Anda di halaman 1dari 27

Cellular manufacturing A time based analysis to the layout problem

SEMINAR REPORT

Submitted in partial fulfillment of the requirement for the award of the degree of

Master of Technology
in

Industrial Engineering and Management


(Mechanical Engineering)

by

VIPIN N.

(Roll No. : M120496ME)

DEPARTMENT OF MECHANICAL ENGINEERING NATIONAL INSTITUTE OF TECHNOLOGY CALICUT NIT CAMPUS PO, CALICUT KERALA, INDIA 673601 APR - 2013

CERTIFICATE

This is to certify that the report entitled Cellular manufacturing A time based analysis to the layout problem is a bonafide record of the seminar presented by Vipin N. (Roll No. : M120496ME), in partial fulfillment of the requirements for the award of degree in Master of Technology in Mechanical Engineering (Industrial Engineering and Management) from National Institute of Technology Calicut.

Dr. T. Radha Ramanan (ME6194 Seminar) Dept. of Mechanical Engineering

Dr. R. Sridharan Head of the Department Dept. of Mechanical Engineering

Place : NIT Calicut Date : 11-04-2013

ABSTARCT

A cellular manufacturing system is an application of group technology principles to production. This involves processing groups of similar components in a dedicated cluster of dissimilar machines. In this paper, an approach that forms the cluster based on the processing time is suggested. For even distribution of workload, workload balancing is carried out in the second phase of the model, i.e., a time-based model. The time-based model is compared with the workload-based model using a commonality score. The performance of the time-based model is compared by means of workload deviation and deviation index. The validity of the approach is tested by application to the problems from the literature and the results are presented. The results indicate that the time-based model gives better even distribution of workload as compared to the workload-based model.

CONTENTS

List of symbols List of tables 1 Introduction 2 Proposed model 2.1 2.2 Problem representation Problem statement 2.2.1 Grouping criteria of the selected algorithms 2.2.2 Data requirements 2.3 Methodology 2.3.1 Algorithm I 2.3.2 Algorithm II 2.4 Analysis of the time-based model

ii iii 1 5 5 5 5 5 6 6 6 12 14 14 17 19 21

3 Examples 3.1 3.2 Workload-based model using commonality Time-based model for even distribution of workload

4 Results, discussion and conclusion References

LIST OF SYMBOLS

Mi Pj tij Cij Tf Tmax s k

machine number part number processing time of part j in machine i flexibility factor commonality score between machines i and j total time of the cell maximum processing time number of similar facility effective time per facility

ii

LIST OF TABLES

1. Machine Part matrix 5x7 2. Incidence matrix with time values 5x7 3. Results of 5x7 matrix using algorithm I 4. Machine Part matrix 7x14 5. Incidence matrix with time values 7x14 6. Results of 7x14 matrix using algorithm I 7. Machine Part matrix 5x6 8. Incidence matrix with time values 5x6 9. Results of 5x6 matrix using algorithm I 10. Machine Part matrix 15x10 11. Incidence matrix with time values 15x10 12. Results of 15x10 matrix using algorithm I 13. Input matrix 5x7 14. Results of 5x7 matrix using algorithm II 15. Input matrix 7x14 16. Results of 7x14 matrix using algorithm II 17. Input matrix 5x6 18. Results of 5x6 matrix using algorithm II 19. Input matrix 15x10 20. Results of 15x10 matrix using algorithm II 21. Detailed test results 22. Additional test results

6 7 7 14 14 15 15 15 15 16 16 16 10 10 17 17 17 18 18 18 19 19

iii

CHAPTER 1 INTRODUCTION
The layout or arrangement of equipments in the work area is an inevitable problem in all industrial plants. The layout decisions receive intensive attention in production and operations management. The type of manufacturing systems, either explosive or implosive, the physical arrangements of machines and equipments, storage area, human work area and other elements critically affect operating efficiency, system capacity and system flexibility. These factors in turn continuously affect the operating costs and degree of satisfaction. Group technology is a manufacturing philosophy whose main idea is to capitalize on similar, recurrent activities. It is a philosophy with broad applicability, potentially affecting all areas of manufacturing organization. The idea behind group technology is to decompose a manufacturing system into sub-systems in order to improve the efficiency of the manufacturing system. One specific application of group technology is cellular manufacturing systems, which involves processing collections of similar parts on dedicated clusters of dissimilar machines or manufacturing processes for increasing productivity.

In the past two decades or so, cellular manufacturing has been emerging as an important manufacturing concept. It has probably had a greater input on increasing manufacturing productivity than any other manufacturing concept. Cellular manufacturing is the application of group technology principles to production parts that have similar processing requirements and/or geometrical shapes that are classified into part families. The equipment requirements for each part family are determined subsequently or simultaneously to the identification of part families. The required equipment may then be moved and grouped into machine cells. These machine cells consist of groups of functionally dissimilar machine types and are dedicated to the production of one or more part families. In addition to the simplification of management control through the creation of smaller subsystems, cellular manufacturing also leads to reduced materials handling, reduced set-up time, reduced work- in-progress, reduced throughout time and improved sequencing and scheduling on the shop floor. Hence, there exists a substantial interest in cellular manufacturing systems as a primary source of improving the productivity of manufacturing operations. One of the first problems encountered in implementing cellular manufacturing is that of cell formation. If the number of cells is high, then the size of the cell is reduced, which increases the intercell moves.
1

For the past two decades, a number of methods have been developed for machine-part cell formation. Most of them used the one-zero incidence matrix. The basis for the incidence matrix was suggested by Burbridge (1963, 1971, 1977, 1982) in his production flow analysis. The methods have relied on both classification and coding approach and direct analysis of production process. The objective of the methods is to form clusters that should have the minimum number of voids (0s inside the diagonal block) and minimum exceptional elements (1s outside the diagonal block) so that the intercell moves are minimized. The approach using a similarity coefficient was first suggested by Mc Auley (1972). The basis of the method is to measure the similarity between each pair of machines and then to group the machines into families based on their similarity measurements. In most of the cases, the Jaccard coefficient (Sokal and Sneatch) is used. That is the number of components that visits both machines divided by the number of components that visits at least one of the machines. Rajagopalan and Batra (1975) developed group theory, which contains three phases. In the first phase, preliminary machine groups are formed using the similarity co-efficient. The second phase uses a graph-partitioning algorithm to merge the machine groups to form cells by minimizing intercell moves. Finally, components are allocated to the cells by evaluating machine loads, set-up times and number of machines required in a cell. The main drawback of this approach is the high density of the graph when large numbers of machines are involved. Seifoddini and Wolfe (1986) improved Mc Auleys (1972) method in three ways. They added a technique that duplicates the machines to eliminate bottleneck parts. They employed an average linking cluster analysis and techniques that reduce the amount of data storage required for their processing. This, however, led to an increase in computational complexity. Vanneli and Ravi Kumar (1986) proposed a method to find the minimum number of bottleneck cells for grouping part- machine families and considered a cell formation that integrates the issue of cell formation and within-cell material flows using the similarity coefficient approach to cluster parts and machines. But, this approach failed to take into account the issues of number of cells and duplication. Purcheck (1974) developed a mathematical approach for solving the grouping problems. The machines and parts can be represented as a path along the edges of a lattice diagram. The lattice diagram grows exponentially as the set is enlarged and hence its usefulness is limited.

The concept of production flow analysis was first introduced by Burbridge (1963, 1971, 1977, 1982). The aim of the technique was to find out the families of machines and parts by a progressive analysis of the information contained in the route sheet. The Burbridge approach consists of three levels of analysis. They are factory flow analysis, group analysis and line analysis. The first stage makes use of the process route number in order to obtain an overall picture of the present state of material flow. In the next stage, information is obtained by sorting components into packs and finally a layout in each group is found which give the nearest approximation to the line flow. Component flow analysis was first used in 1971. McCorrnick et al. (1972) developed a matrix clustering technique named the bond energy algorithm. This maximizes the sum of all products of nearest-neighbour elements in the pennuted matrix. First, it finds the optimal column pennutation and then finds the optimal row pennutation. Clustering is known to be non-polynominally complete and hence, for large- scale problems, heuristic procedures are widely used. King (1980) suggested the rank order clustering method (ROC). This approach sorts rows (and then columns) of a matrix in descending order of their binary weights. He also suggested a relaxation procedure that determines the number of duplicate machines required to eliminate the bottlenecks. The ROC 2 algorithm developed by King and Nakornchai (1982) improves on the original ROC by applying a quicker sorting procedure. Askin and Subramanian (1987) used a binary clustering algorithm for grouping parts and machines. They evaluated the configurations based on fixed and variable machine costs, set-up costs, cycle inventory, work in process inventory and material handling. Ballakur and Stevdel (1987) proposed a within-cell utilization-based clustering. In that, machines are assigned to cells based on workload and cell size, and parts are assigned to cells such that the majority of its operations to be performed are within the cells. This approach also identifies, whether additional machines are needed due to overloads. Logendram (1990) developed an approach that is based on workload to minimize the total intercell and intracell movements. He gave more weightage to the intercell moves for calculating the workload. He considered the processing time also. Mackulak and Cochran (1993) gave a group technology classification and clustering algorithms in cellular manufacturing. Delvale et al. (1994) presented a heuristic workload-based model to form cells by minimizing intercellular movements. Crama and Oasten (1996) in their paper discussed a few models for machine-part grouping in cellular manufacturing.

Most process-oriented techniques use only the information available on the part route sheets. In this paper, a model for even distribution of workload based on commonality score and time is suggested. In the suggested model, the clustering of parts is based on the zero-one matrix. For the time-based algorithm, the input matrix uses the time values instead of 1s. For comparison, however, random time values are given and according to that part grouping is done. In addition, workload deviation and deviation index for comparing the solution of two algorithms are also suggested.

CHAPTER 2 PROPOSED MODEL


2.1. Problem representation The problem in cellular manufacturing consists of information from a route card, arranged in the form of a one-zero matrix. The column of the matrix represents the components and the row of the matrix represents the machines. This one-zero matrix is also called incidence matrix. A one (1) entry in row i and column j indicates that part j has an operation scheduled on machine i. A zero (0) indicates that it does not. 2.2. Problem statement The most important and primary step in cellular manufacturing is to group parts with similar design features or processing requirements into families and form associated machines into machine cells. A number of algorithms have been developed for forming manufacturing cells. The objective of this work is to form cells that should be of even distribution of workload and compare the new approach to the selected model. The two algorithms used are 1. a workload-based model (using a commonality score) and 2. a time-based model for even distribution of workload. 2.2.1. Grouping criteria of the selected algorithms In the first algorithm, the first phase is the formation of machine groups based on the commonality score which indicates similarity between the machines. Final grouping of parts is based on the processing time. In the second algorithm, the grouping of machines and parts is based on the processing time by keeping in mind the even distribution of workload. Finally, balancing the workload of the cells is carried out. 2.2.2. Data requirements These models use input data associated with common machine cell, part family problems. Some of the following data are needed to apply the two algorithms suggested in Section 2.2:
(i) (ii) (iii)

total number of machines (Mi) (i = 1,2,...,M) (rows), total number of parts (Pj) (j = 1,2,...,N) (columns), processing time of the parts (tij),
5

(iv) (v) (vi)

number of cells [C], flexibility factor () >[>1], maximum workload deviation (Max WLD).

2.3.

Methodology Random members that are generated by the computer using C-program are used as input

time values. The zero-one incidence matrixes are taken from the literature or using randomness as an input to the algorithms. Basically, both the algorithms used in the present work are workload based. However, for the algorithm the workload basis is only for part allocation.
2.3.1.

Algorithm I

The steps in the workload-based model (using commonality score) can be grouped into the following three stages.
Stage I: Commonality

Compute the similarity coefficient called commonality score which is defined as

Where Cij is the commonality score between machines i and j, N is the total number of parts aik, ajk are machine and part incidence matrixes where rows and columns represent machines and parts, respectively.

Stage II: Linear cell clustering The various steps involved are: 1. Select the highest commonality score that has not yet been considered in the clustering process. Assume it is the commonality for machine pair (ij). Three states may exist as given below. (a). Neither machine i nor machine j has yet been assigned to a machine cell. In such a case, a new cell is created containing only these two machines. (b). Machine i has been previously assigned to a cell but j has not. In this case, machine j is added to the cell that already contains machine i. (c). Machines i and j are already assigned to the same cell; therefore, this commonality score is redundant and can be ignored. 2. Repeat the above process till all the machines are allotted to any one of the potential clusters. Stage III: Assignment phase The various steps involved are:
1.

For each part evaluate its cumulative processing time (in this case number of operations that a part has in a particular cell).

2.

Assign this part to a cell that contributes to the highest cumulative processing time (in this case maximum number of operations). Repeat step (2) till all the parts are admitted to any one of the existing potential clusters. Compute the workload deviation and deviation index. Algorithm II

3. 4.

2.3.2.

The time-based model develops clusters of machines and parts into cells according to the processing time. The objective of this algorithm is to form clusters that should have even distribution of workloads. For that, selections of time values are in a fashion of maximum and minimum. Total time of the cell (Tf) should be limited within the maximum processing time (T max) in a cell. The other constraints are number of machines and parts per cell. The following assumptions are made to form the model: 1. uniform machine utilization in all the cells is preferred, 2. no intercellular movement of parts. The steps involved are grouped into the following two stages. Stage I: Cell clustering The salient steps are: Step 1: First the possible number of machines and parts to be allotted to a cell is decided. There are two possibilities: 1. If the total number of machines (M) and parts (N) are an integer multiple of the number of machines and parts in a cell, then all F cells in the system will have the same number of machines and parts. 2. If not, the first (F-1) cells will have a certain configuration and the Fth cell is of different configuration. For the first (F-1) cells, m is the M/F (fractional part to be truncated), n the N/F (fractional part to be truncated) where F is the total number of cells, m the number of machines in the first (F- 1) cells and n the number of parts in the first (F-1) cells. For the Fth cell, m'= M [(F 1)m], n' = N [(F 1)n], where m is the number of machines in the Fth cell and n the number of parts in the Fth cell.
8

Step 2: The total processing time of the jobs in the system is

Where tij is the processing time of job j on machine i Step 3: To avoid maximum variability in the total processing time of the parts in all cells. The total processing time of the jobs in a cell = T/F: Tmax = (T/F)(> 1), where Tmax is the total processing time in a cell. The larger the values of the flexibility factor (), the more the flexibility in allocating the operations. The value of a nearer to 1 is preferred. Step 4: 1. Select the maximum processing time element from the input matrix. Allocate the part and machine which corresponds to that element to the cell. 2. Check if the machine is previously allocated to that cluster, then count it column = column + 1, otherwise row = row+1. 3. Check whether the row count is equal to m or the column count is equal to n; then stop the allocation to this cluster. Step 5: Form an array (S1) which contains the time element belonging to the processing times of the allocated (in previous step) parts in various machines and the time elements belonging to the parts which need processing in the allocated machine except the selected time element. Step 6: Check whether the total processing of the allocated parts in the cluster exceeds Tmax; if so, stop the allocation to this cluster. Step 7: Combine the arrays previously formed and the total array (initially the total array is a null array). Select the minimum processing time elements from the array and allocate the part and the machine to the cell.

Step 8: 1. If the machine just allocated is equal to the machine which is already present in the cell, then count column = column+1, otherwise row = row+1. 2. Check whether the row count is equal to m or the column count is equal to n ; then stop the allocation to this cluster.

Step 9: (i) For the minimum time element, the presence of time elements which are in the machine number (row) and in the part number (column) except the allocated time element are selected and form an array S2. (ii) Combine the arrays, array (S2) and already formed array (S1) in step 7. This is the total array. Step 10: Check whether the total processing time in the cluster exceeds or is equal to Tmax then stop allocation to this cell, otherwise select the minimum time element from the total array and allocate the corresponding machine and part to the cluster. Step 11: Form an array (S1) that contains the time elements which are in the machine number (row) and in the part number (column) that is assigned in step 10 except the selected time element.

10

Step 12: 1. If the machine just allocated to the cluster is equal to the machine already present in the cluster, then count column = column+ 1, otherwise row = row+ 1. 2. Check whether the count of column is equal to n and the count of row is equal to m; then stop the allocation to this cluster, otherwise go to step 7. Step 13: For every allocation of machine and part, the processing times are counted. This is the effective processing time of that cell. Check whether allocations for the first (F- 1) cells are over; then allocation is to be done for the final cell. Otherwise repeat from step 4 by initializing the counts. Step 14: For the final cell, the unallocated machines and parts to the first (F-1) cells are allocated to the final cell. The processing timings are counted. Stage II: Balancing of workload If alternative preferences of maximum and minimum time elements are not considered, then there are chances for the first cell to be complete with very few operations with very high elements. Therefore, in order to achieve uniform machine loading, alternate preferences for maximum and minimum time elements are considered, i.e., machine duplication is carried out to balance the workload. In this phase, the degree of workload deviation in the system is reduced within the permissible Max. WLD. This is done by means of increasing the number of facilities (manpower as well as number of similar machines) to the particular cell as per the following steps: 1. (i) (ii) Set i = 1. 2. (i) Number of similar facility (s) = i (teff/tmin) (truncate fractional part). (ii) Effective time per facility (k) = i(teff/s). 3. Check whether workload deviation

11

If the condition is not satisfied, count i i +1 and repeat step 2, otherwise repeat the steps for other cells. 4. Finally calculate deviation index (DI) and workload deviation (WLD) for the balanced cell.

2.4.

Analysis of the time-based model This model forms clusters, taking workload distribution as a primary objective. The

machine parts are grouped into families according to the processing times. Each one of the machine groups are dedicated to any one of the part families. The number of manufacturing cells to be formed depends upon the total cost incurred. The total cost includes set-up cost, cost associated with bottleneck operation and material handling cost. Management should take a decision regarding the number of cells. The decision should optimize the set-up cost with material handling cost and bottleneck operation cost. The size of the cell is the next criterion. In the first phase, the model considers that there is no possibility of duplicating the machines, and bottleneck operations are carried out by outsiders. The maximum number of machines and parts are limited by values [M/F and N/F]. The size of the cell is determined by this (M/F) factor and the availability. The performance of cell formation is measured by workload deviation and deviation index. These figures should be as low as possible in order to attain even distribution of workload. The other factor to limit the size of the cell is the maximum time Tmax (T/F)a to a cell. The second phase minimizes the workload of each cell within the maximum allowable limit (i.e., Max. WLD), which is given by the components. For this purpose, an increase in the number of facilities (machines as well as parts) is carried out. Therefore, cell formation is done by taking number of machines per number of parts per cell, maximum allowable time to a cell and maximum workload deviation as limiting factors. The result from the first phase is improved in the second phase. Moreover, machines and parts are split evenly. For all the test problems, the percentage to be considered for Max. WLD and flexibility factor is 20%.
12

The concept of line balancing is applicable to manufacturing cells. It is important that utilization of the expensive machine is high, even if it means that utilization of the other machines in the cell is relatively low. On the other hand, workload is to be evenly spread among the machines in the cells as much as possible. The expensive machine is referred to as the key machine . The other machines are referred to as supporting machines.

13

CHAPTER 3 EXAMPLES
In this section, four examples from the literature [2] are considered to illustrate the implementation and comparison of the suggested algorithms. Four examples are given in the form of a simple zero-one machine-part matrix. First, the problems are analyzed using a workload-based model using commonality and then the same problems are analyzed using a time-based model for even distribution of workload. 3.1. Workload-based model using commonality The examples considered are of 5 x 7 machine- part, 7 x 14 machine-part, 5 x 6 machinepart and 15 x 10 machine-part problems. Tables 1, 4, 7 and 10 shows a 5 x 7, 7 x 14, 5 x 6 and 15 x 10 machine-part matrix, respectively, and Tables 2, 5, 8 and 11 give the respective incidence matrix with time values. Suggested machine grouping and part grouping in various cell numbers as well as the total processing time for each cell, workload deviation and deviation index have also been calculated and are shown in Tables 3, 6, 9 and 12. Example 1: 5 x 7 machine-part problem. Example 2: 7 x 14 machine-part problem (Tables 4-6). Example 3: 5 x 6 machine-part problem (Tables 7-9). Example 4: 15 x 10 machine-part problem (Tables 10-12).

14

15

16

3.2. Time-based model for even distribution of workload The four example problems considered in Section 3.1 are solved using algorithm II described in Section 2.3.2. The flexibility and maximum workload are uniformly considered as 20 for all the problems. Tables 13, 15, 17 and 19 show the input matrix for 5 x 7, 7 x 14, 5 x 6 and 15 x 10 machine-part problems, respectively, and Tables 14, 16, 18 and 20 show the respective results of cell formation. Example 1: 5 x 7 machine-part problem. Example 2: 7 x 14 machine-part problem (Tables 15 and 16). Example 3: 5 x 6 machine-part problem (Tables 17 and 18). Example 4: 15 x 10 machine-part problem (Tables 19 and 20).

17

18

CHAPTER 4 RESULTS, DISCUSSION AND CONCLUSION


The time-based model and the workload- based model (using commonality score) are tested for the example problems. These problems are taken from the literature and also by generating the matrix through the random number method. The results are tabulated in Table 21. The results indicate that the time-based model for even distribution of workload gives better results than the workload- based model (using commonality score). The minimum value of workload deviation and deviation index indicates the good performance of the cell formation procedure.

For additional computational experience on the two algorithms (workload-based model and time- based model), four more problems are solved. Table 22 shows individual problem sizes, the number of cells and additional test results.

19

The time-based model forms cells with a distributed workload as evenly as possible, for jobs with operation times. The first phase of the two-phase clustering methodology proposed forms machine-part clusters such that the workload is distributed as uniformly as possible. The second phase improves the uniformity in the line loading. The second phase considers the maximum workload deviation specified by the management. The results show the performance measures such as percentage of workload deviation and deviation index for different problems solved using the two algorithms. In the tables, we observe that the performance of the time-based model is best.

20

REFERENCES
[1] Surjit Angra, Rakesh Sehgal, Z. Samsudeen Noori, 2008. Cellular manufacturing-A time based analysis to the layout problem. Int. J. Production Economics 112, 427438 [2] Das, K., Lashkari, R.S., Sengupta, S., 2007. Reliability consideration in the design and analysis of cellular manufacturing systems. International Journal of Production Economics 105, 243262. [3] Burbridge, J.L., 1963. Production flow analysis. Journal of Institution of Production Engineers, London 42, 742-748. [4] Defersha, F.M., Mingyuan, C., 2006. A comprehensive mathematical model for the design of cellular manufacturing systems. International Journal of Production Economics 103, 767-783. [5] Yin, Y., Yasuda, K., 2006. Similarity coefficient methods applied to the cell formation problem: A taxonomy and review. International Journal of Production Economics 101, 329-352.

21

Anda mungkin juga menyukai