis of the chosen fuzzy inference systems. Therefore, future work needs to be focused on the fuzzy set tuning to give a better response in terms of the value of time windows constraint. 9. Uncited references Q5 Chen et al. (2010) and Stadtler et al. (2010). References
Beasley, J. E., & Cao, B. (1998). A dynamic programming based algorithm for the crew scheduling problem. Computers and Operations Research, 25, 567582. Chanas, S., & Kamburowski, J. (1981). The use of fuzzy variables in PERT. Fuzzy Sets and Systems, 5, 1119. Chanas, S., & Zielinski, P. (2001). Critical path analysis in the network with fuzzy activity times. Fuzzy Sets and Systems, 122, 195204. Chen, A., Kim, J., Lee, S., & Kim, Y. (2010). Stochastic multi-objective models for network design problem. Expert Systems with Applications, 37, 16081619. Christopher, M., & Peck, H. (2004). Building the resilient supply chain. International Journal of Logistics Management, 15(2), 113. Cordeau, J. F., Laporte, G., & Mercier, A. (2001). A unied tabu search heuristic for vehicle routing problems with time windows. Journal of the Operational Research Society, 52, 928936.

933 934

935 936 937 938 939 940 941 942 943 944 945 946 947 948

Desrosiers, J., Soumis, F., & Desrochers, M. (1984). Routing with time windows by column generation. Networks, 14, 545565. Dong, M., Wu, C., & Hou, F. (2009). Shortest path based simulated annealing algorithm for dynamic facility layout problem under dynamic business environment. Expert Systems with Applications, 36, 1122111232. Faneyte, D. B. C., Spieksma, F. C. R., & Woeginger, G. J. (2002). A branch-and-price algorithm for a hierarchical crew scheduling problem. Naval Research Logistics (NRL), 49(8), 743759. Federgruen, A., & Zipkin, P. (1984). A combined vehicle routing and inventory allocation problem. Operations Research, 32(5), 10191037. Geoffrion, A. M., & Graves, G. W. (1974). Multi-commodity distribution system design by Bender decomposition. Operations Research, 20(5), 822844. Goetschalcks, M., & Fleischmann, B. (2002). Strategic networks design. In Stadtler, H., & Kilger, C. (Eds.), Supply chain management and advanced planning (pp. 177 133). Klein, C. (1991). Fuzzy shortest path. Fuzzy Sets and Systems, 39, 2741. Ko, M., Tiwari, A., & Mehnen, J. (2010). A review of soft computing applications in supply chain management. Applied Soft Computing, 10, 661674. Kopczak, L., & Lee, H. L. (2004). Hewlett-Packard supply chain: Deskjet printer supply chain (A), Source: <http://www.scilab.nl/Scilab/Downloads_les/ HP%20deskjet%20supply%20chain.pdf>. Accessed: 2 June 2012. Kristianto, Y., Gunasekaran, A., Helo, P., & Sandhu, M. (2012b). A decision support system for integrating manufacturing and product design into the reconguration of the supply chain networks. Decision Support Systems, 52(4), 790801. Kristianto, Y., Helo, P., & Jiao, J. R. (2012c). Mass customization design of engineerto-order products using Benders decomposition and bi-level stochastic programming. Journal of Intelligent Manufacturing. http://dx.doi.org/10.1007/ s10845-012-0692-z, 1-15. Kristianto, Y., Helo, P., Jiao, J. R., & Sandhu, M. (2012a). Adaptive fuzzy vendor managed inventory control for mitigating the bullwhip effect in supply chains. European Journal of Operational Research, 216(2), 346355. Lee, H. L. (2004). The triple a supply chain. Harvard Business Review, 10, 112. Lin, K. P., Chang, P. T., Hung, K. C., & Pai, P. F. (2010). A simulation of vendor managed inventory dynamics using fuzzy arithmetic operations with genetic algorithms. Expert Systems with Applications, 37, 25712579. Mercier, A., Cordeau, J. F., & Soumis, F. (2005). A computational study of Benders decomposition for the integrated aircraft routing and crew scheduling problem. Computers and Operations Research, 32, 14511476. Mingozzi, A., Boschetti, M. A., Ricciardelli, S., & Bianco, L. (1999). A set partitioning approach to the crew scheduling problem. Operations Research, 47, 873888. Niaraki, A. S., & Kim, K. (2009). Ontology based personalized route planning system using a multi-criteria decision making approach. Expert Systems with Applications, 36, 22502259. Park, J. H., & Storch, R. L. (2002). Pipe-routing algorithm development: case study of a ship engine room design. Expert Systems with Applications, 23, 299309. Saharidis, G. K. D., Minoux, M., & Ierapetritou, M. G. (2010). Accelerating Benders method using covering cut bundle generation. International Transactions in Operational Research, 17, 221. Santoso, T., Ahmed, S., Goetschalckx, M., & Shapiro, A. (2005). A stochastic programming approach for supply chain network design under uncertainty. European Journal of Operational Research, 167, 96115. Schmitt, A. J., & Singh, M. (2012). A quantitative analysis of disruption risk in a multi-echelon supply chain. International Journal of Production Economics, 139(1), 2232. Stadtler Aharidis, G. K. D., Minoux, M., & Ierapetritou, M. G. (2010). Accelerating Benders method using covering cut bundle generation. International Transactions in Operational Research, 17, 221. Stojkovic, M., & Soumis, F. (2001). An optimization model for the simultaneous operational ight and pilot scheduling problem. Management Science, 47(9), 12901305. Tang, C. S. (2006). Robust strategies for mitigating supply chain disruptions. International Journal of Logistics Research and Applications: A Leading Journal of Supply Chain Management, 9(1), 3345. Tomaszewski, A., Pioro, M., & Zotkiewicz, M. (2010). On the complexity of resilient network design. Networks. http://dx.doi.org/10.1002/net. 108118.

Please cite this article in press as: Kristianto, Y., et al. A model of resilient supply chain network design: A two-stage programming with fuzzy shortest path. Expert Systems with Applications (2013), http://dx.doi.org/10.1016/j.eswa.2013.07.009