Int J Adv Manuf Technol (2017) 91:1661–1670 DOI 10.1007/s00170-016-9792-9 ORIGINAL ARTICLE Design and optimization of an RFID-enabled automated warehousing system under uncertainties: a multi-criterion fuzzy programming approach Ahmed Mohammed 1 & Qian Wang 1 & Saleh Alyahya 1 & Nick Bennett 1 Received: 27 June 2016 / Accepted: 21 November 2016 / Published online: 16 December 2016 # Springer-Verlag London 2016 Abstract In this paper, we investigated the design and optimization of a proposed radio-frequency identification (RFID)enabled automated warehousing system in terms of the optimal number of storage racks and collection points that should be established in an efficient and cost-effective approach. To this aim, a fuzzy tri-criterion programming model was developed and used for obtaining trade-off decisions by measuring three conflicting objectives. These are minimization of the warehouse total cost, maximization of the warehouse capacity utilization, and minimization of the travel time of products from storage racks to collection points. To reveal the alternative Pareto optimal solutions using the developed model, a new approach was developed and compared with a recently developed fuzzy approach so-called Selim and Ozkarahan (SO). A decision-making algorithm was used to select the best Pareto optimal solution, and the applicability of the developed model was examined using a case study. Research findings demonstrate that the developed model is capable of generating an optimal solution as an aid for the design of the proposed RFID-enabled automated warehousing system. Keywords Automated warehouse . RFID . Design . Fuzzy approach . Multi-criterion optimization * Ahmed Mohammed [email protected] 1 School of Engineering, University of Portsmouth, Portsmouth PO1 3DJ, UK 1 Introduction Warehouses are one of main components consisting of an entire supply chain network in which a warehouse receives and stores merchandising products that are often transported from suppliers to retailers. Hence, accuracy of transportation time plays an important role on the entire supply chain network, which traditionally relies on a well-organized warehouse management [2, 15]. For the last decade, a growing trend has been seen in the application and implementation of automated warehouses aiming to improve the warehouse efficiency and capacity utilization and reduce the material handling time of warehouses. On the other hand, automation of warehouses is subject to additional costs that need to be considered; this led to research interests in optimization of automated warehouse designs by enhancing efficiency and reducing unnecessary costs. There are relatively few studies in optimization of automated warehouse designs in several aspects such as costs and capacity utilization. Lu et al. [9] reviewed some fundamental issues, methodologies, applications, and potentials of applying radio-frequency identification (RFID) techniques in manufacturing sectors. Van Der Berg [13] presented a review on approaches and techniques applied for the warehouse management planning and control. Ma et al. [10] formulated an automated warehouse as a constrained multi-objective model aimed at minimizing the scheduling quality effect and the travel distance. Huang et al. [3] proposed a non-linear mixed integer program under probabilistic constraints for site selection and space determination of a warehouse. The purpose of this work was to minimize the total cost of inbound and outbound transportation and the total cost of warehouse operations in a two-stage network. Lerher et al. [8] developed a multi-objective model for analyzing the design of 1662 an automated warehouse towards the minimization of the travel time of products, the total cost of the warehouse, and the number of material handling devices. Lerher et al. [7] also investigated the design and optimization of the automated storage and retrieval system aiming to minimize the initial investment and annual operating cost of the system using the genetic algorithm. Wang et al. [14] presented a study of an RFID-based automated warehousing mechanism in order to address issues of tighter inventory control, shorter response time, and greater variety of stock keeping units (SKUs), which are the most important challenges for designing future generation warehouses. Lu et al. [9] presented a five-step deployment process aimed at developing a holistic approach for implementing RFID in manufacturing enterprises. Lerher et al. [6] proposed a mono-objective optimization approach for seeking the cost-effective design of an automated warehouse. Ashayeri and Gelders [1] developed a design model of an automated storage and retrieval system incorporating the main influential parameters to minimize costs in investment and operation. Karasawa et al. [5] developed a non-linear mixed integer model aimed at minimizing the cost for an automated warehouse system. A review of the literature reveals that there were no previous studies in applying the fuzzy multi-criterion optimization approach in the context of the warehouse design [8], in particular for the RFID-enabled automated warehousing system. This paper addresses a contribution in developing a fuzzy tricriterion optimization model based on a proposed RFIDenabled automated warehousing system incorporating the uncertainty in varying demands, costs, and item locations. The developed model aims at simultaneously optimizing a number of conflicting criteria including minimization of the total cost, maximization of the warehouse capacity utilization, and minimization of travel time of products. In other words, it aims at obtaining a trade-off that can concurrently maximize the degree of satisfaction and minimize the degree of dissatisfaction at a time for the problem under investigation. The remaining part of the paper proceeds as follows: In Sect. 2, the problem description and the model formulation are presented. In Sect. 3, the optimization methodology is described. Section 4 demonstrates the application and evaluation of the developed multi-criterion model using a case study. In Sect. 5, conclusions are drawn. Int J Adv Manuf Technol (2017) 91:1661–1670 The entire operation of each conveyor system is controlled by a programmable logic controller that communicates with mounted sensors via a local area network. Within the RFID inventory management system, a chosen SKU can be released by the mechanical control system based on a number of assignment policies or rules. These rules include for example the rule of being nearest to a collection point and/or a modular arm which is free or adjacent to the chosen SKU. One of the main issues to be addressed in designing the proposed RFID-enabled automated warehouse includes allocating the optimum number of racks and collection points with respect to three criterion functions: (1) minimization of total cost, (2) maximization of capacity utilization of the warehouse, and (3) minimization of travel time of products from storage racks to collection points. 2.1 Notations The following sets, parameters, and decision variables were used in the formulation of the model: Set: set of nominated storage racks i ∈ I set of nominated collection points j ∈ J set of fixed departure gates k ∈ K I J K Given parameters: C ri C cj C ti C Tjk C lj Rlj N hj W S ri S cj Dj d1 d2 2 Problem description and model formulation d3 Figure 1 illustrates the structure of the proposed RFIDenabled automated storage and retrieval racks (AS/RR) used for this study [14]. The module comprises two types of powered conveyors aligned next to one another; these are input conveyors (storage racks) and output conveyors. djk Sp Spp fixed cost required for establishing an RFID-enabled rack i fixed cost required for establishing a collection point j unit RFID tag cost per item at rack i unit transportation cost per meter from collection point j to departure point k unit labor cost per hour at collection point j working rate (items) per laborer at collection point j minimum required number of working hours for laborers lat collection point j transportation capacity (units) per forklift maximum supply capacity (units) of rack i maximum supply capacity (units) of collection point j demand (units) of collection point j travel distance needed (m) for a pusher from its location to a selected item travel distance (m) of a selected item from its position at a storage rack to an output conveyor travel distance (m) of a selected item from its position at an output conveyor to a collection point travel distance (m) of a selected item from collection point jto departure gate k speed (m/s) of the moving-pusher along d1 speed (m/s) of the moving-pusher to push a selected item onto an output conveyer. Int J Adv Manuf Technol (2017) 91:1661–1670 1663 Items enter ontto a storage rack Fig. 1 Structure of the proposed RFID-enabled AS/RR d2 Thee output conveeyor system Taggeed items Spiral conveyors d3 d1 d3 Storage raack d3 Pusher p Output tto collecon points Sc 2.2.2 Criterion function 2 (F2) speed (m/s) of the output conveyor and the spiral conveyor. The capacity utilization is defined as the used capacity divided by the actual capacity. Thus, maximization of capacity utilization F2 is expressed as follows: Decision variables qij qjk xj yi quantity in units ordered from rack ito collection point j quantity in units dispatched from collection point jto departure gate k required number of laborers at collection point j n 1:if rack i is opened 0:otherwise n yj 0 1 12 X dðC a Þ−ðC u Þe2 A Max F 2 ¼ @ X i i∈I ð2Þ ∑ ∑ qij qij i∈I j∈ J and C ¼ , which refer to the u r ∑ S ri i∈ j j∈ J S i where C a ¼ ∑ ∑ i∈I 1:if collection j point is opened 0:otherwise actual (a) and used (u) capacity (C). 2.2.3 Criterion function 3 (F3) 2.2 Formulation of the multi-criterion optimization problem The three criteria, which include minimization of total cost, maximization of capacity utilization, and minimization of travel time, are formulated as follows: The total travel time (tt) of an in-store item includes tt of a pusher from its location to an item, tt of an item from its location at the storage rack to an output conveyer, and tt of an item onto a conveyer system to the collection point. Thus, minimization of travel time F3 is expressed as follows: X X d1 d2 d3 q þ þ ð3Þ Min F 3 ¼ S p S pp S c ij i∈I j∈ J 2.2.1 Criterion function 1 (F1) 2.3 Constraints In this case, the total cost of establishing the RFID-enabled automated warehouse includes the costs of establishing RFIDenabled racks, collection points, RFID tags, and transportation of products and labors in the warehouse. Thus, minimization of the total cost F1 can be expressed below: The above model was developed under the following constraints: X qij ≤ S ri yi ∀ j∈ J ð4Þ Min F 1 ¼ þ X X i∈I C ri yi þ X j∈ J C cj y j þ XX i∈I j∈ j C ti qij þ XX l m C Tij qjk =W f d jk j∈ J k∈K i∈I X X C lj x j N hj j∈ J ð1Þ qjk ≤ S cj y j ∀k∈K ð5Þ j∈ J i∈I qij ≥D j ∀ j∈ J ð6Þ 1664 Dj ≥ X Int J Adv Manuf Technol (2017) 91:1661–1670 X qjk ∀ j∈ J ð7Þ k∈K qij ≤ x j Rlj ∀i∈I ð8Þ j∈ J qij ; qjk ≥ 0; ∀i; j; k; ð9Þ yi ; y j ∈f0; 1g; ∀i; j; ð10Þ Equations 4 and 5 refer to the flow balance of a product traveling from a storage rack to a collection point and from a collection point to a departure gate. Equations 6 and 7 refer to demands in quantity to be satisfied. Equation 8 determines the required number of labors at a collection point. Equations 9 and 10 limit the decision variables to binary and non-negative. 3. Find the respective satisfaction degree μ(xi) for each criterion as follows: 8 1 > < F 1 ðxÞ−L1 μ1 ð F 1 ðxÞÞ ¼ > : U 1 −L1 0 8 1 > < F 2 ðxÞ−L2 μ2 ð F 2 ðxÞÞ ¼ > : U 2 −L2 0 8 1 > < F 3 ðxÞ−L3 μ3 ð F 3 ðxÞÞ ¼ > : U 3 −L3 0 if F 1 ðxÞ ≥ U 1 if L1 ≤ F 1 ðxÞ≤ U 1 ð17Þ if F 1 ðxÞ ≤ L1 if F 2 ðxÞ ≥ U 2 if L2 ≤ F 2 ðxÞ≤ U 2 ð18Þ if F 2 ðxÞ ≤ L2 if F 3 ðxÞ ≥ U 3 if L3 ≤ F 3 ðxÞ≤ U 3 ð19Þ if F 3 ðxÞ ≤ L3 3 The proposed optimization methodology 4. Transform the crisp model obtained from Sect. 3.2 to a single criterion function using the proposed solution approaches (shown in Sect. 3.3). 3.1 Solution procedures To reveal the alternative Pareto optimal solutions using the developed model, the following procedures were used:1. Convert the developed model into an equivalent crisp model (shown in Sect. 3.2). 2. Find the upper and lower bound (U, L) solution for each criterion function. This can be obtained as follows: 5. Vary the weight combination set consistently for the three criteria to reveal Pareto optimal solutions. Usually, the weight combination set is allocated by decision makers based on the importance of each objective. 6. Select the best Pareto optimal solution using the proposed decision-making algorithm. For upper bound solutions: X X XX Max F 1 ðU 1 Þ ¼ C ri yi þ C cj y j þ C ti qij i∈I j∈ J i∈I j∈ j l m XX X þ C Tij qjk =W f d jk þ C lj x j N hj j∈ J k∈K ð11Þ j∈ J 0 1 12 X dðC a Þ−ðC u Þe2 A Max F 2 ðU 2 Þ ¼ @ X i i∈I X X d1 d2 d3 q þ þ Max F 3 ðU 3 Þ ¼ S p S pp S c ij i∈I j∈ J ð12Þ ð13Þ j∈ J k∈K To incorporate the uncertainty in varying demands, costs, and item locations, the developed tri-criterion model is converted into an equivalent crisp model using the Jiménez method [4]. Accordingly, the equivalent crisp model can be formulated as follows: ð14Þ l . m X qjk W f d jk þ j∈ J ð20Þ j∈ J 1 12 X dðC a Þ−ðC u Þe2 A Min F 2 ðL2 Þ ¼ @ X i i∈I X X d1 d2 d3 q þ þ Min F 3 ðL3 Þ ¼ S p S pp S c ij i∈I j∈ J ! C cpes þ 2C cmos þ C copt j j j yj 4 ! C Tij pes þ 2C Tij mos þ C Tij opt 4 ! lpes lmos C j þ 2C j þ C lopt j x j N hj 4 X C rpes þ 2C rmos þ C ropt X i i i yi þ 4 i∈I j∈ J X X C tpes þ 2C tmos þ C topt XX i i i þ qij þ 4 i∈I j∈ j j∈ J k∈K Min F 1 ¼ For lower bound solutions: X X XX Min F 1 ðL1 Þ ¼ C ri yi þ C cj y j þ C ti qij j∈ j i∈I j∈ J i∈I l m XX X þ C Tij qjk =W f d jk þ C lj x j N hj 3.2 Formulating the uncertainty 0 ð15Þ 0 ð16Þ 1 12 X dðC a Þ−ðC u Þe2 A Max F 2 ¼ @ X i i∈I ð21Þ Int J Adv Manuf Technol (2017) 91:1661–1670 Min F 3 ¼ 1665 X X d pes þ 2d mos þ d opt d pes þ 2d mos þ d opt d pes þ 2d mos þ d opt 1 1 2 3 1 2 3 qij þ 2 þ 3 4S p 4S pp 4S c i∈I j∈ J Subject to X qij ≤S i yi ð22Þ The solution function (F) is formulated as follows: ∀ j∈ J ð23Þ i∈I Min F ¼ 3 3 X X ! ϑn μ f ðxÞ − F d ; n¼1 f ¼1 X Set ϑ*n ¼ qjk ≤ S j y j ∀k∈K ð24Þ j∈ J λ D j1 þ D j2 λ D j3 þ D j4 þ 1− ∀ j∈ J qij ≥ 2 2 2 2 i∈I λ D j1 þ D j2 λ D j3 þ D j4 X þ 1− ≥ qjk ∀ j∈ J : 2 2 2 2 k∈K X X λ x j1 þ x j2 : 2 2 j∈ J λ x j3 þ x j4 l þ 1− Rj 2 2 ð25Þ ð26Þ ϑn F •n F •n − F n ; ϑn ¼ 1 ð30Þ n¼1 then, F d ¼ ϑ*1 F 1 þ ϑ*2 F 2 þ ϑ*3 F 3 ¼ ϑ1 F •1 ϑ2 F • ϑ3 F • F1 þ • 2 FZ þ • 3 F3 • F 1− F 1 F 2− F 2 F 3− F 3 ð31Þ Based on the aforementioned procedures, the above criterion function can be expressed further as follows, qij ≤ x j 3 X Min F ¼ ðϑ1 μ1 −ϑ2 μ2 −ϑ3 μ3 Þ− ϑ1 F •1 ϑ2 F • ϑ3 F • F1 þ • 2 F2 þ • 3 F3 F •1 −F 1 F 2 −F 2 F 3− F 3 ð32Þ ∀i∈I ð27Þ qij ; qjk ≥ 0; ∀i; j; k; ð28Þ yi ; y j ∈f0; 1g; ∀i; j; ð29Þ According to Jiménez’s approach, it is supposed that the constraints in the model should be satisfied with a confidence value which is denoted as λ and it is normally determined by decision makers. Also, mos, pes, and opt are the three prominent points (the most likely, the most pessimistic, and the most optimistic values), respectively [4]. subject to Eqs. 4, 5, 6, 7, 8, 9, and 10. 3.3.2 The SO approach In this approach, the auxiliary crisp model in Sect. 3.2 is converted to a mono-criterion function using the following solution formula [12] Max λðxÞ ¼ γλo þ ð1−γ Þ X θf λf ð33Þ f ∈F subject to 3.3 Optimization approaches 3.3.1 The developed approach With the developed approach, the multi-criterion model can be transformed into a single-criterion model which is formulated by optimizing each criterion individually. This single-criterion model aims to minimize the scalarized differences between each criterion and its optimal value. Undesired deviations are proposed to be subtracted from the single criterion function with the aim to achieve more accurate criterion values. These values are close enough to Pareto optimal solutions which lead to a clear insight of a compromised solution between conflicting criteria for decision makers. λo þ λ f ≤ μðxÞ; f ¼ 1; 2; 3 ð34Þ x∈ F ðxÞ; λo and λ∈½0; 1 ð35Þ in which the value of variable λo = min μ{μ(x)}, which indicates the minimum satisfaction degree for each criterion function. Also, λf refers to the difference between the satisfaction degree of each criterion and the minimum satisfaction degree of criteria (λf = μ(x) − λo). 3.4 The decision-making algorithm The next step after revealing the Pareto solutions is to determine the best trade-off solution. The best Pareto optimal solution can be determined based on the decision maker’s preferences or by using a decision- 1666 Int J Adv Manuf Technol (2017) 91:1661–1670 making algorithm, although there are a number of approaches which can be utilized to determine the best solution in multi-criterion problems. In this study, the technique named TOPSIS (order preference by similarity to ideal solution) was employed for revealing the best trade-off solution. This approach can be used for selecting a solution nearest to the ideal solution but also the farthest from the negative ideal solution [11]. Assume PR − {PR op|o = 1, 2, ... , x (number of pareto solutions); p = 1, 2, ... , y (number of criteria)}refers to the x * y decision matrix, where PR is the performance rating of alternative Pareto solutions with respect to criterion function values. Thus, the normalized selection formula is presented as follows: PRop NPR ¼ o X PRap ð36Þ p¼1 The amount of decision information can be measured by the entropy value as −1 X PRop ln PRop Ep ¼ lnx o¼1 the distance of each alternative from the positive and negative ideal solutions is given as: ﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃ v( ) u y u X þ þ 2 t vop −vo Dp ¼ ; p ¼ 1; 2; :::; x ﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃ v( ) u o¼1 y u X 2 vop −v−o ; p ¼ 1; 2; :::; x D−p ¼ t ð42Þ ð43Þ o¼1 The relative closeness to each of the values of alternative solutions to the value of the ideal solution is expressed as follows: rcp ¼ D−p − ; p ¼ 1; 2; :::; x Dþ p þ Dp ð44Þ where D−p ≥ 0 and Dþ p ≥ 0; then, clearly, rcp ∈ [1, 0]. The trade-off solution can be selected with the maximum rcp or rcp listed in descending order. Figure 2 shows a flowchart of the proposed optimization methodology. x ð37Þ The degree of divergence Dp of the average intrinsic information under p = 1, 2, 3, 4 can be calculated as follows: Dp ¼ 1−Ep ð38Þ The weight for each criterion function value is given by wp ¼ Dp X Dk y ð39Þ 4 Application and evaluation In this section, a case study was used for examining the applicability of the developed tri-criterion model and evaluating the performance of the proposed optimization methodology. A range of application data is presented in Table 1. It is assumed Start Input model parameters Formulate the criteria k¼1 Thus, the criterion weighted normalized value is given by vop ¼ wo PRop ð40Þ where wo refers to the weight in alternatives which are normally assigned by the decision makers. The positive ideal solution (AT+) and the negative ideal solution (AT−) are taken to generate an overall performance matrix for each Pareto solution. These values can be expressed as below: þ þ ATþ ¼ maxðvo1 Þ maxðvo2 Þ max voy ¼ vþ 1 ; v2 ; :::; vy AT− ¼ minðvo1 Þ minðvo2 Þ min voy ¼ v−1 ; v−2 ; :::; v−y ð41Þ A distance between alternative solutions can be measured by the n-dimensional Euclidean distance. Thus, Transform to a crisp model Find the Max and Min solutions for each criterion Calculate membership functions for F1, F2 and F3 Solve the model using the developed approach Solve the model using the SO approach Pareto solutions TOPSIS decision making An optimal design of the RFID-enabled automated warehousing system Fig. 2 Flowchart of the optimization methodology Int J Adv Manuf Technol (2017) 91:1661–1670 Table 1 1667 Application data used for the case study I = 12 Cti = 0.25 £ J = 15 CTjk = 0.4–0.7 £ K=2 Rlj Clj = 6.5–9 £ Cri = 60–90 K£ = 100 djk = 20–45 m d1 = 0.1–4 m Sc = 35 m/s d2 = 0.3 m W = 48 d3 = 7–23 m Table 3 Assignment of weight values for obtaining Pareto solutions using two approaches Number Criterion weights ϑ1, Ɵ1 ϑ2, Ɵ2 ϑ3, Ɵ3 1 1 0 0 2 0.9 0.05 0.05 0.8 0.7 0.1 0.15 0.1 0.15 Si = 25–35 K£ Dj = 6–9 K Sp = 1 m/s 3 4 Sj = 20–29 K£ C cj = 15–18 K£ Spp = 0.8 m/s 5 6 0.6 0.5 0.2 0.25 0.2 0.25 7 0.4 0.3 0.3 8 0.3 0.35 0.35 that (1) the width, length, and height of each rack are W = 0.3 m, L = 18 m, and H = 5 m, (2) the distance between the start of a spiral conveyer and the end of a collection point is 2 m, and (3) the pusher is located at the center of each rack. All these parameters are taken from a real-world automated warehouse design; the prices of RFID equipment and its implementation were estimated based on the marketing prices. The optimizer of the developed tri-criterion model is LINGO11. All computational experiments were conducted on a laptop with a 2.60-GHz CPU and a 4-G memory. and collection points that should be established. Figure 3 illustrates Pareto optimal fronts among the three criterion functions obtained by using the two approaches. 4. Select the best solution using the TOPSIS method; the scored values of Pareto optimal solutions are reported in Table 6. 4.1 Results and discussions This section presents the results which were obtained based on the developed fuzzy tri-criterion model using the proposed fuzzy solution approaches for the problem previously defined. The solution steps of the developed model are described as follows: 1. Obtain the upper and lower values for each criterion function by solving them individually. The results are {U F i ; L F i }) = ({504, 1230}, {0.66, 0.94}, {4.27, 12.25}. 2. Find the respective satisfaction degree μ(xi) for each criterion function. The satisfaction degrees are reported in Table 2. 3. Convert the multi-objective crisp model to a single criterion model using (i) the developed approach by assigning weight values shown in Table 3 and (ii) the Selim and Ozkarahan (SO) approach by assigning the value of which is set as 0.33 by the decision makers who consider a balance in importance of each of the three criteria. The two approaches are compared by assigning different λ levels. Table 4 shows the computational results obtained using the two approaches. Accordingly, Table 5 shows the corresponding optimum numbers of storage racks Table 2 μ(x1) μ(x2) μ(x3) Result of satisfaction degree of each criterion function 0.95 0.7 0.97 0.93 0.78 0.96 0.85 0.83 0.93 0.81 0.88 0.90 0.7 0.92 0.85 0.623 0.97 0.84 0.6 0.98 0.81 0.55 0.99 0.76 As mentioned above, Tables 4 and 5 show the obtained two sets of Pareto optimal solutions, respectively, which were obtained based on the three criterion functions to determine the number of storage racks and collection points that should be established. For instance, solution 1 shown in Table 4 is obtained using the developed approach under an assignment of ϑ1 = 1 , ϑ2 = 0 , and ϑ3 = 0; it gives the minimum total cost of 504 K£, the maximum capacity utilization of 66%, and the minimum travel time for all the requested products of 4.29 h. The result is shown in Table 5; the solution consists of six storage racks and nine collection points, and these trade-off results are obtained based on the three criteria towards the minimization of total cost, the maximization of capacity utilization, and the minimization of travel time. Nevertheless, as shown in Fig. 3, with the Pareto optimal method, it cannot generate a better overall result by gaining one best result based on one criterion function without worsening the results in the other criterion functions, although all Pareto optimal solutions are feasible. It proves the confliction among the three criteria. For instance, an increase in the desired value of criterion 2 (e.g., maximization of capacity utilization) leads to an increase in the undesired value of criterion 1 (e.g., minimization of total cost). It can be noted in Table 4 that by increasing the satisfaction level λ, it leads to an increase in the undesired value of the first and third criterion functions (e.g., minimization of total cost and minimization of travel time, respectively). However, it yields an increase in the desired 1668 Table 4 Int J Adv Manuf Technol (2017) 91:1661–1670 The results obtained by assigning the varying λvalues to each of the three criterion functions 378 non-zero elements, 64 constraints, 129 total variables, 68 integer variables Number λ-level Developed approach SO approach Min F1 (K£) Max F2 (%) Min F3 (h) Run time (s) Min F1 (K£) Max F2 (%) Min F3 (h) Run time (s) 1 0.3 504 0.66 4.29 2 504 0.66 4.29 2 2 3 0.4 0.5 595 678 0.71 0.78 5.31 6.51 2 2 595 681 0.71 0.78 5.31 6.58 3 2 4 0.6 795 0.84 7.75 1 790 0.84 7.69 3 5 6 0.7 0.8 894 978 0.89 0.92 8.92 10.18 3 4 913 1053 0.89 0.93 9.12 11.91 4 3 7 0.9 1064 0.93 11.97 4 969 0.92 10.33 4 8 1 – – – – 1096 0.94 12.19 4 value of the second criterion function (e.g., maximization of capacity utilization). In this case, decision makers have to spend more money to cope with the uncertainties. However, decision makers can vary the weight in importance (ϑn, or f) of each of the three criterion functions and the satisfaction level λ based on their preferences in order to obtain another compromised solution. Through a comparison of the two sets of Pareto optimal solutions shown in Table 4, the values obtained based on the three criterion functions using the developed approach are more balanced than those (of solutions 6–8) using the SO approach. The optimization run time of using the developed approach for the eight iterations was slightly faster than the SO method. It also indicates that there is no feasible solution obtained using the developed approach when the weight for the first criterion (minimization of total cost) is set less than 0.4. This implies that decision makers cannot ignore the importance of cost as it yields an inapplicable warehouse design. In Table 5 The optimal numbers of storage racks and collection points that should be established Number 1 2 3 4 5 6 7 8 Developed approach SO approach Opened storage racks Opened storage racks Opened collection points 6 6 7 9 10 12 11 12 9 9 8 11 13 14 13 15 6 6 7 9 10 11 11 – Opened collection points 9 9 8 11 12 13 13 – other words, with the developed approach, it gives a more realistic and balanced solution. After obtaining a set of Pareto optimal solutions, decision makers may determine a solution depending on their preferences or using a decision-making algorithm. In this work, the TOPSIS method was employed to select the best solution. As shown in Table 6, solution 6 is chosen as the best solution as its score is the highest (rcp = 0.279) with the total cost of 978 K£, 92% capacity utilization, and travel time of 10.18 h. Also, it requires an establishment of 11 storage racks to supply products to 13 collection points. 5 Conclusions In this research, a design of the proposed RFID-enabled automated warehousing system was studied using the multiobjective optimization approach. The work was involved in optimization of the design in terms of (1) allocating the optimal number of storage racks and collection points that should be established and (2) obtaining a trade-off decision between the negative impact of costs and the positive impact of maximization of the warehouse capacity utilization and minimization of travel time of products traveling from storage racks to collection points. To this aim, a tri-criterion programming model was developed and the model was also converted to be a fuzzy programming model for incorporating parameters in varying which include demands, costs, and random locations of items in a warehouse. A two-stage solution methodology was proposed to solve the fuzzy multi-criterion optimization problem. At the first stage, the developed approach and the SO approach were used for obtaining two Pareto optimal sets. The results, which were obtained using the two different approaches, are compared, and it shows that both approaches are appropriate and efficient for the fuzzy multi-criterion model for revealing a trade-off decision among the considered Int J Adv Manuf Technol (2017) 91:1661–1670 1150 1050 1050 Total cost (K£) 1150 Total cost (K£) Fig. 3 Pareto optimal fronts among the three criterion functions obtained by the two approaches 1669 950 850 750 650 550 950 850 750 650 550 450 0.6 0.7 0.8 0.9 450 0.6 1 Using the SO approach 950 950 Total cost (K£) 1050 Total cost (K£) 1050 850 750 650 1 850 750 650 550 550 5.5 7.5 9.5 11.5 450 3.5 13.5 Using the developed approach 12.5 11.5 Travel time (h) 10.5 9.5 8.5 7.5 6.5 5.5 4.5 0.7 0.8 0.9 5.5 7.5 9.5 13.5 12.5 11.5 10.5 9.5 8.5 7.5 6.5 5.5 4.5 3.5 0.6 0.7 0.8 Capacity utilization (%) Capacity utilization (%) Using the developed approach Using the SO approach criteria. Nevertheless, the developed approach has some advantages, which include: (1) the solutions gained using this approach are more balanced than using the SO approach (2) with the developed approach, the run time is slightly faster than using the SO approach, and (3) it gives more realistic solutions for an Table 6 Pareto optimal solutions ranked based on scores using the TOPSIS method Developed approach Solution 1 2 3 4 5 6 7 8 Score 0.245 0.234 0.266 0.245 0.2544 0.279 0.273 SO approach Solution 1 2 3 4 5 6 7 8 Score 0.245 0.234 0.266 0.245 0.2544 0.267 0.273 0.243 11.5 13.5 Travel time (h) Using the SO approach Travel time (h) Travel time (h) 0.9 Using the developed approach 1150 3.5 0.6 0.8 Capacity utilization (%) 1150 450 3.5 0.7 Capacity utilization (%) 0.9 applicable warehouse design. In the second stage, the TOPSIS method was employed to reveal the best Pareto solution. Finally, implementation of a case study demonstrates the applicability of the developed model and the effectiveness of the proposed optimization methodology which can be useful as an aid for optimizing the design of the RFID-enabled automated warehousing system. An interesting research study derived from this work may be a comparison between the RFID-enabled automated warehousing system and the non-RFID-enabled automated warehousing system in terms of these three criteria (e.g., minimization of total cost, maximization of capacity utilization, and minimization of travel time). It was also suggested to compare the developed solution approach with the other available approaches such as e-constraint and augmented e-constraint. Finally, optimizing the developed model by a meta-heuristic algorithm may be useful for handling the large-sized problems in a reasonable time. 1670 Int J Adv Manuf Technol (2017) 91:1661–1670 References 1. 2. 3. 4. 5. 6. 7. Ashayeri J, Gelders LF (1985) A microcomputer-based optimization model for the design of automated warehouses. Int J Prod Res 23(4):825–839 Choi TM, Yeung WK, Cheng TCE (2013) Scheduling and coordination of multi suppliers single-warehouse-operator singlemanufacturer supply chains with variable production rates and storage costs. Int J Prod Res 51(9):2593–2601 Huang S, Wang Q, Batta R, Nagi R (2015) An integrated model for site selection and space determination of warehouses. Comput Oper Res 62:169–176 Jiménez M, Arenas M, Bilbao A, Rodriguez AD (2007) Linear programming with fuzzy parameters: an interactive method resolution. Eur J Oper Res 177:1599–1609 Karasawa Y, Nakayama H, Dohi S (1980) Trade-off analysis for optimal design of automated warehouses. Int J Syst Sci 11(5):567– 576 Lerher T, Potrc I, Sraml M, Sever D (2007) A modeling approach and support tool for designing automated warehouses. Adv Eng 1: 39–54 Lerher T, Potrc I, Sraml M (2010) Designing automated warehouses by minimising investment cost using genetic algorithms. V: ELLIS, Kimberly Paige (ur.). Progress in material handling research: 2010. Charlotte: Mater Handl Ind Am, cop 2013:237–253 Lerher, T., Šraml, M., Borovinšek, M., Potrč, I. (2013) Multiobjective optimization of automated storage and retrieval systems. Ann Faculty of mechanical engineering-International journal of Engineering, 187–194. 9. Lu BH, Bateman RJ, Cheng K (2006) RFID enabled manufacturing: fundamentals, methodology and application perspectives. Int J Agile Syst Manag 1:73–92 10. Ma H, Su S, Simon D, Fei M (2015) Ensemble multi-objective biogeography-based optimization with application to automated warehouse scheduling. Eng Appl Artif Intell 44:79–90 11. Ramesh S, Kannan S, Baskar S (2012) Application of modified NSGA-II algorithmto multi-objective reactive power planning, Appl. Soft Comput 12:741–753 12. Selim H, Ozkarahan I (2008) A supply chain distribution network design model: an interactive fuzzy goal programming-based solution approach. Int J Adv Manuf Technol 36:401–418 13. den Berg V (1999) A literature survey on planning and control of warehousing systems. IIE Trans 31(8):751–762 14. Wang Q, McIntosh R, Brain M (2010) A new-generation automated warehousing capability. Int J Comput Integr Manuf 23(6):565–573 15. Yeung WK, Choi TM, Cheng TCE (2011) Supply chain scheduling and coordination with dual delivery modes and inventory storage cost. Int J Prod Econ 132:223–229 8.