This article was downloaded by: [IIT Indian Institute of Technology - Mumbai]
On: 25 March 2014, At: 09:25
Publisher: Taylor & Francis Informa Ltd Registered in England and Wales Registered Number: 1072954 Registered office: Mortimer House, 37-41 Mortimer Street, London W1T 3JH, UK IIE Transactions Publication details, including instructions for authors and subscription information: http://www.tandfonline.com/loi/uiie20 A heuristic approach to warehouse layout with class- based storage T. NICK LARSON a , HEATHER MARCH b & ANDREW KUSIAK a a Intelligent Systems Laboratory, Department of Industrial Engineering , The University of Iowa , Iowa City, IA, 52242-1527, USA b Hon Industries, Inc , Muscatine, IA, 52761-7109, USA Published online: 30 May 2007. To cite this article: T. NICK LARSON , HEATHER MARCH & ANDREW KUSIAK (1997) A heuristic approach to warehouse layout with class-based storage, IIE Transactions, 29:4, 337-348, DOI: 10.1080/07408179708966339 To link to this article: http://dx.doi.org/10.1080/07408179708966339 PLEASE SCROLL DOWN FOR ARTICLE Taylor & Francis makes every effort to ensure the accuracy of all the information (the Content) contained in the publications on our platform. However, Taylor & Francis, our agents, and our licensors make no representations or warranties whatsoever as to the accuracy, completeness, or suitability for any purpose of the Content. Any opinions and views expressed in this publication are the opinions and views of the authors, and are not the views of or endorsed by Taylor & Francis. The accuracy of the Content should not be relied upon and should be independently verified with primary sources of information. Taylor and Francis shall not be liable for any losses, actions, claims, proceedings, demands, costs, expenses, damages, and other liabilities whatsoever or howsoever caused arising directly or indirectly in connection with, in relation to or arising out of the use of the Content. This article may be used for research, teaching, and private study purposes. Any substantial or systematic reproduction, redistribution, reselling, loan, sub-licensing, systematic supply, or distribution in any form to anyone is expressly forbidden. Terms & Conditions of access and use can be found at http:// www.tandfonline.com/page/terms-and-conditions lIE Transactions (1997) 29, 337-348 A heuristic approach to warehouse layout with class-based storage T. NICK LARSON}, HEATHER MARCH 2 and ANDREW KUSIAK l JIntelligent Systems Laboratory, Department of Industrial Engineering, The University of Iowa, Iowa City, IA 52242-1527, USA 2Hon Industries. Inc., Muscatine, IA 52761-7109, USA Received January 1995 and accepted November 1996 This paper presents a procedure for warehouse layout. It employs the principles ofclass-based storage to increase floor space utilization and decrease material handling. Three phases of the procedure are outlined: (1) determination of aisle layout and storage zone dimensions, (2) assignment of material to a storage medium, and (3) allocation of floor space. An industrial case study illustrates the procedure for a warehouse required to store 739 different products totaling more than 10000 pallets of material. An improved layout was developed and compared with the existing alternative. The proposed layout offered savings of more than 20000 square feet of floor space and an approximately 45% reduction in material handling distance. 1. Introduction Warehousing can be defined by three functions: (1) receiving goods from a source, (2) storing goods until they are needed by a customer (internal or external), and (3) retrieving the goods when requested. Storing material for an internal customer implies the need for work-in- process (WIP) storage, whereas storing goods for an external customer may imply the need for finished product storage. However, the functions of warehousing remain the same and, thus, successful warehouse layouts must, accomplish the following objectives, regardless of the material being stored [I]: (1) maximize the use of space; (2) maximize the use of equipment; (3) maximize the use of labor; (4) maximize accessibility to all items; (5) maximize protection of all items. Although the objectives of warehouse layout and opera- tion are easily recognized, warehouse layout problems are often complicated by large varieties of products needing storage, varying areas of required storage space, and drastic fluctuations in product demand. Optimal approaches to warehouse layout problems often consider a single objective (i.e., maximize floor space utilization) and/or provide a solution to a static problem. Some recent research in warehouse layout is discussed in. Section 2. Warehouse design problems are further complicated by _alternative storage methods and equipment. For ex- 0740-817X 1997 "lIE" . ample, material may be handled by pallet, case, or bulk items and the system may use lift trucks, conveyors, or automated storage and retrieval (AS/R) systems. To limit the scope of problems encountered in practice, this paper focuses on single-command lift truck storage and retrie- val of paJJets (i.e., one storage or one retrieval of a pallet is performed per trip between storage and the inputJ output point). This system is typical of many industries observed by the authors and should be of interest to a large audience. Furthermore, a clear definition of a particular type of warehousing system allows us to present a sensible heuristic procedure and illustrate it with a meaningful case study. As a result, we intend to provide a method for warehouse layout and operation that is attractive to practitioners in a variety of indus- tries. To achieve effective use of floor space and equipment, the layout must maximize floor space utilization and minimize travel time (and/or distance). The layout must also be robust to fluctuations in inventory level and demand. Section 3 presents a heuristic procedure for layout of a warehouse with pallet storage. The class- based storage approach considers average inventory le- vels, frequency of single command operations (i.e., throughput), storage medium (i.e., racks or floor stack- ing), aisle width, row depth (for floor stacking), and fluctuations in inventory and throughput. The case study in Section 4 illustrates how the procedure can drastically improve an' alternative ad hoc layout. Future research issues and concluding remarks are provided in Section 5. D o w n l o a d e d
b y
[ I I T
I n d i a n
I n s t i t u t e
o f
T e c h n o l o g y
-
M u m b a i ]
a t
0 9 : 2 5
2 5
M a r c h
2 0 1 4
338 2. Related research The literature on warehouse layout and operations is extensive. Three areas of research are particularly ap- plicable to the problem of lift truck pallet storage and retrieval: (1) stock location policy selection, (2) travel time and/or distance considerations, and (3) floor space allocation. This section provides a brief overview of some related research. Several stock location policies have been discussed in the literature, including dedicated storage, randomized storage, and class-based storage. Francis et al. [2] pro- vide an overview of these methods, as well as the shared storage location policy. An early paper by Ballou [3] discusses the roll of storage location policies in ware- house layout. Several authors have provided comparisons of storage location policies, including Harmatuck [4], Roll and Rosenblatt [5], and Wilson [6]. Eynan and Rosenblatt [7] applied class-based stock location to the study of automated storage and retrieval (AS/R) systems. Rosenwein [8] used cluster analysis and order picking data to establish class-based storage. The procedure presented in this paper establishes classes based on floor space utilization, frequency of single-command storage or retrieval, and average inventory level. Much of the research in the area of travel time and/or distance reduction has been motivated by the need to improve the efficiency of the order picking operation. The problem has frequently been formulated as a travel- ing-salesman model [9]. However, the problem consid- ered in this paper deals with single-command operations and there is no need for the development of a formal model. The heuristic procedure presented in Section 3 considers the rectilinear (or rectangular) distance from warehouse input point, to storage location, to output point. A variety of topics related to movement .of material handling devices are addressed in [10]. Floor space allocation in storage systems has received less attention in the literature.. Larson and Kusiak [11] presented a generalized transportation problem formula- tion for work-in-process (WIP) space allocation and illustrated the model with a case study. Space allocation within the facility is more generally treated as the facility layout problem [12], in which departments are arranged according to some cost objective and qualita- tive material flow and/or administrative constraints. In the procedure presented in this paper, storage regions are determined for each class and prioritized for the facility layout, The regions are sized in a way that reduces 'honeycombing' [13] and, as a result, the storage regions assume varying areas (see Section 3.2). Note that hon- eycombing occurs when storage locations are partly filled by items or lots that cannot be mixed or are blocked from access. A recent paper by Lacksonen [14] addresses facility layout problems with varying areas. Although recent results in this area are not inte- Larson et al. grated with the heuristic approach presented in this paper, the authors recognize this component of the problem as an opportunity for future research. In addition to the research cited in the areas of stock .location policies, travel time considerations, and floor space allocation, the reader may find the following publications of interest. Lanker [15] provides a detailed description of lift trucks as material-handling devices. The basics of pallet storage systems are discussed in a two-part series by Ross [16,17]. 3. A class-based storage procedure for warehouse layout This section presents a heuristic procedure for detennin- ing warehouse layout in a lift truck pallet storage and retrieval environment. The procedure is motivated by (1) the utilization of lift truck pallet storage in many in- dustries, and (2) the need for a practical method that provides an alternative to the ad hoc layouts that are often implemented in the absence of formal analysis and design. 3.1. Class-based storage rationale Warehouse operations are usually governed by one of three storage policies: randomized storage, dedicated storage, or class-based storage. In randomized storage, inventory is allocated a storage location based on the available space at the time of the storage job. The location is recorded for future retrieval jobs. The dedi- cated storage policy assigns material to a predetermined location based on throughput and storage requirements. All storage jobs of a given material are routed to the dedicated storage location. Class-based storage is a compromise between randomized and dedicated storage. Inventory is assigned a class based on some criteria (i.e., demand, product type, size) and each class is assigned a block of storage locations. Within each block of storage locations, material is stored randomly. A cost tradeoff is often realized between dedicated and randomized storage. Although dedicated storage typically reduces the material handling cost, more sto- rage space is required for each product to prevent over- flow. Alternatively, randomized storage usually requires less storage space; however, the material handling cost is often greater because there is no effort to store fast- moving material in the most desirable storage locations. This paper proposes class-based storage for lift truck pallet storage and retrieval warehouses. The objective is to group material with similar characteristics and allo- cate floor space on the basis of group priority. The classification procedure seeks to increase floor space utilization by limiting the effects of 'honeycombing' [13,16], and decrease material handling by ranking items based on throughput and storage space requirements. D o w n l o a d e d
b y
[ I I T
I n d i a n
I n s t i t u t e
o f
T e c h n o l o g y
-
M u m b a i ]
a t
0 9 : 2 5
2 5
M a r c h
2 0 1 4
Heuristic approach to warehouse layout 339 Each class is assigned to a storage region; however, within the storage region material is stored randomly. Random storage within the region provides flexibility to accommodate variations in inventory level for material assigned to the class. Thus, the proposed class-based storage policy increases the floor space utilization, de- creases the material handling cost, and increases flex- ibility. 3.2. The class-based procedure Before presenting the class-based procedure for ware- house layout, it is necessary to establish several assump- tions. Although these assumptions may appear to be. limiting, the method can be generalized to accommodate most practical scenarios, as illustrated in Section 4. Furthermore, enhancements of the procedure are an expected result of the future research issues discussed in Section 5. The assumptions are summarized below: 1. The facility is rectangular and the dimensions are known. 2. Material enters the facility via a single input point (i.e., a receiving dock). 3. Material exits the facility via a single output point (i.e., a shipping dock). 4. Each storage region consists of only one storage medium (i.e., rack or floor stack). . 5. All primary aisles are parallel and the same length. The procedure can accommodate more complex facil- ities by decomposing the layout problem into rectangular components with single input and output points. In fact, the physical characteristics of the facility are only of interest in determining the direction and length of an initial primary aisle, which is the basis for subsequent primary aisles. Classes are formed and storage regions are sized according to throughput and storage require- ments. The computational procedure in this paper includes the following three phases: Phase 1: determination of aisle layout and dimensions; Phase 2: assignment of material to a storage medium; and Phase 3: allocation of floor space. In Phase 1, the direction and length of the initial aisle are determined, as well as the other necessary dimensions. Each item is assigned a storage medium in Phase 2. CJasses are allocated floor space in Phase 3 and subsequent primary aisles are added to the layout. zone of some storage medium, such as racks or floor stacking. The rack dimensions or row depth determines the size of the storage zone. Aisles and zones of storage are added to the layout until all material that requires storage has been assigned a class and storage zone in the warehouse. It is assumed that material enters the warehouse at a single input point and exits the warehouse at a single output point. Therefore, the most desirable path of travel covers the shortest rectilinear distance from input point to output point. The shortest travel time path is deter- mined so that: (1) the rectilinear distance from input point to output point is minimized, (2) the number of comers on the path is minimized, and (3) the length of the longest arc on the path is maximized. Based on these objectives, the most desirable path is easily determined by examination. The direction and length of the longest arc on the path defines the direction and length of the initial primary aisle. Fig. 1 illustrates three possible paths through a facility with a single input point and a single output point: According to the objectives stated above, path (b) is obviously the best solution. When the input point and output point of the ware- house are at the same location, selection of an initial aisle is somewhat arbitrary. In general, if many classes are desired and it is expected that the corresponding storage zones are of varying sizes (i.e., varying row depths), aisles should be shorter and more frequent. Alternatively) if fewer classes are desired, a lesser number of longer aisles are desired. Fig. 2 shows primary aisle layouts for various input/output point con- figurations. To simplify the classification procedure, it is useful to represent dimensions in units of floor space) or slots, rather than some known unit of measurement (i.e., feet, meters). Let a slot represent the smal lest unit of assign- able floor space. Typically, the slot length 8 1 and slot depth Sd are equal to the pallet dimensions (plus the necessary clearance allowance). Recall that a zone de- fines an area that can be allocated for storage (i.e., racks or floor stacking). The procedure establishes dimensions for zone length L and zone depth Di, where k denotes the kth storage zone. Although the zone depth D k can vary P:nh (a) Fig. 1. Alternative paths through a rectangular warehouse. , Output point , : P:uh(c) Path (b) [.....-. I r ~ Input point Phase 1: Determine aisle layout and dimensions The direction of aisles must first be determined; how- ever, it is not necessary to determine the frequency (or spacing) of aisles a priori. The objective of the heuristic layout procedure is to determine classes that define different storage regions in the facility. It is assumed that each aisle serves the two storage regions on either side of the aisle. A region may also be referred to as a D o w n l o a d e d
b y
[ I I T
I n d i a n
I n s t i t u t e
o f
T e c h n o l o g y
-
M u m b a i ]
a t
0 9 : 2 5
2 5
M a r c h
2 0 1 4
340 Larson et al. \ \ I I
I I I 1 InpurJouput point (a) Output point 1: 1:1 11< I I <I ,I I I 1 I I I' Input point (b) 1'4 J I
Table 1. Floor space consumed and capacity of a storage location for various storage mediums Fig. 2. Primary aisle layouts for various input/output point con- figurations. between zones, it is assumed that the zone length L is the same for all storage zones. The aisle length is used to determine the zone length L. The zone length should exclude any allowances for secondary (or service) aisles at either end of the zone (and possibly even throughout the zone). Thus, storage zone k is said to be L slots long and Die slots deep. Fig. 3 illustrates storage zone and slot dimensions. The zone depth Dk is a function of the storage medium. For each item, the objective is to determine the storage medium (and corresponding zone depth) that minimizes the effects of 'honeycombing'. Honeycomb- ing occurs when storage locations are partly filled by items or lots that cannot be mixed or blocked from access. Two types of storage are typical of lift truck pallet storage and retrieval warehouses, rack and floor stacking. A combination of both types of storage may be used in the warehouse and, furthermore, there may be several different types of racks and/or varying row depths for floor stacking. Before assigning items to Storage zone 1. Identify a path from input point to output point such that: (a) the rectilinear distance from input point to output point is minimized; (b) the number of turns is minimized; (c) the length of the longest arc on the path is maximized. Note: If input/output point is at the same location, consider the number and size of desired classes and arbitrarily choose an initial primary aisle. 2. Establish an initial primary aisle along the longest arc of the path identified in (1). 3. Establish slot dimensions based on pallet dimen- sions, rack dimensions, and the necessary clear- ance allowances. 4. Determine the storage zone length L (in slots) based on the length of the initial primary aisle and any secondary aisles. 5. Identify storage mediums for consideration in Phase 2 (i.e., rack and/or n-deep floor stacking). Fig. 3. Storage zone and slot dimensions. classes, it is necessary to identify the types of storage to consider. For example, the warehouse may use four- tier one-deep rack storage for items that cannot be stacked, and three-deep, four-deep, and five-deep. floor stacking for items that can be stacked. Each item (or lot number) is assigned to a storage location in the ware- house. However, because several different types of sto- rage are used, the size and capacity of the storage location varies. Table 1 lists the floor space consumed and capacity of a storage location for each storage medium. The capacity of a storage location for floor stacking depends on the stack height h of the item stored. Phase 2 of the procedure determines the storage medium for each item such that storage location utiliza- tion increases. The following steps summarize the activities in Phase 1: ' I 3h 4h 5h 0.25 3 4 5 Floor space Capacity per consumed per storage location storage location (slots) (pallets) ... - - -
Aisles ...
- - - - - - ... - - Aisles - - - .. lnput Output point point (e) Four-tier one-deep rack Three-deep floor stack Four-deep floor stack Five-deep floor stack Storage medium D o w n l o a d e d
b y
[ I I T
I n d i a n
I n s t i t u t e
o f
T e c h n o l o g y
-
M u m b a i ]
a t
0 9 : 2 5
2 5
M a r c h
2 0 1 4
d lr F(sj) Heuristic approach to warehouse layout Phase 2.' Determine storage medium The proposed classification procedure is based on storage medium. The algorithm presented below determines the storage medium for an item (or a lot) based on average inventory level, honeycombing, fluctuations in inventory level, and stack height. Targeted at practitioners, the algorithm lends itself to implementation with spreadsheet applications and standard database functions. The fol- lowing data are needed for n items requiring storage: (1) the average inventory level Sj for j = 1, ... , n; (2) the stack height h j for j = 1, ... ,n; (3) the maximum row depth d max for floor stacking; (4) a percentage allowance for honeycombing and inventory fluctuation a; (5) a minimum stack height h min for floor stacking; (6) a minimum row depth d min for floor stacking. The procedure is specifically designed to determine the row depth that maximizes storage location utilization for floor stacking. If rack storage is available, it is necessary to consider a classification criterion for assigning an item to that medium. Two possible criteria are: (1) a mini- mum stack height for floor stacking (i.e., an item with a stack height h j less than an established minimum is assigned to rack storage), and (2) a minimum row depth for floor stacking (i.e., an item with a calculated row depth less than an established minimum is assigned to rack storage). The following notation is used in the row depth algorithm: n the number of items requiring storage; j index for item U= l , ... ,n); h j stack height for item U= I, ... ,n); Sj average inventory level for item U= 1, ... ,n); a percentage allowance for honeycombing; h min minimum stack height for floor stacking; d max maximum row depth for floor stacking; d min minimum row 'depth for floor stacking; N, number of storage locations required for item U= l , ... ,n); optimal row depth for item U= 1, ... , n); current number of storage locations required; a function of the rack design that provides the number of storage locations needed to store Sj pallets. Row depth algorithm: INPUT: n, hj U= 1, ... ,n), Sj U= 1, ... ,n), at h min , d max , d min OUTPUT: Nj, Step O. Let j = 1 Step J. IF h j < h min go to Step 5 ELSE go to Step 2 Step 2. Let Jf = r(l +a)sj/dmaxhjl 341 dj = r(l +O:)Sj/ N*hjl Step 3. IF 11j d min go to Step 4 ELSE go to Step 5 Step 4. Let N, = N* go to Step 6 Step 5. Let d j =0 N, =F(sj) Step 6. IF j < n let j = j + 1 go to Step 1 ELSE STOP. Step 1 checks the stack height to determine whether the item should be assigned to rack storage. The optimal row depth is determined in Step 2. Step 3 checks the optimal row depth to determine if the item should be assigned rack storage. Step 4 determines the number of storage locations needed for the optimal row depth. Step 5 determines the number of storage locations needed for rack storage. The counter is incremen- ted in Step 6. For each item, the row depth algorithm determines the minimum number of storage locations needed for floor stacking and then decreases the row depth until it is necessary to add a storage location. The minimum row depth that does not cause' an increase in the number of storage locations is assigned to the item. This heuristic accomplishes two objectives. First, minimizing the num- ber of storage locations needed to store items maximizes the number of different items that can be stored in a single zone. Thus, the most desirable storage zones can hold larger varieties of items. Secondly, minimizing the number of storage locations per item increases row depths. As a result, fewer aisles are needed and floor space utilization is increased. The following steps summarize the activities in Phase 2: 1. Collect the necessary data for n items requiring storage. 2. Apply the row depth algorithm to obtain: (a) the optimal row depth for each item d j ; (b) the number of storage locations needed to store each item N]. Phase 3: Allocate floor space The third phase of the procedure is to allocate floor space to classes on the basis of storage medium, required number of storage locations, and throughput. The throughput tj for j = 1, ... ,n items is measured by the number of pallets stored and retrieved in a given time period. For each storage medium, items (or lot numbers) are ranked according to the ratio of throughput tj to storage requirement N], denoted as R j . The items wi th the highest R j for a given storage medium are assigned to the D o w n l o a d e d
b y
[ I I T
I n d i a n
I n s t i t u t e
o f
T e c h n o l o g y
-
M u m b a i ]
a t
0 9 : 2 5
2 5
M a r c h
2 0 1 4
342 most desirable storage zones such that the storage zone capacity L is not exceeded. It is assumed that the length of the storage zone L, measured in storage slots (loca- tions), determines the capacity of the zone. The heuristic classification algorithm provides the same solution that would be obtained by solving the linear programming relaxation of the knapsack problem and rounding the fractional assignment to zero. The knapsack can hold only a given weight and, thus, the problem is to maximize the value of material in the knapsack while not exceeding the weight constraint. For the problem considered here, the value of storing a particular item is equal to its throughput tj and the weight is equal to the storage space N, required. For each storage medium, the algorithm solves successive knapsack problems until all items requiring that storage medium are assigned to a storage zone. The classification algorithm provides the classes C k (k = 1, ... ,m), the number of storage locations required for the class Ii, and the storage medium (i.e., row depth) Dk correspond- ing to class k. In addition to what was previously introduced, the following notation is used in the classi- fication algorithm: Ij throughput for item U= I, ... , n); N, storage requirement for item U= I, ... , n); R j ratio of throughput to storage requirement; 'j rank of item U= I, ... ,n); L length of storage zone in number 0' storage locations (i.e., capacity); k index for class; C k set of items in class (k = 1, ... ,m); lk number of storage locations required for class (k= L ... ,m); Dk storage medium (i.e., row depth) for class (k = I, ... , m). Classification algorithm: INPUT: n, t) (j = 1, ... ,n), N) U= 1, ... ,n), ~ (j = 1, ... In), L, d max OUTPUT: Rj; 'j, Ck (k = I, ... ,m), Dk (k = I, ... , m), I k (k = I, ... ,m) Step O. Let k = 0 Step 1. FOR} = 1 to n Let R, = IjlN, Step 2. FORj= 1 to n rank R j in descending order Let 'j = rank of item j Step 3. FOR I = 0 to d max Let k = k + 1 Ck= 0 lk =0 FOR i = 1 to n FORALL j such that 'j = i AND d j = I (a) IF t, +N j s L Let c, = c, Uj Larson et al. lk = lk +Nj k ~ (b) IF t, +N j > L Let k = k + 1 Ck=0 t, = 0 go to (a) Step 4. STOP Step 1 calculates the ratio of throughput to storage require- ment for each item. The items are ranked in descending order of the ratio in Step 2. Step 3 fills up the storage zones (i.e., classes) as needed for each row depth (i.e., storage medium). The classification algorithm assigns items to classes . such that the capacity of the corresponding storage zone is not exceeded. In fact, it is quite probable that each zone will contain a small number of unused storage locations. This is a result of not allowing items to be assigned to multiple classes and not attempting to solve the knapsack problem optimally. However, because a class-based storage policy is being implemented, it is not discouraging to have a small number of unused storage locations in each storage zone to accommodate fluctuations in inventory level. Recall that within each storage zone material is stored randomly; thus, a certain amount of 'overflow' space is desirable. The final step in the class-based layout procedure is to prioritize the classes for floor space allocation. Once again, items are ranked according to the ratio of through- put to storage requirement R]. The average class rank RZ is determined as follows: 1. Calculate R j for each item (j = 1, . . . , n). 2. Rank the items in descending order, where T J IS the rank of item j. 3. Calculate the class rank for each class k: R; = L Tj IICkl, Vj E Cit where ICkl is the cardinality of the set of items in class k. The layout of the warehouse is constructed by placing storage zones corresponding to the classes with the two lowest average class ranks adjacent to the primary aisle, which was determined in Phase 1. Additional storage zones and primary aisles are assigned to the layout according to the average class rank, the storage medium for each class, and the dimensions of the facility (see the case study in Section 4). The following steps summarize the activities in Phase 3: l , Apply the classification algorithm to determine the folJowing: (a) the class assignments for items (i.e., storage D o w n l o a d e d
b y
[ I I T
I n d i a n
I n s t i t u t e
o f
T e c h n o l o g y
-
M u m b a i ]
a t
0 9 : 2 5
2 5
M a r c h
2 0 1 4
Heuristic approach to warehouse layout 343 } 5-dl:eprow Column (storage location) Slot t i s l e ~ 1. Stack height (in pallets) for each brand-code hj. 2. An average inventory level for each brand-code Sj. 3. Throughput (i.e., number of cases stored and retrieved during a 14 week period) of each brand-code tj. 4. Pallet, rack, and aisle dimensions. 5. A CAD drawing of the current layout of the warehouse. } J-deep row Pallet /
A 'snapshot' of the current inventory was used to approx-
imate average inventory levels. Although the data were assumed to be representative of storage requirements, it is recognized that the static model used does not capture the dynamic behavior of inventory levels. However, the class- based heuristic approach to warehouse layout provides the flexibility needed to accommodate the dynamic behavior of inventory levels, given a good static approximation of inventory. To determine the desired measure for throughput, the flow of the customer warehouse process was considered. Once product is received, lift trucks transport it to storage one pallet at a time. Product remains in storage until it is needed in a case-pick station. At that time, an entire pallet of product is transported to case-pick, which was out of the scope of the study. Therefore, whole pallets are stored and retrieved one by one. Thus, throughput was measured in the number of pallets per week. This value was obtained by knowing the demand in case-pick for each brand-code (which came from the number of cases shipped per week) and assuming that steady-state levels of inventory existed for each brand- code. Two types of storage were considered in the ware- house, racks and floor stacking. Two racks (or two zones of floor storage) could be placed back to back, without being separated by an aisle. Racks had three tiers and a capacity of eight pallets per rack; the bottom tier could hold pallets stacked two high and the second and third tiers could not hold stacked palJets. Several important zone assignments) C k ; (b) the storage medium for each class (i.e., row depth) Di; (c) the number of storage locations required for each class h. 2. Prioritize classes according to average class rank s; 3. Construct layout by adding storage zones corre- sponding to classes and additional primary aisles. 4. Case study of a pallet storage and retrieval system The procedure presented in Section 3 was motivated by several warehouse systems observed by the authors in a variety of industries. In this section, an industrial case study is presented to illustrate the class-based storage procedure for warehouse layout. 4.1. Background The company of interest is a manufacturer of personal hygiene products with facilities throughout the USA. The warehouse studied is a stand-alone facility that is adja- cent to one of the company's regional manufacturing sites. Initially, the building was used for consignment warehouse operations, in which whole pallets were stored and shipped to internal and external customers. The proposed use of the facility was for customer ware- house operations, in which case picking and custom packaging would be performed for external customers. To convert the facility from a consignment warehouse to a customer warehouse, two options were being consid- ered. For the first option, the current zones and aisles could be used, which would minimize conversion costs. However, there was a strong feeling within the company that the current layout was not efficient for customer warehouse operations. The second option was to redesign the warehouse to make it more suitable for customer warehouse operations. A redesign of the current ware- house had to answer the following questions: 1. How much rack storage is needed? 2. What is a suitable row depth for floor stacking? 3. How many zones (and/or aisles) are needed? 4. In what direction should aisles run? 5. What storage policy should be used? 6. What products should be stored on racks? Storage was required for n = 739 products, referred to as brand-codes. Product was stored in cases, which were stored on pallets. The size of a case varied between brand-codes and, thus, the number of cases per pallet varied between brand-codes. The predetermined stack height also varied between brand-codes and ranged from two to five pallets high. The following information was collected for the case study: Fig. 4. Top view of aisles, pallets, rows, columns, and slots. D o w n l o a d e d
b y
[ I I T
I n d i a n
I n s t i t u t e
o f
T e c h n o l o g y
-
M u m b a i ]
a t
0 9 : 2 5
2 5
M a r c h
2 0 1 4
344 Larson et a]. Table 2. Important dimensions for the customer warehouse layout dimensions are listed in Table 2. Fig. 4 illustrates aisles, pallets, zones, columns, and slots. Aisle lengths and building dimensions were pre- determined. Two rows of any depth could be placed back to back without an aisle. A column could contain from one to seven slots (determined by the row depth) and was perpendicular to the adjacent aisle. Only one brand- code could be stored in each column and, thus, each column was considered a storage location. A slot is the smallest unit of floor space and can contain from one to five pallets depending on the stack height of the brand- code in the slot. 4.2. Application of the computational procedure In Phase 1 of the class-based layout procedure, the aisle layout and dimensions were to be determined. On the basis of the flow of material through the system, the initial primary aisle was determined by inspection. Three rectangular regions were identified for available storage and the length of proposed storage zones was determined to be L = 74 slots (i.e., storage locations). The most . desirable locations for storage were adjacent to the initial primary aisle. The next most desirable storage locations were in region 1, region 2, and region 3, respectively. The initial primary aisle and proposed Object Aisle Pallet Rack Column Slot Dimensions 14ft wide 4 ft by 4 ft 4 ft by 8 ft (3 tier) 4 ft wide, 4 ft to 28 ft deep 4 ft by 4 ft storage zones are shown in Fig. 5. In Phase 2 of the procedure, the storage medium was determined for each brand-code. The company wanted to provide two types of pallet storage, three-tier rack storage and floor stacking. Furthermore, floor stacking could range from three-deep rows to seven-deep rows. Row depth was a major concern of the company. The existing layout provided seven-deep floor stacking throughout the facility. However, many domain experts, ranging from lift truck operators to managers, felt that seven-deep rows throughout the warehouse would be very inefficient for the customer warehouse operations. The minimum row depth for floor stacking was deter- mined to be d m in = 3. Therefore, brand-codes with a calculated row depth less than three were assigned to rack storage. All brand-codes had stack heights ranging from two to six pallets, and no minimum stack height h min was established for floor stacking. The data were downloaded to a spreadsheet program for implementation of the row depth algorithm. For each of the 739 brand codes, the algorithm calculated the minimum row depth required to store the average in- ventory (plus an Q = 250/0 allowance for honeycombing) in the minimum number of storage locations. The fol- lowing is one iteration of the row depth algorithm for brand code j = 372: INPUT: n =739, h 372 = 3, 5372 = 8, a = 0.25, h min =2, d max = 7, d min = 3 OUTPUT: N 372 , d 372 Step J. Since h 372 " > h min go to Step 2 Step 2. Let N* = r{l +a)sj/dmaxhjl = r(l + 0.25)(8)/(7)(3)1 = I d 372 = r(l + Q)Sj/ N th j1 = f(1 +0.25)(8)/(1)(3)1 = 4 Step 3. Since d 372 > d min go to Step 4 Step 4. Let N372 = 1 I J I I Utilities I Available I I storage I I region 2 I I I II I J I Initial primary aisle Output Proposed storage zone I point I Available .. L =74 slots I storage (i.e .. storage locations) -.., Case-nick II region 3 I department II Available I J 1 storage I IL region I ...J Input point Fig. 5. Initial primary aisle and proposed storage zone dimensions. D o w n l o a d e d
b y
[ I I T
I n d i a n
I n s t i t u t e
o f
T e c h n o l o g y
-
M u m b a i ]
a t
0 9 : 2 5
2 5
M a r c h
2 0 1 4
Heuristic approach to warehouse layout 345 Table 3. Summary of classes for each storage medium Storage medium Number of classes Number of Percentage of Number ofpallets Percentage ofpallets brand-codes brand-codes 3-tier racks 2 417 57 1190 11 3-deep floor 2 98 13 803 8 4-deep floor 1 59 8 887 9 5-deep floor 2 72 10 1506 14 6-deep floor 2 39 5 1513 14 7-deep floor 4 54 7 4705 44 Total 13 739 100 10604 100 Table 5. Floor space requirements for alternative layouts Layout Rack storage Floor stack Aisle space Total space space storage space required (If) required (If) required (If) required (If) 4.4. Comparison of existing and proposed designs To justify the proposed warehouse layout, it was com- pared with the existing layout of the facility. As stated earlier, the existing layout consisted exclusively of seven-deep floor stacking and primary aisles were ar- ranged perpendicular to the proposed layout. Limited rack storage was available for 800 pallets. Furthermore, a random storage policy was suggested for the existing layout. The layouts were compared on the basis of floor space utilization and material handling effort. Table 5 shows the floor space required to store all 739 brand-codes in each warehouse. The current layout provided less rack 4.3. Computational results The resulting layout required 13 storage zones for the 13 classes of brand-codes. The priority and dimensions of each storage zone are listed in Table 4. A letter (i.e., A, B, C, etc.) was added to the name of each storage zone to distinguish between multiple zones (and classes) of the same storage medium. Zone dimensions are given in floor slots, where a floor slot is 4 ft by 4 ft. The storage zones were allocated floor space accord- ing to the priority listed in Table 4. The most desirable locations were adjacent to the initial primary aisle (see Fig. 5). After allocation of floor space to zones 'Rack A' and '3-deep A' zones were allocated floor space in region 1 and region 2, respectively. This method of floor space allocation reduces travel distance for single-com- mand storage or retrieval operations to higher priority classes. The proposed layout is shown in Fig. 6. 126736 104784 32816 29008 90720 71040 3200 4736 Current Proposed Table 4. Storage zones for warehouse layout Priority Zone name Zone length (L) Zone depth (DJ 1 RackA 74 slots (296 ft) 1 slot (4 ft) 2 3-deepA 74 slots 3 slots 3 5-deepA 74 slots 5 slots 4 7-deepA 74 slots 7 slots 5 6-deepA 74 slots 6 slots 6 4-deepA 74 slots 4 slots 7 7-deepB 74 slots 7 slots 8 7-deepC 74 slots 7 slots 9 3-deepB 74 slots 3 slots 10 5-deepB 74 slots 5 slots 11 6-deepB 74 slots 6 slots 12 7-deepD 74 slots 7 slots 13 RackB 74 slots 1 slot Phase 3 of the procedure allocated floor space to classes based on storage medium (i.e., row depth), required number of storage locations, and throughput. The ratio R j was calculated" for each brand-code, and the brand-codes were ranked in descending order for each storage medium. Classes were fanned for each storage medium by adding brand-codes until the number of storage locations required for the class Ik exceeded the storage zone capacity L = 74. A rack consumed two slots of floor space (equivalent to two storage locations of floor stacking) and provided six storage locations, each of which could contain one pallet. Thus, classes of rack storage were formed by adjusting L for the brand-codes assigned to rack storage (i.e., row depth clj < 3). The classification algorithm generated 13 classes of brand- codes, based on storage medium and the ratio R j . The average class rank Ric was determined for each of the 13 classes. Table 3 summarizes the classes for each storage medium. go to Step 6 Step 6. Since 372 < 739 Let j = j + I =372 + 1 = 373 go to Step 1 D o w n l o a d e d
b y
[ I I T
I n d i a n
I n s t i t u t e
o f
T e c h n o l o g y
-
M u m b a i ]
a t
0 9 : 2 5
2 5
M a r c h
2 0 1 4
346 Utilities I I I Avai lable I I storage I I region) I I I I . I L I Receiving dock ,.-------------.., I Excess floor space I ----- -.J 1IIIIIIJIIIIIllIIIflIIII1 11IlI11JIHII;'111111111Illl!ll:!!lill11111l1!:illliilll!11111111111H11!l1111!!lm:!llllfllllllll ease-pick department Larson et al. Fig. 6. Proposed layout of the warehouse. storage than the proposed layout. However, the proposed layout drastically decreased the effects of honeycomb- ing. As a result, the overall floor space requirement (including aisles) was much less for the proposed ware- house. . To compare material handling effort, a random mix of 20 brand-codes was selected. The average weekly throughput (i.e., the number of pallets stored and re- trieved) was obtained for each brand-code. For the current Jayout, the storage location for the brand-code was randomly generated, and the distance traveled to store and retrieve a pallet was measured. The distance traveled to store and retrieve a pallet in the proposed system was determined by the location of the storage zone corresponding to the brand-code's class. Table 6 summarizes the average weekly distance traveled to store and retrieve the 20 randomly selected Table 6. Average weekly distance traveled to store and retrieve 20 randomly selected brand-codes Brand-code Average throughput Current storage and Current average weekly Proposed storage and Proposed average weekly (pallets) retrieval distance (ft) distance (ft) retrieval distance (ft) distance (ft) I 1.67 746 1246 422 705 2 6.56 854 5602 364 2388 3 4.36 433 1888 309 1347 4 50.67 739 37445 249 12617 5 2.48 791 1962 389 965 6 1.64 841 1379 415 681 7 4.33 427 1849 359 1554 8 1.51 787 1 188 543 820 9 13.84 795 11003 321 4443 10 35.68 549 19588 399 14236 11 5.29 456 2412 340 1799 12 1.27 112 142 214 272 13 4.00 430 1720 278 1112 14 1.82 765 1392 521 948 15 22.97 467 10727 329 7557 16 18.93 481 9105 355 6720 17 4.13 441 1821 299 1235 18 11.59 427 4949 293 3396 19 1.34 577 773 509 682 20 3.00 915 2745 537 1611 Total 197.08 12033 118936 7445 65088 D o w n l o a d e d
b y
[ I I T
I n d i a n
I n s t i t u t e
o f
T e c h n o l o g y
-
M u m b a i ]
a t
0 9 : 2 5
2 5
M a r c h
2 0 1 4
Heuristic approach to warehouse layout 4.5. Discussion of results The proposed warehouse layout drastically improved on the current layout. The total floor space required to store all 739 brand-codes was decreased from 126736 ft2 to 104784 ft2, or 17%. The average weekly distance tra- veled by lift truck to store and retrieve pallets for the random mix of 20 brand-codes was decreased from 118936 ft to 65088 ft, or 45%. The floor space reduction was attributed to a decrease in honeycombing. Rather than floor stacking all material in seven-deep rows, a variety of row depths were provided in the proposed layout. The proposed layout also provided additional rack storage to hold the 57% of brand-codes that would be inefficiently stored in floor stacking. The material handling reduction was attributed to the use of the class-based storage policy. Brand-codes were assigned to classes according to throughput. Therefore, bra.nd-codes were assigned to higher priority classes which, In tum, were assigned to the most desir- able storage zones (i.e., the storage zones that minimized the travel distance from input point to output point). The proposed layout was presented to the company. Although the savings in floor space and material hand- ling were attractive to management, a concern was raised about the extra effort needed to manage a class-based system. Because a relatively small number of storage zones were identified, and within each zone material would be stored randomly, class-based storage would be relatively easy to implement. Each lift truck was already equipped with a bar-code scanner and database link. Thus, one alternative was to scan the lot and have the database return the class and storage zone of the pallet. Another alternative was to replace an unused character in the current brand-code with a character storage zone for the pallet. However, WIth the projected savings in floor space and material the company was ,in a position to justify facility and system conversion costs. 5. Conclusion This paper presented a class-based storage procedure for layout. The method is targeted at practitioners of single-command lift truck pallet storage and retrieval. Three phases of the procedure were outlined: Phase 1 d.eterrnination of aisle layout and storage zone Phase 2, assignment of material to a storage medium; and Phase 3, allocation of floor space. Two algorithms were provided for determining floor stack row depth and assigning items to classes. An industrial case study illustrated the procedure for a warehouse required to store 739 different products totaling more than 10000 pallets of material. A new layout was developed' and 347 compared to the existing alternative. The proposed lay- out offered savings of more than 20000 square feet of floor space and an approximately 45% reduction in material handling distance for a random mix of products. Several assumptions were stated for bounding the problem addressed in this paper. Although the method is easily generalized to accommodate most practical problems, the domain of application presented is some- what limited. Therefore, it is necessary to provide several suggestions for extending the procedure to more complex warehouse layout problems. The assumption of a rectangular warehouse is only necessary to simplify the selection of an initial primary aisle. If the warehouse is not rectangular, it should then be partitioned into the minimum number of rectangular sections. The initial primary aisle would satisfy the selection rules in Phase 1 and traverse each of the rectangular sections. Care must be taken to define accurately the length of each storage zone adjacent to the primary aisle. If the ware- house has multiple input and/or output points, the defini- tion of the initial primary aisle must, once again, be In this case, it is necessary to define primary Input and/or output points (most probably based on volume). The initial primary aisle would then connect the input point to the primary output point according to the rules defined in Phase 1 of the proce- dure. Secondary aisles would branch out to the remain- points, bisecting zones along the pnmary aisle. As classes are formed and floor space is allocated in Phase 3, the input and output points for each item being stored are considered. The classification algorithm can be extended to prevent the next item in the list from being added to a storage zone if the zone IS not on the path from its input point to its output These .extensions are relatively simple to apply WIthout loss of generality. The procedure can also be extended to accommodate varying pallet sizes. Recall that the row depth (i.e., storage medium) is determined independently of the pallet size. Therefore, the pallet size can be considered as an additional constraint in the classification problem. As items are allocated to storage zones, the could check to verify the appropriate pallet size. If the pallet size differs from that stored in the current zone, the item remains on the list for allocation to a different zone. Future research will elaborate on issues such as multi- ple . points, alternative' material handling devices, varying pallet sizes, stochastic inventory levels and throughput, and periodic reclassification of items. Nevertheless, the procedure provided in this paper pro- vides a foundation for a warehouse layout method that has displayed encouraging results in practice. D o w n l o a d e d
b y
[ I I T
I n d i a n
I n s t i t u t e
o f
T e c h n o l o g y
-
M u m b a i ]
a t
0 9 : 2 5
2 5
M a r c h
2 0 1 4
348 Acknowledgement This research has been partly funded by the national Science Foundation (Grant No. DDM-9215259) and Procter and Gamble Company. References. I. Tompkins, lA. (1988) The challenge of warehousing. In The Warehouse Management Handbook, Tompkins, l.A. and Smith, J.D. (eds), McGraw-Hili, New York, pp. 1-14. 2. Francis, R.L., McGmnis, L.F., Jr. and White, l.A. (1992) Facility Layout and Location: An Analytical Approach. Prentice-Hall, Englewood Cliffs, NJ. 3. Ballou, R.H. (1967) Improving the physical layout of merchandise in warehouses. Journal ofMarketing, 31(3), 60-64. 4. Harmatuck, OJ. (976) A comparison for two approaches to stock location. Logistics and Transportation Review, 12(4), 282-284. 5. Roll, Y. and Rosenblatt, M.J. (1983) Random vs. grouped storage policies and their effect on warehouse capacity. Material Flow, 1, 199. 6. Wilson, H.G. (1977) Order quantity, product popularity and the location of stock in warehouses. AJIE Transactions, 19(3), 230- 237. 7. Eynan, A. and Rosenblatt, M.J. (1994) Establishing zones in single- command class-based rectangular ASIRS. IJE Transactions, 26( I), 38-46. '8. Rosenwein, M.B. (1994) An application of cluster analysis to the problem of locating items within a warehouse. IJE Transactions, 26(1), 10t-l03. 9. Ratliff, H.D. and Rosenthal, A.S. (1983) Order picking in a rectangular warehouse: a solvable case of the traveling salesman problem. Operations Research, 31(3), 507-521. to. Tompkins, lA. and White, lA. (1984) Factluies Planning. John Wiley, New York. 11. Larson, T.N. and Kusiak, A. (1995) Work-in-process space alloca- tion: a model and an industrial application. IJE Transactions, 27(4), 497-506. t2. Kusiak, A. and Heragu, 5.5. (1987) The facility layout problem. Larson et al. European Journal of Operations Research, 29(3), 229-251. 13. Smith, J.D. and Peters, J.E. (1988) Warehouse space and layout planning. In The Warehouse Management Handbook, Tompkins, J.A. and-Smith, J.D. (eds), New York, pp. 91-114. 14. Lacksonen, T.A. (1994) Static and dynamic layout problems with varying areas. Journal ofthe Operational Research Society, 45( 1), 59-69. 15. Lanker, K.E. (1988) Lift trucks. In The Warehouse Management Handbook, Tompkins, J.A. and Smith, J.D. (eds), New York, pp. 321-371. 16. Ross, P. (1993) The basics of pallet storage systems: part I. Material Handling Engineering (April), 68-69. 17. Ross, P. (1993) The basics of pallet storage systems: part 2. Material Handling Engineering (May), 61-63. Biographies Nick Larson is a Research Assistant and Ph.D. candidate in the Department of Industrial Engineering at the University of Iowa, Iowa City, Iowa. He received a B.A. degree in physics from Central College) Pella, Iowa, in 1991 and an M.S. degree in industrial engineering from the University of Iowa in 1993. His research interests include intelligent systems, concurrent engineering, reengineering design and manufactur- ing processes, and modeling material handling systems. He is a member of lIE, INFORMS, and Alpha Pi Mu. Heather March is working with Hon Industries. lnc., In Muscatine, Iowa. At the time of the research project she was enrolled in the Graduate Program in Industrial Engineering at the University of Iowa. She is interested in manufacturing and distribution systems. Andrew Kusiak: is a Professor in the Department of Industrial Engineer- ing at the University of Iowa, Iowa City, Iowa. He is interested in engineering design and manufacturing. Dr. Kusiak has authored and edited numerous books and published research papers in journals sponsored by vanous societies, including AAAI, ASME, IEEE, liE, INFORMS, and SME. He serves on the editorial boards of a number of journals, edits book series, and is the Editor-in-Chief of The Journal of Intelligent Manufacturing. D o w n l o a d e d