Professional Documents
Culture Documents
net/publication/265702361
CITATIONS READS
8 336
2 authors:
Some of the authors of this publication are also working on these related projects:
Development of efficient algorithms for solving computationally expensive optimization problems View project
All content following this page was uploaded by Ruhul Sarker on 29 September 2014.
Tapabrata Ray
School of Aerospace, Mechanical and Civil Engineering
University of New South Wales
Australian Defence Force Academy, Canberra, Australia
Ruhul Sarker
School of Information Technology and Electrical Engineering
University of New South Wales
Australian Defence Force Academy, Canberra, Australia
Abstract. Machine layout and material flow between machines are crucial
considerations for improving productivity in any manufacturing environment.
The machine layout and the operations assignment problems are both known
to be NP hard problems. In this paper, we consider a combined machine
layout and job assignment problem and introduce an evolutionary algorithm to
solve this combined problem. The usefulness of our approach is demonstrated
through numerical examples.
1. Introduction. Machine layout design has been regarded as the key to improve
productivity in any manufacturing system. An improved layout design reduces the
manufacturing lead time and increases throughput [1, 2, 3]. The primary objective
of machine layout problems is to obtain the most effective machine arrangement
(relative locations of machines) and the path of material flow so that the overall
system cost (usually machining cost plus the material handling cost) is minimized.
The major types of machine arrangements, considered in practice, are process based,
the flow-line (also known as single line), the multi-line, the semi-circular and the
loop layout. The details on these layouts can be found in [4]. Based on the pattern
of material flow, these layouts can be grouped as flow shop and job shop. In the
flow shop layout, all products usually require similar sequence of operations. As
a result, the machines are physically arranged either as single straight line, semi-
circular or in a loop. The flow of material in such arrangements is unidirectional
from the first to the last operation. In job shop, the physical arrangement is usually
multi-line where different jobs require different sequence of operations. As the
frequent rearrangement of machines is an infeasible task, the jobs (with a known
order of operations) need to move in different directions for the required operations
in different machines.
2000 Mathematics Subject Classification. Primary: 58F15, 58F17; Secondary: 53C35.
Key words and phrases. Machine layout, task-machine assignment, optimization, multi-
objective, genetic algorithm.
Corresponding Author: r.sarker@adfa.edu.au, Tel no.: +61 2 6268 8051, Fax no.: +61 2 6268
8581 .
631
632 TAPABRATA RAY AND RUHUL SARKER
The machine layout problem is a well-known NP hard problem [6]. In the past,
a number of attempts have been made to solve such problem using mathematical
programming approaches([12, 11]). Such approaches require a substantial amount
of computational effort when the problem size is large. The purpose of this research
is to develop a general methodology for determining the optimal layout of machines
which is capable of dealing with any layout patterns. Earlier work on layout prob-
lems involving genetic algorithms (GA) is briefly reported here. Mak et al [6] studied
the use of GA in solving machine layout problems by simplifying the material flow
within a manufacturing plant. El-Baz [4] used GA to study different material flow
systems within a manufacturing facility. Ficko et al [7] presented a GA for designing
a flexible manufacturing system with one or multiple rows. Chen and Ho [5] used
GA for solving a production planning problem under flexible manufacturing system
environment. Balakrishnan et al [8] developed a GA based methodology for solving
static plant layout problem with up to 30 departments.
In this paper, we consider a problem where the entire shop floor is divided into
a number of cells of equal area (either square or rectangular) within the layout
arrangement under consideration. The number of cells could be more than the
number of machines in which case there would be empty cells. Each machine can
perform P operations. Our aim is to find out the minimum cost solution where cost
is the sum of the transport cost and the machining cost. The solution would answer
the following questions.
El-Baz [4] considered a problem where the problem is to assign machines to loca-
tions within a given layout arrangement by minimizing the total material handling
costs. However, Chen and Ho [5] assumed that the machine locations are fixed and
given, and each machine is capable of performing multiple operations. Their prob-
lem was to assign the operations of given product(s) to machines by minimizing the
overall flow time and machine workload. In the problem, the flow time is assumed
to be the machining time plus material handling time. In both problems, the order
of operations is given. Our problem is a combined problem of machine layout
and production planning. In other words, it is a generalized problem which
combines both El-Baz [4] and Chen and Ho [5] problems with additional considera-
tions. It is to assign machines of multiple operational capability to locations within
a given layout arrangement and to assign operations to individual machines for mul-
tiple products and the machine sequence for the products while minimizing the sum
of machining and material handling costs. Our problem definition is different from
Urban et al. [15] and much more complex and generalized. In fact, their situation is
a simple case of our problem. Although these two interrelated problems (machine
layout and operations assignment) can be solved sequentially, a combined approach
for determining the machine locations and operations assignment simultaneously
would reduce the chances of obtaining sub-optimal solutions.
EA FOR COMBINED MACHINE LAYOUT AND JOB ASSIGNMENT 633
2. Mathematical model. There are basically two types of problems that we need
to consider
1. Layout problems with conventional machines
2. Layout problems with flexible machines
For the first class of problems with conventional machines, a product may use
either all the machines in the shop floor or a subset of the machines, and it may
visit a given machine more than once before completion. The sequence (/order)
of operations (/machine used) for each product and the operation time of each
product on each machine is known. The material handling equipment is given
(for example, Automated Guided Vehicle, AGV) and the cost of material handling
between machines can be calculated based on distance and/or other criteria. For a
given mix of products, the total material handling cost is dependent on the location
of individual machines. So the layout problem is to assign the machines to cell
locations within a given layout arrangement such that the material handling cost
is minimum. It is noted that the layout problem is a long term decision problem
which is usually made based on aggregate information rather than a short term
schedule.
For the second class of problems i.e. with flexible machines, we assume that
each machine is capable of performing all or a subset of operations required by
the products. However, there is a considerable amount of setup time and cost for
switching from one operation to another and the operation time required for a given
product varies from machine to machine. As a result, it is economic to setup the
machines less frequently. The problem is then to assign the location of the machines
and also their mode of operation. In fact, the first class of problems of machine
layout can be thought as a special case of the second where the mode of operation of
each machine is fixed. As a result, we intend to develop an algorithm to solve both
machine layout and machine operation assignment (production planning) problem.
The layout and production planning problem under consideration can be defined as
follows.
1. The term cell is used to describe a machine centre or location. The number
of cells, cell area and their arrangements are given.
2. The term machine represents a modern machine tool such as Numerical Con-
trolled (NC) machine tools. The machines could be either identical or non-
identical in terms of the type and number of different operations to be per-
formed.
3. A machine is capable of performing several types of operations and an oper-
ation can be performed on alternative machines.
4. A machine can perform only one operation at any time and operations to be
performed in the machine are non-preemptive. Only one operation for each
machine (mode of operation) would be chosen in solving the problem.
5. A product requires a number of operations. A product may choose one or
more machines at each stage of its operations and the transportation of the
products between the machines is handled by a material handling system (we
assume AGV).
634 TAPABRATA RAY AND RUHUL SARKER
6. There may be more than one product (or product type) with different opera-
tions and order of operations.
7. The distances between machines can be calculated using the cell co-ordinates.
Currently it is the sum of traversal in x and y coordinates. The material
handling times can be calculated based on the vehicle velocity and distance.
(Note that one can replace this part with a given time matrix).
8. The times and costs of operations in each machine for any product are deter-
ministic and known.
9. The layout arrangement can be considered of any type.
The machine layout problem addressed here is the assignment of N machines to
M cell locations in a given layout arrangement. The production planning problem
considers performing L operations of each product on N available machines. During
the manufacturing process, the material flows from one machine to the next until all
the operations are completed. So the objective of the general layout and production
planning problem is to minimize the sum of all machining and material handling
costs of the manufacturing system.
For convenience and easy understanding, we number all the cell locations from 1
to N . The examples for numbering for one-line linear, multi-line square and multi-
line rectangular arrangements in a two dimensional floor space(o,p) are shown in
Tables 1-3.
Tables 1-3 show a single line layout with 5 cells, 3 × 5 rectangular layout with
15 cells, and 4 × 4 square layout with 16 cells respectively. As per our problem
definition, if the number of machines is less than the number of cells (i.e., N < M ),
then (M − N ) cells will remain empty.
2.1. Layout Problem 1. As indicated earlier, El-Baz [4] considered a conventional
machine layout problem which is to assign machines to locations within a given
layout arrangement by minimizing the total material handling costs. Each machine
EA FOR COMBINED MACHINE LAYOUT AND JOB ASSIGNMENT 635
would perform only one operation. In this section, we consider a similar problem
except that (i) the layout arrangement can be of any type, (ii) the number of cells
(/locations) can be greater than the number of machines to be located and (iii)
the order of machines that would be visited by each job is given. To formulate the
mathematical model for this problem, the following notations are used.
Q X
X N X
M X
N X
M
Minimize Z = Pqik Cqjl Djl Xij Xkl (1)
q=1 i=1 j=1 k=1 l=1
This is a quadratic function. The value of Pqik will be positive if any positive
quantity of product q required to be transported from a machine in location j to
a machine in location l that is required by product q. Otherwise the value of Pqik
will be zero. Note that the order of operations for each product is known and we
assume that there exists only one path for each product. The values Djl are always
positive. The term (Xij Xkl ) is equal to one if there exist machines in both locations
j and l.
The first constraint ensures that each location contains at best one machine.
N
X
Xij ≤ 1 j = 1, . . . , M (2)
i=1
The second constraint ensures that each machine is assigned to exactly one lo-
cation.
M
X
Xij = 1 i = 1, . . . , N (3)
j=1
The following condition is to make sure that the decision variables are binary (0
or 1) variables.
Xijs = 0 or 1 ∀i, j (4)
636 TAPABRATA RAY AND RUHUL SARKER
1 if machine i is located in cell location j and performing operation s
Xijs =
0 otherwise
The mathematical model is presented below.
The objective is to minimize the sum of machining cost in each machine and
material handling cost over every pair of machines.
Q X
X N X
M X
N X
M X
S X
S
Minimize Z = Pqst (Cqjl Djl + M Cqis ) Xijs Xklt (5)
q=1 i=1 j=1 k=1 l=1 s=1 t=1
This is a quadratic function. The value of Pqst will be positive if any positive
quantity of product q required operation s from machine i located in j would be
transported to machine k in location l for next operation t that is as per the se-
quence of operations for product q. Otherwise the value of Pqst will be zero. Note
that the sequence of operations for each product is known. The values Djl are
always positive. The term (Xijs Xklt ) is equal to one if there exist machines in both
locations j and l, and they perform operations s and t respectively.
The first constraint ensures that each location contains at best one machine.
N
X
Xijs ≤ 1, j = 1, . . . , M and s = 1, . . . , S (6)
i=1
EA FOR COMBINED MACHINE LAYOUT AND JOB ASSIGNMENT 637
The second constraint ensures that each machine is assigned to exactly one lo-
cation.
M
X
Xijs = 1, i = 1, . . . , N and s = 1, . . . , S (7)
j=1
The third constraint ensures that each machine in a given location is to perform
only one operation.
S
X
Xijs ≤ 0, i = 1, . . . , N and j = 1, . . . , M (8)
s=1
The following condition is to make sure that the decision variables are binary (0
or 1) variables.
This constraint can be relaxed by allowing each machine to switch from operation
to operations with switching /setup costs.
The above mathematical programming model is a Quadratic Assignment Prob-
lem (QAP).
2.3. Layout Problem 3. The above two models represent single objective lay-
out problems. For a given mix of jobs and fixed layout, some machines may take
longer to complete their tasks than other machines. This means higher load on
some machines which results a longer makespan for some jobs. So we have consid-
ered another objective to balance the workload on machines which would minimize
the sum of {(Average load or machining time per machine) - (Machining time in
individual machine)}2 . This can be expressed mathematically as follows:
N Q X N X M XN X
M XS X S
X 1 X
Z2 = Pqst M Tqis Xijs Xklt −
N
i=1 q=1 i=1 j=1 s=1 t=1
k=1 l=1
2
Q X
X M X
N X
M X
S X
S
Pqst M Tqis Xijs Xklt (10)
q=1 j=1 k=1 l=1 s=1 t=1
Q X
X N X
M X
N X
M X
S X
S
Minimize Z1 = Pqst (Cqjl Djl + M Cqis ) Xijs Xklt (11)
q=1 i=1 j=1 k=1 l=1 s=1 t=1
638 TAPABRATA RAY AND RUHUL SARKER
N Q X N X M X N X M X S X S
X 1 X
Minimize Z2 = Pqst M Tqis Xijs Xklt −
N
i=1 q=1 i=1 j=1 k=1 l=1 s=1 t=1
2
X Q XM X N XM X S X S
Pqst M Tqis Xijs Xklt (12)
q=1 j=1 k=1 l=1 s=1 t=1
Subject to Costraints 6 - 9
From the mathematical programming point of view, the first few models are
constrained nonlinear integer programming [17] and the last model is a multiobjec-
tive constrained nonlinear integer programming model. Further on multiobjective
programming can be found in [16].
3.1. Initialization. It is important to take note that all possible types of opera-
tions required by the products are atleast available in one machine i.e. the machine
states. This ensures that the process of initialization always creates feasible solu-
tions.
Operation Problem #
1 2 3 4
Number of products 1 1 2 2
Number of operations required 10 5 5 5
Quantity of product 1 100 100 100 100
Quantity of product 2 - - 200 200
2. The second one is for change of states and machine locations. Select a
crossover point which lies in between the second segment of the variable vec-
tor. Create two children from the parents P 1 and P 2. The crossover point is
only considered valid if the machine state vectors in both the children contain
all operations required by the products atleast once. Repair the child solu-
tions states and the machine sequence part of the variable vector to ensure
feasibility.
Column Column
Row 1 2 3 Row 1 2 3
1 7 8 1 1 2 2 2
2 2 3 6 2 2 3 4
3 9 4 5 3 1 2 5
of 0.2, 0.4 and 0.6. The above combinations will result in 96 trials. From these
96 trials, we have obtained consistently the same optimal solutions for all four test
problems. The solutions of the test problems are presented below.
As per the solutions, the machine 9 is located in cell (3, 1), machine 4 in cell (3,
2) and so on (see Table 7(a)) and the machine 9 would perform operation 1, machine
4 operation 2 and so on (see Table 7(b)). The product is using five machines for five
different operations. The machines are 3, 4, 5, 6 and 9 (bold italic in Table 7(a)) and
they perform the operations 3, 2, 5, 4 and 1 (bold italic in Table 7(b)) respectively.
The product is using five machines for five different operations. The machines
are 4, 5, 6, 8 and 9 (bold italic in Table 8(a)) and they perform the operations 2,
3, 4, 5 and 1 (bold italic in Table 8(b)) respectively.
Column Column
Row 1 2 3 Row 1 2 3
1 8 6 3 1 5 4 4
2 4 5 7 2 2 3 5
3 9 1 2 3 1 5 2
Column Column
Row 1 2 3 Row 1 2 3
1 2 7 8 1 1 1 2
2 9 5 4 2 1 5 2
3 3 6 1 3 3 4 1
For the first product, the machines used are 1, 3, 4, 5 and 6 (bold italic in
Table 9(a)) and they perform the operations 1, 3, 2, 5 and 4 (bold italic in Ta-
ble 8(b)) respectively. For the second product, the machines used are 3, 4, 5, 6 and
9 (see Table 9(a)) and they perform the operations 3, 2, 5, 4 and 1 (see Table 8(b))
respectively. Four out of five machines are common in both products.
4.4. Test Problem 4. After running the bi-objective problem, only four non-
dominated solutions, were found. These are: (50833.33, 5458.22), (53166.67, 2972.89),
(59833.333, 2128.00), and (62000.00, 1667.56). These solutions are shown in the
Figure 1.
The detailed results for the above four alternative points on the nondominated
front are given in Table 10.
The machine load balancing objective forces to use as many machines as possible.
As per the bi-objective solutions, three out of four alternative solutions suggest using
of all nine machines and the first solution uses eight out of nine machines, where as
the single objective (as seen above) solutions suggest to use only five machines for
five operations.
4.5. Solution quality. As the complete enumeration would be too expensive, we
have compared our solutions for the single objective test problems with a large
number of randomly generated solutions. We have generated two million random
solutions for the first three test problems and reported the best fitness values in
Table 11. When comparing the best solutions, our algorithm provides superior
solutions for problem 1 and problem 3, but inferior solution for problem 2. In our
experimentation, we allowed a maximum of 20,000 (with population size 100 and
maximum generation number 200) fitness evaluation as opposed to 2 million in
random solutions. By increasing the number of fitness evaluations for problem 2 to
80,000 (with population size 200 and maximum generation number 400), we are able
644 TAPABRATA RAY AND RUHUL SARKER
6000
5500
4500
4000
3500
3000
2500
2000
1500
5 5.2 5.4 5.6 5.8 6 6.2 6.4
Cost 4
x 10
to obtain a fitness value of 16,500 which is same as the best of 2 million random
solutions. This comparison demonstrates the fact that the quality of solutions
produced by our algorithm is quite good.
single objective and one bi-objective test problems. The interesting features of this
research are discussed below.
1. Any layout arrangement can be considered (such as linear, square, rectangular,
circular or any other arrangement) and any size with given material handling
facilities. In our test problems, we have considered a rectangular arrangement.
2. The number of machines can be less than the number of cells available. In our
test cases, we have considered the number of machines equal to the number
of cells. Current work does not put any limit on the number of machines and
/or cells.
3. The machines could be either conventional machines or modern machine tools.
They could be either identical or non-identical in terms of the type and number
of different operations that they can perform. In our test cases, we have
considered non-identical NC machine tools.
4. A product may choose one or more machines at each stage of its operations and
the transportation of the products between the machines. We have considered
that different operations must be performed in different machines.
5. A product may visit a machine more than once (see test problem 1 above)
before completion.
6. There may be more than one product (or product type) with different opera-
tions and order of operations (see test problem 2 above).
7. The distances between machines can be calculated using the cell co-ordinates
if the distance matrix is not given. In our cases, we have calculated based on
cell co-ordinates.
The algorithm is designed to optimize the problem by selecting only one operation
for each machine (mode of operation) irrespective of the number of products to be
processed and their operations requirements. However we intend to incorporate the
option of switching the mode of operations from product to product as our future
work. We would also like to include the scheduling aspect to compute the total
machining time. In addition, we like to extend our current work by using different
objective function such as throughput minimization, imposing constraints such as
capacity and time windows, and treating as a multiobjective problem.
REFERENCES
[1] W. A. Reynolds and M. K. Cheung, The use of industrial engineering in Hong Kong manu-
facturing industry, International Journal of Production Research, 22 (1984), 983–999.
[2] N. L. Hyer and U. Wemmerlov, Group technology in the US manufacturing industry: A survey
of current practice, International Journal of Production Research, 27 (1989), 1287–1304.
[3] U. Wemmerlov and N. L. Hyer, Cellular manufacturing in US industry: a survey of users,
International Journal of Production Research, 27 (2000), 1511–1530.
[4] M. Adel El-Baz, A genetic algorithm for facility layout problems of different manufacturing
environment, Computers & Industrial Engineering, 47 (2004), 233–246.
646 TAPABRATA RAY AND RUHUL SARKER
[5] J. H. Chen and S. Y. Ho, A novel approach to production planning of flexible manufacturing
systems using an efficient multi-objective genetic algorithm, International Journal of Machine
Tools & Manufacture, 45 (2005), 949–957.
[6] K. L. Mak, Y. S. Wong and F. T. S. Chan, A genetic algorithm for facility layout problems,
Computer Integrated Manufacturing Systems, 11 (1998), 113–127.
[7] M. Ficko, M. Brenzocnik and J. Balic, Designig the layout of single and multiple rows flexible
manufacturing system by genetic algorithm, Journal of Materials Processing Technology, 157-
158 (2004), 150–158.
[8] J. Balakrishnan, C. H. Cheng and K. F. Wong, FACOPT: A user friendly FACility layout
OPTimization System, Computers & Operations Research, 30 (2003), 1625–1641.
[9] K. Deb, A. Pratap, S. Agarwal and T. Meyarivan, A fast and elitist multi-objective genetic
algorithm: NSGA-II, IEEE Transactions on Evolutionary Computation, 6 (2002), 181–197.
[10] K. Deb, A. Pratap, S. Agarwal and T. Meyarivan, A fast and elitist multi-objective ge-
netic algorithm: NSGA-II, in “Proceedings of Parallel Problem Solving from Nature, Paris,
France,”(2000), 849–858.
[11] R. F. Love and J. Y. Wong, Solving quadratic assignment problems with rectangular distances
and integer programming, Naval Research Logistics Quarterly, 23 (1976), 623–627.
[12] J. F. Pierce and W. B. Crowston, Tree search algorithms in quadratic assignment problems,
Naval Research Logistics Quarterly, 18 (1971), 1–36.
[13] R. Sarker, J. Kamruzzaman and C. Newton, Evolutionary Optimization(EvOpt): a brief
review and analysis, International Journal of Computational Intelligence and Applications, 3
(1984), 311–330.
[14] R. Sarker, K. Liang and C. Newton, A new evolutionary algorithm for multiobjective opti-
mization, European Journal of Operational Research, 40 (2002), 12–23.
[15] T. L. Urban, W. C. Ching and R. A. Russell, The integrated machine allocation and layout
problem, International Journal of Production Research, 38 (2000), 2911–2930.
[16] X. Yang, X. Yang and K. L. Teo, Higher-order symmetric duality in multiobjective program-
ming with invexity, Journal of industrial and management optimization, 4 (2008), 385–391.
[17] Y. Yang, Z. Wu and F. Bai, A filled function method for constrained nonlinear integer pro-
gramming, Journal of industrial and management optimization, 4 (2008), 353–362.
Received June 2007; 1st revision November 2007; 2nd revision May 2008.
E-mail address: t.ray@adfa.edu.au
E-mail address: r.sarker@adfa.edu.au