Countering the negative impact of intercell flow in cellular manufacturing.
Cellular manufacturing (CM) is an application of group technology (GT) in which similar parts are grouped into part families and are separately processed in manufacturing subsystems called cells. Although a good deal of prior research has been devoted to the classification of parts into families or to the grouping of machines into cells (for example, see Burbidge (1971); McAuley (1972); Carrie (1973); King and Nakornchai (1982); Vakharia (1986); Wemmerlov and Hyer (1986); Kusiak (1987); and Seifoddini (1989)), there has been little research in the process design area that investigates the environments where cellular manufacturing performs better than does a traditional job shop using a process layout (Wemmerlov and Hyer (1987)).
In this paper, we discuss a comprehensive simulation study that tests various environmental attributes that impact relative performance differences between production in a traditional job shop mode and production in a CM mode. Unlike previous studies (see Flynn (1984); Flynn and Jacobs (1986, 1987); and Morris and Tersine (1990)), we explicitly model and test various levels of intercell flow, i.e., the proportion of operations that must be completed for a part outside its assigned cell. Wemmerlov and Hyer's (1989) survey of cellular manufacturing users found that the median level of intercell flow was 10% (with a mean of about 20%) and that only 10% of the surveyed shops processed parts completely within cells. It is also interesting to note that despite the benefits attributed to cellular manufacturing in this survey, almost half of the surveyed firms reported that cells constituted less than 5% of their operation. As firms continue to convert more of their process layout to cellular manufacturing, we believe that intercell flow will become increasingly problematic in practice since relatively fewer machines will be available for the cells, unless additional capital investment is made.
In cellular manufacturing, the reduction in setup time is often cited as a major contributing factor to a reduction in work-in-process inventory (WIP) and in mean flow time (MFT). This is because setups can be simplified by dedicating the machines in a cell to a part family with similar manufacturing attributes. However, when conversion to cellular manufacturing results in intercell flow, some parts must visit more than one cell, eliminating part of the setup time reduction resulting from dedication. At the extreme intercell flow level, the same degree of setups is incurred as in a job shop.
Several disadvantages may arise from the use of cellular manufacturing, including the need for additional machines or a loss of flexibility in dealing with product-mix changes--resulting from the increase in resource usage variance caused by dedication of specific machines to the manufacture of part families. In order to decrease the degradation in performance caused by machine dedication, improvements in other environmental attributes must occur (and often do in practice) such as reduced material handling times, reduced variability of run times, or reduced setup times. The conversion to cellular manufacturing may also result in increased operator responsibility and increased job satisfaction, both of which can increase product quality and worker productivity. These more qualitative benefits further counter performance degradation costs due to machine dedication and should be considered in practice. In our study, we first illustrate the negative impact of intercell flow when operating in a wide range of cellular manufacturing environments, then we indicate how changes in other operating factors caused by the conversion of a job shop to cellular manufacturing may counter the negative impact of intercell flow.
RECENT STUDIES ON CELLULAR MANUFACTURING PERFORMANCE
In earlier prior research of cellular manufacturing process design issues, Flynn (1984) and Flynn and Jacobs (1986, 1987) investigated the situation where a job shop with a process layout was converted to CM. In their studies, they evaluated three job shop environments for possible conversion to cells. They found that CM performed better in terms of average move (material handling) time and average setup time than the original job shops. However, the job shops performed better in terms of queue related variables, including average waiting time. The effect of waiting time outweighed the effects of move time and setup time in their study, resulting in job shops with better MFT and WIP performance than their cellular counterparts. Note that in their model most parts were required to visit many different cells, resulting in large amounts of intercell flow. Although they explicitly optimized facility layout (using CRAFT) and measured move distances, the effect of material handling time on MFT was relatively small in their models. They did not test various levels of intercell flow, nor did they test different levels of move times.
In a very recent paper, Morris and Tersine (1990) used a simulation model to test the effect of setup time levels, variance of part interarrival times, and material handling speed. Eight problem environments were tested for each of the process and cellular layouts. They assumed that all parts were processed within one cell, i.e., no intercell flow existed. Only one environment resulted in CM outperforming the process layout. Due to the limited information given on some of the model's parameter settings, it is difficult to determine the underlying cause of their results. For example, it is not possible to determine the contribution to MFT that is caused by material handling time. Other design issues may have biased CM performance in their study. Due to their assignment method of parts to cells and machines to cells, bottlenecks may have been designed assignment method of parts to cells and machines to cells, bottlenecks may have been designed into the cellular shop were none existed in the job shop. CM with an unbalanced load will likely result in poorer performance. Another reason for their experimental results to favor a process layout lies in their utilization levels. They state that their overall utilization levels were in the 60-70% range. If these levels existed for both the job shop and the cellular shop models, then the dominant job shop performance can be understood. When a job shop models, then average wait time, and, therefore, MFT will be relatively small even though setup times are much larger than those in the cellular shop.
Other research studies have implicitly considered intercell flow, but only as a result of alternate routings in the shop (Ang and Willey (1984); Gupta and Tompkins (1982)). In these studies, workloads from a congested machine in one cell were transferred to an alternate, less congested machine in another cell. Ang and Willey found that intercell transfers could be successfully used to mitigate the problems of workload imbalances in cellular shops. Gupta and Tompkins concluded that by not allowing alternate routings in cellular shops the impact of a short-term demand disturbance would be felt over a longer period of time.
Intercell flow is an important factor to consider when converting to CM, but it is not only the result of alternate job routings. When job shops are converted to cellular shops, some of the cells formed may not completely process all the products assigned to them (e.g., due to an insufficient number of a heavily demanded machine type required to achieve cell independence). In these situations, jobs must go outside their assigned cell to complete their processing. In our experiments, we explicitly control the level of intercell flow caused by the lack of processing capability within a cell. We extend the previous research by testing both the effect of intercell flow and the effect of machine dedication under a wide range of conditions.
Our objective in this research is twofold: 1) to illustrate the negative impact of intercell flow under a wide range of cellular operating environments, and 2) to indicate how improvements in other operating factors resulting from the conversion to CM can counter the negative impact of intercell flow. We study the environment where shifting bottlenecks  occur and where conversion to CM does not cause long-term bottlenecks where none existed in the job shop mode. That is, balanced loads (in terms of run time) are maintained on all machines in all cells and departments. We address the following questions in this research:
a. To what extent does the level of intercell flow affect cellular shop performance, as measured by MFT and WIP?
b. At what level of run time variability will cellular manufacturing provide improved performance over a traditional job shop?
c. Do constraints in batch size, perhaps necessary to achieve stable material flows or low WIP, affect the possible improvement in MFT when converting to cellular manufacturing from a job shop mode?
d. To what extent does the level of setup time affect the change in performance when a job shop is converted to a cellular shop?
e. To what degree must setup time be reduced for a conversion to cellular manufacturing to be beneficial?
f. Does the relative performance of cellular manufacturing improve when material handling time increases?
g. If the product-mix changes over time, will the conversion to cellular manufacturing result in poorer performance that if the firm stayed with a job shop manufacturing mode?
FACTORS AND EXPERIMENTAL DESIGN
A full factorial experiment was designed to answer a through e. The answers to questions f and g were obtained through sensitivity analyses discussed in the section titled "Sensitivity Analyses." The experiment included five factors (independent variables): level of intercell flow (four levels), major setup time (three levels), run time variability (four levels), batch size (four levels), and setup ratio (four levels). A major setup time is the machine setup time in a job shop or, in CM, the machine setup time incurred when a product requires processing outside its assigned cell. We call the machine setup time needed in CM for products processed in their assigned cell a minor setup time. The setup ratio is the ratio of minor to major setup time. The levels of each factor are included in Table 1. A more detailed discussion of our factors follows in the next five sections.
Four levels of intercell flow were considered in the cellular shop model: 0.0, 0.1, 0.2, and 0.3. At a level of 0.1, 10% of the operations (24 of our total of 240 operations) are performed on parts outside of their assigned cell, resulting in increased setup times, increased material handling, and increased overall congestion in the shop.
We limited moves to only one other cell, i.e., parts visit either one or two cells. As the number of cells necessary for processing a part increases, flow times also increase due to the move time associated with intercell transfers. We also limited the location of intercell interruption to the middle stations. Since we are testing balanced shops in our full factorial experiments, the
[TABULAR DATA OMITTED]
location of the interruption will have no effect on system performance. In practice, however, the combination of intercell flows and bottleneck locations should be investigated so that intercell interruptions, where possible, are designed to occur at the non-bottleneck stations.
Run Time Variability
Run time variability was specified at four levels of the coefficient of variation (CV): 0.0, 0.33, 0.67, and 1.00. We found that run time variability can be a good surrogate for variability from other sources (see Monahan and Smunt (1990); and Garza(1990)), such as from machine breakdowns, preventive maintenance, or rework. While it is true that machine breakdowns, for example, introduce forced idleness in the system rather than use capacity at varying levels (like that introduced by run time variability), aggregate performance effects are similar. The intent of this study does not concern the identification of sources of performance degradation due to variance, but rather the overall effect of variance on CM performance. Therefore, there is no need to separately model different sources of variance. However, in practice, it may be necessary to model each source of variance individually so that specific changes for CM performance improvement can be appropriately identified in the actual process.
We tested four levels of batch size: 10, 15, 20, and 25 units. Larger batch sizes lead to reduced total setup time in the shop for a given quantity of work, although it takes longer to process a larger batch. The mean processing time for a unit is 0.1 hours per operation for all products. Therefore, the mean batch processing time per operation is 1, 1.5, 2, or 2.5 hours for batch sizes of 10, 15, 20, and 25, respectively.
Major Setup Time
Three levels of major setup time were considered: 0.2, 0.4, and 0.6 hours. The ratio of major setup time to mean processing time used in this study ranges from .08 to .60. Values within this range have been used in previous research involving shop simulation, including Flynn (1984), Lee (1985), and Jacobs and Bragg (1988). Since setup time is a component of MFT, a decrease in setup time results in reduced MFT and WIP, making the manufacturing system more responsive, and reducing the need for a large finished goods inventory (if in a make-to-stock environment).
We defined setup ratio as the ratio of minor to major setup time. Therefore, it is a measure of the potential setup time savings that can be realized in a cellular shop after dedicating machines to the production of a family of products. The smaller the setup ration, the larger the reduction in setup times when a job shop is converted to a cellular shop. For example, setup ratio of 0.1 translates into a 90% reduction in setup time in the cellular shop each time a minor setup is needed. We tested four levels of setup ratio (ratio of minor to major setup time) in the main experiment: 0.1, 0.3, 0.6, and 0.9. A recent survey of CM users (Wemmerlov and Hyer (1989)) found setup ratios from 0.05 to 0.98, with an average of about 0.68. Therefore, the selected levels fall within actual practice.
Two simulations were run for each set of factors, one for a job shop and one for a cellular shop. The simulation models were written in SIMSCRIPT. a total of 60 products was processed in both models. Each product required four operations with a mean run time of 0.1 hours per operation. The interarrival rate of orders was adjusted to give a shop run time utilization of 60% for every combination of factor level tested. Total utilization (setup + run time) varied between 62% and 95% for the shops mdoeled in this work. This is consistent with typical values found in practice and with levels used in prior research studies. In order to achieve a 60% run time utilization, the demand rate per product was set at 1200 units per simulated year (2000 hours). (See Appendix A for specific assumptions of these simulation models.)
We first modeled the situation where the material handling time required to move a batch of parts between cells and between departments was 0.2 hours. This figure represents a range from 8% to 20% of the mean run time needed to process a batch. Ten observations were gathered for each combination of factors to reduce sample size errors. Therefore, the number of simulation runs for the cellular shop was (4 X 3 X 4 X 4 X 4 X 10) 7680 and for the job shop was (3 X 4 X 4 X 10) 480, for a total of 8160 runs in the full factorial experiment. Note that it was not necessary to tes the intercell flw and setup ratio factors in the job shop setting.
We further tested situations through a sensitivity analysis where material handling time was increased to five times the original value. In these tests, material handling time was equal to 40% to 100% of batch run time. although we do not treat material handling as a separate resource in our simulation models, we do consider reduction in material move times resulting from the use of cells. For this research, we decided to reduce the number of confounding effects so that the effect of setup times and machine dedication could be studied explicitly.
In testing for steady state conditions, we found that the initiatlization period of 36,000 completed jobs (about ten years of operation assuming operation of 2000 hours per year for a batch size of 20 units) was sufficient in all factor combinations. We used the batch means approach (see Law and Kelton (1991) for a detailed description of this method) to collect performance variables for ten observations of 3,600 completed jobs (about one simulated year for a batch size of 20 units). Each observation included sufficient completed jobs to esnsure independence from other observations. The simulation runs required approximately 150 CPU hours on a VAX 8810.
Job Shop Model Specifications
The job shop model groups 24 machines by type in eight departments (i.e., process layout). Each department includes three machines of the same type and has one queue for incoming jobs. Each product type has its own routing and visits four departments. Before visiting a given department, the job goins the department queue and waits for an available machine. When a machine becomes available it is set up for the product type of the next job in the queue (the First-Come, First-Serve (FCFS) priority rule) and the job is then processed. An incoming job searches for a machine already set up for that job (if any). The setup time is zero when the last product processed at a machine is the same as the next to be processed. When this is not the case, a major setup time is incurred. When a job finishes processing in a department, it is moved to the next department in its process routing.
Cellular Shop Model Specifications
The cellular shop model includes the same 24 machines as the job shop model. Four machines of different types are assigned to each of six cells, and each cell is dedicated to one part family. Each of the six part families includes ten parts. The cell size is within the range (four to six machines) used by about half of the cellular manufacturing users surveyed by Wemmerlov and Hyer (1989). Each machine has tooling designed for the part family. This will reduce setup times, but only when members of the assigned part family are processed. This reduced setup time is called a minor setup time. (Since the machines in the job shop are not dedicated, there are no minor setup times in that model.) However, when a dedicated machine processes a job which does not belong to its assigned part family, we assume a major setup time is incurred. As it is the case in the job shop model, the setup time is zero when the last part processed in a machine is the same as the next to be processed.
The intensity of intercell flows is increased by changing the process routings of some products in such a way that more and more products require processing in more than one cell (see Appendix B). This procedure allows us to control the level of intercell flow for each individual experiment. for parts that need processing outside their assigned cell, half of the operations are completed in the cell to which they are assigned, and the other half in a different cell. Each cell is a unidirectional flow line (i.e., no backflow) and each product visits either one or two cells. We kept the number of operations per part and the number of machines per cell equal in a further effort to eliminate long-term bottleneck conditions and confounding effects.
The main performance variables collected were mean flow time and work-in-process inventory. We define flow time as the time between a job arrival and the time when the job finishes processing. MFT is the mean flow time for all jobs finished within the simulation period. WIP is the time-weighted average number of unfinished units of any product type in the system during the simultion period. Both MFT and WIP reduction were selected among the most common reasons for establishing manufacturing cells in a recent survey of cellular manufacturing users (Wemmerlov and Hyer (1989)). In our studies, we found that MFT was highly correlated with WIP, and we do not report WIP results here. However, these results are given in detail in Garza (1990). The theoretical basis for the correlation between MFT and WIP observed in our experiments is given by Little (1961).
In order to determine if a given set of shop factors generated better performance for the cellular shop than for the job, a "percentage improvement in MFT' variable was calculated. For simplicity, we refer to this variable as PIMFT. PIMFT is defined as:
PIMFT = 100 * (MFT in job shop - MFT in cellular shop) / MFT in job shop
A value of PIMFT larger than zero implies that the conversion from job shop to cellular shop resulted in an improvement in MFT.
Analysis of Variance (ANOVA) was used to test significance of PIMFT. (2) All main effects and first order interactions were found to be significant at the .01 level. The Ryan-Einot-Gabriel-Welsch (REGW) multiple range F-test (Schlotzhauer and Littell (1987)) was run for each of the factors. REGW is a multi-stage test that controls the experiment-wise error rate. We chose a .05 experiment-wise significance level and found that, for each factor, these was a significant difference in PIMFT between any pair of levels considered.
The full factorial results are shown by each factor and intercell flow level in Table 2A. Note that only six cells (first-order interactions) in this table indicate positive PIMFT. Five cells appear in the 0.0 intercell flow column for low values of setup ratio, batch size, CV, and for the highest value of major setup time. One occurrence appears in the 0.1 intercell flow column for the lowest CV level. We do not conclude from these results that conversion to CM does not have the potential of improving system performance. Rather, it is important to investigate higher order interactions to gain insight into the types of improvements that must occur in environmental attributes when converting to CM for such conversion to improve performance. Note that the number of cells that indicate positive PIMFT in Table 2B more than doubles, where average PIMFT data is shown only for a setup ratio = 0.1. Clearly, results from any simulation study are driven by the model specifications and the factor levels tested. We believe that our model and choice of factor levels are reasonable, but we also note that it is critical to analyze the detailed data for further explanation of relative system performance.
TABLE 2 PIMFT AVERAGE (*) RESULTS A. Full Factorial Results INTERCELL FLOW FACTOR LEVEL 0.0 0.1 0.2 0.3 AVG. 0.1 +3.0 -8.5 -18.2 -26.5 -12.6 SETUP 0.3 -3.6 -14.5 -23.6 -31.1 -18.2 RATIO 0.6 -15.2 -25.8 -33.8 -40.0 -28.7 0.9 -31.7 -41.5 -47.5 -51.8 -43.1 10 +9.3 -3.7 -14.2 -23.2 -8.0 BATCH 15 -12.3 -23.7 -32.3 -39.2 -26.9 SIZE 20 -19.9 -29.7 -37.0 -42.8 -32.4 25 -24.6 -33.2 -39.6 -44.3 -35.4 0.00 +17.1 +4.1 -4.5 -10.6 1.5 CV 0.33 +2.8 -8.6 -17.5 -24.4 -11.9 0.67 -19.9 -29.7 -38.1 -45.3 -33.3 1.00 -47.6 -56.1 -63.1 -69.2 -59.0 MAJOR 0.2 -24.1 -31.8 -37.6 -41.7 -33.8 SETUP 0.4 -14.0 -25.2 -33.8 -40.3 -28.3 TIME 0.6 +2.5 -10.7 -21.0 -30.1 -14.8 B. SETUP RATIO = 0.1 (only) INTERCELL FLOW FACTOR LEVEL 0.0 0.1 0.2 0.3 AVG. SETUP 0.1 +3.0 -8.5 -18.2 -26.5 -12.6 RATIO 10 +28.8 +16.5 +5.0 -5.4 11.2 BATCH 15 +4.0 +8.5 +18.7 +27.7 -12.7 SIZE 20 -7.1 -18.0 -27.3 -34.7 -21.8 25 -13.9 -24.0 -31.9 -38.0 -27.0 0.00 +26.5 +13.9 +4.6 -2.6 10.6 CV 0.33 +15.3 +3.8 -6.2 -14.4 -0.4 0.67 -3.3 -14.2 -24.3 -33.5 -18.8 1.00 -26.8 -37.6 -46.9 -55.4 -41.7 MAJOR 0.2 -15.7 -24.6 -31.5 -37.1 -27.2 SETUP 0.4 +1.8 -10.7 -21.1 -29.8 -15.0 TIME 0.6 +22.8 +9.7 -2.1 -12.5 4.5 (*) PIMFT values first calculated on the individual simulation results for each combination of factor levels and then averaged for each factor level presented.
When looking at the detailed experimental data (768 cellular shop operating environments, i.e., combinations of factor levels), 158 operating environments (21%) resulted in a positive PIMFT. Furthermore, of these 158 environments, 90 resulted in a PIMFT greater than 10% upon conversion to celular manufacturing. The combinations of factor levels that resulted in CM outperforming the job shop typically occurred for zero or low intercell flow. Of the positive PIMFTs, 51.3% were for scenarios with no intercell flow, 27.8% were for the 0.1 intercell flow case, 13.3% were for the 0.2 intercell flow case, and 7.6% were for the 0.3 intercell flow case.
The level of intercell flow resulting from a conversion to CM is typically a result of initial conditions (number of machines, similarity of parts, etc.) and of the willingness of management to make investments in additional equipment or part redesign. Resulting intercell flow can also be minimized by using larger cell sizes, e.g., using two large cells versus four smaller cells. However, trade-offs exist with this option--large setup time reductions cannot be expected when many different types of parts are processed in the same cell, nor can one expect that operation times will be similar.
In the next four sections, we illustrate specific higher-order interactions and discuss the necessary changes that must occur to various environmental factors for a conversion to CM to prove beneficial.
Run Time Variability
The results of first-order interactions between run time variability and intercell flow are shown in Table 2A. On the average, PIMFT for a processing time variability (CV) of 0.0 (i.e., deterministic processing times) was 13.4, 34.8, and 60.5 percentage points (3) high than PIMFT for CVs of 0.33, 0.67, and 1.0, respectively. When run time variability increases, shifting bottlenecks form, resulting in an increase in MFT. Although MFTs increase with increasing run time variability for both the job shop and the cellular shop, the cellular shop is more sensitive to run time variability than is the job shop. In the job shop, a part may be processed on any of the multiple machines of the same type, providing a reduction in run time variance. When a temporary higher level of demand is placed on one cell, perhaps due to random market forces or random processing times, some machines in this cell will see long queues of WIP while similar machines in other cells remain idle. (See Moily and Stinson (1989) for some experimental results on parallel processing capabilities in a flow shop versus CM with dedicated machines.) Since we do not allow alternate routings in the cellular shop model, the shifting bottleneck problem can cause the performance of CM to be worse than a job shop.
Figure 1 further illustrates the extent of the effect of run time variability across the four levels of intercell flow. (In this and following figures, we show effects where the setup ratio is 0.3, the major setup time is 0.4 hours, and the batch size is ten units since this is an environment that has moderate settings of factor levels and well illustrates the conditions where conversion to CM results in both positive and negative improvements.) When the CV is 0.0 in both the job shop and CM, CM results in extremely high performance improvements for intercell flow levels of 0.0 and 0.1. It remains positive for intercell flow of 0.2, and is slightly negative for intercell flow of 0.3. As the CVs increase simultaneously in both shop modes, the relative performance of CM degrades and is negative for all levels of intercell flow for the high CV level of 1.0.
However, Figure 2 indicates that if a conversion to CM results in lowering the CV, CM is attractive across all intercell levels for cell CVs of 0.0 and 0.33 assuming that the job shop CV remains at 1.0. Even if the cell CV can only be reeduced from 1.0 to 0.67, the CM shop outperforms the job shop for intercell flow levels of 0.0 and 0.1 and has nearly the same performance as the job shop for an intercell flow level of 0.2. That is, when a good deal of variability exists in a job shop and a conversion to a cellular shop reduces that variability, MFTs can decrease even when the conversion results in some intercell flow.
PIMFT for shops operating with a batch size of ten units was, on the average, 18.9, 24.4 and 27.4 percentage points higher than PIMFT results for shops operating with batch sizes of 15, 20 and 25 units, respectively (Table 2A). Job shops are inherently better than cellular shops at producing large batch sizes. Cellular shops can better produce small batch sizes due to lower setup times. For the more detailed factor level combination presented in Figure 3, we note the large difference (about 13 percentage points) between the PIMFT results for batch sizes of 10 and 15 units. This difference is due mainly to a steep incresae in MFT in the job shop caused by decreasing the batch size from 15 to 10 units, since a very high shop utilization level (about 95%) is reached in the job shop when producing in small batches.
Figure 4A illustrates the effect of converting to CM and of reducing batch size simultaneously. When comparing to a job shop producing jobs of size 25 units, CM performance is substantially better when the batch size in the cellular shop is reduced to 10 or 15 units for all levels of intercell flow. We believe that this particular comparison may be an unfair one since the batch size must be the same in both the job shop or cellular shop, if the choice of batch size is marketing driven. The comparison made in Figure 3 are then more appropriate. However, if a firm is able to use any batch size that maximizes performance, then comparisons of the two modes should be made with batch sizes that separately optimize performance for each system. Generally, we found that in our model a batch size of 15 or 20 was best for the job shop. Figure 4B shows that CM resulted in improved performance for most intercell flow levels when its batch size was 10 or 15. When can conclude from these results, that the ability to reduce batch size or the marketing need for batch size reduction is a powerful argument for setup time reduction made possible by conversion to CM.
PIMFT for a major setup time of 0.6 hours was, on the average, 13.5 and 19.0 percentage points higher than PIMFT for major setup times of 0.4 and 0.2 hours, respectively (Table 2A). Decreasing the setup time simultaneously in both modes of production has a detrimental effect on PIMFT. Conversion to CM is most beneficial when the setup times require a large proportion of the machines' utilization in the job shop.
In Figure 5, we illustrate the effects of major setup time for the moderate parameter settings to indicate that conversion to CM can result in positive PIMFT even when the major setup time is fairly low. However, the level of intercell flow after conversion must also be low for this to occur. Since the job shop is running smoothly with low major setup times, conversion to CM provides little potential for improvement. The problems associated with dedicating equipment in CM tend to dominate in producing a negative effect on MFT under these conditions. When major setup times are moderate to high, CM outperforms the job shop even with low to moderate levels of intercell flow.
Overall PIMFT for a setup ratio of 0.1 was 5.6 percentage points higher than PIMFT for a ratio of 0.3, 16.1 percentage points higher than PIMFT for a ratio of 0.6, and 30.5 percentage point higher than PIMFT for a ratio of 0.9 (Table 2A). Capitalizing on setup time reductions is one of the main advantages of cellular manufacturing, and results of this research substantiate this notion. Therefore, production managers of traditional job shops operating with a process layout should be aware that a sizable reduction in setup times must occur before a conversion to cellular manufacturing typically should be considered an option to improve shop performance.
The Effect of Material Handling Time
In our previous experiments, the handling time required to move a batch of parts between departments (in the job shop) and between cells (in the cellular shop) was assumed constant and equal to 0.2 hours. The ratio of handling time to job run time per operation is a function of batch size and varied between 8% and 20%. Although this range was consistent with past research in this area and seems reasonable, we tested the effect of increasing material handling time to five times the original level for the moderate factor levels (setup ratio=0.3, major setup time=0.4, and run time CV = 0.33). We did this to illustrate the queuing effect on flow time that might occur due to limited material handling capacity.
t-tests indicated that the PIMFT results at the base case of material handling time (0.2) are significantly different at the .05 level from the PIMFT results at the higher level of material handling time (1.0). Figure 6 includes PIMFT results for both material handling time levels. Note that for the base case level of 0.2 (Figure 6A) only small batch size scenarios show a positive PIMFT. However, when the material handling time increases (Figure 6B), a majority of the scenarios indicate that a conversion to CM improves MFT performance, even those with high intercell flow. Therefore, a large reduction in handling times that ensure from a conversion to cellular manufacturing might be ample by itself to counter the negative impact of intercell flow. CM performance increases as either setup or material handling time improves.
The Effect of Product-Mix Instability
In our main experiments, we assumed that the expected demand for each product was constant and that the shops were always operating with balanced loads. Changes in product-mix may occur in practice, resulting in workload imbalances. These workload imbalances will develop long-term (versus shifting) bottlenecks and poor system performance.
We expect the negative impact of workload imbalances to be more pronounced in the cellular shops that in the job shops. Due to dedication of machines to part families in a cellular shop, machines of the same type (but located in different cells) may work at very different utilization levels. This cannot occur in a traditional job shop since similar machines are grouped and share the load of a department.
An experiment was designed to explicitly study the effect of product-mix changes on MFT in both a cellular and a job shop setting. The shop operating conditions considered include a batch size of 15 units, a setup time of 0.6 hours, a CV of 0.33 and a setup ratio of 0.3. For this condition, conversion to cellular manufacturing was found to be beneficial under a stable product-mix. To test for the effect of product-mix variations, we increased the average demand of three families (to a "high" demand level) while decreasing the average demand of the other three families (to a "low" demand level) by the same amount, keeping total demand unchanged. The change in the demand of each family constitutes a product-mix change which we measure by a demand ratio, defined as the ratio of high demand to low demand. The larger the demand ratio, the larger the product-mix change. In this experiment, we stated with a scenario where the cellular shop performed better than the job shop, then gradually unbalanced the product-mix in order to study the MFT performance of both shops.
Figure 7A shows the MFT results as the product-mix becomes unbalanced. When the product-mix is balanced (i.e., demand ratio = 1), the cellular shop has better MFT performance than the job shop in this environment. As the demand ratio is increased. MFT performance degrades in both shops. However, the rate of MFT increase with demand ratio is much lower in the job shop. The difference in slopes results in a crossover of the MFT curves at a demand ratio of about 1.9. Figure 7B illustrates the maximum department run time utilization % for a job shop and the maximum cell run time utilization % for a cellular shop as the demand ratio increases. In the cellular shop, a cell working on a part family with high demand will experience sharply increased process utilization. However, in the job shop, a department visited by one or more parts with high demand also experiences increased process utilization, but at a smaller rate. Since the department has three machines, it is not only visited by parts with high demand, but also by parts with low demand which dampens the utilization effect of the product-mix change.
The above evidence substantiates that job shops are better prepared to sustain changes in product mix than are cellular shops, unless provisions are made to allow alternate routings without great increases in major setup times or additional equipment capacity is purchased. Otherwise, cellular shops may be preferred only under conditions of stable product-mix or minor product-mix changes.
Earlier in this paper we described the objective of this research and posed seven questions to help direct our research. We conclude with summary answers to these questions.
a. When a conversion to cellular manufacturing results in intercell flow, performance of the cellular system will likely be worse than that of a traditional job shop with a process layout. Our simulation results confirm that even small amounts of intercell flow can have a substantial negative impact on mean flow times (and WIP levels) for many conditions, especially those associated with high run time variability and large batch sizes.
b. By reducing run time variability by as little as one third the amount in the job shop, cellular manufacturing can outperform a job shop for low intercell flow levels and moderate levels of other operating factors. Further, as run time variability is reduced by 50% or more, cellular manufacturing outperforms the job shop for all intercell flow levels tested under the moderate operating condition.
c. Cellular shops are best suited for small batch size production. This can be expected since the setup times will be lower in the cellular environment. This effect was small, however, when both the job shop and cellular shop were constrained to use the same batch size. In this case, cellular manufacturing outperformed the job shop for small batch sizes and low intercell flow levels. In the situation where batch sizes were large in the job shop and a conversion to cellular manufacturing was accompanied by smaller batch sizes, the performance of the cellular system further improved for all levels of intercell flow.
d. Cellular shops performed better than did job shops when high setup times existed in the job shop. Again, this result is expected since a main advantage of cellular manufacturing is the reduction of existing setup times. However, we also observed that for the moderate operating conditions, a one third reduction of major setup times (from 0.6 to 0.4 hours in our experiment) could result in the job shop going from extremely poor relative performance to better performance than that of a cellular shop for medium to high intercell flow levels.
e. The proportion of setup time that can be reduced by cellular manufacturing is also an important factor in resulting performance. In our experiments, few environments indicated attractive cellular manufacturing performance. Of these, environments with no intercell flow required setup time reductions of 40% (i.e., a setup ratio of 0.6) or better upon conversion to cellular manufacturing. Furthermore, cellular environments with even small amounts of intercell flow usually required setup time reductions of 70% or better. Therefore, sizeable reductions in setup time are needed to mitigate the negative impact of intercell flow.
f. When material handling time is quite high in a job shop, the conversion to cellular manufacturing becomes attractive at most intercell flow levels, especially when producing small batch sizes. If material handling times are a small proportion of MFT, as in our full factorial experiment, only those conditions with small batch size and no or little intercell flow favored cellular manufacturing.
g. The performance of a job shop was found to be more stable than that of a cellular shop when both shops were subjected to a wide range of changes in product-mix. The ability to process a part in any one of several machines in a department in the job shop provides a flexibility that is lost when converting to cellular manufacturing.
In summary, the results of our study indicate that although the existence of intercell flow has a negative effect on cellular manufacturing performance, improvements in other operating factors can counter this negative impact. As Greene and Sadowski (1984) pointed out a few years back, ". . . design or redesign of a job shop to a cellular manufacturing system remains rather difficult and theoretical." Our research does not provide a cookbook approach for converting to cellular manufacturing--each situation in practice will be unique and require its own specific (simulation) analysis. However, our results do illustrate the need to specifically consider a wide range of factors in evaluating the conversion to cellular manufacturing.
In this research we considered move times only, but we did not consider the possible queue times that could occur in shops with substantial material handling equipment constraints. A comprehensive dual resource experiment must be designed to test the effect of constrained material handling resources. In this environment, it is likely that we will find conditions where job shop material handling times are substantially larger than the ones tested in this research, giving further advantage to cellular manufacturing. Additionally, there is a need to test the effect of adding machines to a cell, at an additional investment, as a way to increase its performance.
The FCFS rule was used in the main experiment of this research in both the job shop and cellular shop settings since it provides an upper bound on MFT performance and is directly related to mean lateness, a due date performance measure. We expect, however, that the best scheduling rule in a job shop is different from that in a cellular shop, introducing some bias into our results. For example, there are several group scheduling rules (i.e., a dispatching rule plus a part family queue selection rule) that have been found to outperform traditional dispatching rules in cellular shops (see Mosier, Elvers, and Kelly (1984); Mahmoodi, Dooley, and Starr (1990); and Wemmerlov and Vakharia (1991)).
Ideally, comparisons of cellular and job shop manufacturing performance should be under optimum environments for both. The components of an optimum environment include the layout, the scheduling heuristic, the level of equipment investment, and the batch batch size (if not constrained by external factors). Our reported results are based on the investigations of stable demand conditions and on one condition where product-mix changes. Future research in comparing cellular manufacturing to job shop environments should continue to investigate optimum conditions for either environment and the effect of dynamic product-mixes. Clearly, numerous avenues of cellular manufacturing design research remain untraveled.
(1) These short-term bottlenecks are described in detail and have been called "inplicit shocks" by Monahan and Smunt (1990).
(2) We tested the normality and homogeneity assumptions for ANOVA. Using Cochran's test, we could not confirm the homogeneity of variance assumption for an F-test. However, we plotted the variances from each cell in the experimental design and found that less than 3% of these variances were above the critical level for the Cochran test. All these variances belonged to experiments with high run time variability. However, the F distribution is robust with respect to the violation of the assumption of homogeneity of error variance, provided that the number of observations in the experimental cells is equal, as it is in our case (see Kirk (1968)). We also ran [x.sup.2] tests on the residuals to determine whether or not they came from a normal distribution. Although we obtained a high [x.sup.2] value (significantly different), by observation of the plot of the residuals we found that they were distributed in a leptokurtic manner. Transformations helped reduce concentration of residuals near 0.0, but did not lower the [x.sup.2] sufficiently. Based on the known robustness of the F-test and the fact that our p-values were quite low (.0001 in most cases), we feel that we have significant results.
(3) Percentage points are the difference between two PIMFT values. For example, in Table 2A the average PIMFT for a CV of 0.0 = 1.5%, and for a CV of 0.33 = .11.9%. Therefore, we say that PIMFT for a CV of 0.0 is, on the average, 13.4 [1.5 - (-11.9)] percentage points higher than PIMFT for a CV of 0.33.
Ang, C.L., 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, March/April 1984, 193-233.
Burbidge, J.L. "Production Flow Analysis." The Production Engineer, vol. 50, nos. 4-5, April/May 1971, 139-152. Carrie, A.S. "Numerical Taxonomy Applied to Group Technology and Plant Layout." International Journal of Production Research, vol. 11, no. 4, October 1973, 399-416.
Dudley, N.A. "Work-Time Distributions." International Journal of Production Research, vol. 2, no. 2, June 1963, 137-144.
Flynn, B.B. Group Technology versus Process Layout: A Comparison Using Computerized Job Shop Simulation. Unpublished doctoral dissertation, Indiana University, Bloomington, IN, 1984.
Flynn, B.B., and F.R. Jacobs. "An Experimental Comparison of Cellular (Group Technology) Layout With Process Layout." Decision Sciences, vol. 18, no. 4, Fall 1987, 562-581.
Flynn, B.B., and F.R. Jacobs. "A Simulation Comparison of Group Technology with Traditional Job Shop Manufacturing." International Journal of Production Research, vol. 24, no. 5, September/October 1986, 1171-1192. Garza, O. The Effect of Intercell Flow on the Performance of Cellular Manufacturing Systems. Unpublished doctoral dissertation, Washington University, St. Louis, MO, USA, 1990.
Greene, T.J., and R.P. Sadowski. "A Review of Cellular Manufacturing Assumptions, Advantages and Design Techniques." Journal of Operations Management, vol. 4, no. 2, February 1984, 85-97.
Gupta, R.M., and J.A. Tompkins. "An Examination of the Dynamic Behavior of Part-Families in Group Technology." International Journal of Production Research, vol. 20, no. 1, January/February 1982, 73-86.
Jacobs, F.R., and D.J. Bragg. "Repetitive Lots: Flow-Time Reductions through Sequencing and Dynamic Batch Sizing." Decision Sciences, vol. 19, no. 2, Spring 1988, 281-294.
King, J.R., and V. Nakornchai. "Machine-Component Group Formation in Group Technology: Review and Extension." International Journal of Production Research, vol. 20, no. 2, March/April 1982, 117-133.
Kirk, R.E. Experimental Design: Procedures for the Behavioral Sciences, Brooks/Cole Inc., Belmont, California, 1968. Kusiak, A. "The Generalized Group Technology Concept." International Journal of Production Research, vol. 25, no. 4, April 1987, 561-569.
Law, A.M., and Kelton, D.W. Simulation Modeling and Analysis, McGraw-Hill, Inc., New York, New York, 1991.
Lee, L.C. "A Study of System Characteristics in a Manufacturing Cell." International Journal of Production Research, vol. 23, no. 6, November/December 1985, 1101-1114.
Little, J.D.C. "A Proof for the Queuing Formula: L = [lambda]W." Operations Research, vol. 9, no. 3, May/June 1961, 383-387.
Mahmoodi, F., K.J. Dooley, and P.J. Starr. "An Investigation of Dynamic Group Scheduling Heuristics in a Job Shop Manufacturing Cell." International Journal of Production Research, vol. 28, no. 9, September 1990, 1695-1711.
McAuley, J. "Machine Grouping for Efficient Production." The Production Engineer, vol. 51, no. 2, February 1972, 53-57.
Moily, J.P., and J.P. Stinson. "Effects of Processing Time Variability and Routing Flexibility on Group Technology Implementation in Flow-Shops." In Proceedings of the 1989 Decision Sciences Institute, Atlanta, GA: Decision Sciences Institute, 1989, 898-901.
Monahan, G.E., and T.L. Smunt. "Product-Process Relations in Batch Manufacturing." Working paper, University of Illinois, July 1990.
Morris, J.S. and R.J. Tersine. "A Simulation Analysis of Factors influencing the Attractiveness of Group Technology Cellular Layouts." Management Science, vol. 36, no. 12, December 1990, 1567-1578.
Mosier, C.T., Elvers, D.H. and Kelly, D. "Analysis of Group Technology Scheduling Heuristics." International Journal of Production Research, vol. 22, no. 5, September/October 1984, 857-875.
Schlotzhauer, S.D., and R.C. Littell. SAS System for Elementary Statistical Analysis, SAS Institute Inc., 1987.
Seifoddini, H. "A Note on the Similarity Coefficient Method and the Problem of Improper Machine Assignment in Group Technology Applications." International Journal of Production Research, vol. 27, no. 7, July 1989, 1161-1165.
Vakharia, A.J. "Methods of Dell Formation in Group Technology: A Framework for Evaluation." Journal of Operations Management, vol. 6, no. 3, May 1986, 257-271.
Wemmerlov, U., and N.L. Hyer. "Cellular Manufacturing in the U.S. Industry: A Survey of Users." International Journal of Production Research, vol. 27, no. 9, September 1989, 1511-1530.
Wemmerlov, U., and N.L. Hyer. "Research Issues in Cellular Manufacturing." International Journal of Production Research, vol. 25, no. 3, March 1987, 413-431.
Wemmerlov, U., and N.L. Hyer. "Procedures for the Part Family/Machine Group Identification Problem in Cellular Manufacturing." Journal of Operations Management, vol. 6, no. 2, February 1986, 125-147.
Wemmerlov, U., and A.J. Vakharia. "Job and Family Scheduling of a Flow-Line Manufacturing Cell: A Simulation Study." IIE Transactions, vol. 23, no. 4, December 1991.
ASSUMPTIONS OF THE MODELS
The main assumptions of both models (unless otherwise noted) are:
(1) Job orders of constant size arrive deterministically to the shop. The product type that each job order represents is sampled from the uniform distribution across the whole product-mix (60 products). Therefore, even though the expected total demand for each product during a simulation period is constant, job order arrivals for each product are random.
(2) A First-Come First-Served (FCFS) scheduling rule is used to select jobs from machine (cellular shop) and department (job shop) queues.
(3) Instead of batching orders, it is assumed that the company releases job ordes as received. We do this to be able to explicitly determine the effect of batch size on system performance.
(4) Both major and minor setup times are deterministic. However, the run time per job is stochastic with a Gamma distribution. This distribution is positively skewed for coefficients of variation smaller than one, which correlates to empirical evidence on unpaced task times presented by Dudley (1963). Furthermore, sampling from the Gamma distribution results in non-negative run times.
(5) Handling time is deterministic. Incremental handling time is incurred when jobs are transported from one department to another in the job shop or from one cell to another in the cellular shop. Therefore, in the cellular shop, total handling time is a function of the level of intercell flow. It is further assumed that the time penalty for transporting jobs between departments in the job shops and between cells in the cellular shop is identical. Note that there are no handling requirements for jobs moving within a cell.
(6) There are no alternate routings in the cellular shop. Each product type has only one process routing. In the job shop, any machine of the same type can process a job which requires that machine type.
NOTE: These layouts are not intended to represent the exact physical layouts of the shops but only the machines included in each cell (cellular shop) and in each department (job shop). In the job shop, machines within the same department are similar. For example, machines M5, M13, and M19 in department G are similar. However, when the job shop is converted to a cellular shop, similar machines are included in different cells and dedicated to different families.
Product routings are included for each product (or group of products with the same routing). Machine numbers refer to the above layouts. Product routings included are for the cellular shop. Routings for the job shop are the same except that a part which needs to be processed in a machine included in one department, may be processed in any other machine within the same department (i.e., in any other machine of the same type).
Pij = product number j assigned to cell number i--(cellular shop) Pij = product number (10i + j)--(job shop) (i.e., P35 is product 5 assigned to cell 3 in the cellular shop and in product number 35 in the job shop).
PRODUCT(S) ROUTING P10-P19 M1 M2 M3 M4 P20-P29 M5 M6 M7 M8 P30-P39 M9 M10 M11 M12 P40-P49 M13 M14 M15 M16 P50-P59 M17 M18 M19 M20 P60-P69 M21 M22 M23 M24 PRODUCT ROUTINGS FOR LEVEL OF INTERCELL FLOW = 0.0 PRODUCT(S) ROUTING P10 M1 M2 M7 M8 P11 M9 M10 M3 M4 P12-P19 M1 M2 M3 M4 P20 M5 M6 M11 M12 P21 M1 M2 M7 M8 P22-P29 M5 M6 M7 M8 P30 M9 M10 M3 M4 P31 M5 M6 M11 M12 P32-P39 M9 M10 M11 M12 P40 M17 M18 M15 M16 P41 M13 M14 M23 M24 P42-P49 M13 M14 M15 M16 P50 M21 M22 M19 M20 P51 M17 M18 M15 M16 P52-P59 M17 M18 M19 M20 P60 M13 M14 M23 M24 P61 M21 M22 M19 M20 P62-P69 M21 M22 M23 M24 PRODUCT ROUTINGS FOR LEVEL OF INTERCELL FLOW = 0.1 PRODUCT(S) ROUTING P10 M1 M2 M7 M8 P11 M9 M10 M3 M4 P12 M1 M2 M11 M12 P13 M5 M6 M3 M4 P14-P19 M1 M2 M3 M4 P20 M5 M6 M11 M12 P21 M1 M2 M7 M8 P22 M5 M6 M3 M4 P23 M9 M10 M7 M8 P24-P29 M5 M6 M7 M8 P30 M9 M6 M7 M8 P30 M9 M10 M3 M4 P31 M5 M6 M11 M12 P32 M9 M10 M7 M8 P33 M1 M2 M11 M12 P34-P39 M9 M10 M11 M12 P40 M17 M18 M18 M16 P41 M13 M14 M23 M24 P42 M21 M22 M15 M16 P43 M13 M14 M19 M20 P44-P49 M13 M14 M15 M16 P50 M21 M22 M19 M20 P51 M17 M18 M15 M16 P52 M13 M14 M19 M20 P53 M17 M18 M23 M24 P54-P59 M17 M18 M19 M20 P60 M13 M14 M23 M24 P61 M21 M22 M19 M20 P62 M17 M18 M23 M24 P63 M21 M22 M15 M16 P64-P69 M21 M22 M23 M24 PRODUCT ROUTINGS FOR LEVEL OF INTERCELL FLOW = 0.2 PRODUCT(S) ROUTING P10 M1 M2 M7 M8 P11 M9 M10 M3 M4 P12 M1 M2 M11 M12 P13 M5 M6 M3 M4 P14 M1 M2 M15 M16 P15 M13 M14 M3 M4 P16-P19 M1 M2 M3 M4 P20 M5 M6 M11 M12 P21 M1 M2 M7 M8 P22 M5 M6 M3 M4 P23 M9 M10 M7 M8 P24 M5 M6 M19 M20 P25 M17 M18 M7 M8 P26-P29 M5 M6 M7 M8 P30 M9 M10 M3 M4 P31 M5 M6 M11 M12 P32 M9 M10 M7 M8 P33 M1 M2 M11 M12 P34 M9 M10 M23 M24 P35 M21 M22 M11 M12 P36-P39 M9 M10 M11 M12 P40 M17 M18 M15 M16 P41 M13 M14 M23 M24 P42 M21 M22 M15 M16 P43 M13 M14 M19 M20 P44 M13 M14 M3 M4 P45 M1 M2 M15 M16 P46-P49 M13 M14 M15 M16 P50 M21 M22 M19 M20 P51 M17 M18 M15 M16 P52 M13 M14 M19 M20 P53 M17 M18 M23 M24 P54 M17 M18 M7 M8 P55 M5 M6 M19 M20 P56-P59 M17 M18 M19 M20 P60 M13 M14 M23 M24 P61 M21 M22 M19 M20 P62 M17 M18 M23 M24 P63 M21 M22 M15 M16 P64 M21 M22 M11 M12 P65 M9 M10 M23 M24 P66-P69 M21 M22 M23 M24 PRODUCT ROUTINGS FOR LEVEL OF INTERCELL FLOW = 0.3
|Printer friendly Cite/link Email Feedback|
|Title Annotation:||Special Issue on Group Technology and Cellular Manufacturing|
|Author:||Garza, Oscar; Smunt, Timothy L.|
|Publication:||Journal of Operations Management|
|Date:||Jan 1, 1991|
|Previous Article:||A tool cluster based strategy for the management of cutting tools in flexible manufacturing systems.|
|Next Article:||Overloaded overheads: activity-based cost analysis of material handling in cell manufacturing.|