Intelligent Information Management, 2013, 5, 84-92

http://dx.doi.org/10.4236/iim.2013.53009 Published Online May 2013 (http://www.scirp.org/journal/iim)

Genetic Algorithm for Concurrent Balancing of

Mixed-Model Assembly Lines with Original Task Times of
Models
Panneerselvam Sivasankaran1, Peer Mohamed Shahabudeen2
1
Department of Mechanical Engineering, Rajalakshmi Engineering College, Anna University, Chennai, India
2
Department of Industrial Engineering, College of Engineering, Anna University, Chennai, India
Email: [email protected], [email protected]

Received March 9, 2013; revised April 10, 2013; accepted April 26, 2013

Copyright © 2013 Panneerselvam Sivasankaran, Peer Mohamed Shahabudeen. This is an open access article distributed under the
Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the
original work is properly cited.

ABSTRACT
The growing global competition compels manufacturing organizations to engage themselves in all productivity impro-
vement activities. In this direction, the consideration of mixed-model assembly line balancing problem and implement-
ing in industries plays a major role in improving organizational productivity. In this paper, the mixed model assembly
line balancing problem with deterministic task times is considered. The authors made an attempt to develop a genetic
algorithm for realistic design of the mixed-model assembly line balancing problem. The design is made using the ori-
ginnal task times of the models, which is a realistic approach. Then, it is compared with the generally perceived design
of the mixed-model assembly line balancing problem.

Keywords: Assembly Line Balancing; Cycle Time; Genetic Algorithm; Crossover Operation; Mixed-Model

1. Introduction In this paper, the ALBP 1 problem with mixed-model

is considered. In this problem, there will be many models
The assembly line balancing problem is basically classi-
which are to be produced in batches using the same as-
fied into ALBP 1 and ALBP 2. The ALBP 1 is the type 1
sembly line. The presence of mixed-model makes the
assembly line balancing problem in which the objective
design of the assembly line more complex, in terms of
is to group the tasks into the minimum number of work-
processing times of the tasks and cycle times of the
model. The average processing time T j  of each task
stations for a given cycle time, which in turn maximizes
the balancing efficiency of the assembly line. The ALBP
as given by the following formula is normally taken as
2 is the type 2 assembly line balancing problem, in which
the representative figure for each task time. If tij  0 for
the tasks are grouped into a given number of worksta-
a given i, then the denominator will be decremented by 1.
tions such that the cycle time (the maximum of the sum
M
of the task times of the workstations) is minimized. This
in turn maximizes the production rate.
 tij
Tj  i 1
for j  1, 2,3, , n where tij  0
The cycle time is computed from the given production M
volume per shift using the following formula (Pan- where,
neerselvam 2012). tij is the time of the task j in the model i
Effective time available per shift T j is the average time of the task j
Cycle time 
Production volume per shift M is the number of models
The balancing efficiency of the solution of the line n is the number of tasks
balancing problem is given by the following formula In the single model assembly line balancing problem,
(Panneerselvam, 2012). there will be one cycle time, where as in the mixed-
model assembly line balancing problem, each model will
Sum of all task times
Balancing   100 have a cycle time which is normally computed based on
Number of workstations  Cycle time the desired production volume per shift of that model. In

Copyright © 2013 SciRes. IIM

 P. SIVASANKARAN, P. M. SHAHABUDEEN 85

a problem with two models, if the production volume of works carried out so far in literature, for each task, the
Model 1 is 24 assemblies per shift, then the correspond- authors assumed the average of the times of that task in
ing cycle time is 20 minutes. If the production volume of all the models as the time of that task. If this is followed,
Model 2 is 48 assemblies per shift, then the correspond- after forming the workstations, for some station, there is
ing cycle time is 10 minutes. In the case of single model, a possibility that sum of the original times of the tasks
each cycle time may be considered as such without any which are assigned to that workstation will be more than
modification. In the mixed-model assembly line balance- the cycle time, which will lead to infeasible workstation.
ing problem, the line may be designed for a common Hence, in this paper, the task times of the models are
cycle time which is computed based on the cycle times of used as such in the design of the assembly line without
the individual models. If the cycle times of the models any modification, which is a major contribution of this
are one and the same, then the common cycle time will research, because it introduces perfection in the solution.
be equal to the single value; otherwise, the average of the
cycle times will be treated as the common cycle time for 2. Literature Review
the problem. Consider Model 1 and Model 2 whose
This section presents the review of literature of the
precedence networks are shown in Figures 1 and 2, re-
mixed-model assembly line balancing problem. Gokcen
spectively. The number by the side of each node repre-
and Erel (1998) have developed a binary integer formu-
sents the respective task time. The precedence network
lation for of the models. Kim and Kim (2000) have con-
for the combined model is shown in Figure 3. In other
sidered the mixed-model assembly line balancing prob-
lem. They developed a convolutionary algorithm for ba-
lancing and sequencing the assembly line. Matanachai
and Yano (2001) have developed a heuristic based on
filtered beam search for balancing the mixed-model as-
sembly line to reduce work overload as well as to main-
tain reasonable workload balance among the stations. Jin
and Wu (2002) have considered the mixed-model assem-
bly line balancing problem and developed a heuristic
called “variance algorithm” to balance the assembly line.
Bukchin and Rabinowitch (2006) have developed a
Figure 1. Precedence network of Model 1. branch and bound based solution approach for the
mixed-model assembly line balancing problem to mini-
mize the number of workstations and task duplication
costs. Noorul Haq, Zayaprakash and Rengarajan (2006)
have developed a hybrid genetic algorithm approach to
mixed-model assembly line balancing problem in which
the objective is to minimize the number of workstations
for a given cycle time. Su and Lu (2007) have considered
the mixed-model assembly line balancing problem in
which the objective is to design the assembly line to
smooth the workload balance within each workstation.
They developed a genetic algorithm to find the sequence
Figure 2. Precedence network of Model 2. of models which will minimize the cycle time. They car-
ried out a simulation experiment. Bock (2008) has used
distributed search methods for balancing mixed-model
assembly lines in the auto industry, in which the object-
tive is to minimize the number of workstations. Bai,
Zhao and Zhu (2009) have considered mixed-model as-
sembly line balancing problem for which they developed
a new hybrid genetic algorithm for finding good solution
of the problems.
Ozcan, Cercioglu, Gokcen and Toklu (2010) have con-
sidered the balancing and sequencing of parallel mixed
assembly lines in which more than one assembly line are
balanced together. They have developed a simulated an-
Figure 3. Combined model of Model 1 and Model 2. nealing algorithm for this problem to minimize the num-

Copyright © 2013 SciRes. IIM

86 P. SIVASANKARAN, P. M. SHAHABUDEEN

ber of workstations and attain equalization of worksta- The cyclic crossover method presented by Senthilku-
tions among workstations. Zhang and Han (2012) have mar and Shahabudeen (2006) gives two offspring from
developed an improved differential evolution algorithm the chromosome 1 and the chromosome 2 as shown in
for the mixed model assembly line balancing problem Table 3.
applied to car manufacturing industry.
From these literatures, it is observed that the research- 4.1. Construction of Ordered Vector
ers used mathematical models, heuristics, genetic algo
The genes of each chromosome are to be ordered (rear-
rithm, simulated annealing algorithms, etc. to design the ranged) such that the serial assignment of the genes from
assembly line of mixed-models. In all the heuristics as the ordered vector to workstations does not violate the
well as meta-heuristics, the combined model of the mod- precedence constraints as shown in the Figure 1 as well
els is derived and the average time for each task is com- as in the Figure 2. The steps of constructing ordered vec-
puted. Then the design of the assembly line is done for a tor for a given chromosome/offspring are presented in
given cycle time common to all the models based on the Annexure 1.
average task times. Later, the allocation of the tasks to The application of the steps to the chromosome 1
different stations of each model is carried out based on shown in Table 1 gives an ordered vector as shown in
the design of the combined model, which is considered to Table 4.
be unrealistic. Hence, in this paper, an attempt has been
made to design the assembly line for the mixed models 4.2. Evaluation of Fitness Function of
based on the original timings of the models, which is Chromosome
more realistic. Further, a genetic algorithm is designed
to balance the mixed-model assembly line, in which the The fitness function of the chromosome 1, namely bal-
objective is to minimize the number of stations/ maximize ancing efficiency is obtained by assigning the tasks seri-
the balancing efficiency. ally from left to right from its ordered vector 1-5-3-4-8-
7-9-2-6-10 into workstations for a given cycle time of 20
3. Problem Statement units as shown in Table 5. It is assumed that the cycle
time of the model 1 as well as that of the model 2 as 20
The problem considered in this paper is the mixed-model minutes. Hence, the cycle time of the combined model is
assembly line balancing problem. The objective of this also 20 minutes, which is the average of the cycle times
research is to group the tasks into a minimum number of of both the models.
workstations without violating precedence constraints for While assigning a task into a workstation, that task
a common cycle time, which is derived, based on the pertaining to all the models should be assigned to the
cycle times of the models. Generally, the common cycle same workstation. If a task is available in only one model
time is the average of the cycle times of the models. In then that can be independently assigned to the current
this paper, a genetic algorithm is designed for the mixed- workstation. The allocations of the tasks to different
model assembly line balancing problem, in which the workstations of both the models are pictorially shown in
original times of the tasks are used while concurrently Figure 4.
designing the workstations of the models.
4.3. Steps of GA Based Heuristic with Cyclic
4. Genetic Algorithm Crossover Method
This section presents a genetic algorithm for designing The steps of GA based heuristic with cyclic crossover
mixed-model assembly line balancing problem. It uses
cyclic crossover to obtain offspring. Table 2. Chromosome 2.
Consider the combined precedence network of assem- Position 1 2 3 4 5 6 7 8 9 10
bling the Model 1 and Model 2, which has ten tasks, as Chromosome 2 3 6 9 2 5 10 4 7 1 8
shown in Figure 3. Each chromosome in the original
population is generated by randomly assigning the tasks Table 3. Offspring of chromosomes 1 and 2.
to different gene positions in the chromosome. Two
Position 1 2 3 4 5 6 7 8 9 10
sample chromosomes for the tasks in Figure 3 are shown
Offspring 1 3 7 9 2 5 10 4 6 1 8
in Tables 1 and 2.
Offspring 2 5 10 8 3 7 4 9 1 6 2
Table 1. Chromosome 1.
Table 4. Ordered vector of chromosome 1.
Position 1 2 3 4 5 6 7 8 9 10
Gene Position J 1 2 3 4 5 6 7 8 9 10
Chromosome 1 1 5 8 3 7 4 9 6 10 2 Chromosome 1 1 5 3 4 8 7 9 2 6 10

Copyright © 2013 SciRes. IIM

 P. SIVASANKARAN, P. M. SHAHABUDEEN 87

Table 5. Solution for the ordered vector 1-5-3-4-8-7-9-2-6- method to group the tasks of the mixed-model assembly
10. line balancing problem into a minimum number of work-
Assigned Tasks Unassigned Time
stations are presented in Figure 5. Here, the balancing
Workstation efficiencies of the individual models and the balancing
Model 1 Model 2 Model 1 Model 2 efficiency of the combined model are maximized.
1 1 12 10
I 5. Comparison of Proposed Ga Results with
- 5 12 5 Conventional Ga Results
II 3 3 13 8
This section gives the comparison of the results of the
4 4 6 4 proposed GA algorithm with that of the generally per-
III ceived genetic algorithm using the example given in Sec-
- 8 6 0
tion 1. In the genetic algorithm, the number of iteration is
7 7 11 13 20. The size of the population and that of the sub-popu-
IV - 9 11 7

2 - 1 7

6 - 15 20
V
10 10 4 7

Balancing efficiency 64% 73%

Combined efficiency 68.5%

Figure 4. Assignment of tasks to workstations.

Copyright © 2013 SciRes. IIM

88 P. SIVASANKARAN, P. M. SHAHABUDEEN

Copyright © 2013 SciRes. IIM

 P. SIVASANKARAN, P. M. SHAHABUDEEN 89

lation are 50 and 20, respectively.

5.1. Results of Proposed Genetic Algorithm

The application of the proposed genetic algorithm to the
example problem shown in Section 2 gives the best chro-
mosome and the corresponding ordered vector as shown
in Table 6. The details of the stations of the Model 1 and
Model 2 are shown Table 7. The allocations of the task
to different workstations of both models are shown in
Figure 6.

5.2. Results of Generally Perceived Genetic

Algorithm
Generally all the researchers use the combined prece-
dence network, average processing times of the tasks to
design the assembly line for the common cycle time of
the combined model. The average time of the tasks are
shown in Table 8. It should be noted that the average
time of a task is computed by dividing the sum of the
task times in all the models of that task where it is present,
by the number of models in which that task is present.
But, in past researches, for a given task, the average task
time is obtained by dividing the sum of that task times in

Table 6. Best chromosome and corresponding ordered vec-

tor.

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

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

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

Table 7. Solution using the proposed genetic algorithm with

individual task times of the models.

Assigned Tasks Unassigned Time

Workstation
Model 1 Model 2 Model 1 Model 2

1 1 12 10

I 2 - 2 10

- 5 2 5

4 4 6 4
II
- 8 6 0

3 3 13 8

III 6 - 8 8

- 9 8 2
7 7 11 13
IV
10 10 0 0
Balancing efficiency 80% 91.25%
Figure 5. Steps of GA based heuristic with cyclic crossover
Combined efficiency 85.625%
method.

Copyright © 2013 SciRes. IIM

90 P. SIVASANKARAN, P. M. SHAHABUDEEN

all the models by the number of models, which is a wr- in Figure 7. From the Figure 7, it is clear that the bal-
ong approach. ancing efficiencies of the individual models using P-GA
The application of the generally perceived genetic al- are better than that that of the individual models using
gorithm with cyclic crossover gives the best chromosome GP-GA. The comparison of combined balancing effi-
and the corresponding ordered vector as shown in Table ciencies of GP-GA and P-GA is shown in Figure 8.
9. From the Figure 8, it is seen that the proposed GA per-
For the ordered vector shown in Table 9, the assem- forms better than the generally perceived GA, in terms of
blyline balancing (ALB) results using the generally per- combined balancing efficiency.
ceived genetic algorithm with average task times are al-
ready shown in Table 5. The allocations of the tasks to 6. Conclusions
different workstations of both models for this solution
The mixed-model assembly line balancing gives a greater
are already pictorially shown in Figure 5.
challenge to the designers in industries in terms of deal-
The comparison of balancing efficiencies of the gener-
ing with different models. The researchers use the aver-
ally perceived GA (GP-GA) and proposed GA (P-GA),
which is shown in Tables 5 and 7, respectively is shown age times of the tasks of the models as the timings of the
tasks of the combined model while designing the line.
But, this is an approximation method of dealing with the
data. Hence, in this paper, the design of the assembly line
of the models is done concurrently for all the models
using their original task times. Such an attempt gives a
realistic design of the mixed-model assembly line.
A genetic algorithm is designed to balance the assem-
bly lines of the models of the mixed-model assembly line
balancing problem. In this algorithm, a cyclic crossover

GP-GA P-GA
Figure 6. Assignment of tasks to workstations.
Figure 7. Comparison of balancing efficiencies of models
using GP_GA and P-GA.
Table 8. Average task times of the combined model.

Task j 1 2 3 4 5 6 7 8 9 10
Task time in
8 10 7 14 0 5 9 0 0 11
Model 1
Task time in
10 0 12 16 5 0 7 4 6 13
Model 2
Average task
9 10 9.5 15 5 5 8 4 6 12
time

Table 9. Best chromosome and ordered vector using gener-

ally perceived genetic algorithm.

Gene Position 1 2 3 4 5 6 7 8 9 10
Best Chromosome 1 8 7 2 3 9 5 6 4 10
Figure 8. Comparison of combined balancing efficiencies of
Ordered vector 1 2 3 5 6 4 8 7 9 10
GP-GA and P-GA.

Copyright © 2013 SciRes. IIM

 P. SIVASANKARAN, P. M. SHAHABUDEEN 91

method is used to perform crossover operation between 159-169. doi:10.1016/S0360-8352(02)00190-0

chromosomes. The superiority of this approach is dem- [6] Y. K. Kim and J. Y. Kim, “A Co-Evolutionary Algorithm
onstrated using a numerical example. The proposed ge- for Balancing and Sequencing in Mixed-Model Assembly
netic algorithm with cyclic crossover and with individual Lines,” Applied Intelligence, Vol. 13, No. 3, 2000, pp.
247-258. doi:10.1023/A:1026568011013
task times of the models is a unique contribution to lit-
erature, because in the past all the researchers used aver- [7] S. Matanachai and C. A. Yano, “Balancing Mixed-Model
age times of the tasks of the combined model to form the Assembly Lines to Reduce Work Overload,” IIE Trans-
actions, Vol. 33, No. 1, 2001, pp. 29-42.
stations which is an approximate method of balancing. doi:10.1080/07408170108936804
[8] A. N. Ha, J. Jayaprakash and K. Rengarajan, “A Hybrid
7. Acknowledgements Genetic Algorithm Approach to Mixed-Model Assembly
The authors thank the anonymous referees for their con- Line Balancing,” International Journal of Advanced Ma-
nufacturing Technology, Vol. 28, No. 3-4, 2006, pp.
structive suggestions, which have improved the paper.
337-341. doi:10.1007/s00170-004-2373-3
[9] U. Ozcan, H. Cercioglu, H. Gokcen and B. Toklu, “Bal-
REFERENCES ancing and Sequencing of Parallel Mixed-Model Assem-
bly Lines,” International Journal of Production Research,
[1] Y. Bai, H. Zhao and L. Zhu, “Mixed-Model Assembly Vol. 48, No. 17, 2010, pp. 5089-5113.
Line Balancing Using the Hybrid Genetic Algorithm,” doi:10.1080/00207540903055735
International Conference on Measuring Technology and
Mechatronics Automation, Zhangjiajie, 11-12 April 2009, [10] R. Panneerselvam, “Production and Operations Manage-
pp. 242-245. ment,” 3rd Edition, PHI Learning Private Limited, New
Delhi, 2012.
[2] S. Bock, “Using Distributed Search Methods for Balanc-
ing Mixed-Model Assembly Lines in the Automotive In- [11] P. Senthilkumar and P. Shahabudeen, “GA Based Heuris-
dustry,” OR Spectrum, Vol. 30, No. 3, 2008, pp. 551-578. tic for the Open Shop Scheduling Problem,” International
doi:10.1007/s00291-006-0069-9 Journal of Advanced Manufacturing Technology, Vol. 30,
No. 3-4, 2006, pp. 297-301.
[3] Y. Bukchin and I. Rabinowitch, “A Branch-and-Bound doi:10.1007/s00170-005-0057-2
Based Solution Approach for the Mixed-Model Assembly
Line-Balancing Problem for Minimizing Stations and [12] P. Su and Y. Lu, “Combining Genetic Algorithm and
Task Duplication Costs,” European Journal of Opera- Simulation for the Mixed-Model Assembly Line Balanc-
tional Research, 174, No. 1, 2006, pp. 492-508. ing Problem,” 3rd International Conference on Natural
doi:10.1016/j.ejor.2005.01.055 Computation (ICNC 2007), Vol. 4, Haikou, 24-27 Au-
gust 2007, pp. 314-318.
[4] H. Gokcen and E. Erel, “Binary Integer Formulation for
Mixed-Model Assembly Line Balancing Problem,” Com- [13] X. M. Zhang and X. C. Han, “The Balance Problem
puters & Industrial Engineering, Vol. 34, No. 2, 1998, pp. Solving of the Car Mixed-Model Assembly Line Based
451-461. doi:10.1016/S0360-8352(97)00142-3 on Hybrid Differential Evolution Algorithm,” Applied
Mechanics and Materials, Vol. 220-223, 2012, pp. 178-
[5] M.-Z. Jin and S. D. Wu, “A New Heuristic Method for 183. doi:10.4028/www.scientific.net/AMM.220-223.178
Mixed-Model Assembly Line Balancing Problem,” Com-
puters & Industrial Engineering, Vol. 44, No. 1, 2002, pp.

Annexure 1. Steps of Construction of cessors in Figure 3 is 3.

Ordered Vector  Number of tasks (genes of the chromosome), N
 Initialize gene position of the ordered vector, K  1
The steps of constructing ordered vector are presented in
this appendix.
Step 2: Set the gene position of the chromosome,
Step 1: Input the chromosome.
J 1
 Let the chromosome I be as shown in Table A1 along Step 3: If the STATUS of the gene J of the chromo-
with the values for the STATUS row as zero. If the some I is equal to 1, then go to Step 9; else, go to Step 4.
value of the STATUS for a gene (task) is zero, then it Step 4: If all the immediate predecessors values in the
signifies that it is not assigned to any workstation; Table 9 with respect two the task at the gene position J
otherwise, it signifies that it is assigned to some of the chromosome I are zero, then go to Step 5; other-
workstation. wise, go to Step 9.
 Form the immediate predecessor(s) matrix of the Step 5: Assign the task at the gene position J of the
tasks of the combined model as shown in Table A2. chromosome I to the gene position K of the ordered vec-
The maximum of the number of immediate prede- tor. OVK  CIJ

Copyright © 2013 SciRes. IIM

92 P. SIVASANKARAN, P. M. SHAHABUDEEN

Step 6: Set the status of the gene position J of the Table A2. Immediate predecessor(s) matrix [IPpq].
chromosome I to 1. STATUS J  1
Task (p) Immediate Predecessor(s) (q)
Step 7: Change the value of CIJ in immediate prede-
cessor matrix to zero, wherever it appears in it. 1 Nil
Step 8: Increment the gene position (K) of the ordered 2 1
vector by 1 and go to Step 2.
Step 9: Increment the gene position (J) of the chro- 3 1
mosome by 1. 4 1
Step 10: If J  N , then go to Step 3; else go to Step
5 1
11.
Step 11: Stop. 6 2 3

7 3 4
Table A1. Chromosome I.
8 4 5
Gene Position J 1 2 3 4 5 6 7 8 9 10
9 8
STATUS 0 0 0 0 0 0 0 0 0 0
10 6 7 9
Chromosome 1 1 5 8 3 7 4 9 6 10 2

Copyright © 2013 SciRes. IIM