Вы находитесь на странице: 1из 19

April 24, 2006 17:5 WSPC/180-JAMS 00071

Journal of Advanced Manufacturing Systems


Vol. 5, No. 1 (2006) 27–44
c World Scientific Publishing Company

A GENETIC ALGORITHM APPROACH FOR MACHINE


CELL FORMATION

RAVISHANKAR RAJAGOPALAN∗ and DANIEL J. FONSECA†


Department of Industrial Engineering, The University of Alabama
Tuscaloosa, AL 35487 — 0288, US
∗rajagopalan.18@osu.edu
†dfonseca@coe.eng.ua.edu

Cellular Manufacturing Systems (CMS) have been the key to the success of manufac-
turing industries in the recent past. Machine cell design, which involves formation of
machine cells and component groups, represents the most important step in the design
of CMS. Even though a tremendous amount of research has been conducted in this
area, the gap between theoretical research and practice is widening primarily because
of the lack of consideration of key production data such as production volumes, oper-
ations sequences, machine sequence inside the cells, processing times, setup times, and
machine costs during the cell design stage. This paper discuses the development of a
Genetic Algorithm Model (GAM), designed to assist in the formation of manufacturing
cells. The GAM aims at the minimization of the material handling and the penalty costs
while considering the effects of inter-cell, intra-cell, backtracking, and machine skipping
movements.

Keywords: Genetic algorithms; cellular manufacturing; cell formation.

1. Introduction
Increased competition and fluctuating market demands have driven many manu-
facturing firms to consider novel approaches to improve productivity and eliminate
waste. In the past two decades, manufacturing industries have undergone a revo-
lution, widely considered as the third industrial revolution.1 Many innovative con-
cepts have surfaced, and only a few among them has been successful. The concept
of Group Technology (GT) is one of such successful principles embraced by most
industries.
Group Technology (GT) is the exploitation of the similarities among processes
and component designs in such a way that it increases the utilization of resources,
and eliminates/reduces nonvalue added activities, i.e., material handling, scraps,
downtime, etc. GT exploits similarities in three different ways: (1) by performing
alike activities together, (2) by standardizing similar tasks, and (3) by efficiently
storing and retrieving information about recurring problems.2 GT forms the basis

27
April 24, 2006 17:5 WSPC/180-JAMS 00071

28 R. Rajagopalan & D. J. Fonseca

of Cellular Manufacturing Systems (CMS). Cellular manufacturing is a practice


that involves grouping similar machines into cells, simultaneously grouping sim-
ilar components into groups. The CMS offer a great deal of benefits including
reduction in material handling times, setup times, batch sizes, and throughput
times.3

2. Drawbacks of Current Cell Formation Techniques


The formation of machine cells and component groups has been the main focus of
most of the research performed in the field of CMS. The complexity, in terms of
the size of the problem, makes it difficult for researchers to present a comprehen-
sive method that simulates the exact manufacturing system. The initial methods
that attempted to solve the problem of machine and component grouping failed
to consider any practical constraints.4–7 Most of these methods work with the 0-1
incidence matrix, in which the entries are either 0 or 1 depending on whether a
part is processed in a machine or not, respectively. In these methods, the incidence
matrix is rearranged to obtain a diagonal structure that facilitates the identifica-
tion of machine cells and their corresponding component groups. Another group
of solution techniques known as the similarity coefficient methods involves com-
puting the value of similarity coefficients between all the machines or components
and grouping the ones with the highest similarity.3,4 This process is repeated until
all the machines or components are grouped into cells. These methods, though
quick in obtaining machine cells and component groups, do not consider sev-
eral important production data that have significant effect on the cell formation
process.
Since the late eighties, researchers have included several production data such as
production volume, processing sequence, processing times, alternate process plans,
and capacity of machines etc., during the cell design stage. References 8–14 are the
few of the works that have taken into consideration the effects of some of the previ-
ously mentioned production data during machine cell formation. These production
data have been found to impact the machine cells and component groups formed.
The last decade has seen a steep rise in the number of publications considering
various production data in arriving at a practical solution to the cell formation
problem. Many of the techniques proposed by these researchers have the objective
of minimizing the material handling costs while forming the machine cells and com-
ponent groups. The composition of material handling costs affects the cells designed.
The early researchers15,16 considered the effect of inter-cell material handling alone
and neglected the effects of intra-cell material handling. Even though, the magni-
tude of the intra-cell material handling is less when compared to inter-cell material
handling, the frequency of intra-cell material handling is very high when com-
pared to inter-cell material handling. Hence, considering the intra-cell movements
in the material handling costs calculation makes a significant impact on the machine
cells formed. Several researchers have attempted to capture this intra-cell material
April 24, 2006 17:5 WSPC/180-JAMS 00071

A Genetic Algorithm Approach for Machine Cell Formation 29

handling in their material handling cost calculation.9,17 However, many of them


have failed to capture precisely the different components of intra-cell material
handling namely the intra-cell, backtracking, and machine skipping movements.
Reference 12 made an attempt to capture these factors in their research. However,
the nonconsideration of machine sequence inside the cells, another factor repeatedly
neglected by the researchers, makes the material handling costs calculated unreal-
istic. This machine sequence inside the cells has a significant impact on the three
components of the intra-cell material handling cost, and hence, on the cell design
process.
Considering these factors simultaneously in the cell formation stage will cer-
tainly affect the machine cells and component groups formed. Hence, there is a
need to develop a one step method that will address all these issues effectively in a
timely manner. The Genetic Algorithm Model (GAM) discussed here is a one-step
model that captures all these factors effectively and provides the optimal/near-
optimal machine cells, component groups and machine sequences inside the cells.

3. The Genetic Algorithm Model


The main goal of the study was to develop a Genetic Algorithm (GA) based model
(GAM) to tackle the machine cell formation problem in CMS, considering various
production data such as production volume of components, processing sequences,
machine sequence inside the cell, processing times, setup times, and machines
costs. The computer code developed in Visual C++ enables the determination
of optimal/near-optimal solutions in a timely manner.

3.1. Assumptions behind the GAM


The GAM has been designed based on the following assumptions:
1. The number of cells is predefined.
2. There is only one machine of each type, i.e., no duplicate machines are allowed.
3. Each component visits a machine only once during its processing.
4. Every component has a fixed operation sequence.
5. The layout of the cells is assumed to be linear and has a single row.
6. The transfer batch size is one.
7. Each machine can belong to one and only one cell.

3.2. Notation used in the GAM


Following is the notation used in the development of the model:
1. IRC: Total inter-cell cost ($).
2. IAC: Total intra-cell cost ($).
3. BC: Total backtracking cost ($).
April 24, 2006 17:5 WSPC/180-JAMS 00071

30 R. Rajagopalan & D. J. Fonseca

4. MSC: Machine skipping cost ($).


5. PC: Total penalty cost ($).
6. ST: Setup time per batch (min).
7. MR: Machine rate per hour ($/h).
8. PT: Processing time per unit (min).
9. PV: Production volume (units).

3.3. The model


The GAM identifies the machine cells, machine sequences inside each cells, and
component groups. The objective of the GAM is to minimize the sum of material
handling and penalty costs. The material handling cost encapsulates the various
material handling movements namely the inter-cell, intra-cell, backtracking, and
machine skipping. The penalty cost consists of the setup and the processing costs
and accounts for the formation of component groups. Its impact on machine cells
and component groups formation is explained in detail in the subsequent section.

3.4. Penalty cost


The total material handling cost controls the formation of machine cells, but has
no influence on the component group formation. For a specific machine cell formed,
there might be a number of different component groups. For instance, consider a
component that requires only two operations. If these two operations are performed
in two different cells, the only material handling cost associated with it is the inter-
cell cost. All other costs namely intra-cell, backtracking, and machine skipping are
zero. However, assigning the components to either of these cells would not influence
the calculated inter-cell material handling cost. Hence, the same material handling
cost will be arrived at irrespective of whether the component is assigned to one
cell or another. To overcome this problem of multi-solution component grouping,
a penalty cost is associated with each component that visits a cell other than the
one to which it is assigned. Such a cost factor is required to effectively account
for the grouping of components. The penalty cost is the sum of the setup and the
processing costs. In this regard, each cell is considered as a separate cost center.
Hence, any processing in the cell to which the part is assigned will incur in no extra
cost. However, if the part is assigned to another cell, the setup and the processing
costs are treated as extra costs. This in turn restricts most of the operations on a
part to be assigned to the same cell, thus reducing the inter-cell movement. The
penalty cost is calculated as given in Eq. (1).
PC = (ST × MR/60) + (PT × MR × PV/60). (3.1)
It thus can be said that the objective function of the study involves obtaining
the machine cells, component groups, and machine sequences in such a way that
the overall cost, which is the sum of the material movement cost and the penalty
cost for all the components, is minimized.
April 24, 2006 17:5 WSPC/180-JAMS 00071

A Genetic Algorithm Approach for Machine Cell Formation 31

3.5. Overall cost calculation heuristic


The overall cost calculation is the most important step in the GAM since it captures
all relevant production data such as the production volume, the processing sequence,
the processing time, the setup time, and the machine cost per hour. The heuristic
calculates the inter-cell, intra-cell, backtracking, and the penalty costs separately.
These costs are added together to arrive at the overall cost. The calculated overall
cost forms the fitness function for the genetic algorithm which guides the search
toward the optimal solution.
The developed heuristic shown in Fig. 1 is illustrated with a numerical example
as follows. Consider a two cell problem with five machines and six components.
Figure 2 shows the sequence matrix, the processing time matrix, production vol-
umes, machine setup times, and machine costs for each machine. Apart from the
above inputs, the following costs are also assumed:

Inter-cell cost = PV × Inter-cell cost per unit (3.2)


Intra-cell cost = PV × MSC × Intra-cell cost per unit (3.3)
Backtracking cost = PV × MSC × Backtracking cost per unit. (3.4)

Figure 3 shows a feasible solution generated for the problem under consideration.
The calculation of the overall cost using the heuristic is discussed as follows.
Component 1 has a production volume (PV) of 100, and its sequence of opera-
tions is 1-2-3. As per the solution shown in Fig. 3, component 1 belongs to cell 1. For
this component, there is no inter-cell cost since all the machines required to process
it are located in cell 1 itself. There is an intra-cell movement associated with the
movement from machine 1 to 2 that involves a machine skipping. A backtracking
cost, without machine skipping, is associated with this component as it moves from
machine 2 to 3. This move is considered as backtracking because the direction of
part movement (machine 2–3) is opposite to that of the direction in which these
two machines are located (machine 3 followed by machine 2). The penalty cost is
zero since all the machines required for component 1 are located in cell 1, the cell
to which component 1 is assigned. Therefore:

Inter-cell cost = PV × Inter-cell cost per unit = 0


Intra-cell cost [1-2] = PV × MSC × Intra-cell cost per unit
= 100 × 2 × 1 = 200
Backtracking cost [2-3] = PV × MSC × Backtracking cost per unit
= 100 × 1 × 2 = 200
Setup cost = (ST × MR)/60 = 0
Processing cost = (PT × PV × MR)/60 = 0
Penalty cost = Setup cost + Processing cost = 0.
For Each Component Repeat These Steps; 32
PC=0; IRC=0;IAC=0; BC=0

compcellno=cell number to which component belongs;


operation no=1

machno=machine in which operation is carried out


machcellno=cell to which this machine belongs

YES NO
operation no
=1

YES machcellno=co YES machcellno= NO


April 24, 2006 17:5 WSPC/180-JAMS

mpcellno
R. Rajagopalan & D. J. Fonseca

compcellno
operation no= operation no+1
NO NO machcellno(current)= YES
machcellno(previous)
PC=PC+(ST*MR/60)+(PT*MR*PV/60)
00071

PC=PC+(ST*MR/60)+(PT*MR*PV/60)
Current Machine Postion
operation no= operation no+1 IRC=IRC+(PV*Intercell_cost)
> Previous Machine
YES Position NO
NO YES
machcellno(current)= operation no= operation no+1
PC=PC+(ST*MR/60)+(PT*MR*PV/60) PC=PC+(ST*MR/60)+(PT*MR*PV/60)
machcellno(previous)
IAC=IAC+(Current Position-Previous BC=BC+(Previous Position-Current
IRC=IRC+(PV*Intercell_cost) Position)*Intra_cell_cost*PV Position)*Backtrack_cost*PV
Step-3
Current Machine Postion
operation no= operation no+1 operation no= operation no+1
operation no= operation no+1 > Previous Machine
YES Position NO

Step-3 IAC=IAC+(Current Position-Previous BC=BC+(Previous Position-Current Step-3 Step-3


Position)*Intra_cell_cost*PV Position)*Backtrack_cost*PV

operation no= operation no+1


operation no= operation no+1

Step-3
Step-3

Fig. 1. Cost calculation heuristic.


April 24, 2006 17:5 WSPC/180-JAMS 00071

A Genetic Algorithm Approach for Machine Cell Formation 33

5 Machine, 6 Component Problem


Sequence Matrix

Machines
1 2 3 4 5

1 1 2 3 0 0

2 1 2 0 3 0

3 2 1 0 0 0

4 0 0 0 1 2

5 0 0 1 2 3

6 0 0 0 2 1

Production Volume = {100, 300, 200, 250, 150, 175}

Processing Time Matrix (Minutes)


Machines
1 2 3 4 5

1 5 4 3 0 0

2 6 2 0 3 0

3 4 5 0 0 0
Components
4 0 0 0 4 6

5 0 0 4 3 2

6 0 0 0 3 4

Setup Time in Each Machine (Minutes) = {120, 75, 100, 60, 100}

Machine Cost for Each Machine ($/Hour) = {25, 35, 50, 20, 40}

Fig. 2. Overall cost calculation example.

5 Machine – 6 Component Problem


2 Cell Solution
Cells Machines in Order Components
1 1, 3, 2 1, 2, 3
2 4, 5 4, 5, 6

1 3 2 4 5
Cell 1 Cell 2

Fig. 3. Solution to problem in Fig. 4.


April 24, 2006 17:5 WSPC/180-JAMS 00071

34 R. Rajagopalan & D. J. Fonseca

Machines 1 2 3 4 5 6

1 2 1 1 2 1
Cells

Fig. 4. Sample chromosome for machine cells.

Similarly, component 2 has a production volume of 300, and follows a processing


sequencing of 1-2-4. From Fig. 4, machines 1 and 2 are located in cell 1, whereas
machine 4 is located in cell 2. Moreover, component 2 belongs to cell 1. This compo-
nent incurs in an inter-cell material handling while being moved from machine 2 to
machine 4. An intra-cell cost, along with a machine skipping factor of 2, is incurred
when the component is transferred from machine 1 to 2. However, there are no
backtracking costs associated with this part’s movement. Moreover, component 2
belongs to cell 1, but machine 4 is located in cell 2; hence, its processing in machine
4 implies a penalty cost. In such a case, the setup and processing costs in machine
4 are considered as penalty costs. The Setup (ST) and Processing Times (PT) for
component 2 in machine 4 are 60 and 3 min, respectively, thus:

Inter-cell cost [2-4] = 300 × 5 = 1500


Intra-cell cost [1-2] = 300 × 2 × 1 = 600
Backtracking cost = 0
Setup cost = (60 × 20)/60 = 20
Processing cost = (3 × 300 × 20)/60 = 300
Penalty cost = 300 + 20 = 320.

In a similar fashion, the different cost elements can be calculated for all the
components as shown in Table 1. The overall cost incurred by all the components
is $5628.33. This cost is treated as the fitness function for the GAM, and this
effectively guides the GAM toward the optimal/near-optimal solution.

Table 1. Overall cost calculation summary.

Comp No. Material Handling Costs ($) Penalty Costs ($) Overall Cost ($)
Inter Intra Back Total Setup Process Total
1 0 200 200 400 0 0 0 400
2 1500 600 0 2100 20 300 320 2420
3 0 0 800 800 0 0 0 800
4 0 250 0 250 0 0 0 250
5 750 150 0 900 8.33 500 508.33 1408.33
6 0 0 350 350 0 0 0 350
4800 828.33 5628.33
April 24, 2006 17:5 WSPC/180-JAMS 00071

A Genetic Algorithm Approach for Machine Cell Formation 35

4. Genetic Algorithm Design for the GAM


4.1. Coding scheme for the GAM
In the GAM, the objective is to find optimal machine cells, machine sequences inside
each cell, and component groups. The coding is designed to properly capture these
three decision variables. A three-set coding scheme is used in this model. One set
to code the machine cells, one to code the machine sequences inside the cells, and
the other to code the component groups. A set of real numbers, strictly integers,
are used to code the sets.
Figure 5 shows a sample chromosome that represents the machine cells. In this
chromosome, the position of each bit represents the machines and its value cor-
responds to the cell number to which the machine belongs. The total number of
positions is equal to the number of machines available. The shaded chromosome
shown in Fig. 5 corresponds to a 6-machine, 2-cell scenario with machines 1, 3, 4,
and 6 placed in cell 1, and machines 2 and 5 placed in cell 2. This set is formed
randomly by generating values between 1 and the number of cells. A subroutine is
used to check and remove empty cells, if present.
The second coding corresponds to the machine sequence inside the cells. Figure 5
shows a sample chromosome representing the machine sequences. The number of
bits corresponds to the number of machines and the value of these digits pro-
vides the sequence of machines inside each cells. However, the interpretation of
this chromosome requires knowledge on the machine cell chromosomes. Figure 6
illustrates the decoding of the machine sequences chromosome in conjunction with
the machine cells chromosome.
The decoding (interpretation) of the machine sequence chromosome is done in
two steps. In the first step, the machines are rearranged according to the cells they
belong to, as well as their machine sequence values. In Fig. 6, machines 1, 3, 4,
and 6 which belong to cell 1, are grouped together, as well as their corresponding
machine sequence values 1, 5, 2, and 6. Afterward, each cell has one set of machines
and sequences associated with it. In stage two, the machine sequence values in a
cell are arranged in ascending order, and thus, the machines are too. This machine
arrangement, interpreted in the ascending order, gives machine sequence inside each
cell. In Fig. 6, machine sequence numbers 1, 5, 2, and 6 in cell 1 are arranged in
ascending order as 1, 2, 5 and 6. Hence, machines are rearranged as 1, 4, 3 and 6,

Machines 1 2 3 4 5 6

Machine
1 4 5 2 3 6
Sequence

Fig. 5. Sample chromosome for machine sequences.


April 24, 2006 17:5 WSPC/180-JAMS 00071

36 R. Rajagopalan & D. J. Fonseca

Machines 1 2 3 4 5 6

Cells 1 2 1 1 2 1

Machine
Sequence 1 4 5 2 3 6

Machines (Stage 1) 1 3 4 6 2 5
Cells 1 1 1 1 2 2
Grouped

Corresponding
1 5 2 6 4 3
Sequence

Machines (Stage 2) 1 4 3 6 5 2
Cells
1 1 1 1 2 2
Grouped

Sequence in
Ascending
Order 1 2 5 6 3 4

Final Machine
Sequence 1 4 3 6 5 2

Cell 1 Cell 2

Fig. 6. Decoding of machine sequence chromosome.

which corresponds to the sequence of machines inside the cell 1. Similarly, the
sequence of machines inside cell 2 is 5 and 2.
The coding scheme used for component grouping is very similar to the one
defined for the machine cells. The only difference being that the total num-
ber of positions is equal to the number of components and not the number of
machines.
April 24, 2006 17:5 WSPC/180-JAMS 00071

A Genetic Algorithm Approach for Machine Cell Formation 37

4.2. Fitness function design for the GAM


The GAM’s main objective is to minimize the overall cost, which is the sum of
the material handling and penalty costs as defined previously. The total material
handling cost consists of the inter-cell cost, the intra-cell cost, and the backtracking
cost. The penalty cost is made out of the setup and processing costs. The fitness
function in the GAM corresponds to the overall cost. The heuristic designed (Fig. 1)
is used to calculate the fitness function. The fitness function values are computed for
every member of the population, and these values are used to perform the selection
operation.

4.3. Selection of mating chromosomes for the GAM


A selection operator exploits the information available about the fitness of each
chromosome, and it is responsible for propagating good chromosomes thorough the
subsequent generations. The most widely used selection operators are the roulette-
wheel selection and the tournament selection.18 Roulette-wheel selection is best
suited for maximization problems. Tournament selection is easier to apply and
more efficient than roulette-wheel selection for minimization problems. Since the
problem at hand involves minimization of the overall cost, tournament selection
was chosen.
In tournament selection, four chromosomes are chosen at random from the
machine cell population. Each of these four chromosomes has an overall cost
associated with it. A tournament is conducted among the four chromosomes to
detect the one with the least overall cost. The chromosome corresponding to the
minimum overall cost is copied to the subsequent generation.18 This process is
repeated as many times as the population size to obtain a set of machine cells. Sim-
ilar tournament selection is conducted to obtain a population of machine sequences,
and component groups.

4.4. Crossover operator design for the GAM


Genetic Algorithms are mainly driven by the crossover operator. The crossover
operator is responsible for effective exploration of the solution space, and hence, it
is applied with a high probability. Crossover is complicated in the case of the GAM
since three crossover operators need to be applied: one for the machine cell pop-
ulation, one for the machine sequence population, and another for the component
group population.
A single point crossover was designed for the machine cell population.18 In
it, two chromosomes are selected at random from the machine cell population. A
crossover position, between 1 and the number of machines, is randomly selected,
and applied to the two chromosomes. The tails, length of the chromosome after
the crossover site, of the two chromosomes are swapped. While using this operator,
care needs to be taken to avoid the creation of empty cells. In the GAM, a checking
April 24, 2006 17:5 WSPC/180-JAMS 00071

38 R. Rajagopalan & D. J. Fonseca

procedure is implemented to check for empty cells every time the crossover operation
is implemented. If an empty cell is detected, the chromosome is removed, and
the crossover operator is applied again to generate a valid chromosome. The same
crossover operator with a similar checking procedure is applied for the component
population.
For the machine sequence population, a crossover operator called Partially
Matched Crossover (PMX) is applied. This operator ensures that the continuity
of number sequence is maintained. In this PMX, two chromosomes are randomly
selected for crossover, and two crossover sites are picked at random along the length
of the strings. A matching section is formed between the two crossover sites, and a
position-by-position exchange is executed to perform the crossover operation along
the matching section. PMX crossover is illustrated in Fig. 7. In this position wise
exchange, 5 from chromosome 1 is exchanged with 2 of chromosome 2. To avoid
repetition of sequence numbers, 5 also replaces 2 in chromosome 1. In this way,
PMX crossover is carried out between each position in the matching zone. The
probability of applying these crossover operators dictates the effectiveness of the
GA under consideration. A crossover probability of 0.8 was found to be effective in
most of the problems.

4.5. Mutation operator design for the GAM


A mutation operator, even though applied sparingly in a GA, induces fresh genetic
material, and hence, aids in the effective exploration of the solution space.18 As in
the case of crossover, three mutation operators were designed in the GAM: a muta-
tion operator for the machine cell population, a mutation operator for the machine
sequence population, and a mutation operator for the component group popula-
tion. In the GAM, a simple exchange mutation operator is applied in all the three
cases. This operator ensures that only valid offspring are generated. In the simple
exchange mutation, two bits of a chromosome are chosen randomly, and the posi-
tions of the bits are exchanged. The same mutation operator is applied for machine
sequence and component group populations. While applying this to the machine
sequence population, care needs to be taken to ensure that corresponding machines

Before Crossover

9 8 4 5 6 7 1 3 2 10
8 7 1 2 3 10 9 5 4 6

After Crossover

9 8 4 2 3 10 1 6 5 7

8 10 1 5 6 7 9 2 4 3

Fig. 7. PMX crossover for the machine sequence population.


April 24, 2006 17:5 WSPC/180-JAMS 00071

A Genetic Algorithm Approach for Machine Cell Formation 39

are also exchanged along with their sequences. The probability of mutation usually
is kept very low, and in the GAM, a probability of mutation of 0.001 was found to
be effective.

4.6. Termination condition for the GAM


The termination condition for a GA is usually specified as a pre-established number
of generations. If the number of generations is less, the solution obtained might be
suboptimal, and if larger than necessary, significant amount of computation time
would have been wasted. In the GAM, a run of 5000 generations was found effective
to obtain satisfactory solutions. The number of generations can be modified by
the user.

5. Model Validation
For validating the GAM, several examples from the literature were used. The results
obtained from one such example are discussed here. The problem under considera-
tion is a 10 — machine, 16 — component problem.19 Figure 8 shows the sequence
matrix of this problem. This defines the sequence of operations followed by each
component. For example, component 1 is processed in machine 3 first, then in
machine 10, and finally in machine 2.
The 3-cell solution generated by the conventional methods, which do not take
into account any of the production data discussed before, is shown in Fig. 9. Accord-
ing to the solution, the designed cell accounts for one inter-cell movement as part 15
moves from machine 5 to 8. The number of voids, number of zeros, in the main diag-
onal is 16. The resulting grouping efficacy20 is 68.51%.
Figure 10 shows the inputs used for the GAM, and Fig. 11 displays the solution
generated by the GAM. Figure 12 shows the machine arrangements inside each

MC1 MC2 MC3 MC4 MC5 MC6 MC7 MC8 MC9 MC10
PR1 3 1 2
PR2 1 2
PR3 1 3 2
PR4 1 2
PR5 1 3 2
PR6 2 1
PR7 1 2
PR8 1 2
PR9 2 1
PR10 3 1 2
PR11 1 2
PR12 2 1
PR13 1 2
PR14 1 2
PR15 1 2 3
PR16 2 1 3

Fig. 8. Reported 10-machine 16-component problem data.


April 24, 2006 17:5 WSPC/180-JAMS 00071

40 R. Rajagopalan & D. J. Fonseca

MC10 MC3 MC6 MC2 MC5 MC4 MC1 MC8 MC7 MC9
PR1 3 1 0 2
PR7 0 1 2 0
PR5 2 1 3 0
PR3 0 3 2 1
PR6 0 0 1 2
PR4 2 0 1
PR9 0 1 2
PR16 3 1 2
PR2 2 1 0
PR11 0 2 1
PR15 2 1 0 3
PR13 2 1 0
PR8 1 0 2
PR12 0 2 1
PR10 1 3 2
PR14 0 1 2

Fig. 9. Reported 10-machine 16-component problem solution.

Process Time Matrix (Minutes)

MC1 MC2 MC3 MC4 MC5 MC6 MC7 MC8 MC9 MC10
PR1 3 4 5
PR2 5 6
PR3 7 4 3
PR4 8 3
PR5 2 4 3
PR6 2 2
PR7 3 2
PR8 8 7
PR9 3 4
PR10 7 2 2
PR11 2 4
PR12 3 3
PR13 2 3
PR14 2 2
PR15 3 7 4
PR16 4 2 3

Production Volume
{100, 50, 150, 250, 600, 400, 200, 75, 300, 250, 50, 100, 500, 250, 150, 100}
Setup Time in Each Machine (Minutes)
{0, 60, 75, 100, 80, 45, 70, 120, 75, 100, 45,}
Machine Rate per Hour ($/Hr)
{25, 20, 15, 30, 35, 25, 40, 45, 20, 25}
Material Handling Cost per Unit
Inter-Cell Material Handling Cost per Unit = $ 5.00
Intra-Cell Material Handling Cost per Unit = $1.00
Backtracking Material Handling Cost per Unit = $2.00

Fig. 10. Inputs to the GAM.


April 24, 2006 17:5 WSPC/180-JAMS 00071

A Genetic Algorithm Approach for Machine Cell Formation 41

MC3 MC10 MC6 MC2 MC8 MC7 MC9 MC4 MC1 MC5
PR1 1 2 0 3
PR3 3 0 2 1
PR5 1 2 3 0
PR6 0 0 1 2
PR7 1 0 2 0
PR8 1 0 2
PR10 1 2 3
PR12 0 2 1
PR13 2 1 0
PR14 0 1 2
PR2 1 0 2
PR4 0 1 2
PR9 1 2 0
PR11 2 1 0
PR15 3 1 0 2
PR16 1 2 3

Fig. 11. Solution generated by the GAM.

M/c M/c M/c M/c M/c M/c M/c


3 10 6 2 8 7 9

Cell 1 Cell 2

M/c M/c M/c


4 1 5

Cell 3

Fig. 12. Cell configurations generated by the GAM.

cell as generated by the GAM. As far as the machine cells and the component
groups are concerned, the solution generated by the GAM is the same as that
generated by the conventional method. The number of inter-cell moves is 1 due
to part movement from machine 5 to 8; and the number of voids inside the main
diagonal is unchanged from the conventional- method solution at 16, resulting in
the same grouping efficacy of 68.51%.
However, the effect of intra-cell, backtracking, and machine skipping costs, and
the consideration of machine sequences inside the cells is evident in the difference
between the overall costs incurred by the two methods. Table 2 summarizes the vari-
ous material handling movements, and incurred costs. The inclusion of the machine
sequences has resulted in the reduction of the backtracking moves from 12 to 4
April 24, 2006 17:5 WSPC/180-JAMS 00071

42 R. Rajagopalan & D. J. Fonseca

Table 2. Comparison of conventional method versus GAM.

Sl. No. Description Conventional Method’s Results GAM Result


Inputs
1 Number of machines 10 10
2 Number of components 16 16
3 Number of cells 3 3
4 Inter-cell cost per unit($) 5 5
5 Intra-cell cost per unit ($) 1 2
6 Backtracking cost per unit ($) 2 2
7 Crossover rate — 0.8
8 Mutation rate — 0.001
9 Number of generations — 10000
10 Population size — 50
Results
11 Number of inter-cell movements 1 1
12 Number of voids 16 16
13 Number of backtracking moves 12 4
14 Number of machine skipping 7 6
15 Total inter-cell cost ($) 750 750
16 Total intra-cell cost ($) 3300 4350
17 Total backtracking cost ($) 5500 2700
18 Total penalty cost ($) 506.25 506.25
19 Overall cost ($) 10056.25 8306.25

as well as the number of machine skippings from 7 to 6. This reduction is reflected


in the reduction of the overall cost from $10,056.25 to $8,306.25. Even though
there is no change in the machine cells formed by the conventional method and the
GAM, the reduction in the overall cost justifies the superiority of the GAM over
the conventional methods. It can be concluded that this difference in the overall
cost is not considerable enough to bring about changes in the machine cells formed.
However, a significant difference in the overall cost would make a marked change
in the machine cells formed. As the GAM captures all the costs effectively, it is
expected to provide better results than the existing methods as the problem size
and complexity increases.
The number of generations to arrive at this result was approximately 2000.
Data on the number of generations is required to assess whether the number of
generations chosen are sufficient enough for the GAM to reach the optimal/near-
optimal solution. In this case, the number of generations chosen was 10 000, which
was more than sufficient for the GA to arrive at the optimal solution.

6. Conclusions
The GAM has several advantages over the existing methods. In the GAM, the
consideration of the machine sequences at the cell design stage itself gives a more
realistic estimate of the material handling costs. Most of the existing methods
consider the sequencing of machines in the layout stage; and hence, the material
handling costs calculated by most of these methods are far from being realistic.
April 24, 2006 17:5 WSPC/180-JAMS 00071

A Genetic Algorithm Approach for Machine Cell Formation 43

Also, most of the existing cell formation techniques do not consider the effects of
backtracking and machine skipping in arriving at the material handling costs. Many
of them just consider the inter-cell costs alone. The GAM, however, considers all
the elements of the material handling costs in the right magnitude. Moreover, the
methods developed earlier involved stages of complex computations. The GAM,
on the other hand, is a one-step method that is computationally less tedious, and
provides effective solutions in minimal time.

References
1. J. T. Black, The Design of the Factory with a Future (McGraw-Hill Inc., New York,
1991).
2. N. Singh, Systems Approach to Computer-integrated Design and Manufacturing (John
Wiley & Sons Inc., New York, 1996).
3. N. Singh and D. Rajamani, Cellular Manufacturing Systems: Design, Planning and
Control (Chapman & Hall, London, 1996).
4. J. McAuley, Machine grouping for efficient production, The Production Engineer 51
(1972) 53–57.
5. J. R. King, Machine-component grouping in production flow analysis: an approach
using rank order clustering algorithm, International of Journal of Production Research
18 (1980) 213–232.
6. H. M. Chan and D. A. Milner, Direct clustering algorithm for group formation in
cellular manufacture, Journal of Manufacturing Systems 1(1982) 65–74.
7. M. P. Chandrasekharan and R. Rajagopalan, MODROC: An extension of rank order
clustering for group technology, International Journal of Production Research 24
(1986) 1221–1233.
8. S. Sankaran and R. G. Kasilingam, An integrated approach to cell formation and
part routing in group technology manufacturing systems, Engineering Optimization
16 (1990) 235–245.
9. R. Logendran, A workload based model for minimizing total inter-cell and intra-cell
moves in cellular manufacturing, International Journal of Production Research 28
(1990) 913–925.
10. R. Logendran, Impact of sequence of operations and layout of cells in cellular manu-
facturing, International Journal of Production Research 29 (1991) 375–390.
11. B. R. Sarker and C. V. Balan, Cell formation with operation times of jobs for even
distribution of workloads, International Journal of Production Research 34 (1996)
1447–1468.
12. P. Verma and F. Y. Ding, A sequence-based materials flow procedure for design-
ing manufacturing cells, International Journal of Production Research 33 (1995)
3267–3287.
13. G. K. Adil, D. Rajamani and D. Strong, Cell formation considering alternate routings,
International Journal of Production Research 34 (1996) 1361–1380.
14. Y. Won and K. C. Lee, Group technology cell formation considering operation
sequences and production volumes, International Journal of Production Research 39
(2001) 2755–2768.
15. G. Harhalakis, R. Nagi and J. M. Proth, An efficient heuristic in manufacturing
cell formation of group technology applications, International Journal of Production
Research 28 (1990) 185–198.
April 24, 2006 17:5 WSPC/180-JAMS 00071

44 R. Rajagopalan & D. J. Fonseca

16. O. G. Okogbaa, M. T. Chen, C. Changchit and R. L. Shell, Manufacturing system cell


formation and evaluation using new inter-cell flow reduction heuristic, International
Journal of Production Research 30 (1992) 1101–1118.
17. T. L. Lin, M. M. Dessouky, K. Ravikumar and S. M. Ng, A heuristic-based proce-
dure for the weighted production-cell formation problem, IIE Transactions 28 (1996)
579–589.
18. D. E. Goldberg, Genetic algorithms in search, optimization, and machine learning
(Addison-Wesley Publication Co., New York, 1989).
19. L. Salum, The cellular manufacturing layout problem, International Journal Produc-
tion Research 38 (2000) 1053–1069.
20. K. R. Kumar and M. P. Chandrasekharan, Grouping efficacy: A quantitative crite-
rion for goodness of block diagonal forms of binary matrices in group technology,
International Journal of Production Research 28 (1990) 233–243.

Вам также может понравиться