Assignment Flexibility in a Cellular Manufacturing System - Machine Pooling versus Labor Chaining -

Transcription

1 1050 Flexible Automation and Intelligent Manufacturing, FAIM2004, Toronto, Canada Assignment Flexibility in a Cellular Manufacturing System - Machine Pooling versus Labor Chaining - Jos A.C. Bokhorst 1, Jannes Slomp, and Gerard J.C. Gaalman Production Systems Design Group, Faculty of Management and Organization University of Groningen P.O. Box 800, 9700 AV Groningen, The Netherlands ABSTRACT In this paper, we compare the effects of machine pooling and labor chaining by means of a simulation study. In this study, the independent variables are: (1) the level of machine pooling (no. of jobs that can be produced in more than one cell), (2) the level of labor chaining (no. of workers able to work in more than one cell), and (3) labor utilization/machine utilization. The dependent variable is the mean flow time of jobs. Major outcome of our study is that, within our simulation setting, machine pooling is more important than labor chaining. All interaction effects, however, appear to be significant. As a consequence of our study, we suggest that managers should support the development of procedures for inter-cell movements of jobs. If there is a certain level of routing flexibility over cells, it is in most CM environments not needed to support the possibility of moving workers from one cell to another cell. 1. INTRODUCTION Producers of goods are under intense pressure to improve their operations by enhancing productivity, quality, customer responsiveness, and reducing manufacturing costs. The adoption of cellular manufacturing (CM) has consistently formed a central element of many of these efforts and has received considerable interest from both practitioners and academicians. In a CM system, the manufacturing department of a firm is split in relative autonomous groups of workers and machines, each responsible for a set of part types. Within most CM systems, not all equipment is manned at all times. In these manufacturing cells, total machine capacity is higher than the total labor capacity of the team operating the machines within the cell. This is a general characteristic of the so-called Dual Resource Constrained (DRC) systems (see, e.g., [1]). Cellular manufacturing is known to offer several major advantages, including reduction in lead times and work-in-process inventories, full accountability for jobs by teams of workers, etc. [2]. Other advantages include reduction of setup times due to similarity of part types produced in manufacturing cells. A critical element in CM is the inability to cope with changes in product mix. In a dynamic environment, the various cells face imbalances. Identical machines split among more than one cell may have different loads, which results in negative effects on the performance of queue-related variables. In order to cope with workload imbalances between the various cells, due to changes in the product mix, management of a CM system may take several measures. One possibility is to move the demand (i.e., jobs) from one cell to the other. This means that flexibility with respect to assigning jobs is increased. Another possibility is to move the supply (i.e., workers) from one cell to the other. This entails increased flexibility with respect to assigning workers. The first possibility basically means that identical machines, located in more than one cell, are regarded as one resource pool within the production control system. We call this machine pooling. The second possibility basically means that autonomous teams become connected by the assignment of workers of these teams to more than one cell. We call this labor chaining, since a path is created connecting the teams/cells. In case of labor chaining between cells, all workers and machines of the cells are connected, directly or indirectly. Machine pooling and labor chaining are alternative concepts that increase assignment flexibility in order to cope with workload imbalances in a CM-system. In this paper, we focus on the performance effects of machine pooling and labor chaining. In most CM systems, both measures can be implemented to some extent. The question is what (combination of) measures results in good performance. For example, is it better to invest in machine pooling than in labor chaining? Or, is it worthwhile to 1 Corresponding author: Tel.: (+31) ; Fax: (+31) ; j.a.c.bokhorst@bdk.rug.nl

2 Assignment Flexibility in a Cellular Manufacturing System - Machine Pooling versus Labor Chaining 1051 invest in both types of assignment flexibility? This paper addresses these questions by means of a simulation study using the package EM-Plant. In this study, the independent variables are: (1) the level of machine pooling (i.e., the number of jobs that can be produced in more than one cell), (2) the level of labor chaining (i.e., the number of workers able to work in more than one cell), and (3) labor utilization/machine utilization. The dependent variable is the mean flow time of jobs. Section 2 reviews pooling literature and labor chaining literature. Section 3 describes the design of a simulation study comparing the performance effects of labor chaining with those of machine pooling, the results of which are presented in Section 4. Section 5 gives the conclusions of our study, indicates the practical relevance of the outcomes, and suggests topics for further research. 2. ASSIGNMENT FLEXIBILITY In the introduction, we specified two types of assignment flexibility: machine pooling and labor chaining. In this section, the literature on these types of flexibility will be reviewed. 2.1 MACHINE POOLING First of all, it is important to clarify what is meant by the term machine pooling. Suresh [3, 4] analyzed the effects of partitioning work centers for group technology. Analytical models showed that partitioning leads to adverse effects on performance. Later, Suresh and Meredith [5] labeled these adverse effects the loss of pooling synergy and they investigated the extent to which these adverse effects can be overcome through several measures. Our starting point is not a functional layout that is converted into a cellular layout, which involves the partitioning of work centers (e.g., [3, 4, 6]). Instead, we look at a cellular layout and group operations of jobs that can be processed on identical machines within different cells. We use the term machine pooling, which results in pooling synergy, to denote the opposite of the partitioning of work centers. Note that machine pooling does not necessarily imply that similar machines are placed in close proximity, as in a functional layout. If operations of a job can be (and are allowed to be) processed on alternate machines, and a choice between these machines can be made till the moment of processing, then these alternate machines are pooled for this operation, irrespective of the physical proximity of these machines. In other words, routing flexibility in its most liberal form can be regarded as machine pooling, with similar effects. Of course, if the choice between machines can be delayed till the moment of processing, there will most often be a single queue of jobs for the pooled machines. In that case, these machines will probably not be located too far apart from each other due to reasons of practicality. Ang and Willey [7] studied the effects of inter-cell workload transfers on system performance in group technology. If a machine center is congested in one manufacturing cell, (part of) the workload from that machine center is transferred to a less congested work center in another cell. They showed that inter-cell workload transfer improves shop performance and most of the improvements can be achieved with a small ratio of inter-cell transfers to the total number of job transfers. Benjaafar et al. [8] also found the relationship between routing flexibility and performance to be of the diminishing return type. In their paper, routing flexibility is represented by the number of processing options for a part. They claim that for most operating scenarios, increases beyond two or three routing options result in only marginal performance improvements. However, both papers [7, 8] do not include labor constraints. In other words, the CM systems modeled in these papers are not dual resource constrained and therefore possibilities of transferring labor instead of operations of jobs have not been explored. 2.2 LABOR CHAINING Jordan and Graves [9] stressed the importance of chaining in the case of limited flexibility. In terms of graph theory, a chain is a connected bipartite graph. Jordan and Graves [9] studied the effect of process flexibility, which they define as the ability of plants to produce different types of products. In their paper, the two subsets of vertices of the bipartite graph are products and plants. The edges or arcs represent product assignment decisions. Their findings show that most of the benefits of total flexibility, where all pairs of vertices are adjacent (i.e., a complete graph), can be achieved by chaining. Flexibility should thus be added in such a way that a path is created that connects every pair of vertices. Process flexibility as in [9] is conceptually equivalent to labor flexibility, which refers to the ability of workers to operate different machines. With labor flexibility, the two subsets of vertices are workers and machines, instead of products and plants. The edges or arcs represent the worker skills. Brusco and Johns [10] recognized this and used the term chaining to explain the preference of some of their cross-training patterns. They presented a linear programming model that minimizes costs associated with workforce staffing, subject to the satisfaction of minimum labor requirements across a planning horizon. They used their model to evaluate eight cross-training structures across various patterns of labor requirements, reaching the important

3 1052 Flexible Automation and Intelligent Manufacturing, FAIM2004, Toronto, Canada conclusion that chaining of employee skill classes across work activity categories is a basic element of successful cross-training structures. Hopp et al. [11] studied two cross-training strategies for serial production systems with flexible servers. One of these strategies was a chaining strategy. The D-skill chaining strategy enables workers to help over-utilized stations indirectly by means of paths through other workers. Under the D-skill chaining strategy, each worker is cross-trained for an initial station and for the next D-1 stations in the line. Hopp et al. [11] showed that the 2-skill chaining strategy is potentially robust and efficient in obtaining workforce agility in serial production lines. Chaining provides the ability to shift work from a worker with a heavy workload to a worker with a lighter workload, leading directly or indirectly to a more balanced workload. Chaining therefore supports the efficient use of labor capacity and provides sufficient agility to respond to changes in demand, thus enabling fluctuations in the mix of work to be absorbed. As explained in the previous subsection, another possibility of dealing with workload fluctuations is to move (some of) the workload of heavily loaded machines to other machines able to perform the operation. To the best of our knowledge, trade-offs between chaining and machine pooling have not been investigated in the literature. This paper deals with this issue. 3. SIMULATION STUDY This section describes the simulation study used to compare the effects of labor chaining with the effects of machine pooling. First, we explain the experimental design. Second, we discuss the performance measures and provide details on how the simulation experiments were performed. 3.1 EXPERIMENTAL DESIGN For our simulation experiment, we model two manufacturing cells. In order to isolate the effects of assignment flexibility between the cells, we focus on the identical machines that are split between these two cells and consider those workers who are able to operate these machines. We assume that the two cells have five machines in common, which can be operated by three workers within each cell. Each of these three operators can thus operate all five machines within the cell. In other words, there is total flexibility or full cross-training within the cells. Figure 1a represents this situation. One cell comprises machines 1 to 5 and workers A, B, and C, the other cell comprises machines 6 to 10 and workers D, E, and F. Further, identical machine pairs are (1,6), (2,7) (5,10). This system is Dual Resource Constrained (DRC), since there are fewer workers than machines. Labor capacity as well as machine capacity may limit the output of the system. Jobs for each machine within each cell arrive according to a Poisson distribution. We assume the arrival rate of jobs to each machine to be identical. In our simulation, the routing of each job only includes a visit to one machine. Other possible routing steps of these jobs are assumed to be at specific machines within each cell (i.e., machines that are unique to the cell and cannot be found in the other cell). The processing times at machines are distributed according to a negative exponential distribution with a mean of one. Because the CM system is dual resource constrained, labor allocation rules must be specified. Traditional labor allocation rules in DRC systems are the when labor assignment rule, which determines when a worker is eligible for transfer, and the where labor assignment rule, which determines where a worker should be transferred to once he or she is eligible for transfer. Additionally, we specified a who labor allocation rule, which is often neglected in prior literature, but is necessary for proper operation of the system. The who-rule decides which worker should be transferred to a machine when multiple workers are available to perform an operation at a machine, and no other machines are waiting for (one of) these workers. We assume a centralized when-rule, meaning that workers are allowed to transfer to another machine after they have finished a job at a machine. By contrast, a decentralized when-rule requires that workers finish all jobs in front of the machine before they are allowed to transfer to another machine. Further, a First-In-System-First-Served (FISFS) where-rule is used to assign workers who are eligible for transfer to the machine with the oldest job in the queue. Jobs are also dispatched according to the FISFS rule, meaning that the oldest job at a machine is processed first. We used the longest idle time who-rule, which assigns the worker who has been idle the longest time when more workers are available for a job. This rule was also employed in [12]. Machine pooling and labor chaining are examined as main experimental factors with different levels. Additionally, we included labor utilization as an experimental factor. Benjaafar [8] for example showed that routing flexibility has a greater impact under conditions of high system loading. By including utilization as an experimental factor, interactions between this factor and machine pooling as well as labor chaining can be examined.

4 Assignment Flexibility in a Cellular Manufacturing System - Machine Pooling versus Labor Chaining 1053 In this experiment, we distinguished four levels of machine pooling: 0%, 33%, 67%, and 100%. At 0% machine pooling, all jobs arriving at a machine within a cell must be processed on that machine within that cell. There is no inter-cell movement of jobs allowed. At 33%, 67% and 100% machine pooling, 33%, 67% and 100% of the jobs arriving at a machine within a cell, respectively, are allowed to be processed on the identical machine in the other cell. Note that the number of processing options per job is thus either one or two. Another way of looking at it is that with 0% machine pooling, each machine has its own queue. With 33% machine pooling for instance, 33% of the jobs are in a common queue for two machines, while the machines are dedicated for the remaining 67% of the jobs. For labor chaining, we also distinguished four levels: 0%, 33%, 67%, and 100%. At 0% labor chaining, workers are only allowed to work in their own cell. Figure 1a represents this situation, where an arc denotes the assignment possibility of a worker. We see that workers A, B, and C are allowed to operate machines 1-5, and workers D, E, and F are allowed to operate machines Note that even though the workers are skilled to work in the other cell, since the cells are identical, they are not permitted to transfer to machines outside their own cell. At 33%, 67% and 100% labor chaining, one, two or all worker(s) from each cell, respectively, is/are allowed to transfer to the other cell. Figure 1b, 1c, and 1d represent these levels of labor chaining, respectively. 1 A 6 1 A 6 2 B 7 2 B 7 3 C D 8 3 C D 8 4 E 9 4 E 9 5 F 10 5 F 10 (a) (b) 1 A 6 1 A 6 2 B 7 2 B 7 3 C D 8 3 C D 8 4 E 9 4 E 9 5 F 10 5 F 10 (c) (d) Figure 1: The two cells with machines 1-10 and workers A-F with 4 different levels of labor chaining Finally, we considered labor utilizations of 75%, 80%, and 85%. The full-factorial design resulted in 48 unique experimental cells (4x4x3). 3.2 PERFORMANCE MEASURES AND SIMULATION DETAILS Performance is measured as the mean flow time of jobs through the system (MFT). Almost all simulation studies include MFT as a major performance measure. The simulation models were written in the object-oriented simulation software package EM-Plant Version 5.5 (Stuttgart: Technomatix). The simulation models used are stochastic, steady-state, and non-terminating. We used the replication-deletion approach [13] to estimate the steady-state means

5 1054 Flexible Automation and Intelligent Manufacturing, FAIM2004, Toronto, Canada of the output parameters. We employed Welch s method in order to determine the warm-up period. Other graphical approaches were used to gain insight into the number of replications and the run length required. Each experiment consisted of 35 replications with a run length of 18,000 time units and a warm-up period of 4,000 time units. Different seeds were used for each replication to maximize sampling independence. 4. RESULTS The data were analyzed by means of Analysis of Variance (ANOVA) with labor utilization (LU), machine pooling (MP) and labor chaining (LC) as independent variables and mean flow time (MFT) as dependent variable. The ANOVA results are summarized in Table 1. The homogeneity of variance assumption does not hold (Levene test, p < 0.05). The F statistic, however, is robust against heterogeneous variances, as long as the group sizes are equal [14] or approximately equal [15]. Because between-subjects ANOVAs are conducted with equal cell sizes, we continue the analyses. Table 1 shows the main effects and all interaction effects to be significant (p < 0.05). Table 1: ANOVA results Mean Flow Time Source F p-value LU p < MP p < LC p < LU x MP p < LU x LC p < MP x LC p < LU x MP x LC p < Tukey post-hoc tests were performed to find out which of the levels within each main effect differ significantly from each other with respect to MFT. Tukey results indicated that for all main effects, MFT for each level differs significantly (p < 0.05) from the other levels. Table 2 shows the main effects. Obviously, MFT increases when labor utilization (LU) is increased. The main effect of machine pooling (MP) shows that the higher the level of MP is, the lower the MFT. More specifically, MFT reduction diminishes if the level of MP is increased. In going from no machine pooling to 100% machine pooling, MFT is reduced by 30%. Finally, the main effect of labor chaining (LC) shows that increasing the level of LC decreases MFT. In going from no labor chaining to 100% labor chaining, MFT is reduced by 8.7%. However, the first introduction of LC (i.e., from no labor chaining to 33% labor chaining) already reduces MFT by 5.8%. To conclude, the main effect of machine pooling on MFT is stronger than the main effect of labor chaining on MFT. Table 2: Main effects LU MFT MP MFT LC MFT 75% % % % % % % % % % % 1.90 The largest interaction effect (F ) is that of machine pooling (MP) with labor chaining (LC). The effect of labor chaining (LC) on machine pooling (MP) is illustrated in Figure 2. The simple main effects showed us that the effect of labor chaining on mean flow time is significant (p < 0.05) within all levels of machine pooling. These simple main effects of labor chaining were further analyzed by pairwise comparisons using the Sidak adjustment for multiple comparisons. If the level of machine pooling is 0%, 33%, or 67%, all levels of labor pooling differ significantly from each other, where 100% labor pooling results in the lowest MFT and 0% labor pooling results in the highest MFT. If the level of machine pooling is 100%, the difference between 0% labor chaining and 100% labor chaining is significant as well as the difference between 33% labor chaining and 100% labor chaining. All other differences are not significant. The interaction effect shows that the size of the effect of labor chaining on MFT depends on the level of machine pooling. Under 0% MP, the MFT reduction when going from no LC to 100% LC is

6 Assignment Flexibility in a Cellular Manufacturing System - Machine Pooling versus Labor Chaining %, under 33% MP it is 6.7%, under 67% MP it is 2.9%, and under 100% MP it is 0.83%. Labor chaining is thus most beneficial in the case of no machine pooling. Most of the benefits are again achieved in going from no LC to 33% LC. Further, from the viewpoint of machine pooling, the MFT decrease in going from no MP to 100% MP is 39,1% under no LC, 28,3% under 33% LC, 25,9% under 67% LC, and 25.2% under 100% LC. Even though the effect of machine pooling on MFT diminishes when the level of labor chaining increases, machine pooling remains beneficial under all levels of labor chaining MFT: Mean Flow Time LC 0% 33% % 1.6 0% 33% 67% 100% 100% MP Figure 2: Interaction effect of labor chaining (LC) and machine pooling (MP) The interaction effect of labor utilization with machine pooling (LU x MP) is significant as well. The effect of machine pooling on MFT is significant (p < 0.05) within all levels of labor utilization. Further, pairwise comparisons using the Sidak adjustment for multiple comparisons showed that within each level of labor utilization, each level of machine pooling differs from each other. However, the effect of machine pooling on MFT seems to be larger if labor utilization is higher. This is mainly visible when going from no machine pooling to 33% machine pooling. A MFT decrease of about 30% is reached in going from no machine pooling to 100% machine pooling, irrespective of the level of labor utilization. The interaction effect of labor utilization with labor chaining (LU x LC) is significant and shows that the effect of labor chaining on MFT seems to be larger if labor utilization is higher. The effect of labor chaining is significant (p < 0.05) within all levels of labor utilization. In going from no labor chaining to 100% labor chaining, MFT decreases with 6,8% under 75% labor utilization, with 8,9% under 80% labor utilization, and with 10,1% under 85% labor utilization. Finally, we interpret the significant three-way interaction effect of labor utilization, machine pooling and labor chaining (LU x MP x LC) by means of simple interaction effects and simple, simple main effects. The interaction effect of machine pooling and labor chaining (MP x LC) is significant at 80% and 85% labor utilization. Under 75% labor utilization there is no significant interaction effect (p = 0.33) of machine pooling and labor chaining. The interaction effect of labor utilization with machine pooling (LU x MP) is only significant (p < 0.05) at 0% labor chaining. At 0% labor chaining, a MFT decrease of 36,3% is reached in going from no machine pooling to 100% machine pooling under 75% labor utilization, a MFT decrease of 38.7% under 80% utilization, and a MFT decrease of 41.3% under 85% labor utilization. At the other levels of labor chaining, the effect of machine pooling on MFT remains constant or even seems to decrease somewhat. Finally, the interaction effect of labor utilization with labor chaining (LU x LC) is only significant (p < 0.05) at 0% machine pooling. At 0% machine pooling, a MFT decrease of 14,1% is reached in going from no labor chaining to 100% labor chaining under 75% labor utilization, a MFT decrease of 18.7% under 80% utilization, and a MFT decrease of 22.9% under 85% labor utilization. The mean flow time results for labor chaining by machine pooling, and labor utilization are shown in Table 3. Pairwise comparisons of the simple, simple effects of labor chaining have been performed (α = 0.05) using the Sidak adjustment for multiple comparisons.

7 1056 Flexible Automation and Intelligent Manufacturing, FAIM2004, Toronto, Canada Table 3: Results for labor chaining by machine pooling, and labor utilization. Pairwise comparisons of the simple, simple effects of labor chaining have been performed (α = 0.05) using the Sidak adjustment for multiple comparisons. LU 75% LU 80% LU 85% MP LC MFT MP LC MFT MP LC MFT 100% 100% % 100% % 100% % % % % % % % % % % 100% % 100% % 100% % % % % % % % % % % 100% % 100% % 100% % % % % % % % % % % 100% % 100% % 100% % % % % % % % % % DISCUSSION AND CONCLUSIONS Confronted with imbalances between cell loads, managers may try to even the load of the cells by transferring jobs or workers, or both, between the cells. This paper compares the flow time effects of machine pooling versus labor chaining in a cellular manufacturing system. We performed a simulation study in which the model focused on five pairs of identical machines split between two manufacturing cells and three workers per cell able to operate these machines. Experimental factors were machine pooling (4 levels), labor chaining (4 levels), and labor utilization (3 levels). A major outcome of this study is that, within the setting studied, mean flow time is reduced more by machine pooling than by labor chaining. It is thus more beneficial to transfer jobs than to transfer workers between cells. An explanation for this is that machine pooling by itself may be beneficial for machine constraints as well as for labor constraints in a cell, while labor chaining by itself may only be beneficial for labor constraints in a cell. That is, if work queues up in front of a machine in a cell due to the non-availability of labor, one possibility is to transfer an additional worker from the other cell, but another possibility is to transfer jobs waiting in the queue to the identical machine in the other cell. If work queues up in front of a machine due to machine constraints (the machine is manned and there may even be idle workers in the cell), the only possibility is to transfer jobs waiting in the queue to the identical machine in the other cell. The outcomes further suggest that machine pooling will be worthwhile to consider irrespective of the level of labor chaining that is permitted. As we just argued, machine pooling has the additional advantage of dealing with machine constraints. By contrast, the influence of labor chaining on mean flow time is small in the case there already is some machine pooling. However, labor chaining results in significant mean flow time decreases (up to 22.9% under 85% labor utilization) if machine pooling is not allowed. Labor utilization only interacts with machine pooling if there is no labor chaining. Also, labor utilization only interacts with labor chaining if there is no machine pooling. In these cases, the effect of machine pooling or labor chaining (mainly the step of going from 0% to 33% MP or LC, respectively) on MFT is larger if labor utilization increases. Even though machine pooling seems to be the better option to reduce mean flow time, it may be easier to realize a worker transfer in practice. Within a practical situation, the ease of transferring workers versus the ease of creating a common queue for identical machines plays a role in the decision of what assignment flexibility to use. The results of this paper are based on a simulated setting with two cells including 5 pairs of identical machines and 6 workers. Despite the limitations of this setting, we believe that the general conclusions as presented in this

8 Assignment Flexibility in a Cellular Manufacturing System - Machine Pooling versus Labor Chaining 1057 section hold for a broader spectrum of situations, although this should obviously be investigated in future research. More complex and larger systems may be considered in which there may be identical machines not only within different cells, but also within a single cell. This may enhance the value of labor transfers, which may also depend on the staffing level in the cells. Further, with larger systems, routing flexibility could be extended to create more than two processing options per operation. Another issue is the effect of the number of routing steps per job (in this study, jobs only have one routing step) and the type of routing (job shop, flow) through the cell. For this, the unique machines within the cells must be included. Finally, we modeled full labor flexibility within cells in our study. Modeling limited flexibility probably influences the value of labor chaining. REFERENCES [1] M. Treleven: A review of dual resource constrained systems research, IIE Transactions, Vol.21, pp , [2] J.L. Burbidge, The introduction of group technology, Wiley and Sons, [3] N.C. Suresh: Partitioning work centers for group technology: insights from an analytical model, Decision Sciences, Vol.22, pp , [4] N.C. Suresh: Partitioning work centers for group technology: analytical extension and shop-level simulation investigation, Decision Sciences, Vol.23, pp , [5] N.C. Suresh and J.R. Meredith: Coping with the loss of pooling synergy in cellular manufacturing systems, Management Science, Vol.40, No.4, pp , [6] S.M. Shafer and J.M. Charnes: Offsetting lower routeing flexibility in cellular manufacturing due to machine dedication, International Journal of Production Research, Vol.35, No.2, pp , [7] C.L. Ang and P.C.T. Willey: A comparative study of the performance of pure and hybrid group technology manufacturing systems using computer simulation techniques, International Journal of Production Research, Vol.22, No.2, pp , [8] S. Benjaafar, J. Talavage and R. Ramakrishnan: The effect of routeing and machine flexibility on the performance of manufacturing systems, International Journal of Computer Integrated Manufacturing, Vol.8, No.4, pp , [9] W.C. Jordan and S.C. Graves: Principles on the benefits of manufacturing process flexibility, Management Science, Vol.41, pp , [10] M.J. Brusco and T.R. Johns: Staffing a multiskilled workforce with varying levels of productivity: An analysis of crosstraining policies, Decision Sciences, Vol.29, pp , [11] W.J. Hopp, E. Tekin and M.P. Van Oyen: Benefits of skill chaining in serial production lines with cross-trained workers, Management Science, Vol.50, No.1, pp.83-98, [12] R. Rochette. and R.P. Sadowski: A statistical comparison of the performance of simple dispatching rules for a particular set of job shops, International Journal of Production Research, Vol.14, No.1, pp.63-75, [13] A.M. Law and W.D. Kelton, Simulation Modeling and Analysis, McGraw-Hill, [14] G.V. Glass, P.D. Peckham and J.H. Sanders: Consequences of failure to meet assumptions underlying the fixed effects analysis of variance and covariance, Review of Educational Research, Vol.42, No.3, pp , [15] J. Stevens, Applied multivariate statistics for the social sciences, Lawrence Erlbaum Associates Inc.,1996.