Int J Adv Manuf Technol (2006) 28: 766–774 DOI 10.1007/s00170-004-2427-6 ORIGINAL ARTICLE Riccardo Manzini · Mauro Gamberi · Alberto Regattieri Design and control of an AS/RS Received: 10 May 2004 / Accepted: 21 September 2004 / Published online: 25 May 2005 © Springer-Verlag London Limited 2005 Abstract Automated storage/retrieval systems (AS/RSs) are a combination of equipment and controls which automatically handle, store and retrieve materials (components, tools, raw material and subassemblies) with great speed and accuracy. Consequently, they are widely used in industrial companies to manage products with cost-effective utilization of time, space and equipment. This paper presents a multi-parametric dynamic model of a product-to-picker storage system with class based storage allocation of products. Thousands of what-if scenarios are simulated in order to measure the impact of alternative design and operating configurations on the expected system performance and to identify the most critical factors and combinations of factors affecting the response of the system. Class based storage proves to be a very effective way of both reducing the picking cycle time and maximizing the throughput of the system. The rapid effectiveness of visual interactive simulation (VIS) in supporting the design (redesign) and control of new (existing) warehouses emerges, responding to the need for flexibility which modern companies need in order to adapt to strongly changing operating conditions quickly. Keywords Automated storage and retrieval system (AS/RS) · Class based storage · Factorial analysis · Order picking system (OPS) · Warehousing system · What-if scenarios 1 Introduction New, global and extended markets are forcing companies to process and manage increasingly differentiated products with shorter life cycles, low volumes and reduced customer delivery times. In today’s global marketplace companies need to R. Manzini (u) · M. Gamberi · A. Regattieri Department of Industrial Mechanical Plants, University of Bologna, Viale Risorgimento, 2., 40136 Bologna, Italy E-mail: riccardo.manzini@mail.ing.unibo.it Tel.: +39-51-2093406 Fax: +39-51-2093411 be able to deliver products on time, maintain market credibility and introduce new products and services faster than competitors. In particular, recent growth and strong development of e-commerce has brought a new focus on warehousing facilities and especially on the design and management of order picking systems (OPS) where typically thousands of customers’ orders have to be processed per day. This is especially true in e-fulfillment (e.g., business to consumer – B2C fulfillment), because internet consumers generally order one to two products in small quantities achieving a one-day cycle time for orders [1]. Order picking (OP) can be defined as the retrieval of items from their warehouse locations in order to satisfy demands from internal or external customers: it is a process of gathering requested stock keeping units one order at a time. Picking operations are carried out by a great many large and medium-sized companies, which belong to different industrial and service sectors. OPSs can be classified as picker to part (or product) systems when the picker travels to picking locations, and part (or product) to picker if materials are automatically brought to the picker. A well-know example of part to picker OPS is represented by automatic storage/retrieval systems (AS/RS), which are a response to the increasing trend towards both the automation of warehousing operations and the increasing labor costs [2]. Other related available automated technological options and devices are the automated guided vehicles (AGV), pick list generation software, bar coding, etc. The aim of this study is to identify the most critical factors affecting the response (i.e., the performance) of an AS/RS based on the prevailing activity of picking. AS/RSs are computer-directed storage and transport facilities for large capacity and high volumes of handled materials. They consist of storage racks erected along aisles with unique or un-unique cell conveyors, input/output (I/O) stations for receiving and sending items, and storage/retrieval (S/R) machines for providing transport between I/O stations and storage cells. The automated stacker crane (i.e the S/R vehicle) travels within an aisle performing storage and retrieval operations. 767 When considering the retrieval activity, there are two principal macro classes of OPSs: • Unit load systems. Materials are moved and stored by devices capable of moving and storing only a single-unit handling load (e.g., pallets, odette boxes). • Less than unit load systems. With multiple stops per trip. Vehicles are capable of handling multiple unit loads simultaneously. Order pickers, who can be identified with operatorsaboard retrieval vehicles (in picker to part systems) or with S/R automated machines in AS/RSs, the object of this study, retrieve sets of items or multiple handling units of the same item on a single OP cycle. They visit different slots in the warehouse before returning to the I/O or depot areas. Each picker is responsible for picking a complete pool of customer orders during a mission [3]. The aim of this study is to identify and measure the combined and not combined effects of different system parameterizations on the performance of both unit load and less than unit load part to picker OPS. The following section discusses the principal AS/RS and OPS studies in the literature. Section 3 presents the storage location policies and particularly the class-based policy for the optimization of warehousing systems. Section 4 discusses batching procedures for the reduction of picking cycle time and presents the proposed genetic algorithm based on cellular automation. Sections 5 and 6 present a dynamic model and related results based on simulation for the what-if analysis of different OPS configurations that work in different operating scenarios. 2 Literature overview Manzini et al. [4] and Ferrari et al. [5, 6] discuss the importance of flexibility, i.e., the ability of a system to adapt to changing market demands in terms of both product variations or changes (capability flexibility) and product quantity (capacity flexibility). Consequently, the importance of flexibility is progressively emerging in inventory management and in picking activities (in particular in automated S/R systems). Several studies discuss the importance of integrating AS/RS with other flexible devices, such as flexible manufacturing systems (FMSs) and AGVs. They present algorithms and procedures to support the optimization (e.g., storage allocation of materials, storage and retrieval sequencing) of the whole production system [7–10]. A large set of studies discuss picker to part OPSs [2, 3, 11– 24]. In particular, the authors of this paper present in-depth studies of the design and management of flexible picker to part OPSs [21–24]: principal factors affecting the performance of the picking activities are identified and their effects are measured and compared. The literature presents many studies supporting the design and control of an AS/RS. In particular, Sharker and Babu [25] present a review and comparison of analytical travel-time models in unit-load AS/RS. Other significant studies on unit-load AS/RS, for both single command and dual command with interleaving cycles (i.e., cycles performed on an opportunistic basis where a retrieval transaction is combined with a storage transaction), are presented by [26–30]. All these studies are generally based on analytical models and involve a small number of multi-level parameters because of the NP-hard computational complexity of the optimizing problem [21, 23, 24, 31, 32]. This is especially true for the less than unit load OPS optimization problem (OPSP), which is the object of this study. The most important planning issues concerning the design and management of both picker to part and part to picker OPSs can be addressed on two levels (i.e., the level of policies and the level of operations) and grouped into three main operating strategies [21–24, 33]: • Storage location assignment and fulfillment policies, i.e., the assignment of items to storage locations; • Order consolidation, i.e., the transformation of customer orders into picking orders; • Routing and sequencing, i.e., the identification of the routing of pickers through the warehouse. The purpose of Sects. 3 and 4 is to present different alternative policies for each of these groups of strategies. 2.1 OPS performance evaluation There are a great many different criteria for the evaluation of an OPS performance, but in the literature travel time (picking cycle time) is commonly used: this is the expected amount of time for the S/R machine to perform a list of storage or retrieval operations. In fact, the throughput of OPS is normally the inverse of the travel time. The literature demonstrates that the OP travel time accounts for about 50% of all OP activities [34], while system cost accounts for over 65% of total operating costs for a typical warehouse [13, 14, 35]. Reference [3] lists the principal contributors to the total picking cycle time: • Administrative time at the I/O point, at the start and end of a tour. • Processing time the time spent extracting items and documenting the picking activity. • Travel time between pick locations. The importance of reducing picking variable traveling cycle time emerges: it is a monotone increasing function of the distance traveled. For this reason the measure of the mean distance traveled in a picking cycle is a significant indicator of the performance of the generic OPS. This distance is adopted as the response (i.e., the performance) of the generic OPS in order to compare different configurations (i.e., system parameterizations) and optimize the management of the whole system. 3 Storage location policies The literature discusses storage location assignment strategies at great length [24, 36]. It is possible to discriminate two macroclasses of assignment storage location rules: 768 • Storage location assignment policies. The selection of open locations for incoming unit loads are subject to imposed constraints. • Open locations selection rules (e.g., closest open location, nearest neighbor) when the policy does not make a unique selection. Warehouse assignment-dispatching rules are differentiated by the objective they address; known examples are [3, 4, 12, 15, 33, 34, 36–41]: • Class based storage policy. It partitions all products into a number of classes and reserves a physical portion within racks for each class. This storage policy is generally managed according to the dispatching rule based on the cube per order index (COI) introduced by [42]. It is defined as the ratio of the number of storage addresses allocated to an item and the number of transactions per period. This rule is applied by routing incoming items with lowest values to the most accessible (nearest to I/O points) storage addresses of a facility. In class-based storage policy each class is assigned to a block of storage locations: for this reason the size of classes and distributions of products among them have to be properly designed. Within each block of storage locations, material is generally stored randomly. The CL-K notation denotes the number of classes (K ) according to a class-based policy. Several authors demonstrate that storage policy based on COI and its correlated assignment rule are the most effective with respect to random storage in unit load picking systems [43–45]. Studies on picker to part less than unit load OPS with a class based storage policy are presented by [21, 22], and [23]. Any shape is possible for each: rectangular, triangular etc. [38]. Park and Webster [37] present some algorithms for minimizing travel times according to a class-based storage assignment and in three dimensional storage systems: they refer to L-shaped classes in automated OPS. Reference [40] studies the impact of combining storage policy, tour construction heuristic and storage rack configuration on AS/RS travel time. Reference [3] develops a study on the OPS optimization based in particular on the shape and trend of the COI-based ABC curve. • Turnover based storage policy. Items with highest transaction demand are stored in the most accessible addresses. In particular, in continuous policy all rack locations are classified according to the following non-decreasing value: t jin + t jout (1) where t jin and t jout are the travel time from generic location j to the input and output stations respectively; and all products are classified on non-increasing demand per reserved location. The aim is to prevent the generic unit-loads from being assigned to locations designated for faster moving products. • Randomized storage. It allows products to be stored anywhere in the storage area. • Duration of stay storage. This is a rule for randomized systems and is based on the duration of stay. • Dedicated storage. Materials are assigned to predetermined locations based on throughput and storage requirements. • Service level assignment. The aim of this policy is to determine the minimum size that satisfies a service-level objective (expressed as a probability of a space shortage) in randomized or dedicated storage. • Correlated storage policy. This is based on correlations between products, and reduces OP travel times by storing correlated products close to each other. Randomized and dedicated storage are extreme cases of the class-based storage policy: the former only considers one class while dedicated storage policy assigns a class to each product; in other words, class-based storage is a compromise between the randomized and dedicated storage policies. 4 Batching and routing procedures The order consolidation policies rearrange customer orders into picking orders. In single order picking each customer order, generally made of multiple items to be picked within the system (less than unit load OPS), is directly taken as a picking order. In order batching picking several customer orders are combined into a batch, i.e., a single picking order of requests (picking lines in the picking list) belonging to different customers. The literature presents a great many studies proposing clustering procedures to solve the batching problem [24, 45–49]. In the proposed study the order batching problem is defined as follows: given a set of orders, each consisting of a number of orderlines, how can the amount of orderlines be batched into picking orders according to the capacity of the picking device? Reference [24] identifies and discusses three principal families of heuristic order batching algorithms: 1. Priority rule-based algorithms, 2. Seed algorithms, 3. Saving algorithms. The group of research compares two different algorithms. Firstly, a large set of scenarios are simulated according to the first come first serve (FCFS) batching rule, i.e., customer orders and related orderlines are assigned to batches one by one, picking orders of a defined number of orderlines (i.e., a parameterized number of slots to be visited) according to the application of the shortest path (SP) routing policy [50]. Secondly, a heuristic rule based on genetic programming and cellular automation [51] applied to the traveling salesman problem (TSP) with Chebyshev distances [52, 53] is proposed. In this the picking requests (i.e., the amount of orderlines belonging to an input orderlist, which represents the mission assigned to an AS/R vehicle) are rearranged within the orderlist (TSPmission algorithm) or within pools of orderlines which saturate the vehicle capacity (TSPtrip algorithm). In other words, the generic mission is composed of different trips within the storage area according to the finite capacity of the picking vehicle: the rearrangement of locations to be visited is applied to the entire mission (TSPmission ) or to each vehicle trip (TSPtrip ). In the latter condition, TSPtrip is applied many times within the same mission. 769 The hypothesis is that a set of picking missions, generated randomly and according to a defined storage allocation policy, are assigned to the generic AS/R truck. This total number of picking and stochastic requests, which generate random outputs and are assigned to one vehicle, is calculated in order to predict a picking cycle time (the performance of the system) for a single mission with a high level of acceptance (model validation). 3. 5 Dynamic AS/RS model This section presents the proposed dynamic model of an AS/RS and concentrates on the multi-level factors which parameterize the system. The product to picker OPS has been modeled by the application of an object-oriented visual interactive simulation (VIS) tool [54]. This model involves more than 40 interrelated dynamic processes (e.g., initializing, ordering, batching, routing) and 2000 functional entities (e.g., control points, loads, vehicles, routes). Firstly, by the analysis of real warehousing systems based on picking activity the most significant parameters have been collected in order to define the list of principal factors which could affect the response of an AS/RS. These factors are the object of the system parameterization process. This set of factors is composed of the following: 1. AREA (A). This is related to the storage capacity of the warehousing system and is proportional to the number of locations capable of receiving picking items. 2. CURVE (y). The class based storage location assignment policy is adopted and is based on three classes (A, B and C). The generic class is identified by different COI values, according to a COI – Pareto ABC curve that is related to physical stocks and movement frequencies. The notation x/y, attributed to this factor, indicates that x% of cumulated stor- Fig. 1. Automated storage and retrieval system with class based storage allocation. Classes A, B, C 4. 5. 6. 7. age commits y% of material movements. x is assumed to be equal to 20. CLASS (a/b/c). This factor deals with the physical dimension of each class of storage item. The notation a/b/c and a, b, c values indicate the portion of the total storage capacity (in percentage), which is associated with classes A, B and C respectively. If (a, b, c) = (0.10, 0.25, 0.65), then 10% of the most requested items belong to the first class of movement (A), which is located very closed to the I/O area. Figure 1 exemplifies the distribution of items within the three classes for a defined value of class factor. DIMORD. This is the number of lines of a picking list and it is associated with a mission assigned to a retrieval vehicle. The generic picking vehicle begins and ends its mission at the I/O zone, when it reaches the end of the list. The picking mission is made of several routes (trips) which begin and end at the I/O: every time the vehicle saturates its capacity it has to reach the depot (I/O) area, unload the items collected during the current trip and continue the mission by initializing a new trip. CAPACITY. This factor relates to the capacity of the picking machine and to the average number of items picked at each stop during a vehicle trip. In particular, it represents the average number of stops per route (i.e., trip). SHAPE (p&q). This factor is the ratio between the length and the height of the generic rack (see Fig. 1). K . This is the ratio between the length ( p ) and the height (q ) of the area dedicated to class A in accordance with a rectangular shape. This value can differ from the ratio p/q (shape factor). B and C classes’ configuration is L-Shape (Fig. 1). The previously described batching procedures were tested in order to quantify the effect of the rearrangement of the pick- 770 ing requests on the response of the system, i.e., the normalized picking cycle time (T _Normal). T _Normal is the response of the system recorded during the what-if analysis and is defined as the ratio between the generic picking cycle time and the worst (i.e., the longest) cycle time obtained by the same system configuration but using a random allocation of products. In fact, the what-if analysis demonstrates that the greatest values of picking cycle time are the result of a non-application of the class based storage policy (i.e., c = random in Eq. 3). The T _Normal value is defined according to the following: T _NormalArea= A (c, l, s, r, d, k) CTime(c, l, s, r, d, k/a = A) = ≤1 max {CTime(c, l, s, r, d, k)i /a = A} (2) i max (c,l,s,r,d,k) {CTime(c, l, s, r, d, k/a = A)} = max {CTime(a = A, c = random, s, r = 1, d = 1, k)} s,d,k (3) where CTime(c, l, s, r, d, k) is the picking cycle time for the mission associated with the picking list; a = AREA value, c = CURVE value, l = CLASS value, s = SHAPE value, r = CAPACITY value, d = DIMORD value, k = K value. T _Normal is calculated for different storage capacities (AREA value) and the relationship 2 offers the opportunity to compare all values obtained in accordance with different system configurations, storage capacities and operating scenarios. This response value is capable of measuring the impact of different policies and factor values (e.g., class dimensions, COI-curve, number of picks per picking list, vehicle capacity) on the system performance. 6 Results and factorial analysis A multi-level factorial analysis was conducted in order to measure the impact of each factor and combinations of factors on the throughput capacity of the system (i.e., the inverse of the mean Fig. 2. Plot of main effects on normalized picking cycle time (T _Normal) picking cycle time, T _Normal). Moreover, thanks to the collection of a large portfolio of simulation results, the simulation steps and accompanying statistical analysis provide several OPS design guidelines for the choice of the best picking strategies (free parameters) given a subset of system constraints (unfree parameters). The collected output of every simulation run is the global picking cycle time for a set of more than 1000 stops (orderlines) belonging to different trips and picking cycles. All of the principal results are presented in Fig. 2 (plot of main effects) where the picking cycle time (i.e., travel distance in agreement with the definition introduced in Sect. 2.1) is normalized in comparison with the cycle time based on a random storage policy. In this graph the data means of T _Normal (the normalized cycle time) are a measure of the system performance: the average response of the system is seen when each of the free-parameter (AREA, RATIO, SHAPE, etc.) changes its value. The dashed line identifies the mean value obtained from the total number of simulation runs. From this figure the importance of certain factors affecting the picking cycle time T _Normal, compared to the other factors, is emphasized. In particular, Fig. 2 demonstrates an insignificant influence of AREA factor on system performance; as a consequence, all results presented in this manuscript can be generalized for a warehousing system with a generic storage capacity. The effect of each factor and each combination of factors on the response of the system is measured more accurately by applying the design of experiment (DOE) analysis to the normalized response time of the modeled part to picker OPS. The DOE analysis is a two-level factorial analysis (i.e., the effect of each factor is measured according to two levels) which is based on the following statistical model [55]: Y = α+ βi X i + βij X i X j + βijk X i X j X k + . . . + u i i, j i< j i, j,k i< j<k (4) where Y is the response of the system, α is a fixed parameter, X i is the value assumed by the factor i, βi is a multiplying parameter 771 associated with factor i and is a measure of the effect of i on the response Y (similarly for βij , βijk , etc.) and u is the error on the expected response of the system. In particular, the effect of a factor (e.g., CLASS, CAPACITY) or a combination of factors (e.g., CLASS*CAPACITY, CLASS*CAPACITY*CURVE) on Y is quantified by the following standardized effect (defined for the factor i in Eq. 5) in accordance with a t-Student distribution of values: β̂i − βo t= V(β̂i ) Factor AS/R - OPS levels AREA DIMORD CAPACITY SHAPE ( p/q) CURVE (x/y) CLASSES (A/B/C) K A, A/2 5, 10, 15, 20, 25, 30 [1, 3, 5, . . . , 31] 1&1, 2&1, 4&1, 8&1 50/20, 60/20, 70/20, 80/20, 90/20, 95/20 5/40/55, 10/50/30, 20/30/50, 20/50/30 0.3, 1.7, 3.5 (5) where β̂i and β0 are respectively the expected value and the mean value of βi , and V(β̂i ) is the variance of βi . A standardized effect defined for a generic combination of factors can be introduced in the same way. Hypothesis testing is conducted on the value of β0 according to both the t-Student distribution of t, and the following null hypothesis: H: β0 = 0 . (6) If t (Eq. 5) is a great value the influence of the related factor i (or combination of factors) on the system’s response becomes critical. This 2k factorial analysis is the basis for a multi-level n ki factorial analysis where n i is the number of levels assumed by factor i: a large number of two-level factorial analyses were conducted for different couples of levels belonging to the values reported in Table 1. The total amount of two-level factorial analyses on the whole portfolio of simulated scenarios (on values of Table 1) is: n i 2 6 16 4 = × × × 2 2 2 2 2 i Table 1. Factorial analysis. Factors’ levels i=AREA i=DIMORD 6 4 × × 2 2 i=CURVE i=CAPACITY i=CLASSES i=SHAPE 3 × = 291600 . 2 (7) i=k Figure 3 present two final classifications of factors, and combinations of factors, which affect the system’s response for a fixed value of K (equal to 1.7) and AREA (according to the demonstration of its non-influence). This classification is obtained by analyzing several different two-level factorial studies; the values of t, expression 5, have been collected and cumulated. Figure 3a shows the mean value of t (standardize cumulated value) calculated in agreement with the following expression: Std_ti = ti, j (8) i, j where i identifies a factor or a combination of factors, j identifies a factorial study (i.e., an operating scenario), ti, j is calculated in accordance with Eq. 5. An absolute standardized cumulated value for each factor and combination of factors, collected by Fig. 3b, is introduced in accordance with the following expression: ti, j . Std_ti, ABS = (9) i, j Figure 3a and b shows that CURVE is the most critical factor; followed by CAPACITY. Then the third critical factor in terms of cumulated effects is DIMORD, while SHAPE is the third critical factor in terms of the measure of the absolute cumulated effect (i.e., in accordance with the absolute value of t). Figure 4 presents the simulated picking cycle times values obtained by the application of different batching procedures Fig. 3. (a,b) Pareto analysis of cumulative standardized and absolute cumulative standardized factor effects 772 Fig. 4. TSPtrip algorithm and batch procedures’ evaluation Fig. 5. TSPtrip algorithm and class based storage policy (batch x, batch y, batch xy and TSP) when the dimension of the pickinglist (i.e., DIMORD value) is 30. In particular routines batch x and batch y rearrange the picking requests in accordance with a non-decreasing value of the following times calculated from I/O point: x vx y vy (10) (11) where vx and v y are the velocity of the AS/R vehicle along x and y (as defined in Fig. 1); (x, y)i is the location of slot i. Batch xy rearrangement is based on the following Chebyshev distance: xi yi max , . (12) v v x y The effect of the batching rearrangement in terms of T _Normal reduction is not great, especially when the class based policy is not applied (Fig. 5). 773 7 Conclusions and further research The design and control of a part to picker OPS is one of the most critical issues in planning and optimization of modern industrial facilities. The paper presents significant results, obtained by using a dynamic multi-factorial analysis, capable of supporting the management of AS/RSs operating in flexible conditions. In particular, a classification of parameters’ effects on the system performance is presented. The analysis proves to be innovative for the large number of factors involved in accordance with the NP-hard complexity of the optimization problem. Specifically, the integrated use of simulation modeling and statistical factorial analysis proves to be effective. New in-depth studies on AS/RSs and picking activities (e.g., quantifying and minimizing global operating costs, changing the position of I/O area, etc.) are achieved. The measure of the effects of OPS design on the management of the other supply chain activities (e.g., packaging and loading processes, purchasing activity, geographical location of factories and warehousing facilities) is a critical issue that is not as yet sufficiently discussed in the literature. References 1. Lederer AL, Mirchandani DA, Sims K (2001) The search for strategic advantage from the world wide web. Int J Electron Commer 5(4):117–133 2. Gray AE, Karmarkar US, Seidmann A (1992) Design and operation of an order-consolidation warehouse: Models and application. Eur J Oper Res 58:14–36 3. Caron F, Marchet G, Perego A (2000) Optimal layout in low-level picker-to-part systems. Int J Prod Res 38:101–117 4. Manzini R, Gamberi M, Regattieri A, Persona A (2004) Framework for the designing a flexible cellular assembly system. Int J Prod Res 42(17):3505–3528 5. Ferrari E, Manzini R, Regattieri A, Persona A (2001) Application of Operations Research topics and techniques for problem solving in industrial plants. Proc 6th International Symposium on Operational Research in Slovenia, Slovenia, Preddvor, pp 37–43 6. Ferrari E, Gamberi M, Manzini R, Pareschi A, Persona A, Regattieri A (2003) An integrated optimization process for the production planning and control of a flexible manufacturing system. Proc Business and Industry Symposium, Advanced Simulation Technologies Conference, Orlando, FL, pp 79–84 7. Jawahar N, Aravindan P, Ponnambalam SG (1998) Optimal random storage allocation for an AS/RS in an FMS. Int J Adv Manuf Technol 14(2):116–132 8. Chetty Krishnaiah OV, Reddy Sarveswar M (2003) Genetic algorithms for studies on AS/RS integrated with machines. Int J Adv Manuf Technol 22(11–12):932–940 9. Chincholkar AK, Krishnaiah Chetty OV (1996) Simultaneous optimisation of control factors in automated storage and retrieval systems and FMS using stochastic coloured Petri nets and the Taguchi method. Int J Adv Manuf Technol 12(2):137–144 10. Sankar SS, Ponnanbalam SG, Rajendran C (2003) A multiobjective genetic algorithm for scheduling a flexible manufacturing system. Int J Adv Manuf Technol 22(3–4):229–236 11. Yoon CS, Sharp GP (1996) A structured procedure for analysis and design of order pick systems. IIE Trans 28:379–389 12. Larson TN, March H, Kusiak A (1997) A heuristic approach to warehouse layout with class based storage. IIE Trans 29:337–348 13. Van den Berg JP, Zijm WHM (1999) Models for warehouse management: Classification and examples. Int J Prod Econ 59:519–528 14. Vaughan TS, Petersen CG (1999) The effect of warehouse cross aisles on order picking efficiency. Int J Prod Res 37:881–897 15. Malmborg CJ, Altassan KM (1998) Analysis of storage assignment policies in less than unit load warehousing systems. Int J Prod Res 36:3459–3475 16. Rouwenhorst B, Reuter B, Stockrahm V, van Houtum GJ, Mantel RJ, Zijm WHM (2000) Warehouse design and control: framework and literature review. Eur J Oper Res 122:515–533 17. Roodbergen KJ, de Koster R (2001) Routing order pickers in a warehouse with middle aisle. Eur J Oper Res 133:32–43 18. Zerangue NF, Bodner DA, Govindaraj T, Karathur KN, McGinnis LF, Goetschalckx M, Sharp GP (2001) A process model of expertise in the design of warehousing and distribution systems. 2001 IEEE Int Conference on Systems, Man and Cybernetics 5:3403–3408 19. Kim BI, Graves RJ, Heragu SS, St.Onge A (2002) Intelligent agent modeling of an industrial warehousing problem. IIE Trans 34:601–612 20. Ferrari E, Gamberi M, Manzini R, Pareschi A, Persona A, Regattieri A (2002) Effectiveness of dynamic simulation supporting and optimizing design and management of warehouse facilities. Proc Business and Industry Symposium, Advanced Simulation Technologies Conference, San Diego, CA, pp 76–81 21. Manzini R, Ferrari E, Gamberi M, Pareschi A, Persona A, Regattieri A (2003) The optimization of a picker to product Order Picking System: a supporting decision tool based on a multi-parametric simulation approach. Proc 21st System Dynamics Conference, New York, NY 22. Manzini R, Pareschi A, Regattieri A, Basso A (2003) Design and control of flexible orderpicking systems (FOPS), (La progettazione e il controllo di sistemi Flessibili di Order Picking (OPS)). Proc 30th national conference ANIMP. Rapallo, Italy 23. Manzini R, Gamberi M, Regattieri A (2005) Design and control of a flexible order-picking systems (FOPS). J Manuf Tech Manage 16(1):18–35 24. Wascher G (2004) Order picking: A survey of planning problems and methods. Dyckhoff H, Lackes R, Reese J (eds) Supply chain management and reverse logistics. Springer, Berlin Heidelberg New York 25. Sharker BR, Babu PS (1995) Travel time models in automated storage/retrieval systems: A critical review. Int J Prod Econ 40:173–184 26. Malmborg CJ (2003) Interleaving dynamics in autonomous vehicle storage and retrieval systems. Int J Prod Res 41(5):1057–1069 27. Eldemir F, Graves RJ, Malmborg CJ (2003) A comparison of alternative conceptualising tools for automated storage and retrieval systems. Int J Prod Res 41(18):4517–4539 28. Pan CH, Wang CH (1996) A framework for the dual command cycle travel time model in automated warehousing systems. Int J Prod Res 34(8):2099–2117 29. Kim J, Seidmann A (1990) A framework for the exact evaluation of expected cycle times in automated storage systems with fullturnover item allocation and random service requests. Comput Ind Eng 18(4):601–612 30. Hwang H, Lee SB (1990) Travel-time models considering the operating characteristics of the storage and retrieval machine. Int J Prod Res 28(10):1779–1789 31. Lawler EL, Lenstra JK, Rinnooy Kan JK, Shmoys AHG (1993) Sequencing and scheduling: algorithms and complexity. In: Graves SC et al. (eds) Handbooks in OR & MS, vol 4, pp 445–522, Elsevier, Amsterdam 32. Lawler EL, Lenstra JK, Rinnooy Kan JK, Shmoys AHG (1985) The travelling salesman problem. Wiley, New York, NY 33. Caron F, Marchet G, Perego A (1998) Routing policies and COI-based storage policies in picker-to-part systems. Int J Prod Res 36:713–732 34. Brynzer H, Johansson MI (1996) Storage location assignement: using the product structure to reduce order picking times. Int J Prod Econ 46– 47:595–603 35. Tompkins JA, White JA, Bozer YA, Frazelle EH, Tanchoco JMA, and Trevino J (1996) Facilities planning. Wiley, New York 36. Malmborg CJ, Krishnakumar B (1989) Optimal storage assignment policies for multiaddress warehousing systems. IEEE Trans Syst, Man, Cybern 19:197–204 37. Park YH, Webster DB (1989) Design of class-based storage racks for minimizing travel time in a three-dimensional storage system. Int J Prod Res 9:1589–1601 38. Francis RL, McGinnis LF, White JA (1992) Facility layout and location: an analytical approach. Prentice-Hall, Englewood Cliffs, NJ 39. Van den Berg JP (1999) A literature survey on planning and control of warehousing systems. IIE Trans 31:751–762 774 40. Dallari F, Marchet G, Ruggeri R (2000) Optimisation of a manon-board automated storage/retrieval system. Integr Manuf Syst 11:87–93 41. Van den Berg JP, Gademann AJRM (2000) Simulation study of an automated storage/retrieval system. Int J Prod Res 38:1339–1356 42. Haskett JL (1963) Cube-per-order index–a key to warehouse stock location. Transp Distrib Manage 3:27–31 43. Kallina C, Lynn J (1976) Application of the cube-per-order index rule for stock location in distribution warehouse. Interfaces 7:37–46 44. Frazelle EH, Hackman ST, Platzman LK (1989) Improving order picking productivity through intelligent stock assignment planning. Proc Council of Logistics Management, pp 353–371 45. Gibson DR, Sharp GP (1992) Order batching procedures. Eur J Oper Res 58:57–67 46. Tang LC, Chew EP (1997) Order picking systems: batching and storage assignment atrategies. Comput Ind Eng 33:817–820 47. Hwang H, Beak W, Lee M (1988) Clustering algorithms for order picking in an automated storage and retrieval system. Int J Prod Res 26:189–201 48. Elsayed EA, Stern RG (1983) Computerized algorithm for order processing in automated warehousing systems. Int J Prod Res 27:1097–1114 49. De Koster MBM, van der Poort ES, Wolters M (1999) Efficient orderbatching methods in warehouse. Int J Prod Res 37:1479–1504 50. Papadimitriou C, Steiglitz K (1982) Combinatorial optimization. Prentice-Hall, Englewood Cliffs, NJ 51. Gerhardt M, Schuster H (1989) Cellular automation describing the formation of spatially ordered structures in chemical systems. Physica D: Nonlinear Phenom 36(3):209–221 52. Heragu SS, Mazacioglu B, Fuerst KD (1994) Meta-heuristic algorithms for the order picking problem. Int J Ind Eng-Appl Pract 1: 67–76 53. Makris PA, Giakoumakis IG (2003) k-Interchange heuristic as an optimization procedure for material handling applications. Appl Math Model 27:345–358 54. Brooks Automation (2001) AutoMod User’s Manual v 10.0. Chelmsford, MA 55. Montgomery DC (2001) Design and analysis of experiment. Wiley, New York