You are on page 1of 13

IIE Transactions (2003) 35, 73–85

Copyright  2003 ‘‘IIE’’


0740-817X/03 $12.00+.00
DOI: 10.1080/07408170390116670

A weighted approach for assembly line design with station


paralleling and equipment selection

JOSEPH BUKCHIN1 and JACOB RUBINOVITZ2; *


1
Department of Industrial Engineering, Faculty of Engineering, Tel-Aviv University, Tel-Aviv, Israel 69978
E-mail: bukchin@eng.tau.ac.il
2
Faculty of Industrial Engineering and Management, Technion - Israel Institute of Technology, Haifa, Israel 32000
E-mail: ierjr01@ie.technion.ac.il

Received August 2001 and accepted May 2002

This paper studies the problem of assembly line design, focusing on station paralleling and equipment selection. Two problem
formulations, minimizing the number of stations, and minimizing the total cost, are discussed. The latter formulation is dem-
onstrated by several examples, for different assembly system conditions: labor intensive or equipment intensive, and with task times
that may exceed the required cycle time. It is shown that the problem of assembly system design with parallel stations can be
treated as a special case of the problem of equipment selection for an assembly line. A branch and bound optimal algorithm
developed for the equipment selection problem is adapted to solve the parallel station problem. Experiments are designed to
investigate and demonstrate the influence of system parameters, such as assembly sequence flexibility and cycle time, on the
balancing improvement due to station paralleling. An ILP formulation is developed for the combined problem of station paral-
leling with equipment selection, and an optimal solution of an example problem is presented.

1. Introduction task into a sequence of workstations, so that the assembly


time required at each workstation is approximately the
A considerable proportion of manufacturing activities same. The assembly work is completed on the line as the
and costs are devoted to the assembly of products. As the parts pass each station in sequence, with every station
life cycle of products becomes shorter, with rapid design adding its work content to the assembly task. The cycle
changes and growing product complexity, the assembly time of the assembly line is determined by the workstation
systems must adapt to the change. A technological solu- with the maximum work content time. Two formulation
tion for these changes is the use of Flexible Assembly types are possible for this problem (Mastor, 1970): type I
Systems (FAS), which include programmable automation attempts to minimize the number of workstations for a
and robots. The use of these flexible systems in a required cycle time, while type II attempts to minimize the
changing environment requires methods for efficient de- cycle time for a given number of stations. The grouping of
sign and re-design of the assembly system in which they work elements into stations must be done without viola-
are used. This assembly system design is focused on issues tion of precedence relations between the work elements.
of assembly system or line configuration, balancing, and Extensive research was done in the 1960s, 1970s and
equipment selection. The use of programmable equip- 1980s, resulting in many effective solution techniques to
ment and automation, along with human operators, ne- the two formulations of the line balancing problem and
cessitates design solutions that deal with optimal use of their extensions. Several comprehensive survey papers of
equipment and consider equipment costs, along with the the many methods have been published, including Baybars
objective of a balanced assembly line. (1986) that surveys the exact (optimal) methods, Talbot
Most of the work related to assembly lines concentrates et al. (1986) that compares and evaluates the heuristic
on the line balancing problem. The traditional approach to methods developed, Erel and Sarin (1998) that present a
the assembly line balancing problem deals with grouping comprehensive review of the assembly line balancing pro-
and assigning non-divisible work elements of the assembly cedures for single model and multi-mixed-model assembly
lines and Ghosh and Gagnon (1989) that present a com-
prehensive review and analysis of the different methods for
* Corresponding author design, balancing and scheduling of assembly systems.
74 Bukchin and Rubinovitz

Most of the attention of the research work on assembly the parallel stations). Udomkesmalee and Daganzo (1989)
line balancing and design was given to the configuration focus on a flexible (in fact, mixed-model) assembly line
of a sequential organization of stations on the line. Rel- with parallel stations. An undesirable effect that may oc-
atively few works explored alternative configurations, cur in such a situation, where task times (for different
and in particular incorporating a certain number of models) are allowed to vary, is that jobs (specific work
parallel equivalent stations into the assembly line. This is orders for assembly of models) may get out of the initially
in spite of several important benefits that can be achieved planned processing sequence. This in turn affects the pre-
by allowing stations to perform tasks in parallel (Buxey, planned supply of parts and materials to the assembly line.
1974; Bard, 1989). One benefit is the potential improve- The work analyzes the effect of variation in process times
ment of balance efficiency (reduction of station idle time), on the job sequences in a given parallel-processing as-
due to the fact that each duplicated station has a cycle sembly line. To solve the problem, the use of either ma-
time which is equal to the original cycle time multiplied terial or job buffers is suggested to re-sequence materials
by the number of identical parallel stations. As a result, a or jobs, as needed, and the work provides analytical
better fit of work element times assigned to the station models to determine the size of the buffers needed. In
within the cycle time is likely. Another benefit (and even a recent work, McMullen and Frazier (1997) suggest a
necessity) of using stations in parallel, is meeting required simple heuristic procedure to solve a mixed-model as-
high production rates (resulting in short cycle times) sembly line problem with stochastic task times when
when some work element times exceed the required cycle paralleling of tasks is permitted. In another work, they use
time. Last but not the least benefit of a line design with a simulated annealing heuristic to solve the same problem
stations in parallel is increased reliability. In a serial line, for a multi-objective combined mainly of the total cost of
a failure of a station stops the entire line, while a failure labor and equipment and the balance efficiency (McMul-
of a parallel station allows continuing the line operation len and Frazier, 1998). Askin and Zhou (1997) propose a
at a reduced production rate. nonlinear integer program model for a mixed-model
Of the few works to suggest algorithms or solve prob- production line with parallel stations. Their model in-
lems related to assembly line configuration with parallel corporates a cost trade-off in the objective function, be-
stations, Buxey (1974) suggested two heuristic algorithms, tween balance quality and equipment duplication in
one based on the ranked positional weight method, and parallel stations. To solve the problem, a heuristic method
the other on random generation of sequences. Both al- is developed, based on a task assignment rule and a station
gorithms incorporate a limit on the number of work ele- paralleling rule. Suer Gursel (1988) suggests a simple
ments per station, and use stations in parallel only for the heuristic for designing parallel assembly lines for high
longer elements. Pinto et al. (1975) developed a branch production rates, with the objective to minimize man-
and bound procedure for selecting tasks to be paralleled, power required. The heuristic is based on a three-phase
with the objective to minimize total cost (labor, including methodology that balances the assembly line, determines
overtime, and equipment duplication costs). Their work parallel stations, and determines parallel lines.
suggests that only certain tasks are duplicated (in different Pinnoi and Wilhelm (1997) propose a unified classifica-
stations), a procedure which is difficult to implement and tion for the design of deterministic assembly systems.
control in practice. In a later work, Pinto et al. (1981) Their classification accommodates tasks with processing
extend their branch and bound algorithm to include times longer than the cycle time, positional constraints on
possible paralleling of two stations. Nanda and Scher task processing, and station configurations of single, par-
(1975) present two models for designing assembly lines allel, collateral or collaborative machines. The purpose of
with overlapping work stations (task paralleling), and this classification is to suggest a unified framework of hi-
demonstrate that this line design is more balanced, and erarchical models for the design of assembly systems,
has increased output over the serial work station design. that would be solved by a technique such as strong cutting
In a sequel work (Nanda and Sher, 1976) they present an plane methods. However, later implementations of this
algorithm which incorporates technological constrains approach (Pinnoi and Wilhelm, 1998) are limited to the
that may limit simultaneous (overlapping) performance of single-product assembly system design problem, attempt-
certain work elements. Their computational procedure is ing optimal cost solution for the assignment of machines to
based on disjunctive graph theory. Sarker and Shantiku- stations at the required cycle time, and without exploring
mar (1983) suggest a general approach that can be applied different station configurations, such as stations in parallel.
for both serial or parallel line balancing. Bard (1989) de- The objective of our work is to provide a comprehen-
velops a dynamic programming algorithm that attempts sive approach for the assembly line design for minimizing
to meet the required cycle time with the minimal total the associated cost. In the studied system several equip-
number of stations, while improving the line efficiency ment types as well as a human operator that are capable
(reducing ‘dead’ time at stations) by using parallel sta- of performing the assembly operations are considered
tions, and selecting the parallel configuration with mini- and paralleling of stations is allowed. In such systems that
mum cost (due to the equipment duplication necessary at may be highly mechanized the associated costs is highly
Weighted approach for assembly line design 75
8
dependent on the number of parallel stations, due to the > 1 if task i is assigned at stage j (to a single sta-
<
equipment duplication. Hence we first analyze the parallel xij =
tion or to several identical parallel stations
station problem, and add weights which are associated >
: opened at this stage),
with the system costs to the objective function. These 0 otherwise;
weights are also considered as control parameters, and by Pi = set of tasks that must precede task i due to
changing these weights, different line configurations, as- technological constraints;
sociated with different types of assembly systems (human Wk = weight (cost factor) for each paralleling situation
intensive, capital intensive), are obtained. Next we show of k identical parallel stations;
that the parallel station problem is analogous to the as- yjk = a binary variable, which equals one when there
sembly line design with equipment selection, discussed in are exactly k parallel stations in stage j;
Bukchin and Tzur (2000). Moreover, we show that the Jmax = the maximal number of stages;
former problem is a special case of the latter, and hence Kmax = the maximum number of stations in parallel;
can be solved by the branch and bound algorithm de- n = the number of tasks.
veloped by Bukchin and Tzur (2000). Eventually the
combined problem (parallel stations and equipment du- The mathematical program (P1) can now be formulated
plication) is addressed and solved by the same procedure. as follows:
In the next section the model is developed, starting
from a basic formulation that leads to a full model and Jmax X
X Kmax

definition of the different weights appropriate for the ðP1Þ min Wk yjk ; ð1Þ
j¼1 k¼1
different paralleling situations. The solution algorithm is
reviewed and adapted to our problem in Section 3. A subject to
comprehensive analysis of the problem variations, and
X
Jmax X
Jmax
examples of the effect of different weights on the problem r  xgr  l  xhl 8g; 8h; subject to g 2 Ph ; ð2Þ
solutions is presented in Section 4. In this section, a de- r¼1 l¼1
tailed analysis of the relation between problem parame-
ters and the balancing improvement that can be achieved X
Jmax
xij ¼ 1 8i; ð3Þ
by paralleling is also conducted. Section 5 extends the j¼1
problem to deal with multiple equipment types option,
and selection of equipment for the parallel stations. X
n X
Kmax

Summary and discussion are presented in Section 6. ti xij  kyjk C 8j; ð4Þ
i¼1 k¼1

X
Kmax
2. Problem description yjk  1 8j; ð5Þ
k¼1
2.1. Model development xij ¼ 0; 1 8i; j;
ð6Þ
In this section, models for the assembly line design yjk ¼ 0; 1 8j; k:
problem are developed. We start with a basic model that
minimizes the number of stations, while allowing stations The objective function (1) minimizes the weighted
in parallel. Further, this model is reformulated to incor- product Wk yjk , of k identical parallel stations at stage j.
porate cost/weight factor for different paralleling situa- Constraint set (2) handles the precedence relationship
tions. The formulation of this model is similar, in part, to between tasks that result from the technological con-
the model suggested by Askin and Zhou (1997). However, straints. Constraint set (3) ensures that each task is per-
in lieu of incorporating just the equipment cost and sta- formed exactly once. The capacity constraint set (4)
tion fixed operating cost per period, the model uses more ensures that the work content assigned to each stage does
general cost/weight factors. These factors can express not exceed the associated capacity. Constraint set (5)
different costs related to the line design, and enable the ensures that if a station is opened in stage j, the number of
setting of unique weight values to solve different design parallel stations at this stage is unique.
problems associated with the particular nature of the The weight factor Wk , represents the cost of a stage with
assembly system environment. k parallel stations, taking into consideration the addi-
tional equipment cost. It determines the outcome of line
design and the allocation of parallel stations. Selecting
2.2. Problem formulation
different values for the weight factor Wk aids in achieving
Notation: different line design objectives while using parallel sta-
tions. When the only objective is to minimize the total
C = cycle time; number of stations, without preferring sequential stations
ti = duration of task i; to parallel stations the weight values are:
76 Bukchin and Rubinovitz

k X
Kmax X
n X
Kmax
Wk ¼ Wk1 ; k ¼ 2::Kmax : ti xijk  kyjk C 8j:
k1
k¼1 i¼1 k¼1

In this case, the weight values are proportional to the


The new constraint can be also written as
capacity content of the station. For example, if W1 ¼ 1
(for a single station), then, W2 ¼ 2 (two parallel stations), Xn

W3 ¼ 3 (three stations in parallel), and so on. This means ti xijk  kyjk C 8j 8k; or alternatively
i¼1
that the cost (weight factor) of n stations in parallel is
exactly n times larger than the cost of a single station. It X
n
ti
xijk  yjk C 8j; 8k: ð4aÞ
also means that creating an additional identical station in k
i¼1
parallel is equivalent to using an additional new sequen-
tial station on the line. This is usually the case in manual The capacity constraint (4a) is analogous to a case where
assembly, where the manually operated cells do not re- there are several equipment alternatives for the assembly
quire special (expensive) tools or equipment. process, and only one of them should be selected for each
When trying to minimize the total number of stations, stage. Each equipment alternative, k, performs task i, in
while trying to minimize the number of parallel stations duration of ti =k. In other words, adding parallel stations
as a secondary objective, the weights should be adjusted is equivalent to replacing the assembly equipment with
as follows: more efficient equipment, which may perform the as-
sembly operations at a faster rate. Hence, the above
k problem can be considered as a special case of the line
Wk ¼ Wk1 þ e; k ¼ 2::Kmax ; ð7Þ
k1 design with equipment selection problem, addressed in
where e is a small constant for breaking tie situations in Bukchin and Tzur (2000). In order to complete the
order to prefer a smaller number of parallel stations. The modification of the model to the multi-equipment selec-
tion problem (P2), constraint sets (2) and (3) from (P1)
constant e can also be interpreted as a small penalty cost
should be adapted as follows:
for using identical parallel stations. This is usually the
case when creating an additional identical station in X
Jmax X
Jmax
parallel requires purchasing (duplicating) some tools or r  xgrk  l  xhlk 8g; 8h; 8k; subject to g 2 Ph ;
equipment that were not required when using an addi- r¼1 l¼1
tional new sequential station on the line. Nevertheless, in ð2aÞ
this case the additional cost is assumed to be much Kmax X
X Jmax
smaller than the cost associated with opening an addi- xijk ¼ 1 8i: ð3aÞ
tional station. k¼1 j¼1
When the cost of equipment duplication is very large,
only essential parallel stations are opened. In this case, Based on the analogy between the two problems, a
the only reason for adding stations in parallel is ‘long’ branch and bound algorithm developed in Bukchin and
tasks, namely, tasks larger than the pre-defined cycle Tzur (2000) for the multi-equipment selection problem is
time, the values of the weights are: adopted to handle the current problem. The algorithm
principles are described in the next section.
Wk ¼ A  Wk1 ; k ¼ 2::Kmax ; ð8Þ
where A is a large constant. In this case, the total number 3. Branch and bound algorithm
of stations is minimized, but only essential parallel sta-
tions are formed, as needed to meet the production rate As was shown in the previous section, (P1) is a special
when tasks with times longer than the required cycle time case of the line balancing problem with equipment al-
are present. This is usually the case when creating an ternatives for minimizing equipment costs. In the general
additional identical station in parallel requires purchasing problem, each task can be performed by at least one
(duplicating) some expensive tools or equipment that equipment type, with a duration that is dependent on the
were not required when using an additional new se- equipment type. In the current model, k stations in par-
quential station on the line. allel are analogous to an equipment type that can per-
form task i in ti =k time units, with an equipment cost of
Wk .
2.3. Analogy to the multi-equipment selection
A frontier search branch and bound algorithm was
Let us define a new binary variable xijk , which equals one developed for the general problem (Bukchin and Tzur,
if task i is assigned at stage j with a configuration of k 2000). Throughout the algorithm, workstations are
parallel stations. Consequently, constraint (4) is modified opened sequentially, equipment types are selected and
as follows: placed in any newly opened workstation, and tasks to be
Weighted approach for assembly line design 77

performed by the selected equipment are assigned to the durations in the second column are equal to half of the
last workstation opened in a given partial solution. The time in the first column. The same rule is applied to a
algorithm ends when all tasks are assigned to worksta- stage with three parallel stations, and the resulting task
tions, and the obtained solution value is no larger than times are presented in the third column. It is important to
the lower bound of all partial solutions. note that the decision on how many parallel stations are
The branch and bound algorithm requires large com- allowed may be made for each task separately. The de-
puter resources in order to solve very large problems, and signer may decide that some of the tasks require expen-
therefore a heuristic is required for most real world sive assembly equipment, and therefore, in order to avoid
problems. According to the branch and bound procedure, equipment duplication, they should be performed in a
the node with the smallest lower bound is extended at single station. In this case, he can leave the task duration
each iteration (jumptracking approach). However, some times in the second and third column empty, allowing
of these nodes have a very small probability of eventually such tasks to be assigned only to a single sequential sta-
providing the optimal solution, and their extension is tion. The problem was optimally solved for three sets of
essential only for proving the optimality of the final so- weights, and the solutions are presented in Table 1.
lution. In the proposed heuristic, the node selection rule is The solution of each problem is presented in three
modified, in order to avoid the extension of such nodes. A columns: the first includes the number of stations with
heuristic control parameter is defined, which determines and without paralleling, the second consists of the task
how many nodes of the tree may be skipped. The detailed assignments to stations, and the third presents the accu-
algorithm and lower bounds developments are presented mulated assembly time assigned to each station. Case C1
in Bukchin and Tzur (2000). includes a set of weights where W1
W2
W3 (the values
of 100, 2000 and 30 000 were chosen here but any other
values that satisfy the above condition could be used as
well), which imposes a solution without stations in par-
4. Analysis of problem variations
allel, resulting in a configuration of 14 stations. Case C2
represents indifference between parallel and sequential
4.1. Balancing improvement and design trade-offs
stations, by weighting each stage proportionally to its
The introduction of identical parallel stations to the line capacity, Wk ¼ ðk=k  1ÞWk1 (using the weight values
design leads to reduction of idle time and allows to im- 100, 200 and 300). In this case the objective is to minimize
prove the line balance. This design alternative can be very the capacity (total number of stations), while being in-
effective when high production rates are required from an different to whether sequential or parallel stations are
assembly. To meet the high production rates, the line used. The solution obtained includes 13 stations, most of
must be balanced for short cycle times, where obtaining a them in parallel. This solution is an improvement to the
good balance is more difficult. Using parallel stations solution of C1, but some of the parallel stations may not
may provide good balancing solutions in such an envi- be essential. In case C3, we use an objective function with
ronment, and is more cost effective than constructing a a small penalty for the paralleling situation, in favor of
few separate lines with longer cycle times, an option solutions with sequential stations (the penalty used is
which is too expensive due to tooling and machines du- e ¼ 1, resulting in weight values of 100, 201, 302.5). In-
plication. The benefits of using parallel stations to reduce deed, the resulting solution includes 13 stations, as in the
station idle times have been pointed out by numerous solution of C2, but with fewer paralleling situations (only
researchers (for example Buxey (1974)). Others indicated one parallel station instead of six parallel stations in C2).
additional possible trade-offs and benefits of this design
alternative (for example Bard (1989); Askin and Zhou
4.2. Long task constraints
(1997)).
We now present a small example, which illustrates the Parallel stations are essential where there are tasks with a
advantages of using parallel stations in line design, and duration longer than the required cycle time. Consider for
demonstrates the use of weights in the objective function example a final assembly line of televisions, where in one
in order to control the results. The objective is to mini- of the last stages the TV set should be operated for a
mize the number of stations for a given cycle time of 60 couple of hours as a testing phase. It is clear that if a
time units. Figure 1 presents the input data of a problem single sequential station will be used at this stage, the
with 20 tasks, where at most three stations in parallel are testing phase will be the bottleneck of the line, and will
allowed. The precedence diagram of the product to be cause low throughput and high idle time in other stations.
assembled is presented on the right-hand side, and a table In the actual TV assembly environment, we will probably
of task durations is on the left-hand side. find at this stage many TV sets being tested in parallel,
A stage with two stations in parallel is considered here implementing a situation of many stations in parallel.
as a single station, where it takes half of the time to Figure 2 shows such a situation, where stage j is the
perform each task assigned to this stage. Hence, the task testing stage, and the line cycle time is C time units.
78 Bukchin and Rubinovitz

Fig. 1. Input data of the paralleling example.

The capacity of stage j is m times C, where there are m allel stations to meet the required cycle time. Yet, by
stations in parallel. If, for example, the line daily applying the different weight factors to control design
throughput is 200 TV sets in an 8 hours shift (the cycle objectives, three different objective functions can be
time is equal to 2.4 minutes), and the testing time required achieved, as presented by the cases in Table 2.
for each TV is 2 hours, then 50 stations of type j are In case C4, the problem objective enforces only essen-
required to prevent blockage in stage j and meet the re- tial paralleling, as required to meet the cycle time by the
quired production rate. long tasks. As a result, the optimal solution includes only
By making small changes in the input data of the two parallel stations, one in stage 9 that performs task 10
previous example, we can illustrate the paralleling situa- (along with task 12), and the other in stage 11 that per-
tion with tasks having a duration longer than the required forms task 17 (along with task 18). A total of 15 stations
cycle time. Let the duration of task 10 be 72 time units are established in the solution of case C4. Case C5 is
and the duration of task 17 be 90 time units. Both tasks equivalent to case C2, where the objective is to minimize
are now longer than the cycle time (60 time units). In this capacity subject to a given cycle time, a configuration of
modified problem, a feasible solution must include par- up to three stations in parallel is allowed, and there is
Weighted approach for assembly line design 79
Table 1. Optimal solutions of the example problems C1–C3
Stage C1 C2 C3
k k
W1
W2
W3 Wk ¼ k1 Wk1 , k ¼ 2, 3 Wk ¼ k1 Wk1 þ e, k ¼ 2, 3

Parallel Tasks Station Parallel Tasks Station Parallel Tasks Station


stations time stations time stations time
1 1 1,4,7 42 2 2,5,8 120 1 2 60
2 1 2 60 2 1,3,9 108 1 5,8 60
3 1 5,8 60 1 6 48 1 1,9 54
4 1 9 42 1 11 60 1 3 54
5 1 3 54 2 4,7,10,13 114 1 11 60
6 1 6 48 3 12,14,15,16,17,18 180 1 6,14 60
7 1 11 60 2 19,20 102 1 4,16 60
8 1 14,16 48 2 7,10,12 114
9 1 10 60 1 13,15 60
10 1 12 48 1 17,18 48
11 1 13,15 60 1 19 48
12 1 17,18 48 1 20 54
13 1 19 48
14 1 20 54
14 stations (no paralleling) 13 stations (six parallel1) 13 station (one parallel1)
1
Note: the number of parallel stations counts additional identical stations used in parallel with the ‘base’ station (e.g., one parallel station indicates
the existence of two identical stations at the stage).

solving this model is capable of handling a wide range of


problems.

4.3. The effect of the problem parameters on the balancing


improvement
In this section we examine how different problem para-
meters influence the balancing improvement that can be
achieved by using parallel stations. The purpose of this
examination is to check the potential for improvement of
Fig. 2. Parallel stations for a long task situation. the line efficiency, by using parallel stations, for a wide
range of problems with different characteristics. All ex-
indifference between stations in parallel and sequential periments were performed for problems with 20 tasks,
stations. The solution for this case includes five sets of and four factors were defined for the experimentation:
parallel stations and 14 stations in total, one station less
than in the solution for case C4. The solution of case C6, 1. F-ratio: The F-ratio is a measure for the flexibility in
in which a small penalty is applied for using parallel creating different assembly sequences for a K elements
stations (as in case C3), also has a total of 14 stations as assembly task. It was described by Dar-El (1973), and
in C5, but there are only two sets of parallel stations, with can be defined as follows:
three identical stations in each set. The two sets of parallel Let pij be an element of a precedence matrix P, such
stations resolve the problem posed by tasks 10 and 17, that: n
that have times longer than the cycle time. pij ¼ 1 if task i precedes task j,
These examples illustrate the way in which the weight 0 otherwise.
(cost) parameters can be used in order to change the
design objective of the problem to be solved. This use Then, F-ratio ¼ 2Z=nðn  1Þ, where Z is the number
provides the line designer with the flexibility to review of zeroes in P, and n is the number of assembly tasks.
solutions for different design objectives, for a wide range The F-ratio value is therefore between zero, when
of problems involving parallel stations. The different de- there are no precedence constraints between tasks (any
sign objectives are all represented by the same model (in sequence is feasible), and one, when only a single as-
which only the weight parameters are modified), and as a sembly sequence is feasible. Assembly tasks are often
result the branch and bound algorithm designed for characterized by relatively low F-ratios. Hence, pre-
80 Bukchin and Rubinovitz
Table 2. Optimal solutions of the example problems C4–C6
Stage C4 C5 C6
k k
W1
W2
W3 Wk ¼ k1 Wk1 , k ¼ 2, 3 Wk ¼ k1 Wk1 þ e, k ¼ 2, 3

Parallel Tasks Station Parallel Tasks Station Parallel Tasks Station


stations time stations time stations time
1 1 1,4,7 42 2 2,5,8 120 1 2 60
2 1 2 60 2 1,3,9 108 1 5,8 60
3 1 5,8 60 1 11 60 1 1,9 54
4 1 9 42 3 4,6,7,10,13 174 1 3 54
5 1 3 54 1 12,14 60 1 11 60
6 1 6 48 3 15,16,17,18 180 3 4,6,7,10,13 174
7 1 11 60 2 19,20 102 1 12,14 60
8 1 14,16 48 3 15,16,17,18 180
9 2 10,12 120 1 19 48
10 1 13,15 48 1 20 54
11 2 17,18 108
12 1 19 48
13 1 20 54
15 stations (two parallel) 14 stations (seven parallel1) 14 stations (four parallel1)
1
Note: the number of parallel stations counts additional identical stations used in parallel with the ‘base’ station (e.g., one parallel station indicates
the existence of two identical stations at the stage).

cedence diagrams with F-ratios of 0.1, 0.3 and 0.5 were For each experiment, three different precedence dia-
generated in this study. grams were generated, each with two instances of task
2. Average number of tasks per station (ATS): This para- durations (total of six replications). In the first set of
meter is equal to the ratio between the number of tasks experiments, the effect of the first three parameters on
in the assembly and the minimal number of stations the balancing improvement is examined. The dependent
required to meet the production rate. (The number of variable, the balancing improvement, is equal to the ratio
stations can be calculated by the ratio of total as- between the number of stations obtained with paralleling
sembly time required and the cycle time.) The pa- and the number of stations obtained with no parallel
rameter was set to two and four tasks per station. stations. Since an optimal solution was obtained for both
3. Variability of task duration (VTD): The duration of cases, this value is always smaller than or equal to one.
every task was generated from a uniform distribution. Table 3 shows the ANOVA results, where the rows in
We examined a distribution with a small variance, bold type denote the significant factors (p-value smaller
U ð0:8l, 1:2lÞ, and a distribution with a high variance than 0.05). We can see that the effect of the F-ratio and
U ð0:4l, 1:6lÞ, where l is the expected value of the task ATS is quite significant, as well as the interaction of the
duration. two. As can be expected, the results show that parallel
4. Maximal number of stations in parallel (MSP): Each stations are mostly desired in situations where a good
problem was solved for a maximal number of two, balance is difficult to achieve (low F-ratio and small cycle
three and four parallel stations. time).

Table 3. ANOVA table of the balancing improvement


General MANOVA Summary of all effects; 1-F-ratio, 2-ATS, 3-VTD
Effect df effect MS effect df error MS error F-ratio p-value
1 2 0.015 750 204 0.003 749 4.2013 0.016 292
2 1 0.595 439 204 0.003 749 158.8323 0.000 000
3 1 0.000 329 204 0.003 749 0.0879 0.767 198
12 2 0.022 669 204 0.003 749 6.0469 0.002 810
13 2 0.004 240 204 0.003 749 1.1309 0.324 755
23 1 0.002 893 204 0.003 749 0.7718 0.380 686
123 2 0.001 009 204 0.003 749 0.2692 0.764 257
Weighted approach for assembly line design 81

Looking at Fig. 3, which demonstrates the F-Ratio- contribution of the paralleling to the balance solution
ATS interaction, we can see that the effect of the F-ratio decreases with the number of parallel stations. In fact, we
seems much more significant for small ATS. No clear can say that in this case the significant contribution is
effect is identified for large ATS (in particular we can see obtained for adding the first parallel station (average
that at the upper plot in Fig. 3). The reason for this may improvement of 8.4%), and there is very little contribu-
be that for relatively large ATS values good balance may tion beyond that (additional improvement of 0.95 and
be obtained without allowing parallel stations, and the 0.26% for adding the second and third parallel stations
improvement due to paralleling is very small (less than respectively). We can assume that those values are de-
6% in our experiments). In this small range the effect of pendent on the problem parameters, and that another
the F-ratio is minor. In small ATS, on the other hand, problem set may have resulted in different values of the
good balance is difficult to achieve, and the effect of the improvements. However, the trend is quite intuitive, and
F-ratio is much clearer. we can assume that this behavior of the balancing im-
We also examined the effect of the limitation on the provement will remain the same for almost any set of
number of parallel stations on the balancing improve- problems. We then may conclude that when the reason
ment. Seventy-two problems were solved for each limi- for paralleling is balancing improvement we should start
tation, giving a total of 288 problems. The graph in Fig. 4 solving the problem for a small value of the maximal
shows the balancing improvement as a function of the number of parallel stations, and then increase this value
maximal number of parallel stations allowed. until no significant improvement is obtained.
Each point of the graph represents an average value of
the 72 problems solved. We can see that the marginal
5. Modification to the multi-equipment line balancing
problem with paralleling

The equivalence between the line balancing problem with


paralleling and a special case of the line balancing prob-
lem with equipment selection was discussed in Section 2.
In this section the integration of the two problems is
addressed, and an example is given.
Assume that there are m optional equipment types to
be assigned to stations. In each stage we can assign either
a single piece of equipment or several identical equip-
ment units. The problem is to select the required equip-
ment type for each stage and the number of parallel
stations containing this equipment, and to assign the
tasks to stations. The objective is to minimize the total
equipment costs. A few notations should be defined for
Fig. 3. Balancing improvement as a function of the F-ratio and the formulation of the multi-equipment problem with
the ATS. paralleling:

Wkm = the cost of k parallel stations with identical


equipment type m;
tim = the duration of task i when performed by equip-
ment type m;
8
< 1 if there are exactly k parallel stations in
yjkm = stage j with equipment type m,
:
0 otherwise;
8
< 1 if task i is performed by equipment type
xijm = at stage j,
:
0 otherwise;
M = the number of different equipment types.

The new formulation of the problem (P3) is then:


Jmax X
X Kmax X
M

Fig. 4. Balancing improvement as a function of the paralleling ðP3Þ min Wkm yjkm ; ð1aÞ
limitation. j¼1 k¼1 m¼1
82 Bukchin and Rubinovitz
subject to performing a large number of assembly tasks (all tasks
X
Jmax X
M X
Jmax X
M in this example). A special case of this equipment type
j  xgjm  l  xhlm 8g; 8h; subject to g 2 Ph ; may be a human being, which will be considered from
j¼1 m¼1 l¼1 m¼1 now on without loss of generality.
ð2bÞ 2. Equipment type E1 : A fast assembly piece of equip-
ment characterized by short tasks’ duration.
X
Jmax X
M
xijm ¼ 1 8i; ð3bÞ 3. Equipment type E2 : The least expensive equipment.
j¼1 m¼1
The maximal number of parallel station of equipment
X
n X
M Kmax X
X M
types M, E1 , and E2 in each stage was set to three, two,
tim xijm  kyjkm C 8j; ð4bÞ
and two respectively. The input data of the task duration,
i¼1 m¼1 k¼1 m¼1
which is presented in Table 4 also refers to the maximal
X
Kmax X
M
number of parallel stations of each equipment type.
yjkm  1 8j; ð5aÞ Columns ‘M1 ’, ‘M2 ’ and ‘M3 ’ refer to task durations in a
k¼1 m¼1
single station, two and three parallel stations of the hu-
xijm ¼ 0; 1 8i; 8j; man worker; columns ‘E11 ’ and ‘E21 ’ refer to a single sta-
yjkm ¼ 0; 1 8j; 8k; 8m: tion and two parallel stations of equipment type E1 , and
columns ‘E12 ’ and ‘E22 ’ refer to a single station and two
parallel stations of equipment type E2 . In this example,
The formulation is a modification of original formu- every task may be performed in parallel stations where
lation of the assembly line balancing with paralleling the maximal number of parallel stations is limited by the
(P1). The objective function (1a) expresses the total equipment. In general, we can limit the maximal number
equipment cost of the assembly system. Constraint sets of parallel stations for each task independently by en-
(2b), (3b), (4b) and (5a) replace constraint sets (2), (3), (4) tering empty cells in the appropriate places in the dura-
and (5) respectively. tion matrix. We can see that the duration of three out of
The line balancing problem with multi-equipment and the 30 tasks (tasks 7, 18 and 28) is larger than the cycle
paralleling can be solved by the branch and bound al- time for all optional equipment types, namely, all feasible
gorithm, which was presented in Section 3. The weights in configurations should include parallel stations.
the objective function in addition of being control pa- The costs of the human worker and the two equipment
rameters have now also a practical meaning of the types are $100 000, $100 000 and $60 000 respectively. (It
equipment costs. In order to demonstrate the identical should be noted that although the cost of the human
solving procedure of both problem types, the following worker is different than the cost of automated assembly
example is presented. equipment, both can be translated into a total cost over
An assembly line has to be balanced for a product the life cycle of the assembly system.)
containing 30 assembly tasks while minimizing the The problem was solved according to case C6 (see
equipment costs. The precedence diagram of the product Table 2), where tasks longer than the cycle time exist and
is presented in Fig. 5. The line cycle time is 250 time units, the weights’ values are set according to Equation (6). In
as derived from the required production rate. this case the minimal equipment costs is obtained in the
Three equipment types are available for each station: optimal solution with a minimal number of parallel sta-
tions.
1. Equipment type M: A highly flexible piece of equip- An optimal solution for the problem was obtained, and
ment, namely, a piece of equipment that is capable of the minimal cost configuration is shown in Fig. 6. The

Fig. 5. Precedence diagram.


Weighted approach for assembly line design 83
Table 4. Task duration
M1 M2 M3 E11 E21 E12 E22
Task 1 54 27 18 64 32
Task 2 144 72 48 70 35
Task 3 138 69 46 160 80
Task 4 144 72 48 86 43 142 71
Task 5 24 12 8
Task 6 102 51 34
Task 7 300 150 100 380 190
Task 8 114 57 38 58 29 108 54
Task 9 36 18 12 14 7
Task 10 114 57 38 70 35 110 55
Task 11 84 42 28
Task 12 72 36 24 44 22
Task 13 72 36 24 28 14 70 35
Task 14 24 12 8 12 6
Task 15 90 45 30
Task 16 72 36 24 42 21
Task 17 60 30 20 68 34
Task 18 360 180 120
Task 19 36 18 12 42 21
Task 20 114 57 38 64 32
Task 21 96 48 32 86 43
Task 22 24 12 8 14 7
Task 23 30 15 10 30 15
Task 24 78 39 26 48 24
Task 25 12 6 4 12 6
Task 26 270 135 90 290 145
Task 27 78 39 26 50 25 94 47
Task 28 90 45 30
Task 29 150 75 50 154 77
Task 30 36 18 12 22 11

the fifth column where in stages with parallel stations the


effective station time, defined as the division of the ac-
cumulated station time by the number of station, appears
in the parenthesis. The equipment cost of each stage is
shown in the sixth column and the total equipment costs,
Fig. 6. Optimal assembly line configuration.
$1100 000 is presented at the bottom.

assembly line contained 10 stages; seven stages include a


single station and thee stages include two parallel sta- 6. Summary and conclusions
tions. All optional equipment is used, where in four stages
(six stations) a manual worker is assigned, in four stages The assembly line design with station paralleling and
(five stations) the less expensive equipment is assigned equipment selection has been addressed in this paper.
and in other two stages (three stations) the fast equipment After presenting the basic formulation of the station
is selected. paralleling, a modified model that takes into consider-
The quantitative data of the optimal solution are pre- ation costs associated with the trade-off between serial
sented in Table 5. The first three columns show the stage and parallel stations is presented. Changing the costs
number, the selected equipment for each stage and the parameters enables us to address a typical assembly line
number of stations in each stage. The assembly tasks design problem, in which the assembly system may be
assigned to each stage are presented in the fourth column, labor or equipment intensive and where tasks may be
and not surprisingly we can see that all three tasks with a larger than the cycle time. An analogy between the par-
duration larger than the cycle time (tasks 7, 18 and 26) are allel station problem and the assembly system design with
assigned in stages with two parallel stations (stages 4, 6 equipment selection is shown, where the former problem
and 8). The accumulated time in each station appears in is found to be a special case of the latter. The advantage
84 Bukchin and Rubinovitz
Table 5. Optimal solution
Stage Equipment type Number of stations Tasks Acc. time Cost (· $000)
1 E2 1 1,2 224 60
2 M 1 5,6,9 234 100
3 E1 1 2,4,10 226 100
4 E2 2 7,8 488 (244) 120
5 M 1 11,15,17 234 100
6 M 2 14,18,20 498 (249) 200
7 E2 1 13,21,23 186 60
8 M 2 16,19,22,24,25,26 492 (246) 200
9 E2 1 27,29 248 60
10 M 1 28,30 126 100
Total cost: $1100
(· $000)

of this observation is two-fold; a solution approach for mixed-model environment, stochastic assembly times and
the equipment selection problem can be used for solving machine breakdown.
the parallel station problem, and moreover, the two
problems can be combined and be efficiently solved. The Acknowledgement
balancing improvement associated with applying parallel
stations is examined through a wide range of experiments. This work was performed in part while Dr. Bukchin was
Results show that the flexibility in the order of the as- on a sabbatical at the Grado Department of Industrial
sembly operations (expressed by the F-ratio measure) and and Systems Engineering, Virginia Tech, and Dr. Rubi-
the cycle time (expressed by the ATS measure) have sig- novitz was on leave at the Department of Industrial En-
nificant effects on the balancing improvement. We can gineering, Tel Aviv University. The authors wish to thank
state that parallel stations are especially needed when a both departments for the hospitality during their stay.
good balance is difficult to obtain due to small cycle times
or low flexibility in the assembly process. Another inter-
esting observation concerns the relationship between the References
paralleling limitation (maximal number of stations in
parallel in a single stage) and the balancing improvement. Askin, R.G. and Zhou, M. (1997) A parallel station heuristic for the
mixed-model production line balancing problem. International
Results show that the first station in parallel contributes Journal of Production Research, 35(11), 3095–3105.
the main improvement (8.4% on average) while the Bard, J.F. (1989) Assembly line balancing with parallel workstations
contribution of additional parallel stations is quite small and dead time. International Journal of Production Research,
(0.95 and 0.26% are associated with the third and fourth 27(6), 1005–1018.
parallel stations respectively). The importance of this Baybars, I. (1986) A survey of exact algorithms for the simple assem-
bly line balancing problem. Management Science, 32(8), 909–932.
observation results from the fact that the branch and Bukchin, J. and Tzur, M. (2000) Design of flexible assembly line to
bound algorithm run time is highly dependent on this minimize equipment cost. IIE Transactions, 32(7), 585–598.
parameter. For practical problems, based on this result of Buxey, G.M. (1974) Assembly line balancing with multiple stations.
diminishing returns one may experiment with low values Management Science, 20(6), 1010–1021.
(of one or two) stations in parallel. In cases where many Dar-El, E.M. (1973) MALB - A heuristic technique for balancing large
single-model assembly lines. AIIE Transactions, 5(4), 343–356.
parallel stations are needed due to long task times, such Erel, E. and Sarin, S.C. (1998) A survey of assembly line balancing
as in the testing stage of the TV line example, it will be procedures. Production Planning and Control, 9(5), 414–434.
reasonable to solve this part of the problem separately, Ghosh, S. and Gagnon, R.J. (1989) A comprehensive literature review
without combining additional tasks with short task times and analysis of the design, balancing and scheduling of assembly
into such stations. systems. International Journal of Production Research, 27(4), 637–
670.
The combined problem, which incorporates the station Mastor, A. (1970) An experimental investigation and comparative
paralleling with equipment selection problem is discussed evaluation of production line balancing techniques. Management
in the paper, an ILP formulation is developed and an Science, 16(22), 728–745.
optimal solution of an example problem is presented. We McMullen, P.R. and Frazier, G.V. (1997) Heuristic for solving mixed-
believe that the design approach presented here is suitable model line balancing problems with stochastic task durations and
parallel stations. International Journal of Production Economics,
for handling various practical problems. Nevertheless, 51(3), 177–190.
much further research should be done in combining more McMullen, P.R. and Frazier, G.V. (1998) Using simulated annealing to
aspects of practical design problem, such as the multi- solve a multiobjective assembly line balancing problem with par-
Weighted approach for assembly line design 85
allel workstations. International Journal of Production Research, Biographies
36(10), 2717–2741.
Nanda, R. and Scher, J.M. (1975) Assembly lines with overlapping Joseph Bukchin is a faculty member of the Department of Industrial
work stations. AIIE Transactions, 7(3), 311–318. Engineering at Tel Aviv University. He received his B.Sc., M.Sc. and
Nanda, R. and Scher, J.M. (1976) Nonparallelability constraints in D.Sc. degrees in Industrial Engineering from the Technion, Israel In-
assembly lines with overlapping work stations. AIIE Transactions, stitute of Technology. He is a member of the IIE and INFORMS. He
8(3), 343–349. was a visiting professor at the Grado Department of Industrial and
Pinnoi, A. and Wilhelm, W.E. (1997) Family of hierarchical models for Systems Engineering at Virginia Tech. His main research interests are
the design of deterministic assembly systems. International Journal in the areas of assembly systems design, assembly line balancing, fa-
of Production Research, 35(1), 253–280. cility design, design of cellular manufacturing systems, operational
Pinnoi, A. and Wilhelm, W.E. (1998) Assembly system design: a scheduling as well as work station design with respect to cognitive and
branch and cut approach Management Science, 44(1), 103–118. physical aspects of the human operator.
Pinto, P., Dannenbring, D.G. and Khumawala, B.M. (1975) A branch
and bound algorithm for assembly line balancing with paralleling. Jacob Rubinovitz is a Senior Lecturer at the Faculty of Industrial
International Journal of Production Research, 13(2), 183–196. Engineering and Management of the Technion. He holds a B.Sc and
Pinto, P., Dannenbring, D.G. and Khumawala, B.M. (1981) Branch M.Sc. in Industrial Engineering from the Technion, and a Ph.D. in
and bound heuristic procedures for assembly line balancing with Industrial Engineering from The Pennsylvania State University. He is a
paralleling of stations. International Journal of Production Re- senior member of the Institute of Industrial Engineers (IIE), the Society
search, 19(4), 565–576. of Manufacturing Engineers (SME), and the Israel Society for Com-
Suer Gursel, A. (1998) Designing parallel assembly lines. Computers puter Aided Design and Manufacturing. He established the Robotics
and Industrial Engineering, 35(3–4), 467–470. and Computer Integrated Manufacturing Lab at the Faculty of In-
Sarker, B.R. and Shantikumar, J.G. (1983) Generalized approach for dustrial Engineering and Management at the Technion, and served as
serial or parallel line balancing. International Journal of Produc- its head between 1988–1995. He has held visiting appointments at the
tion Research, 21(1), 109–133. Industrial Engineering departments of the University of Pittsburgh
Talbot, F.B., Patterson, J.H. and Gehrlein, W.V. (1986) A comparative (1995–1996) and the Tel Aviv University (2001–2002). The research
evaluation of heuristic line balancing techniques. Management focus of Dr. Rubinovitz is optimization of design and operation of
Science, 32(4), 430–454. flexible computer-integrated manufacturing systems and assembly
Udomkesmalee, N. and Daganzo, C.F. (1989) Impact of parallel systems.
processing on job sequences in flexible assembly systems. Inter-
national Journal of Production Research, 27(1), 73–89. Contributed by the Facilities Layout and Material Handling Department

You might also like