You are on page 1of 16

PLEASE SCROLL DOWN FOR ARTICLE

This article was downloaded by: [National Institute of Technology]


On: 20 October 2010
Access details: Access Details: [subscription number 919625261]
Publisher Taylor & Francis
Informa Ltd Registered in England and Wales Registered Number: 1072954 Registered office: Mortimer House, 37-
41 Mortimer Street, London W1T 3JH, UK
International Journal of Production Research
Publication details, including instructions for authors and subscription information:
http://www.informaworld.com/smpp/title~content=t713696255
Development of the MAIN algorithm for a cellular manufacturing machine
layout
W. M. Chan
a
; C. Y. Chan
a
; C. K. Kwong
a
a
Department of Industrial & Systems Engineering, The Hong Kong Polytechnic University, Kowloon,
Hong Kong
To cite this Article Chan, W. M. , Chan, C. Y. and Kwong, C. K.(2004) 'Development of the MAIN algorithm for a cellular
manufacturing machine layout', International Journal of Production Research, 42: 1, 51 65
To link to this Article: DOI: 10.1080/00207540310001598456
URL: http://dx.doi.org/10.1080/00207540310001598456
Full terms and conditions of use: http://www.informaworld.com/terms-and-conditions-of-access.pdf
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., 2004, vol. 42, no. 1, 5165
Development of the MAIN algorithm for a cellular manufacturing
machine layout
W. M. CHANy, C. Y. CHANy* and C. K. KWONGy
When a static machine layout is simply optimized to a xed quantitative demand,
it may not be able to cope with demand uctuations. The development of a
heuristic approach called the MAIN (Machines Allocation INter-relationship)
algorithm for intracellular machine layout design is introduced. It begins by
analysing a single period with a xed quantitative demand and machine assign-
ment is based on a set of proposed objective functions together with merging
techniques. To simulate the demand uctuations in a multiple periods planning
horizon, layouts are generated to t dierent demand proles. Each layout is then
processed further. First, at every periodic transition, there is re-layout to minimize
the material-handling cost. Second, we will examine the condition of using a
layout right through all the periods to keep away from the machine rearrange-
ment cost with some scarications of the material-handling cost. Comparisons
will be done to seek out good machine layout(s). In terms of a single-period
layout, the experimental results show that the proposed heuristic algorithm can
achieve an average deviation of 4.8% in contrast with optimal solutions.
However, to opt for re-layout or not depends very much on the relations between
the machine rearrangement and material-handling costs. Moreover, the variations
in the qualitative aspects in material handling will also be considered as it has
signicant inuence on the design of the machine layout.
1. Introduction
Group Technology (GT) has been applied to generate machine groups and
corresponding part families for cellular manufacturing (CM). In an ideal situation,
each cell is independent of another and, in practice, the main idea is to reduce the
intercellular movements. This results in an intracellular ow dominance arrange-
ment, and there are often similarities in processes in a cell (Chan et al. 1999,
Yasuda and Yin 2001). To have the full benets of CM, ecient layout design is
a key element (Massoud 1999, Taho and Brett 1999). In a manufacturing cell, each
part family has its own operating sequences and, in reality, the quantitative demands
can also vary periodically. This paper is to investigate the Dynamic Machine Cellular
Layout (DMCL) in an environment with multiple periods planning horizon.
Initially, the machine layout will be generated based on a xed quantitative
demand prole and this is referred to the Static Machine Cellular Layout (SMCL).
Each static layout makes a good t for a planning period but, in real situation, the
product mix is a function of demand, and it may well change as demand uctuates
Revision received June 2003.
y Department of Industrial & Systems Engineering, The Hong Kong Polytechnic
University, Hung Hom, Kowloon, Hong Kong.
*To whom correspondence should be addressed. e-mail: mfcychan@inet.polyu.edu.hk
International Journal of Production Research ISSN 00207543 print/ISSN 1366588X online # 2004 Taylor & Francis Ltd
http://www.tandf.co.uk/journals
DOI: 10.1080/00207540310001598456
D
o
w
n
l
o
a
d
e
d

B
y
:

[
N
a
t
i
o
n
a
l

I
n
s
t
i
t
u
t
e

o
f

T
e
c
h
n
o
l
o
g
y
]

A
t
:

0
4
:
4
6

2
0

O
c
t
o
b
e
r

2
0
1
0
(Seifoddini and Djassemi 1997). Nicol and Hollier (1983) also mentioned that the
changes of radical layout happened frequently and management should always con-
sider this in the forward planning. DMCL is an attempt to address the demand
changes with unchanged product operating sequences in a cell.
The qualitative ow dominance represents considerable changing attributes
in parts whilst the quantitative ow dominance means that the movements (e.g.
transportation frequencies and distances) of the parts between machines are more
signicant. In fact, most of the current approaches pay little attention to the quali-
tative aspects. In this paper, the MAIN (Machines Allocation INter-relationship)
algorithm is formulated to optimize the machine layout by considering both the
qualitative and the quantitative characteristics of a cell. This algorithm will address
the problem in view of the machine rearrangement cost and the part travelling cost.
In addition, the total part travelling score will be used as the SMCL performance
indicator and the dynamic layout(s) in DMCL is obtained by searching the layouts
generated by SMCL to cope with the multiple periods planning horizon.
In brief, the objective of this research is to exploit the machine layout with
varying periodic demand in a CM environment. The intention is to minimize the
total part travelling cost by taking into the consideration of the closeness of
machines, the handling eciency, the part travelling distance, the materials ow
frequency and the machine rearrangement cost in a dynamic environment. First,
we will look into how to assign machines within a cell eectively based on several
factors like the operation sequences, the part ow frequencies, and the customer
demands of a part family in order to determine the required closeness of machines
in the layout. Then, the total travelling score will be calculated by accumulating
values of all related pair-wise machines, each of which incorporates information
such as the distance, the frequency of part ow, and the part handling eciency.
As mentioned before, this score is also used to represent the SMCL performance for
comparison purpose. In terms of DMCL, the ultimate goal is to shrink the overall
layout cost and this can be achieved by having a good balance between the total
machine rearrangement cost and the total part travelling cost.
2. Literature review
Static machine layout problems have being studied for decades and the solution
layout is only good t for a planning period. Currently, most of the available
techniques tackle static layout problems by minimizing the total travelling cost of
parts. This can be done by working out a suitable arrangement for machines under
predened locating zones (Tanchoco and Lee 1999, Urban et al. 2000). For example,
the Quadratic Assignment Problem (QAP) technique uses the number of locations as
the number of facilities, and Koopmans and Beckmann (1957) are the pioneers of the
QAP technique. However, the QAP is NP-complete, and the size of a problem that
can be solved by optimal methods is very limited (Kusiak and Heragu 1987). In
SMCL, if n machines are grouped into a cell, there are as many as n! potential
layouts. In case of DMCL, the same machine group may have to be rearranged to
suit for m periods in a planning horizon. As the result, the probable combina-
tions are as large as (n!)
m
and it is dicult to obtain the optimal solution based
on QAP approach (Rosenblatt 1986, Balakrishnan and Chun 1998). Indeed, purely
mathematical iterations will not be a feasible method for dealing with this sort of
problems (Jajodia et al. 1992) and, moreover, QAP usually works only with little
constraints. Instead, the heuristic approach can be useful in SMCL and DMCL.
52 W. M. Chan et al.
D
o
w
n
l
o
a
d
e
d

B
y
:

[
N
a
t
i
o
n
a
l

I
n
s
t
i
t
u
t
e

o
f

T
e
c
h
n
o
l
o
g
y
]

A
t
:

0
4
:
4
6

2
0

O
c
t
o
b
e
r

2
0
1
0
Although heuristic approach cannot guarantee optimal result, it requires relatively
little time and eort to achieve an acceptable solution (Heragu and Kusiak 1988).
Apart from using the QAP technique, Montreuil (1990) developed a static layout
solving method using mixed-integer linear programming, which focuses on the load/
unload point along aisles. Tam and Li (1991) also formulated a three-phase
hierarchical approach that attempted to minimize the geometric constraint.
Nevertheless, the key limitation of static layout is that it only focuses on a single
planning period and will not be able to full the changing customer demands today.
The dynamic layout problem usually assumes that machines could be economi-
cally relocated in order to cope with a new demand prole. Rosenblatt (1986) rst
introduced a layout problem-solving technique for multiple planning periods. His
dynamic programming model attempts to nd out the best layout for each prede-
termined period. The aim is to minimize the sum of deterministic rearrangement cost
through the entire planning periods. One limitation of Rosemblatts approach is the
cost of rearrangement, which does not consider the distance from a machine location
zone to another. Secondly, the discrete best layouts may not give a good overall
performance and the assumption that machines could always economically relocated
is also debatable. To resolve the problem, Kouvelis et al. (1992) presented a similar
algorithm to determine a layout, which would suit for multiple planning periods,
and, of course, this resulted in some scarications in the material-handling cost.
Rather than switching from two extremes, there are often some rooms in between
worth exploring. Actually, constraints always exist in layout design. For example,
Balakrishnan et al. (1992) added the budget constraint for machine rearrange-
ments and solved the problem by using the shortest path algorithm. In fact, other
realistic constraints such as the location restrictions, designers preferences, etc. are
indispensable in determining machine layout (Massoud 1999). Still, there is not much
concern in the product attributes. Clearly, the design of material-handling devices
plays a very important role in the material-handling system and the operational
eciencies can quite vary with dierent handling facilities. Thus, the eort spent
on moving parts should not be uniformly assumed. In terms of layout determination,
Conway and Venkataramanan (1994) applied the genetic search technique to evolve
a population with a premature initial solution. Optimal solution layouts were also
computed for comparison purpose. Two cases were tested by this approach, and an
optimum could be obtained in the case with six machines. It hints that with a
reasonable complicated problem, looking for optimal solution will be impractical.
Again, heuristic algorithms are usually more suitable for cases of larger size.
Based to the studies, intracellular machine layout problem associated with
changing in part attributes in both SMCL and DMCL will be essential to look
into, and the signicance of machine rearrangement cost in a multiple periods
planning horizon will also be a sensible issue to be examined. In addition, the
material-handling eciency notably aecting the layout design should be included
to reect the real-world situations.
3. Problem denitions
This research aims on developing a method to undertake the static and the
dynamic machine layouts in a manufacturing cell where the machine location site
is a regular grid; similar to Tanchoco and Lee (1999) and Urban et al. (2000), each
machine cluster contains machines requiring equal oor space areas. The maximum
number of machines can be assigned into a grid is limited to nine machines as a
53 MAIN algorithm for a cellular manufacturing machine layout
D
o
w
n
l
o
a
d
e
d

B
y
:

[
N
a
t
i
o
n
a
l

I
n
s
t
i
t
u
t
e

o
f

T
e
c
h
n
o
l
o
g
y
]

A
t
:

0
4
:
4
6

2
0

O
c
t
o
b
e
r

2
0
1
0
3 3 grid is going to be used and there are only nine potential machine location
zones on it (gure 1). This is considered to be adequate for a machine cell as the
average machine size in a cell is around seven machines. For the model, we also
assume that a machine can be assigned to any location and the part travelling
distance is estimated by measuring the rectilinear distance from the centre of the
source machine to the centre of the destination machine. This implies the travelling
distance between machines j and k will be identical, and the size of any machine
location zone (x, y) should be able to accommodate the largest machine in a cell.
SMCL is designed to work with a xed quantitative demand for a given part
family. Besides targeted on the part ow between pair-wise machines such as
machines with larger number of in between ow would be allocated more closely,
the variations in part handling eciency will also counted. As mentioned, it is
important as the attributes of a part will change from process to process. For
example, in an assembly cell, a part can have changes in size, weight, shape and
so on. It is not a surprise to see an initial 1 kg part increased to 5 kg after some
assembly operations or, the opposite, a nished part may reduce in weight in com-
parison with its earlier state if there is a material removal action involved. As a
result, even though the quantitative demand of a part remains unchanged, the best
possible layout can be dierent if the part handling eciency is taken into account.
DMCL is the extension of SMCL. It aims at obtaining machine layout that goes
well with periodic variations. In the MAIN algorithm, DMCL is further divided
into two branches namely the multiple-DMCL (mDMCL) and the single-DMCL
(sDMCL). mDMCL works with changing layouts whilst sDMCL keeps a unique
layout over periods. In the former case, machine locations will be rearranged as
the part ow frequencies change so as to maintain the lowest total part travelling
scores in the subsequent periods (the lowest the score the better the layout).
Generally, four potential quantitative demand proles in a planning horizon
would have four possible optimal layouts ideally. However, owing to the similarity
in processing natures in a cell, we may have identical layout for a succeeding plan-
ning period. Of course, this depends on the conditions but the potential is always
here. To measure the performance of the proposed algorithm, the total cell layout
cost is used as the performance indictor in the dynamic stage while the total part
travelling score is employed in the static stage. The formulation of the total cell
layout cost includes two basic elements: the total machine rearrangement cost and
j k
Measuring
points
Measuring
distance
n = 3
Machine
m = 3
x
y
Maximum size of
machine location
Figure 1. Nine zones location site.
54 W. M. Chan et al.
D
o
w
n
l
o
a
d
e
d

B
y
:

[
N
a
t
i
o
n
a
l

I
n
s
t
i
t
u
t
e

o
f

T
e
c
h
n
o
l
o
g
y
]

A
t
:

0
4
:
4
6

2
0

O
c
t
o
b
e
r

2
0
1
0
the total part travelling cost over all periods. The total part travelling cost is devised
from the part travelling score, which is determined by the proposed objective func-
tions to be discussed in the next section. In terms of the machine rearrangement cost,
gure 2 shows nine machines at the rst period ( p1). If machines E and A are going
to be swapped to suit for the second planning period ( p2), the rearrangement cost
(R
1 !2
) must be introduced into the total machine rearrangement cost at the
dynamic state.
In some cases, machine relocation may not be practical due to prohibitive
rearrangement cost on moving machines and so on. Consequently, only the same
machine layout will be used for throughout all periods. Under these circumstances,
some relinquishes in part moving distances have to be made in order to keep the
layout and this is where the sDMCL takes place. Obviously, there is no machine
rearrangement cost but the part travelling cost may be higher.
4. MAIN algorithm
This section provides a brief description of the mathematical model presented by
Chan et al. (2002); the interested reader can refer to Chan et al. for details and
explanations about the handling of the SMCL. Basically, SMCL covers the data
collection and matrix-based data manipulations along with the application of the
formulated objective functions. Then, pair-wise machines are inserted into the 3 3
space grids in accordance with the ranking orders. Possible layouts are evaluated and
the total part travelling scores calculated to represent the preferences of the machine
layouts. SMCL can produce a good layout for a single period while DMCL needs
a series of static layouts for multiple planning periods correspondingly, or a parti-
cular layout in case of machine rearrangement is impractical. Figure 3 shows the
skeleton of the MAIN model.
R
total
= R
1 2
+R
2 3
+R
3 4
(> 0)
p-1
A B C
D E F
G H I
p-2
E B C
D A F
G H I
p-3
F B C
D E A
G H I
p-4
A F C
D E G
B H I
Changing in quantitative demands on parts
There is machine rearrangement costs incurred in four periods (p-1 to p-4).
Figure 2. mDMCL with four planning periods.
SMCL mDMCL
sDMCL
Dynamic Stage Static Stage
Figure 3. Skeleton of the conceptual MAIN model.
55 MAIN algorithm for a cellular manufacturing machine layout
D
o
w
n
l
o
a
d
e
d

B
y
:

[
N
a
t
i
o
n
a
l

I
n
s
t
i
t
u
t
e

o
f

T
e
c
h
n
o
l
o
g
y
]

A
t
:

0
4
:
4
6

2
0

O
c
t
o
b
e
r

2
0
1
0
4.1. Static stage: SMCL
The static stage of the MAIN algorithm is to determine the total part travelling
score (
p
) in a cell in a period ( p) with a particular layout. The purpose is to minimize
the total travelling score and this is analogue to the cost on making part movements.
The objective function is dened as:
min
p

X
m
j1
X
m
kj1
D
j $k
F
pj $k
1
where
D
j $k
X
j
X
k

Y
j
Y
k

: 2
The Manhattan method has been chosen for calculating the distance (D
j $ k
)
between two points where (X
j
, Y
j
) and (X
k
, Y
k
) are the coordinates of the measuring
points of machines j and k with reference to the 3 3 grid.
F
p, j $x
T
p, j !k
T
p, k !j
, 3
where F
p, j $ k
is the merged part ow weight for period p. It combines the unidirec-
tional basic part ow weights (T
p, j !k
and T
p,k !j
) to form a generic correlation of
two machines. A basic part ow weight (e.g. T
p, j !k
) is the mean by which parts are
transported from one machine to another in unidirectional (say, from machine j to
machine k) within a cell. The characteristics of intracellular part ows are that they
are usually rushed and short distances. These weights directly aect the positioning
of machines. In other words, a pair of machines with higher merged part ow weight
should be more closely placed. The determination of the basic and the merged part
ow weights are intimately related to the quantitative demand of a part (i) in a family
in a period (Q
p,i
) and the part transportation quantity per move (H
i, j !k
). The later is
needed because some transportation devices may have the ability to carry several
items in one transaction. This suggests that the transportation frequency is equal to
the quantitative demand of a part divided by the transportation quantity per move.
Lastly, the part-handling factor (l
i, j !k
) is introduced to signify the levels of
diculty in moving a part between a machine pair. In this research, a simple relative
mode is used to quantify this factor. For examples, 1 represents the easiest in
transporting and 2 means double the eort of 1, etc. The equation for determine
the basic part ow weight is:
T
p, j !k

X
n
i1
Q
p, i
H
i, j !k
l
i, j !k
4
4.2. Dynamic stage: DMCL
At this point, two situations would be faced, namely the mDMCL and the
sDMCL in a multiple periods planning horizon. Figure 3 illustrates the two
alternative routes of the MAIN algorithm. In mDMCL, the part travelling cost
and the machine rearrangement cost are needed to be determined. In case of the
travelling cost, it will be obtained by converting the total travelling scores generated
in SMCL into monetary terms; fortunately, this is a linear conversion only. The
function used for calculating the total cell layout cost () over multiple periods is:

X
P
p1

p
!
X
P1
p1
R
p $p1
5
56 W. M. Chan et al.
D
o
w
n
l
o
a
d
e
d

B
y
:

[
N
a
t
i
o
n
a
l

I
n
s
t
i
t
u
t
e

o
f

T
e
c
h
n
o
l
o
g
y
]

A
t
:

0
4
:
4
6

2
0

O
c
t
o
b
e
r

2
0
1
0
There are at most three machine rearrangement costs involved in a four periods
planning horizon. ! represents the cost per unit travelling score and it is a constant.
However, the establishment of this value must be taken reference to the formulation
of the part-handling factor. In the period-to-period machine rearrangement cost
(R
p !p 1
), it consists of two portions; they are the basic cost for machine relocation
such as the set-up/installation cost (M
j
) and the cost per unit machine movement
(C
j
). The calculation of this cost is based on:
R
p !p1

X
m
j1
M
j
D
j $k
C
j
6
Note that R
p !p 1
is equal to 0 if all machines remain in the same locations
because M
j
0 and D
j $ k
0.
Remember, sDMCL tends to keep all machines stationary. This signies that the
layout determined by sDMCL may not be the optimal solution in all periods. In the
MAIN algorithm, the cell layout cost derived from sDMCL is the aggregation of the
part travelling cost over all periods. This is done by evaluating every static layout
generated in SMCL to nd out the best one, which is going to be used as the nal
layout for a cell. The equation for working out the total part travelling cost for each
SMCL generated layout is:
Total part travelling cost !
X
n
p1

0
p
7
For example, if there are four periods, then SMCL may generate four layouts, each
of which ts for a particular period. The iteration of using each of these four layouts
exclusively throughout all periods yields new sets of part travelling scores (
0
p
), and
the new part travelling costs can be found. Normally, one of them will have the
lowest value and this becomes the selected layout of sDMCL.
5. Numerical analysis establishment
Numerical information provided by established research cases in machine layout
is used to evaluate the proposed MAIN algorithm. Some related data are obtained
from Yaman et al. (1993) and Tang and Abdel-Malek (1996). In essence, both
Yaman et al. and Tang and Abdel-Malek used the same data set. Besides, Yaman
et al. also used a 3 3 grid for locating machines. In the process operation side, there
were ve parts processed by nine machines. For each part in this family, the opera-
tional sequence and quantitative demands in a ve-period planning horizon were
proposed (tables 1 and 2). Although they provided some basic requirements, there
was still information missing for the MAIN algorithm to operation. These include
the cost per unit part movement, the part transportation quantity per move, the part-
handling factor and those that related to machine rearrangement, etc. Therefore, we
assigned a xed cost (! $10/unit travelling score) to deal with part movements. To
test the eect of the part-handling factor, a set of part-handling factors had also been
proposed (table 3); parts 1 and 3 gradually decease in part-handling factors, whilst
the others increased. Table 4 shows the machine rearrangement costs. They are
required in the conditions with multiple planning periods.
In the static stage, all the part transportation quantity per move and the part-
handling factors were set to 1s. This is required with the purpose of fullling the
needs of the MAIN and also is capable of going for comparisons to see the eect of
57 MAIN algorithm for a cellular manufacturing machine layout
D
o
w
n
l
o
a
d
e
d

B
y
:

[
N
a
t
i
o
n
a
l

I
n
s
t
i
t
u
t
e

o
f

T
e
c
h
n
o
l
o
g
y
]

A
t
:

0
4
:
4
6

2
0

O
c
t
o
b
e
r

2
0
1
0
having taken into the consideration of part-handling eort. As a result, the part ow
weights (F
p, j $ k
) generated by applying the static stage of the MAIN when dealing
with the period ( p1) are presented in the third column in table 5. After the part ow
weights were ranked, the layout designer can start pulling the machines to form a
3 3 machine layout grid according the rankings. Once the pulling operations have
been completed, the distances among machines (D
j $ k
) can be obtained by referring
to the Manhattan method (equation 2). Then, the travelling score for each machine
pair can be determined accordingly and the summation of these travelling scores give
Part
Period
p-1 p-2 p-3 p-4 p-5
1 10 35 90 40 55
2 30 50 25 65 20
3 45 15 40 70 15
4 70 80 55 90 85
5 85 60 70 20 30
Table 2. Demand proles in Yaman et al.s and Tang et al.s cases.
Part Machine operational sequence
1 01 !03 !05 !07 !02 !07 !09
2 01 !04 !02 !05 !06 !08 !09
3 01 !05 !07 !08 !05 !06 !02 !09
4 01 !02 !04 !06 !07 !08 !02 !03 !09
5 01 !07 !06 !04 !02 !08 !03 !05 !06 !09
Table 1. Original data of Yamana et al.s and Tang et al.s cases.
Part Part-handling factor
1 6 !5 !4 !3 !2 !1
2 1 !2 !3 !4 !5 !6
3 6 !5 !4 !4 !3 !2 !1
4 1 !1 !2 !2 !2 !3 !3 !4
5 1 !2 !3 !4 !5 !5 !5 !5 !6
Table 3. Proposed part-handling factors for
Yaman et al.s and Tang et al.s cases.
Machine 01 02 03 04 05 06 07 08 09
Basic cost for machine relocation (M
j
) for case ve
($) 150 100 200 320 180 90 240 130 90
Cost per unit machine movement (C
j
) for case ve
($) 15 10 20 32 18 9 24 13 9
Table 4. Proposed machine rearrangement costs for Yaman et al.s and
Tang et al.s cases.
58 W. M. Chan et al.
D
o
w
n
l
o
a
d
e
d

B
y
:

[
N
a
t
i
o
n
a
l

I
n
s
t
i
t
u
t
e

o
f

T
e
c
h
n
o
l
o
g
y
]

A
t
:

0
4
:
4
6

2
0

O
c
t
o
b
e
r

2
0
1
0
the total travelling score (
p
) of this layout. The second iteration begins by examining
the part travelling scores. For example, table 5 shows that machines 0708 (F
1,7 $ 8
)
contributed a signicant inuence to the part travelling score (see the greyed row)
and this was due to the lengthy distance (D
7 $ 8
) in between. As the result, it is likely
to make some improvements by bringing machines 07 and 08 closer. The rule of
thumb for doing this part of the job is that that one should always check for those
with high merged part ow weights and far distances (say, D
j $ k
>2) to seek for
improvements. To simplify the discussions, the layouts for all ve periods are shown
in tables 6 and 7 for the rst and the second iterations, respectively. Note that
layouts suggested by the second iteration may not always be better than the rst
iteration, such as in p4 and p5. The layout designer can make the selection easily by
looking at the gures and, moreover, based on our experiences, two iterations are
adequate to provide very good solutions.
Two more test cases had been created arbitrarily with 10 parts and nine
machines. In each test case, ve periods with varying quantitative demands were
generated to simulate the production uctuations (tables 8 and 9). In table 10,
operational sequences and part-handling factors were also proposed for the self-
developed test cases. Finally, the basic costs for machine relocation and cost per
unit machine movement were assumed as in tables 11 and 12.
Ranking ( j $k) F
p, j $k
D
j $k
Iteration 1
(F
p, j $k
D
j $k
) D
j $k
Iteration 2
(F
p, j $k
D
j $k
)
1 02,04 185 1 185 1 185
2 05,06 160 1 160 1 160
3 02,08 155 1 155 1 155
04,06 155 1 155 1 155
06,07 155 1 155 1 155
4 07,08 115 4 460 2 230
5 03,05 95 1 95 1 95
6 01,07 85 1 85 1 85
03,08 85 1 85 1 85
06,09 85 3 255 2 170
7 01,02 70 2 140 2 140
02,03 70 2 140 2 140
03,09 70 1 70 2 140
8 05,07 55 2 110 2 110
9 01,05 45 1 45 3 135
02,06 45 2 90 2 90
02,09 45 3 135 4 180
05,08 45 2 90 2 90
10 01,04 30 3 90 1 30
02,05 30 1 30 3 90
06,08 30 3 90 1 30
08,09 30 2 60 3 90
11 02,07 20 3 60 3 60
12 01,03 10 2 20 4 40
07,09 10 2 20 1 10

1
2980 2850
* Greyed values are machines 07, 08 with high part-travelling score at Iteration 1.
Table 5. Part travelling scores in the rst planning period
( p-1, l
i, j $k
1, H
i, j $k
1).
59 MAIN algorithm for a cellular manufacturing machine layout
D
o
w
n
l
o
a
d
e
d

B
y
:

[
N
a
t
i
o
n
a
l

I
n
s
t
i
t
u
t
e

o
f

T
e
c
h
n
o
l
o
g
y
]

A
t
:

0
4
:
4
6

2
0

O
c
t
o
b
e
r

2
0
1
0
Part
Period
Total p-1 p-2 p-3 p-4 p-5
1 100 200 300 700 900 2200
2 200 610 800 440 650 2700
3 340 300 980 670 320 2610
4 450 450 450 450 450 2250
5 590 370 740 870 1100 3670
6 500 210 180 120 80 1090
7 770 800 900 530 790 3790
8 1120 1500 1400 1600 1210 6830
9 960 300 200 720 530 2710
10 440 110 680 90 880 2200
Table 8. Demand proles for self-developed case one [S1].
Part
Period
Total p-1 p-2 p-3 p-4 p-5
1 1200 1450 1670 1890 1940 8150
2 400 520 460 930 210 2520
3 980 460 270 380 740 2830
4 450 450 450 450 450 2250
5 610 220 1000 880 370 3080
6 270 160 150 340 130 1050
7 650 790 530 460 230 2660
8 1720 1440 1380 1690 2130 8360
9 810 150 130 240 1170 2500
10 410 320 60 700 530 2020
Table 9. Demand proles for self-developed case two [S2].
p-1 p-2 p-3 p-4 p-5
09 07 07 09 03 05 09 08 06 07 05 06 05 07 06
03 05 06 01 07 06 01 03 05 08 02 04 08 02 04
08 02 04 08 02 04 04 02 07 09 01 03 03 01 09

1
2980
2
2880
3
3710
4
3165
5
2275
* Greyed values are the lowest total part-travelling score at each period.
Table 6. First iterative results of the MAIN algorithm ( p-1 to p-5, l
i, j $k
1, H
i, j $k
1).
p-1 p-2 p-3 p-4 p-5
02 04 01 02 04 01 01 08 03 01 02 04 01 02 04
08 06 07 08 06 07 02 07 05 09 08 06 09 08 06
03 05 09 03 05 09 04 09 06 03 07 05 03 05 07

1
2850
2
2790
3
3160
4
3225
5
2475
* Greyed values are the lowest total part-travelling score at each period.
Table 7. Second iterative results of the MAIN algorithm ( p-1 to p-5, l
i, j $ k
1,
60 W. M. Chan et al.
D
o
w
n
l
o
a
d
e
d

B
y
:

[
N
a
t
i
o
n
a
l

I
n
s
t
i
t
u
t
e

o
f

T
e
c
h
n
o
l
o
g
y
]

A
t
:

0
4
:
4
6

2
0

O
c
t
o
b
e
r

2
0
1
0
6. Results and analysis
For evaluation purposes, MATLAB was used to work out the optimal solutions
for static layouts using an exhaustive search. However, there is little chance to get the
optimal layouts for DMCL due to the very large number of combinations involved.
This is also why a heuristic approach is the feasible technique to determine the
dynamic layouts. Referring to the information presented in tables 6 and 7, the
static layouts generated by the SMCL of MAIN for all ve periods are given in
table 13 based on the data sets from Yaman et al.s and Tang and Abdel-Maleks
cases. Table 14 summarizes the results from various approaches. Note that all
preliminary results generated by the proposed MAIN algorithm were better than
Yaman et al.s approaches, although some results might not be more superior to
Tang and Abdel-Maleks method such as in planning periods p1 and p4. Among all
tested heuristics, the proposed algorithm was best in overall performance and the
average error was least, i.e. it was only 4.8% in comparison with the optimal
solutions. Furthermore, in operating MAIN, one important feature was virtually
disabled at this stage as all part-handling factors were set to 1 s. This was done
Part
type Operational sequence Part-handling factor
1 01 !04 !06 !04 !05 !03 !09 5 !5 !4 !3 !2 !1
2 01 !06 !04 !07 !08 !09 4 !2 !2 !1 !1
3 01 !05 !03 !02 !05 !09 1 !1 !2 !3 !4
4 01 !02 !03 !05 !04 !06 !09 1 !2 !3 !3 !4 !4
5 01 !08 !04 !06 !05 !02 !03 !05 !09 1 !1 !1 !1 !1 !2 !2 !2
6 01 !03 !04 !07 !09 7 !6 !5 !4
7 01 !06 !05 !02 !06 !08 !03 !09 1 !1 !1 !2 !2 !3 !3
8 01 !02 !04 !06 !05 !03 !05 !09 1 !1 !1 !1 !1 !2 !2
9 01 !04 !05 !07 !03 !08 !09 4 !3 !2 !1 !1 !1
10 01 !07 !08 !02 !06 !04 !05 !03 !09 5 !4 !3 !2 !1 !2 !3 !4
Table 10. Operational sequences and part-handling factors for [S1] and [S2].
M/C 01 02 03 04 05 06 07 08 09
Basic cost for machine relocation (M
j
) for S1
($) 780 650 930 820 620 670 770 910 860
Cost per unit machine movement (C
j
) for S2
($) 78 65 93 82 62 67 77 91 86
Table 11. Machine rearrangement cost for [S1].
M/C 01 02 03 04 05 06 07 08 09
Basic cost for machine relocation (M
j
) for S1
($) 940 640 1100 450 390 780 990 420 520
Cost per unit machine movement (C
j
) for S2
($) 94 64 110 45 39 78 99 42 52
Table 12. Machine rearrangement cost for [S2].
61 MAIN algorithm for a cellular manufacturing machine layout
D
o
w
n
l
o
a
d
e
d

B
y
:

[
N
a
t
i
o
n
a
l

I
n
s
t
i
t
u
t
e

o
f

T
e
c
h
n
o
l
o
g
y
]

A
t
:

0
4
:
4
6

2
0

O
c
t
o
b
e
r

2
0
1
0
because other approaches had not considered the aect of this parameter and the
comparison might not be fair if it was included.
The sDMCL looks for a sole layout for all periods by picking the most
appropriate static layout. Since there are ve periods, ve static layouts produced
by the SMCL will be examined. Naturally, the one with a minimum overall total
part-travelling score should be chosen; to ease the comparison with mDMCL, we
multiplied them with the cost per unit travelling score (!) to yield the total cell layout
cost. Referring to table 15, the smallest value is 145 700; it was the static layout
generated by SMCL for period p5 and it is called SMCL
5
in this case.
In mDMCL, all SMCL generated static layouts were used to minimize the parts
travelling due to periodic changes and the machine rearrangement cost is brought in
to nance the re-layout. We also observed that in some cases it might not be
necessary to rearrange the machines if the proposed successive layout was also
identical to the one in use. The chance of having this is always there as parts assigned
to a cell are from the same family. Similarly, in this case, there was no machine
rearrangement cost incurred between p1 and p2 (table 16).
For better illustration, the overall results are summarized in table 17. Note
that the total cell layout cost of the mDMCL (146 351) was greater than the best
layout generated by sDMCL (145 700) when using Yaman et al.s and Tang and
Abdel-Maleks data sets, in which part-handling factors were set to 1s. Therefore,
sDMCL (grey) should be applied in this case. Moreover, the gures generated by
including part-handling factors were signicantly larger (up to two-to-three times in
our cases) than those with the part-handling factors disabled. This also indicates that
the eect of the part-handling factor can be striking and this will surely inuence
planning of the machine layout. In addition, referring to the rst self-developed case
[S1] in table 17, it is clear that if the part-handling factors were taken into account,
the decision would be quite dierent. In this case, the most suitable layout strategy
changed from sDMCL to mDMCLs.
Method
Part travelling score
Per cent of
deviation
p-1
(
1
)
p-2
(
2
)
p-3
(
3
)
p-4
(
4
)
p-5
(
5
)
Total
(
P
)
Optimal 2780 2640 2950 3020 2200 13 590
Yaman et al. (1) 3630 3180 3690 3975 3045 17 520 28.9
Yaman et al. (2) 3470 3350 3570 4065 2975 17 430 28.3
Tang et al. 2820 2980 3200 3100 2355 14 455 6.4
MAIN 2850 2790 3160 3165 2275 14 240 4.8
Table 14. Part traveling scores generated by various approaches.
p-1 p-2 p-3 p-4 p-5
02 04 01 02 04 01 01 08 03 07 05 06 05 07 06
08 06 07 08 06 07 02 07 05 08 02 04 08 02 04
03 05 09 03 05 09 04 09 06 09 01 03 03 01 09

1
2850
2
2790
3
3160
4
3165
5
2275
Table 13. Static layouts generated by MAIN (l
i, j $ k
1, H
i, j $ k
1).
62 W. M. Chan et al.
D
o
w
n
l
o
a
d
e
d

B
y
:

[
N
a
t
i
o
n
a
l

I
n
s
t
i
t
u
t
e

o
f

T
e
c
h
n
o
l
o
g
y
]

A
t
:

0
4
:
4
6

2
0

O
c
t
o
b
e
r

2
0
1
0
7. Discussion
Although we did not incorporate the data for part transportation quantity in this
research, the establishment of the part-handling factor could imitate the importance
of the part transportation quantity. In practice, both can be valuable when reecting
true situations. In view of the MAIN algorithm, the number of machines and the
orientation of the location zones should not be the denite constraints. In addition,
more than nine machines and irregular machine disposition zones are also possible.
However, these require further investigation. Readers interested in this area can
refer to Massoud (1999) and Taho and Brett (1999).
Total part travelling
cost ( !) ($)
Period
Total
(a) p-1 p-2 p-3 p-4 p-5
Static layout SMCL
1
28 500 27 900 35 100 36 100 25 300 152 900
SMCL
2
28 500 27 900 35 100 36 100 25 300 152 900
SMCL
3
31 400 29 700 31 600 35 150 24 650 152 500
SMCL
4
32 600 30 200 35 900 31 650 24 550 154 900
SMCL
5
29 400 28 200 31 700 33 650 22 750 145 700
Minimum total
cell layout cost ()
(b) Total machine rearrangement cost 0
(a) (b)
145 700
*Grayed values are generated static layouts by SMCL.
Table 15. Total cell layout costs for sDMCL.
Data Set
Total cell layout cost
sDMCL
(l 1)
mDMCL
(l )
sDMCL
(l vary)
mDMCL
(l vary)
Average
variations between
(l 1) and (l vary)
Tang et al.s
Yaman et al.s
145 700 146 351 456 550 457 326 3.13 times
S1 3 243 600 3 548 260 7 060 400 6 983 514 2.07 times
S2 2 899 000 3 180 960 6 052 000 6 057 874 1.99 times
* Greyed values are the lowest total cell layout costs.
Table 17. Summary of total cell layout costs.
Period p-1 p-2 p-3 p-4 p-5 Total
Total part
travelling cost
( !) ($)
28 500 27 900 31 600 31 650 22 750 142 400 (a)
Machine
rearrangement
cost (R
p !p 1
) ($)
g 0 1829 1660 1671 g 3951 (b)
Minimum total
cell layout cost ()
(a) (b) 146 351
Table 16. Total cell layout cost from mDMCL.
63 MAIN algorithm for a cellular manufacturing machine layout
D
o
w
n
l
o
a
d
e
d

B
y
:

[
N
a
t
i
o
n
a
l

I
n
s
t
i
t
u
t
e

o
f

T
e
c
h
n
o
l
o
g
y
]

A
t
:

0
4
:
4
6

2
0

O
c
t
o
b
e
r

2
0
1
0
Referring to the analysis, one may observe that the ratios of total part travelling
cost to total cell layout cost and of total machine rearrangement cost to total cell
layout cost are very important factors when deciding on layout tactics (e.g. sDMCL
or mDMCL). Hence, future research could involve the investigation of these
ratios and the development of a methodology for the quick determination of a
proper layout strategy could save much calculation eort. Additionally, Taho and
Brett (1999) presented an idea of putting the probability of occurrence on demand
proles. This can be valuable when designing a machine layout if there are some
kinds of uncertainties. In fact, this always happens in real life. Basically, MAIN
can absorb probability into the objective functions and this is surely another area
worth investigating.
At rst, we tended to take out the machine rearrangement cost from the layout
from the dynamic layouts to yield a lower bound to assist in the searching for better
solutions. However, it was also noticed that the eect of the machine rearrangement
cost was not signicant in our cases. Moreover, due to similarities in processing
a part family, it was also observed that in a lot of cases an identical layout could
be used eectively throughout multiple periods. However, this can act against our
intention of expanding the use of the MAIN algorithm to handling more than nine
machines. This is because a layout with more than nine machines cannot follow the
characteristics of CM anymore and, thus, the similarity in processes cannot exist in
these cases.
8. Conclusions
Machine layout problems in manufacturing have received considerable
attention. For instance, using a computer can hardly solve problems with more
than 15 machines in the static stage with limited constraints, according to a literature
review. MAIN can overcome this problem as it is based on a heuristic approach.
The proposed MAIN algorithm addresses the problems in both static and
dynamic layouts by incorporating practical factors such as the part-handling
factor, the basic cost for machine relocation and the cost per unit machine
movement, etc. MAIN makes a best compromise of the above-mentioned factors
with the closeness of machines in order to minimize the total cell layout cost. This
algorithm also caters for the possibility of allowing a single layout for multiple
periods with demand uctuation. The algorithm works well for a maximum of
nine machines, which are assigned into a 3 3 matrix-like layout by using a pulling
technique based on a set of objective functions developed under MAIN. Referring to
the experimental results, the MAIN algorithm achieved an average of 95.2%
performance (4.8% deviation) in comparison with optimal solutions in the case of
a static layout. In terms of multiple planning periods, the selection of either sDMCL
or mDMCL depends on the proportion of the total machine rearrangement cost
and the total part travelling cost. Moreover, the role of the part-handling factor in
the layout design is inevitable. The dierence in cell layout cost could be more than
three times in the cases we studied when the part-handling factor was involved. In
addition, it was assumed that the transportation quantity per move (H
i, j !k
) kept
constant and relied on the part-handling factor to simulate the eect. Finally, it can
be observed that the layout decision can be totally dierent if this factor is taken
into consideration.
64 W. M. Chan et al.
D
o
w
n
l
o
a
d
e
d

B
y
:

[
N
a
t
i
o
n
a
l

I
n
s
t
i
t
u
t
e

o
f

T
e
c
h
n
o
l
o
g
y
]

A
t
:

0
4
:
4
6

2
0

O
c
t
o
b
e
r

2
0
1
0
Acknowledgements
The work was substantially supported by a grant from the Department of
Industiral & Systems Engineering, The Hong Kong Polytechnic University,
Research Grants Council of the Hong Kong Special Administrative Region
(Project No. G-V524).
References
BALAKRISHNAN, J. and CHUN, H. C., 1998, Dynamic layout algorithms: a state-of-the-art
survey. Omega, International Journal Management Science, 26, 507521.
BALAKRISHNAN, J., JACOBS, F. R. and VENKATARAMANAN, M. A., 1992, Solution for the con-
strained dynamic facility layout problem. European Journal of Operational Research,
57, 280286.
CHAN, C. Y., LAM, F. W. and LEE, C. P., 1999, Considerations for using cellular manufactur-
ing. Journal of Materials Processing Technology, 96, 182187.
CHAN, W. M., CHAN, C. Y. and IP, W. H., 2002, A heuristic algorithm for machine assignment
in cellular layout. Computer and Industrial Engineering, 44, 4973.
CONWAY, D. G. and VENKATARAMANAN, M. A., 1994, Genetic search and the dynamic facility
layout problem. Computers Operations Research, 21, 955960.
HERAGU, S. S. and KUSIAK, A., 1988, Machine layout problem in exible manufacturing
systems. European Journal of Operations Research, 36, 258268.
JAJODIA, S., MINIS, I., HARHALAKI, S. G. and PROTH, J. M., 1992, CLASS: Computerized
Layout Solutions using simulated annealing. International Journal of Production
Research, 30, 95108.
KOOPMANS, T. C. and BECKMANN, M., 1957, Assignment problems and the location of eco-
nomic activities. Econometrica, 25, 5376.
KOUVELIS, P., KURAWARAWALA, A. A. and GUTIERREZ, G. J., 1992, Algorithms for robust single
and multiple period layout planning for manufacturing systems. European Journal of
Operational Research, 63, 287303.
KUSIAK, A. and HERAGU, S. S., 1987, The facility layout problem. European Journal of
Operations Research, 29, 229251.
MASSOUD, B.-L., 1999, Layout designs in cellular manufacturing. European Journal of
Operational Research, 112, 258272.
MONTREUIL, B., 1990, A modelling framework for integrating layout design and ow network
design. In Proceedings of the 1990 Material Handling Research Colloquium, Material
Handling Institute.
NICOL, L. M. and HOLLIER, R. H., 1983, Plant layout in practice. Material Flow, 1, 177188.
ROSENBLATT, M. J., 1986, The dynamics of plant layout. Management Science, 32, 7685.
SEIFODDINI, H. and DJASSEMI, M., 1997, Determination of a exibility range for cellular man-
ufacturing systems under product mix variations. International Journal of Production
Research, 35, 33493366.
TAHO, Y. and BRETT, A. P., 1998, Flexible machine layout design for dynamic and uncertain
production environments. European Journal of Operational Research, 108, 4964.
TAM, K. Y. and LI, S. G., 1991, A hierarchical approach to the facility layout problem.
International Journal of Production Research, 29, 165184.
TANCHOCO, J. M. A. and LEE, A. C., 1999, Cellular machine layout based on the segmented
ow topology. International Journal of Production Research, 37, 10411062.
TANG, C. and ABDEL-MALEK, L. L., 1996, A framework for hierarchical interactive generation
of cellular layout. International Journal of Production Research, 34, 21332162.
URBAN, T. L., CHIANG, W. C. and RUSSEL, R. A., 2000, The integrated machine allocation and
layout problem. International Journal of Production Research, 13, 29112930.
YAMAN, R., GETHIN, D. T. and CLARKE, M. J., 1993, An eective sorting method for facility
layout construction. International Journal of Production Research, 31, 413427.
YASUDA, K. and YIN, Y., 2001, A dissimilarity measure for solving the cell formation problem
in cellular manufacturing. Computer and Industrial Engineering, 39, 117.
65 MAIN algorithm for a cellular manufacturing machine layout
D
o
w
n
l
o
a
d
e
d

B
y
:

[
N
a
t
i
o
n
a
l

I
n
s
t
i
t
u
t
e

o
f

T
e
c
h
n
o
l
o
g
y
]

A
t
:

0
4
:
4
6

2
0

O
c
t
o
b
e
r

2
0
1
0

You might also like