Computers & Industrial Engineering 62 (2012) 527535

Contents lists available at SciVerse ScienceDirect

Computers & Industrial Engineering

journal homepage: www.elsevier.com/locate/caie

Storage assignment problem with travel distance and blocking considerations

for a picker-to-part order picking system
Jason Chao-Hsien Pan a, Po-Hsun Shih b,, Ming-Hung Wu a
a
Department of Business Administration, Takming University of Science and Technology, Taiwan, No. 56 Huanshan Road, Section 1, Taipei 11451, Taiwan, ROC
b
Department of Information Management, Vanung University, No.1 Van-Nung Road, Chung-Li, Tao-Yuan 320, Taiwan, ROC

a r t i c l e i n f o a b s t r a c t

Article history: Order picking is a key operation in managing a warehouse efciently. Most previous studies on picking only
Received 11 February 2011 considered single-picker operation; however, many pickers frequently work concurrently in the same
Received in revised form 31 October 2011 region. Since congestion may occur in such a multi-picker system, waiting time must be taken into account
Accepted 1 November 2011
together with travel time and distance when evaluating the efciency of picking operations. The picking
Available online 9 November 2011
model under investigation can be formulated as a queueing network, and a heuristic storage assignment
policy that considers both the travel time and the waiting time simultaneously by minimizing the average
Keywords:
order fulllment time is developed in the paper. An approximation method and a simulation model using
Storage assignment policy
Order picking
eM-plant software are presented to implement the proposed heuristic algorithm and to compare the mean
Warehouse management travel time for different storage assignment polices as well. The results indicate that the proposed heuristic
Multi-picker operations policy outperforms existing storage assignment policies in a multi-picker warehouse environment.
2011 Elsevier Ltd. All rights reserved.

1. Introduction tion from many pickers working in the same areas slows down the
processing (Gue, Meller, & Skufca, 2006). In practice, the distribu-
Order picking is the most costly part of the warehouse opera- tor congestion lead to decreasing worker productivity, more work
tions. According to Coyle, Bardi, and Langley (1996), 5075% of stress, and higher laborer turnover (De Koster & Yu, 2006). There-
the total operating costs of a warehouse can be attributed to the fore, an optimal assignment policy has to consider the trade-off be-
order-picking operation. A picker-to-part order picking system tween the travel distance reductions against the delay caused by
where items are stored on racks or bins and a picker travels to stor- congestion under such circumstances.
age locations by walking or driving along the aisles to retrieve all This paper addresses the storage assignment problem in a pick-
the items specied in a picking list is extensively used in industrial er-to-part system in which a picker travels to storage locations to
settings (De Koster, Le-Duc, & Roodbergen, 2007). Bartholdi and retrieve order items and investigates how congestion affects the
Hankman (2007) indicated that travel time is waste since it costs operations in a multi-picker warehouse. Such a picking model can
labor hours but does not add value. Warehouses have no choice be treated as a closed tandem queueing network with nite buffer,
but to improve their order fulllment operations through better and each aisle as a workstation. The exact analysis of this queueing
storage, batching, and routing strategies (Gagliardi, Ruiz, & Renaud, network is complicated and time-consuming, so the approximation
2008). Consequently, order picking has been a major study topic procedure proposed by Dallery and Frein (1989) is adopted and a
and most research have focused mainly on the development of tra- heuristic storage assignment policy is developed. In order to evalu-
vel time or distance models for various storage assignment, picking ate the performance of the proposed heuristic algorithm, this study
routing and order batching policies. compares the efciency of different storage policies by the simula-
Most of related studies considered only single-picker operations tion models. Moreover, this study analyzes the sensitivity of results
and are therefore adequate to measure order picking efciency by to various levels of input parameters including order size, number
travel distance as congestion never takes place in such systems. of racks and demand distribution of items.
However, multiple pickers frequently work concurrently in the
same region of a real-world warehouse system and congestion
inevitably occurs (Gu, Goetschalckx, & McGinnis, 2007). Especially, 2. Literature review
operations with large numbers of orders will nd that the conges-
A storage assignment policy (SAP) attempts to provide an
effective way of locating products in an order in a warehouse in or-
Corresponding author. Tel.: +886 3 4515811 280. der to reduce the order picking efforts and has been extensively
E-mail address: shih@vnu.edu.tw (P.-H. Shih). investigated in recent decades. It inuences almost all key

0360-8352/$ - see front matter 2011 Elsevier Ltd. All rights reserved.
doi:10.1016/j.cie.2011.11.001
528 J.C.-H. Pan et al. / Computers & Industrial Engineering 62 (2012) 527535

Nomenclature

m number of aisles in the warehouse ki arrival rate of order that enters the aisle i where the rate
N number of pickers in the warehouse follows a Poisson process, i = 1, . . ., m
k number of racks in an aisle li service rate of aisle i in which the service time is expo-
X throughput of the warehouse nentially distributed, i = 1,. . ., m
Wa the width of an aisle (including the width of racks) Bi the buffer capacity of aisle i, i = 1,. . ., m
Wi mean waiting time of a picker in aisle i, i = 1,. . ., m Li the average number of pickers in aisle i, i = 1,. . ., m
H the length of an aisle Qi the maximum number of pickers allowed in the aisle i,
Pij probability of the item in rack j of aisle i is in a picking i = 1, . . ., m
order, i = 1,. . ., m, j = 1,. . ., k ai(n) the probability that an arrival nds n pickers already in
Di the travel distance of a picker in aisle i, i = 1, 2,. . ., m aisle i, i = 1, . . ., m, n = 1, . . ., Qi
Ii picking time for all items of an order in aisle i, i = 1, 2,. . ., Pi(n) the steady-state probability of having n pickers in aisle i,
m i = 1,. . ., m, n = 1, . . .., Qi
Ti service time of picking in aisle i, i = 1, 2, . . ., m li(n) the service rate of aisle i when n pickers are present,
t average picking time of an item i = 1, . . ., m, n = 1, . . ., Qi
v average velocity of a rider pallet truck

performance indicator of a warehouse such as order picking item tions. Frazelle and Sharp (1989) developed a heuristic assignment
and cost (Muppani & Adil, 2008a). SAP is in the class of NP-hard procedure to minimize the total picking time for dependent de-
problems (Frazelle & Sharp, 1989); therefore, many heuristic meth- mand items.
ods for SAP have been proposed. Petersen and Gerald (2004) classi- Class-based storage policies classify items into classes and usu-
ed storage assignment policies into three broad categories: ally assign a xed area to each class in which items are randomly
randomized, volume-based and class-based. The random storage allocated in that area. Bynzer and Johansson (1996) provided a
policy is widely adopted in practice because it is straightforward storage assignment strategy derived from the product framework
to use, generally requires less space than other storage methods, and used variant characteristics as picking information to reduce
and improves the usage level of all picking aisles (Petersen & Ger- the picking time. Eynan and Rosenblatt (1994) presented a one-
ald, 2004). dimensional search algorithm to derive the boundaries for any de-
Volume-based storage policy attempts to distribute items sired number of classes in an automated warehouse, and found it
according to their demand rates or volumes to locations in a ware- can solve most practical problems effectively. Tang and Chew
house. Heskett (1963) presented the COI (cube-per-order index) (1997) consider batching and storage allocation strategies in a
policy which assigns item with the lowest ratio of the required stor- manual order picking system of small parts which processes high
age space to the order frequency to the location closest to the input/ volume of orders. Muppani and Adil (2008) modeled storage class
output (I/O) point. Caron, Marchet, and Perego (1998) evaluated formation considering area reduction, handling cost and storage
and compared different routing policies with storage by COI, and space cost and to come up with an efcient optimal solution pro-
concluded that the traversal policy in COI-based storage systems cedure. Chan and Chan (2011) presented a simulation study of a
outperforms the basic return policy for a range of picking densities real case regarding storage assignment problem of a manual-pick
and the COI-based ABC curves most frequently found in real-world and multi-level rack warehouse. They indicate that the key to
applications. Caron, Marchet, and Perego (2000) developed an ana- effective implementation of a storage assignment system is to
lytical approach to the layout design of the picking area in picker- match the types of warehouse storage system and the variety of
to-part systems by COI-based and random storage policies and items in the customer order. Ronald, Peter, and Sunderesh (2007)
the analytical results demonstrate that layout preference is appar- developed the Order Oriented Slotting (OOS) strategy which stores
ently strongly inuenced by decisions concerning the adoption of a the items in such a way that the total time needed to pick all the
COI-based storage policy. Kallina and Lynn (1976) presented some order is minimized.
practical conclusions collected from experience in using the COI Schwarz, Graves, and Hausman (1978) studied the performance
rule to help design warehouse layouts. Hwang, Oh, and Cha of an automatic warehouse system adopting a storage policy based
(2003) proposed density-turnover index (DTI) rule with the aim on item picking frequency. Hsieh and Tsai (2001) presented a bill of
of minimizing the energy consumption of the material handlers material (BOM) oriented class-based storage assignment method
working in a picker-to-part warehouse system. The DTI rule is for an automated storage/retrieval system (AS/RS) and found it to
based on the weight, space requirement, demand rate and the effec- be more efcient than random storage assignment. Larson, March,
tive distance measure of storage location. Hwang et al. (2003) com- and Kusiak (1997) presented a class-based storage warehouse lay-
pared the DTI rule with the COI rule and demonstrated that the DTI out procedure to increase oor space usage and decrease material
rule is substantially better than COI in terms of human safety with handling operations. Jane and Laih (2005) suggested a clustering
some sacrice of throughput. Muppani and Adil (2008b) presented algorithm for item assignment in a synchronized zone order pick-
to solve an integer programming model for class formation and ing system in which all zones process the same order simulta-
storage assignment that considers all possible product combina- neously by balancing the workload among all pickers to reduce
tions, storage-space cost and order-picking cost. the time required to fulll each requested order. Hausman, Sch-
Jarvis and McDowell (1991) studied the necessary and sufcient warz, and Graves (1976) proposed an optimal storage assignment
conditions of optimally for locating a product in a class of symmet- algorithm and suggested that turnover assignment policies can de-
ric warehouses, and found that simply assign the most frequently crease crane travel times in automatic warehousing systems more
picked items to the nearest aisles does not necessarily minimize effectively than random policy.
the average travel distance if the aisles are not symmetrical. In Petersen and Gerald (2004) analyzed the degree the three pro-
addition, they developed a heuristic based on these optimal condi- cess decisions of picking, storage and routing affect order picker
 J.C.-H. Pan et al. / Computers & Industrial Engineering 62 (2012) 527535 529

CAPS ceeds through the aisles to perform picking operations with the
Buffer
Platform Back aisle aid of CAPS and takes a transversal route by entering an aisle con-
taining a product in the order from one end and leaving from the
other (Frazelle & Sharp, 1989; Jarvis & McDowell, 1991 and Ruben
k-1 k
& Jacobs, 1999). A single-order picker picks only the items on a sin-
gle order in each tour in the facility. Moreover, with the platform
on the right side of the warehouse, a pickers tour is assumed to
commence at the input point and end at the platform. Products
are delivered by rider pallet trucks, which are loaded at the plat-
H form. The rider pallet truck has unlimited capacity of carrying
items.
The number of pickers in a warehouse is limited. Each picker
Aisle Aisle Aisle picks by order according to the information displayed by the CAPS
3 1 4 2 m
which shows all the picking for a picker in the aisle that he/she is
1 2
entering. Each picker proceeds by not exceeding the one in front of
him/her and returns to the input point after an order is nished.
Wa Therefore, a picking model can be regarded as a closed tandem
Front aisle queueing network, and each aisle as a server station. The time
Input Point spent by a picker in one aisle is regarded as the service time based
on the contents of the picking list, as shown Fig. 2.
Fig. 1. A warehouse model. Though following the optimal rule proposed by Jarvis and
McDowell (1991) will minimize the total distance traveled for sin-
gle-picker environment, collecting the most frequently picked
travel, and indicated that both class-based and volume-based stor- items into a few aisles may increase the possibility of congestion
age policies provides almost the same level of savings as batching between order pickers if there is more than one picker working
policies, but are less sensitive to the order size. Gibson and Sharp concurrently in the same region and consequently decreases the
(1992) showed that signicant reductions in traveling distance operation efciency. Therefore, the waiting time and the total dis-
can be achieved by locating high picking frequency items close to tance traveled must be simultaneously considered in order to min-
the I/O point. Ruben and Jacobs (1999) presented batch construc- imize the average order completion time in a multi-picker system.
tion heuristics under three storage assignment policies and found
that batches construction method and storage space assignment 3.3. Approximation of waiting time for picking system
method have signicantly impact on order retrieval efforts in a
warehouse. The exact analysis of average time is usually complicated in a
tandem queue network (Bolch, Greiner, De Meer, & Trivedi,
2006). Hence, this study proposes to calculate the mean waiting
3. Problem statement and solution methods time by approximation method. Dallery and Frein (1989) pre-
sented an approximation technique based on decomposition for
3.1. Warehouse system the analysis of closed queueing networks with blocking. Although
the fundamental concept of the decomposition method is quite
Consider a multi-picker warehouse limited to a two-dimen- general, Dallery and Frein (1989) restricted their attention to
sional storage allocation. The warehouse has m aisles where each closed tandem queueing networks with exponential service times,
aisle has two sides and there is a rack on each side. All these m in which the rst station has innite capacity. Furthermore, their
aisles are assumed to have an equal number of k racks. Only one method considers the transfer blocking mechanism where a server
item type can be stored in a rack and the number of racks is as- is blocked if the destination queue of a picker is full after a picking
sumed to be exactly the same as the number of item types. Dummy service has completed. The queueing network under study is
items with no demand can be created to guarantee this relation- decomposed into a set of individual queues aiming to approximate
ship if fewer item types exist than racks. the ow in the queueing network buffers. Furthermore, each indi-
For information support system, a computer aided picking sys- vidual queue is modeled as an M/M/1/K queue with load-depen-
tem (CAPS) or electronic paperless pick-to-light system can be uti- dent service rate. The further constraint on the number of
lized in practice in a picker-to-part system. The CAPS automatically pickers because of the closed model is considered by prescribing
guides order pickers quickly to the pick locations, and shows the that the sum of the average number of pickers in the individual
exact numbers of Storage Keeping Unit (SKU) to be picked. The queues equals the closed network population.
advantages of the CAPS include effectively improving the picking Consider a closed tandem queueing network comprising m sta-
productivity by 50% or more, and reducing the picking task error tions and N pickers, where each station comprises an aisle and a
as well (Jane & Laih, 2005). Additionally, CAPS simplies the train- server. The service time at station i is exponentially distributed
ing for pickers, thus cuts down the training cost. with rate li. The rst station has an innite buffer capacity, while
Finite buffers are allocated between aisles. If a picker intends to others have nite buffer capacitates. After completion at station i, a
enter an aisle and nds another one present, he/she would wait in picker must then proceed to station (i + 1), i = 1, 2, . . ., m  1. After
the buffer until the aisle is clear. Moreover, the buffer in front of completion at station m, a picker returns to station 1. Notably, sta-
the rst aisle is assumed to be innite; hence, all pickers can wait tion i has one preceding station (i  1), and one succeeding station
in this buffer until their turns to enter the intended aisles. (i + 1). By convention, station (i  1) is called station m if i = 1, and
station (i + 1) is called station 1 if i = m.
3.2. Operation assumption and order attribution
3.3.1. Determination of the service rate
The routing policy is assumed to be sequential one-way travel, The service rate is the reciprocal of mean service time. In
as shown in Fig. 1. A picker starts from the input point and pro- the warehouse under investigation, the expected service time
530 J.C.-H. Pan et al. / Computers & Industrial Engineering 62 (2012) 527535

Fig. 2. Flow diagram of a closed queueing network for the picking operations.

of each aisle includes the traveling time and the picking time. 3.3.2. An approximate algorithm for the queueing network
To compute the traveling time, the distance traveled in aisle i, Dallery and Frein (1989) suggested to decomposing the queue-
Di, can be divided into the distances traveled in aisle i, qi and ing network into a set of m single queues with nite capacity. Let E
across aisle i. Since a picker must cross every aisle, the be the set of stations i such that Bi > N. Note that station 1 always
distance traveled across is equal to the aisle width Wa for belongs to E, and Qi can be dened as:
any aisle i.
Q i N for any i 2 E
Each aisle contains k racks and each rack contains a different 6
item. The probability of picking each item in an order affects the Q i Bi 1 for any i R E
expected picking distance in this aisle (qi), and once an aisle is en- For the picking operation system under study, the service rate
tered it must be completely traversed. Therefore, the expected va- of each aisle is dependent on the number of pickers at the aisle.
lue of qi is: Hence, the system ts the load-dependent rate condition stated
in Dallery and Frein (1989). Moreover, the labor level stays un-
Eqi H  Probability of entering aisle i changed during a batch operation since every picker returns to
H  f1  1  Pi1  1  Pi2      1  P ik1  1  Pik g the input point after nishing a trip and starts a new order. Conse-
quently, the constraint that the sum of the average number of pick-
!
Yk ers in the individual queues equals the closed network population
H  1  1  P ij for i 1; 2; . . . ; m 1 is satised as the number of pickers in the warehouse system is al-
j1 ways constant.
Therefore, aisle i can be modeled as an M/M/1/Qi queue with
The expected distance traveled within aisle i (Di) can then be calcu- load-dependent service rates. The queueing characteristics of aisle
lated as: i can be derived from the parameters using the standard M/M/1/K
! formulas as:
Y
k
!
EDi W a Eqi W a H  1  1  Pij for i Qi Y
X n
ki
j1 pi 0 1= 1 7
n1 j1
lj n
1; 2; . . . ; m 2
Y
n
ki
The average time of picking all items within aisle i, E(Ii), can
pi n pi 0 8
j1
lj n
also be found from the picking probability of each item in an
order as: ki X=1  pi Q i 9
ai n pi n=1  pi Q i 10
X
k
EIi t Pij 3
j1
Following Dallery and Frein (1989), the values of X and Li can be
found by the following algorithm.
By (6) and (7), the mean service time of picking within aisle i, E(Ti),
can be calculated as: Step 0: Set Xmin = 0 and Xmax = min{l1, l2, . . ., lM} as the lower
ET i EDi =v EIi bound and upper bound, respectively, for the throughput X.
!!, Step 1: Select a current value of X within the bounds.
Y
k X
k
Wa H  1  1Pij v t Pij 4
j1 j1
Set i = m.

Consequently, the service rate of aisle i, li, is given by Step 2: Calculate li(n), for n = 1,. . ., Qi:
li 1=ET i
, !!, ! If any i 1 R E and n  N  Bi1 ; then li n
Y
k X
k
1 Wa H  1  1  Pij v t Pij 5 li li1
; 11
j1 j1 li1 li ai1 Bi1
 J.C.-H. Pan et al. / Computers & Industrial Engineering 62 (2012) 527535 531

else use Eq. (5). This study assumes that an aisle only contains one picker.
Determine the arrival rate ki using Eq. (9). Therefore, the workload of N or more aisles must be balanced to re-
If i E, calculate ai(Bi) using Eq. (10). duce the waiting time when there are N pickers in warehouse. For
instance, if a warehouse with m aisles has N pickers, then the
Step 3: if i > 1, set i to i  1, and go to step 2. workload needs to be balanced from aisle m to aisle (m  N) while
Step 4: Calculate the average queue length Li, for i = 1,. . ., m. other aisles will keep the optimal assignment proposed by Jarvis
and McDowell (1991) in order to preserve the minimum travel
X
N distance as much as possible. This study partitions the aisles from
Li npi n for any i 2 E; 12 m to (m  N) into k/2 zones as shown in Fig. 3. The proposed heu-
n0 ristic policy considers the picking frequency and the average wait-
B
X i 1
ing time in an aisle to assign each item to the rack according to the
Li npi n Bi pi Bi pi Bi 1 for any i 1 R E; 13 zone sequence. Since pickers could be congested in the warehouse,
n0
the workload of more than N aisles should be balanced in order
X
m
L Li 14 improvement the throughput of the warehouse. Fig. 4 outlines
i1 the main aspects of the proposed heuristic assignment method
and the procedure proceeds as follows.
Step 5: If L is close enough to N, then stop;
Step 0: Use the rule provided by Jarvis and McDowell (1991) as
else if L > N, set Xmax = X, otherwise set Xmin = X.
follows:
Go to step 1.
In step 1, a current value of X can be selected as X = (Xmin + Xmax)/
2.
Moreover, the mean waiting time can be obtained by

W i Li =X 15
Zone k / 2

3.4. A Heuristic assignment procedure . .

According to the above description, a mathematical model for . .
the storage assignment problem (SAP) under study can be formu- . .
lated as follows:
!! Zone 2
X
m X
k Y
k X
r
Minimize Wi Wa H 1  1  Pij xijz =
i1 j1 j1 z1
Zone 1
X
k X r Aisle 1 Aisle 2 Aisle 3 Aisle 4 Aisle 5
v t Pij xijz 16
j1 z1
Fig. 3. The zoning of a warehouse with two pickers.
Subject to

X
m X
k
xijz  1; z 1; . . . ; r: 17
i1 j1
k X
X r
xijz  k; i 1; . . . ; m: 18
j1 z1

xijz 0 or 1; i 1; . . . ; m; j 1; . . . ; k: and z 1; . . . ; r 19

where xij = 1, if item z is assigned to rack j of aisle i; and 0, other-

wise. The constraint set (17) ensures that item z is located to rack
j of aisle i only. Assign item z in rack j of aisle i incurs the mean wait-
ing time, the mean travel time and the mean picking time. The ob-
ject function (16) is to minimize the expected time of orders
fullled in one tour for each picker. The constraint set (18) is used
to restrict the sum of the number of items assigned to an aisle does
not exceed the number of racks in the aisle. Since SAP is in the class
of NP-hard problem (Frazelle & Sharp, 1989), this paper develops a
heuristic method accordingly.
The purpose of the proposed heuristic policy is to minimize the
average time of an order completed and it not only takes the travel
distance of an order nished, but also the congestion consideration
into account. This heuristic initially applies Jarvis and McDowell
optimal rule (1991) to allocate each item to the storage location
and obtains the optimal assignment for the single-picker environ-
ment. Then, to relieve the congestion for multi-picker operation,
the items assigned in the aisle are adjusted according to the num-
ber of pickers in the warehouse. Fig. 4. The proposed heuristic procedure.
532 J.C.-H. Pan et al. / Computers & Industrial Engineering 62 (2012) 527535

Arrange the items into m groups according to the picking fre- Table 2
Throughput of the three storage policies on different number of pickers.
quencies by assigning the most frequently picked item groups to
the most distant aisle, the second most frequently picked items No. of pickers (N) J&M Random Heuristic
to the second most distant aisle and so on until the least frequently 1 0.0601 0.0441 0.0601
picked item groups are assigned to the nearest aisle. 2 0.0970 0.0770 0.0981
3 0.1227 0.1076 0.1296
4 0.1344 0.1286 0.1424
Step 1: Set n = N, X0 = 0. 5 0.1430 0.1497 0.1581
Step 2: Divide aisle i into k/2 zones, i = m, m  1, . . ., m  n. 6 0.1479 0.1686 0.1726
Step 3: Rearrange the items within k/2 zones of aisle i according 7 0.1508 0.1851 0.1861
to descending picking frequency, i = m, m  1, . . ., m  n. 8 0.1524 0.1987 0.1997
9 0.1534 0.2104 0.2123
Step 4: Set j = 1 and calculate Wi using Eq. (19) i = 1, 2, 3, . . ., m.
10 0.1540 0.2214 0.2232
Step 5: At iteration j, assign the most frequently picked item to
the racks of aisle i with the smallest Wi, and the second most
frequently picked item group to the next smallest aisle and so
on until all of the racks of zone j are completely assigned. Since the order-picking model resembles a closed tandem
Step 6: Re-calculate Wi, i = 1, 2, 3, . . ., m. queue, the procedure presented in Section 2.3.2 is used to calculate
the throughput of the three assignment policies for different num-
If all racks are assigned, then estimate throughput X using the ber of pickers, as listed Table 2 where J&M stands for Jarvis and
computational algorithm and go to step 7; else set j = j + 1 and McDowell method. Performance on the mean time measure is pre-
return to step 5. sented in Table 3, and can be obtained as the reciprocal of the
throughput listed in Table 2. A comparison on the throughputs or
Step 7: if n < m and X > X0 , mean order fulllment times of the three storage policies tested
under all levels of number of pickers indicates that the heuristic
set n = n + 1, X = X and record the location of all items assigned is superior to the other two storage assignment policies. Since
and go to step 2; else stop assignment procedure and restore the the storage assignment policy of Jarvis and McDowell method
previous locations of all items assigned. (1991) groups the items with the highest demand in the front
aisles, most pickers have to wait to enter these aisles. Although
the average order fulllment time reduces as the number of pick-
4. Experimental design and simulation model ers rises, the degree of reduction is restricted by the increasing
blocking effect. Jarvis and McDowell method (1991) has the worst
4.1. Description of the warehouse conguration performance among all examined policies when the warehouse has
more than four pickers, as shown in Table 3.
A warehouse layout has 10 picking aisles with front and back Fig. 5 shows the results on total distance measure. The random
cross-aisles. The picking aisles are two-sided (Petersen & Gerald, storage policy achieves a more balanced workload in each aisle
2004). The warehouse installs a CAPS and there are 680 items lo- than the Jarvis and McDowell rule (1991), but with a longer travel
cated in these 10 aisles (Jane & Laih, 2005). Since only one item distance. When a warehouse has only a few pickers, the blocking in
can be stored in a rack, each aisle has 68 racks. The input point each aisle is not signicant. Excessively balanced workloads of
of the warehouse is located in its lower left part. Moreover, the de- each aisle increases the total travel distance. Therefore, the perfor-
mand for the items is based on an 80/20 distribution, that is, 20% of mance using random storage policy is less efcient than the Jarvis
the items account for 80% of the picking activities. Each picking and McDowell rule (1991), when there are fewer than four pickers.
probability is generated randomly. The order size, the number of Total distance increases as the number of pickers rises for the
items picked in a tour, is assumed as 20. The travel rate of proposed heuristic policy since the number of balanced aisles is
0.75 m per second and picking time per item of 20 s of a picker based on the number of pickers in the warehouse. However, the
are assumed to be constant. The picking time includes all item han- heuristic method performs better than the other two storage poli-
dling and administration activities. Table 1 shows the warehouse cies. Fig. 6 illustrates the improvement percentage of the heuristic
specications. policy over other two assignment policies by using heuristic ap-
proach as the basis. The improvement in performance of heuristic
4.2. Computation results method over the Jarvis and McDowell rule (1991) increases and

This section compares the Jarvis and McDowell (1991), random

and heuristic assignment methods, and attempts to study their ef-
ciencies using numerical data analysis with approximation method.

Table 1
Related data of the warehouse.

Parameter Specication
Number of aisles (m) 10
Order size 20
Number of racks in an aisle (k) 68
Aisle length (H) 75 m
Aisle width (Wa) 3m
Average velocity of truck (v) 0.75 m/s
Time of picking an item (t) 20 s
Maximum queue capacity of aisles First queue is innite, the others equal
110 two
Fig. 5. Total traveling distances for the three storage policies on different number of
pickers.
 J.C.-H. Pan et al. / Computers & Industrial Engineering 62 (2012) 527535 533

20

15
Random
J&M

Minute
Heurisitc
10

0
1 2 3 4 5 6 7 8 9 10
Number of pickers
Fig. 6. The improvement percentage of the heuristic policy over the other two
Fig. 7. Mean order fulllment time for the three storage policies on different
polices on different number of pickers. Note:  Indicates signicant improvements
number of pickers.
(P < 0.05, number of samples is 30).

over the random assignment decreases as more order pickers are The various numbers of orders fullled in an experiment are
added to the warehouse. As noted earlier, the level of congestion implemented in the simulation. Table 4 shows output data from
is highest under the Jarvis and McDowell rule (1991) while the the simulation. The results reveal that the heuristic policy is better
warehouse has 10 pickers. Thus, the heuristic assignment yields
the highest improvement over Jarvis and McDowell rule (1991)
by 31.7%. Additionally, the highest improvement percentage of Table 4
heuristic policy over random assignment policy is 26.71% when Total time (minutes) to fulll all orders time for the three storage policies.
the warehouse only has one picker. Fig. 6 also indicates signicant
No. of Storage No. of pickers (N)
improvements of the proposed method in comparison to Jarvis and orders policies
McDowell rule (1991) on three or more pickers. Fig. 6 also reveals 1 2 3 4 5 6 7 8 9 10
that the proposed heuristic assignment is obviously the best policy 100 Random 1790 914 632 482 405 351 316 293 280 274
of the three between three and six pickers. J&M 1398 743 581 553 553 553 553 553 553 553
Being an approximation method, the results generated may be Heuristic 1398 728 527 427 364 328 304 281 267 247
prone to have deviations with the exact method. Moreover, the 300 Random 5391 2760 1881 1452 1200 1038 921 852 813 798
environment of a real warehouse is quite complicated and the J&M 4236 2286 1752 1680 1680 1680 1680 1680 1680 1680
approximation process is time-consuming. Another way to evalu- Heuristic 4236 2232 1572 1263 1077 960 879 816 765 705
ate the performance of the three assignment policies is by simula- 500 Random 9080 4640 3160 2435 2010 1735 1545 1425 1360 1330
tion. Simulation can be used quickly and easily to analyze various J&M 7125 3815 2945 2845 2845 2845 2845 2845 2845 2845
Heuristic 7125 3730 2650 2105 1805 1600 1455 1355 1270 1165
aspects of order picking systems in a warehouse.
800 Random 14544 7432 5080 3888 3208 2776 2464 2272 2176 2136
J&M 11432 6112 4712 4568 4568 4568 4568 4568 4568 4568
Heuristic 11432 5984 4232 3368 2880 2560 2328 2152 2024 1864

4.3. The simulation model and results

An object-oriented software model, eM-plant (2000), is used to Table 5

The improvement percentage of the heuristic policy over the other two polices on
build the warehouse model and implement the proposed assign-
different order size and number of pickers.
ment storage policy. eM-plant has been used in various environ-
ments, such as manufacturing, warehousing, material handling Order Storage No. of pickers (N)
size policies
processes and semiconductor manufacturing, and has proven to
1 (%) 2 (%) 3 (%) 4 (%) 5 (%) 6 (%) 7 (%) 8 (%) 9 (%) 10
be a highly effective simulation program. (%)
20 Random 21.66 19.67 16.46 12.82 10.17 7.40 4.96 4.61 6.11 11.68
Table 3 J&M 0.00 2.18 10.01 24.92 35.83 42.83 47.66 51.46 54.32 57.91
Mean order fulllment time (minutes) for the three storage policies on different 25 Random 19.19 17.81 15.00 11.36 8.93 5.99 4.72 3.56 6.02 11.51
number of pickers. J&M 0.00 1.82 9.35 24.84 36.35 43.77 49.23 52.77 55.79 59.30

No. of pickers (N) J&M Random Heuristic 30 Random 16.37 14.90 12.09 8.75 6.78 4.37 3.62 2.61 5.96 11.40
Jarvis 0.00 1.23 8.12 23.91 36.25 43.94 49.99 53.52 56.93 60.42
1 16.6267 22.6875 16.6267
2 10.3052 12.9904 10.1985 35 Random 14.12 12.74 9.93 6.98 5.26 3.52 2.53 2.02 5.14 11.22
3 8.1505 9.2976 7.7190 J&M 0.00 1.24 7.10 23.74 36.66 44.73 50.66 54.56 57.90 61.45
4 7.4414 7.7763 7.0220 40 Random 12.07 10.59 8.31 5.71 4.10 3.19 2.25 1.71 5.44 11.23
5 6.9943 6.6799 6.3235 J&M 0.00 1.08 6.68 24.21 37.25 45.60 51.52 55.44 59.03 62.25
6 6.7605 5.9299 5.7925
7 6.6312 5.4015 5.3721 45 Random 9.72 8.11 6.53 4.04 2.41 1.60 1.21 1.45 4.97 11.03
8 6.5606 5.0329 5.0070 J&M 0.00 0.96 6.53 24.37 37.61 46.09 52.13 56.34 59.79 62.94
9 6.5187 4.7536 4.7100 50 Random 8.03 6.85 5.50 3.31 1.81 0.93 0.63 1.25 4.91 11.27
10 6.4933 4.5170 4.4798 J&M 0.00 0.73 6.51 24.86 38.15 46.87 52.84 57.06 60.38 63.65
534 J.C.-H. Pan et al. / Computers & Industrial Engineering 62 (2012) 527535

Table 6
The improvement percentage of the heuristic policy over the other two polices on different demand distribution of the items and number of pickers.

Demand distribution of the items Storage policies No. of pickers (N)

1 (%) 2 (%) 3 (%) 4 (%) 5 (%) 6 (%) 7 (%) 8 (%) 9 (%) 10 (%)
80/20 Random 19.19 17.81 15.00 11.36 8.93 5.99 4.72 3.56 6.02 11.51
J&M 0.00 1.82 9.35 24.84 36.35 43.77 49.23 52.77 55.79 59.30
60/20 Random 8.94 7.44 6.43 4.06 3.18 2.77 1.68 2.61 3.08 6.41
J&M 0.00 0.60 1.85 6.53 17.37 28.54 36.21 41.87 45.24 49.02
40/20 Random 1.81 1.53 2.07 0.39 1.29 0.60 0.09 1.22 0.71 1.54
J&M 0.00 0.58 1.16 1.11 1.23 2.95 6.38 12.26 16.64 20.95

Table 7
The improvement percentage of heuristic policy over other two polices with 1000 racks in the warehouse.

Order size Storage policies No. of pickers (N)

1 (%) 2 (%) 3 (%) 4 (%) 5 (%) 6 (%) 7 (%) 8 (%) 9 (%) 10 (%)
30 Random 21.57 18.09 13.93 10.48 6.73 3.76 0.55 1.95 2.33 3.15
J&M 0.00 0.34 0.45 8.04 20.47 30.04 35.99 42.39 46.40 48.48
35 Random 18.47 15.51 12.47 8.73 5.47 3.60 0.85 1.93 2.30 3.55
J&M 0.00 0.15 0.44 10.02 23.33 33.15 39.68 45.70 49.50 51.56
40 Random 15.94 13.26 9.55 7.54 4.28 2.18 1.55 2.11 2.34 3.47
J&M 0.00 0.16 0.78 11.65 25.02 35.15 42.68 48.13 51.77 53.82
45 Random 14.04 11.55 8.24 5.98 3.13 1.61 1.77 2.09 2.51 3.31
J&M 0.00 0.25 1.38 13.91 27.39 37.38 44.99 50.31 53.75 55.57

than the other two storage assignments under all numbers of congestion while the rest of aisles hold the items of low demand
orders and pickers tested. Fig. 7 shows the mean order fulllment rates. Consequently, the pickers not only avoid traversing some
times for the three storage policies with different number of pick- aisles but also have shorter waiting times.
ers. Although the mean order fulllment time is short under all The probability that pickers pass each aisle goes up as the order
workforce level compared to the approximation method, the sim- size increases. Under this circumstance, the waiting times and the
ulation results also indicate that the heuristic model has the best travel distances of all pickers in each aisle rise as well. Since the
performance under any workforce level. workload of all aisles under the heuristic is more balanced than
The simulation results also indicate that the average order ful- the other two, the blocking effect is thus reduced. Consequently,
llment time can not be improved under the Jarvis and McDowell each picker enters the aisle as early as possible to pick the items
rule (1991) with more than four pickers. Since items with high de- in the orders.
mand rates are concentrated on the rst three aisles, the order To study the impact of demand distributions, a sensitivity anal-
picking efciency cannot be increased by adding more pickers. ysis was performed to evaluate 60/20 and 40/20 distributions of
Moreover, due to increasing congestion level, the marginal produc- the items in addition to the 80/20 distribution and the results
tivity of additional order pickers decreases for Jarvis and McDowell are shown in Table 6. When the picking frequency of each item
rule (1991). is overly closed, no storage policies cause the apparent difference
of performance on order picking in a warehouse. Table 6 indicates
4.4. Sensitivity analysis that the 40/20 item distribution yields minimal improvements un-
der any conditions. A more evenly distributed demand item patter
Further analysis was conducted to study the sensitivity of the tends to benets the random assignment as expected.
results to various levels of input parameters including order size Moreover, Table 7 shows the results for the number of racks in-
and demand distribution of items. Table 5 shows the effect of the creased to 1000. The importance of travel distance rises as the
order size on performance. As the order size increases, the length of each aisle expands by 40 m. The Jarvis and McDowell rule
improvement percentage of the heuristic policy over the random (1991) performed better at 1000 racks than at 680 racks, since this
policy falls except with 10 pickers in the warehouse. This outcome rule only considers the total distance traveled and the length of
is reasonable because the waiting time of the pickers in the aisle each aisle added is thus advantageous. However, the proposed
rises as the density of items picked from each aisle increases. With heuristic rule is still superior to the others two when the conges-
the increased congestion, the workload of each aisle needs to be tion effect is considered in the warehouse.
further balanced. When the congestion exceeds a predetermined
level, the waiting time has more impact than the travel distance 5. Conclusions
for the order picking efciency, even without considering the inu-
ence of the travel distance. This study presents a storage assignment policy that improves
The proposed heuristic is clearly better than random policy the order picking efciency with the consideration of multiple
when the order size is small. Because the random policy distributes pickers in a warehouse. Most previous related studies have focused
all items into each aisle equally, pickers need to go through the on reducing travel distance; however, congestion may occur as
most number of aisles. Although the proposed heuristics assigns more order pickers occupy a xed storage space. Thus, the waiting
the items with high demand rates to some specic aisles, it strives time caused by aisle blocking could signicantly affect the picking
to balance the workload of these aisles according to the level of process efciency.
 J.C.-H. Pan et al. / Computers & Industrial Engineering 62 (2012) 527535 535

Although the random storage policy generally uses the entire Eynan, A., & Rosenblatt, M. J. (1994). Establishing zones in single-command class-
based rectangular AS/RS. IIE Transactions, 26, 3846.
picking area and consequently reduces worker congestion impact,
Frazelle, E. H., & Sharp, G. P. (1989). Correlated assignment strategy can improve
doing so may generate unnecessary travel distance. The Jarvis and any order-picking operation. Industrial Engineering, 21, 3337.
McDowell rule (1991) can minimize the travel distance, but may Gagliardi, J. P., Ruiz, A., & Renaud, J. (2008). Space allocation and stock
increase congestion with multiple pickers in a warehouse. There- replenishment synchronization in a distribution center. International Journal
of Production Economics, 115, 1927.
fore, the heuristic approach presented by this study simultaneously Gibson, D. R., & Sharp, G. P. (1992). Order batching procedures. European Journal of
considers travel distance and waiting time to reduce the order fulll- Operational Research, 58, 5767.
ment time. The proposed heuristic algorithm can determine the Gu, J., Goetschalckx, M., & McGinnis, L. F. (2007). Research on warehouse operation:
A comprehensive review. European Journal of Operational Research, 177, 121.
number of aisles needs to be balanced in order to reduce blocking Gue, K. R., Meller, R., & Skufca, J. D. (2006). The effects of pick density on order
time and travel distance according to the level of workforce. picking areas with narrow aisles. IIE Transactions, 38, 859868.
Hausman, W. H., Schwarz, L. B., & Graves, S. C. (1976). Optimal storage assignment
in automatic warehouse system. Management Science, 22, 629638.
Acknowledgment Heskett, J. L. (1963). Cube-Per-Order Index A key to warehouse stock location.
Transportation and distribution Management, 3, 2731.
This research was supported by the National Science Council of Hsieh, S., & Tsai, K. C. (2001). A BOM oriented class-based storage assignment in an
automated storage/retrieval system. International Journal of Advanced
Taiwan, Republic of China, Contract NSC99-2221-E-147-001-MY3. Manufacturing Technology, 17, 683691.
Hwang, H., Oh, Y. H., & Cha, C. N. (2003). A stock location rule for a low level picker-
References to-part system. Engineering Optimization, 35, 285295.
Jane, C. C., & Laih, Y. W. (2005). A clustering algorithm for item assignment in a
synchronized zone order picking system. European Journal of Operational
Bartholdi, J. J., & Hankman, S. T. (2007). Warehouse and distribution science.
Research, 166, 489496.
<http://www.tli.gatech.edu/research/warehousing/>.
Jarvis, J. M., & McDowell, E. D. (1991). Optimal product layout in an order picking
Bolch, G., Greiner, S., De Meer, H., & Trivedi, K. S. (2006). Queueing networks and
warehouse. IIE Transactions, 23, 93102.
markov chains: Modeling and performance evaluation with computer science
Kallina, C., & Lynn, J. (1976). Application of the Cube-Per-Order index rule for stock
application. New York: John Wiley.
location in distribution warehouse. Interfaces, 7, 3746.
Bynzer, H., & Johansson, M. I. (1996). Storage location assignment: Using the
Larson, T. N., March, H., & Kusiak, A. (1997). A heuristic approach to warehouse
product structure to reduce order picking times. International Journal of
layout with class-based storage. IIE Transactions, 29, 337348.
Production Economics, 595603.
Muppani, V. R., & Adil, G. K. (2008a). A branch and bound algorithm for class based
Caron, F., Marchet, G., & Perego, A. (1998). Routing policies and COI-based storage
storage location assignment. European Journal of Operational Research, 189,
policies in picker-to-part systems. International Journal of Production Economics,
492507.
36, 713732.
Muppani, V. R., & Adil, G. K. (2008b). Efcient formation of storage classes for
Caron, F., Marchet, G., & Perego, A. (2000). Optimal layout in low-level picker-to-
warehouse storage location assignment: Simulated annealing approach. Omega,
part systems. International Journal of Production Economics, 38, 101117.
36, 609618.
Chan, F. T. S., & Chan, H. K. (2011). Improving the productivity of order picking of a
Petersen, C. G., & Gerald, A. (2004). A comparison of picking, storage, and routing
manual-pick and multi-level rack distribution warehouse through the
policies in manual order picking. International Journal of Production Economics,
implementation of class-based storage. Expert Systems with Applications, 38,
92, 1119.
26862700.
Ronald, J. M., Peter, C. S., & Sunderesh, S. H. (2007). Order oriented slotting: A new
Coyle, J. J., Bardi, E. J., & Langley, C. J. (1996). The management of business logistics. St.
assignment strategy for warehouses. European Journal of Industrial Engineering,
Paul, MN: West.
1, 301316.
Dallery, Y., & Frein, Y. (1989). A decomposition method for the approximate analysis
Ruben, R. A., & Jacobs, F. R. (1999). Batch construction heuristics and storage
of closed queueing network with blocking Queueing networks with blocking,
assignment strategies for walk/ride and pick systems. Management Science, 45,
North Holland.
575596.
De Koster, R., Le-Duc, T., & Roodbergen, K. J. (2007). Design and control of
Schwarz, L. B., Graves, S. C., & Hausman, W. H. (1978). Scheduling policies for
warehouse order picking: A literature review. European Journal of Operational
automatic warehousing systems: Simulation results. AIIE Transactions, 10,
Research, 182, 481501.
260270.
De Koster, R., & Yu, M. (2006). Makespan minimization at Aalsmeer ower auction.
Tang, L. C., & Chew, E. P. (1997). Order picking systems: Batching and storage
In 9th International material handling research colloquium, Salt Lake City, Utah.
assignment strategies. Computers & Industrial Engineering, 33, 817820.
eM-plant (2000). Object manual version 4.6 Teconmatix Technologies.