You are on page 1of 9

EUROPEAN

JOURNAL
OF OPERATIONAL
RESEARCH
ELSEVIER European Journal of Operational Research 87 (1995) 65-73

Theory and Methodology

A heuristic procedure for the single-row facility layout problem


K. Ravi Kumar a. , , G e o r g e C. H a d j i n i c o l a b, T i n g - l i L i n c
a Department of Information and Operations Management, School of Business Administration, University of Southern California,
LosAngeles, CA 90089-1421, USA
b Department of Public and Business Administration, School of Economics and Management, Universityof Cyprus, Kallipoleos 75,
P.O. Box 537, Nicosia, Cyprus
c Department of Industrial and Systems Engineering, School of Engineering, University of Southern California, Los Angeles,
CA 90089-0193, USA
Received March 1991; revised July 1993, January 1994

Abstract

The single-row facility layout problem arises when assigning service rooms along a corridor, storing files on the
cylinders of a disk, and in flexible manufacturing systems where the efficiency of automated guided vehicles forces
the layout to be linear. This paper describes a constructive heuristic which provides solutions to the single-row
facility layout problem so as to minimize the materials handling cost. The heuristic assigns the facilities with the
largest number of moves between them to adjacent locations on the line. The heuristic differs from earlier
algorithms since at any stage of the process, more than two facilities can be added to the solution sequence. The
performance of the heuristic is compared to other methods proposed for the single-row facility layout problem.
Results show that the heuristic performs well in terms of computational efficiency and solution quality.

Keywords: Facility layout problem; Single-row layout problem; One dimensional assignment problem; Heuristic

1. Introduction linear integer programming problem (Lawler,


1963), mixed-integer programming problem
The facility layout problem concerns the place- (Kaufman and Broeckx, 1978; Bazaraa and Sher-
ment of a number of facilities on the factory floor ali, 1980; Burkard and Bonniger, 1983), or
so as to minimize the amount or cost of materials quadratic set covering problem (Bazaraa, 1975).
movement between the facilities. Koopmans and Love and Wong (1976a) presented a linear
Beckman (1957) first formulated the problem of mixed-integer program which was not a lineariza-
assigning facilities to an equal number of loca- tion attempt of the QAP. Heragu and Kusiak
tions as a Quadratic Assignment Problem (QAP). (1991) also present a linear mixed-integer pro-
Subsequent formulations of the facility layout gramming formulation which they solve through
problem attempted to linearize the QAP to a unconstrained minimization techniques. Hall
(1970) and Drezner (1980) formulated the facility
layout problem as a non-convex mathematical
* Corresponding author. program.

0377-2217/95/$09.50 © 1995 Elsevier Science B.V. All rights reserved


SSDI 0377-2217(94)00062-H
66 K.R. Kumar et al. / European Journal of Operational Research 87 (1995) 65-73

The single-row facility layout problem is a spe- procedure that extracts the eigenvectors of a
cial case where facilities of equal or unequal transformed flow matrix. Hall (1970) suggests that
dimensions are arranged on a line. Picard and the plot of the two eigenvectors associated with
Oueyranne (1981), besides solving the problem the smallest eigenvalues could be used to extract
with dynamic programming, discuss a number of the sequence with which facilities could be placed
applications of the single-row facility layout prob- in a single-row layout. Simmons (1969, 1971) pre-
lem. For example, the single-row layout problem sented a branch and bound algorithm to solve the
arises in the layout of machines in a Flexible linear ordering problem. In computing his upper
Manufacturing System (FMS) which is served by bound, he proposed an algorithm that generates
an Automated Guided Vehicle (AGV) (Kusiak a sequence of machines which Heragu and Ku-
and Heragu, 1988), since linear layouts enable siak (1988) applied to solve the general single-row
the efficient functioning of AGVs (Muller, 1988). facility layout problem. This stepwise algorithm,
In practice, the single-row layout problem is en- referred to by Kusiak (1990) as the Modified
countered in two forms: the linear single row and Spanning Tree (MST) algorithm, selects the pair
circular single row problem (Kusiak and Heragu, of facilities with the largest element in the flow
1988). The linear placement of facilities of differ- matrix and then proceeds by adding facilities to
ent but known dimensions on a straight line, is the left and right of this pair.
also known as the one dimensional space alloca- Neghabat (1974) proposed the linear place-
tion problem, which Simmons (1969) states is ment algorithm, a stepwise procedure which con-
more of an ordering problem. An even more structs the facility layout by inserting one facility
specific case of the ordering problem is when the at a time in the solution set. Heragu and Kusiak
facilities are of the same length, and as a result, (1991) modeled the single-row facility layout
the distance between adjacent locations for plac- problem as a linear mixed-integer program which
ing the facilities on the line can be assumed to be they solved using the penalty method for uncon-
one unit. This case is referred to by Adolphson strained programs. Heragu and Alfa (1992) pro-
and Hu (1973) as the optimal linear ordering posed an algorithm known as the Hybrid Simu-
problem. lated Annealing algorithm which improves upon
A number of models and algorithms have been an initial solution using a simulated annealing
proposed for the single-row facility layout which algorithm.
is known to be an NP-complete problem In this paper, we propose a greedy heuristic
(Beghin-Picavet and Hansen, 1982). Karp and that looks at strong relationships between mutu-
Held (1967) developed a dynamic programming ally distinct pairs of facilities. The heuristic is
algorithm for the solution of the module place- such that at any stage of the process, more than
ment problem which is equivalent to the optimal one facility can be added to the solution se-
linear ordering problem. Picard and Oueyranne quence, which can also be disjointed. This differs
(1981) extended the algorithm developed by Karp from earlier stepwise single-row layout heuristic
and Held (1967) for the general one-dimensional methods which increment the solution sequence
space allocation problem. Adolphson and Hu one facility at a time. Our algorithm's perfor-
(1973) suggested an algorithm for the optimal mance is compared to other stepwise and simul-
linear ordering problem to extract the linear facil- taneous procedures which provide solutions to
ity sequence for problems whose flow matrix can the single-row layout problem, such as those pro-
be represented as a single rooted tree. Picard and posed by Hall (1970), Neghabat (1974), Heragu
Oueyranne (1981) extended their algorithm to and Kusiak (1988), Heragu and Kusiak (1991),
include facilities of dissimilar dimensions. Love and Heragu and Alfa (1992). The computational
and Wong (1976b) presented a linear mixed-in- experience shows that the proposed heuristic per-
teger program for the single-row facility layout forms well in terms of computational efficiency
problem. and solution quality.
Hall (1970) and Drezner (1987) presented a The organization of the paper is as follows.
K.R. Kumar et al. / European Journal of Operational Research 87 (1995) 65-73 67

Section 2 presents a QAP formulation of the 1979; Bazaraa and Kirca, 1983; Kaku and
single-row facility layout problem under the re- Thompson, 1986) and cutting plane algorithms
laxation that the distance between adjacent loca- (Bazaraa and Sherali, 1980; Burkard and Bon-
tions is one unit and also describes the construc- niger, 1983). QAP is an NP-hard problem
tive greedy heuristic. Section 3 includes the per- (Burkard, 1984) and a number of suboptimal
formance testing of the heuristic as compared to techniques have been proposed which can be
other single-row facility layout methods. Finally, classified into four categories: construction algo-
Section 4 contains a summary and our conclu- rithms (Hillier and Connors, 1966; Lee and
sions. Moore, 1967; Kusiak and Heragu, 1988), im-
provement algorithms (Armour and Buffa, 1963;
Vollman et al., 1968; Khalil, 1973), hybrid algo-
2. Mathematical formulation and greedy heuristic rithms (Drezner, 1980, 1987; Bazaraa and Sherali,
1980; Bazaraa and Kirca, 1983), and graph theo-
Heragu and Kusiak (1988) pointed out that if retic algorithms (Seppanen and Moore, 1970,
the facilities have dissimilar dimensions, then the 1975; Foulds and Robinson, 1978; Foulds et al.,
facilities layout problem can not be formulated as 1985; Giffin and Foulds, 1987). More recent
a QAP due to the undefined locations where methods involve the use of simulated annealing
facilities should be placed. We provide a QAP algorithms (Heragu and Alfa, 1992). Foulds
formulation for the case where a number of dis- (1983), Burkard (1984), Kusiak and Heragu (1987),
tinct facilities are to be assigned to an equal and Heragu (1992) provide extensive reviews of
number of equi-distant locations which lie on a the literature on the QAP and the facilities lay-
straight line, so as to minimize the materials out problem. Rosenblatt (1979) and Malakooti
handling cost. To formulate the problem, let m and D'Souza (1987) provide reviews on multiple
be the number of facilities and locations, and fij criteria QAP formulations.
the number of moves per time period in the work Below we present a constructive greedy heuris-
flow from facility i to facility j. In addition, we tic for the single-row layout problem. The heuris-
define tic does not take into account the dimensions of
the facilities in the construction of the single-row
xik = 1 if facility i is assigned to location k; 0 layout. For the description of the heuristic, the
otherwise following notation and terminology are used:
The mathematical formulation of the above S: A set whose elements are sequences of two or
problem is given by: more facilities; no two elements contain the same
m m
types of facilities.
min£ ~ ~ £ xikxjk,fulk-k'l (1) P: A set whose elements are pairs of facilities
i=l j=l k=l k'=l which have the same fu value.
Links: The connection between two adjacent fa-
s.t.
cilities. (A facility in a single-row layout can be
m
x,k= l, k = l, 2 . . . . . m, (2) placed adjacent to at most two other facilities.
i=l When two facilities are placed adjacently, then
one of their links is said to have been occupied.)
~xik=l, i = 1 , 2 .... ,m, (3)
k=l
Heuristic
Xik ~ { 0 , 1}, i,k = 1, 2 . . . . . m. (4) Step 1. Transform the flow matrix Fmx m into
Optimal algorithms for solving the QAP have an F'xm (lower triangular) such that
been classified by Kusiak and Heragu (1987) in
[fij +fji, i >j,
two categories: branch and bound algorithms
fiJ ~ 0, i < j.
(Lawler, 1963; Land, 1963; Bazaraa and Elshafei,
68 K.R. Kumar et al. / European Journal of OperationalResearch 87 (1995) 65-73

Step Z Set S = {¢ }. tion, but this would be at the expense of the


Step 3. Repeat until all m facilities are in S. computational efficiency of our heuristic method.
Step 3.1. Assign the pair(s) of facilities with the
largest f'j to P. Example. Consider a case where seven equal-area
Step 3.2. Replace the fo-(s) of the above pair(s) facilities are to be assigned to locations on a
in the flow matrix with zero. straight line layout so as to minimize the materi-
Step 3.3. Repeat until the current P has no als handling cost. Furthermore, assume that the
elements. When done, go to Step 3. lower triangular matrix shown below is the trans-
Step 3.3.1. Either randomly or by comparative formed flow matrix of the above seven facilities,
search select an element from P. Remove this i.e. that the first step of the heuristic has already
element from P. been completed.
Step 3.3.2. If one facility in the element se- 4 5 6 7
1 2 3
lected from P is in S and both of its links are 1 0 0 0 0 0 0 0
occupied, discard the element and go to Step 3.3;
else go to Step 3.3.3. 2 69 0 0 0 0 0 0
Step 3.3.3. If the element consists of two facili- 3 78 42 0 0 0 0 0
ties that belong to the same element of S, discard
the element and go to Step 3.3; else go to Step 4 5 58 6 0 0 0 0
3.3.4. 5 51 31 30 99 0 0 0
Step 3.3.4. If the facilities in the element se-
lected from P connects the facilities of two dis- 6 76 88 22 98 99 0 0
joint elements in S, establish the corresponding 7 59 97 29 26 2 85 0
link and go to Step 3.3; else go to Step 3.3.5.
Step 3.3.5. If the facilities in the element se- The iterations below describe the process that
lected from P link a facility in an element in S the heuristic follows for the construction of the
and the element, establish the link and go to Step single-row layout. The numbers in parentheses
3.3; else go to Step 3.3.6. indicate the heuristic's step responsible for such
Step 3.3.6. Insert the element in S as a distinct an action to be taken.
element and go to Step 3.3. Iteration 1: S = {¢} (2). P = {5 ~ 4, 6 ~ 5} (3.1).
Select {5 ,~, 4} to enter S, P = {6 ~ 5} (3.3.1). S =
Unlike Heragu and Kusiak's (1988), and {5 ~ 4} (3.3.6). Select {6 ~ 5} to enter S, P = {¢}
Neghabat's (1974) heuristic methods, our heuris- (3.3.1). S = {6 ~ 5 ,~, 4} (3.3.5).
tic looks at and retains strong relationships be- Iteration 2: P = { 6 ~ 4} (3.1), discard element
tween mutually distinct pairs of facilities. Wher- (3.3.3).
ever possible, these are put in S, often forming Iteration 3: P = {7 ~ 2} (3.1). Insert the element
disjointed sequences of two or more facilities, in S (3.3.6), S = {6 ,--, 5 ,--, 4, 7 ~ 2}.
unlike the other heuristic methods that minimally Iteration 4: P = {6 ~ 2} (3.1). The element in P
increment the sequence by one facility at a time. connects two elements in S (3.3.4), S = {7 ~ 2
Nevertheless, in our heuristic, these disjointed 6,~5 ~4}.
sequences form constraints for the selection of Iteration 5: P = {7 ~ 6} (3.1), discard element
subsequent facility pair allocations. Neghabat's (3.3.3).
procedure, while allowing initial facility orderings Iteration 6: P = {1 ,~, 3} (3.1). Insert element in S
to be split apart, has the disadvantages of incre- (3.3.6), S = {7 <--}2--, 6,--, 5 ",-}4, 1 ,--, 3}.
menting the solution sequence by a single facility, Iteration 7: P = { 6 ~ 1} (3.1), discard element
and requiring a great deal of computation time, (3.3.2).
especially for problems of higher dimensions. Iteration 8: P = {2,--, 1} (3.1), discard element
Clearly, we could use a Neghabat-type search (3.3.2).
procedure in Step 3.3.1 rather than random selec- Iteration 9: P = {7 ~ 1} (3.1). The element in P
K.R. Kumar et al. / European Journal of Operational Research 87 (1995) 65-73 69

connects two elements in S (3.3.4). S = {4 o 5 (1974), Heragu and Kusiak (1988), Heragu and
6~2~7~1~3}. Kusiak (1991), and Heragu and Alfa (1992). The
At this point the heuristic terminates since all algorithms by Hall (1970), Neghabat (1974), Her-
seven facilities are in S. The single element in S agu and Kusiak (1988), and the heuristic pre-
indicates the order in which the facilities should sented in this paper were implemented using the
be placed in order to obtain the single-row facil- C programming language and run on a Sun 3/260
ity layout. If the distance between adjacent loca- computer. The performance testing comprises of
tions is one unit, then if facility 4 is assigned to three sets of problems found in the literature.
the first location, facility 5 on the second, etc., The first set of problems includes nine prob-
the resulting materials handling cost is 2285. A lems of 4 facilities. The flow matrices and dimen-
complete enumeration of all 7! possible layouts, sions of these problems can be found in Heragu
to search for the one with minimum materials and Kusiak (1988). The clearance between facili-
handling cost, indicates that the solution con- ties was assumed to be one unit. For these small
structed by the heuristic is the optimal one. dimensionality facility layout problems, all meth-
Note that in iteration 3, our heuristic inserts ods have proven to be computationally efficient
two facilities in S. This is not feasible in Heragu and yield optimal or near optimal solutions and
and Kusiak's and Neghabat's heuristic methods hence the results are not reported here. The
which increment their solution sequences by one heuristic proposed in this paper is slightly supe-
facility at a time. In general, the heuristic at any rior to Heragu and Kusiak's (1988) algorithm, but
point of its operation may insert any non-negative comparable to Hall's (1970) and Neghabat's (1974)
number of facilities in the solution sequence. algorithms.
Also, in the above example, at the end of the The second set of problems consists of 8 prob-
heuristic's execution, S contained a single ele- lems with 5 to 30 facilities. The flow matrices of
ment due to the nonzero elements of the flow the problems can be found in Nugent et al. (1968),
matrix. In cases where a number of elements in whereas their dimensions are reported in Heragu
the flow matrix are zero, the heuristic might and Kusiak (1988). The clearance between facili-
construct a solution sequence with several se- ties was assumed to be 0.01 units. Table 1 con-
quences of facilities with no overlap of facilities. tains the costs and computational times required
In other words, if we construct a graph whose by the algorithms tested. It is evident that Negha-
nodes are the facilities and whose arcs represent bat's algorithm is computationally more expensive
the flow between two facilities, and two or more than the heuristic presented in this paper. In
disjoint sub-graphs are created, then the heuristic addition, for problems of small dimensions (12 or
views the sub-graphs' flow matrices as distinct less facilities) Neghabat's algorithm yields the best
cases and provides single-row facility layouts in- cost (except in problem 2). Nevertheless, his solu-
dependent from one another. tion quality is heavily dependent on computa-
tional cost, a fact which is particularly true for
problems of higher dimensions. For problems of
3. Performance testing 20 and 30 facilities Neghabat's algorithm yields
inferior solutions, even after half an hour of
The performance of the heuristic was tested execution. The algorithm by Heragu and Kusiak
on a number of problems with 4 to 30 facilities. consistently yields inferior solutions in terms of
In all problems considered, the dimensions of the cost, but is computationally less expensive than
facilities were dissimilar. To assess the effective- our heuristic.
ness of the heuristic, the materials handling cost For problems with 12 or less facilities, our
and computational time for each problem was heuristic is less computationally expensive than
compared to the solution provided by other sin- Hall's method. For problems of higher dimen-
gle-row facility layout heuristic methods such as sions, Hall's algorithm is the least computation-
the ones proposed by Hall (1970), Neghabat ally expensive among all methods. We would like
70 K~I~ Kumar et al. / European Journal of Operational Research 87 (1995) 65-73

to point out that in our implementation of Hall's Simmons (1969). The flow matrix for problems 7
algorithm we did not account for the transforma- and 8 can be found in Nugent et al. (1968) and
tions required to be performed on the flow ma- the dimension of the facilities in Heragu and
trix. Nevertheless, we believe that the incorpora- Kusiak (1991). For these eight problems the
tion of the transformations in the software would clearance between facilities was assumed to be
not result in significant changes in the computa- zero units.
tional efficiency of Hall's algorithm. Table 1 also Table 2 contains the costs and processing times
shows that the cost provided by our heuristic in 4 of the above 8 problems for the heuristic and the
of the 8 problems (problems 3, 4, 5, and 7) algorithms proposed by Heragu and Kusiak (1991)
outperforms the solution by Hall. Hall's algo- and Heragu and Alfa (1992). The table includes
rithm yields solutions with better cost in 2 of the two sets of columns for costs and computation
8 problems (problems 6 and 8). times associated with our heuristic. In the first
We believe that the performance of our set, the solutions provided by our heuristic were
heuristic can deteriorate due to two primary rea- improved by one iteration of a greedy pairwise
sons. First, the greedy nature of the heuristic on exchange algorithm, whereas in the second, the
pairs of facilities with higher interaction, ignores solutions provided by our heuristic were im-
the aggregate effect of the remaining pairwise proved by a greedy pairwise exchange algorithm
interactions. Second, random selection of ele- until no improvement in cost could result from
ments from P (Step 3.3.1) can mislead the heuris- any further facility exchange. These sets of results
tic. Both of these issues can be alleviated to some were used for comparisons with the results ob-
extent by using a Neghabat-type local compari- tained by Heragu and Kusiak (1991) and Heragu
son, which will certainly increase the computa- and Alfa (1992), respectively.
tional cost of the heuristic. From the first set in Table 2 we see that the
The third set of problems consists of 8 prob- heuristic uniformly dominates the method by
lems. The facility dimensions and flow matrix for Heragu and Kusiak (1991) in terms of processing
problem 1 can be found in Beghin-Picavet and time. The heuristic obtains the optimal cost in
Hansen (1982), for problems 2 and 6 in Love and problems 1, 2, and 4 and provides better costs for
Wong (1976b), and for problems 3, 4 and 5 in problems 3, 5, 7, and 8 in less time than Heragu

Table 1
Computational comparisons of the heuristic four heuristic methods of the single-row facility layout problem
Previously best Heragu & Kusiak Kumar, Hadjinicola
known solution a Hall (1970) Neghabat (1974) (1988) & Lin
Problem Number Cost CPU Cost CPU b Cost CPU Cost CPU Cost CPU
number of
facilities
1 5 1.165 0.04 1.140 0.018 1.100 0.009 1.165 0.007 1.140 0.008
2 6 2.085 0.04 1.990 0.034 2.030 0.015 2.085 0.008 1.990 0.010
3 7 5.420 0.04 5.530 0.052 4.730 0.500 5.420 0.017 5.150 0.033
4 8 7.995 0.04 7.035 0.069 6.295 5.500 7.995 0.035 6.775 0.067
5 12 31.525 0.04 27.805 0.120 24.675 1800.00 31.525 0.067 24.715 0.100
6 15 62.624 0.05 49.375 0.153 51.500 1800.00 62.624 0.100 54.830 0.600
7 20 178.149 0.05 153.260 0.255 169.920 1800.00 178.149 0.600 141.040 4.200
8 30 n.a. n.a. 395.770 0.624 466.800 1800.00 414.400 4.800 405.520 14.800
CPU = central processing unit time in seconds. All CPU times were obtained on a Sun 3/260 computer.
n.a. means information not available.
a The previously best known solutions were obtained from Heragu and Kusiak (1988).
b The CPU time for Hall corresponds to the time needed to derive the eigenvalues using the software MATLAB and the time to
compute the cost. The time reported does not account for the matrix transformations required by Hall's algorithm.
K.R. Kumar et al. / European Journal of Operational Research 87 (1995) 65-73 71

and Kusiak (1991). The heuristic does not domi- Such a problem arises when assigning service
nate Heragu and Kusiak's (1991) method since it rooms in one side of a corridor to minimize the
yields higher cost in problem 6. total distance travelled, arranging books on a
From the second set in Table 2 we see that the shelf to minimize the movements for book re-
heuristic attains the optimal cost for problems 1, trieval, and assigning files on the cylinders of a
2, and 4 in significantly less time than the method disk so as to minimize the movement of the
proposed by Heragu and Alfa (1992). In addition, mechanical head in retrieving the files. In the
in problem 3, it obtains the same cost but in less context of flexible manufacturing systems, the
processing time. The heuristic does not provide single-row layout problem arises in the assign-
improved cost for problems 5 and 6 even though ment of machines to minimize the movement for
the solution it provides is computationally less an automated guided vehicle.
expensive. It is interesting to note that the heuris- We propose a constructive heuristic which as-
tic yielded improved costs for problems 7 and 8, signs facilities to locations on a straight line by
the higher dimension problems in significantly capitalizing on the pairs of facilities which have
less time. the largest number of moves between them. The
heuristic is tested on three sets of problems for
cases where the facilities have different dimen-
sions. Its performance is compared to existing
4. Summary and conclusion
algorithms specializing on single-row layout such
as those proposed by Hall (1970), Neghabat
A special case of the general facility layout (1974), Heragu and Kusiak (1988), Heragu and
problem is the single-row facility layout problem. Kusiak (1991) and Heragu and Alfa (1992).
Table 2
Comparative performance testing of the heuristic with the m e t h o d proposed by Heragu and Kusiak (1991) and Heragu and Alfa
(1992)
Heragu & Kusiak (1991) c
with infeasible with feasible Heragu & Alfa Kumar, Hadjinicola & Lin
initial solution initial solution (1992) Set 1 c Set 2 d
Problem N u m b e r of Cost a CPU b Cost a CPU b Cost a CPU b Cost CPU Cost CPU b
number facilities

1 4 78.0 0.08 78.0 0.09 78.0 1.624 78.0 0.007 78.0 0.007
2 5 151.0 0.08 151.0 0.13 151.0 10.351 151.0 0.010 151.0 0.010
3 8 2 342.5 0.36 2 341.5 0.59 2 324.5 11.803 2 324.5 0.080 2 324.5 0.080
4 10 2781.5 1.11 2781.5 0.84 2781.5 19.815 2781.5 0.100 2781.5 0.100
5 11 7041.5 0.96 7274.5 2.18 6933.5 23.103 6953.5 0.120 6953.5 0.120
6 11 6933.5 0.98 6933.5 0.95 6933.5 29.176 7265.5 0.120 7265.5 0.120
7 20 16 265.0 10.68 16 109.0 7.82 15 602.0 603.376 15 971.0 5.300 15 549.0 8.600
8 30 46139.0 36.43 46 454.0 35.74 45 111.0 585.947 45 308.5 25.800 44 466.5 91.800
C P U = central processing unit time in seconds.
a T h e costs and computational times obtained by Heragu and Kusiak (1991) are equal to or better than the previously best known

solutions (up to 1991), also reported in the same paper. T h e only exception is problem 5 whose previously best known cost, which is
also the optimal, is 6933.5 with 9.84 sec. of computational time on an IRIS 80 C.I.I.-H.B. computer (Beghin-Picavet and Hansen,
1982). The costs obtained by H e r a g u and Alfa (1992) are equal to or better than the previously best known solutions. T h e costs
reported for problems 1, 2, and 4 are also those of the optimal solutions.
b C P U time for H e r a g u and Kusiak (1991) on an A M D A H L 5870 computer. C P U time for the H S A algorithm was obtained on a
V A X 6420 mainframe computer. C P U time for the heuristic presented in this paper was obtained on a Sun 3 / 2 6 0 computer.
c T h e solutions of the heuristic presented in this paper and the ones produced by H e r a g u and Kusiak (1991) were improved by one
iteration of a greedy pairwise exchange algorithm. T h e costs and C P U times reported account for such improvement.
d T h e solution produced by the heuristic presented in this paper was improved by a greedy pairwise exchange algorithm until no
further improvement in the cost could be obtained.
72 K.R. Kumar et al. / European Journal of Operational Research 87 (1995) 65-73

The computational experience from the first Armour, G.C., and Buffa, E.S. (1963), "A heuristic algorithm
and simulation approach to relative location of facilities",
set of problems indicates that for problems of low
Management Science 9/2, 294-309.
dimensions (4 facilities), the heuristic's perfor- Bazaraa, M.S., and Kirca, O. (1983), "A branch-and-bound-
mance is slightly superior to Heragu and Kusiak's based heuristic for solving the quadratic assignment prob-
(1988) algorithm and comparable to Hall's (1970) lem", Naval Research Logistics Quarterly 30/2, 287-304.
and Neghabat's (1974) algorithms. Bazaraa, M.S., and Sherali, H.D., (1980), "Benders' partition-
ing scheme applied to a new formulation of the quadratic
The second set of problems consisted of prob-
assignment problem", Naval Research Logistics Quarterly
lems of higher dimensions. In these problems, the 27/1, 29-41.
Heragu and Kusiak (1988) algorithm, although Beghin-Picavet, M., and Hansen, P. (1982), "Deux problemes
being the most computationally efficient, resulted d'affectation non lineaires", RAIRO Recherche Op~ratio-
in inferior costs, whereas the algorithm by Negha- nelle 16/3, 263-276.
Burkard, R.E. (1984), "Quadratic assignment problem", Eu-
bat (1974) required extremely long computational
ropean Journal of Operational Research 15/3, 283-289.
times. The method by Hall was the least compu- Burkard, R.E., and Bonniger, T. (1983), "A heuristic for
tationally expensive for problems of more than 20 quadratic Boolean programs with applications to quadratic
facilities. Nevertheless, out of the 8 problems that assignment problems", European Journal of Operational
Hall's method was tested on, the heuristic ob- Research 13/4, 374-386.
Drezner, Z. (1987), "A heuristic procedure for the layout of a
tained superior solutions in 4 problems in shorter
large number of facilities", Management Science 33/7,
processing time. 907-915.
In the third set of problems, the performance Foulds, L.R. (1983), "Techniques for facilities layout: Decid-
of the heuristic is compared to the methods by ing which pairs of activities should be adjacent", Manage-
Heragu and Kusiak (1991) and Heragu and Alfa ment Science 29/12, 1414-1426.
Foulds, L.R., and Robinson, D.F. (1978), "Graph theoretic
(1992). Both methods are more computationally
heuristics for the plant layout problem", International
expensive than our heuristic. Our heuristic ob- Journal of Production Research 16/1, 27-37.
tained the optimal (when known) or improved Foulds, L.R., Gibbons, P.B., and Giffin, J.W. (1985), "Facili-
costs in less time in 7 of the 8 problems tested ties layout adjacency determination: An experimental
when compared to Heragu and Kusiak (1991) and comparison of three graph theoretic heuristics", Opera-
tions Research 33/5, 1091-1106.
6 of the 8 problems tested when compared to
Giffin, J.W., and Foulds, L.R. (1987), "Facilities layout gener-
Heragu and Alfa (1992). alized model solved by n-boundary shortest path heuris-
Our research can be extended in two direc- tics", European Journal of Operational Research 28/3,
tions. The heuristic can be enhanced by devising 382-891.
a Neghabat-type local comparison procedure Hall, K.M. (1970), "An r-dimensional quadratic placement
algorithm", Management Science 17/3, 219-229.
which does not become 'too' computationally ex-
Heragu, S.S. (1992), "Recent models and techniques for solv-
pensive. Yet another research avenue is to gener- ing the layout problem", European Journal of Operational
alize our heuristic to two-dimensional layout Research 57/2, 136-144.
problems. Heragu, S.S., and Alfa, A.S. (1992), "Experimental analysis of
simulated annealing based algorithms for the layout prob-
lem", European Journal of Operational Research 57/2,
190-202.
Acknowledgments Heragu, S.S., and Kusiak, A. (1988), "Machine layout prob-
lem in flexible manufacturing systems", Operations Re-
The authors thank Professor Sunderesh S. search 36/2, 258-268.
Heragu and the two anonymous referees whose Heragu, S.S., and Kusiak, A. (1991), "Efficient models for the
comments greatly improved this paper. facility layout problem", European Journal of Operational
Research 53/1, 1-13.
Hillier, F.S., and Connors, M.M. (1966), "Quadratic assign-
ment problem algorithms and the location of indivisible
References facilities", Management Science 13/1, 42-57.
Kaku, B.K., and Thompson, G.L. (1986), "An exact algorithm
Adolphson, D., and Flu, T.C. (1973), "Optimal linear order- for the general quadratic assignment problem", European
ing", SIAM Journal on Applied Mathematics 25/3, 403-423. Journal of Operational Research 23/3, 382-390.
K.R. Kumar et al. / European Journal of Operational Research 87 (1995) 65-73 73

Karp, R.M., and Held, M. (1967), "Finite state processes and programming for the Quadratic Assignment Problem",
dynamic programming", SlAM Journal on Applied Mathe- International Journal of Production Research 25/2, 285-
matics 15/3, 693-718. 300.
Kaufman, L., and Broeckx, F. (1978), "An algorithm for the Muller, T. (1983), Automated Guided Vehicles, IFS Publica-
quadratic assignment problem using Benders' decomposi- tions, Kempton, Bedford, UK.
tion", European Journal of Operational Research 2/3, 207- Neghabat, F. (1974), "An efficient equipment layout algo-
211. rithm", Operations Research 22/3, 622-628.
Khalil, T.M. (1973), "Facilities relative allocation technique Nugent, C.E., Vollmann, T.E., and Ruml, J. (1968), "An
(FRAT)", International Journal of Production Research experimental comparison of techniques for the assignment
11/2, 183-194. of facilities to locations", Operations Research 16/1, 150-
Koopmans, T.C., and Beckmann, M. (1957), "Assignment 173.
problems and location of economic activities", Economet- Picard, J., and Queyranne, M. (1981), "On the one-dimen-
rica 25/1, 53-76. sional space allocation problem", Operations Research
Kusiak, A. (1990), Intelligent Manufacturing Systems, 29/2, 371-391.
Prentice-Hall, Englewood Cliffs, NJ. Rosenblatt, M.J. (1979), "The facilities layout problem: A
Kusiak, A., and Heragu, S.S (1987), "The facility layout prob- multigoal approach", International Journal of Production
lem", European Journal of Operational Research 29/3, Research, 17/4, 323-332.
229-251. Seppanen, J., and Moore, J.M. (1970), "Facilities planning
Lawler, E.L. (1963), "The quadratic assignment problem", with graph theory", Management Science 17/4, B242-
Management Science 9/4, 586-599. B253.
Love, R.F., and Wong, J.Y. (1976a), "Solving quadratic as- Seppanen, J.J., and Moore, J.M. (1975), "String processing
signment problems with rectangular distances and integer algorithms for plant layout problems", International Jour-
programming", Naval Research Logistics Quarterly 23/4, nal of Production Research 13/3, 239-254.
623-627. Simmons, D.M. (1969), "One-dimensional space allocation:
Love, R.F., and Wong, J.Y. (1976b), "On solving a one-di- An ordering algorithm", Operations Research 17/5, 812-
mensional space allocation problem with integer program- 826.
ming", INFOR 14/2, 139-143. Simmons, D.M. (1971), "A further note on one-dimensional
Malakooti, B., and D'Souza, G.I (1987), "Multiple objective space allocation", Operations Research 19/1,249.

You might also like