Download as pdf or txt
Download as pdf or txt
You are on page 1of 8

Available online at www.sciencedirect.

com

Applied Mathematical Modelling 32 (2008) 2954–2961


www.elsevier.com/locate/apm

Simple and U-type assembly line balancing problems


with a learning effect
M. Duran Toksarı a,* _ ßleyen b, Ertan Güner c, Ömer Faruk Baykoç
, Selçuk K. Is c

a
Erciyes University, Engineering Faculty, Industrial Engineering Department, Kayseri, Turkey
b
Atatürk University, Engineering Faculty, Industrial Engineering Department, Erzurum, Turkey
c
Gazi University, Engineering and Architecture Faculty, Industrial Engineering Department, Ankara, Turkey

Received 9 February 2007; received in revised form 4 October 2007; accepted 10 October 2007
Available online 17 October 2007

Abstract

In this paper, we introduced learning effect into assembly line balancing problems. In many realistic settings, the pro-
duced worker(s) (or machine(s)) develops continuously by repeated the same or similar activities. Therefore, the produc-
tion time of product shortens if it is processed later. We show that polynomial solutions can be obtained for both simple
assembly line balancing problem (SALBP) and U-type line balancing problem (ULBP) with learning effect.
 2007 Elsevier Inc. All rights reserved.

Keywords: Simple assembly line balancing; U-type line balancing; Learning effect

1. Introduction

Assembly line balancing (ALB) problem was first introduced by Bryton [1] and the first scientific study was
published by Salveson [2]. The assembly line balancing problem consists of assigning tasks to an ordered
sequence of stations such that the precedence relations among the tasks are satisfied and some performance
measure is optimized [3]. Due to computational complexity of the problem-ALB is known to be an NP-hard
problem [4]. The problem is NP-hard, since a special version of the problem, i.e. the one involving no prece-
dence relations among tasks is a bin-packing problem which is NP-hard in the strong sense [3]. Faaland et al.
[5] state that the existing procedures attempting to find the optimal solution have a complexity of at least 2N.
Hence, developing heuristic procedures to solve problems of a practical size remains the only option for the
researchers. The optimum-seeking algorithms can only be used to evaluate the performance of the heuristic
procedures by providing the optimal solutions to the problems of restricted sizes [3].
In recent years, many manufacturers have adopted a just-in-time (JIT) approach to manufacturing, finding
that it improves their productivity, profits, and product quality. JIT is beneficial for companies engaged in

*
Corresponding author.
E-mail address: dtoksari@erciyes.edu.tr (M.D. Toksarı).

0307-904X/$ - see front matter  2007 Elsevier Inc. All rights reserved.
doi:10.1016/j.apm.2007.10.007
M.D. Toksarı et al. / Applied Mathematical Modelling 32 (2008) 2954–2961 2955

repetitive, job shop, or process manufacturing. One of the important changes resulting from JIT implementa-
tion is the replacement of the traditional straight lines with U-shaped production lines [6]. The main benefits of
the U-line as compared to a straight line include reduction in the wasted movement of operators and work-in-
process inventory, improved productivity [7], easier implementation of zero-defects campaign, higher flexibil-
ity in workforce planning in the face of changing demand [8], and improvement in material handling [9].
Most researchers have studied on line balancing problems for more than 50 years. Most of them assumed
that the time of task is independent from learning of worker(s) for repetition tasks. However, learning effect
may arise in a workstation which produces similar tasks. Thus, the actual task time of task performed in the
workstation is shorter if it is scheduled later, rather than earlier in the sequence. Mosheiov [10] determined
that this phenomenon is known in the literature as a ‘‘learning effect’’. The learning effect has been examined
on assembly line balancing problems by a few researchers up to now [11–13].
The learning effect has been examined on scheduling problems such as by researchers in recent years [14],
Wang and Cheng [15,10,16–20]. Biskup [14] was the first to investigate the learning effect in the scheduling
problems. He assumed that the production time as a function of the number of repetitions of the production
of a single item with a learning effect decreases. As in Biskup [14], we assume that the task time of task j with a
learning effect if assigned in position r, is given by
tjr ¼ tj ra ð1Þ

where a 6 0 is a constant learning index.


This study investigates SALBP and ULBP with learning considerations, using learning curve introduced by
Biskup [14]. We proposed that new problem is assignment of tasks to each workstation in line balancing prob-
lems, which aim minimization of the number of workstations for a given cycle time, with a learning effect. The
same learning rate for all workstations was used. However, learning for each workstation restarts because
worker(s) (or machine(s)) into it may be different from each other.
In Sections 2 and 3, we present polynomial solution for SALBP and ULBP, respectively.

2. Simple assembly line balancing problem (SALBP) with a learning effect

Assembly lines consist of successive workstations at which products are processed. Workstations are
defined as places where some tasks (operations) on products are performed. Products stay at each workstation
for the cycle time (C), which corresponds to the time interval between successively completed units [21].
SALBP may aim two goals which are minimization of the number of workstations for a given cycle time
and minimization of cycle time for a given number of work stations. We use first goal in this study. However,
SALBP with a learning effect is new problem.
Theorem 1. Task i through alternative tasks to a workstation in SALBP with a learning effect assigns according
to the shortest task time (STT) rule.

Proof. Task i through alternative tasks to a workstation in SALBP with a learning effect assigns according to
the shortest task time (STT) rule.
The sum of flowtimes of tasks taken place in a workstation must been minimized to maximize number of
tasks in the workstation. Thus, the goal, which is minimization of the number of workstations for a given
cycle, is attained. The sum of flowtimes of tasks taken place in a workstation is minimized by the STT rule.
Learning for other workstation restarts because worker(s) (or machine(s)) into it may be different from each
other. If we assume that ti and tj (ti 6 tj) are task time of tasks i and j, respectively, two states can be examined.
State 1. Task i is assigned before the task j. Note that if c is cycle time, it must be ti + tj 6 c. Otherwise,
second task (j) to the workstation cannot assign.
State 2. Task j is assigned before the task i. Note that it must be tj + ti 6 c. Otherwise, second task (i) to the
workstation cannot assign.
Total flowtime for state 1 (Ci  Cj)state1 then
a
ðC i  C j Þstate1 6 ðti ra Þ þ ðti ra þ tj ðr þ 1Þ Þ
2956 M.D. Toksarı et al. / Applied Mathematical Modelling 32 (2008) 2954–2961

Total flowtime for state 2 (Cj  Ci)state2 then


a
ðC j  C i Þstate2 ¼ ðtj ra Þ þ ðtj ra þ ti ðr þ 1Þ Þ:
Hence,
a
ðC j  C i Þstate2  ðC i  C j Þstate1 P ðtj  ti Þð2ra  ðr þ 1Þ Þ P 0 for ðti 6 tj Þ: 

Algorithm 1. Apply below two steps for each workstation


Step 1: (remaining time  ti) 6 0.10 · (c) is examined for all alternative tasks in the related workstation. It
means that algorithm allows assigning directly the job, if remaining time is equal or less than (0.1 · c). Since,
90%-performance for a workstation is an acceptable performance level. 10%-tolerance gets system reliability
against unexpected defects. Thus, proposed theorem guarantees good utilization rates, the average being
higher than 90% except for last workstation.
Step 2: If (remaining time  ti) > 0.10 · (c) for the related workstation is then assignment is performed
according to Theorem 1.
Algorithm is very simple and robust. We only have an alternative in each assignment. This alternative
determine either by Step 1, which selects the job gets condition (remaining time  ti) 6 0.10 · (c), or by Step 2,
which selects the job gets both condition (remaining time  ti) > 0.10 · (c) and condition STT. So, the time
complexity of the algorithm is given as O(n) where n is number of tasks. CPU times obtained for test problems
in Section 4 show that the time complexity of the algorithm is so short.
The result of the proposed algorithm is again in the following example.

Illustrative example 1. Problem is Jackson 11 which is best-known line balancing problem in literature. A pre-
cedence diagram of problem is given in Fig. 1. Table 1 presents the performance times of the tasks. If cycle
time assume as 11, minimum number of workstations for Jackson 11 without a learning effect are 5. While
solving the problem with a learning effect, cycle time assume as 11 and learning takes place by the 80%-learn-
ing curve, thus a = 0.322. The goal of problem is minimization of the number of workstations.
According to above theorem, selected task through alternative tasks to a workstation in SALBP with a
learning effect is assigned by using the shortest task time (STT) rule if it is not (remaining time  ti) 6 0.90(c).
Table 2 shows results for Jackson 11 problem with a learning effect. Number of workstations are 4 and total
spare time is 4.7.

3. U line balancing problem (ULBP) with a learning effect

U-type line balancing was first introduced by Miltenburg and Wijngaard [22]. The U-type assembly line is
an attractive alternative for assembly production systems since operators become multi skilled by performing
tasks located on different parts of assembly line [23].

2 6 8 10

1
11
4 7 9

Fig. 1. Precedence diagram for Jackson 11 problem.


M.D. Toksarı et al. / Applied Mathematical Modelling 32 (2008) 2954–2961 2957

Table 1
Task times for Jackson 11 problem
Task No. Task time
1 6
2 2
3 5
4 7
5 1
6 2
7 3
8 6
9 5
10 5
11 4

Table 2
SALB results for Jackson 11 problem with a learning effect
Workstation Assigned task(s) Remaining spare time
1 1–3 0.3
2 5–2–6–8–10 0.0
3 4–7 1.6
4 9–11 2.8

Table 3
ULBP results for Jackson 11 problem with a learning effect
Workstation Assigned task(s) Remaining spare time
1 1–3 0.3
2 5–2–6–11–4 0.12
3 7–9–10 0.5
4 8 5.0

ULBP may aim two goals which are minimization of the number of workstations for a given cycle time and
minimization of cycle time for a given number of work stations. We use first goal in this study. However,
ULBP with a learning effect is new problem. When selecting task through alternative tasks to assign to the
workstation for ULBP with a learning effect, above proposed algorithm should use due to goal of ULBP is
the same with SALB. On the other words, if (remaining time  ti) > 0.10(c) is then selected task through alter-
native tasks to a workstation in ULBP with a learning effect assigns according to the STT rule.
Illustrative example 2 (continued). While solving the problem with a learning effect, cycle time assumes as 11
and learning takes place by the 80%-learning curve, thus a = 0.322. The goal of problem is minimization of
the number of workstations.

According to proposed theorem, selected task through alternative tasks to a workstation in ULBP with a
learning effect is assigned by using the shortest task time (STT) rule.
Table 3 shows ULBP results for Jackson 11 problem with a learning effect. Number of workstations are 4
and total spare time is 5.92.

4. Computational results

For large scale problems, some heuristic approaches needed. In Section 3, we proposed algorithm to solve
large scale SALBP and ULBP with learning effect. Proposed algorithm is solved for test problems, which
are Jackson 11, Heskia 28, Gunther 35, Killbridge 45, Warnecke 58 and Mukherje 94, under three different
2958 M.D. Toksarı et al. / Applied Mathematical Modelling 32 (2008) 2954–2961

learning rates. Indeed, for all problems, we used two different cycle times, which were obtained by accounting
total 15% and 25% of task time of each problem, respectively. Table 4 presents theoretical minimum number
of workstations for all test problems.
Table 5 shows that proposed algorithm obtains the same results with the theoretical minimum number of
workstations for Jackson 11 when learning effect is 90%. Furthermore, it obtains less number of workstation
than the theoretical minimum number of workstations when learning rate increases. Because each workstation
can take more tasks due to high learning rate. ULBP produced better solution than SALBP when cycle time
and learning effect equal to 7% and 80%, respectively.
For problem Heskia 28 by using proposed algorithm, in Table 6, ULBP produced better solutions than
SALBP when learning effect were 90% and 80% for both values of cycle time. Proposed algorithm obtains less
number of workstation than the theoretical minimum number of workstations for all test configurations
except solution of SALBP when learning effect is 90%.
Table 7 shows that the proposed algorithm obtains less number of workstations than the theoretical min-
imum number of workstations for Gunther 35 when learning effect is 80% and 70%. ULBP produced better
solution than SALBP when learning effect is 90% and 80%.
For problem Kilbridge 45 by using proposed algorithm, in Table 8, ULBP produced better solutions than
SALBP when cycle time equaled 82 for both values of learning effect. Proposed algorithm obtains less number
of workstation than the theoretical minimum number of workstations for all test configurations except solu-
tions of SALBP and ULBP when learning effect is 90%.

Table 4
Theoretical minimum number of workstations for all test problems
Problem No. of task Cycle time No. of station
Jackson 11 7 7
12 4
Heskia 28 154 7
256 4
Gunther 35 73 7
121 4
Kilbridge 45 82 7
136 4
Warnecke 58 233 7
387 4
Mukherje 94 632 7
1052 4

Table 5
Balance with proposed algorithm of SALBP and ULBP with effects of learning for Jackson 11
Cycle Theoretical minimum Learning Obtained minimum CPU Obtained minimum CPU time (s)a
time number of effect (%) number of workstations time number of workstations
workstations for SALBP (s)a for ULBP
7 7 90 7 0.20 7 0.22
80 7 0.21 6 0.24
70 6 0.20 6 0.22

12 4 90 4 0.18 4 0.20
80 3 0.18 3 0.19
70 3 0.17 3 0.19
a
We used Core Duo 1.73 GHz CPU and 1 GB RAM to test. Results are the average of five runs.
M.D. Toksarı et al. / Applied Mathematical Modelling 32 (2008) 2954–2961 2959

Table 6
Balance with proposed algorithm of SALBP and ULBP with effects of learning for Heskia 28
Cycle Theoretical minimum Learning Obtained minimum CPU Obtained minimum CPU time (s)a
time number of effect (%) number of workstations time number of workstations
workstations for SALBP (s)a for ULBP
154 7 90 7 0.33 6 0.35
80 6 0.28 5 0.30
70 5 0.31 5 0.33
256 4 90 4 0.24 3 0.28
80 3 0.26 2 0.30
70 2 0.26 2 0.28
a
We used Core Duo 1.73 GHz CPU and 1 GB RAM to test. Results are the average of five runs.

Table 7
Balance with proposed algorithm of SALBP and ULBP with effects of learning for Gunther 35
Cycle Theoretical minimum Learning Obtained minimum CPU Obtained minimum CPU time (s)a
time number of effect (%) number of workstations time number of workstations
workstations for SALBP (s)a for ULBP
73 7 90 7 0.53 6 0.58
80 6 0.53 5 0.57
70 5 0.49 5 0.52

121 4 90 4 0.44 4 0.51


80 3 0.43 3 0.50
70 2 0.41 1 0.50
a
We used Core Duo 1.73 GHz CPU and 1 GB RAM to test. Results are the average of five runs.

Table 8
Balance with proposed algorithm of SALBP and ULBP with effects of learning for Kilbridge 45
Cycle Theoretical minimum Learning Obtained minimum CPU Obtained minimum CPU time (s)a
time number of effect (%) number of workstations time number of workstations
workstations for SALBP (s)a for ULBP
82 7 90 7 0.74 6 0.77
80 6 0.72 5 0.76
70 5 0.70 3 0.76

136 4 90 4 0.68 4 0.72


80 3 0.68 2 0.71
70 2 0.67 1 0.71
a
We used Core Duo 1.73 GHz CPU and 1 GB RAM to test. Results are the average of five runs.

Table 9
Balance with proposed algorithm of SALBP and ULBP with effects of learning for Warnecke 58
Cycle Theoretical minimum Learning Obtained minimum CPU Obtained minimum CPU time (s)a
time number of effect (%) number of workstations time number of workstations
workstations for SALBP (s)a for ULBP
233 7 90 6 1.02 6 1.08
80 5 1.01 4 1.08
70 3 1.01 2 1.07
387 4 90 3 0.99 3 1.04
80 3 0.97 2 1.02
70 2 0.95 1 1.01
a
We used Core Duo 1.73 GHz CPU and 1 GB RAM to test. Results are the average of five runs.
2960 M.D. Toksarı et al. / Applied Mathematical Modelling 32 (2008) 2954–2961

Table 10
Balance with proposed algorithm of SALBP and ULBP with effects of learning for Mukherje 94
Cycle Theoretical minimum Learning Obtained minimum CPU Obtained minimum CPU time (s)a
time number of effect (%) number of workstations time number of workstations
workstations for SALBP (s)a for ULBP
632 7 90 6 1.41 6 1.45
80 4 1.38 4 1.42
70 2 1.36 1 1.42

1052 4 90 3 1.32 3 1.39


80 2 1.30 2 1.39
70 2 1.30 1 1.38
a
We used Core Duo 1.73 GHz CPU and 1 GB RAM to test. Results are the average of five runs.

Tables 9 and 10 show that proposed algorithm obtains less number of workstations than the theoretical
minimum number of workstations for all configurations of learning effect and cycle time.
If problem scale or cycle time increases, learning effect will be more dominant to increase the number of
tasks at each workstation.

5. Conclusions

It is shown in this paper that the simple assembly line balancing problem and U-type line balancing prob-
lem with the consideration of learning effects remains polynomially solvable. More computational effort than
the effort required for solving the original problems is required to solve line balancing problems with a learn-
ing effect. Assembly line balancing with a learning effect are clearly interesting and significant topics for future
research on assignment problems.
We also tested proposed algorithm for large scale SALBP and ULBP with learning effect. Proposed algo-
rithm obtained solutions equal and less their theoretical minimum numbers of workstations. Furthermore,
more computational effort than the effort required for solving the original problems is required to solve line
balancing problems with a learning effect. Assembly line balancing problems with learning effect are clearly
interesting and significant topics for future research on assignment problems.

References

[1] Bryton, B., 1954. Balancing of a Continuous Production Line, M.S. Thesis, Northwestern University, Evanston, IL.
[2] M.E. Salveson, The assembly line balancing problem, Journal of Industrial Engineering 6 (3) (1955) 18–25.
[3] E. Erel, S.C. Sarin, A survey of the assembly line procedures, Production Planning & Control 9 (5) (1998) 414–434.
[4] R.M. Karp, Reducibility among combinatorial problems, in: R.E. Miller, J.W. Thatcher (Eds.), Complexity of Computer
Computations, Plenum Press, New York, 1972, pp. 85–133.
[5] B.H. Faaland, T.D. Klastorin, T.G. Schmitt, A. Shtub, Assembly line balancing with resource dependent task times, Decision
Sciences 23 (1992) 343–364.
[6] W.C. Chiang, T.L. Urban, The stochastic U-line balancing problem: a heuristic procedure, European Journal of Operational
Research 175 (2006) 1767–1781.
[7] H. Hirano, JIT Factory Revolution, Productivity Press, Cambridge, MA, 1988.
[8] Y. Monden, Toyota Production System, Industrial Engineering and Management Press, Norcross, GA, 1983.
[9] K. Sekine, One-Piece Flow, Productivity Press, Portland, OR, 1992.
[10] G. Mosheiov, Scheduling problems with a learning effect, European Journal of Operational Research 132 (2001) 687–693.
[11] Y. Cohen, G. Vitner, S.C. Sarin, Optimal allocation of work in assembly lines for lots with homogenous learning, European Journal
of Operational Research 168 (2006) 922–931.
[12] Y. Cohen, M.E. Dar-El, Optimizing the number of stations in assembly lines under learning for limited production, Production
Planning and Control 9 (3) (1998) 230–240.
[13] A.K. Chakravarty, A. Shtub, Modeling the effects of learning and job enlargement on assembly systems with parallel lines,
International Journal of Production Research 26 (2) (1988) 267–281.
[14] D. Biskup, Single-machine scheduling with learning considerations, European Journal of Operational Research 115 (1999) 173–178.
M.D. Toksarı et al. / Applied Mathematical Modelling 32 (2008) 2954–2961 2961

[15] G.Q. Wang, T.C.E. Cheng, Single machine scheduling with learning effect considerations, Annals of Operations Research 98 (2000)
273–290.
[16] G. Mosheiov, J.B. Sidney, Scheduling with general job dependent learning curves, European Journal of Operational Research 147
(2003) 665–670.
[17] W.H. Kuo, D.L. Yang, Minimizing the total completion time in a single machine scheduling problem with a time dependent learning
effect, European Journal of Operational Research 174 (2006) 1184–1190.
[18] W.H. Kuo, D.L. Yang, Single machine group scheduling with a time dependent learning effect, European Journal of Operational
Research 33 (2006) 2099–2112.
[19] T. Eren, E. Güner, Minimizing total tardiness in a scheduling problem with a learning effect, Applied Mathematical Modelling 31
(2007) 1351–1361.
[20] C. Koulamas, G.J. Kyparisis, Single machine and two machine flowshop scheduling with general learning functions, European
Journal of Operational Research 178 (2007) 402–407.
[21] K. Ağpak, H. Gökçen, Assembly line balancing: two resource constrained cases, International Journal of Production Economics 96
(2005) 129–140.
[22] J. Miltenburg, J. Wijngaard, The U-line balancing problem, Management Science 40 (10) (1994) 1378–1388.
[23] H. Gökçen, K. Ağpak, C. Gencer, E. Kizilkaya, A shortest route formulation of simple U-type assembly line balancing problem,
Applied Mathematical Modeling 29 (2005) 373–380.

You might also like