18th European Symposium on Computer Aided Process Engineering – ESCAPE 18 Bertrand Braunschweig and Xavier Joulia (Editors) © 2008 Elsevier B.V./Ltd. All rights reserved. Supply Chain Design Considering Plants Performance Gabriela Corsano* and Jorge M. Montagna INGAR- Instituto de Desarrollo y Diseño - Avellaneda 3657- (3000) Santa FeArgentina Abstract Most of the Supply Chain (SC) design models have focused on the integration problem, where links among nodes have to be determined in order to allow an efficient operation of the whole system. In this level, all the problem elements are modeled like black boxes, and the optimal solution determines the nodes allocation and their capacity, and the connections among nodes. In this work, a MILP model is proposed where decisions about plant design are simultaneously taken with operational and planning variables. The model considers unit duplication and the allocation of storage of discrete sizes. The solved examples show the impact of plant design in the overall SC operation. Keywords: supply chain design, multiproduct batch plant, MILP. 1. Introduction A supply chain (SC) is a network of facilities (e.g. plants, warehouses, customers) that perform a set of operations ranging from the acquisition of raw materials to the distribution of finished goods to customers. The design of the SC involves making several decisions like number, size and location of plants, products to be produced in each plant, allocation of suppliers to plants and plants to distribution centers, etc. Most of the design models have focused on the SC integration (Tsiakis et al., 2001, Ryu and Pistikopoulos, 2005), where links among nodes have to be determined in order to allow an efficient operation of the whole system. In this level, all the problem elements are modeled like black boxes, with a maximum capacity, and the optimal solution determines the nodes allocation and their capacity, and the connections among them. Recently, Guillén et al. (2006) have addressed the SC design problem under demand uncertainty through a multistage stochastic formulation, using a decomposition technique that combines genetic algorithms and mathematical programming tools. Shah (2005) has described the state of the art in SC modeling. remarking that there are not published works dealing with the connection between process design and SC operation. In this work, a mixed-integer linear programming (MILP) model is proposed for the optimization of a SC where decisions about plant design are simultaneously taken with operational and planning variables. The model considers unit duplication in phase and out of phase and the allocation of intermediate storage of discrete sizes. Several examples are solved to show the impact of plant design in the overall SC operation. 2. Problem Description Four echelons in the SC are considered in this formulation: the raw material sites, the manufacturing plants, the warehouses and customer zones. At each raw material site s (s =1,…, Ns), one or more types of raw materials r (r = 1, …, Nr) are available to be delivered to plants l (l = 1, …, Nl), that operate during the time horizon Hl. Each 2 G. Corsano and J. M. Montagna multiproduct batch plant has stages j (j = 1, …, N jl ) for producing i products (i = 1, …, Np). Parallel units duplication in phase and out of phase are allowed as well as the allocation of intermediate storage between two batch stages. The warehouses m (m = 1, …, Nm) have different stock capacity and the product demand of the customer zones k, Dik, are known for k = 1, …, Nk. Moreover, the following data is assumed to be known: (i) the set of raw materials and their availability at each raw material site (ii) the set of available discrete sizes for units in the batch plants (iii) the cost parameters associated to: plants and warehouses installation, investment and production, transportation, raw material and maintenance. The problem is to determine: (a) the allocation of each plant and each warehouse (b) production of each product in each plant (c) the structure of the plant considering parallel units duplications and allocation of storage tanks, and unit sizes (d) the SC structure: flows among SC nodes. in order to minimize the total annual cost given by fixed cost, investment cost, production cost, stock and maintenance cost, and transportation cost. 3. Model Formulation The problem involves optimizing the SC performance simultaneously with the plant design in order to minimize the total cost. Following, the basic constraints are posed. 3.1. SC Network Constraints These constraints are mass balances between the different nodes in the SC. Q represent total amounts, for example Qsril are the kg of r from s to l to produce i, fr the raw material conversion factor, and zil is a binary variable equal to 1 if product i is produced in plant l, and zero otherwise: Ns Q s 1 sril frilr Qil r , i,l zil QilLO Qil zil QilUP Q sril i,l Q r , s, and Qil i,l UP rs (1) (2) Qsril zil Q UP rs r , s, i, l (3) i ,l Nm Q m 1 ilm (4) Let ym be the binary variable equal to 1 if the warehouse m is installed, zero otherwise, and Qmmax the capacity of warehouse m. Q ilm Q Qmmax ym and imk i ,l Nl Q l 1 Qmmax ym m (5) i,k Nk ilm Qimk Nm i,m (6) k 1 Dik Qimk i,k (7) m 1 3.2. Design Equations Let SVjl={VFjl1, VFjl2, …, VFjlP} be the set of available discrete sizes for stage j of plant l. p = 1,…, P represents the index for the different unit sizes. If the plant is not installed, then all the unit sizes of that plant have to be zero. Let exl be the binary variable that Supply chain design considering plants performance 3 takes value 1 if plant l is installed, and zero otherwise, and vjlp is a binary variable equal to 1 if unit j of plant l has size p. Then, the size for unit j of plant l, Vjl, is given by: P V jl v jlpVFjlp j , l (8) p 1 pSV jl v jlp exl j , l and zil exl i, l (9) The sizing equations that relate the unit size with the batch size (Bijl) are: Sijl Bijl V jl i, j , l NPjl Qil Nbil Bijl i , j , l (10) (11) NPjl represents the number of in phase units for stage j in plant l, Nbil the number of batches of product i in plant l, and Sijl the size factor. In order to avoid no linear constraints, the binary variable xxjld is defined and it takes value 1 if stage j of plant l has d units in phase, and zero otherwise, so NPjl NPjlUP d xx d 1 NPjlUP and jld xx d 1 jld exl j, l (12) Then, replacing Eq. (11) in Eq. (10) Sijl Qil Nbijl i, j, l V jl NPjl And substituting Eq.(13) for Eq. (8) and Eq. (12) Sijl Qil Nbijl eijlpd d , p VFjlp d (13) (14) Q if vilp and xx jld are equal to 1 With eijlpd il 0 otherwise UP i, j, l, p eijlpd Qil vilp (15) (16) d e QilUP xx jld ijlpd i, j, l, d (17) p P Qil eijlp i, l (18) p 1 3.3. Timing Constraints Let tijl be the processing time for product i in stage j of plant l, TLil the cycle time of product i in plant l, and NTjl the number of out of phase units for stage j of plant l. Then Til ≥ Nbijl TLil, represents the total time for producing i in plant l, so that Til H l . The i cycle time is defined by NT jlUP tijl TLil Nbijl x jln Nbijl i, j, l , n 1 n NT jl NT jlUP nx n 1 jln j, l (19) NT jlUP with x n 1 jln exl (20) 4 G. Corsano and J. M. Montagna Where xjln is equal to 1 if n out of phase units are used in stage j of plant l, and zero otherwise. In order to avoid no linear constraints, a new variable is defined as wijln = Nbil xjln, and replaced in (20). 3.4. Intermediate Storage Constraints According to Voudouris and Grossmann (1993) the upper bound for the storage vessels can be defined by VT jl 2 STijl Bijl s jl i , j 1, 2, , N j 1, l (21) VT jl 2 STijl Bi , j 1,l s jl i, j 1, 2, , N j 1, l (22) STijl represents the size factor for storage tanks and sjl is a binary variable equal to 1 if a tank is allocated after batch stage j, and zero otherwise. Let STFjl VTFj1l ,VTFj 2l ,...,VTFjG jl l be the set of available discrete sizes for the tank allocated after stage j. Let stjgl be the binary variable that takes value 1 if a tank of size g is allocated. VTFj1l is equal to zero to represent the no tank allocation. Using Eq.(11) in (21) and (22) the storage constraints are: STijl Qil (23) Nbijl 2 st jgl i, j 1, 2,..., N j 1, l g 1 VTFjgl STijl Qil (24) Nbi , j 1,l 2 st jgl i, j 1, 2,..., N j 1, l g 1 VTFjgl (25) st jgl exl j 1, 2, , N j 1, l g Using the continuous variable fijgl=Qil stjgl, constraints (23) and (24) become linear and: (26) fijgl QilUP st jgl i, g, j 1, 2,..., N j 1, l Qil fijgl i, j 1, 2, , N j 1, l (27) g The bounds for number of batches can be stated by 1 Nbijl 1 st jgl Nbijl Nbi , j 1,l Nbijl 1 st jgl Nbijl i, j 1,2, , N j 1, l (28) g 1 g 1 Where is a constant value corresponding to the maximum ratio allowed between consecutive batch sizes. The nonlinearity in these equations is eliminated with the definition of a new variable ijgl such that ijgl NbijlUP st jgl and Nbijl ijgl i, j 1,..., N j -1 , l (29) g 3.5. Objective Function The objective function is the minimization of the total annual cost given by fixed cost (plant and warehouse installation), investment cost (batch units and storage vessels), production cost (raw material consumption and operating cost), stock and maintenance costs, and transportation cost (between the SC nodes). 4. Example The example considers a SC that consists of 2 sites with 3 different raw materials, 3 possible plants with 3 stages for producing 3 products, 3 possible warehouses and 3 customer zones. Tables 1-3 show the most relevant model parameters. A set of 5 discrete sizes is provided for process units and 4 discrete sizes for storage tanks. Table 4 Supply chain design considering plants performance 5 shows these sizes and their costs. The problem was solved with GAMS via CPLEX solver with a 0% optimality gap. Table 1. Model parameters for the example Product (for all plants) fril R1 i1 i2 i3 0.8 0.7 0.6 R2 R3 0.85 0.7 0.65 0.7 0.7 0.65 tijl (h) J1 14 12 16 Table 2. Product demands (x103 kg) Product i1 i2 i3 k1 150 130 150 k2 200 220 250 J2 5 6 8 Sijl J1 J3 7 4 5 J2 0.9 0.6 0.7 J3 0.6 0.5 0.5 0.4 0.4 0.4 Table 3. Raw material availability (x103 kg) k3 115 130 150 Sites R1 R2 R3 S1 S2 3,200 3,500 3,000 3,200 1,950 1,800 Table 4. Discrete sizes and costs of units and storage tanks Cost jlV jl jl Discrete units sizes (l) Stages J1 J2 J3 J1 J2 P1 P2 300 500 300 500 300 500 Storage Tanks G1 G2 0 1500 0 1500 P3 P4 750 750 750 1000 1000 1000 G3 5000 5000 G4 10000 10000 P5 (x 103) 1200 1200 1200 L1 6 8 7 0.5 0.5 L2 9 6 10 L3 5 7 7 0.6 0.6 0.7 0.6 0.6 The optimal solution consists in only one plant that produces the 3 products. The plant and the unit sizes are shown in Fig. 1 as well the number of batches (Nbi) and the total production (Qi). A storage tank is allocated between stages 2 and 3, that is not used by product 1. All the consumed raw materials arrive from site 1 and two warehouses are installed. In the first row of Table 5 the investment and operative costs of this optimal solution (S1) are given. In Table 6 the optimal network distribution is showed. The same model without consider unit duplications and storage tank allocation is solved in order to compare the significance of plant design in the overall SC problem optimization. The optimal solution for this model consists in two plants: Plant 1 produces 392,920 kg of product 2 and 480,570 kg of product 3 with unit sizes of 1200, 1000 and 750 for stages 1, 2 and 3 respectively, and Plant 3 produces 465,000 kg of product 1, 87080 kg of product 2, and 69430 kg of product 3 with unit sizes equal to 1000, 750 and 500 for stages 1, 2 and 3 respectively. In Table 5 the economical results are shown (S2). The total annual cost is equal to $1406300 representing 28.6% bigger that the optimal solution considering plant design. 5. Conclusions This work has proposed a MILP model for the optimal production and distribution SC problem taking into account plant design in the overall model. Unit duplication in phase 6 G. Corsano and J. M. Montagna and out of phase as well as the allocation of intermediate storage tank where considered, using discrete unit and tanks sizes in order to avoid nonlinearities. The model was applied to a SC with four echelons and compared with a model where plant configuration is not taken account. The importance of considering plant configuration has been highlighted by using an example in which considerable savings on the overall annual costs have been achieved, given not only by the investment costs but also by the SC design. Acknowledgment The authors are grateful for financial support from CONICET and ANPCyT. References G. Guillén, F.D. Mele, A. Espuña, L. Puigjaner, 2006, Addressing the design of chemical supply chains under demand uncertainty, Ind. Eng. Chem. Res., 45, 7566-7581. J. Ryu, E. Pistikopoulos, 2005, Design and operation of enterprise-wide process network using operation policies, Ind. Eng. Chem. Res., 44, 2174-2182. N. Shah, 2005, Process industry supply chains: Advances and challenges, Comp. and Chem. Engn., 29, 1225-1235. P. Tsiakis, N. Shah, C.C. Pantelides, 2001, Design of multi-echelon supply chain networks under demand uncertainty, Ind. Eng. Chem. Res., 40, 3585-3604. V.T. Voudouris and I.E. Grossmann, 1993, Optimal synthesis of multiproduct batch plants with cyclic scheduling and inventory considerations, Ind. Eng. Chem. Res., 32, 1962-1980. Table 5. Investment and operative annual costs ($) S1 S2 Plant Investment 519,820 795,859 Tanks Production 18,643 - 116,200 131,150 Raw material 84,182 100,200 Warehouse Transport Total 109,450 109,450 245,500 269,670 1,093,795 1,406,329 Table 6. SC optimal design and product distribution for solution S1 (x10 3 kg) From sites to Plant 1 Products R1 R2 R3 From Plant 1 to warehouses M1 M3 i1 i2 i3 372 336 330 395 336 357 325 336 357 465 0 400 0 480 150 From warehouses to customers K1 K2 K3 M1:150 M3:130 M1:150 M1:200 M3:220 M1:250 M1:115 M3:130 M3:150 V=1200 l V=1000 l V=500 l V=1200 l V=5000 l Nb1 = 372 Nb2 = 240 Nb3 = 357 Nb1 = 372 Nb2 = 240 Nb3 = 357 Nb1 = 372 Nb2 = 384 Nb3 = 440 Figure 1. Plant design and production planning Q1 = 465,000 kg Q2 = 480,000 kg Q3 = 550,000 kg