Formation of machine groups and part families

:
a modified SLC method and comparative study
Hassan M. Selim
Department of Business Administration, College of Business and Economics,
United Arab Emirates University, Al Ain, United Arab Emirates
Reda M.S. Abdel Aal
Faculty of Engineering, King Abdulaziz University, Jiddah, Saudi Arabia
Araby I. Mahdi
Faculty of Engineering, Helwan University, Cairo, Egypt
Introduction
International competitiveness and market
demand for rapid response have led many
low to medium volume manufacturing firms
to consider alternative approaches to the
design and control of manufacturing
systems. One approach that appeared in the
1970s is the application of cellular
manufacturing (CM). In CM, similar parts
are grouped into part families and dissimilar
machines are grouped into machine groups
or manufacturing cells. The ideal
manufacturing cell:
.
is independent and requires no inter-cell
movements, i.e. part family(ies) are
completely produced within single
manufacturing cell;
.
has minimum setups; and
.
requires minimal backtracking.
The result is simplified scheduling, control,
and implementation of automation
(Vakharia and Selim, 1994). CM provides the
benefits of a mass manufacturing system for
a discrete batch manufacturing system
including reduced setup times, work in
progress, throughput time, and material
handling as well as encouraging improved
product quality.
The problem of designing CM systems, i.e.
identification of machine groups and
corresponding part families, is known as
manufacturing cell formation (MCF)
problem. Several taxonomies and reviews of
the MCF problem exist in the literature (e.g.
Shambu et al., 1996; Joines et al., 1996; Selim
et al., 1998). Over the past three decades, a
number of different approaches to solve the
MCF problem have been proposed (e.g.
Stanfel, 1985; Joines et al., 1996; Selim et al.,
1998; Vakharia et al., 1999). One of the MCF
solution approaches is to deal with the MCF
problem as a formation of clusters in a (0, 1)
machine-part incidence matrix or using
similarity or dissimilarity measures between
machines or parts. This category of
approaches is referred to as cluster analysis
or cluster formation based techniques.
This paper introduces an enhancement of
one of the MCF cluster formation-based
techniques. The modified heuristic has been
compared and evaluated against three
well-established MCF methods using four
published performance measures. A total of
35 problem data sets have been used in the
comparative and evaluative study. In the
next section, the literature of the MCF cluster
formation-based methods is reviewed.
Literature review
The MCF cluster formation based techniques
can be classified as:
.
array-based clustering techniques;
.
hierarchical clustering techniques; and
.
non-hierarchical clustering techniques.
The array-based clustering techniques try to
allocate machines to cells and parts to
corresponding part families by appropriately
rearranging the order of rows and columns,
which represent machines and parts
respectively, to find a block diagonal form of
the entries in the machine-part matrix. The
literature yields at least seven array-based
clustering heuristics, namely:
1 bond energy analysis (McCormick et al.,
1972);
2 rank order clustering (King, 1980a, b; King
and Nakornchai, 1982);
3 modified rank order clustering
(Chandrasekharan and Rajagopalan,
1986a);
4 direct clustering analysis (Chan and
Milner, 1982);
5 occupancy value method (Khator and
Irani, 1987);
6 cluster identification method (Kusiak and
Chow, 1987); and
The research register for this journal is available at
http://www.emeraldinsight.com/researchregisters
The current issue and full text archive of this journal is available at
http://www.emeraldinsight.com/0957-6061.htm
[ 123]
Integrated Manufacturing
Systems
14/2 [2003] 123-137
# MCB UP Limited
[ISSN 0957-6061]
[DOI 10.1108/09576060310459429]
Keywords
Cellular manufacturing,
Cluster analysis, Groups,
Effectiveness, Measurement
Abstract
This paper introduces a modified
single linkage clustering heuristic
(MOD-SLC). The proposed
MOD-SLC objective is to test the
application of Baroni-Urban and
Buser (BUB) similarity coefficient
to the manufacturing cell
formation (MCF) problem instead
of Jaccard's similarity coefficient.
The MOD-SLC has been compared
and evaluated against three
cluster formation-based heuristics
for MCF. The three heuristics are:
the single linkage clustering,
enhanced rank order clustering,
and direct clustering algorithm.
The MCF methods considered in
this comparative and evaluative
study belong to the cluster
formation approach of solving the
MCF problem. The comparison and
evaluation are performed using
four published performance
measures. A total of 25 published
and ten hypothetical and randomly
generated problem data sets are
used in the proposed evaluative
study. Results analysis is carried
out to test and validate the
proposed BUB based MOD-SLC.
Finally the pros and cons of each
method are stated and discussed.
Received July 2001
Revised March 2000
Accepted May 2002
7 Hamiltonian path heuristic (Askin et al.,
1991).
In hierarchical clustering, the data in the
machine-part incidence matrix are not
partitioned into manufacturing cells in one
step. Rather a similarity coefficient matrix
(for machines or parts) is formed from the
binary machine-part matrix. Then, from the
formed similarity matrix, machines (parts)
with maximum similarity are grouped to
form machine cells (part families).
Essentially, hierarchical techniques may be
subdivided into agglomerative methods that
proceed by a series of fusions of the machines
or the parts into groups, and divisive
methods which partition the set of machines
or parts successively into finer groups. All
agglomerative hierarchical techniques
ultimately reduce the data to a single cluster
containing all the machines (parts), and
divisive techniques will finally split the
entire set of machines (parts) into
one-machine (one-part) cells. Hierarchical
classifications may be represented by
inverted tree structure or dendrograms,
which are 2D diagrams illustrating the
fusions or divisions that have been made at
each successive stage of the analysis. Few
researchers applied a divisive method to the
MCF problem (Stanfel, 1985); therefore,
attention is focused on agglomerative
clustering algorithms. The most widely used
technique is the single linkage clustering
(SLC) (Sneath, 1957a, b; McAuley, 1972;
Carrie, 1973; Waghodekar and Sahu, 1984).
The SLC method merges clusters based on
the maximum similarity of their members.
Vakharia and WemmerloÈ v (1990) proposed
the average linkage clustering (ALC)
heuristic. The ALC method merges clusters
based on the average similarity of their
members.
Non-hierarchical clustering methods are
iterative methods and they begin with either
an initial partition of the data set or the
choice of a few seed points. In either case, one
had to decide the number of clusters (C) in
advance. Arbitrariness in the choice of seed
points (or initial partition of data) could lead
to unsatisfactory results. Several researchers
have developed non-hierarchical procedures
(Lemoine and Mutel, 1983; Chandrasekharan
and Rajagopalan, 1986a, b; Srinivasan and
Narendran, 1991).
Several research studies have been devoted
to the comparison and evaluation of
clustering algorithms. Seifoddini (1989a)
compared the SLC and ALC methods only.
Chu and Tsai (1990) compared three
array-based machine-part grouping methods:
ROC, DCA and BEA. Miltenburg and
Zhang (1991) compared nine cell formation
methods including similarity measure
method, non-hierarchical clustering, and
rank order methods. Cheng (1992) analyzed
several MCF clustering techniques by
example. Shafer and Rogers (1993a) validated
a newly developed similarity measure by
comparing four clustering methods.
Kaparthi and Suresh (1994) compared the
performance of several procedures for
part-machine grouping using data sets
of wide-ranging sizes and perfection.
Cheng et al. (1995) provided a narrative
comparison and used a real-life example to
demonstrate the effectiveness of various
clustering algorithms. Park and WemmerloÈ v
(1995) studied several cell formation
techniques with different data
requirements and illustrated some of the
problems related to securing compatibility.
Dimopoulos and Mort (2001) used the
grouping efficacy performance measure in
evaluating and comparing a genetic
programming based SLC method to five other
procedures.
The enhanced rank order clustering
(ROC2) (King and Nakornchai, 1982), DCA
(Chan and Milner, 1982), and SLC (McAuley,
1972) are among the popular cluster analysis-
based MCF problem solution methods.
Although each one of these methods uses a
different strategy for cluster formation, their
relative performance has often been
compared in terms of the number of inter-cell
moves they generate. The MCF problem
cluster analysis-based solution approaches
consist of two phases. The first is to develop a
similarity coefficient and the second phase is
to apply a solution methodology to solve the
MCF problem. This paper focuses on the first
phase by using a similarity measure that has
not been applied to the MCF cluster analysis-
based approaches. In this paper, a modified
SLC (MOD-SLC) method is proposed. The
MOD-SLC is compared to three cluster
formation methods selected from the
literature, namely ROC2, DCA, and SLC. The
comparison and evaluation are based on four
different performance measures selected
from the literature (Chu and Tsai, 1990; Islam
and Sarker, 2000; Sarker and Khan, 2001).
In the next section, the two selected
array-based cluster formation and the
hierarchical SLC methods are explained
using examples. An explanation of the MOD-
SLC method is introduced in the fourth
section. The fifth section presents a
comparison and evaluation of the solutions
obtained by ROC2, DCA, SLC, and MOD-SLC.
Finally, concluding remarks are presented in
the last section.
[ 124]
Hassan M. Selim,
Reda M.S. Abdel Aal and
Araby I. Mahdi
Formation of machine groups
and part families: a modified
SLC method and comparative
study
Integrated Manufacturing
Systems
14/2 [2003] 123-137
Cluster formation methods
In this section, the selected two array-based
cluster formation methods (ROC2 and DCA)
followed by the hierarchical cluster
formation method (SLC) are illustrated by
example in order to contrast them with the
MOD-SLC method.
Array-based cluster formation methods
The array-based cluster formation methods
that are addressed in this paper can be stated
as follows (Veramani and Mani, 1996): Given
a (0, 1) incidence matrix, partition it into a
maximal number (C) of clusters, C > 2, such
that the number of exceptional elements is
minimized.
If the matrix is a machine-part incidence
matrix, then C represents the number of
manufacturing cells. The exceptional
elements are cases where either the parts
must move from one cell to another for one or
more operations or a machine has to be
replicated in another cell.
ROC is a well known cluster formation
technique that attempts to create a block
diagonal form by repeatedly reallocating the
columns and rows of a machine-part matrix
according to binary values (King, 1980a, b).
The binary values are calculated by reading
the pattern of cell entries as a binary word.
Each row and column is assigned a weight
that is the decimal equivalent of its binary
word. Although ROC is easy to apply, it has
several disadvantages. First, the quality of
the results is strongly dependent on the
initial disposition of the machine-part
matrix. Second, the binary value that is used
for the reallocation restricts the size of the
problem that the technique can handle. This
results in a storage problem. A revised
version, called ROC2, has been developed by
King and Nakornchai (1982) to overcome this
limitation and increase computational
efficiency. The ROC2 algorithm is depicted in
Figure 1.
To illustrate the ROC2 method, an
application to the problem represented by
machine-part matrix given in Table I is to be
considered.
Table II shows the results of ordering the
machine rows based on the positive entries
in their associated part columns. The first
row in Table II represents part-column 8 in
Table I that shows positive entries in
machine-rows 2, 3, 5 and 7 respectively
(shown in italic in Table II). Accordingly,
rows 2, 3, 5 and 7 are moved to the top of the
matrix and the procedure is repeated on part-
column 7 represented by the second row in
Table II. The last row of Table II shows the
final row order after applying the procedure
on the eight part columns. Table III shows the
results of ordering the part-columns using
the same procedure. The last row of Table III
shows the final part-column order. The whole
process is repeated until two consecutive
machine-part matrices are identical. Table IV
shows the final solution of applying ROC2 to
the matrix shown in Table I. ROC2 tends to
produce one cluster in the northwest corner
leaving the rest of the matrix relatively
disorganized.
DCA rearranges the rows with the left-
most positive cells to the top and the
columns with the top-most positive cells to
the left of the machine-part matrix (Chan
and Milner, 1982). After several iterations,
all the positive cells will form diagonal
blocks from the top left corner to the
bottom-right corner of the matrix. The DCA
algorithm including the correction made by
WemmerloÈ v (1984) is shown in Figure 2
which is depicted from Chu and Tsai (1990).
This method allows more flexibility in the
problem size. Furthermore, the sensitivity
of the ROC and ROC2 algorithms to the
initial matrix is eradicated because DCA
initiates the procedure by counting the
number of positive cells instead of
depending on intuition. The same machine-
part matrix shown in Table I is used to
illustrate the DCA method. Table V shows
rows and columns ranked in descending
order of K (number of positive cells in each
row and column). Table VI shows the final
machine-part matrix and the clusters are
clearly visible. As shown in bold in Table
VI, there are three inter-cell moves
(exceptional elements) in the final matrix.
SLC formation methods
The SLC first developed by Sneath (1957a, b)
for the classification of bacteria, was
followed by McAuley (1972) in the context of
the MCF problem for forming machine cells
through iterative process. SLC method
groups two individuals i and j with highest
level of similarity into one cluster, ij. The
combined cluster behaves as if it is a single
individual. The similarity between this
cluster and individual k, as defined by the
SLC algorithm, is the maximum of the
similarities between k and the component
members of the cluster ij. Iterations continue
to merge the groups with the largest
similarity coefficient until a single group
exists. The SLC algorithm is shown in
Figure 3.
Several researchers have proposed
different similarity measures for the
cluster formation problem, but the
measures in most cases are suitable for
specific problems (Sneath and Sokal, 1973;
[ 125]
Hassan M. Selim,
Reda M.S. Abdel Aal and
Araby I. Mahdi
Formation of machine groups
and part families: a modified
SLC method and comparative
study
Integrated Manufacturing
Systems
14/2 [2003] 123-137
Romesburg, 1984; Seifoddini and Djassemi,
1991; Askin et al., 1997). Among these
measures, is the Jaccard similarity
measure by Jaccard (1908). The property of
this measure is that there exists a
maximum similarity of one between a pair
of machines/parts when both of the
machines/parts have identical binary
values in the machine-part incidence
matrix. This measure is sensitive to the
direction of coding, meaning that
interchanging the elements 1 and 0 in an
incidence matrix usually changes the
similarity values between machines/parts
(Islam and Sarker, 2000).
The SLC method uses Jaccard similarity
measure in evaluating the machine pair
similarity based on the common part-
operations between them. Jaccard similarity
measure can be expressed as follows:
SJ
ij
= X
iy
=(X
i
÷X
j
÷X
ij
) where:
SJ
ij
= Jaccard similarity between
machine i and machine j, 0 < SJ
ij
_ 1.
X
ij
= number of parts processed by both
of machines i and j (number of
matches).
Figure 1
The ROC2 cluster formation method
Table II
Row ordering
Columns Row order
8 1 2 3 4 5 6 7
7 2 3 5 7 1 4 6
6 2 7 1 4 6 3 5
5 2 7 3 5 1 4 6
4 5 2 7 3 1 4 6
3 1 4 6 5 2 7 3
2 5 3 1 4 6 2 7
1 1 2 7 5 3 4 6
Final 2 5 7 1 3 4 6
Table I
Initial ROC2 machine-part matrix
Part
Machine 1 2 3 4 5 6 7 8
1 1 1
2 1 1 1 1 1
3 1 1 1
4 1 1
5 1 1 1 1 1
6 1 1
7 1 1 1 1 1
Table III
Column ordering
Rows Column order
6 1 2 3 4 5 6 7 8
4 4 7 1 2 3 5 6 8
3 4 7 1 2 3 5 6 8
1 3 6 8 4 7 1 2 5
7 4 7 2 3 6 8 1 5
5 7 6 8 1 4 2 3 5
2 6 8 1 3 5 7 4 2
Final 6 8 1 2 3 5 7 4
Table IV
Final ROC2 machine-part matrix
Part
6 8 1 7 2 3 5 4
2 1 1 1 1 1
7 1 1 1 1 1
5 1 1 1 1 1
3 1 1 1
1 1 1 1
4 1 1
6 1 1
[ 126]
Hassan M. Selim,
Reda M.S. Abdel Aal and
Araby I. Mahdi
Formation of machine groups
and part families: a modified
SLC method and comparative
study
Integrated Manufacturing
Systems
14/2 [2003] 123-137
X
i
= Number of parts processed by
machine i only.
X
j
= Number of parts processed by
machine j only.
Y
ij
= Number of parts that are not
processed by either machine i nor
machine j (number of misses).
The most common way to display the
hierarchy of clusters generated by the SLC
algorithm is the form of a dendrogram. The
CM designer must choose a similarity level or
threshold value in order to define the number
of clusters. As the threshold value increases,
the number of cells increases while the size of
the cells decreases. When SLC is used in
machine grouping, it is assumed that part
operations have been pre-assigned to machine
types or to individual machines. Table VII
shows the machine similarity matrix created
from the machine-part incidence matrix
shown in Table I using Jaccard similarity
coefficient. The similarity matrix is
symmetric. The dendrogram shown in Figure
4 shows the machine groups at different
values of Jaccard similarity coefficient. For
example, at similarity coefficient value of 0.67,
there are four machine groups as indicated in
Figure 4 by the vertical dotted line
intersections with the horizontal machine/
group lines. The similarity scale indicates the
amount of similarity between clusters at each
branching of the tree. The threshold value of
the similarity measure is selected based on
the preferred number of cells and the
Figure 2
The DCA cluster formation method
Table V
DCA initial matrix
Part
Machines 7 6 8 1 2 4 3 5 K
7 1 1 1 1 1 5
5 1 1 1 1 1 5
2 1 1 1 1 1 5
3 1 1 1 3
1 1 1 1 3
6 1 1 2
4 1 1 2
K 5 4 4 3 3 3 2 1
Table VI
DCA final matrix
Part
Machine 7 4 2 8 6 1 3 5
6 1 1
4 1 1
1 1 1 1
2 1 1 1 1 1
7 1 1 1 1 1
5 1 1 1 1 1
3 1 1 1
[ 127]
Hassan M. Selim,
Reda M.S. Abdel Aal and
Araby I. Mahdi
Formation of machine groups
and part families: a modified
SLC method and comparative
study
Integrated Manufacturing
Systems
14/2 [2003] 123-137
minimum intercell moves. Table VIII shows
the machine-part matrix at similarity
measure of 0.43 and the clusters are clearly
visible.
Modified single linkage cluster
(MOD-SLC) formation
In this section, the MOD-SLC formation
method is explained. The MOD-SLC method
is proposed in order to test a non-Jaccardian
similarity measure instead of Jaccard's
measure used in the SLC method. The
similarity measures used in solving the MCF
problem as a cluster formation problem have
been surveyed by several researchers
(Kusiak and Cho, 1984; Shafer and Rogers,
1993a, b; Mosier et al., 1997; Islam and Sarker,
2000; Selim, 2002). Sarker (1996), Sarker and
Islam (1999), and Islam and Sarker (2000)
analyzed most of the existing similarity
measures in the literature based on a set
of important properties developed by
Baroni-Urban and Buser (1976) and modified
by Islam and Sarker (2000).
Similarity coefficients are either
Jaccardian or non-Jaccardian, with respect
to the Jaccard similarity coefficient (Jaccard,
1908). The Jaccardian similarity coefficients
are expressed as a measure of level of
matches, in which the number of matches
(X
ij
) is divided by a normalized quantity
Table VII
The SLC similarity matrix
Machine
Machine 1 2 3 4 5 6 7
1 0.33 0.67 0.67 0.33
2 0.33 0.17 0.43 0.17 1.00
3 0.60 0.33
4 1.00 0.17
5 0.43
6 0.17
7
Table VIII
SLC final matrix
Part
Machine 4 7 1 2 3 5 6 8
4 1 1
6 1 1
1 1 1 1
2 1 1 1 1 1
7 1 1 1 1 1
3 1 1 1
5 1 1 1 1 1
Figure 4
The SLC dendogram
Figure 3
The SLC cluster formation method
[ 128]
Hassan M. Selim,
Reda M.S. Abdel Aal and
Araby I. Mahdi
Formation of machine groups
and part families: a modified
SLC method and comparative
study
Integrated Manufacturing
Systems
14/2 [2003] 123-137
usually represented by the expected number
of matches. Non-Jaccardian similarity
coefficients have an additional term, the
number of misses (Y
ij
), appears in the
numerator and then divided by the
normalizing term. The status of the number
of misses (Y
ij
) in similarity coefficients
applied to the CM problem is ambiguous. It
refers to the number of parts not processed
by either machine or the number of
machines not needed by either part type. The
researchers who adopt Jaccardian similarity
coefficients assume that the similarity
coefficients measure the degree of
commonality between the two machines in
terms of parts processed (McAuley, 1972;
Rajagopalan and Batra, 1975; Kusiak and
Cho, 1984; Waghodekar and Sahu, 1984;
Mosier and Taube, 1985; Seifoddini and
Wolfe, 1986; Wei and Kern, 1989; Seifoddini,
1989b). Therefore, the number of misses does
not contribute to the machine pair similarity
coefficient. On the other hand, a significant
part of the literature (Mosier, 1989; Islam and
Sarker, 2000) shows that Jaccardian
similarity coefficients are unable to reflect
the true values of similarity, as the
Jaccardian measures do not consider the
number of misses (Y
ij
). The number of
misses, in the CM context, represents the
similarity of a machine pair indicated by the
number of parts that both are not capable of
processing or the similarity of a product pair
indicated by the number of attributes that
both products do not have.
Baroni-Urban and Buser (1976) defined a
set of properties of similarity coefficients and
applied these properties to the several
similarity coefficients. There does not exist
any similarity coefficient which follows all
the properties defined by Baroni-Urban and
Buser (1976). Islam and Sarker (2000)
modified the properties proposed by Baroni-
Urban and Buser (1976) and stated them as
follows (S
ij
is the machine i and machine j
similarity coefficient):
1 No mismatch, S
ij
÷ 1 for X
i
= X
j
= 0.
2 Minimum match, S
ij
÷ 0 for X
ij
, Y
ij
÷ 0.
3 No match, S
ij
÷ 0 for X
ij
, = 0.
4 Complete match, S
ij
= 1 for X
ij
= number
of parts.
5 Maximum match, S
ij
÷ 1 for X
ij
+ Y
ij
÷
number of parts.
The similarity measure developed by Baroni-
Urban and Buser (1976) ± BUB measure ± has
conformed to the five properties. This
similarity coefficient has superior properties
of distribution compared to other coefficients
because the distribution of its values are
more normal and continuous (Sarker, 1996).
The BUB similarity coefficient is defined
as follows:
SB
ij
=
X
ij
÷

X
ij
Y
ij
p
X
i
÷X
j
÷X
ij
÷

X
ij
Y
if
p (1)
where SB
ij
= BUB similarity between
machine i and machine j, 0 _ SB
ij
_ 1.
In order to justify the application of non-
Jaccardian similarity coefficients to the MCF
problem, Islam and Sarker (2000) used
properties 2 and 5 to conclude that both
matches (X
ij
) and misses (Y
ij
) must be
included in the numerator of the defining
similarity coefficient. To satisfy properties 2,
3, 4, and 5, the product X
ij
Y
ij
is considered in
addition to X
ij
in the numerator. The square
root is used to maintain the order
consistency (Baroni-Urban and Buser, 1976).
When there are no misses (Y
ij
= 0), BUB
measure is reduced to Jaccard's measure
which is the ratio of the number of parts
processed by both machines to the total
number of parts processed by both or one of
the machines. If (Y
ij
) ~ the BUB coefficient
value increases to reflect the real similarity
of machine/part pairs. Islam and Sarker
(2000) modified BUB measure by adding the
number of misses (Y
ij
) to the denominator
and called it ``relative matching measure''.
The Jaccard measure has conformed to
only three out of the same five properties
namely, properties 1, 3 and 4. The Jaccard
similarity measure has several additional
limitations that have been discussed for
years by a number of researchers (Shafer and
Rogers, 1993a):
.
The first limitation occurs when there is a
high similarity between two machines and
the Jaccard measure does not detect it.
Consider the machine-part matrix given in
Table IX, even though machine two can
perform 100 percent of the operations
performed by machine one (operations 4
and 5), SJ
12
= 2/(0 + 2 + 2) = 0.5.
.
The second limitation is that it does not
include the parts that do not need
processing by machine pairs (Islam and
Sarker, 2000; Yasuda and Yin, 2001).
.
The third limitation is that it is sensitive
to the direction of coding, meaning that
interchanging the elements 1 and 0 in an
incident matrix usually changes the
similarity values between machines/parts
(Islam and Sarker, 2000).
For the same case example shown in Table IX
that explained the deficiency in Jaccard
Table IX
Machine-part matrix
Part
Machine 1 2 3 4 5 6 7 8
1 1 1
2 1 1 1 1
[ 129]
Hassan M. Selim,
Reda M.S. Abdel Aal and
Araby I. Mahdi
Formation of machine groups
and part families: a modified
SLC method and comparative
study
Integrated Manufacturing
Systems
14/2 [2003] 123-137
measure, BUB measure yielded a higher
similarity value of SB
ij
= 0.71 compared with
SJ
ij
= 0.50 and a value of 0.446 resulted from
the modified BUB (relative matching
measure) developed by Islam and Sarker
(2000). This indicates that the original BUB
measure more closely reflects true similarity
between machine/part pairs, overcomes
Jaccard's measure drawbacks, and satisfies
the five properties defined by (Islam and
Sarker, 2000).
The MOD-SLC method replaces Jaccard
measure with BUB similarity measure to
initialize the SLC method shown in Figure 3.
The modified initialization of the SLC
method enhances the accuracy of the
similarity matrix. The machine-part
incidence matrix shown in Table I is used to
illustrate the MOD-SLC method. Table X
shows the BUB similarity values of all
machine pairs of Table I. The resulting
dendrogram is shown in Figure 5, which
indicates different machine groups and
higher similarity values compared to
Figure 4. The minimum number of machine
groups resulting from the MOD-SLC is three
groups (see Figure 5). The machine-part
incidence matrix for a similarity measure of
0.75 is shown in Table XI, which shows a
three-cluster (cells) solution to the example
under consideration.
In the next section, ROC2, DCA, SLC, and
MOD-SLC cluster formation methods will be
compared and evaluated using 35 data sets
and four published performance measures.
Comparative and evaluative
analysis
In this section, a general procedure for
comparing and evaluating the MCF problem
solutions using the four cluster formation
methods is presented. The ROC2, DCA, SLC,
and MOD-SLC are coded in the C
programming language and run on an IBM
Pentium III 800MHz Station.
Performance measures
The performance of cluster formation methods
can be evaluated either according to
computational efficiency or according to
clustering effectiveness (Chu and Tsai, 1990).
Clustering efficiency is normally measured in
terms of program execution time, the amount
of memory needed, and the complexity of the
algorithm. In this paper, four measures were
selected due to their wide usage in the
literature (Seifoddini and Djassemi, 1991; Islam
and Sarker, 2000).
Number of exceptional elements (PE)
The number of off-diagonal positive entries
(exceptional elements) in the final machine-
part incidence matrix can measure the
quality of the cluster formation method
(King, 1980a, b; Chan and Milner, 1982). PE
can be computed as:
PE = e
0
(2)
where e
0
is the number of exceptional
elements or the off-diagonal positive entries.
Machine utilization (MU)
MU indicates the percentage of times the
machines within clusters (cells) are used in
production. MU can be computed as
(Chandrasekharan and Rajagopalan, 1986a):
Mu =
e
d
P
C
i=1
m
k
n
k
(3)
where e
d
is the number of positive entries in
the diagonal blocks, m
k
is the number of
machines in the kth cell, n
k
is the number of
parts in the kth cell, and C is the number of
Table X
The MOD-SLC similarity matrix
Machine
Machine 1 2 3 4 5 6 7
1 0.50 0.84 0.84 0.50
2 0.50 0.33 0.54 0.33 1.00
3 0.75 0.50
4 1.00 0.33
5 0.54
6 0.33
7
Table XI
MOD-SLC final matrix
Part
Machine 4 7 1 2 8 5 6 3
4 1 1
6 1 1
1 1 1 1
2 1 1 1 1 1
7 1 1 1 1 1 1
3 1 1 1
8 1 1 1 1 1
Figure 5
The MOD-SLC dendrogram
[ 130]
Hassan M. Selim,
Reda M.S. Abdel Aal and
Araby I. Mahdi
Formation of machine groups
and part families: a modified
SLC method and comparative
study
Integrated Manufacturing
Systems
14/2 [2003] 123-137
cells. The higher the value of MU, the better
the machines are being utilized.
Grouping efficiency (GE)
GE is an aggregate measure that takes both
the number of exceptional elements and the
machine utilization into consideration. A
convex combination of both terms is
considered to reveal the relative importance
of each term. GE can be defined as
(Chandrasekharan and Rajagopalan, 1986a):
GE = MU ÷(1 ÷)
e
0
M v N ÷
P
c
k=1
m
k
n
k
(4)
where is a weight; 0 _; _ 1, M is the total
number of machines, N is the total number of
parts, as a general rule, the higher the
grouping efficiency, the better the clustering
results.
Grouping efficacy (GC)
GC overcomes the problems of selecting
and the limiting range of GE. GC has the
requisite properties like non-negativity, zero
to one range and is not affected by the size of
the machine-part matrix. GC is defined by
Kumar and Chandrasekharan (1990) and
Sandbothe (1998) as in equation (5):
GC =
e ÷e
o
e

(5)
where e

is the number of zeros in the
diagonal blocks. Generally, the higher the
GC, the better the cluster formation results.
Data problems overview
Most of the cluster formation methods are
data dependent. The testing data can be
either randomly generated via computer or
collected from the literature. In order to
evaluate and compare the performance of the
four cluster formation methods and validate
the applicability of the BUB similarity
coefficient in designing CM systems, 25 data
problems have been selected from the
literature in addition to ten randomly
generated data problems. Jaccard's
coefficient three limitations mentioned in the
previous section exist in most of the 35 data
problems used in the comparative study.
Table XII shows the data problems, sources,
sizes, and densities (D = e/(M
.
N) that are
calculated by dividing the total number of
ones in the machine-part incidence matrix
(e) by M
.
N.
Experimental framework
The 35 data sets have been classified into
three groups based on the number of
machines (M), three groups based on the
number of parts (N), and three groups based
on the density level (D). Table XIII shows the
value ranges of M, N and D for each group.
The selected density range values are based
on the selected data sets and specific
implementation in the literature. Densities
between 0.10 and 0.30 represent the different
scenarios adequately.
Computational results
The four performance measures, PE, MU, GE,
and GC are computed for each data set in
each group of the nine groups. The
performance measures are averaged for each
group.
PE results
Table XIV summarizes the computational
results of PE for each data set group. The first
column shows the data set group labels and
each cluster formation method's PE average
values are shown in the succeeding columns.
The last row shows the overall performance
of each cluster formation method based on
average PE performance measure values.
In Table XIV, all the cluster formation
methods under consideration performed very
well with medium number of machines
(represented by group M2), which yielded the
best PE values. Both ROC2 and DCA
methods' PE values increase with the
increase in the number of parts. The SLC and
MOD-SLC performed very well with high
number of parts as indicated by the PE
values of group N3. As the matrix density
increases, the PE measure associated with
both SLC and MOD-SLC methods increases.
This indicates that both SLC and MOD-SLC
methods do not perform well on dense
matrices. The MOD-SLC was the best in the
overall performance over all the problem
data sets. As shown in Figure 6, MOD-SLC
achieved the lowest value of PE at the
aggregated level over all the data sets
considered.
MU results
Table XV summarizes the computational
results of MU average values for each data set
group.
In Table XV, as the number of machines
increases the machine utilization decreases
which is intuitive. For groups M1 and M2,
MOD-SLC method dominated the other
cluster formation methods and for group M3,
the SLC method generated the maximum
MU. The group N3 that represents the
maximum number of parts generated the
minimum MU for all the four cluster
formation methods, this is due to the
increase in the number of cells generated to
process the high number of parts. The
results of matrix density were intuitive for
groups D1, D2, and D3. As the matrix density
[ 131]
Hassan M. Selim,
Reda M.S. Abdel Aal and
Araby I. Mahdi
Formation of machine groups
and part families: a modified
SLC method and comparative
study
Integrated Manufacturing
Systems
14/2 [2003] 123-137
increases the machine utilization increases
accordingly. As shown in Figure 7, SLC and
MOD-SLC yielded the highest values of MU
at the aggregated level.
GE results
Table XVI summarizes the computational
results of GE average values for each data set
group. The GE values are calculated using
= 0.5. Generally, as the number of machines
and number of parts increase the grouping
efficiency GE decreases. All the four methods
generated their minimum GE values at the
highest level of M that is represented by
Table XIII
The data set groups and factor ranges
Factor Group label Value range Data problems
M M1 M _ 8 10
M2 8 < M _ 16 12
M3 M > 16 13
N N1 N _ 10 13
N2 10 < N _ 25 12
N3 N > 25 10
D D1 D _ 0.2 14
D2 0.2 < D _ 0.3 10
D3 D > 0.3 11
Table XIV
PE values for each data set group
Group ROC2 DCA SLC MOD-SLC
M1 8.8197 7.0179 9.8054 7.6404
M2 5.7583 4.9013 2.6786 3.6402
M3 15.5804 14.348 6.0133 6.0508
N1 6.0653 6.7029 3.3804 4.1899
N2 11.0154 6.9867 8.3567 7.2227
N3 14.8807 14.4546 5.5058 5.5103
D1 8.2717 6.5401 3.6776 3.5758
D2 12.1430 12.1979 4.3256 4.9919
D3 11.1461 9.2712 10.0365 9.4346
Overall 10.2812 9.0150 5.8155 5.6444
Table XII
The data problems used in the evaluative study
Problem Source M N D
1 Kumar et al. (1986) 30 41 0.104
2 Chandrasekharan and Rajagopalan (1987) 24 40 0.136
3 Chandrasekharan and Rajagopalan (1987) 24 40 0.135
4 Chandrasekharan and Rajagopalan (1987) 24 40 0.136
5 Chandrasekharan and Rajagopalan (1987) 24 40 0.136
6 Chandrasekharan and Rajagopalan (1986a) 20 35 0.193
7 Randomly generated 20 35 0.200
8 Randomly generated 20 35 0.204
9 Randomly generated 20 35 0.215
10 Randomly generated 20 35 0.211
11 Harhalakis et al. (1990) 20 20 0.197
12 Shafer and Rogers (1993b) 20 20 0.147
13 Randomly generated 20 20 0.210
14 Chan and Milner (1982) 15 10 0.306
15 Chan and Milner (1982) 15 10 0.330
16 Balasubramanian and Panneerselvam (1993) 15 10 0.280
17 Randomly generated 15 10 0.193
18 Askin et al. (1991) 14 24 0.181
19 McAuley (1972) 12 10 0.316
20 Srinivasan et al. (1990) 10 20 0.245
21 Randomly generated 10 20 0.195
22 Askin et al. (1991) 10 15 0.326
23 Mukhopadhyay and Golpalakrishnan (1995) 10 10 0.240
24 Randomly generated 10 10 0.190
25 Arvinh and Irani (1994) 10 8 0.325
26 Srinivasan and Narendran (1991) 8 20 0.381
27 Kusiak et al. (1993) 8 9 0.236
28 Randomly generated 8 9 0.194
29 Mukhopadhyay et al. (1994) 7 11 0.270
30 Randomly generated 7 11 0.194
31 Mukhopadhyay et al. (1994) 7 9 0.412
32 Kusiak and Cho (1984) 6 8 0.458
33 Seifoddini (1989c) 5 18 0.470
34 Mukhopadhyay et al. (1994) 5 18 0.511
35 King and Nakornchai (1982) 5 7 0.400
[ 132]
Hassan M. Selim,
Reda M.S. Abdel Aal and
Araby I. Mahdi
Formation of machine groups
and part families: a modified
SLC method and comparative
study
Integrated Manufacturing
Systems
14/2 [2003] 123-137
group M3 and the highest level of N
represented by group N3.
As shown in the three rows representing
D1, D2 and D3 in Table XVI, the general trend
shows that as the matrix density increases
the GE values increase, this is due to the
increase in the within cell utilization. As
shown in Figure 8, MOD-SLC yielded the
highest values of GE at the aggregated level.
GC results
Table XVII summarizes the computational
results of GC average values for each data
set group.
The best performance of GC was with
medium number of machines and small
number of parts as shown in Table XVII.
Generally, the GC measure was high at a
small number of parts. The best GC values
were obtained at high matrix density values.
As shown in Figure 9, MOD-SLC yielded the
highest values of GC at the aggregated level.
MOD-SLC performed poorer than SLC when
high number of machines and number of
parts were used as indicated by groups M3
and N3.
In this section, the performance of the
four cluster formation methods was
evaluated and compared using 25 published
Figure 7
MU overall values
Table XVI
GE values for each data set group
Group ROC2 DCA SLC MOD-SLC
M1 0.8250 0.8000 0.8387 0.8420
M2 0.8642 0.8467 0.8373 0.8554
M3 0.7033 0.7022 0.7927 0.7887
N1 0.8432 0.8257 0.8212 0.8322
N2 0.8160 0.7471 0.8396 0.8482
N3 0.7011 0.6952 0.7983 0.7875
D1 0.7614 0.7498 0.7904 0.7887
D2 0.7580 0.7546 0.8110 0.8180
D3 0.8658 0.8405 0.8707 0.8827
Overall 0.7933 0.7797 0.8207 0.8265
Figure 8
GE overall values
Table XVII
GC values for each data set group
Group ROC2 DCA SLC MOD-SLC
M1 0.6580 0.5541 0.6517 0.6869
M2 0.7210 0.6916 0.6695 0.7036
M3 0.4279 0.4250 0.6011 0.5710
N1 0.6864 0.6576 0.6606 0.6643
N2 0.6361 0.6164 0.6592 0.6910
N3 0.4239 0.4135 0.6165 0.5736
D1 0.5273 0.5083 0.5958 0.5723
D2 0.5357 0.5257 0.6101 0.6332
D3 0.7323 0.7010 0.7225 0.7628
Overall 0.5941 0.5738 0.6388 0.6489
Table XV
MU values for each data set group
Group ROC2 DCA SLC MOD-SLC
M1 0.7154 0.6625 0.7310 0.7439
M2 0.7545 0.7173 0.6876 0.7281
M3 0.4551 0.4514 0.6269 0.5824
N1 0.7208 0.6927 0.6591 0.6910
N2 0.6895 0.6427 0.7208 0.7390
N3 0.4480 0.4384 0.6142 0.5877
D1 0.5428 0.5176 0.6124 0.5847
D2 0.5635 0.5566 0.6274 0.6537
D3 0.8085 0.7536 0.8047 0.8332
Overall 0.6321 0.6029 0.6759 0.6748
Figure 6
PE overall values
[ 133]
Hassan M. Selim,
Reda M.S. Abdel Aal and
Araby I. Mahdi
Formation of machine groups
and part families: a modified
SLC method and comparative
study
Integrated Manufacturing
Systems
14/2 [2003] 123-137
problem data sets and ten randomly
generated. The 35 problem data sets were
classified into nine groups based on three
dimensions: M, N and D. Four different
performance measures were used in the
evaluative and comparative analysis. The
results showed that MOD-SLC method
slightly improved the average values of the
four performance measures used. This
indicate that, the replacement of Jaccard
similarity measure with the BUB similarity
measure improved the SLC method
performance in addition to the accuracy
improvement in the initial similarity
matrix.
Conclusion
This paper has proposed a cluster formation
method that is adapted from the SLC method.
The MOD-SLC method has been tested
against three MCF solution methods using
the cluster formation approach, namely ROC,
direct clustering algorithm, and SLC. The
paper also demonstrated an evaluative and
comparative analysis using four different
performance measures, percentage of
exceptional parts, within cell machine
utilization, grouping efficiency, and grouping
efficacy.
The following results have been obtained
based on 35 problem data sets that have been
classified into nine groups based on the
number of machines, number of parts, and
machine-part incidence matrix:
.
The MOD-SLC improved the average
values of the four performance
measures, PE, MU, GE and GC. This
indicates that the modification made to
the SLC to overcome the deficiencies of
Jaccard measure by using BUB non-
Jaccardian measure has resulted in
better performance by the MOD-SLC
method.
.
The performance of the two array-based
cluster formation methods considered
(ROC2 and DCA) is poorer than the
hierarchical methods used (SLC and
MOD-SLC).
Further research is needed to add more
dimensions to the cluster formation methods
when applied to the MCF problem. The
dimensions include machine loading, part
volumes, multiple machines of each type, the
arrangement of machines in each cell, and
worker assignments.
References
Arvinh, B. and Irani, S.A. (1994), ``Principal
component analysis for evaluating the
feasibility of CM without initial machine-part
matrix clustering'', International Journal of
Production Research, Vol. 32 No. 8,
pp. 1909-38.
Askin, R.G., Selim, H.M. and Vakharia, A.J.
(1997), ``A methodology for designing flexible
cellular manufacturing systems'',
IIE Transactions, Vol. 29.
Askin, R.G., Creswell, J.B., Goldberg, J.B. and
Vakharia, A.J. (1991), ``A Hamiltonian path
approach to reordering the machine-part
matrix for cellular manufacturing'',
International Journal of Production Research,
Vol. 29 No. 6, pp. 1081-1100.
Balasubramanian, K.N. and Panneerselvam, R.
(1993), ``Covering technique-based algorithm
for machine grouping to form manufacturing
cells'', International Journal of Production
Research, Vol. 31 No. 6, pp. 1479-1504.
Baroni-Urban, C. and Buser, M.W. (1976),
``Similarity of binary data'', Zoology, Vol. 25,
pp. 251-9.
Carrie, A.S. (1973), ``Numerical taxonomy applied
to group technology and plant layout'',
International Journal of Production Research,
Vol. 11 No. 4, pp. 399-416.
Chan, H.M. and Milner, D.A. (1982), ``Direct
clustering algorithm for group formation in
cellular manufacturing'', Journal of
Manufacturing Systems, Vol. 1 No. 1, pp. 65-74.
Chandrasekharan, M.P. and Rajagopalan, R.
(1986a), ``MODROC: an extension of rank
order clustering for group technology'',
International Journal of Production Research,
Vol. 24 No. 5, pp. 1221-33.
Chandrasekharan, M.P. and Rajagopalan, R.
(1986b), ``An ideal seed non-hierarchical
clustering algorithm for cellular
manufacturing'', International Journal of
Production Research, Vol. 24 No. 1, pp. 451-64.
Chandrasekharan, M.P. and Rajagopalan, R.
(1987), ``ZODIAK: an algorithm for concurrent
formation of part families and machine cells'',
International Journal of Production Research,
Vol. 25 No. 2, pp. 835-50.
Cheng, C.H. (1992), ``Algorithms for grouping
machine groups in group technology'',
OMEGA-International Journal of
Management Science, Vol. 20 No. 4,
pp. 493-501.
Figure 9
GE overall values
[ 134]
Hassan M. Selim,
Reda M.S. Abdel Aal and
Araby I. Mahdi
Formation of machine groups
and part families: a modified
SLC method and comparative
study
Integrated Manufacturing
Systems
14/2 [2003] 123-137
Cheng, C.H., Kumar, A. and Motwani, J. (1995), ``A
comparative examination of selected cellular
manufacturing clustering algorithms'',
International Journal of Operations &
Production Management, Vol. 15 No. 12,
pp. 86-97.
Chu, C.H. and Tsai, M. (1990), ``A comparison of
three array-based clustering techniques for
manufacturing cell formation'', International
Journal of Production Research, Vol. 28 No. 8,
pp. 1417-33.
Dimopoulos, C. and Mort, N. (2001), ``A
hierarchical clustering methodology based on
genetic programming for the solution of
simple cell-formation problems'',
International Journal of Production Research,
Vol. 39 No. 1, pp. 1-19.
Harhalakis, G., Nagi, R. and Proth, J.M. (1990),
``An efficient heuristic in manufacturing cell
formation for group technology applications'',
International Journal of Production Research,
Vol. 28 No. 1, pp. 185-98.
Islam, K.M. and Sarker, B.P. (2000), ``A similarity
coefficient measure and machine-parts
grouping in cellular manufacturing systems'',
International Journal of Production Research,
Vol. 38 No. 3, pp. 699-720.
Jaccard, P. (1908), ``Nouvelles recherches sur la
distribution florale'', Bull. Soc. Vaud. Sce,
Vol. 44, pp. 223-70.
Joines, J.A., King, R.E. and Culbreth, C.T. (1996),
``Comprehensive review of production
oriented cell formation techniques'',
International Journal of Flexible Automation
and Integrated Manufacturing, Vol. 3
Nos. 3/4, pp. 161-200.
Kaparthi, S. and Suresh, N.C. (1994),
``Performance of selected part-machine
grouping techniques for data sets of wide
ranging sizes and imperfection'', Decision
Sciences, Vol. 25 No. 4, pp. 515-39.
Khator, S.K. and Irani, S.A. (1987), ``Cell formation
in group technology: a new approach'',
International Journal of Computers and
Industrial Engineering, Vol. 12 No. 2,
pp. 131-42.
King, J.R. (1980a), ``Machine-component group
formation in group technology'', OMEGA ±
International Journal of Management Science,
Vol. 8 No. 2, pp. 193-9.
King, J.R. (1980b), ``Machine-component grouping
in production flow analysis: an approach
using rank order clustering algorithm'',
International Journal of Production Research,
Vol. 18 No. 2, pp. 213-32.
King, J.R. and Nakornchai, V. (1982), ``Machine-
component group formation in group
technology ± review and extension'',
International Journal of Production Research,
Vol. 20 No. 2, pp. 117-33.
Kumar, C.S. and Chandrasekharan, M.P. (1990),
``Grouping efficacy: a quantitative criterion
for goodness of block diagonal forms of binary
matrices in group technology'', International
Journal of Production Research, Vol. 28
No. 2, pp. 233-43.
Kumar, K.R., Kusiak, A. and Vannelli, A. (1986),
``Grouping of parts and components in
flexible manufacturing systems'', European
Journal of Operational Research, Vol. 20,
pp. 387-97.
Kusiak, A. and Cho, M. (1984), ``Similarity
coefficient algorithm for solving the group
technology problem'', International Journal
of Production Research, Vol. 30 No. 11,
pp. 2633-46.
Kusiak, A. and Chow, W.S. (1987), ``Efficient
solving of the group technology problem'',
Journal of Manufacturing Systems, Vol. 6
No. 2, pp. 117-24.
Kusiak, A., Boe, W.J. and Cheng, C.H. (1993),
``Designing cellular manufacturing systems:
branch and bound and AI approaches'',
IIE Transactions, Vol. 25 No. 4, pp. 46-56.
Lemoine, Y. and Mutel, B. (Eds) (1983), Automatic
Recognition of Production Cells and Part
Families, North-Holland, Amsterdam.
McAuley, J. (1972), ``Machine grouping for
efficient production'', The Production
Engineer, Vol. 51, February, pp. 53-7.
McCormick, W.T., Scweitzer, P.J. and White, T.W.
(1972), ``Problem decomposition and data
reorganization by a cluster technique'',
Operations Research, Vol. 20, pp. 993-1009.
Miltenburg, J. and Zhang, W. (1991), ``A
comparative evaluation of nine well-known
algorithms for solving the cell formation
problem in group technology'', Journal of
Operations Management, Vol. 10, pp. 44-72.
Mosier, C.T. (1989), ``An experiment investigating
the application of clustering procedures and
similarity coefficients to the group
technology cell formation problem'',
International Journal of Production Research,
Vol. 27 No. 10, pp. 1811-35.
Mosier, C.T. and Taube, L. (1985), ``Weighted
similarity coefficient heuristics for the group
technology machine clustering problem'',
International Journal of Management Science,
Vol. 13 No. 6, pp. 577-83.
Mosier, C.T., Yelle, J. and Walker, G. (1997),
``Survey of similarity coefficient based
methods as applied to the group technology
configuration problem'', OMEGA ±
International Journal of Management Science,
Vol. 25 No. 1, pp. 65-79.
Mukhopadhyay, S.K. and Gopalakrishnan, A.
(1995), ``A vector analytic (VECAN) method
for solving the machine-part grouping
problem in group technology'', International
Journal of Production Research, Vol. 33 No. 3,
pp. 795-818.
Mukhopadhyay, S.K., Sarker, B.P. and
Panda, R.P. (1994), ``Machine-component
grouping in cellular manufacturing by
multidimensional scaling'', International
Journal of Production Research, Vol. 32 No. 2,
pp. 457-77.
[ 135]
Hassan M. Selim,
Reda M.S. Abdel Aal and
Araby I. Mahdi
Formation of machine groups
and part families: a modified
SLC method and comparative
study
Integrated Manufacturing
Systems
14/2 [2003] 123-137
Park, T. and WemmerloÈ v, U. (1995), ``A
comparative study of cell formation
techniques: experimental findings on data-
dependency and solution quality'', Decision
Sciences, Vol. 26 No. 4, pp. 475-502.
Rajagopalan, R. and Batra, J. (1975), ``Design of
cellular production systems ± a graph
theoretic approach'', International Journal of
Production Research, Vol. 13 No. 6, pp. 56-68.
Romesburg, H.C. (1984), Cluster Analysis for
Researchers, Lifetime Learning Publications,
Wadworth, Inc., Belmont, CA.
Sandbothe, R.A. (1998), ``Two observations on the
grouping efficacy measure for goodness of
block diagonal forms'', International Journal
of Production Research, Vol. 36 No. 11,
pp. 3217-22.
Sarker, B.P. (1996), ``The resemblance coefficients
in group technology: a survey and
comparative study of relational metrics'',
International Journal of Computers and
Industrial Engineering, Vol. 30 No. 1,
pp. 103-16.
Sarker, B.P. and Islam, K.M. (1999), ``Relative
performance of similarity and dissimilarity
measures'', International Journal of
Computers and Industrial Engineering,
Vol. 37, pp. 769-807.
Sarker, B.P. and Khan, M.A. (2001), ``Comparison
of existing grouping efficiency measures and
a new weighted efficiency measure'',
IIE Transactions, Vol. 33, pp. 11-27.
Seifoddini, H. (1989a), ``Single linkage versus
linkage clustering in machine cells formation
applications'', International Journal of
Computers and Industrial Engineering, Vol. 16
No. 3, pp. 419-26.
Seifoddini, H. (1989b), ``Duplication process in
machine cells formation in group
technology'', IIE Transactions, Vol. 21,
pp. 382-8.
Seifoddini, H. (1989c), ``A note on the similarity
coefficient method and the problem of
improper machine assignment in group
technology applications'', International
Journal of Production Research, Vol. 27 No. 7,
pp. 1161-5.
Seifoddini, H. and Djassemi, M. (1991), ``The
production data-based similarity coefficient
versus Jaccard's similarity coefficient'',
13th Annual Conference on Computers and
Industrial Engineering, Vol. 21, pp. 263-6.
Seifoddini, H. and Wolfe, P.M. (1986), ``Application
of the similarity coefficient method in group
technology'', IIE Transactions, Vol. 18 No. 2,
pp. 271-7.
Selim, H.M. (2002), ``The evaluation of similarity
measures performance in cellular
manufacturing: survey and analysis'',
6th International Conference on Production
Engineering and Design for Development,
Cairo, February.
Selim, H.M., Askin, R.G. and Vakharia, A.J.
(1998), ``Cell formation in group technology:
review, evaluation and directions for future
research'', International Journal of
Computers and Industrial Engineering, Vol. 34
No. 1, pp. 3-20.
Shafer, S.M. and Rogers, D.F. (1993a), ``Similarity
and distance measures for cellular
manufacturing Part II: an extension and
comparison'', International Journal of
Production Research, Vol. 31 No. 6, pp. 1315-26.
Shafer, S.M. and Rogers, D.F. (1993b), ``Similarity
and distance measures for cellular
manufacturing Part I: a survey'',
International Journal of Production Research,
Vol. 31 No. 5, pp. 1133-42.
Shambu, G., Suresh, N.C. and Pegels, C.C. (1996),
``Performance evaluation of cellular
manufacturing systems: a taxonomy and
review of research'', International Journal of
Operations & Production Management, Vol. 16
No. 8, pp. 81-103.
Sneath, P.H. (1957a), ``The application of
computers to taxonomy'', Journal of General
Microbiology, Vol. 17, pp. 201-26.
Sneath, P.H. (1957b), ``Some thoughts of bacterial
classification'', Journal of General
Microbiology, Vol. 17, pp. 184-200.
Sneath, P.H. and Sokal, R.R. (1973), Numerical
Taxonomy, 1st ed., Freeman,
San Francisco, CA.
Srinivasan, G. and Narendran, T. (1991),
``GRAGICS ± a nonhierarchical clustering
algorithm for group technology'',
International Journal of Production Research,
Vol. 29 No. 3, pp. 463-78.
Srinivasan, G., Narendran, T. and Mahadevan, B.
(1990), ``An assignment model for the part-
families problem in group technology'',
International Journal of Production Research,
Vol. 28 No. 1, pp. 145-52.
Stanfel, L.E. (1985), ``Machine clustering for
economic production'', Engineering Costs and
Production Economics, Vol. 9, pp. 73-81.
Vakharia, A.J. and Selim, H.M. (1994), ``Group
technology'', in Dorf, R.C. and Kusiak, A.
(Eds), Handbook of Design, Manufacturing
and Automation, 1st ed., John Wiley & Sons,
New York, NY, pp. 435-60.
Vakharia, A.J. and WemmerloÈ v, U. (1990),
``Designing a cellular manufacturing system:
a material flow approach based on operation
sequences'', IIE Transactions, Vol. 22 No. 1,
pp. 84-97.
Vakharia, A.J., Askin, R.G. and Selim, H.M.
(1999), ``Flexibility considerations in cell
design'', in Irani, S.A. (Ed.), Handbook of
Cellular Manufacturing Systems, 1st ed.,
John Wiley & Sons, New York, NY,
pp. 249-74.
Veramani, D. and Mani, K. (1996), ``Optimal
clustering in vertex-tree graphic matrices'',
International Journal of Production Research,
Vol. 34 No. 9, pp. 2587-2611.
Waghodekar, P.H. and Sahu, S. (1984), ``Machine-
component cell formation in group
technology'', MACE, International Journal of
Production Research, Vol. 22 No. 6, pp. 937-48.
[ 136]
Hassan M. Selim,
Reda M.S. Abdel Aal and
Araby I. Mahdi
Formation of machine groups
and part families: a modified
SLC method and comparative
study
Integrated Manufacturing
Systems
14/2 [2003] 123-137
Wei, J.C. and Kern, G.M. (1989), ``Commonality
analysis: a linear cell clustering algorithm for
group technology'', International Journal of
Production Research, Vol. 27 No. 12,
pp. 2053-62.
WemmerloÈ v, U. (1984), ``Comments on direct
clustering algorithm for group formation in
cellular manufacturing'', International
Journal of Production Research, Vol. 22 No. 3,
pp. vii-ix
Yasuda, K. and Yin, Y. (2001), ``A dissimilarity
measure for solving the cell formation
problem in cellular manufacturing'',
International Journal of Computers and
Industrial Engineering, Vol. 39 No. 1,
pp. 1-17.
[ 137]
Hassan M. Selim,
Reda M.S. Abdel Aal and
Araby I. Mahdi
Formation of machine groups
and part families: a modified
SLC method and comparative
study
Integrated Manufacturing
Systems
14/2 [2003] 123-137