You are on page 1of 18

int. j. prod. res.

, 15 september 2004,
vol. 42, no. 18, 3873–3889

An evaluation of routing policies for order-picking operations in low-level


picker-to-part system

H. HWANGy*, Y. H. OHz and Y. K. LEEz

This paper evaluates the performance of three routing policies in the order-
picking process, i.e. return, traversal, and midpoint policy. It is assumed that items
are assigned to storage locations on the basis of the cube-per-order index (COI)
rule in a low-level picker-to-part warehousing system. First, for the three policies,
analytical models are developed for the total expected travel distance of the order
picker considering the number of the stocking aisles is even or odd. Then the
developed models are compared with simulation results to show the validity.
Finally, the performance of the three policies is examined by varying the param-
eter value of the COI-based ABC curve, number of picks in the pick list, and ratio
of the length to the width of the warehouse.

1. Introduction
In low-level picker-to-part systems, the picker performs the retrieval of items
on the pick list from their storage locations to satisfy customer orders. This process
is known as the order-picking process and constitutes 65% of the total operating
costs for a typical warehouse (Coyle et al. 1996). The efficiency of the order-picking
process is closely related with operating policies on batching (grouping of customer
orders for pick lists), routing (sequencing the retrieval orders in a pick list), and
storage (assignment of storage space to inventory items). This paper focuses on
routing policies assuming that both pick list (a list of items and their locations
to be visited in a tour) and storage locations of the items are given.
Several routing heuristics including optimal algorithms were presented for
a rectangular warehouse (Ratliff and Rosenthal 1983, Goetschalcks and Ratliff
1988, Hall 1993). For real world applications, Goetschalcks and Ratliff (1988) intro-
duced the traversal routing policy and Hall (1993) developed the largest gap policy.
Storage policies based on the ratio of the required storage space to the order
frequency of the item, i.e. cube-per-order index (COI), were suggested (Heskett
1963, Kallina and Lynn 1976) and several researchers proved that the COI policy
is an efficient assignment methodology (Gibson and Sharp 1992, Kallina and Lynn
1976, Jarvis and McDowell 1991). Focusing on traversal policy, Jarvis and
McDowell (1991) proposed a popularity-based method of assigning items.
Petersen II and Schmenner (1999) evaluated routing heuristics in popularity-based
storage policies using simulations. Roodbergen and Koster (2001) constructed
a dynamic programming algorithm for calculating order-picking tours of minimal
length in warehouses. Caron et al. (1998, 2000) developed analytical models for

yDepartment of Industrial Engineering, Korea Advanced Institute of Science and


Technology (KAIST), 373-1, Guseong-dong, Yuseong-gu, Daejeon, Korea.
zDepartment of Industrial Engineering, KAIST.
*To whom correspondence should be addressed. e-mail: harkhwang@kaist.ac.kr

International Journal of Production Research ISSN 0020–7543 print/ISSN 1366–588X online # 2004 Taylor & Francis Ltd
http://www.tandf.co.uk/journals
DOI: 10.1080/00207540410001696339
3874 H. Hwang et al.

W
Back aisle wc

L l
... ...

w X Front aisle wc
P/D

Figure 1. Warehouse layout (with an even number of aisles).

traversal and return policies under the COI-based ABC curve and then provided
a framework for layout design. In their studies, a rectangular layout was considered
with two sections, picking aisles running parallel to the warehouse front-end where
the pick-up/deposit (P/D) point is located and an even number of aisles per section.
The objective of this paper is to evaluate and compare the performance of three
routing policies, namely, return, traversal, and midpoint policies in a warehouse as
shown in figure 1. It is assumed that items are stored in the warehouse on the basis
of the cube-per-order (COI) rule. Analytical models are developed for the expected
travel distance of order picker under each policy and then we investigate the effects
of the system parameters, i.e. the demand skewness, the size of a pick list, and the
ratio of length to width of the warehouse. The remainder of this paper is organized
as follows. The next section introduces the configuration of the warehouse, basic
picking operations, and assumptions that are necessary for this study. Section 3
develops the expected travel distance models and Section 4 discusses the computa-
tional results of the analytical models under various values of the system parameters.
Finally, conclusions appear in Section 5.

2. Background and assumptions of the expected travel distance models


When a layout is developed for storage and warehousing, those layouts are
sought that will maximize space utilization as well as maximize the level of service
provided (Tompkins and White 1984). Thus item and order-picker accessibility
becomes a critical factor in developing the layout. Aisles must be planned to be
wide enough for efficient material handling and be located so that every face of
a storage rack has aisle access. All main aisles should be oriented so that the majority
of materials are stored along the long axis of the storage area. The warehouse layout
considered in this study is shown in figure 1, which satisfies the above requirements
and has been adopted by many research papers in the literature (Ratliff et al. 1983,
Gibson et al. 1992, Hall 1993, Rosenwein 1996, Petersen 1997, and Koster et al.
1999). It is rectangular with the stocking aisles running perpendicular to the P/D
point and with cross aisles (front and back) only at the aisle ends. L and W denote
the length and width of the warehouse, respectively. There are P stocking aisles
where the P/D point is located in the centre of the front aisle. The length of stocking
racks is l and the width of stocking area including a stocking aisle and both sides
of the storage area is w. Starting from the P/D point, a picker performs retrieval
Routing policies for order-picking operations 3875

x x
P/D P/D
(1) Return (2) Traversal

x
P/D
(3) Midpoint

Figure 2. Routing policies.

of items on the pick list and takes the items back to the P/D point for packing and
shipment. In this study, it is assumed that the horizontal travel distance within
stocking aisles is negligible.
Among various routing policies in the literature, return, traversal, and midpoint
policies are prevalent in real world applications due to the convenience of their imple-
mentation. Figure 2 shows how these policies work when the shaded areas denote
pick locations.
(1) Return policy
A picker enters those (picking) aisles containing picks from the front cross
aisle only, performs the pick, and then returns to the front cross aisle.
(2) Traversal policy
A picker starts at the P/D point, traverses each aisle fully by entering it from
one cross aisle and exiting it at the other cross aisle, and finally ends at the
P/D point.
(3) Midpoint policy
Each length of each picking aisle is divided into two equal parts at its
midpoint. A picker entering a picking aisle from a cross aisle traverses
only up to its midpoint to perform the picks in that section of the aisle
and returns to the same cross aisle. Depending on the items to be picked
in an aisle, a picker may have to enter the same picking aisle again from the
other cross aisle.
The cube-per-order index (COI) rule proposed by Heskett (1963) utilizes both
the turnover rate and size of item in determining its storage location. The COI of an
item is simply the quotient of the space requirement and the order frequency of the
item. In the COI rule, items with the lowest COI values are assigned to the locations
3876 H. Hwang et al.

closest to the warehouse P/D point. From the well-known ‘ABC’ phenomena for
inventories, the COI-based ABC curve can be expressed by the following function
(Bender 1979 and Caron 1998):
ð1 þ SÞx
FðxÞ ¼ , 0  x  1, FðxÞ  0, S  0 and S þ x 6¼ 0: ð1Þ
Sþx
In equation (1), x indicates the ratio of required storage space to total storage space,
corresponding to the items whose order frequency represents a fraction F(x) of total
warehouse activity. S is the shaping factor, which determines the skewness of the
curve. As the curve skews more, S becomes smaller.
There are several different ways of implementing the COI-based storage strategy
such as across-aisle, within-aisle, and perimeter storage sequence (strategy) (Peterson
et al. 1999). They are illustrated in figure 3 with the high (light grey), medium
(medium grey), and low (dark grey) value of COI items. For return policy, across-
aisle strategy is known to perform better because it can reduce the probability of
retrieving distant picks from the entering point of the aisle resulting in the minimum
expected within-aisle travel distance. In the case of traversal policy, it is desirable
to reduce the number of aisles visited so that within-aisle strategy is preferable. It is
known that perimeter strategy is effective with midpoint policy, which can be easily
inferred from the characteristics of the policy. In the next section, the expected travel
distance models are developed for return, traversal, and midpoint policy adopting
across-aisle, within-aisle, and perimeter strategy, respectively. For the models, the
time required to pick items is assumed to be independent of the location height. Thus
this study only considers the travel distance according to the length of picker’s walk
on the floor. Also, it is assumed that the distance travelled crossing a picking aisle
from one side to the other is negligible compared to the distance travelled along the
centre-line of the aisle. As a result, minimizing the total expected order picking time
is equivalent to minimizing the summation of the expected travel distance along the
centre-line within each aisle and the expected travel distance across aisles. Although

P/D P/D
(1) Across-aisle (2) Within-aisle

P/D
(3) Perimeter

Figure 3. COI-based storage policy.


Routing policies for order-picking operations 3877

the real order-picking operation is performed in discrete rack openings, all the
approaches in the study are developed under continuous approximation.

3. Model developments
Notation
P the number of stocking aisles
N the number of picks in a tour (order size)
L the length of the warehouse (m)
W the width of the warehouse (m)
l the length of a stocking aisle (m)
w the width of a stocking aisle (including shelves’ width) (m)
wC the width of the front and back aisles (m)
D the expected total travel distance (m)
DW the expected within-aisles travel distance (m)
DA the expected across-aisles travel distance (m)

3.1. Travel model for the return policy with across-aisle strategy
The expected total travel distance (DR) under return policy is the sum of the
expected within-aisles travel distance (DR W ) and the expected across-aisles travel
distance (DRA ).
First, the within aisles component of DR will be determined following the
approaches of Caron et al. (1998, 2000). Under the across-aisle strategy, every stock-
ing aisle has an equal probability to be visited and thus the COI-based ABC curve in
equation (1) becomes identical in each aisle. Since each stocking aisle has the same
probability to contain picks, the probability that a pick exists in a specific aisle is 1/P
and the probability that a specific aisle does not have any pick becomes {1  (1/P)}N.
Then the expected number of picking aisles (v) can be expressed as
"   #
1 N
v¼ 1 1 P: ð2Þ
P
The expected number of picks per visited aisle (n) is:
N
n¼ : ð3Þ
v
The return route distance within an aisle having n picks depends on the farthest
pick location. These n pick locations can be treated as random variables whose
cumulative distribution function is given as F(x). Then the farthest location in
each picking aisle becomes a random variable distributed as the largest order statis-
tic. From the distribution of the order statistics, the probability density function of
the farthest pick location is
dF n ðxÞ
Pðxmax ¼ xÞ ¼ , 0  x  1: ð4Þ
dx
The expected value of the largest order statistic, R(n), is obtained as
Z1 n
dF ðxÞ
Eðxmax Þ ¼ RðnÞ ¼ x dx: ð5Þ
0 dx
Since the expected distance of the farthest pick location is a fraction R(n) of the
aisle length l, the expected travel distance within aisles can be expressed as
DR
w ¼ v½wC þ 2lRðnÞ: ð6Þ
3878 H. Hwang et al.

Back aisle

2 1 0 1 2
... ...

X Front aisle
P/D
j = (P−1)/2 ... j=2 j=1 k=1 k = 2 ... k = (P−1)/2

Figure 4. Aisle numbering with an odd number of P.

To develop the expected across-aisles travel distance model, two cases are con-
sidered in terms of the number of stocking aisles, odd and even number. For the case
of an odd number of stocking aisles, the stocking aisles are numbered with the indices
of j, k as shown in figure 4. Suppose that every pick location is at the right side of
the aisle 0. In this case the total distance across-aisle becomes 2kw. The probability
that the kth aisle, k > 0, is the rightmost picking aisle can be expressed as:
"   N #
kþ1 N k
 : ð7Þ
P P
Note that the same results are obtained for the case where every pick location
is in the aisles to the left of the centre aisle. Thus the expected across-aisles travel
distance for the above two cases is as follows:
"   N #
X
ðP1Þ=2
kþ1 N k
2 2kw  : ð8Þ
k¼1
P P
Consider the case where all pick locations are located between the jth aisle and
kth aisle ( j  k 6¼ 0)). Then the travel distance becomes (2j þ 2k)w. The probability
that all pick locations are between the jth aisle and kth aisle with the condition that
at least one pick location exists in each of the jth and kth aisle can be expressed as
follows:
ð j þ k þ 1ÞN 2ð j þ kÞN ð j þ k  1ÞN
 þ : ð9Þ
PN PN PN
Then the expected travel distance can be found from equation (9):
X ðP1Þ=2
ðP1Þ=2 X ð j þ k þ 1ÞN  2ð j þ kÞN þ ð j þ k  1ÞN
ð2j þ 2kÞw : ð10Þ
j¼1 k¼1
PN

With the odd number of stocking aisles, the total expected travel distance across-
aisle is obtained as follows:
"   N #
R
X
ðP1Þ=2
kþ1 N k
DA ¼ 2 2kw 
k¼1
P P
X ðP1Þ=2
ðP1Þ=2 X ð j þ k þ 1ÞN  2ð j þ kÞN þ ð j þ k  1ÞN
þ ð2j þ 2kÞw : ð11Þ
j¼1 k¼1
PN
Routing policies for order-picking operations 3879

Similarly, DRA can be developed for the case of an even number of stocking aisles
(figure 1). The result is given in equation (12):
"    #
XP=2
k N k1 N
R
DA ¼ 2 ð2k  1Þw 
k¼1
P P
P=2 X
X P=2
ð j þ kÞN  2ð j þ k  1ÞN þ ð j þ k  2ÞN
þ ð2j þ 2k  2Þw : ð12Þ
j¼1 k¼1
PN

The expected travel distance for return policy with across-aisle strategy is found
by adding equation (11) to equation (6) for an odd number of stocking aisles and
equation (12) to equation (6) for an even number of stocking aisles.

3.2. Travel model for the traversal policy with within-aisle strategy
Under within-aisle storage policy, the right and left sections of the stocking
area divided by the vertical centre-line have pick locations that are symmetrically
distributed. The probability of having a pick location in an aisle is determined by
F(x) of equation (1) and is identical for the two aisles in symmetric position. Also,
note that in a given aisle, pick locations are uniformly distributed.
Before developing the travel model, a slight modification is needed on the tra-
versal policy for the case of an odd number of picking aisles in which a redundant
traversal route occurs from the back aisle to the front. In this case, the order picker is
assumed to traverse all the picking aisles except the last aisle, in which return policy
is adopted. Furthermore, the expected travel distance for the return route is assumed
to be l.
Note that the expected number of picking aisles has a slightly different expression
depending on whether the number of stocking aisles is odd or even. When P is odd
(refer to figure 4), the probability that an aisle has a pick location can be expressed
as F(1/P) when k ¼ 0 and 1/2[F((2k þ 1)/P)  F((2k  1)/P)] when k ¼ 1, . . ., (P  1)/
2. For the jth aisle, j ¼ 1, . . ., (P  1)/2, it has the same probability as the one in the
symmetric position. With P being even, the probability that an aisle has a pick
location becomes 1/2[F(2k/P)  F((2k  2)/P)] when k ¼ 1, . . ., (P  1)/2.
Now, the expected within-aisle travel distance is developed. Let Pk be the
probability that a pick location exists in the aisle k. The probability that the aisle
has no pick locations is (1  Pk)N. Thus the probability that there is at least one pick
location in the aisle which the operator has to traverse becomes 1  (1  Pk)N.
Substituting an appropriate form of Pk developed above, the expected within-aisle
travel distance, (l þ wC) (the expected number of picking aisles), is obtained as
follows:
(   N ðP1Þ=2 (      N ))
1 X 1 2kþ1 2k1
T
DW ¼ ðl þwC Þ 1 1F þ2 1 1 F F
P k¼1
2 P P
when P is odd ð13Þ
(      N )
X
P=2
1 2k 2k  2
DTW ¼ 2ðl þ wC Þ 1 1 F F when P is even:
k¼1
2 P P
ð14Þ
3880 H. Hwang et al.

Now, the expected across-aisles travel distance is in order. Suppose that all the
pick locations are to the right side of the aisle 0 with the kth aisle being the rightmost
one. In this case, the expected across-aisle travel distance becomes 2kw when P is odd
and (2k  1)w when P is even. And, the probability that the farthest picking aisle is
the kth aisle can be expressed as follows:
         
1 1 1 2k þ 1 N 1 1 1 2k  1 N
F þ F  F þ F when P is odd ð15Þ
2 P 2 P 2 P 2 P
  N   
1 2k 1 2k  2 N
F  F when P is even: ð16Þ
2 P 2 P

Combining the two elements, the expected travel distance is obtained when
every pick location is located either to the right side or the left side of the warehouse
and they are
(          )
X
ðP1Þ=2
1 1 1 2k þ 1 N 1 1 1 2k  1 N
2 2kw F þ F  F þ F
k¼1
2 P 2 P 2 P 2 P
when P is odd ð17Þ

(       )
X
P=2
1 2k N 1 2k  2 N
2 ð2k  1Þw F  F
k¼1
2 P 2 P
when P is even: ð18Þ

Suppose that all the pick locations are in between the jth and kth aisle with the
condition that at least one pick location exists in each of the two aisles. In this case,
the expected travel distance is 2( j þ k)/w when P is odd and 2( j þ k  1)/w when
P is even. With the associated probability found in the same way as that of return
policy, the expected travel distance can be written as
(         
X ðP1Þ=2
ðP1Þ=2 X 1 2k þ 1 1 2j þ 1 N 1 2k  1 1 2j þ 1 N
2ð j þ kÞw F þ F  F þ F
j¼1 k¼1
2 P 2 P 2 P 2 P
          )
1 2k þ 1 1 2j  1 N 1 2k  1 1 2j  1 N
 F þ F þ F þ F
2 P 2 P 2 P 2 P
when P is odd ð19Þ

(   N     
X
P=2 X
P=2
1 2j 1 2k 1 2j 1 2k  2 N
2ð j þ k  1Þw
F þ F  F þ F
j¼1 k¼1
2 P 2 P 2 P 2 P
    N     N )
1 2j  2 1 2k 1 2j  2 1 2k  2
 F þ F þ F þ F
2 P 2 P 2 P 2 P
when P is even: ð20Þ

The expected travel distance model for traversal policy is the sum of equations
(13), (17) and (19) for an odd number of stocking aisles and the sum of equations
(14), (18) and (20) for an even number of stocking aisles.
Routing policies for order-picking operations 3881

Back aisle

1 2 3 ... k ... P

Front aisle

Figure 5. Aisle numbering with midpoint policy.

3.3. Travel model for midpoint policy with perimeter strategy


In the midpoint policy with perimeter strategy, the warehouse is divided hori-
zontally into two sections of equal area, the lower and upper sections (see figure 5).
Since pick locations in each section can be described by an identical COI-based ABC
curve, they can be considered separately. Note that the travel model in midpoint
policy is very similar to that in return policy with the only difference being the
existence of a traversal route instead of a return route in the rightmost and leftmost
picking aisles. Thus the travel model can be easily developed by utilizing the results
in Section 3.1. For instance, the substitution of 1/2P into1/P for the probability that
a pick location exists in an aisle and 2P into P for the number of picking aisles in
equation (2), the expected number of picks per picking aisle is obtained as
"   #
1 N
v¼ 1 1 2P: ð21Þ
2P

The expected travel distance within aisles (DM


W ) is determined through some
adjustments of equation (22):
 
M l
D0 ¼ v wC þ 2 RðnÞ : ð22Þ
2
Note that equation (22) comes from equation (6) where wC þ 2l/2R(n) is the
expected travel distance within an aisle. Also, note that the probability that a
given aisle becomes the leftmost or rightmost picking aisle is identical in the upper
and lower sections. Let Prob (L ¼ i, R ¼ j) be the probability that the ith aisle
becomes the leftmost picking aisle (L) while the jth aisle, j > i, the rightmost picking
aisle (R). Prob (L ¼ 1) can be found through the following steps.
 N  N  N
4 3 1
Prob ðL ¼ 1, R ¼ 2Þ ¼ 2 þ2 ð23Þ
2P 2P 2P
 N      
2k 2k  1 N 2k  3 N 2k  4 N
Prob ðL ¼ 1, R ¼ kÞ ¼ 2 þ2 
2P 2P 2P 2P
where k ¼ 2, 3, . . . , P ð24Þ
"        #
XP
2k N 2k  1 N 2k  3 N 2k  4 N
Prob ðL ¼ 1Þ ¼ 2 þ2 
k¼2
2P 2P 2P 2P
 N  N  N  N  N
2P  2 2P 2 2P  1 1
¼ þ  2 þ2 : ð25Þ
2P 2P 2P 2P 2P
3882 H. Hwang et al.

Similarly, Prob (L ¼ j ) is obtained as follows:


"        #
XP
2ðkj þ1Þ N 2ðkj þ1Þ1 N 2ðkj þ1Þ3 N 2ðkj þ1Þ4 N
2 þ2 
k¼jþ1
2P 2P 2P 2P
 N  N  N  N  N
2ðPjÞ 2ðPj þ1Þ 2 2ðPj þ1Þ1 1
¼ þ  2 þ2 : ð26Þ
2P 2P 2P 2P 2P
Let PL and PR be the probability that the two sections have an identical value
of L and R, respectively. To facilitate the estimation of DMW , the situation is ignored
where all pick locations are concentrated in one aisle, which tends to be zero in
probability as N becomes large. Equation (27) shows an approximated value of
PL and PR:
"
P1 
X     N
L R 2ðP  jÞ N 2ðP  j þ 1Þ N 2
P ¼P ¼ þ 
j¼1
2P 2P 2P
   N #
2ðP  j þ 1Þ  1 N 1
2 þ2 : ð27Þ
2P 2P
When the two sections have only one common value in either L or R, DM W is
obtained by subtracting three return routes from DM 0 of equation (22) after adding
two traversal routes. With no common value in L and R, two return routes are
subtracted from DM 0 after adding two traversal routes. If the two sections have
an identical value in both L and R, two traversal routes are added after subtracting
four return routes.
Therefore,
DM M L R R L
W ¼ D0  P P ð4lRðnÞ  2lÞ  P ð1  P Þð3lRðnÞ  2lÞ

 PL ð1  PR Þð3lRðnÞ  2lÞ  ð1  PL Þð1  PR Þð2lRðnÞ  2lÞ


¼ DM L
0  2ðP þ 1ÞlRðnÞ þ 2l: ð28Þ
Since the total expected travel distance of across aisle (DM
A ) depends only on pick
locations and each aisle has an equal probability of having a pick location due to the
perimeter storage policy, it has a similar expression as those in return policy and
"   N #
M
X
ðP1Þ=2
kþ1 N k
DA ¼ 2 2kw 
k¼1
P P
X ðP1Þ=2
ðP1Þ=2 X ð j þ k þ 1ÞN  2ð j þ kÞN þ ð j þ k  1ÞN
þ ð2j þ 2kÞw
j¼1 k¼1
PN
when P is odd ð29Þ
"    #
X
P=2
k N k1 N
DM
A ¼2 ð2k  1Þw 
k¼1
P P
P=2 X
X P=2
ð j þ kÞN  2ð j þ k  1ÞN þ ð j þ k  2ÞN
þ ð2j þ 2k  2Þw
j¼1 k¼1
PN
when P is even: ð30Þ
Routing policies for order-picking operations 3883

The total expected travel distance for midpoint policy is the sum of equation
(28) and (29) for an odd number of stocking aisles and the sum of equation (28) and
(30) for an even number of stocking aisles.

4. Performance evaluation of the routing policies


The performance of the routing policies depends on the particular operating
condition of the system under study. To examine the effects of the system param-
eters, the total expected travel distance was computed varying the order size (N),
popularity skewness (S), and ratio (b) of the length to the width of the warehouse.
Altogether, 160 different cases, i.e. eight different order sizes from 4 to 80, four
different popularity skewnesses, and five types of layouts, were tested for each
routing policy. In the COI-based ABC curve of equation (1), S with 50/20
implies that 20% of items represent approximately 50% of the total number of
order pickings. 50/50 implies that items are randomly stored along stocking aisles
and so pick locations are assumed to be uniformly distributed. In addition to
random (50/50), low (50/20), medium (60/20), and high (70/20) were considered
with the S value of 0.33, 0.20, and 0.12, respectively. Table 1 shows the five different
layouts of an equal stocking area where b is the ratio of the length to the width of
the warehouse.
Prior to the evaluation study, the analytical models developed were empirically
tested for accuracy. Specifically, the results obtained from the models were compared
with those obtained from a simulation model programmed in Visual Basic. For the
test, 32 cases, i.e. four different popularity skewnesses and eight different order
sizes, were considered in layout No.1, and 400 simulation runs were performed for
each case of the system parameters. For the simulation study, let AN(i) and LW(i)
denote the aisle number and location within the aisle of pick order i, respectively.
The simulation software generates a set of (AN(i), LW(i)), i ¼ 1, 2, . . ., N, for each
run on the basis of the routing policies and associated storage strategies. For
instance, with return policy with across-aisle strategy, AN is chosen randomly
among the numbers between 1 and P, while LW is obtained from the inverse func-
tion of equation (1) and a random number x, 0  x  1. The opposite holds for
traversal policy with within-aisle strategy where AN is found from the COI-based
ABC curve and LW is chosen randomly. For midpoint policy with perimeter strat-
egy, considering the existence of two sections in the aisles, AN becomes a random
integer among the numbers between 1 and 2P and LW follows the COI-based ABC
curve. Table 2 shows the difference between the analytical models and simulation
results. The differences in general are less than 6%, which indicate that the analytical
models give a relatively good approximation of the expected total travel distances.
Compared with traversal policy, the models with the other two policies show

Layout L W b l P

1 45.75 80 0.57 42.75 16


2 51.86 70 0.74 48.86 14
3 60.00 60 1.00 57.00 12
4 71.40 50 1.43 68.40 10
5 88.50 40 2.21 85.50 8

Table 1. Layout configurations tested.


3884 H. Hwang et al.

S N return traversal midpoint

Random (50/50) 4 1.6194 0.6975 0.8690


8 2.7002 0.2662 1.0464
16 4.5507 0.0943 3.5339
24 5.2073 0.2552 2.3410
32 5.2217 0.4471 1.5976
48 4.5980 0.0389 2.4500
64 3.6538 0.1310 4.0615
80 2.7719 0.0669 4.1267
0.33 (50/20) 4 1.4730 0.2509 0.9951
8 2.3978 0.8304 0.0319
16 3.9759 0.2313 1.6153
24 4.4166 0.3167 1.7846
32 4.8415 0.3505 2.4940
48 4.6093 0.9683 1.1844
64 4.0303 0.1231 3.1754
80 3.2603 0.3054 2.6378
0.20 (60/20) 4 1.3786 0.8513 2.2000
8 2.1987 0.4544 1.5503
16 3.5879 1.0934 0.2248
24 3.9182 0.1243 2.3640
32 4.4150 0.4947 1.3974
48 4.3155 0.3693 2.9629
64 3.8730 0.2521 3.3952
80 3.1824 0.6506 3.7372
0.12 (70/20) 4 1.2638 1.0504 0.6388
8 1.9515 1.1335 1.2728
16 3.1224 1.0921 1.1937
24 3.2903 0.6824 2.6855
32 3.8694 0.8731 1.6846
48 3.8913 0.4447 2.2179
64 3.6088 0.9112 1.2913
80 2.9762 0.6653 0.9071

Table 2. Percentage difference between analytical models and simulation (%) for layout 1.

substantially larger differences. A major source of the differences is attributable to


R(n) of equation (5).
The computational results from the analytical models are shown in table 3 and
figure 6 for the layout No. 1. For the other layouts, very similar results are
obtained. For almost every order size tested, midpoint routing policy gives
better performance except when the order size is either very small (N ¼ 4) or
very large (N ¼ 80). For small size of order, return policy works better due to
shorter within-aisle travel distance. With traversal policy, the order picker has to
traverse the aisle once he enters a picking aisle, resulting in non-productive travel.
When pick locations tend to exist in the upper part, midpoint policy also causes
inefficient travel from the front aisle to the back aisle. This inefficiency becomes
more prominent when the order size is small. As the order size becomes very large,
within-aisle storage strategy with traversal policy provides smaller across-aisles
travel distance.
Figure 7 visualizes the effect of the popularity skewness. It can be observed that
the higher the popularity skewness, the smaller the expected travel distance. This is
Routing policies for order-picking operations 3885

S N (Order size) Return Traversal Midpoint

Random (50/50) 4 271.5080 264.1829 231.5978*


8 448.4268 418.8906 341.7760*
16 706.2866 610.8862 507.2570*
24 889.0205 721.2465 639.8552*
32 1020.5820 786.3479 750.2365*
48 1185.7600 848.0162* 920.7764
64 1277.1830 869.9074* 1041.9490
80 1332.1670 877.6951* 1128.8620
0.33 (50/20) 4 201.6523* 218.8311 215.5812
8 320.6902 349.3123 294.2877*
16 493.5516 505.4319 405.4674*
24 623.3170 598.7661 494.1280*
32 724.8284 661.8773 569.4863*
48 870.0916 741.5414 691.4204*
64 967.2238 788.4121 785.0039*
80 1036.5630 817.9624* 858.2330
0.20 (60/20) 4 184.8135* 202.2078 211.7551
8 289.1888 317.3845 282.8261*
16 438.5840 454.4482 380.3064*
24 551.6744 539.8593 457.2749*
32 641.4486 600.5786 522.6956*
48 773.9524 682.5125 629.4035*
64 866.2732 735.2397 712.5776*
80 934.9582 771.5236* 778.9128
0.12 (70/20) 4 168.4728* 182.9048 208.0592
8 258.4046* 279.4332 271.6805
16 384.0569 395.8260 355.7152*
24 479.2777 472.4365 420.9737*
32 555.9186 529.2964 476.2787*
48 672.0656 610.0876 566.8288*
64 756.3042 665.7452 638.1797*
80 821.248 706.6656 695.9571*
*The best routing policies for given conditions.
Table 3. Total expected travel distance for layout No. 1 (P ¼ 16, L ¼ 45.75m).

due to the fact that pick locations tend to be closer to the P/D point with a high value
of the popularity skewness.
As the order size becomes larger, the effects of the skewness tend to increase
on the performance of return and midpoint policy, while they decrease on that of
traversal policy. If the order size is so large that every aisle becomes a picking aisle,
then in traversal policy the order picker has to traverse all aisles. In this case, the
popularity skewness does not have any significant role. In return and midpoint
policy with a given order size, the largest order statistics is closely related with the
skewness, which in turn affects the expected within-aisles travel distance. That
explains the shape of figure 7.
Also, the effect of the ratio of the length to the width of the warehouse is
examined. Layout No. 1 turns out to be the most desirable one among the various
warehouse configurations. Especially, midpoint policy in layout No. 1 gives the
shortest expected travel distance irrespective of the order size and popularity
skewness. Figure 8 illustrates the above findings with the low popularity skewness.
3886
1200

expected travel distance(m)


1200

expected travel distance(m)


1000
1000
800
return 800 return
600 traversal
600 traversal
400 midpoint
midpoint
400
200
200
0
0
4 8 16 24 32 48 64 80
4 8 16 24 32 48 64 80

H. Hwang et al.
order size(N) order size(N)

(1) Low skewness (50/20) (2) Medium skewness (60/20)

expected travel distance(m)


900
800
700
600
return
500
traversal
400
300 midpoint
200
100
0
4 8 16 24 32 48 64 80
order size(N)

(3) High skewness (70/20)

Figure 6. Expected travel distance for layout No. 1.


1400 1000

expected travel distance(m)

expected travel distance(m)


900
1200
800
1000 700
random rando m
800 600

Routing policies for order-picking operations


low low
500
600 medium medium
400
high high
400 300
200
200
100
0 0
4 8 16 24 32 48 64 80 4 8 16 24 32 48 64 80
order size(N) order size(N)

(1) Return policy (2) Traversal policy

1200

expected travel distance(m)


1000

800 random
low
600
medium
400 high

200

0
4 8 16 24 32 48 64 80
order size(N)

(3) Midpoint policy

Figure 7. Expected travel distance with the layout No. 1.

3887
3888 H. Hwang et al.
1000

900

800
expected travel distance(m)

700

600 Layout 1
Layout 2
500 Layout 3
Layout 4
400 Layout 5

300

200

100

0
4 8 16 24 32 48 64 80
order size(N)

Figure 8. Expected travel distance under low skewness.

5. Conclusions
This paper compared the performances of three well-known routing policies,
return policy with across-aisle strategy, traversal policy with within-aisle strategy
and midpoint policy with perimeter strategy in a low-level picker-to-part system.
First, based on the COI-based item assignment in the warehouse, the analytical
travel distance models were developed. Then through comparison studies with the
results of a simulation model, the validity of the models was illustrated. Finally,
varying the system parameter values, the total expected travel distances were com-
puted for each policy. It is observed that for very small order size, i.e. 4, the return
policy shows better performance, while for very large order size, i.e. 64–80, traversal
policy performs better. In general, midpoint policy outperforms the other two.
The results from the sensitivity analysis indicate that the high skewed popularity
can reduce the travel distance significantly regardless of the routing policies,
which is consistent with our expectation. Also, it is observed that a most desirable
configuration of the warehouse in figure 1 is the one whose width is about one half of
its length.

Acknowledgements
This work was supported by grant No. R01-1999-00323 from the Korea Science
and Engineering Foundation.

References
BENDER, PAUL S., 1979, Mathematical Modeling of the 20/80 rule: theory and practice. Journal
of Business Logistics, 2, 139–157.
CARON, F., MARCHET, G. and PEREGO, A., 1998, Routing policies and COI-based storage
policies in picker-to-part systems. International Journal of Production Research,
36, 713–732.
CARON, F., MARCHET, G. and PEREGO, A., 2000, Optimal layout in low-level picker-to-part
systems. International Journal of Production Research, 38, 101–117.
COYLE, J. J., BARDI, E. J. and LANGLEY, C. J., 2002, The Management of Business Logistics
(Mason: South-Western College Pub).
Routing policies for order-picking operations 3889

GIBSON, D. R. and SHARP, G. P., 1992, Order batching procedures. European Journal of
Operational Research, 58, 57–67.
GOETSCHALCKX, M. and RATLIFF, H. D., 1988, Order picking in an aisle. IIE Transactions,
20, 53–62.
HALL, R. W., 1993, Distance approximations for routing manual pickers in a warehouse.
IIE Transactions, 25, 76–87.
HESKETT, J. L., 1963, Cube-per-order index – a key to warehouse stock location.
Transportation and Distribution Management, 3, 27–31.
JARVIS, J. M. and MCDOWELL, E. D., 1991, Optimal product layout in an order picking
warehouse. IIE Transactions, 23, 93–102.
KALLINA, C. and LYNN, J., 1976, Application of the cube-per-order index rule for stock
location in distribution warehouse. Interfaces, 7, 37–46.
KOSTER, M. B. M., VAN DER POORT, E. S. and WOLTERS. M., 1999, Efficient orderbatching
methods in warehouse, International Journal of Production Research, 37, 1479–1504.
PETERSEN, C. G., 1997, An evaluation of order picking routing policies. International Journal
of Operation & Production Management, 17, 1096–1111.
PETERSEN II, C. G. and SCHMENNER, R. W., 1999, An evaluation of routing and volume-based
storage policies in an order picking operation. Decision Sciences, 30, 481–501.
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, 507–521.
ROODBERGEN, K. J. and KOSTER, R., 2001, Routing order pickers in a warehouse with a middle
aisle. European Journal of Operational Research, 133, 32–43.
ROSENWEIN, M. B., 1996, A comparison of heuristics for the problem of batching orders for
warehouse selection. International Journal of Production Research, 34, 657–664.
TOMPKINS, J. A. and WHITE, J., 1984, Facilities Planning (New York: John Wiley).

You might also like