You are on page 1of 10

ARTICLE IN PRESS

Robotics and Computer-Integrated Manufacturing 24 (2008) 488–497


www.elsevier.com/locate/rcim

CLASS: An algorithm for cellular manufacturing system and layout


design using sequence data
Iraj Mahdavia,, B. Mahadevanb
a
Mazandaran University of Science and Technology, P.O. Box 734, Babol, Iran
b
Indian Institute of Management Bangalore, Bannerghatta Road, Bangalore 560 076, India
Received 25 March 2007; accepted 6 July 2007

Abstract

Cell formation problem in CMS design has received the attention of researchers for more than three decades. However, use of
sequence data for cell formation has been a least researched area. Sequence data provides valuable information about the flow patterns of
various jobs in a manufacturing system. Therefore, it is only natural to expect that use of sequence data must result in not only
identifying the part families and machine groups but also the layout (sequence) of the machines within each cell. Unfortunately, such an
approach has not been taken in the past while solving CMS design problem using sequence data. In this paper, we fill this gap in the
literature by developing an algorithm that not only identifies the cells but also the sequence of machines in the cells in a simultaneous
fashion. The numerical computations of the algorithm with the available problems in the literature indicate the usefulness of the
algorithm. Further, it also points to the untapped potential of such an approach to solve CMS design and layout problem using sequence
data.
r 2007 Elsevier Ltd. All rights reserved.

Keywords: CMS design; Sequence data; Layout design

1. Introduction various objectives, such as maximizing utilization, mini-


mizing material handling cost, and minimizing load
Cellular manufacturing is an application of the Group unbalance, have been employed in assessing the quality
Technology philosophy to designing manufacturing sys- of the solution. These approaches can be found in Ballakur
tems. The main idea of Group Technology is to identify and Steudel [7], Wei and Gaither [8], Shafer and Rogers [9],
similar manufacturing processes and features, where and Lee and Chen [10].
machines are grouped into machine cells based on their In the last three decades of research in cell formation,
contribution to the production process. One of the first researchers have mainly used zero–one machine compo-
problems encountered in implementing cellular manufac- nent incidence matrix as the input data for the problem.
turing is that of cell formation. Over the last three decades, However, of late efforts are being made to use other data
many solution methods such as mathematical program- structures such as interval data [5] and ordinal data
ming, heuristics, optimization procedures, and clustering (consisting of sequence of processing, see for instance
techniques have been proposed to address the cell [11–14]). Different data structures provide different set of
formation problems [1–3]. Most of the cell formation information and enable the cell designers to make
studies have focused on the independence of cells, and the appropriate use of them while solving the cell formation
number of inter-cell movements is commonly viewed as an problem.
indicator of that (see for instance [4–6]) In addition, Use of sequence data for cell formation problem pro-
vides additional information to the cell designer. Sequence
Corresponding author. Tel.: +98 911 113 1380; fax: +98 111 229 0118. data identifies the order in which jobs get processed in a
E-mail addresses: irajarash@rediffmail.com (I. Mahdavi), manufacturing setup. Therefore, this information could be
mahadev@iimb.ernet.in (B. Mahadevan). used not only for identifying part families and machine

0736-5845/$ - see front matter r 2007 Elsevier Ltd. All rights reserved.
doi:10.1016/j.rcim.2007.07.011
ARTICLE IN PRESS
I. Mahdavi, B. Mahadevan / Robotics and Computer-Integrated Manufacturing 24 (2008) 488–497 489

groups but also to arrive at the layout of machines within algorithm to identify machine cells. Suresh et al. [18]
each cell based on dominant flow patterns within each cell. utilized the fuzzy neural network approach to cell
Despite this simple truth, traditionally, the cell design formation using sequence data. The other notable works
problem and the layout problem are treated in a in cell formation using sequence data include Nair and
discontinuous fashion. Cell formation methodologies Narendran [12], Vakharia and Wemmerlov [6], Won and
proposed in the literature using sequence data have often Lee [13], Harhalakis et al. [5], and Kang and Wemmerlov
confined only to the cell design issue. On the other hand, [19]. Park and Suresh [20] compared the performance of
layout designers have not taken into consideration the cell fuzzy neural network with other clustering methods.
formation aspect. The solution is either taken as given or One approach to solve the grouping and layout problem
arrived independently without taking the similarity aspects has been to assume a certain structure of the layout and
between jobs in a shop. In such cases where the machine then develop solution methodologies for arriving at the
grouping and the intra-cell layout problems have been solution. Vakharia and Wemmerlov [6] proposed a
jointly considered, the solution methods utilize a sequential heuristic approach for the machine cell design, where
approach. The machine cells are found first, and then the machines within each cell are arranged along a linear flow
intra-cell layout is constructed based on the given cell line. Irani et al. [21] used maximal spanning arborescence
formation. Clearly in such an approach, the quality of the as a graphic structure to integrate the machine grouping,
final solution largely depends on initial cell formation. the intra-cell layout, and the inter-cell layout. Sequential,
We advocate that while using a sequence data for cell two-phase mathematical programming models were pro-
formation problem, it is important to also address the posed to decompose the joint problem. Liao [22] presented
layout, as the required information for arriving at the a sequential three-stage procedure, to determine the best
layout is already available with the cell designer. In this part routing, machine cells, and inter-cell layout for a line-
paper, the cell formation and layout design are considered type cellular manufacturing system. Approaches that are
simultaneously. We propose a new algorithm, Cell and more complicated have also been developed to address the
LAyout Solution using Sequence data (CLASS), that joint problem. Arvindh and Irani [23] presented an iterative
utilizes the sequence data as input to the problem and approach to design a cellular manufacturing system, where
identifies machine cells and the layout of machines within one iterative loop deals with machine and part grouping,
each cell. The objective is to make use of the valuable and another iterative loop varies the number of cells to find
information about the flow patterns of various jobs in a the best design. Akturk and Balkose [24] described a multi-
manufacturing system and obtain relevant performance criterion clustering approach that considers the manufac-
measures for the cell design and layout problem. The turing attributes, the operational sequences of part, and the
approach is illustrated by an example and a comparison within-cell layout. Due to the modeling complexity, a
has been done with other algorithms in the literature. heuristic approach is developed to find the part-machine
The paper is organized as follows: in the next section we grouping.
motivate our research by a review of the literature Heragu and Kakuturi [25] also attempted to integrate
pertaining to the use of sequence data for cell formation machine grouping and layout problems. The machine cells
and methodologies used for arriving at layout of the cells. are first formed by a heuristic, and near-optimal intra-cell
In Section 3 we develop the logic for our algorithm by and inter-cell layouts are constructed by a hybrid simulated
analyzing Nair and Narendran [12] solution. We develop annealing algorithm. Lee and Chiang [26] considered the
appropriate performance measures for the problem in this joint clustering-layout problem where machine cells are to
section. The CLASS algorithm is laid out in Section 4 be located along the bi-directional linear flow layout. They
and a numerical analysis of CLASS and its performance seek to minimize the actual inter-cell flow cost, instead of
vis-à-vis other algorithms is shown in Section 5. Finally in the typical measure that minimizes the number of inter-cell
Section 6 we draw some conclusions. movements. A three-phase approach, using the cut tree
network model, is developed to solve this joint problem.
2. Literature review Chiang and Lee [27] developed a genetic-based algorithm
with the optimal partition approach for the cell formation
Research in cellular manufacturing using sequence data in bi-directional linear flow layout, where the objective is to
has mainly confined to two areas; one is to identify minimize the actual inter-cell flow cost, instead of the
appropriate similarity measures for the cell formation typical measure that optimizes the number of inter-cell
problem and the other is to develop new algorithms for cell movements.
formation using sequence data. Selvam and Balasubrama- On the other hand, the layout problem has been long
nian [15] first reported the use of sequence data for cell addressed by researcher (see Kusiak and Heragu [28] and
formation using a heuristic procedure based on set cover- Heragu [29]). Houshyar and McGinnis [30] applied a
ing technique to identify cells. Choobineh [16] and Tam graphic approach to assign the facilities in a layout, where
[17] proposed new similarity coefficients using sequence the travel distance of work in process is to be minimized.
data for cell formation. Using Tam’s similarity coefficient, Among other works, Afentakis [31] addressed the unidir-
Kiang et al. [11] developed a neural network based ectional uni-cyclic manufacturing layout problems.
ARTICLE IN PRESS
490 I. Mahdavi, B. Mahadevan / Robotics and Computer-Integrated Manufacturing 24 (2008) 488–497

There are some more recent works regarding the use of Table 1
sequence data to address the cell formation problem, for Nair and Narendran [12]—8  20 machine–part matrix
example the researches done by Kim et al. [32], Yin and Machines
Yasuda [33], Yin et al. [34] and Xambre and Vilarinho [35].
We make several inferences based on the review of the 1 2 3 4 5 6 7 8
literature. First, known methods for cell formation
Parts
problem using sequence data have not addressed the 1 2 1
layout issue explicitly. As we will show in the next section, 2 1 2
while the solution methodologies might have correctly 3 2 1 5 3 4
identified the machines for each cell, arbitrarily laying 4 1 2 3 4
5 2 1
machines within each cell may result in a poor layout of the
6 1 2 5 3 4
machines in the shop. It is hard to find out from the 7 4 2 3 1
solution the order in which the machines in every group 8 1 2
needs to be laid out. In several of the methods described 9 1 3 2
above, we note that the problem and solution complexity 10 2 3 1
11 3 2 1
are high, thereby posing limitations to solving large
12 1 3 2
problem sizes. We also notice that the measure of 13 1 2
performance for a cell design cum cell layout problem is 14 1 2 3
not the pure ‘‘inter-cell’’ moves. While an appropriate 15 1 2
measure of inter-cell moves will indicate the quality of the 16 1 2
17 3 1 2
cells, other measures pertaining to flow patterns within
18 2 1 4 3
each cell is required for the layout problem. In this paper, 19 1 2
we develop an algorithm and an appropriate performance 20 2 1 3 4 5
measure that takes into consideration these requirements.
Before we state the algorithm we proceed to emphasize the
need for the algorithm and the performance measure by
taking an example from the literature. the complexity of planning and control. Therefore, one
way to understand how appropriate is the order of
3. Performance measures for the cell design and layout machines within each cell is to compute the number of
problem consecutive forward flows within the cell. If we examine
the pair of alternatives for cell 1, viz., (3,1) and (1,3),
Let us examine the solution provided by [12] Nair and and compute the number of consecutive forward flows
Narendran [12] for example 4 in their paper. Table 1 shows we find that in the case of (3,1) it is just one, whereas
the sequence data pertaining to the problem consisting of 8 in the case of (1,3) the number of consecutive flows is
machines and 20 parts and Table 2 is the rearranged form five. Therefore, we conclude that (1,3) is a better layout
of the input data as per the final solution using the CASE than (3,1).
algorithm proposed in the paper. The final solution has
three cells with machine groupings for each cell as Cell 1: One can identify appropriate ordering of machines
(3,1); Cell 2: (4,7,8,2); and Cell 3: (5,6). A closer look at within each cell using the above logic. Table 3 provides
Table 2 provides several interesting clues with respect to an alternative version of the input data of Nair and
the layout of the problem. The salient among them are as Narendran problem with the machines rearranged differ-
follows: ently. This arrangement provides the best measure of the
number of consecutive forward flows within each cell for
(a) The existing solution does not indicate anything with the Nair and Narendran solution. One can extend this logic
respect to the sequence in which the machines needs to to arrive at appropriate performance measures for the cell
be placed in each of these three cells. This is evident on design cum layout problem.
examining cell 1. If we consider the existing solution The existence of voids and exceptions (inter-cell move-
(3,1) as the layout of the machines, it is obvious that it ments) are typical of CMS design and cell designers
is inappropriate. We see from Table 2 that there is a endeavor to minimize them. On the other hand, in the
dominant flow of jobs from machine 1 to machine 3. case of layout design the focus is more on intra-cell
Therefore (1,3) is the correct order in which the movements. Layout design emphasizes the need for
machines are to be arranged. adjacency of processes as it can reduce material handling
(b) Layout design emphasizes the need for adjacency of costs. As pointed out earlier, from a production planning
processes as it can reduce material handling costs. and control perspective, reverse flows within a cell and
Furthermore, from a production planning and control skipping of workstations are undesirable, as they tend to
perspective, reverse flows within a cell and skipping of increase the complexity of planning and control. We define
workstations are undesirable, as they tend to increase two measures of performance for evaluating the goodness
ARTICLE IN PRESS
I. Mahdavi, B. Mahadevan / Robotics and Computer-Integrated Manufacturing 24 (2008) 488–497 491

Table 2
Nair and Narendran [12]—CASE solution

Parts

2 8 9 11 13 14 16 17 19 3 4 6 7 18 20 1 5 10 12 15

Machines
3 2 2 3 2 2 3 2 1 2
1 1 1 1 3 1 1 1 3 1 2
4 5 2 2 2 1 1 2
7 1 3 3 3 3 4 4 2
8 4 4 4 1 3 5
2 2 1 1 1 4 2 2
5 2 5 2 2 3 1 1
6 2 3 1 1 1 3 2

Table 3
Nair and Narendran [12] solution with rearranged machines

Parts

2 8 9 11 13 14 16 17 19 3 4 6 7 18 20 15 1 5 10 12

Machines
1 1 1 1 3 1 1 1 3 1 2
3 2 2 3 2 2 3 2 1 2
2 2 1 1 1 4 2 2
4 5 2 2 2 1 1 2
7 1 3 3 3 3 4 4 2
8 4 4 4 1 3 5
6 2 3 2 1 1 1 3
5 2 5 1 2 2 3 1

of the solution. The proposed flow measures addresses all machines as follows:
these requirements.
Total number of operations,
We define the following notations with respect to the X
CMS cell and layout design problem and develop relevant N ops ¼ max sjk . ð1Þ
j
performance measures: k

Therefore; total number of flows,


Notations: N flow ¼ N ops  n. ð2Þ

P ¼ 1,2,3, y, n represent the parts in the system Total number of flows within a cell c,
M ¼ 1,2,3, y, m represent the machines in the system N tc ¼ ðnc mc Þ  vc  nc . ð3Þ
S ¼ [sjk] represent the machine-component incidence ma-
trix using sequence data (Sjk is the sequence of the
3.1. Cell flow index (CFI)
kth part in the jth machine, 0 if the part does not
require the machine, 40 and an ordinal number
The CFI for cell c, CFIc is defined as the ratio of the
indicating the position in the overall sequence)
number of consecutive forward flows to the total number
nc number of parts in cell c
of flows within a cell. Cell average flow index (ACFI) is the
mc number of machines in cell c
weighted average of CFIs weighted by the number of parts
vc number of voids in cell c
in the cells. The expressions for CFI and ACFI are given
Nfc number of consecutive forward flows within
below:
cell c
Number of consecutive forward flows within the cell
CFIc ¼
Number of flows within the cell
We first calculate the total number of operations to be N fc
¼ . ð4Þ
performed and the total number of flows between a pair of N tc
ARTICLE IN PRESS
492 I. Mahdavi, B. Mahadevan / Robotics and Computer-Integrated Manufacturing 24 (2008) 488–497

P
nc CFIc 3.3. Flow matrix
c
Average Cell Flow Index; ACFI ¼ . (5)
n Since consecutive forward flows is a good indicator of
the goodness of the solution, we develop a flow matrix on
3.2. Overall flow index (OFI) the basis of the number of consecutive forward flows
between a pair of machines and use it as the basic input to
Another measure of interest is the OFI. OFI is defined as the grouping and layout problem. We define the flow
the ratio of the number of consecutive forward flows in all matrix as follows:
the cells to the total number of flows required to process all Let us consider a pair of machines, machine i and
the parts in the system: machine j. If a part k visits machine i and machine j in
immediate succession, then there is one unit of consecutive
Sum of consecutive forward flows in all the cell forward flow between the pair machines. Therefore, by
OFI ¼
Total number of flows in the system assessing the nature of flow of all the parts visiting a pair of
P
N fc machines, one can estimate the extent of consecutive
¼ c . ð6Þ forward flows between the pair of machines in the
N flow production system. We define this measure as the flow
From an examination of the CFI measure, we find that a coefficient fij between machines i and j. The expression for
high value of CFI is possible when either the number of fij is obtained as follows: we first define Xijk, the unit flow
voids in the cell is minimum or the number of consecutive coefficient for a part k between machines i and j as
forward flows is higher. Therefore, CFI indicates the
X ijk ¼ 0 if S ik ¼ 0;
quality of the solution with respect to the intra-cell
movements as well as the number of voids in the cell. On X ijk ¼ 0 if Sjk ¼ 0;
the other hand, OFI indicates the extent of inter-cell X ijk ¼ 1 if Sjk  S ik ¼ 1;
(7)
movement required by the parts to complete processing. X ijk ¼ 0 otherwise;
This is obvious because Nflow is the summation of the total P
f ij ¼ X ijk :
number of flows within cell (N tc ) and exceptions (inter-cell k
moves). Since N tc and inter-cell moves are complementary
to each other, for a given value of the number of By a pair-wise comparison of all the machines in the
consecutive forward flows (Nfc), a higher value of OFI system in this fashion, one can arrive at the flow matrix
could be achieved only by having fewer inter-cell F ¼ [fij]. The flow matrix F is an m  m asymmetric matrix.
movements. It is clear from the above that while CFI
addresses the issue of intra-cell issues OFI focuses on 4. CLASS algorithm
inter-cell issues.
At outset, it may appear that OFI is sufficient to assess The proposed algorithm aims at simultaneously provid-
the goodness of the solution. However, it can be shown ing solutions pertaining to the number of cells and their
that the measures together provide the complete picture respective membership of parts and machines as well as the
about the layout and CMS design solution. A higher value layout (sequence) of machines in each cell. This is a
of OFI, by itself does not ensure a high value of CFI. It construction algorithm that identifies the maximum value
merely indicates that the proposed solution has fewer inter- of fij for selecting the next machine to include in the cells.
cell moves. It is possible that the machines are not arranged The algorithm begins by selecting a pair of machines that
in proper sequence, i.e. layout issues are not addressed with could be placed side by side on account of high value of fij.
respect to each cell. In such a case, while the solution may After the machine pair is identified, partial solution is
have a high OFI, it may result in relatively lower CFIs. We examined to find if one of them is already part of the
demonstrate this by computing these measures for solu- partial solution. In such a situation, the improved solution
tions arrived in earlier studies to CMS problems using is obtained by inserting the other machine into that cell
sequence data that did not explicitly address the layout appropriately. If the pair is not found in any of the cells
issues within the cells in the section on comparative already obtained a new cell is formed. The procedure stops
analysis of our algorithm with others. when all the machines are assigned to the cells. The
By assessing the solution on the basis of consecutive corresponding part families are identified. The final
forward flows, we take care of the layout design and solution is not only the group membership of the machines
adjacency requirements. Furthermore, the definitions and in each cell, but also the order in which the machines have
the above discussions for the flow measures show that the to be placed to maximize consecutive forward flows. The
two flow measures together address the traditional steps of the algorithm are outlined below:
requirements of minimization of exceptions and voids for Step 1: Initialization
a CMS design. In this manner, we have provided a
quantitative measure for assessing both the layout and the Read S ¼ [sjk].
CMS designs of the proposed algorithm. Compute F ¼ [fij] using Eq. (7).
ARTICLE IN PRESS
I. Mahdavi, B. Mahadevan / Robotics and Computer-Integrated Manufacturing 24 (2008) 488–497 493

Number of cells formed, C ¼ 0. Step 7: New cell formation


Number of machines assigned, nm ¼ 0.
Let us define Lc as the list of machines assigned to C ¼ C+1,
cell c in the algorithm as a partial solution to the LC ¼ [m1,m2],
problem. nm ¼ nm+2,
f m1j ¼ 0 8j ¼ 1; 2; 3; . . . ; m; f im2 ¼ 0 8i ¼ 1; 2; 3; . . . ; m;
Step 2: Stopping criterion for the algorithm fm2,m1 ¼ 0.
Go to step 2.
If nm ¼ m, go to step 9.
Step 8: Reminder cell formation
Step 3: Identify the candidate machine pair for cell
formation Let Lf denote the list of unassigned machines in the
shop.
Select the largest value, f max ¼ max F ¼ ½f ij . Break ties C ¼ C+1,
i;j
arbitrarily. LC ¼ Lf,
Let m1 ¼ row index and m2 ¼ column index of fmax. nm ¼ m.
If fmax ¼ 0, go to step 8. Go to step 9.
Else go to step 4.
Step 9: Part family identification
Step 4: Addition of machines to cells
Examine each part with respect to the cells formed and
If m1 2 Lc , for any c ¼ 1,2, y, C, assign parts to the cells on the basis of the following
Let Lm1 denote the cell to which m1 belongs. rules
Lm1 ¼ [m11,m12y, m1n,m1]. (a) Assign a part to a cell, where it requires maximum
If m2 2 Lc , for any c ¼ 1,2, y, C, number of machines for its operations.
Let Lm2 denote the cell to which m2 belongs. (b) If ties exist, resolve the ties by assigning the part to the
Lm2 ¼ [m2, m21,m22y, m2n]. cell where its first operation will be performed.
Examine if Lm1 and/or Lm2 exists. Print the final solution and stop the procedure.
If one of them exists go to step 5.
If both exist, go to step 6. 5. Computational analysis of CLASS
If none exist, then go to step 7.
5.1. Numerical illustration of CLASS
Step 5: Insertion of a machine to existing cells
We first begin with a step-by-step numerical illustration
If m1 2 Lm1 , of CLASS. We use the 20 components, 8 machines problem
Update Lm1 as Lm1 ¼ [m11,m12ym1n,m1,m2], considered by Nair and Narendran [12] for numerical
fm2,m11 ¼ 0, fm2,m12 ¼ 0, y, fm2,m1n ¼ 0; fm2,m1 ¼ 0. illustration of CLASS algorithm. Table 1 has the machine–
If m2 2 Lm2 , component incidence matrix with sequence data, S ¼ [sjk].
Update Lm2 as Lm2 ¼ [m1,m2,m21,m22ym2n], Using Eq. (7) we compute the flow matrix for the eight
fm2,m1 ¼ 0; fm21,m1 ¼ 0; fm22,m1 ¼ 0;y, fm2n,m1 ¼ 0. machines F ¼ [fij]. Table 4 has the flow matrix for the
nm ¼ nm+1, problem. With this data we demonstrate the application of
f m1j ¼ 0 8j ¼ 1; 2; 3; . . . ; m; f im2 ¼ 0 8i ¼ 1; 2; 3; . . . ; m. CLASS.
Go to step 2. Step 1: We read S ¼ [sjk] and compute F ¼ [fij]. C ¼ 0,
nm ¼ 0; m ¼ 8.
Step 6: Merging of existing cells
Table 4
Flow matrix for Nair and Narendran [12]—8  20 matrix
6(a) Merge cells Lm1 and Lm2 by adding all machines in
Lm2 into Lm1. m1 m2 m3 m4 m5 m6 m7 m8
Update Lm1 ¼ [m11,m12ym1n,m1,m2, m21,m22ym2n].
m1 1 5 1 1
6(b) fm2,m11 ¼ 0, fm2,m12 ¼ 0, y, fm2,m1n ¼ 0; fm2,m1 ¼ 0.
m2 1 1 2 1 1
6(c) Repeat 6(b) for all machines to the right of m2 in m3 1 1
cell Lm2. m4 2 1 3
6(d) Delete cell Lm2 m5 1 1 1
C ¼ C1; m6 1 1 2 1
m7 1 1 1 4
f m1j ¼ 0 8j ¼ 1; 2; 3; . . . ; m;
m8 2 1 1
f im2 ¼ 0 8i ¼ 1; 2; 3; . . . ; m; fm2,m1 ¼ 0.
ARTICLE IN PRESS
494 I. Mahdavi, B. Mahadevan / Robotics and Computer-Integrated Manufacturing 24 (2008) 488–497

Step 2: As nmo8, we proceed to step 3 of the algorithm. Nair and Narendran [12]. We also notice that the CLASS
Step 3: From the flow matrix F, we identify the following: solution is the same as that we obtained by rearranging the
machines while improving the Nair and Narendran
fmax ¼ 5; m1 ¼ 1; m2 ¼ 3. solution (see Table 3). However, it must be noted that in
our solution the order in which the machines are specified
Step 4: Since no partial solutions are currently available, in the final solution is sacrosanct. It specifies the layout of
the procedure jumps to step 7. the machines of the cells in the shop floor. In order to show
Step 7: C ¼ 1, L1 ¼ [1,3], nm ¼ 2. We replace all f1j the significance of this and the usefulness of the algorithm,
values, fi3 values and f31 by 0. The procedure returns to step we provide a comparative study of our algorithm with
2 for a second iteration of the algorithm. CASE, proposed by Nair and Narendran for the above
During the second iteration, in a similar manner we problem. Table 5 has the details of the comparative study.
identify fmax ¼ 4; m1 ¼ 7; m2 ¼ 8 to form the second cell While the machine groups and part families are identical in
with the following details: C ¼ 2, L2 ¼ [7,8], nm ¼ 4. We both the CASE and the CLASS solution, the ACFI and
replace all the relevant fij values as before. However, the OFI have varied markedly. In the case of CLASS, the
third iteration has a machine insertion as shown below in a arrangement of machines in the right order has ensured
step-by-step fashion: that the CFIs for all the three cells are much higher
Step 2: As nmo8, we proceed to step 3 of the algorithm. compared to the CASE solution. This has resulted in much
Step 3: From the flow matrix F, we identify the following: higher values for both ACFI and OFI, indicating a more
appropriate solution for the combined cell formation and
fmax ¼ 3; m1 ¼ 4; m2 ¼ 7, L1 ¼ [1,3], L2 ¼ [7,8]. layout problem.
Ill-structured matrices pose challenges and provide an
alternative perspective on the performance of an algorithm
Step 4: On comparison of the partial solutions obtained
in cell formation studies. Solution methodologies some-
so far with m1 and m2, we get the following information:
times provide unrealistic solutions to ill-structured matrices
m2 2 L2 , therefore we have Lm2 ¼ 2 and the procedure
and limit the usefulness of the algorithm. Therefore, we
jumps to step 5.
have analyzed the 20  45 matrix provided in Nair and
Step 5: As per the logic of the procedure, we insert
Narendran [12] and compared it with their solution.
machine 4 in cell 2 to the left of machine 7 and update the
solution as follows: C ¼ 2, L1 ¼ [1,3], L2 ¼ [4,7,8], nm ¼ 5. Table 6 has the comparative performance of CASE and
CLASS. It is clear from the table that CLASS has
We replace all f4j values, fi7 values and f74 and f84 by 0. The
outperformed CASE in the case of an ill-structured matrix.
procedure returns to step 2 for a fourth iteration of the
The CASE solution has provided 8 cells with some
algorithm.
unrealistic solutions. For instance, the number of machines
Proceeding in this fashion, we insert machine 2 to the left of
in cell 1 is twice the number of the number of parts assigned
machine 4 in the next iteration and in the fifth iteration form a
to the cell. In the absence of information on production
third cell with machines 5 and 6. As all machines are assigned,
volume and process times, it may be difficult to make an
the procedure jumps to the last step of the algorithm.
Step 9: The final machine groups are as follows: assessment of the practical feasibility of such solutions.
However, in general it is understood that such solutions
may result in severe underutilization and inter-cell moves.
L1 ¼ [1,3], L2 ¼ [2,4,7,8] and L3 ¼ [6,5]. On examina-
On the other hand, the CLASS solution has fewer cells. A
tion of the part 1, we find that it requires machines 5 and
closer examination reveals that cells 6 and 1 of CASE
6. Therefore, we include it in group 3. Similarly, part 2
solution have been merged to form a single cell in the case
requires processing in machines 1 and 3. Therefore, we
of CLASS. Similar is the case with respect to the other
include it in group 1. In the case of part 3, we find that a
majority of its requirements are from the machines in
group 2. Therefore, we include the part in group 2. Table 5
Proceeding in this fashion, we get the final solution to Performance measure comparison between CASE and CLASS solutions
the problem as follows:
Cell No CASE solution to the CLASS solution to the
problem problem
Machines Parts
nc N fc N tc CFI (%) nc N fc N tc CFI (%)
Cell 1: [1,3] [2,8,9,11,13,14,16,17,19]
Cell 2: [2,4,7,8] [3,4,6,7,18,20] 1 9 1 9 11.1 9 5 9 55.6
Cell 3: [6,5] [1,5,10,12,15] 2 6 7 18 38.9 6 9 18 50.0
3 5 1 5 20.0 5 2 5 40.0
Sum 20 9 20 16
5.2. Comparison of CLASS with CASE
N flow ¼ 41
ACFI (%) 21 50.0
The machine groups obtained by our algorithm are the
OFI (%) 22.0 39.0
same as those obtained by CASE algorithm proposed by
ARTICLE IN PRESS
I. Mahdavi, B. Mahadevan / Robotics and Computer-Integrated Manufacturing 24 (2008) 488–497 495

Table 6
Nair and Narendran [12] 25  40 matrix—comparison of CASE and CLASS solutions

Cell No CASE solution to the problem CLASS solution to the problem

Machines in the cell Parts in the cell Machines in the cell Parts in the cell

1 14,13,15,22 18,32 9,8,10,18,7,4,16,1,2,17 1,2,5,7,10,15,16,17,19,21,22,28,30,36,38


2 18,4,7,16 1,5,7,16,17,30 24,20,3,11,25 3,9,12,13,14,33
3 19,5 8,15,23,24,31 21,6,5,19 8,11,23,24,25,29,31,35,40
4 20,3,11,25 3,9,13,14,33 23,12,22,15,14,13 4,6,18,20,26,27,32,34,37,39
5 21,6 11,25,27,29,35,40
6 23,12 4,6,20,26,34,37,39
7 1,2,17,24 2,12,36
8 9,8,10 10,19,21,22,28,38
ACFI 47% 52%
OFI 26% 34%

Table 7
A comparison of CLASS with other algorithms in the literature

Algorithms for comparison Criterion for Data sets for comparison


comparison
1 2 3 4

Tam [17]— Harhalakis et al. [5]— Nair and Narendran [12]— Nair and Narendran [12]—
12  19 20  20 25  40 8  20

CLASS No. of cells 2 4 4 3


ACFI 65% 65% 52% 50%
OFI 50% 41% 34% 39%
Park & Suresh (Fuzzy Art) No. of cells 2 4 7
ACFI 49% 42% 38%
OFI 36% 34% 27%
Park & Suresh (Best No. of cells 2 4 8
Heirarchical) ACFI 48% 42% 37%
OFI 45% 34% 22%

Nair & Narendran (CASE) No. of cells 8 3


ACFI 47% 22%
OFI 26% 22%
Others No. of cells 2a 4b
ACFI 47% 46%
OFI 45% 37%
a
Kiang et al. [11] solution to the data set.
b
Harhalakis et al. [5] solution to the data set.

cells. Due to these, CLASS solution has better ACFI and and Park and found to be representative enough for our
OFI compared to CASE. study [20]. Moreover, we have been able to obtain multiple
solutions for comparison for these data sets. Therefore, the
5.3. Comparative study of CLASS with other algorithms performance of CLASS could be sufficiently analyzed in
this context. While using the solutions arrived by the earlier
In order to understand the performance potential of studies, we have retained the ordering of the machines as
CLASS, we have made a comparison of the CLASS provided by them for the purpose of computing ACFI and
solution with several others reported in the literature. Park OFI.
and Suresh [20] made a detailed comparative study of the Table 7 has a comparative study of CLASS with the
performance of known algorithms with respect to the other solutions. The number of cells identified by CLASS
sequence data. The algorithms considered included fuzzy was the same as in the case of all other algorithms in all
ART neural network and other traditional hierarchical data sets tested, with the exception of the Nair and
clustering methods. The study utilized three data sets; Tam Narendran [12] 25  40 matrix. As we have already
[17] used by Kiang et al. [11], Harhalakis et al. [5] and Nair discussed, this matrix is ill structured and CLASS provides
and Narendran [12]. These data sets were used by Suresh a better solution. In all the cases, the OFI and ACFI of
ARTICLE IN PRESS
496 I. Mahdavi, B. Mahadevan / Robotics and Computer-Integrated Manufacturing 24 (2008) 488–497

CLASS are much higher than that of others. Clearly, the [4] Askin RG, Subramanian SP. A cost-based heuristic for group
study has established the need for focusing on the issue of technology configuration. Int J Prod Res 1987;25:101–13.
layout in addition to mere cell formation while using [5] Harhalakis G, Nagi R, Proth JM. An efficient heuristic in manu-
facturing cell formation for group technology applications. Int J Prod
sequence data. Res 1990;28:185–98.
[6] Vakharia AJ, Wemmerlov U. Designing a cellular manufacturing
6. Conclusions system: a materials flow approach based on operations sequences. IIE
Trans 1990;22:84–97.
Sequence data provides additional information to the [7] Ballakur A, Steudel HJ. A within-cell utilization based heuristic for
designing cellular manufacturing systems. Int J Prod Res 1987;25:
cell designer than what she requires for cell formation. This 639–65.
pertains to the dominant flow pattern in the identified cells, [8] Wei JC, Gaither N. An optimal model for cell formation decisions.
which could be a sound basis for solving the layout Decision Sci 1990;21:416–33.
problem in cellular manufacturing system design. Unfortu- [9] Shafer SM, Rogers DF. A goal programming approach to the cell
nately, cell designers using sequence data have not made formation problem. J Oper Manage 1991;10:28–43.
[10] Lee S-D, Chen Y-L. A weighted approach for cellular manufacturing
use of this information. This paper’s main contribution is
design: minimising inter-cell movement and balancing workload
to extend the application of sequence data for the layout among duplicate machines. Int J Prod Res 1997;35:1125–46.
problem also. This paper addresses the joint problem of the [11] Kiang MY, Kulkarni UR, Tam KY. Self-organizing map network as
cell formation and the layout. An algorithm, CLASS has an interactive clustering tool—an application to group technology.
been developed for cellular manufacturing system and Decision Support Syst 1995;15:351–74.
layout design using sequence data. Two measures of [12] Nair GJ, Narendran TT. CASE: a clustering algorithm for cell
formation with sequence data. Int J Prod Res 1998;36:157–79.
performance have been proposed for evaluating the good-
[13] Won Y, Lee KC. Group technology cell formation considering
ness of solution. operation sequences and production volumes. Int J Prod Res 2001;
A comparison of CLASS with other known algorithms 39:2755–68.
in the literature has been made using available sequence [14] Jayaswal S, Adil GK. Efficient algorithm for cell formation with
data based data sets. The computational analysis points to sequence data, machine replications and alternative process routings.
the usefulness of the approach and the promise that it has Int J Prod Res 2004;42:2419–33.
[15] Selvam RP, Balasubramanian KN. Algorithmic grouping of opera-
in jointly solving the cell design and the layout problem. tion sequences. Eng Costs Prod Econ 1985;9:125–34.
While the proposed logic has a good potential for [16] Choobineh F. A framework for the design of cellular manufacturing
application, it is too early to assess the potential of the systems. Int J Prod Res 1988;26:1161–72.
proposed algorithm. Several issues need to be addressed to [17] Tam KY. An operation sequence based similarity coefficient for part
improve the application potential of the proposed logic. family formations. J Manuf Syst 1988;9:55–68.
[18] Suresh NC, Slomp J, Kaparthi S. Sequence-dependent clustering of
These include understanding the behavior of the algorithm
parts and machines: a Fuzzy ART neural network approach. Int J
with respect to a variety of ill-structured matrices and its Prod Res 1999;37:2793–816.
effectiveness in solving large real life data. Another area of [19] Kang SL, Wemmerlöv U. A work load-oriented heuristic methodol-
research worth pursuing is to improve the algorithm. For ogy for manufacturing cell formation allowing reallocation of
instance, the construction algorithm could be supplemen- operations. Eur J Oper Res 1993;69:292–311.
ted with an improvement algorithm to examine second- [20] Park S, Suresh NC. Performance of Fuzzy ART neural network and
hierarchical clustering for part-machine grouping based on operation
order refinements to the solution. Moreover, instead of a sequences. Int J Prod Res 2003;41:3185–216.
pair-wise comparison of machines and selection based on [21] Irani SA, Cavalier TM, Cohen PH. Virtual manufacturing cells:
the maximum flow among them, it is worthwhile to devise exploiting layout design and intercell flows for the machine sharing
algorithms that consider more than two machines at a time. problem. Int J Prod Res 1993;31:791–810.
Use of some meta-heuristics could provide some additional [22] Liao TW. Design of line-type cellular manufacturing systems for
minimum operating and material-handling costs. Int J Prod Res
insights into the quality of the solution. Despite all these
1994;32:387–97.
improvement opportunities and scope for further research [23] Arvindh B, Irani SA. Principal component analysis for evaluating the
in the area, we believe that the basic surmise that using feasibility of cellular manufacturing without initial machine-part
more information from the sequence data will be beneficial clustering. Int J Prod Res 1994;32:1909–38.
for cellular manufacturing system design in the long run [24] Akturk MS, Balkose HO. Part-machine grouping using a multi-
will hold good. We motivate further research in this objective cluster analysis. Int J Prod Res 1996;34:2299–315.
[25] Heragu SS, Kakuturi SR. Grouping and placement of machine cells.
unexplored area.
IIE Trans 1997;29:561–71.
[26] Lee S-D, Chiang C-P. A cut-tree-based approach for clustering
References machine cells in the bidirectional linear flow layout. Int J Prod Res
2001;39:3491–512.
[1] Sarker BR, Xu Y. Operation sequences-based cell formation [27] Chiang C-P, Lee S-D. A genetic-based algorithm with the optimal
methods: a critical survey. Prod Plan Control 1998;9:771–83. partition approach for the cell formation in bi-directional linear flow
[2] Singh N. Design of cellular manufacturing systems: an invited review. layout. Comput Integr Manuf 2004;17:346–75.
Eur J Oper Res 1993;69:284–91. [28] Kusiak A, Heragu SS. The facility layout problem. Eur J Oper Res
[3] Wemmerlov U, Hyer NL. Procedures for the part-family/machine 1987;29:229–51.
group identification problem in cellular manufacturing. J Oper [29] Heragu SS. Recent models and techniques for solving the layout
Manage 1986;6(2):125–47. problem. Eur J Oper Res 1992;57:136–44.
ARTICLE IN PRESS
I. Mahdavi, B. Mahadevan / Robotics and Computer-Integrated Manufacturing 24 (2008) 488–497 497

[30] Houshyar A, McGinnis LF. A heuristic for assigning facilities to [33] Yin Y, Yasuda K. Reconsidering generalized similarity coefficient via
locations to minimize WIP travel distance in a linear facility. Int a sequence ratio. Int J Ind Eng—Theory Appl Practice 2004;11:
J Prod Res 1990;28:1485–98. 140–50.
[31] Afentakis P. A loop layout design problem for flexible manufacturing [34] Yin Y, Yasuda K, Hu L. Formation of manufacturing cells based on
systems. Int J Flexible Manuf Syst 1989;1:175–96. material flows. Int J Adv Manuf Technol 2005;27:159–65.
[32] Kim CO, Baek JG, Baek JK. A two-phase heuristic algorithm for cell [35] Xambre AR, Vilarinho PM. A simulated annealing approach for
formation problems considering alternative part routes and machine manufacturing cell formation with multiple identical machines. Eur J
sequence. Int J Prod Res 2004;42:3911–27. Oper Res 2003;151:434–46.

You might also like