This article was downloaded by: [NUS National University of Singapore]

On: 05 June 2014, At: 15:20

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.tandfonline.com/loi/tprs20

Similarity coefficient algorithms for solving the group

technology problem
a a
A. KUSIAK & M. CHO
a
Intelligent Systems Laboratory, Department of Industrial Engineering , College of
Engineering, University of Iowa , Iowa City, IA, 52242, USA
Published online: 07 May 2007.

To cite this article: A. KUSIAK & M. CHO (1992) Similarity coefficient algorithms for solving the group technology problem,
International Journal of Production Research, 30:11, 2633-2646, DOI: 10.1080/00207549208948181

To link to this article: http://dx.doi.org/10.1080/00207549208948181

PLEASE SCROLL DOWN FOR ARTICLE

Taylor & Francis makes every effort to ensure the accuracy of all the information (the “Content”) contained in the
publications on our platform. However, Taylor & Francis, our agents, and our licensors make no representations
or warranties whatsoever as to the accuracy, completeness, or suitability for any purpose of the Content. Any
opinions and views expressed in this publication are the opinions and views of the authors, and are not the
views of or endorsed by Taylor & Francis. The accuracy of the Content should not be relied upon and should be
independently verified with primary sources of information. Taylor and Francis shall not be liable for any losses,
actions, claims, proceedings, demands, costs, expenses, damages, and other liabilities whatsoever or howsoever
caused arising directly or indirectly in connection with, in relation to or arising out of the use of the Content.

This article may be used for research, teaching, and private study purposes. Any substantial or systematic
reproduction, redistribution, reselling, loan, sub-licensing, systematic supply, or distribution in any
form to anyone is expressly forbidden. Terms & Conditions of access and use can be found at http://
www.tandfonline.com/page/terms-and-conditions
 INT. J. PROD. RES., 1992, VOL. 30, NO. 11, 2633-2646

Similarity coefficient algorithms for solving the group technology

problem

A. KUSIAKH and M. CHOt

This paper discusses two formulations of the group technology (GT) problem.
The first one involves basic and alternative process plans without bottleneck parts.
The second formulation considers basic process plans and bottleneck parts or
Downloaded by [NUS National University of Singapore] at 15:20 05 June 2014

bottleneck machines. The algorithms developed in the paper are illustrated with
numerical examples. Computational results are also provided.

1. Introduction
Group technology (GT) has been recognized as one of the key factors to improve
productivity of manufacturing systems. The basic idea of GT is to group together
parts require that similar operations and machines corresponding to these operations.
For detailed review of GT and its advantages, see Fazakeriay (1974), Ham et al.
(1985), and Gallagher and Knight (1986).
One of the frequently used representations of the GT problem is a machine-part
incidence matrix [aij] which consists of' I' (empty) entries, where an entry 1 (empty)
indicates that machine i is used (not used) to process part j. Typically, when an initial
machine-part incidence matrix [a;J is constructed, clusters of machines and parts
are not visible. A clustering algorithm allows transforming the initial incidence matrix
into a structured (possible block diagonal) form. To illustrate the clustering approach
to GT, consider the machine-part incidence matrix (1).

Part number

1 2 3 4 5

[:
1

J
2 I Machine
[aij] = number (1)
3 1
4 1

Rearranging rows and columns in matrix (1) results in matrix (2). Two machine
cells (clusters) Me-l = {2, 4} and MC-2 = {I, 3}, and two corresponding part families
PF-I = {I, 3} and PF-2 = {2, 4, 5} are visible in matrix (2).

Received November 1991.

t Intelligent Systems Laboratory, Department of Industrial Engineering, College of Engin-
eering, University of Iowa, Iowa City, IA 52242, USA
t To whom correspondence should be addressed.
0021>-7543/92 S3W © 1992 Taylor & Francis Ltd.
 2634 A. Kusiak and M. Cho

PF-I PF-2

~3
I
I I
) 2 4 5

)
-
MC-I-C 2
4 )
(2)
MC-2 -e) ) ) )
3 ) ) )
-
Downloaded by [NUS National University of Singapore] at 15:20 05 June 2014

Clustering of a binary incidence matrix may result in the' following two categories
of clusters:
(1) Mutually separable clusters
(2) Partially separable clusters
Mutually separable clusters are shown in matrix (2), while partially separable clusters
are presented in matrix (3).

PF-l PF-2

~
) 3
n
2 4 5

) ) ) (3)
MC-I-C ; ) )

MC- 2 -e 3) )
) )
)
)
-
Matrix (3) cannot be separated into two disjoint clusters because of part 5, which
is to be machined in two cells, MC-) and MC-2. Removing part 5 from matrix
(3) results in the decomposition of matrix (3) into two separable machine cells,
MC-/={),2} and MC-2={3,4} and two part families, PF-I={1,2} and
PF-2 = {3,4}. The two clusters are called partially separable clusters and the
overlapping part is called a bottleneck part (i.e. the part that is processed on machines
belonging to more than one machine cell).
A way to eliminate a bottleneck part is to use an alternative process plan. For
example, assuming that an alternative process plan for part 5 in matrix (3) involves
machines 2 and 4 would result in two mutually separable machine cells. Alternative
process plans are frequently available for many parts. It is obvious that grouping of
parts with alternative routes increases the likelihood of generating ideal machine cells.
Analogous to the bottleneck part, a bottleneck machine is defined. A bottleneck
machine is a machine that processes parts belonging to more than one cell, i.e. it
does not allow for decomposition of a machine-part incidence matrix into disjoint
submatrices. For example, machine 3 in matrix (4) does not permit decomposition
of that matrix into two machine cells and two part families.
 Similarity coefficient algorithms for GT 2635

Part number
2 3 4 5 6

1 1 1
Machine 2 1 1 (4)
number 3 1 1 1 1 1
4 1 1 1 1
5 1 1 1

A way to decompose matrix (4) into two disjoint submatrices is to use an additional
Downloaded by [NUS National University of Singapore] at 15:20 05 June 2014

copy of machine 3. The latter leads to the transformation of matrix (4) into matrix
(5). Two machine cells, MC-l = {1, 2, 3(l)} and MC-2 = {3(2), 4, 5} and two corre-
sponding part families, PF-l = {1, 2} and PF-2 = p,4, 5, 6} are shown in matrix (5).

PF-I PF-2

n
1 2
I
3 4
I
5
I
6
-
1 1 1
MC-I-[ 2 1 1
3(1) 1 1 (5)
3(2) 1 1 1
MC-2-[ 4 1 1 1 1
5 1 1 1
'--

To solve the matrix formulation of the group technology problem, the following
heuristic approaches have been developed:
• Similarity coefficient methods (McAuley 1972, Seifoddini and Wolfe 1986).
• Sorting-based algorithms (King 1980, Chan and Milner 1982).
• Bond energy algorithms (McCormick et 01. 1972, Slagle et al. 1975, Bhat and
Haupt 1976).
• Cost-based methods (Askin and Subramanian 1987, Kusiak and Chow 1987).
• Extended cluster identification algorithm (Kusiak 1990).
• Within-cell utilization based heuristic (Ballakur and Steudel 1987).
• Non-hierarchical clustering algorithm (Chandrasekharan and Rajagopalan
1987).
In the following section, two similarity measures are defined and incorporated in two
heuristic algorithms. Most of the GT algorithms employ a binary machine-part
incidence matrix and do not use similarity measures. Many existing similarity
measures (Table 1) are suitable for problems that do not involve alternative process
plans and bottleneck parts or machines. The two similarity measures discussed next
are suitable for solving the latter cases of the GT problem that have not been
well studied in the literature.
 2636 A. Kusiak and M. Cho

Coefficient Reference Formula Value Form

Minkowski Arthanari and Dodge (1981) Ct,la" - a jk lP) 'Ip 2 Integer

,
L .5'(a'k> a jk)
k=1
McAuley McAuley (1972) , 0·6 0<;;5<;;1
L .5"(a'k' ajk)
k= I nij
Jaccard Leskowsky et a1. (1987) 0·6 0<;;5<;;1
(n, + nj - n,)
Downloaded by [NUS National University of Singapore] at 15:20 05 June 2014

Dulla er a/. Dulla et a1. (1986) nj + nj - 2nij 2 Integer

2nij
Dice-Sorensen Chu and Pan (1988) 0·8 0<;;5<;;1
(n, + n)

Dot-product Chu and Tsai (1989) 0·4 0<;;5<;;1

Chu and Lee Chu and Lee (1989) 2 Integer

if a ik ;?; ail or
Kusiak and Cho This paper aik ~ Q j h for all i O. I
0 otherwise

b'(di~' a jl) = I, if ail = ajl = I; 0, otherwise.

.5"(a", aJ, ) = 0, if a" = aJ, = 0; I. otherwise.
nj = number of machines visited by part i.
nu = number of machines visited by part i and j.
Table 1. Similarity measures and their values for matrix (7).

2. Similarity measures
In order to develop clustering heuristics. two new similarity measures are
proposed. For any two parts (process plans) represented with column k and I in the
incidence matrix [aij]. the zero-one similarity measure is defined as follows:
I if a'k ;;;: ail or a'k ,;; ail for all i
SkI = {0 (6)
otherwise
The value of the similarity measure SkI = I indicates that the part vector (process
plan) I is a subset of the part vector (process plan) k or vice versa. where
part vector k = [a'b a 2k' ...• amk]T. and
part vector I = [al/. a 21> •••• am,]T

Comparison of the existing similarity measures with the zero-one similarity measure
SkI for matrix (7) is shown in Table 1.
 Similarity coefficient algorithms for GT 2637

Part number
k

1 1
1 0
Machine 1 1 (7)
number o 0
1 0
o 0
1 1
Downloaded by [NUS National University of Singapore] at 15:20 05 June 2014

In addition to the similarity measure (6), a modified zero-one similarity measure

sW for the part is defined:
sW = {Io CX p ;;.

otherwise
t
for parts

where the parameter t is to be specified by the user, tE [0, 1]

m

L .1(aik, ail)
i= 1 if aik = ail = 1
otherwise

The modified similarity measure can be used for parts or machines for which the
similarity measure Ski would have value O.
To measure the quality of clusters, the following grouping efficiency GE measure
has been introduced by Chandrasekharan and Rajagopalan (1989):
GE = qn, + (1 - q)n 2
number of entries' 1' in the diagonal blocks
where n, = ----------------'=------
total number of elements in the diagonal blocks
number of entries '0' in the off-diagonal blocks
n2 =
total number of elements in the off-diagonal blocks
q =a weighting factor having a value between 0 and 1.
Chandrasekharan and Rajagopalan (1986) suggested the value of q = 0·5. This
function is non-negative and its range is zero to one. There is no good way of selecting
the value of q published in the literature. If the value of GE is 1 it means that the
matrix has a perfect block diagonal form, and zero is the opposite case.

3. Algorithms
To solve the group technology problem with alternative process plans, a branch-
ing algorithm is developed. Alternative process plans increase the probability of
forming mutually separable machine cells and families. Algorithm I presented below
solves the GT problem with alternative process plans.
 2638 A. Kusiak and M. Cho

The terms used in algorithm 1 are defined as follows:

node represents a process plan
node degree number of arcs at a node
arc represents similarity of 1 between two nodes
clique a closed tour between process plans representing different parts

3.1. Algorithm I
Step O. (Initialization): begin with the generalized matrix [aij] with basic and
Downloaded by [NUS National University of Singapore] at 15:20 05 June 2014

alternative process plans for each part and set level number L = O.
Step 1. (Transformation): transform matrix [aij] into zero-one similarity matrix
[sij].
Step 2. Represent the current matrix [sij] with a transition graph.
Step 3. (Branching): determine maximum cliques in the transition graph, i.e. a
part on a node graph (see Kusiak 1990, pp. 223-224). Assign level number
L = L + 1 to each clique and branch on a clique with the maximum total
sum of the node degrees. Delete from [sij] the rows and columns
corresponding to the parts that have process plans included in the clique
selected and go to step 4.
Step 4. (Fathoming): exclude from further consideration a node which produces
a larger number of machine cells (process families) than the current
solution.
Step 5. (Stopping rule): stop when the current solution is acceptable; otherwise
backtrack to the nearest unfathomed node at level L and repeat from
step 2.
The most computationally complex steps in algorithm 1 are step 1 and step 3. The
computational time complexity of step 1 is O(m 2p ), where p is the number of columns
in the incidence matrix. The complexity of step 3 is dominated by the clique problem
which is NP complete. For example, a pairwise counting heuristic solves the clique
problem in O(p2) time. Algorithm 1 is illustrated with example 1.
Example 1. Consider the incidence matrix (8) with alternative process plans.
Step O. Begin with the matrix (8) and set level number L = O.

Part number
1 2 3 4 5
l----"l n n n n
Process plan number
0 o 0 0 0 0 000
1 1
1 23456 789 0 1
(8)

~[
1 1
Machine
[aij] =
4 1 1 1 1 1
1 1
1
J number
 Similarity coefficient algorithms for GT 2639

Step 1. Transform matrix (8) into the zero-one similarity matrix (9).

Process plan number

0 0 0 0 0 0 o 0 0 1 1 Node
1 2 3 4 5 6 7 8 9 0 1 degree

01 000 0 o 0 0 001 0 1
02 000 000 1 o 0 0 0 1
03 000 o 0 o 0 000 1 1
04 000 o 0 o 0 o 0 0 0 0
000 o 0 o 0 o 1 0 1
Downloaded by [NUS National University of Singapore] at 15:20 05 June 2014

[sij] = 05 2 (9)
06 000 o 0 o 0 J 001 2
07 010 o 0 o 0 o 0 0 0 1
08 000 o 0 1 0 000 1 2
09 000 o 1 o 0 000 1 2
10 100 o 0 o 0 o 0 0 0 1
11 001 o 1 1 0 1 100 5

Step 2. The transition graph representing matrix (9) is shown in Fig. 1.

Figure 1. Transition graph for matrix (9).

Step 3. Two cliques (6,8,11) and (5,9,11), each of size three, are determined (see
level L = 1 in Fig. 2). The clique (5,9, 11) with the corresponding total or
node degree of 9 is selected. The process plans 4, 5, 8, 9, 10 and 11
corresponding to parts 2, 4, 5 are removed from matrix (9) which results in
matrix (10).

Level
L--Q

1.>=1

1.>=2

Figure 2. Branching tree.

 2640 A. Kusiak and M. Cho

Process plan nurnher

0 0 0 0 0 Node
1 2 3 6 7 degree

01 0 0 0 0 0 0
02 0 0 0 0 1 I
[sij] = 0 (10)
03 0 0 0 0 0
06 0 0 0 0 0 0
07 0 1 0 0 0 I
Downloaded by [NUS National University of Singapore] at 15:20 05 June 2014

Step 5. Steps 2 and 3 are repeated for the similarity matrix (10) resulting 10
the clique (2, 7) shown at level L = 2 in Fig. 2.

The outcome of algorithm 1 is shown in matrix (11) where two machine cells and
two process families are visible. Branching on the node 6-8-11 in Fig. 2 does not
produce a solution of better quality than the one in matrix (11), namely three process
families should be produced.

Process Process
plan plan
family 1 family 2

2 7 5 9 11
-
2 1 Machine (11)
4 1 cell - 1
1 1 1 Machine
3 1 1 cell-2
-
Algorithm 1 is perhaps the only heuristic developed to date for solving the
GT problem with the presence of alternative process plans. Further research with
the algorithm should be directed towards defining tight bounds that could produce
an efficient branch-and-bound procedure.
Algorithm 2 presented next uses the modified zero-one similarity measure for
solving' the GT problem. It is applicable for the problem with basic process plans
only where the ideal diagonal structure is not embedded into the incidence matrix.
Depending on the type of the similarity measure used sW (for parts) or sl~) (for
machines), in addition to machine cells and part families, bottleneck parts or bottle-
neck machines are determined. Algorithm 2 is an extension of the cluster identification
algorithm presented in Kusiak and Chow (1987) in that it uses as the input a similarity
coefficient matrix rather than a machine-part incidence matrix. The quality of the
solution generated by algorithm 2 depends on the threshold value.

3.2. Algorithm 2
Step O. (Initialization): begin with the matrix [aij], select a threshold value t and set
the iteration number k = 1.
 Similarity coefficient algorithms for GT 2641

Step 1. Transform matrix [au] into the modified similarity matrix [s\}')].
Step 2. Select any row i ofthe modified similarity matrix [sj}')](k) and draw horizontal
line h, through it. ([S!),Jk) is read: modified similarity matrix [sl}')] for parts
at iteration k.)
Step 3. For each non-zero entry crossed by the horizontal line hi, draw vertical
lines Vj'
Step 4. For each non-zero entry crossed-once by vertical line Vj' draw a hor-
izontalline s;
Step 5. Repeat steps 3 and 4 until there are no more 'crossed-once' non-zero
entries in the matrix.
Step 6. Transform [sl}')Jk) into [S\}')Jk + I) by removing rows and columns corre-
Downloaded by [NUS National University of Singapore] at 15:20 05 June 2014

sponding to all the horizontal and vertical lines drawn in steps 2-4. The
removed rows and columns from a part family.
Step 7. [fthe modified similarity matrix [s\}')Ju 1) = 0, stop; otherwise set k = k + I
and to step 1. (0 denotes a modified similarity matrix with all elements equal
to zero.) The computational time complexity of algorithm 2 is determined
by step 2 and is O(m 2n).

Example 2. Consider the machine-part incidence matrix (12).

StepO. Begin with the matrix (12) at iteration k =1
Part number
1 2 3 4 5 6 7 8

1 1 1 1
2 1 1 1 1 1 (12)
3 1 1 Machine
4 1 number
5 1 1 1 1
6 1 1

Step 1. Assume the threshold value t = 0'6 and transform matrix (12) into the
similarity matrix (13).

Part number
2 3 4 5 6 7 8

1 0 0 1 0 1 1 0 1
2 0 0 0 0 0 0 1 0
3 1 0 0 0 1 1 0 1
4 0 0 0 0 0 0 1 0 Part
5 1 0 1 0 number (13)
0 1 0 1
6 1 0 1 0 1 0 0 1
7 0 1 0 1 0 0 0 0
8 1 0 1 0 1 1 0 0
 2642 A. Kusiak and M. Cho

Step 2. Select any row, for example 1, and draw a horizontal line hi'
Part number
1 234 5 678

1 ·····e····~·····I·····~·····I·..·..i·····e·····i·· ...... hi
200000010
310001101
400000010 (14)
510100101
610101001
701010000
Downloaded by [NUS National University of Singapore] at 15:20 05 June 2014

810101100

Step 3. For the non-zero entries crossed by hi' draw vertical lines V 3, vs, V6,
and Va including VI corresponding to the starting horizontal line h, (see
matrix (15)).
Part number
123 4 567 8

1··..·~·
•
..·e....·t..
•
·..G····+·..
••
·-l...··-o--·J·f ...... hI
2 CloCloOOICl
3 io~oiloi
4 ~O~O~~I~ (15)
5 ioio~ioi
6 ioloi~oi
7 ~I~I~OO~
8 ioioiio~
~ i ~ i :
Vi V3 Vs V6 VB

Step 4. For the entries I crossed by VI' V 3, V S, V6, and Va, horizontal lines
are drawn in matrix (16).
Step 5. Since there are no single-crossed non-zero entries in matrix (16), go to
step 6.
Step 6. Part family PF-l = {I, 3, 5, 6, 8} is formed. Matrix (16) is transformed
into matrix (17).
Part number
12345678

1·..·-{j-..··e··..+-e-+-J···..-o-·_·l hI
2 bobo~blP
3 ....··i.....
•
O.~• ....o.....:
l-.-l.-O---i..._-..
. :
h3 (16)
4 CJ 0 Cl 0 ~ Cl 1 ~
5 ..+_·-o-·..·t-·-o-·_~-+-G .. - -.h s ·+..
6 ..·..·l·_·-e····+..
• •
··e··-+-··C!····-o-·_·j.·
.' •
....... h.6
7 ClICJI~ClO~
8 ..·-i·_··o-·i-·..~--j-·l·-~·-9· ..-- h a
I ! I ~ :
 Similarity coefficient algorithms for GT 2643

Step 7. Set iteration number k = 2 and go to step 2.

Part number
247

o
42[00 o (17)
7 1 1

The result of step 2 and 3 in the second iteration is shown in matrix (18).
Part number
Downloaded by [NUS National University of Singapore] at 15:20 05 June 2014

2 4 7

;ErrJjh' ,
V2
.
V7
(18)

In step 4, for the non-zero entry crossed once by the vertical line v7 ,
horizontal line h4 is drawn.
Part number
247

2
4
t·.·-t~·. ·._····-e-(} _.~~. ··-h
-t h.
2
(19)
7 i 1 E>
~ :
V2

Repeating step 3 and 4 would result in a horizontal line h 7 and vertical line v4 .
As a result of clustering, part family PF-2 = {2, 4, 7} is formed. Applying algorithm
2 to matrix (12) results in matrix (20) which indicates two machine cells and two part
families. The machine cells have been generated from matrix (12) based on the parts
in each of the two part families. The solution in matrix (20) shows bottleneck parts
2 and 7 with group efficiency GE = 87·49 for q = 0·5 and t = 0·6.
Part number
13568247
-
2 I 1 1 1 1 1 1
3 1 1 1 Machine
5 1 1 1 1 1 number ("'VI
1 1 1 1
4 1 1
6 1 1
-
4. Computational results
The two algorithms have been programmed in PASCAL, and run on a Prime
9955 computer. Twenty problems taken from the literature were tested. The computa-
 N

t
Downloaded by [NUS National University of Singapore] at 15:20 05 June 2014

Reference algorithm Algorithm 2

Problem Size
number (m x n) Reference PF BP GE PF BP GE CPU (s)

1 5x 7 King and Nakomchai (1982) 2 0 91-18 2 0 9\-l8 0-012

2 5x 7 Waghodekar and Sahu (1984) 2 2 85-62 2 2 85-62 0-015
3 5x 7 Waghodekar and Sahu (1984) 2 2 85-62 2 2 85-62 0-015
4 8x 7 Kusiak (1992) 3 1 89-38 3 1 89-38 0-021
5 7x 8 Kusiak and Chow (1987) 3 0 82-50 3 0 82-50 0-027
6 7x II Kusiak and Chow (1987) 2 3 70-95 2 3 70-95 0-048
7 10 x 8 Ham et al. (1985) 3 2 92-58 3 2 92-58 0-033
8 9x 12 Seifoddini (1989) 3 3 93-75 3 3 93-75 0-061
9. 9x 12 Seifoddini (1989) 3 4 90-70 3 4 90-70 0-063 :....
10 9x 12 Seifoddini (1989) 3 3 92-48 3 3 92-48 0-060
11 12 x 10 McAuley (1972) 3 5 88:\3 3 5 88-\3 0-060 ~
12 15 x 10 Chan and Milner (1982) 3 0 96-00 3 0 96-00 0-066 '"is-
;>;-
13 8x 20 Chandrasekharan and 3 9 95-83 3 9 95-83 0-155
I:l
Rajagopalan (1986) ::
14 20 x 20 Harhalakis et ai_ (1990) 4 11 78-21 4 11 79-62 0-367 "'-
15 14 x 24 King (1980) 4 2 83-81 3 0 80-73 0-397 ~
16 40 x 24 Chandrasekharan and 7 0 100-00 7 0 100-00 1-594
Rajagopalan (1989) Q
c
17 16 x 43 King (1980) 4 2 74-95 5 3 80-71 1-421
18 30 x 41 Kumar and Vannelli (1987) 3 6 66-89 3 6 66-89 2-425
19 30 x 50 Stanfel (1985) 4 0 68-33 4 0 68-33 4-273
20 40 x 100 Chandrasekharan and 10 37 90-33 10 37 90-33 2-000
Rajagopalan (1987)

m: number of machines
n: number of parts
PF: number of part families
BP: number of bottleneck parts
GE: group efficiency
Table 2_ Performance of algorithm 2 for selected problems.
 Similarity coefficient algorithms for GT 2645

tional results were evaluated in terms of CPU time, grouping efficiency, and the
number of bottleneck parts. The computational results show that value of parameter
t in the modified similarity measure defined in this paper has an impact on the quality
of clusters.
The computational results are summarized in Table 2. The quality of results is
measured in the number of bottleneck parts (BP) and grouping efficiency (GE) for
q = 0'5 and t = 1. Note that the value of q is selected arbitrarily. For the two quality
measures, the solutions generated by the two algorithms are at least as good as the
solutions provided by the reference algorithm. The CPU time requirement of
algorithm 2 is modest. Since in the literature there is no data for the GT problem
with alternative process plans, algorithm I was not tested. The computational
Downloaded by [NUS National University of Singapore] at 15:20 05 June 2014

complexity of algorithm 1 is dominated by calculating the similarity matrix. Assuming

that the clique problem is solved heuristically, its time efficiency should be close to
that of algorithm 2.

5. Conclusion
In this paper, two new similarity measures were proposed. The first measure is
applicable for the group technology (GT) problem with basic and alternative process
plans. It works well for cases where a block diagonal structure is embedded into the
machine-part incidence matrix. The second similarity measure generalizes the first
one and is applicable to non-decomposable incidence matrices.
Two heuristic algorithms were discussed for solving the GT problem. One is
suitable for the GT problem with basic and alternative process plans. The second
algorithm is the clustering identification algorithm (Kusiak and Chow 1987) applied
to the similarity matrix. This study showed that similarity coefficients have a
significant effect on the quality of clusters. Algorithm 2 was tested on 20 problems
available in the literature. It provides solutions of good quality and requires modest
CPU time. One of the advantages of algorithm 2 is flexibility due to the similarity
measure it uses. Changing the threshold value t in the extended zero-one similarity
measure provides different solutions.

Acknowledgment
This research has been partially supported by a grant from Rockwell International
Corporation.

References
ARTHANARI, T. S., and DODGE, Y., 1981, Mathematical Programming in Statistics (New York:
Wiley).
ASKIN, R., and SUBRAMANIAN, S., 1987. A cost-based heuristic for group technology configura-
tion. International Journal of Production Research, 25(1), 101-114.
BALLAKUR. A., and STEUDEL, H. J., 1987, A within-cell utilization based heuristic for designing
cellular manufacturing systems. International Journal of Production Research, 25(5),
639-665.
BHAT, M. Y., and HAuPT, A., 1976, An efficient clustering algorithm. IEEE Transaction on
Systems, Man, and Cybernetics, SMC-6(I), 61-64.
CHAN, H. M., and MILNER, D. A., 1982, Direct clustering algorithm for group formation in
cellular manufacture. Journal of Manufacturing Systems, 1(1), 63-74.
CHANDRASEKHARAN, M. P.,and RAJAGOPALAN, R., 1986, An ideal seed non-hierarchical clustering
algorithm for cellular manufacturing. International Journal of Production Research, 28(2),
451-464.
 2646 Similarity coefficient algorithms for GT

CHANDRASEKHARAN, M. P., and RAJAGOPALAN, R., 1987, ZODIAC-an algorithm for concurrent
formation of part-families and machine-cells. International Journal of Production Re-
search, 25(6), 835-850.
CHANDRASEKHARAN, M. P., and RAJAGOPA~AN, R., 1989, Groupability: an analysis of the
properties of binary data matrices for group technology. International Journal of
Production Research, 27(6), 1035-1052.
CHU, C. H., and LEE, W., 1989, An efficient heuristic for grouping part families/machine cells.
Research Paper, Iowa State University, Ames, Iowa.
CHU, C. H., and PAN, P., 1988, A comparison of hierarchical clustering techniques for.
manufacturing cellular formation. Proceedings of the Third International Conference on
CAD/CAM, Robotics, and Factory of the Future, edited by B. Prasad (New York:
Springer Verlag).
CHU, C. H., TSAI, M., 1989, Clustering analysis in manufacturing cellular formation. OMEGA,
Downloaded by [NUS National University of Singapore] at 15:20 05 June 2014

(17) (3), 289-295.

DUTrA, S. P., LASHKARI, R. S., NADOLI, G., and RAVI, T, 1986, A heuristic procedure for
determining manufacturing families from design-based groupings for flexible man-
ufacturing systems. Computers and Industrial Engineering, 10(3), 193-201.
FAZAKERLAV, G. M., 1974, Group technology: social benefits and social problems. The
Production Engineer, October, 383-386.
GALLAGER, C. C., and KNIGHT, W. A., 1986, Group Technology: Production Methods in
Manufacture (New York: John Wiley).
HAM, I., HITOMI, K., and YOSHIDA, T, 1985, Group Technology: Application to Product
Management (Boston, MA: Kluwer-Nijhofl).
HARHALAKIS, G., NAGI, R., and PROTH, M., 1990, An efficient heuristic in manufacturing cell
formation for group technology applications. International Journal of Production Re-
search, 28(2), 185-198.
KING, J. R., 1980, Machine-component grouping in production flow analysis: an approach
using a rank order clustering algorithm. International Journal of Production Research,
18(2),213-232.
KING, 1. R., 1980, Machine-component group formation in group technology. The International
Journal of Management Science, 8(2), 193-199.
KUMAR, K. R., and VANNELLI, A., 1987, Strategic subcontracting for efficient disaggregated
manufacturing, International Journal of Production Research, 25(12), 1715-1728.
KUSIAK, A., 1990, Intelligent Manufacturing Systems (Engelwood Cliffs, NJ: Prentice Hall).
KUSIAK, A., 1992, Group technology. Intelligent Design and Manufacturing, edited by A. Kusiak
(New York: John Wiley), pp. 289-302.
KUSIAK, A., and CHOW, W. S., 1987, Efficient solving of the group technology problem. Journal
of Manufacturing Systems, 6(2), 117-124.
LESKOWSKY, L., LOGAN, L., and VANNELLI, 1987, Group technology decision aids in an expert
system for plant layout. Modern Production Management Systems, edited by A. Kusiak
(Amsterdam: North Holland), pp. 561-585.
McAuLEV, J., 1972, Machine grouping for efficient production. The Production Engineer,
February, 53-57.
MCCORMICK, W. T. SCHWEITZER, P. J., and WHITE, T E., 1972, Problem decomposition and
data reorganization by a cluster technique. Operations Research, 52, 993-1009.
SIEFODDlNI, H., 1989, A probabilistic approach to machine cell formation in group technology.
International Industrial Engineering Conference & Societies' Manufacturing and Pro-
ductivity Symposium Proceedings, Toronto, Canada, pp. 625-629.
SEIFODDINI, H., and WOLFE, P. M., 1986, Application of the similarity coefficient method in
group technology. liE Transactions, 18(3), 2710277.
SLAGLE, 1. L., CHANG, C. L., and HELLER, S. R., 1975, A clustering and data reorganization
algorithm. IEEE Transactions on System, Man, and Cybernetics, SMC-5, 125-128.
STANFEL, L. E., 1985, Machine clustering for economic production. Engineering Costs and
Production Economics, 9, 73-81.
WAGHODEKAR, P. H., and SAHU, S., 1984, Machine-component cell formation in group
technology: MACE. International Journal of Production Research, 22(6), 937-948.