Load smoothing by the planning and order review/release systems: a simulation experiment.
Most research studies investigating the impact of Order Review/Release (ORR) mechanisms on shop performance have focused strictly on managing the flow of orders to the shop floor once they have been generated by the planning system. This approach ignores the possibility that the manufacturing planning system may be managing the incoming work flow so that peaks and valleys in the work load have been smoothed out.
This study shows, via a simple computer simulation of a random job shop, that such smoothing by the planning system can improve system performance and enhance the effects of Order Review/Release. The two "filtering" mechanisms of planning system smoothing and Order Review/Release have a complementary impact on the system, with smoothing working to reduce flow time and flow time variability and ORR working to reduce work in process and work in process variability. The combination of smoothing with ORR results in shorter and more consistent lead times, lower and more stable work-in-process levels, and better delivery performance. This results in a system which is very stable and predictable.
Further, the study shows that the combined effects of smoothing and ORR can improve the performance of simple shop floor dispatching rules like first come-first served to the point where they are competitive with more sophisticated, due date oriented rules. This raises the possibility of simplifying control mechanisms on the shop floor by doing a better job of work load planning and order release.
Within the context of the shop floor control (SFC) system, the order review/release (ORR) system occupies a special place. It is the interface between the manufacturing planning system and the shop floor. It controls the flow of work between the planning system and the shop floor. To many managers, an effective ORR system is a critical element in a successful SFC system.
However, the ORR system is at the center of a controversy, To some writers familiar with the practice of shop floor control (e.g., Harty (1969); Melnyk and Carter (1987); Nicholson and Pullen (1972); Sandman and Hays (1980); Wight (1970)), ORR plays a critical role in the operation of the SFC system. Yet many researchers (e.g., Baker (1984); Bertrand( 1983); Ragatz (1985)) have found the evidence in support of ORR to be weak at best. Several studies (e.g., Melnyk and Ragatz (1989)) have tried to resolve this controversy by developing and presenting several paradigms for the study of ORR mechanisms. Their success has been limited.
Most of the past studies on the operation of ORR mechanisms have ignored the nature of the planning system and the work schedules it generates. Typically, these schedules have been generated by a stochastic process and exhibit a high degree of variability. It can be argued that in most practical settings, ORR will be implemented in conjunction with a planning system that smoothes some of the overloads and underloads in the schedule. Such smoothing may simplify the ORR task and enhance the effectiveness of ORR. This study will examine the impact of such load smoothing on the operation of the ORR system and on the performance of a simulated job shop.
Specifically, this paper has the following objectives:
* to examine the impact of load smoothing on the operation of ORR;
* to determine if load smoothing should be applied only to overloads, underloads, or both;
* to examine the interrelationship among load smoothing, the ORR mechanism, and
* to identify potential directions for future research.
THE ROLE OF ORR IN SHOP FLOOR CONTROL
ORR is one component of a complete shop floor control system. In addition to ORR, such a system, as described by Melnyk, Carter, Dilts, and Lyth (1985), consists of detailed scheduling, data collection/monitoring, control/feedback, and order disposition.
Order review/release precedes these other activities. It determines what orders are released to the floor, when they are to be released, and the conditions of release. ORR is both a filtering mechanism and a capacity management tool. It tries to release to the floor only that work which needs to be released and which has a good chance of being completed on time and within costs. It also tries to ensure that there is a balance between the load released to the floor and the capacity available for processing that load.
Conceptually, ORR mechanisms should have a positive impact on the operation of the shop floor. By controlling the flow of work and ensuring that the shop is not overloaded, ORR can help generate stable queues, which are consistent with stable lead times. Furthermore, by reducing the number of jobs on the floor, dispatching should be simplified. There are fewer jobs to prioritize. Finally, since only good jobs are released, the shop is always working on jobs which can be completed in a timely and cost effective manner.
Researchers have found limited support for the use of ORR mechanisms. In general, simulation based studies have found that, while queues are stabilized by the use of ORR and the time spent by jobs on the floor is reduced, overall lead times are not reduced (see, for example, Ragatz (1985) and Melnyk and Ragatz (1989)). The time saved by spending less time on the floor is offset by the increased time spent by jobs waiting in the pool. As a result, if the goal is to reduce overall lead times, mean tardiness and proportion tardy, the most effective tactic is not to filter the orders through a ORR mechanism but to release them immediately to the floor.
One reason for this apparent contradiction may lie in the nature of the models that have been constructed to study the effect of ORR mechanisms on shop performance. In practice, ORR is a linkage between the planning system and the shop floor. The models used to study ORR have generally looked at the shop floor in detail, but treated the planning system as a stochastic process which is outside the control of the system being studied, and produces work schedules that may be highly variable.
THE PLANNING SYSTEM AND ORR
In most manufacturing settings, the planning system embodies all of the activities that are required to generate a feasible and valid production schedule. Included are such activities as master production scheduling, material and priority planning, and capacity planning. It is the planning system which generates the schedule of orders, assigns routings, processing times, order quantities and order due dates. It is also the planning system (and specifically the capacity planning module) that determines the variation in the total work load in the plans released to the shop floor control system.
As noted by Vollmann, Berry and Whybark (1988), an effective manufacturing planning system must develop schedules which avoid subjecting the floor to conditions of either excess capacity or excess work. One method of avoiding these situations is to "smooth" the schedules. That is, the work load-in each period of the schedule is reviewed, and any "peaks" or "valleys" in the load are identified and eliminated by either pulling work forward or pushing it back.
The importance of such smoothed schedules can be better appreciated if the entire manufacturing planning and control system is envisioned as a series of filters. These filters are arranged from coarsest to finest. The coarsest is found at the production planning stage. Next comes the master production schedule, capacity planning and, finally, ORR. This last stage is the finest of the filters.
These filters are interdependent in that each subsequent system depends on the previous stage to identify and eliminate major capacity problems. In turn, each stage fine tunes the schedule passed from upper level filters by evaluating its capacity implications in greater detail. This is an ongoing process of filtering which is regularly repeated (e.g., week by week).
A potential problem is created when this progressive process of filtering the schedules breaks down. If an unfiltered schedule is passed to a lower level, it could present that system with more variation than it is designed to manage, and the performance of the shop floor deteriorates. This situation may account for the weak performance of ORR mechanisms in prior studies.
This study examines the way in which smoothing by the planning system affects the operation of the ORR system and the subsequent performance of the shop. In so doing, the study addresses the issue of whether an ORR system can be examined independently of the planning system that feeds it or whether the ORR system and the planning system must be treated as one.
The basis for this study is a computer simulation model of a simple random job shop. The simulation model was programmed in SIMAN (Pegden (1987)). All statistical analysis was carried out using SYSTAT Version 4. 1. In all cases, [alpha] = 0.05 was used in evaluating statistical significance.
The discussion of the research methodology is presented in six sections. In the first section, we lay out the major features of the simulation model. In subsequent sections, we discuss the simulated planning system smoothing procedure, the shop performance measures, the experimentation strategy, the experimental factors, and the experimental design.
The Simulation Model
The model used in this study follows the form and structure of the simulated job shop described in Melnyk and Ragatz (1989). That is, the shop consists of six work centers operating 40 hours per week. Each work center contains a single machine and can process only one job at a time. No preemption is permitted. Job routings are random, with no return visits to a work center.
The number of operations per job is uniformly distributed between two and six, and operation times are drawn from a normal distribution with a mean of 1.5 hours and a standard deviation of 0.1 hours. Jobs arrive according to a Poisson distribution with a mean of 35 jobs per week, yielding an overall shop utilization level of approximately 87%.
Each job is assigned a due date on arrival. This due date is equal to the arrival time plus a multiple of the total work content, or:
Due [Date.sub.j] = Arrival [Time.sub.j] + K*[TWK.sub.j]
where K is a multiplier and [TWK.sub.j] is the total operation time for order j. A K value of 7 was used, which creates about 20% tardy jobs in the base case of immediate job release and first come-first served dispatching.
Due dates are rounded up or down so that they coincide with the beginning of the closest week. This practice is similar to that found in production systems where orders are released by the planning system to the shop floor on a regular weekly basis, such as in MRP systems (Orlicky (1975)).
After assignment of the due date, each job is placed in the order release pool. The movement of orders from the pool to the shop floor is managed by the order release mechanism alone.
The Planning System Smoothing Procedure
In this study, work load smoothing is treated in a simplified fashion. A schedule of work is created once every 40 hours. This schedule is generated from a Poisson distribution which determines the number of jobs to be released. Before the schedule of jobs is placed in the order release pool, it is evaluated to determine if it exceeds the ceiling or if it is less than the floor. The ceiling and floor represent the maximum and minimum amounts of work that may be sent to the shop each week. The closer together the ceiling and floor are set, the smoother the input of work to the shop. The values set for the ceiling and the floor are experimental values and are stated in terms of hours.
The smoothing begins by testing against the ceiling. The schedule generated from the Poisson distribution initially sets the number of jobs to be created for the next 40 hours. The individual jobs are then defined. That is, each job's characteristics (i.e., number of operations, routing, and operation times) are established, and the job is added to the total work load for the period. If the total load to be released for the period is less than the planning ceiling, that job is placed in the backlog pool. When the load created by an additional job exceeds the planning ceiling, the excess jobs, including the one that caused the total work load to exceed the ceiling are pushed off to the next week.
If the load created by all of the planned jobs does not at least meet the planning floor, additional jobs are created (i.e., pulled forward from the next planning period). The schedule for the next period is subsequently adjusted to reflect the pulling forward of these jobs so that in the long run, the average workload sent to the system is not affected by smoothing.
The smoothing mechanism used in this study differs from that developed by Irastorza and Deane (1974). Their procedure was based on a mixed integer programming formulation. In contrast, the mechanism used in this study is very simple. It does not try to manage the planned load in detail. Rather, it manages the planned load for extremes (i.e., peaks and valleys). This approach to smoothing is similar to Capacity Planning Using Overall Factors (CPOF) as described by Schmitt, Berry and Vollmann (1985). This procedure has been found to be as effective under many conditions as capacity bills and resource profiles.
To identify the upper and lower limits, a Monte Carlo simulation of the schedule generation process was run. A total of 5000 weeks worth of scheduled work load was generated and a cumulative distribution of the resulting weekly loads generated. Based on this distribution, the following values were identified:
Fractile of Workload Distribution Value (Hours) 10% 162 20% 177 80% 240 90% 260
Shop Performance Measures
Six measures of shop performance are collected in the simulation. These measures fall into three categories: measures of delivery performance (mean tardiness and proportion tardy), measures of shop load (mean and variance of work in process), and measures of lead time (mean and variance of flow time through the system).
The Experimentation Strategy
The study involves a two stage strategy for investigating the impact of planning system smoothing on the system. In the first stage, the study focuses on different types of smoothing. Smoothing might be applied to the peaks (those loads which exceed a preset ceiling) alone, the valleys (the loads which are less than a preset floor) alone, or both. The questions in this stage of the study are whether work load smoothing by the planning system improves shop performance, and if so, whether the benefits can be achieved by focusing on smoothing peaks or valleys alone. In the second stage, the study focuses on the degree of smoothing (i.e., the fractiles of the work load distribution chosen for the ceiling and floor).
The Experimental Factors
The primary focus of this study is to examine the effect of smoothing of scheduled work load on the performance of the shop, and to analyze possible interactions between planning system smoothing, ORR, and shop floor dispatching. As a result, three experimental factors are examined via separate analyses of variance on the six system performance measures:
* work load smoothing;
* order release mechanisms; and,
* dispatching rules used on the shop floor.
Work Load Smoothing. The first experimental factor deals with the way in which the planning system smoothes shop work load. As mentioned above, in the first stage of the study, the concern is with the type of smoothing. In this stage of the study, the floor and ceiling values are set at the 10th and 90th percentiles of the unsmoothed weekly workload distribution. These limits are combined to create four levels of the smoothing factor (SMOOTH): none, floor, ceiling, and both.
In the second stage of the study, the concern is with the degree of smoothing - that is, the specific percentile values chosen for the ceiling and floor. This factor, DSMOOTH has three levels: 0,100 (no smoothing); 10,90 (moderate smoothing); and 20,80 (extensive smoothing). Based on the results of the first stage of the study, only symmetrical smoothing is considered in the second stage.
Order Release Mechanisms. The examination of order release focuses on the method used to manage the release of jobs from the order release pool. Two levels are considered for this factor (ORR):
* Immediate release (IMM): Under this approach, all jobs in the pool are released at the
beginning of the week.
* Maximum load limit (MAX): This approach releases work to the shop floor until the current
work load on the floor reaches a maximum load limit. In this study, the limit was set at 144
hours of work, or 300% of daily shop capacity.
Dispatching Rules. The dispatching rules are used in scheduling operations at the various work centers on the shop floor. Three levels of this factor (DISPATCH) are considered:
* First-come-first-served (FCFS),
* Shortest processing time (SPT); and,
* Minimum slack (MinSlk).
For the first-stage experiments, a full factorial design is used with 24 cells and 10 observations per cell. The second stage requires a full factorial design of 18 cells. To generate the observations for one cell, the simulation model is run until 11 batches of 1600 job completions are collected. The first batch is discarded to eliminate any starting condition effects. To reduce variance, common random number streams are used. However, to reduce any problems due to non-independence between cells, Mihram's procedure is followed (Mihram (1974)). That is, common streams are used in all instances except when generating operation times. The random number for this activity is synchronized across cells. Furthermore, the batch numbers (NBATCH) are used as a blocking factor during the subsequent ANOVA analysis.
DISCUSSION OF RESULTS
The results of the simulation runs are summarized in Tables 1-4, with Tables 1 and 2 containing data pertaining to the first-stage experiment and Tables 3 and 4 applying to the second stage.
Stage One Results
Tables 1 and 2 offer evidence of the impact different types of work load smoothing exert on shop performance. The most important observation from the stage one results is that combined floor and ceiling smoothing ("both") yields the best, or ties for the best, shop performance for both of the tardiness-related measures and both of the flow time related measures, regardless of dispatch-ORR combination. Also, with only two exceptions, combined floor and ceiling smoothing yields the best, or ties for the best, performance on the two work in process related measures. The two exceptions are for mean work in process under MinSlk-MAX and variance of work in process under FCFS-MAX. In these two cases, the difference between the result for "both" smoothing and the best type of smoothing is so small as to have no practical significance. In the presence of FCFS or SPT dispatching, ceiling smoothing generally provides second-best results, and in the presence of MinSlk dispatching, floor smoothing provides results very close to "both."
[TABULAR DATA OMITTED]
The results of the stage one experiment clearly indicate that work load smoothing by the planning system can improve system performance. The results regarding the effects of the different types of smoothing led to the decision to eliminate floor only and ceiling only smoothing from consideration in the stage two experiment. The asymmetrical types of smoothing appeared to offer no advantage over smoothing both peaks and valleys in work load.
The stage one experiment also suggests that planning system smoothing and ORR make their impact felt on different aspects of shop performance. Planning system smoothing seems to influence primarily the flow time measures, whereas ORR affects the work in process measures more strongly.
Stage Two Results
Tables 3 and 4 summarize the results of the stage two experiment. The results in Table 3 indicate that as the degree of smoothing increases, there generally is monotone improvement in shop performance. Moreover, the magnitude of the improvement appears to be greater in moving from (10,90) to (20,80) smoothing than from (0,100) to (10,90). This is apparently because the move from (0,100) to (10,90) reduces work load variance by about 29%, and the move from (10,90) to (20,80) reduces the work load variance by an additional 39%.
[TABULAR DATA OMITTED]
The effect of smoothing is particularly strong on the tardiness related measures for the nondue date oriented dispatching rules (FCFS and SPT) and for the flow time variance measure, regardless of dispatching rule. As in the first stage experiment, smoothing has some impact on work in process, but overall, controlled order release affects work in process to a greater extent than smoothing does.
One exception to the monotonic improvement with greater smoothing is in the variance of work in process measure in the presence of the MAX ORR mechanism. At the highest level of smoothing, the variance of work in process actually increases. This is due to the fact that at this highest level of smoothing, with controlled job release, the shop becomes completely empty with some regularity. The variance of work in process could probably be reduced by lowering the load limit used in the MAX releasing mechanism so that the peaks in work in process would be reduced.
Apart from the impact of greater degrees of smoothing, the results of the stage two experiments show more clearly than the stage one experiment that the combined effect of work load smoothing by the planning system and controlled job release by the ORR system is to dramatically reduce variability in the system. Improvement in delivery performance accompanies the reduction in system variability. It is also evident that the combination of controlled job release and planning system smoothing reduces the performance difference between MinSlk and the two non-due date oriented dispatching rules.
The results of the experiments reported here support the idea that the effectiveness of ORR can be enhanced by the presence of a planning system that does some coarse smoothing of scheduled work load. The ORR system can then act as a finer filter to further stabilize the system. The two filters are complementary tools for reducing variance in the system - that is, planning system smoothing strongly affects job flow times, while ORR strongly affects work in process levels.
The combination of work load smoothing and controlled job release also diminishes the performance advantage the due date oriented MinSlk dispatching rule holds over the simpler FCFS rule. This suggests that careful use of the filtering mechanisms in the planning and ORR systems can allow the use of a much simpler dispatching mechanism for control of jobs once they are released to the floor. And, while controlled job release still results in somewhat longer total flow time than immediate release, it is notable that controlled release with work load smoothing yields shorter flow times than immediate release with no smoothing.
The results of this study help somewhat to explain the contradiction between the value placed on ORR in practice and the experimental results reported in the literature on the effect of ORR on system performance. This study indicates that a complementary relationship exists between the two filtering mechanisms of the planning system and the ORR system. Studies of ORR which overlook this relationship may not capture the full effect of the ORR function.
Further work is warranted on work load smoothing and its interaction with ORR. Only fairly crude smoothing and ORR mechanisms were considered in this study. More sophisticated methods may provide even greater improvement in system performance. In particular, better smoothing and ORR methods may further reduce the advantage of sophisticated dispatching rules, which might enable a shop to run quite effectively with very simple control mechanisms on the shop floor itself.
Baker, K.R. "The Effects of Input Control on the Performance of a Simple Scheduling Model." Journal of Operations Management, vol. 4, 1984, 99-112. Bertrand, J.W.M. "The Use of Workload Information to Control Job Lateness in controlled and Uncontrolled Release Production Systems." Journal of Operations Management, vol. 3, 1983, 67-78. Conway, R.W., W.L. Maxwell, and L.W. Miller. Theory of Scheduling. Reading, MA: Addison-Wesley Publishing Company, 1967. Harty, J.D. "Controlling Production Capacity" American Production and Inventory Control Society 12th Annual Conference Proceedings, 1969, 60-64. Irastorza, J.C., and R.H. Deane. "A Loading and Balancing Methodology for Job Shop Control." AIIE Transactions, vol. 6, 1974, 302-307. Melnyk, S.A., P.L. Carter, D.M. Dilts, and D.M. Lyth. Shop Floor Control. Homewood, IL: Dow Jones-Irwin, 1985. Melnyk, S.A., and P.L. Carter. Shop Floor Control: Principles, Practices and Case Studies, 1987 Falls Church, VA: American Production and Inventory Control Society. Melnyk, S.A., and G.L. Ragatz. "Order Review/Release: Research Issues and Perspectives." International Journal of Production Research, vol. 27, no. 7, 1989, 1081-1096. Mihram, G. Arthur. "Blocking in Simulation Experimental Designs." Journal of Statistical Computing and Simulation, vol. 3, 1974, 29-32. Nicholson T.A.J., and R.D. Pullen. "A Practical Control System for Optimizing Production Schedules." International Journal of Production Research, vol. 16, 1972, 219-227. Orlicky, J.A. Material Requirements Planning. New York, NY: McGraw-Hill Publishing Company, 1975. Pegden, C. Dennis. Introduction to SIMAN. State College, PA: Systems Modeling Corporation, 1987. Ragatz, G.L. An Evaluation of Order Release Mechanisms for Job Shops. Unpublished Ph.d. dissertation, Indiana University, Bloomington, IN, 1985. Sandman, W.E., and J.P Hays. How to Win Productivity in Manufacturing. Yellow Book of Pennsylvania, 1980. Schmitt, T., W.L. Berry, and T.E. Vollmann. "An Analysis of Capacity Planning Procedures for a Material Requirements Planning System." Decision Sciences, vol. 15, no. 4, 1985, 522-541. Vollmann, T.E., W.L. Berry, and D. Clay Whybark. Manufacturing Planning and Control Systems. 2nd ed. Homewood, IL: Irwin, 1988. Wight, O.W. "Input/Output Control: A Real Handle on Lead Times." Production and Inventory Management, vol. 11, 1970), 9-30.
|Printer friendly Cite/link Email Feedback|
|Author:||Melnyk, Steven A.; Ragatz, Gary L.; Fredendall, Lawrence|
|Publication:||Journal of Operations Management|
|Date:||Oct 1, 1991|
|Previous Article:||A volume and material handling cost based heuristic for designing cellular manufacturing cells.|
|Next Article:||On apportioning costs to customers in centralized continuous review inventory systems.|