Professional Documents
Culture Documents
To cite this Article Wang (Jian) , Farhad Azadivar John(2000) 'Facility layout optimization using simulation and genetic
algorithms', International Journal of Production Research, 38: 17, 4369 — 4383
To link to this Article: DOI: 10.1080/00207540050205154
URL: http://dx.doi.org/10.1080/00207540050205154
This article may be used for research, teaching and private study purposes. Any substantial or
systematic reproduction, re-distribution, re-selling, loan or sub-licensing, systematic supply or
distribution in any form to anyone is expressly forbidden.
The publisher does not give any warranty express or implied or make any representation that the contents
will be complete or accurate or up to date. The accuracy of any instructions, formulae and drug doses
should be independently verified with primary sources. The publisher shall not be liable for any loss,
actions, claims, proceedings, demand or costs or damages whatsoever or howsoever caused arising directly
or indirectly in connection with or arising out of the use of this material.
INT. J. PROD. RES., 2000, VOL. 38, NO. 17, 4369± 4383
Traditionally, the objective of a facility layout problem has been to minimize the
material handling cost of the manufacturing system. While it is important to
reduce the amount of material handling, the traditional methods do not address
the actual time at which the material is transported. In today’s short cycle time
production environments, the timing of material movement may have a bigger
impact on the productivity of the system than its cost. In this paper, a facility
layout optimization technique is presented that takes into consideration the
dynamic characteristics and operational constraints of the system as a whole,
and is able to solve the facility layout design problem based on a system’s per-
formance measures, such as the cycle time and productivity. Each layout solution
Downloaded By: [Brunel University] At: 20:56 21 July 2010
1. Introduction
The facility layout problem in a manufacturing setting is de® ned as the
determination of the relative locations for, and allocation of, the available space
among a given number of workstations. Although most facility layout solutions
have, in the past, focused on minimizing the amount of transportation , the eŒect
of a given layout design on the production function of a manufacturing system is
much more than just the cost of material handling. While material handling cost
remains critical, shorter cycle times have become much more important in today’s
manufacturing systems. In other words, when a certain material is moved is as
important, if not more important, as how much it costs to move it. Rapid develop-
ment of new products, coupled with short delivery times demanded by customers,
are the bases of the time-based competitive strategies rapidly being adopted by
leading ® rms in many industries. Responsive delivery without ine cient excess
inventory and short manufacturing cycle times are the practical considerations
that have strong impacts on the layout design and should be incorporated into the
layout design process as genuine concerns.
use an integrated solution procedure that optimizes facility layout designs using
simulation as the means of evaluating the objective function. This provides an addi-
tional ¯ exibility in optimization because, in addition to the usual quantitative vari-
ables, evaluation by simulation allows consideration of qualitative decision variables
that analytical objective functions are not equipped to incorporate.
One of the promising methods of optimizing problems whose performances are
evaluated by a simulation model, especially when qualitative variables are involved,
is the use of Genetic Algorithms (GA). Azadivar and Tompkins (1999) proposed a
simulation model generator with a GA-based optimum seeking algorithm capable of
optimizing simulation models whose performances are functions of qualitative and
structural decision variables of the system. Zhang (1997) extended the technique to
more general ¯ exible manufacturing systems. The work presented here is a method-
ology that is based on this approach for facility layout design where the objective
function is a measure of an actual system performance rather than just the volume of
materials handled.
2. Problem statement
Consider a manufacturing system consisting of m workstations in which n types
of parts, each requiring a set of tasks (operations), are to be processed. A work-
station may consist of a single machine, a cell of several machines, an inspection
centre, a paint booth, etc. The parts require processing on diŒerent subsets of the m
workstations and have diŒerent processing times in each workstation. Each work-
station has its own queuing discipline and breakdown distribution. The system is
either a pull or a push type. In addition, let the area of the shop ¯ oor, the area
required by each workstation, the time delay in each workstation, capacity and speed
of the material handling devices, and the precedence constraints of tasks be given. A
desired design for the system requires an arrangement of these workstations into the
shop ¯ oor such that a certain measure of performance is optimized.
The main assumptions for this problem are as follows.
. The work areas of workstations are rectangular in shape and their orientations
are known.
. Every workstation works only one part at a time.
. Every transporter carries only one type of part at a time.
Facility layout optimization 4371
width xi and length yi, must be large enough to accommodate one workstation Mi
plus its clearances.
serial searches, and randomized search methods to perform better than greedy or
enumerative searches. Genetic algorithms combine both of these attributes in a
parallel, stochastic heuristic.
As a powerful and broadly applicable stochastic search and optimization tech-
nique, genetic algorithms have successfully been applied in various areas of industrial
engineering, such as production scheduling and sequencing, reliability design, vehicle
routing and scheduling, group technology, transportation , and many others. The
technique has also been applied to the facility layout problem (Tate and Smith 1995,
Cheng and Gen 1996, Meller and Gau 1996, Tam 1992). However, these published
works are mostly material handling cost driven and do not put enough emphasis on
the performance measures that are time driven and are complex functions of the
layout design (e.g. production rate, cycle time). To evaluate these complex measures,
simulation modelling is often the only feasible method. The approach proposed here,
which combines simulation modelling and a genetic algorithm, provides a unique
opportunity to address this issue.
Downloaded By: [Brunel University] At: 20:56 21 July 2010
1 2
1 2 2 1
2 1
(a) Slicing Structure
* * + +
1 2 2 1 1 2 2 1
(b) Slicing Tree
structure. Let the operation of a horizontal cut and a vertical cut be denoted by the
position symbols * and ‡, respectively. Each symbol is explained pictorially in ® gure
1(b). Each internal node represents the way a rectangular partition is cut. Partitions
reserved for workstations reside in the leaves of the tree. Each leaf is assigned a
unique integer corresponding to the identi® er (id) of a workstation. If we recursively
print out the left subtree and the right subtree, then the position symbol of a slicing
tree, a post® x expression called Reverse Polish Expression, is obtained (® gure1(c)).
This representation method yields itself very well to the coding scheme of chro-
mosomes in genetic algorithms. A chromosome will then have m diŒerent work-
station numbers and …m ¡ 1† position symbols where m is the number of
workstations. Slicing structures comprising m given workstations can be represented
by
P slicing trees or Reverse Polish Expressions over the symbol set
ˆ f1; 2; . . . ; m; *; ‡g (Gen and Chen 1997). Figure 2 demonstrates the process
of constructing a layout from its Reverse Polish Expression for a 6-station facility
layout problem.
Downloaded By: [Brunel University] At: 20:56 21 July 2010
3.2. Crossover
We employed a special form of crossover operation as suggested by Cheng and
Gen (1996) to preserve the feasibility of solutions. In this operation, an oŒspring
chromosome is generated by adopting workstation numbers from one parent and
position symbols from the other. An example of crossover operation is shown in
® gures 3 and 4. Suppose we have two parents, p1 and p2. The crossover operator
copies the workstation numbers from parents p1 into the corresponding positions in
an oŒspring o. Then it copies position symbols from p2, by scanning from left to
right, to complete the oŒspring o.
3.3. Mutation
Random altering, inverting and swapping are used as a mutation operation
(that is, altering a position symbol to the opposite one (® gure 5), inverting a
sequence of adjacent position symbols or a sequence of adjacent facility numbers
(® gure 6), swapping two adjacent position symbols or two adjacent facility numbers
(® gure 7). Mutation performed in this way can also guarantee to generate legal
oŒspring.
3.4. Selection
The task of selection in the genetic algorithms is to allocate the reproductive
opportunities to each chromosome such that the chromosomes with higher ® tness
value are more likely to survive to the next generation. Selection directs a genetic
algorithm search toward promising regions in the search space. The degree to which
the better chromosomes are favoured is de® ned as the selection pressure. Typically,
higher selection pressure indicates that more of the high ® tness chromosomes are
selected.
Tournament selection (Brindle 1981, Goldberg 1989) is a selection approach with
both random and deterministic sampling features. This method randomly chooses a
set of chromosomes and picks out the one with the highest ® tness value (the winner)
for reproduction. The number of chromosomes in the set is called tournament size.
Usually, tournaments are held between pairs of chromosomes (tournament size ˆ 2),
and the selection process is repeated until a desired size of reproduction set has been
formed. The tournament selection is e cient, simple to code, has no scaling problem,
4374 F. Azadivar and J. Wang
+
cut-point
+ *
12+345+6**+ 1 2 3 *
+ 6
4 5
+ *
1 2 3 *
345+6**
(12+)
+ 6
4 5
Downloaded By: [Brunel University] At: 20:56 21 July 2010
*
3
1 2 + 6
45+6*
4 5
3 +
1 2 45+
4 5
1 2 4 5
and is capable of adjusting its selection pressure. This selection method has been
employed in this work.
In a facility layout design process based on simulation and genetic algorithm, the
® tness functions are almost always stochastic. The ® tness value of the chromosome,
which is the output of a simulation experiment, could be viewed as one realization of
a random variable whose mean corresponds to the presumed true response. Since the
selection process is based on ® tness values, random ® tness functions cause the selec-
tion process itself to be random as well. However, the precision of the ® tness values
can be improved by replicating the simulation experiment and obtaining a narrower
Facility layout optimization 4375
P1 1 2 * 3 4 5 * 6 + + *
o 1 2 + 3 4 5 * 6 * * +
P2 3 6 1 + 2 5 4 * * * +
* +
m2
Downloaded By: [Brunel University] At: 20:56 21 July 2010
m1
* + 3 *
m5
m2
1 2 3 + m3 2 *
m4
m4 5
m3 * 6 *
m6 m6 m1
m5 4 +
4 5
(a) Parent 1 p1 6 1
(b) Parent 2 p2
m3 + *
*
1 2 3 *
m1 m2 m4
* 6
m5
4 5
m6
(c) Offspring
con® dence interval. The more replications are performed the better will be the accu-
racy of the solution obtained. Welsh’ s (1938) statistical comparison was employed
for determining the winners of the tournaments.
Since simulation experiments are usually very time consuming, a variable number
of replications per solution are used. This allows us to make fewer replications when
the diŒerence in the ® tness values is relatively large, and to save a larger number of
replications for points where the signal-to-noise ratio is small. The process for using
a variable sample size is as follows.
4376 F. Azadivar and J. Wang
3 + *
1 2 4 5 1 2 3 *
6 + 6
4 5
Parent p1 (12+345+6**+ )
1 2 + 3 4 5 + 6 * * +
Downloaded By: [Brunel University] At: 20:56 21 July 2010
1 2 * 3 4 5 + 6 * * +
(a) Altering operation
3 * *
1
4 5 1 2 3 *
2 6
+ 6
4 5
(b) Offspring o1 after altering
Figure 5. Altering operation.
1 2 + 3 4 5 + 6 * * +
1 2 + 5 4 3 + 6 * * +
(a) Inverting operation
5 + *
1 2 4 3 1 2 5 *
6 + 6
4 3
1 2 + 3 4 5 + 6 * * +
1 2 + 3 4 5 + 6 * + *
(a) Swapping operation
1 2 + +
4 5 1 2 3 *
3
6 + 6
4 5
(b) Offspring o3 after swapping
Downloaded By: [Brunel University] At: 20:56 21 July 2010
. The ® tness values of two chromosomes are ® rst compared based on ® ve repli-
cations of their corresponding simulation models. A test of signi® cance is
performed to make sure the diŒerence between ® tness values is signi® cant. If
so, the chromosome with the inferior ® tness value is considered the loser and
does not move on to the next generation.
. If the diŒerence is not signi® cant, one more replication is made for each
chromosome and the comparison is repeated. This process continues until
either the diŒerence becomes signi® cant (as a result of a reduction in the con-
® dence interval due to the larger number of replications) or a limit of 20
replications per point is reached.
start
initial
comparing
population
chromosomes
yes
find? get fitness evaluation
Downloaded By: [Brunel University] At: 20:56 21 July 2010
no
no
run simulation
converged ?
yes
end
4. System architecture
The system consists of a GA package, a simulation package, an automatic
simulation model generator, and a graphical user interface. The graphical user inter-
face is used to input the information on workstations and parts, dimensional
constraints of the shop ¯ oor and GA parameters. The simulation is considered a
function evaluator (objective function). The genetic algorithm systematically
searches and generates alternative layout designs according to the decision criterion
speci® ed by the user. The simulation model generator then creates and executes
simulation models recommended by the GA and returns the results to the GA.
This iteration between the generator and GA continues until all the chromosomes
in the generation converge to one structure, or the limit on the number of genera-
tions to consider (set according to the available time and resources) is reached.
5. Numerical example
The test problem is described as follows: A manufacturing system consists of
eight workstations, and two lift trucks. Four diŒerent types of parts come into the
system randomly with an inter-arrival time following a certain distribution. The
parts require processing on diŒerent subsets of eight workstations and have diŒerent
distributions of processing times on each operation. Lift trucks are used to move
parts from one workstation to another according to the pre-de® ned processing
sequences for each part type. The area requirements of each workstation are given
in table 1. Tables 2, 3 and 4 provide more information about the manufacturing
system. The shop ¯ oor is a 90 £ 90 (m) square area. The objective is to ® nd a free
layout design for the system to result in an overall shorter average cycle time for all
Facility layout optimization 4379
1 9 17 1 3 10 £ 20
2 18 18 2 2 20 £ 20
3 26 16 4 4 30 £ 20
4 32 9 8 1 40 £ 10
5 45 28 5 2 50 £ 30
6 16 8 4 2 20 £ 10
7 36 17 4 3 40 £ 20
8 16 9 4 1 20 £ 10
Speed
Number (m/min) Policy Capacity
Downloaded By: [Brunel University] At: 20:56 21 July 2010
2 10.0 FCFS 1
1 EXPONENTIAL(12) 1 100 0 4
2 EXPONENTIAL(14) 1 100 0 4
3 CONSTANT(8) 1 100 0 3
4 EXPONENTIAL(14) 1 100 0 4
parts. The evolutionary environment for this GA experiment is given as follows. The
population size is 30, the crossover rate is 0.60, the mutation rate is 0.008, and the
maximum number of generations allowed is 50. A typical solution (the best chromo-
some obtained from one pass of the genetic algorithm) is shown as follows:
2; 1; *; 3; *; 7; 6; *; 4; *; 8; ¤; 5; *; ‡
m5
Downloaded By: [Brunel University] At: 20:56 21 July 2010
m3
m8
Width 90
m4 80
m1
m6
m2 m7
80
Length 90
Evolutionary Process
1800
1600
1400
1200
cycle time
1000 average
cycle time
800
minimum
600 cycle time
400
200
0
1
11
16
21
26
31
36
41
46
gener at io n
From the evolutionary process, we can see that before the 11th generation, the
solution with the minimum cycle time still has a cycle time greater than the ® nal
minimum and the whole generation has an average cycle time of above 1200 minutes.
The worst cycle time found during this period has an average cycle time of 2101.53
minutes. Under pressure of selection, chromosomes evolve gradually while under the
guidance of genetic algorithm the solutions in the search process move slowly to
some promising regions. The average cycle time improves from the initial value of
1665.81 to 948.67 in the 20th generation. The solutions in the set converge gradually
and, after the 25th generation, the variation among the chromosomes in the set
diminishes and all the solutions converge to only one alternative. As shown in the
history of the best ® tness values, the best chromosome is found in the 21st genera-
tion, with an average cycle time of 940.88, which is less than half of the average cycle
time that resulted from the worst system.
Downloaded By: [Brunel University] At: 20:56 21 July 2010
2000
1800
System 1 800
1500
600
1000
Syste 2 400
500 Three Transporters
200
0
0
1 11 21 31 41
1 11 21 31 41
Number of Generations Number of Generations
Downloaded By: [Brunel University] At: 20:56 21 July 2010
(a) (b)
Figure 11. The eŒect of change in system parameters on the optimum layout.
Similar tests were conducted with material handling resources. This time the
manufacturing system was tested with diŒerent numbers of material handling
resources. In the two test cases, the number of transporters was varied while all
other parameters were kept the same. The test results showed that, again the
optimum layout changed with this change. The evolutionary process for this test
is shown in ® gure 11(b).
7. Conclusions
This paper presented an approach for solving facility layout optimization
problems for manufacturing systems with dynamic characteristics and qualitative
and structural decision variables. The proposed approach integrates genetic algor-
ithms, computer simulation and an automated simulation model generator with a
user-friendly interface. Since GA is capable of solving the combinatorial optimiza-
tion problems, and the simulation is capable of modelling and evaluating the per-
formance of complex systems, this combination enables us to optimize e ciently the
facility layout design of such systems. The proposed method considers the opera-
tional policies, resources and time requirements of all aspects of the process to
overcome the limitations of traditional layout optimization methods.
Although this method cannot guarantee an optimum solution, empirical tests indi-
cate it is able to make considerable improvements in the value of the objective
function.
Additional work in this area can improve the performance of the process. In
particular, to preserve the feasibility of a solution, the search proposed in this work
uses a particular crossover strategy that allows only the position symbols to trade
places. Other techniques can be investigated to take full advantage of all crossover
methods that can still preserve feasibility. Furthermore, other methods of mutation
can also be investigated.
Facility layout optimization 4383
References
A ZADIVAR, F . and T OMPKINS, G ., 1999, Simulation optimization with quantitative variables
and structural model changes: a genetic algorithm approach. European Journal of
Operational Research, 113, 169± 182.
BRINDLE, A ., 1981, Genetic algorithms for function optimization. PhD dissertation,
University of Alberta.
C HENG, R . and G EN , M ., 1996, Genetic search for facility layout design under inter¯ ows
uncertainty. Japanese Journal of Fuzzy Theory and Systems, 8, 335± 346.
G EN , M . and C HENG, R ., 1997, Genetic Algorithms and Engineering Design (Wiley).
G OLDBERG , D . E ., 1989, Genetic Algorithms in Search, Optimization, and Machine Learning
(MA: Addison-Wesley).
H OLLAND, J . H ., 1975, Adaptation in Natural and Arti® cial Systems (Ann Arbor: The
University of Michigan Press).
M ELLER, R . D . and G AU, K . Y ., 1996, The facility layout problem: recent and emerging
trends and perspectives. Journal of Manufacturing Systems, 15, 351± 366.
R OSENBLATT, M ., 1986, The dynamic of plant layout. Management Science, 32, 76± 85.
T AM, K . Y ., 1992, Genetic algorithms, function optimization, and facility layout design.
European Journal of Operational Research, 63, 322± 346.
Downloaded By: [Brunel University] At: 20:56 21 July 2010
T ATE, D . and SMITH , A ., 1995, A genetic approach to the quadratic assignment problem.
Computers and Operations Research, 22, 73± 83.
W ELCH, B. L ., 1938, The signi® cance of the diŒerence between two means when the popula-
tion variances are unequal. Biometrika, 25, 350± 362.
Z HANG, J ., 1997, Genetic algorithm based simulation optimization of ¯ exible manufacturing
systems. PhD dissertation, Kansas State University.