You are on page 1of 10

Int J Adv Manuf Technol (1999) 15:577-586 9 1999 Springer-Verlag London Limited

The Intwmation=l Journal of

Rdvanced manufacturing Technologu

A Comparative Evaluation of Assembly Line Balancing Heuristics


S. G Ponnambalam ~, P. Aravindan 2 and G. Mogileeswar Naidu ~
~Department of Mechanical Engineering, PSG College of Technology, Coimbatore, India; and "-Regional Engineering College, Tiruchirapalli, India

A comparative evaluation is presented of six popular assembly line balancing heuristics, namely, ranked positional weight, Kilbridge and Wester, Moodie and Young, Hoffman precedence matrix, immediate update first fit, and rank and assign heuristic. The evaluation criteria used are the number of excess stations given, line efficiency, smoothness index and CPU time. The trade and transfer phase of the Moodie and Young method is applied to all heuristics, then the results are compared based on the evaluation criteria. Twenty problems each with 5 different cycle times are used for the comparative evaluation of heuristics. All six heuristics are coded in C-H- language. Among the six considered heuristics the Hoffmann enumeration procedure performs best; but, the execution time for the Hoffmann procedure is longer because this procedure enumerates all the feasible alternative sets of tasks for the stations.

The number of work elements cannot be greater than the number of work stations. Also the minimum number of work stations is one. The cycle time (amount of time available at each station as well as the time between successive units coming off the line) is greater than or equal to the maximum time for any station time and for the time of any work element Ti. The station time should not exceed the cycle time.

1.1

Types of Simple ALB Problems

Two types of simple assembly line balancing problems are: 1. In type I problems, the required production rate (i.e. cycle time), assembly tasks, tasks times, and precedence requirements will be given. Our objective is to minimise the number of work stations. A line with fewer stations has lower labour costs and reduced space requirements. Type I problems generally occur when designing new assembly lines. In this type of problem, for forecast demand we have to reduce the number of workstations, whereas expansion (when demand is more) can also be treated as a type I problem. 2. In type II problems, when the number of work stations or production employees is fixed, the objective is to minimise the cycle time. This will maximise the production rate. Type II balancing problems generally occur when the organisation wants to produce the optimum number of items by using the fixed number of work stations without purchasing new machines or without expansion. Here, we can identify precedence, and zoning constraints. While balancing the main line, we have to consider subassembly lines also, e.g. the cycle times of subassembly lines. Type I problems are more common than type II, which can be iteratively solved using type I methods. All exact algorithms become intractable with increasing problem size. While reasonable progress has been made in the development of exact or optimal approaches, considerable advances have been reported in the development of heuristic or inexact approaches to solve the single model deterministic task times problem. The reason is perhaps twofold: the understanding that highly efficient optimal approaches do not exit; and the need to solve large

Keywords: Heuristics; Line balancing; Simple assembly lines

1.

Introduction

An assembly line balancing (ALB) consists of a finite set of work elements or tasks, each having an operation processing time and a set of precedence relations, which specify the permissible orderings of the tasks. One of the problems inherent in organising for mass production is how to group work tasks to be performed on work stations so as to achieve a desired level of performance. Line balancing is an attempt to allocate an equal amount of work to each work station along the line. The fundamental line balancing problem is to assign a set of tasks to an ordered set of work stations, so that the precedence relations are satisfied and some measure of performance is optimised [1]. While designing an assembly line, the following restrictions must be imposed on groupings of work elements. Precedence relationship.

Correspondence and offprint requests to: Dr. S. G. Ponnambalam, Department of Mechanical Engineering, PSG College of Technology, PB 1611, Peelamedu, Coimbatore, 641004, India. E-mail: psgct@giasmd01 .vsnl.net.in

578

S. G Ponnambalam et al.
between columns are totally free as long as the elements are not connected by arrows. The cycle time CT is then determined by finding all combinations of the primes of E ~ T i (the total elemental times). Select a feasible cycle time CT. The permissible number of stations is K = (E~T/CT) Assign work elements to the station such that the sum of the elemental time does not exceed CT. Delete the assigned elements from the total number of work elements and repeat step 3. If the station exceeds CT owing to the inclusion of a certain work element, this element should be assigned to the next station. Repeat steps 3 to 5 until all elements are assigned to workstations.

problems. Most inexact approaches use the priority ranking or tree search logic. The ALB problems fall into the NP hard class of combinatorial optimisation problems [2]. Thus, numerous attempts have been directed towards the development of heuristic algorithms (e.g. [3-8]). A comprehensive review of ALB literature has been carried out by many workers (e.g. [1,9,10]) but a comparative evaluation of the various heuristic algorithms has not been carried out. Thus, the primary purpose of this paper is to evaluate the six popular heuristic algorithms, available in the literature, based on various performance measures.

2.

3. 4. 5.

2.

Heuristic Methods of Line Balancing

6.

In this section, we described several methods for solving line balancing problems. These methods are heuristic approaches, meaning that they are based on logic and common sense rather than on a mathematical proof. None of the methods guarantees an optimal solution, but they are likely to result in good solutions which approach the true optimum.

2.3

Moodio--Young Method

The Moodie-Young Method [5] consists of two phases:

Phase 1 2.1 Positional Weight Method


Work elements are assigned to consecutive workstations on the assembly line by the largest candidate rule (maximum task time rule). Use the matrix P (immediate predecessors of each element) and matrix F (the immediate followers) for the assignment procedure.

The steps involved in the Helgeson-Birnie [3] positional weight method are as follows: 1. Determine the positional weight (PW) for each task. (Time of the longest path from the beginning of the operation through the remainder of the network.) 2. Rank the work elements based on the PW. The work element with the highest PW is ranked first. 3. Proceed to assign work elements (tasks) to the workstations, where elements of the highest positional weight and rank are assigned first. 4. If at any workstation additional time remains after assignment of an operation, assign the next succeeding ranked operation to the workstation, as long as the operation does not violate the precedence relationships, and the station times do not exceed the cycle time. 5. Repeat steps 3 and 4 until all elements are assigned to the workstations.

Phase 2
Attempt to distribute the idle time equally to all stations through the mechanism of trades and transfers of elements.

Steps of phase 2 (trade and transfer)


1. Determine both the largest and smallest station times from the balance of phase 1. 2. Call half the difference between these two values GOAL = (STmax-S)/2. 3. Determine all single elements in STmax that are less than twice the value of GOAL and will not violate precedence restrictions if transferred to ST,,~n. 4. Determine all possible trades of single elements from STm,x for single elements from STrum such that the reduction in ST,,~x and subsequent gain in STm~, will be less than 2 GOAL. 5. Carry out the trade or transfer indicated by the candidate with the smallest absolute difference between itself and GOAL. 6. If no trade or transfer was possible between the largest and smallest stations, attempt trades and transfers between the ranked stations in the following order: with N (N-ranked stations have the greatest amount of idle time), N-1 . . . . . 3, 2,1. 7. If a trade or transfer is still not possible, drop the restrictions imposed by the value of GOAL and attempt, via the first six steps, to get a trade or transfer that will not increase the station time of any station beyond that of the original cycle time.

2.2

Kilbridge and Wester Heuristic

Numbers are assigned to each operation describing how many predecessors it has. Operations with the lowest predecessor number are assigned first to the workstations. The following procedures to be followed are proposed by Kilbridge and Wester [4]: 1. Construct the precedence diagram for the work elements. In the precedence diagram, list in column I all work elements that need not follow others. In column 1I, list work elements that must follow those in column I. Continue to the other columns in the same way. There exist many orderings that satisfy the precedence diagram. Movements

Assembly Line Balancing Heuristics


Table 1. Numerical score functions used in IUFF and RA.

579

n Name 1 Positional weight 2 3 4 5 6 7

Description

highest to the lowest numerical score, and assignment of tasks to stations is based on this rank. We summarise the steps of the RA heuristic as follows: 1. Assign a numerical score to each task using the functions given in Table 1. 2. Rank tasks from the highest to the lowest numerical score. 3. Assign tasks successively to the first station in which both the precedence and capacity constraints are met.

Sum of task times for x and all tasks that must follow it Reverse positional weight Sum of the task times for x and all tasks that precede it Number of tasks that follow task x Number of followers Number of tasks that immediately Number of immediate follow task x followers Number of tasks that precede task x Number of predecessors Task time of x Work element time Sum of the task times for x and all Backward recursive tasks in paths having x as its root positional weight

2.6

Hoffmann (Precedence Matrix)

Hoffmann [8] proposed an ALB algorithm using a precedence matrix. The procedure is described below: Starting with station 1, a precedence feasible list of tasks is maintained from which the combination of tasks which will minimise station idle time is found via complete enumeration. These tasks are assigned to station 1; the process continues with station 2 using an updated precedence feasible list. This procedure is repeated for each station in numerical order, until all tasks have been assigned. Hoffmann uses a special zero--one precedence matrix and vector to implement the enumeration procedure. This is a square matrix, consisting of zeros and ones, in which the rows are labelled with consecutive element numbers and the columns are labelled in the same order. Entries in the matrix are as follows. If the element of row i immediately precedes the element of column j, a 1 is placed in row i, column j. All other entries are zero. (Note that only immediate, 1 > > 3 relationships are stated explicitly. If 1 > > 3 > > 4 a one (1) is not entered in row 1, column 4.)

2.4

Immediate Update First-Fit (IUFF)

The immediate update first-fit (IUFF) heuristic was proposed by Hackman et al. [6]. It depends on numerical score functions that have been proposed in the literature, as shown in Table 1. We use the notation IUFF, n = 1,2,...,7, to signify which of the seven functions is used in the IUFF heuristic. The steps of the heuristic are as follows: 1. Assign a numerical score n(x) to each task x. 2. Update the set o f available tasks (tasks whose immediate predecessors have been assigned). 3. Among the available tasks, assign the task with the highest numerical score to the first station in which the capacity and precedence constraints will not be violated. Go to step 2.

2.5

Rank and Assign (RA) Heuristic

The rank-and-assign (RA) heuristic is similar to the IUFF heuristic, with the exception t h a t t h e tasks are ranked from the
Table 2. Data set used for the analysis.

Reference

Number of Cycle times tasks 12 9 19 9 12 24 14 10 10 8 11 9 7 9 11 12 11 9 45 50 8 7 8 10 0.8 2.0 0.65 1.I 0.9 1.6 31 7 10 36 0.35 51 48 20 69 100 10 8 10 12 0.9 2.5 0.8 1.2 1.0 1.8 40 8 11 40 0.4 60 52 25 92 150 13 9 12 13 1.0 3.0 1.0 1.4 1.1 2.0 50 10 12 50 0.5 70 62 30 138 200 17 10 15 16 1.1 4.0 1.1 1.5 1.3 2.5 75 12 14 75 0.7 80 64 38 184 250 25 13 20 24 1.4 6.0 1.3 2.0 1.4 3.0 80 14 21 80 1.0 105 97 40 276 300

Optimal number of stations for each cycle time 7 6 10 5 5 5 6 6 5 4 5 6 5 5 6 4 4 4 8 24 5 5 8 4 5 4 5 5 5 4 4 5 4 4 5 4 4 3 6 16 4 5 7 4 4 3 4 5 4 3 3 4 4 3 4 3 3 3 4 12 3 4 6 3 4 3 4 4 4 3 3 4 3 2 3 3 3 2 3 10 2 3 4 2 3 2 3 3 3 2 2 3 2 2 2 2 2 2 2 8

Elsayed and Boucher [11] (354 p) Hoffmann [8] Hoffmann [8] Elsayed and Boucher (348 p) Groover [12] (147 p) Groover [12] (169 p) Groover [12] (168 p) Groover [12] (168 p) Groover [12] (166p) Groover [12] (167 p) Askin and Standridge [13] (63 p) Gutjahr and George [2] Hoffmann [8] Askin and Standridge [13] (64 p) Magazine and Wee [7] Askin and Standridge [13] (40 p) Mansoor [ 141 Talbot and Patterson [15] Kilbridge and Wester [4] Generated problem

580

S. G Ponnambalam et al. 7. Subtract the remaining cycle time (the slack time) from the slack time of the previous combination generated (If this is the first, then subtract from the cycle time).

Table 3. Table form of the problem (Fig. 1). Task number


1

Task time
0.2

Immediate predecessor tasks


--

2 3 4 5 6 7 8 9 10
11

0.4 0.7 0.t 0.3 0.11 0.32 0.6 0.27 0.38


0.5

12

0.12

-1 1,2 2 3 3 3, 4 6, 7, 8 5, 8 9, I0 11

8. If zero or negative go to step 9. 8a. If positive, then this set of elements just generated becomes the new combination for this station. Go to step 10. 9. Go back one code number and go back to step 1 starting one element to the right of the element which had been selected from the code number. Repeat this procedure until the last column of the first code number has been tested; the result is that the last combination generated by step 8 is the one having the maximum elemental time f o r this station.

Table 4. Positional weights and ranking of the tasks. Rank


1

10. Replace the first code number with the last code number corresponding to the previous result. (This eliminates from further consideration the elements already selected.) 11. Repeat the previous steps until all the elements have been assigned. (Code number is entirely negative.)

Task number
1

Positional weight 2.5 2.3 2.1 1.7 1.6 1.3 1.21 1.0 1.0 O.89 0.62 0.12

2 3 4 5 6 7 8 9 10 I1 12

3 2 4 8 5 7 10 6 9 11 12

3. Comparison of the Algorithms


3.1 Data Set

In order to generalise the results, problems taken from the published literature and also randomly generated problems are solved. All the 6 heuristics are tested for 100 problems (20 networks and each solved for 5 different cycle times). Problem sizes are varied from 7 tasks to 50 tasks networks. Details of this data set is shown in Table 2. 3.2 Performance Measures

To use this matrix in generating all the feasible permutations, each column of the matrix is summed and these sums form another row adjoined to the bottom row of the matrix. The new row in the augmented matrix is termed a "code number". Next, the diagonal of the matrix is labelled with any arbitrary value (D). This first code number, K~, consists of a integers (a being the number of elements to be balanced), at least one of which is zero. The elements heading the columns in which there are zeros in K~, are candidates for the first position in the list of feasible permutations and only those elements can be candidates. The scheme for generating the feasible combinations and balancing the line, station by station, is as follows. 1. 2. 3. 4. Search left to right in the code number for a zero. Select the element which heads the column in which zero is located. Subtract the element's time from the cycle time remaining. If the result is positive go to step 5.

Perfect balance of the line means to combine the elements of the work to be clone in such a manner that at each station the sum of the elemental times just equals the cycle time. When a perfect balance cannot be achieved, we measure the effectiveness of the balance by the following.

Number of Excess Stations. It is the main measure considered by many researchers. A line with fewer work stations produces lower labour costs and reduces the space requirements, so it will produce a more cost effective plan. Further, if the number of stations is less then capacity, utilisation (LE) is generally more.
Line Efficiency (LE). This is the ratio of total station time to the cycle time multiplied by the number of work stations. It shows the percentage utilisation of the line. It is expressed as LE - ] ~ S T , 10 CT K where K = total number of work stations and CT = cycle time. S m o o t h n e s s Index (SO. This is an index to indicate the relative smoothness of a given assembly line balance. A smoothness index of 0 indicates a perfect balance. A smaller SI results in smoother lines, thereby reducing in-process inventory.

4a. If the result is negative go to step 6. 5. Subtract from the code number the row corresponding to the element selected and use this result as a new code number. Go to step 6. 6. Go to step 1 and start search one element to the right of the one just selected and repeat steps 1-6 until all the columns have been examined, then go to step 7.

Assembly Line Balancing Heuristics

581

0.il

Table 5. Backward recursive positional weight and ranking of the tasks. Rank Task number Backward recursive positional weight 10.99 7.4 5.29 3.19 3.09 1.3 1.21 1.0 1.0 0.89 0.62 0.12

1 2 3 4 5 6 7 8 9 10 11 12

1 3 2 4 8 5 7 6 10 9 11 12

Fig. 1. The precedence diagram of assembly network.

3.3

Experimental Conditions and Implementation

< ...... I U F F ........... >

< ........R & A .......... >

P
W

&
W

& Y

M X

Fig. 2. Heuristic rule vs. number of excess stations. Notations and numbers in the x-axis: RPW K & W M & Y PMX IUFF Ranked positional weight method Kilbridge-Wester method Moodie and Young method Hoffmann's precedence matrix method Immediate update first fit method 1 -- Minimum reverse positional weight rule 2 = Maximum number of follower tasks rule 3 = Maximum number of immediate follower tasks rule 4 = Minimum total number of predecessor tasks rule 5 = Maximum task time rule 6 = Maximum backward recursive positional weight rule Rank and assign method 1 = Minimum reverse positional weight rule 2 = Maximum number of follower tasks rule 3 = Maximum number of immediate follower tasks rule 4 = Minimum total number of predecessor tasks rule 5 = Maximum task time rule 6 = Maximum backward recursive positional weight rule

R & A

SI =

(STmax-ST)

where ST,..x = maximum station time and STi = station time of station i.

Execution (CPU) Time. This measure is also considered by many researchers, since the execution time is directly tied to the efficiency of the algorithm selected.

All the algorithms are coded using the C++ language. The computer system used is HCL-HP Pentium with 32 MB RAM, 2 GB HDD, and 133 MHz speed. All the methods are programmed to read from the same data set in an identical manner. Various object oriented programming (OOP) concepts, constructors, call by reference, pointers, dynamic memory allocation (new and delete) functions have been used in order to increase the speed of execution of the programs, and the effective use of the memory. The main advantage of using classes is that various information and associated functions about an entity (task or station) can be stored under a name which makes addressing easy. Classes make use of memory in a dynamic fashion and can free the memory allocated during the execution of the program, making it possible for some entities to share the memory space. In all the methods, we used two classes namely "task" and "station". Task class can hold all the details relating to the tasks, (e.g. task numbers, task times, precedence tasks, available tasks for the assignment, ranking of the tasks based on weights, etc.) and the associated functions. Similarly "class station" holds the details like station number, station time, station tasks and the functions relating to assignment of the tasks, and performing the trade and transfer phase. Pointers have been used extensively in the program and hence transfer of data between functions has been reduced to a bare minimum and hence duplicate copies of the same value are not made, as far as possible. Also the use of pointers increases the computational efficiency as the data is accessed directly and not through an intermediate step. Friend functions are also used here, so that a different object of the same class (declared as friend class) can use the same friend class object. However, if we use the derived class concept, then the data in the base class must be a protected type, so that the derived class members can make use of it. When we create objects of a derived class, then for each object, the base class data will be duplicated and initialisation problems will occur. The private data of a friend class can be used by reference to its object. All these concepts are used in our programs.

582
Table

S. G Ponnambalam et al.
6. ALB using positional weight method. Table 7. ALB using the Moodie and Young method. (a) Station-task assignment: before trade and transfer phase Station Task number Task time (Te) Station time (STy)
1 1

(a) Station-task assignment (CT=I.0): before trade and transfer phase Station Task number Task time (7",.) Station time (STy) CT-ST,
1 1

CT-ST~

3 4 5

3 2 4 5 6 8 7 10 9 11 12

0.2 O.7 0.4 0.1 0.3 0. l 1 0.6 0.32 0.38 0.27 0.5 0.12

0.9

0.1 2

0.91 0.92 0.65 0.62

0.09 3 0.08 4 0.35 5 0.38

2 4 5 3 6 8 10 7 9 11 12

0.2 0.4 0.1 0.3 0.7 0.11 0.6 0.38 0.32 0.27 0.5 0.12

1.0 0.81 0.98 0.59 0.62

0 0.19 0.02 0.41 0.38

Line efficiency = 80% Smoothness index = 0.404 (b) Revised assignment (CT=0.9): after trade and transfer phase Station Task number Task time (T,)
1 l

Line efficiency = 80% Smoothness index = 0.591 (b) Revised assignment (CT = 0.92): after trade and transfer phase Station Task number Task time (T~)
I 1

Station time (ST,)

CT-ST~

Station time (STy)

CT-ST~

2 3 4 5

3 2 4 7 5 8 6 10 9
11

0.2 0.7 0.4 0.1 0.32 0.3 0.6 0.11 0.38 0.27 0.5 0.12

0.9 0.82 0.90 0.76 0.62

0.1 2 0.08 3 0 4 0.14 5 0.28

2 5 4 3 8 7 6 10 9 11 12

0.2 0.4 0.3 0.1 0.7 0.6 0.32 0.11 0.38 0.27 0.5 0.12

0.9 0.8 0.92 0.76 0.62

0.02 0.12 0 0.16 0.30

12

Line efficiency = 88.889% Smoothness index = 0.323

Line efficiency = 86.957% Smoothness index = 0.361

4.

Numerical Illustration
$o 7o

A n illustration problem with 12 tasks is described in the precedence diagram in Fig. 1. The table form o f the network is s h o w n in Table 3. Tables 4 and 5 indicate the positional weight and backward recursive positional weights o f the tasks and the ranking based on these weights, e.g. positional weight of task 8 = Max {0.6+0.27+0.5+0.12, 0.6+0.38+0.5+0.12} = 1.6. Backward recursive positional weight o f task 2 = {{0.4+0. I}+{0.6+0.27+0.5+0.12}+{0.6+0.38+0.5+0.12}+{0.4+0.3 +0.38+0.5+0.12}} = 5.29. Tables 6-11 s h o w s the tasks allocated to different workstations, line efficiency, and s m o o t h n e s s index results for the illustrated problem using different heuristics. The results show that the trade and transfer phase reduces the s m o o t h n e s s index and thereby results in a s m o o t h e r line and higher efficiency.

R P

K &

M &

P M

< I

...... I U F F ...........> 2 3 4 5 6

< ........R & A I 2 3

..........> 4 5 6

Fig. 3. Line efficiency before trade and transfer (LEBT) vs. heuristic rule. (See Fig. 2 for key.)

5.

Results and Discussion

Table 12 shows a s u m m a r y o f results for all six heuristics based on 20 networks each solved with 5 different cycle times

(thus the total = 100 problems). The performance o f the six heuristic algorithms are evaluated based on a number o f performance measures. T h e results are discussed in the tbllowing sections.

Assembly Line Balancing Heuristics


Table

583

9. ALB using R and A heuristic: using task time rule.

25

(a) Station-task assignment: before trade and transfer phase Station Task number Task time (T,) 0.2 0.4 0.3 0.7 0.32 0.11 0.10 0.6 0.38 0.27 0.5 0.12 Station time (STi)

20

CT-ST~

2 3
R K M P <

P & & M W W Y X

...... IUFF ........... > <........ R & A .......... > I 2 3 4 5 6 I 2 3 4 5 6 4 5

Fig. 4. Smoothness index before trade and transfer (SIBT) vs. heuristic rule. (See Fig. 2 for key.) Table8. ALB weight rule. using IUFF: using backward recursive positional

2 5 3 7 6 4 8 10 9 11 12

0.9 0.7 0.53 0.98 0.89

0.1 0.3 0.47 0.02 0.ll

Line efficiency = 80% Smoothness index = 0.544 (b) Revised assignment (CT = 0.9): after trade and transfer phase Station Task number Task time (T,) 0.4 0.3 0.2 0.7 0.1 0.6 0.11 0.32 0.38 0.27 0.5 0.12 Station time (STI)

(a) Station-task assignmen': before trade and transfer phase Station Task number Task time (Te) Station time (STy)
1 1

CT-STi
0.1

3 4 5

3 2 4 5 6 8 7 10 9 II 12

0.2 0.7 0.4 0.1 0.3 0.11 0.6 0.32 0.38 0.27 0.5 0.12

CT-ST~
0.2 0 0.2 0.09 0.01

0.9

2 0.91 0.92 0.65 0.62 0.09 0.08 0.35 0.38 5 3 4

Line efficiency = 80% Smoothness index = 0.404 (b) Revised assignment (CT = 0.9): after trade and transfer phase Station Task number
1 1

5 1 3 4 8 6 7 10 9 11 12

0.7 0.9 0.7 0.81 0.89

Line efficiency = 88.889% Smoothness index = 0.297 well. The average rank and assign (RA) m e t h o d gives very poor results c o m p a r e d to the others (see Fig. 2 and Table 12).

Task time (7",) 0.2 0.7 0.4 0.1 0.32 0.3 0.6 0.11 0.38 0.27 0.5 0.12

Station time (STi)

CT-ST~
0 0.08 0 0.14 0.28

2 3 4 5

3 2 4 7 5 8 6 10 9 11 12

0.9 0.82 0.90 0.76 0.62

5.2 Average Line Efficiency (Le): Before Trade and Transfer Phase
Here also the H o f f m a n n precedence matrix procedure gives good results. In the case o f IUFF, the m a x i m u m task time rule gives good results; these results are close to the ranked positional weight method. Here also the R A based heuristic rules performed poorly. A m o n g the R A heuristic rules, the reverse positional weight rule performs best (see Fig. 3 and Table 12).

Line efficiency = 88.889% Smoothness index = 0.323

5.3 Average Smoothness Index (Si): Before Trade and Transfer Phase 5.1 Number of Excess Stations
The H o f f m a n n precedence matrix procedure gives good results with an average smoothness index o f 11.54. In the case o f IUFF, the m a x i m u m task time rule p e r f o r m s better with an average smoothness index o f 15.835. Here also R A based rules give poor result (see Fig. 4 and Table 12).

A m o n g the six heuristic algorithms, the precedence matrix procedure o f H o f f m a n n generally performs best. A m o n g the six heuristic rules c o n s i d e r e d in the I U F F m a x i m u m task time, m a x i m u m recursive positional weight rules perform reasonably

584

S. G Ponnambalam et al. Table 11. ALB using Hoffmann precedence matrix method. (a) Station-task assignment: before trade and transfer phase Station Task number Task time (Te) Station time (STy) CT-ST,
1 1

Table 10. ALB using Kilbridge and Wester method. (a) Station-task assignment: before trade and transfer phase Station Task number Task time (T~) Station time (STi)
1 1

CT-ST~)

2 3 4 5

2 4 5 3 6 8 7 10 9 11 12

0.2 0.4 0.1 0.3 0.7 0.11 0.6 0.32 0.38 0.27 0.5 0.12

1.0 0.81 0.92 0.65 0.62

0 0.19 0.08 0.35 0.38

2 4 5 3 6 8 10 7 9 11 12

0.2 0.4 0.1 0.3 0.7 0.11 0.6 0.38 0.32 0.27 0.5 0.12

1.0 0.81 0.98 0.59 0.62

0 0.19 0.02 0.41 0.38

Line efficiency = 80% Smoothness index = 0.556 (b) Revised assignment (CT = 0.9): after trade and transfer phase Station Task number Task time (Te)
1 1

Line efficiency = 80% Smoothness index = 0.591 (b) Revised Assignment (CT = 0.92): after trade and transfer phase Station Task number
1 1

Station time (STy)

CT-ST~

Task time (T~) 0.2 0.4 0.3 0.1 0.7 0.6 0.32 0.11 0.38 0.27 0.5 0.12

Station time (STy) CT-ST~

2 3 4 5

2 5 3 4
8

0.2 0.4 0.3 0.7 0.1


0.6

0.9 0.7
0.7

0 0.2
0.2

2 3 4

6 7 10 9
11

12

0.11 0.32 0.38 0.27 0.5 0.12

0.81 0.89

0.09 5 0.01

2 5 4 3 8 7 6 10 9 11 12

0.9 0.8 0.92 0.76 0.62

0.02 0.12 0 0.16 0.30

Line efficiency = 88.889% Smoothness index = 0.297

Line efficiency = 86.957% Smoothness index = 0.361

5.4 Average Line Efficiency (Le): After Trade and Transfer Phase
The average line efficiency of the Hoffmann enumeration procedure after the trade and transfer phase is 90.92%. Ranked positional weight, Moodie and Young, Kilbridge and Wester methods also perform very well. Among the IUFF, the maximum backward recursive positional weight rule performs best with an efficiency of 89.982%. Efficiencies of RA rules are also comparable with the other methods (see Fig. 5 and Table 12).

5.6

Total Execution Time (s)

All the rank and assign rules were executed within the minimum CPU time. Each rule takes 1.50 seconds on average to solve 100 problems. For RA with the maximum task time rule, the CPU time is 1.246 seconds. Except for the Hoffmann precedence matrix procedure, all other rules solve the problems in less than 2 seconds. The Hoffmann procedure takes more CPU time (18.35s) because this procedure enumerates all feasible combinations of tasks to be performed on stations.

5.5 Average Smoothness Index (Si): After Trade and Transfer Phase
The Hoffmann enumeration procedure performs best on average with an average m i n i m u m SI of 6.582. Ranked positional weight, the Moodie and Young method, and the Kilbridge and Wester heuristic also perform best. A m o n g the IUFF rules, the maximum task time rule performs better with a minimum SI of 9.214. RA rules also perform best, even though these produce more stations and less efficiency (see Fig. 6 and Table 12).

6.

Conclusions

Six heuristic assembly line balancing methods are implemented using the C-H- language, each of which is designed to assign tasks to workstations given a prespecified cycle time, or production rate. All these methods are tested on 100 problems found in the published literature, and also on randomly generated problems. In order to increase the line efficiency and smoothness of the line, the trade and transfer phase of the Moodie-Young method is applied to all six heuristics algor-

Assembly Line Balancing Heuristics Table 12. Comparative results of various heuristic algorithms Method Number of excess stations 59 64 68 51 69 78 72 87 64 69 101 106 115 107 106 105 Average LEBT Average SIBT Average LEAT Average SIAT

585

Total execution time (s) 2.25 1.54 1.98 18.35 1.593 1.703 1.978 1.484 1.429 1.758 1.590 1.39 1.429 1.494 1.246 1.538

Ranked positional weight method Kilbridge and Wester heuristic Moodie and Young method Hoffmann precedence matrix IUFF heuristic 1. Minimum reverse positional weight 2. Maximum number of follower tasks 3. Maximum number of immediate follower tasks 4. Minimum number of predecessor tasks 5. Maximum task time 6. Maximum backward recursive PW R and A heuristic 1. Minimum reverse positional weight 2. Maximum number of follower tasks 3. Maximum number of immediate follower tasks 4. Minimum number of predecessor tasks 5. Maximum task time 6. Maximum backward recursive PW

82.66 82.18 81.29 84.39 81.27 79.57 80.80 78.34 82.18 81.44 76.93 76.36 75.06 76.07 76.25 76.68

18.84 15.84 21.93 11.54 20.96 23.16 20.25 21.32 15.83 22.29 22.11 23.63 25.10 24.12 22.50 21.91

90.36 89.89 89.59 90.92 89.553 88.81 88.954 88.35 89.89 89.98 87.87 86.92 86.76 88.04 87.04 86.64

8.22 9.21 9.37 6.58 9.884 10.12 9.69 10.1 t 9.21 9.77 10.61 11.76 12.03 10.84 11.73 11.47

cedure performs best; but the execution time for the Hoffmann procedure is greatest because this procedure enumerates all the feasible alternative sets of tasks at the stations.

iilllllllllm
R K M P < .... IUFF ........ > < ...... R&A

Acknowledgement
This work was partially supported by the grant for the major research project supported by University Grants Commission, New Delhi, India (F.14-34/96 (SR-I)) dated 5 April 1996.
6

....... >

P & & W W Y

M X

Fig. 5. Line efficiency after trade and transfer (LEAT) vs. heuristic rule. (See Fig. 2 for key).

References
1. Soumen Ghosh and Roger J. Gagnon, "A comprehensive literature review and analysis of the design, balancing and scheduling of assembly lines", International Journal of Production Research, 27(4), pp. 637-670, 1989. 2. Allan L., Gutjahr and George L. Nemhauser, "An algorithm for the line balancing problem", Management Science, 11(2), pp. 308315, 1964. 3. W. P. Helgeson and D. P. Birnie, "Assembly line balancing using the ranked positional weight technique", Journal of Industrial Engineering, 12(6), pp. 394-398, 1961. 4. M. D. Kilbridge and L. Wester, "A heuristic method of assembly line balancing", Journal of Industrial Engineering, 12(4), pp. 292298, 1961. 5. C. L. Moodie and H. H. Young, "A heuristic method of assembly line balancing for assumptions of constant or variable work element times", Journal of Industrial Engineering, 16(1), pp. 2329, 1965. 6. Steven T. Hackman, Michael J. Magazine and T. S. Wee, "Fast, effective algorithms for simple assembly line balancing problems", Operations Research, 37(6), pp. 916-924, 1989. 7. M. J. Magazine and T. S. Wee, "An efficient branch and bound algorithm for an assembly line balancing problem Part 1. The minimum number of workstations", working paper no. 150, University of Waterloo. Ontario, Canada, June 1981. 8. Thomas R. Hoffmann, "Assembly line balancing with a precedence matrix", Management Science, 9(4), pp. 551-562, 1963. 9. I. Baybars, "A survey of exact algorithms for the simple assembly line balancing problem", Management Science, 32(8). pp. 909932, t986.

12

10

i'llllmlillill
R K M P < ...... I U F F ........... > < ........ R & A .......... >

P W

& & W Y

M X

Fig. 6. Smoothness index after trade and transfer (SIAT) vs. heuristic rule. (See Fig. 2 for key). ithms. It is observed that the smoothness index of the line depends upon the heuristic method used. All the methods are compared for the number of excess work stations, line efficiency, smoothness index before and after the trade and transfer phase, and CPU time. A comparison of results is presented in the form of tables and charts. Among the six considered heuristics the Hoffmann enumeration pro-

586

S. G P o n n a m b a l a m et al.

10. Brian F. Talbot, James H. Patterson and William V. Gehrlein, "A comparative evaluation of heuristic line balancing techniques", Management Science 32(4), pp. 430-454, 1986. 11. A. E. Elsayed, Thomas O. Boucher, Analysis and Control of Production Systems, Prentice-Hall, NJ, 1994. 12. MikeU P. Groover, Automation, Production Systems, and Computer Integrated Manufacturing, Prentice-Hall of India, February 1996.

13. Ronald G. Askin and Charles R. Standridge, Modelling and Analysis of Manufacturing Systems, John Wiley, 1993. 14. E. M. Dar-el Mansoor, Assembly line balancing - an improvement on the ranked positional weight technique, Journal of Industrial Engineering, 15(2), pp. 73-77, 1964. 15. Brian F. Talbot and James H. Patterson, "An integer programming algorithm with network cuts for solving the assembly line balancing problem", Management Science, 30(1), pp. 85-99, 1984.

You might also like