222 ..

2547

Heuristic-Based Optimization Models for Assembly Line Balancing in Garment Industry

Natayanee KETMATEEKAROON Jaramporn HASSAMONTR Department of Production Engineering,King Mongkut Institute of Technology North Bangkok, 1518 Pibulsongkram Rd., Bangsue, Bangkok, 10800 Thailand tel. 66-02-913-2500, fax 66-02-587-0029

Abstract.
While assembly line balancing problem has been investigated by researchers for decades, its solution techniques have been used sparingly in practice. In garment industry, for example, the number of workers, their skills and the number of operations and their constraints involved often lead to too large integer linear programming models to solve by a commercial ILP solver. In this research, a heuristic method is developed based on greedy strategy to solve the assembly line balancing problem where there is more than one machine per station and there may be parallel stations. The method is directly implemented on a commercial ILP solver. Case studies from a nylon bag manufacturing company are used to illustrate its capabilities. The parameters influencing line balancing performance, such as target cycle time, the number of workers and parallel station decision criteria are investigated and discussed. The method is currently tested by the nylon bag manufacturing company to perform line balancing for a team of workers.

1. Introduction
Garment industry is a very labor-intensive industry. Its productivity is predominantly driven by workers' sewing skills. Typically it takes 1-2 months to train a worker extensively from not knowing anything about sewing to the level that he/she can work efficiently in the production line. Unfortunately very few companies are able to afford such training in practice. Instead they employ on-the-job training approach and have sewing supervisor teaches necessary skills to new workers. Consequently, workers are trained by performing the operations on a large number of actual workpieces. The sewing supervisor is usually the one to assign operations to workers so as to balance the overall line cycle time. As the industry now moves toward fashionable products, it must adjust their operations to work on much smaller lot sizes. This unavoidably leads to shorter preparation time for each production lot. This research focuses on developing a practical methodology to assist sewing supervisor perform effective line balancing. In garment industry a product is manufactured through a series of operations. Each operation must be performed on a machine(sewing machine) with a specific machine setting, i.e. yarn color, machine attachment. Manufacturing a product always requires different types of sewing machines and different yarn colors, making it difficult to assign a worker to perform operations on just a single machine. There is a maximum number of machines that each worker can use for a particular product. Figure 1, for examle,

 ..2547 223

denotes the line configuration of the problem considered in this research of which each worker can use at most three different machines. For the ease of working, identical machines of different settings will be treated as different machines. The worker therefore needs not to adjust the setting every time he/she performs an operation. The decision making scheme developed in this research is to assign operations and their corresponding machines to a team of workers so that the line cycle time is minimized. The optimization model takes into account workers skill levels as well as the constraint on the number of machines at each station(worker). Each operation can be classified as a skill type. Each worker in the team is evaluated for all these skills on standardized tests. The ratings based on time required to perform such skill to meet acceptable quality level is given to each worker for each skill. This rating system allows for incompetent workers who cannot perform certain skills as well. The solution approach is divided into two phases. In the first phase, a multi-stage integer programming model is developed to assign operations, corresponding machines and their settings to stations considering standard operation times, station by station. Parallel stations are allowed so as to improve overall line cycle to as well as to use the required number of workers. Then in the second phase, another integer programming model is used to assign workers to stations based on their aptitudes to minimize the overall line cycle time.

Fig. 1. Assembly line consists of a team of workers performing tasks. Each task must be performed using operators skill and a machine Both phases are implemented on a spreadsheet software interfacing with a commercial integer programming solver software , LINGO. Case studies are used to illustrate the proposed solution procedures capability. The effects of parallel station criteria and the number of workers on line balancing performance are discussed. It is currently used by a company manufacturing nylon bags.

224 ..2547

2. Literature Review
Assembly line balancing problems have been investigated by researchers well over 4 decades. The problem was first introduced as an integer programming problem. [1] and [3] had proposed well-known heuristic methods to group operations into stations. The linear programming model was introduced by [2]. The dynamic programming formulation for assembly line balancing problems was proposed and solved by [4]. [5] proposed the formulation in zero-one integer programming and solved the problem using Fibonacci search. An improvement on branch-and-bound method to solve the assembly line balancing can be found in [6]. [8] applied a backtracking technique to the network precedence diagram. Several researchers investigated the problems of assembly line design. [16] and [17] provided good general overview on the topic. [13] focused on designing assembly line for modular products. While much work has been done in solving the assembly line balancing problem effectively, the techniques are difficult to apply in practice. For example, the number of workers available must be taken into account in the assembly line balancing. [7] and [12] addressed the issues of assigning more than one worker to the station. [11] considered the problem of assigning tasks to a fixed number of stations using assembly line mapping techniques. [9] and [14] also tried to determine the minimum number of workers for the assembly line. Other aspects in assembly line balancing research include stochastic task times as considered by [15]. The actual applications of assembly line balancing have been limited, except in automotive industry. [10] reported an interesting application of line balancing in home appliance manufacturing. In the paper, several aspects not generally considered in research community were addressed. For example, some of the tasks are considered float. They can be assigned to certain fixed tasks only. Some tasks may be incompatible with other tasks as well. The work presented in this paper is along the same line as the work in [10] in that it focuses on machine or tooling restrictions. It is, however, inspired by the sewing operations in nylon bag manufacturing industry where human workers skill plays an important role in determining the operation time. Furthermore, the number of workers to be used is fixed.

3. Problem Formulation
Similar to garment industry, bags are made by sewing operations. However, the sewing operations for nylon bags require several types of sewing machines, making the skill required to perform individual operation varies. The following are assumptions associated with the problem formulation considered in this research. Skills to perform an operation can be classified into a limited number of categories. For example, in the case study considered, there are 49 skill types. A operation is classified as one skill type only. Operator's aptitude toward each skill is evaluated and known in advance. There are 4 levels of skill aptitude. An operator receives an A rating for a skill if he/she can complete standardized task of that skill with at least 20% faster than the specified standard time. The B rating is given to an operator who can complete the task of that skill within 20% of the standard time. The C rating is given to an operator who

 ..2547 225

takes at least 20% longer than the standard time to complete the task of that particular skill. If the worker cannot or never perform the task of that particular skill, a zero rating will be given. In practice, such zero rating implies that the worker may receive on-the-job training to perform the task and, therefore, takes significantly longer time than the C-rating worker. In this research, it is assumed that the zero rating implies that the worker takes twice as long as the standard time to complete the task. There are several types of sewing machines. In the case considered, there are 6 types, each will have only one setting. For the sewing machine, the setting represents yarn colors, attachment for the sewing task. A typical product would have about three different yarn colors. Thus, with the combination of both sewing machines and yarn color, there will be up to 18 different kinds of machines. Machine 1 is of type 1 with color setting 1, machine 2 of type 1 with color setting 2, machine 3 of type 1 with color setting 3, and so on. Each operator can use at most three machines in a station. This constraint is due to the U-shaped line configuration as shown in Fig.1. Target cycle time, Tc, is pre-specified. All product information, such as operation details, its corresponding skill type, standard time, required machine and immediately preceding operations, is given. The formulation of assembly line balancing problem can be explained as follows. First, the time required by each worker to perform each operation is estimated deterministically. Let gik be an aptitude rating of worker i on skill k and fjk a binary parameter to indicate whether operation j is designated as skill type k. Note that the summation of fik for all k's would be unity since each operation can be classified by one skill type only. The time factor for worker i to perform skill k, hik, can be calculated from
0.8 if gik = A 1 if g = B ik . hik = 1.2 if gik = C 2 if gik = 0

(1)

Then the estimated time for worker i to perform operation j, tij, can be calculated from t ij = w j hik f jk ,
k

(2)

where wj is the standard time to perform operation j. The assembly line balancing considered here is more complicated than those discussed in the literature since more decision variables are necessary to ensure that constraints on workers and machines are met. Fig. 2 defines all binary decision variables necessary. For example, binary variables xij are to indicate whether worker i is assigned to perform operation j. yjm and zms are defined similarly. Only when an operation j is assigned to station s, js will be set to one. Otherwise it is set to zero. Similarly, when a worker i is assigned to machine m, im will be set to one. Dummy variables js are used to ensure appropriate assignments between operation j and station s whereas im to ensure appropriate assignments between worker i and

226 ..2547 machine m respectively. Furthermore, let aij be a binary parameter to indicate whether worker i can perform operation j, that is aij = 1 if tij < 2wj and 0 otherwise. All aijs can be set to unity if one insists that each worker can perform any task. And also let bjm be a binary parameter to indicate whether operation j can be performed on machine m.

3.1 Traditional Formulation

The assembly line balancing for these sewing operations can be represented as follows. minimize max i
i

(3) (4) (5) (6) (7) (8) (9)

subject to
a ij x ij 1, j
i

b jm y jm = 1, j
m

i = tij xij , i
j

js + 2 js = max (y jm + zms ) , j , s
m

s =1

s j1s s j2 s , ( j1 << j2 )
s =1

im 3, i
m

im + 2 im = max xij + 2 y jm , i ,m
m

(10) (11)

xij, yjm, zms = 0 or 1, i, j, m, s im, im, js, js = 0 or 1, i, j, m, s

The objective function (3) is to minimize the maximum time required by any worker i, i.e. line cycle time. Constraint (4) states that all operation j's must be performed by at least one worker while constraint (5) that all operation j's must be assigned to a machine. Constraint (6) is to provide cycle time calculation for each worker i or station. Constraints (7) and (8) ensure that precedence constraints are met. The notation j1 << j2 means that operation j1 immediately precedes operation j2. Note that when operation j is assigned to station s, js is set to unity, otherwise it is set to zero. Constraints (9) and (10) require that each worker can operate on at most 3 machines. While it is possible to formulate the problem differently, the number of constraints and variables involved still prohibits actual application in industrial environment. For instance, in actual manufacturing setting in the case study in section 4, the problem must be solved every day for several teams of workers. The number of workers in each team ranges from 11 to 15. A typical product requires up to 100 operations. The above formulation involves 7110 binary variables and roughly 2100 constraints, assuming that product's precedence diagram is of a tree structure. It generally takes hours to solve even a smaller problem using commercially available optimization software.

 ..2547 227

Fig. 2. Decision variables are all binary variables indicating whether to assign one node to the other

3.2 Heuristic Approach

In order to provide line balancing solution in a reasonable time, the sacrifice on solution's quality must be made. The greedy strategy is employed in this research. The solution procedure is divided into two phases. In phase I, operations are assigned to stations based upon standard times. Phase II then assigns all workers in the team to stations obtained from phase I. It should be noted that by grouping the operations without considering the workers' limitations in phase I, the solution obtained in the heuristic method cannot ensure the optimal solution. Phase I line balancing. This phase involves line balancing formulation with an additional constraint to limit the number of machines per station and parallel station criteria. In order to allow for parallel stations, a few more variables are introduced. Suppose there are S workers in the team, the average cycle time for the line, Tavg , and maximum standard time, Wmax, can be calculated from standard times wj of all operations. Then the maximum number of workers, r, that can be assigned to any station can be calculated from
max w j W j . r = max = Tavg 1 S wj j

(12)

Also, let J* be a set of operations j with wj [vWmax, Wmax] where v is a value between 0 and 1. The value v represents criteria for assigning more than one worker to a station, when considering any operation that belongs to set J*. The value v should be selected according to the distribution of standard times of all operations necessary to complete the task. Figure 3 denotes an example of histogram of standard time distribution for a product with maximum standard time of 145 s. and minimum standard time of 24 s. Intuitively v should be selected in such a way that all the operations with significantly large standard time be included in set J*. In most cases v can be safely set at 0.5 To decompose the problem to smaller sizes, the problem is formulated as an ILP for each station s. The ILP formulation can be solved progressively from station 1 up to station S. The objective function of this phase

228 ..2547 is to assign operations to each station such that the cycle time at each station, s, is maximized while still not exceeding the given target cycle time Tc as shown in constraint (15). Note that if there exists an operation j in J* included in the station s, i.e. yjs = 1, the station cycle time is bounded by rTc, else it is bounded by Tc. maximize subject to
s =
jJ *

s
w j y js

(13) (14) (15) (16) (17) (18) (19)

s max (y js rTc + (1 y js )Tc )

u j1 + y j1s y j2 s , ( j1 << j2 ) zms b jm y js Mzms , m
j

j u j =0

z ms 3
m

yjs, zms = 0 or 1, j, m

S ta nda rd Tim e Distribution

Tavg = 103 s

10
num be r of ope ra tions

8 6 4 2 0
2 4 -3 6 3 6 -4 8 4 8 -6 0 6 0 -7 2 7 2 -8 4 8 4 -9 6 9 6 -1 0 8 108120 120132 132144 144156

sta nda rd tim e ra nge

Fig. 3. Example of standard time distribution of all operations required for to make a nylon bag Additional binary variables uj's in constraint (16) are introduced to keep track of all the operations that have already been assigned. At the beginning, all ujs are zero. They are updated at the end of each stage(stations) until, at the end of phase I, all values of ujs are unity. The updating procedure is used to examine all operation js with yjs = 1, then set the corresponding ujs to unity. The precedence constraints (16) are applied to all arcs in the precedence diagram. In constraint (17), binary values zms is set to one when the

 ..2547 229

machine m is used in station s. Then the total number of machines used in any station must be less than 3 as shown in constraint (18). For any given Tc, the number of workers at each station s, qs, can be calculated from
y js w j j qs = Tc

(20)

Then the total number of workers required can be calculated by combining the value of qs over all stations. Due to the discrete nature of line balancing problem, Tc must be varied to identify local optimium. The searching procedure is assisted by script utility found in many ILP solvers interfacing with Visual Basic Programming Languages. In this research the lower and upper bounds on Tc must be be given by the user so that each integer value of Tc is explored. Only the one that satisfies the number of worker requirement with minimal line cycle time is reported. If such value is not found, the program will inform the users to enter another range of Tc. Phase II station assignment. The ILP formulation to assign workers to stations can be solved for each team of workers. First, the time required for each worker i to perform all tasks in station s, Tis is estimated. This can be computed from Tis = t ij , i , s (21)
j y js =1

Then the following optimization problem can be solved. Its objective function is to maximize the overall line cycle time, CT. (22) minimize CT subject to
2 x is Tis CT q s , s i

(23) (24) (25)

xis = q s , s
i

x is = 1, i
s

(26) xis = 0 or 1, i, s Constraints (23) determine the overall line cycle time. Note that in the case of parallel stations, the cycle time from these stations must be averaged twice in order to identify the combined station time. Constraints (24) state that each station must have appropriate number of workers assigned to it whereas constraints (25) imply that each individual worker must be assigned to a station. It should be noted that the resulting line cycle time obtained in this phase is different from that obtained in phase I since now the worker's skill are included.

230 ..2547

4. Case Studies
Both optimization phases are implemented on an ILP commercial software, LINGO, interfacing with an Excel spreadsheet software and Visual Basic Programming Dialog box. An example of product information from a nylon bag manufacturing company is used to illustrate the solution procedures capability. The company has several teams of workers to assemble (sew) products. Each team is assigned a product to assemble every day. Depending on the products order quantity, the team may have to work up to two products in a day. Table 1 denotes an example of Team 1 workers skill-by-skill evaluation. Note that most assembly line balancing problems considered by previous works assume that each worker is equally adept to perform any operations. In practice, this isnt always the case. There are 13 workers in the team. For illustration purposes, however, let us suppose that worker 10 is absent. The program can skip worker 10 and consider only the remaining 12 workers available. In the interest of limited space, the skill evaluation of other teams is omitted here even though such information can be used in comparing different teams' performance. Two case studies from two different products are used to illustrate the effects of line balancing parameters. Tables 2 and 3 illustrate product information used in case studies 1 and 2. The standard time for each operation step, wj, the skill type required to perform the operation, k, the machine type required to perform such operation, m, and its immediately preceding operations are also given. The standard time distribution of product 1 is shown in Figure 4. Table 1. Team 1s skill-by-skill evaluation skill \ worker 1 2 3 4 5 6 7 8 : 49 1 c b c c a b c c : a 2 a a b c a a b b : a 3 b a b c a b a b : a 4 0 a b 0 a b c b : a 5 a a a b a a b a : a 6 0 b 0 0 b b 0 c : b 7 0 c 0 a b b 0 c : c 8 0 c 0 c b b 0 c : c 9 0 c 0 0 b b 0 a : c 10 11 12 13 : 0 0 0 : 0 0 0 : 0 0 0 : 0 0 0 : c c 0 : c c 0 : 0 0 0 : 0 c 0 : : : : : 0 c c

 ..2547 231

Table 2. Product 1 Specification (12 workers, Tavg = 252 s.) used for case study 1 j 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 wj (s.) 60 90 60 76 38 18 44 14 36 116 96 103 72 116 116 72 120 51 28 145 k 31 36 31 33 16 16 36 42 32 36 33 33 24 27 37 13 34 27 6 36 m 5 5 5 5 2 2 5 5 5 5 5 5 5 5 5 3 6 3 4 5 preceding operations 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 19 j 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 wj (s.) 116 72 120 51 90 96 36 96 76 38 56 36 36 120 60 24 24 207 40 51 98 k 31 13 34 21 4 32 36 31 27 24 31 32 42 44 32 43 43 35 45 32 21 preceding operations 5 20 3 21 6 22 3 23 8 5 25 5 26 5 27 7 28 7 29 5 30 5 31 5 32 4 33 4 34 4 35 4 36 5 18,24,40 5 37 5 38 100 39 M

232 ..2547 Table 3. Product 2 Specification (12 workers, Tavg = 103 s.) used for case study 2 j 1 2 3 4 5 6 7 8 9 10 11 12 13 wj (s.) 36 40 51 36 25 32 40 40 48 72 25 42 40 k 5 45 24 5 45 24 31 38 31 1 31 5 5 preceding operations 4 6 1 100 2 5 6 4 6 5 6 6 6 7 6 8 5 9 6 10 5 5 12 m j 14 15 16 17 18 19 20 21 22 23 24 25 26 wj (s.) 48 24 64 36 60 48 48 48 48 30 145 51 64 k 10 47 3 16 32 31 36 31 27 25 35 16 24 M 5 3 4 2 6 6 6 6 6 6 5 2 5 preceding operations 13 14 15 16 17 18 20 21 22 3,11,19,23 24 25

4.1 Case study 1: maximum standard time is lower than average cycle time
When the maximum standard time, i.e. 207 s., is lower than the average cycle time, 252 s., as shown in Figure 4, phase I always results in operation grouping with no parallel stations as shown in Table 4. From systematic search for target cycle time that will yield exact number of workers with locally minimum balance delay, the target cycle time is found to be 269 s. Variation of the decision criteria, v, to make parallel stations will not have any effect on the result. The estimated balance delay in phase I is the maximum difference between target cycle time and estimated station time among all stations, before assigning workers to stations. For example, in this case, the estimated delay is 269-98 = 171 s. for the line. The last station will have the most free time of 171 seconds to spare, or 63.5% of target cycle time. This balance delay seems high, but the line balancing problem considered here is also restricted by the number of machines at each station, making it difficult to assign workloads more efficiently. When applying phase II, however, the overall line cycle time is increased to 319.2 s. Table 5 denotes the result of worker assignment to each station.

 ..2547 233
S ta nda rd Tim e Distribution

Tavg = 2 52 s

9 num be r of ope ra tions 8 7 6 5 4 3 2 1 0

1 0 -3 0 3 0 -5 0 5 0 -7 0 7 0 -9 0 9 0 -1 1 0 110130 130150 150170 170190 190210 210230 230250 250270

S ta nda rd tim e ra nge

Fig. 4. Standard time distribution of all operations required for product 1

Table 4. Result of line balancing in Phase I and Phase II station operations performed 1 1,2,19,25 2 3,4,26,27 3 5-10 4 11,28,29 5 12,13,30,31 6 14,15,32 7 33-37 8 20,21 9 16,17,22 10 23,24,39,40 11 18,38 12 41 estimated cycle time Phase I station time(s) 268 268 266 268 269 268 264 261 264 262 258 98 269 machines required 4,5,8 5 2,5 5,7 5,7 5 4,5 5 3,6 3,5,6 3,5 100 number of workers 1 1 1 1 1 1 1 1 1 1 1 1 total 12 Phase II worker actual station assigned time (s.) 7 250 3 248.8 12 319.2 4 252.8 2 215.2 9 291.2 6 223.2 13 313.2 5 211.2 8 286 1 226.8 11 117.6 319.2

4.2 Case study 2: maximum standard time is higher than average cycle time For product 2 with its information as shown in Table 3, it is found that the maximum standard time is larger than average cycle time. The standard time distribution for this product is as shown in Figure 3. The decision cretiria for making parallel stations, v, of at least 0.5 will result in target cycle of 115 s. Setting the

234 ..2547 value of v below 0.5 leads to higher target cycle times which is not desired and will not be shown here. Table 5 denotes the result of operation assignment and worker assignment in both Phase I and II. Predicted balance delay as shown in Phase I is 37.4% due to operation assignment in station 9. After assigning workers to stations in Phase II, the resulting line cycle time is 129 s. with the balance delay of 55.3% One may argue whether the methodology of assigning parallel stations is justified as compared to having only one worker in each station. To investigate such approach, the maximum number of worker per station, r, is manually set to unity. Using Phase I leads to the use of 9 workers with target cycle time of 145 s. as shown in Table 6. It should be noted that since the maximum standard time is 145 s., the minimum target cycle time for the line that can be used is 145 s. After phase II the resulting cycle time yields 144 s. Indeed, 3 remaining workers can work on other assignments as needed. The company should explore the trade-offs between the cycle time reduction and workers' utilization as the situation calls for. Table 5. Result of line balancing in Phase I and Phase II (parallel stations allowed) Phase I station time(s) 114 113 112 115 112 96 96 96 72 100 115 115 Phase II worker actual station assigned time (s.) 4 91.2 7 113 9 97.6 3 115 6 112 12 115.2 13 115.2 8 105.6 2 57.6 5,11 129 1 92 129

station

operations performed 1 1,4,12 2 5,13,20 3 2,14,15 4 3,16 5 6-8 6 17,18 7 9,21 8 19,22 9 10 10 11,23,24 11 25,26 estimated cycle time

machines required 4,5 5,6 3,5,6 4,100 6 2,6 6 6 5 5,6 2,5

number of workers 1 1 1 1 1 1 1 1 1 2 1 total 12

 ..2547 235

Table 6. Result of line balancing in Phase I and Phase II (no parallel station) station operations performed 1 1,4,5,20 2 2,3,21 3 12-14 4 6,7,15,22 5 8,16,17 6 9-11 7 18,19,23 8 24 9 25,26 estimated cycle time Phase I station time(s) 145 139 130 144 140 145 138 145 115 145 machines required 4,5,6 6,100 5 3,6 2,4,6 5,6 6 5 2,5 number of workers 1 1 1 1 1 1 1 1 1 total 9 Phase II worker actual station assigned time (s.) 3 116 6 131 9 120.9 4 115.2 5 112 2 116 7 144 1 116 12 138 144

6. Conclusion
In this research, a heuristic methodology to perform line balancing with parallel stations is proposed. The restrictions on the number of machines per station and workers skill level are taken into account. It consists of two phases. In phase I, operations are grouped into stations such that the target cycle time is not exceeded, parallel stations are allowed, precedence constraints are satisfied, and the number of machines per station is limited to 3. It is the most time-consuming stage with locally and manually adjusted step search. The criteria to make parallel stations depends on the maximum standard time relative to the average cycle time. As the number of available workers changes, as usually the case in practice, the average cycle time changes, leading to different line configurations, i.e. machine layouts, operation-station assignment. In phase II, the stations generated in phase I are assigned to workers in such a way that the overall line cycle time is minimized. While such greedy approach does not necessarily provide an optimal solution, it is able to generate a reasonably good solution in short time as compared to traditional line balancing formulation. The target cycle and the number of workers affect how operations are grouped into stations. The tradeoffs between cycle time reduction and workforce utilization are necessarily performed by the user. The comparisons between globally optimal solutions and those provided by this system are necessary and under way.

Acknowledgement
Author would like to thank the Faculty of Engineering, King Mongkut Institute of Technology North Bangkok for their partial support in this research.

236 ..2547

Reference
[1] J.R. Jackson, A Computing Procedure for a Line Balancing Problem, Management Sci. 2 (1956) 261-272 [2] E.M. Bowman, Assembly-Line Balancing by Linear Programming, Opns. Res. 8 (1960) 385389 [3] M.D. Kilbridge and L. Wester, A Heuristic Method of Assembly Line Balancing, J. Industrial Engineering. 12 (1961) 292-298 [4] M. Held, R.M. Karp and R. Sharaeshian, Assembly Line Balancing-Dynamic Programming with Precedence Constraints, Opns. Res. 10 (1963) 442-460 [5] J.H. Patterson and J.J. Albracht, Assembly-Line Balancing: Zero-One Programming with Fibonacci Search, Opns. Res. 23 (1975) 166172 [6] F.V. Assche and W.S. Herroelen, An Optimal Procedure for the Single-Model Deterministic assembly Line Balancing Problem, E. J. of Opn. Res. 3 (1978) 142-149 [7] F. Akagi, H. Osaki and S. Kikuchi, A Method for Assembly Line Balancing with More Than One Worker in Each Station, Int. J. Prod. Res. 21 (1983) 755-770 [8] F.B. Talbot and J.H. Patterson, An Integer Programming Algorithm with Network Cuts for Solving the Assembly Line Balancing Problem, Management Sci. 30 (1984) 85-99 [9] R.R. Inman and W.C. Jordan, Integrated Assembly Line Loading, Design, and Labor Planning, J. of Mfg. Sys. 16 (1997) 315-322 [10] K. Park, S. Park and W. Kim, A Heuristic for an Assembly Line Balancing Problem with Incompatibility, Range, and Partial Precedence Constraints, Computers Ind. Engng. 32 (1997) 321-332 [11] H.F. Ugurdag, R. Rachamadugu and C.A. Papachristou, Designing Paced Assembly Lines with Fixed Number of Stations, E. J. of Opn. Res. 102 (1997) 488-501 [12] G. Suer, Designing Parallel Assembly Lines, Computers Ind. Engng. 35 (1998) 467-470 [13] D.W. He, and A. Kusiak, Designing an Assembly Line for Modular Products, Computers Ind. Engng. 34(1998) 37-52. [14] K. Nakade and K. Ohno, An Optimal Worker Allocation Problem for a U-shaped Production Line, Int. J. Production Economics 60-61 (1999) 353-358 [15] S.C. Sarin, E. Erel and E.M. Dar-El, A Methodology for Solving Single-model, Stochastic Assembly Line Balancing Problem, Omega Int. J. Mgmnt. Sci. 27 (1999) 525-535 [16] B. Rekiek et al., State of Art of Optimization Methods for Assembly Line Design, Annual Reviews in Control 26 (2002) 163-174 [17] A. Khan, and A.J. Day, A Knowledge Based Design Methodology for Manufacturing Assembly Lines. Computers Ind. Engng. 41 (2002) 441-467