INT. J. PROD. Rhs.. 2003, VOL. 4 1 , NO. 9, 2021 - 2 0 3 5 _ Taylor & Francis \^y Revising the master production schedule in sequence dependent processes JAMES A. HILLt*, WILLIAM L. BERRYt and DAVID A. SCHILLINGI Revising ihe master production schedule (MPS) in a roiling planning horizon environment to improve plant performance is critical in process industries with sequence dependeni changeovers. Plant performance is defined in terms of process changeover time, total shortages, and finished goods inventory. Two heuristics for revising the MPS are introduced and tested against a rolling horizon MPS thai does not take sequence dependent changeovers into consideration. These two heuristics represent two dilTerent approaches to heuristic design: local interchange (SWAP) and global interchange (3OPT). Simulation experiments are reported that test the performance of the two different approaches lo revising the MPS in process industries with sequence dependent changeovers. The MPS designs were tesled under plant conditions that are frequently encountered in process industries. The results indicate that revising the MPS in process industries significantly improves plant performance. In addition, 30PT provides a major improvement in changeover time, while SWAP provides improvement in total shortages. 1. Introduction Changing markets are placing new business reqtiirements on many process industry plants, including the need for greater product variety wilh improved changeover time and delivery performance (Leschke 1995. Olliff and Burch 1985). Supporting these requirements is especially difticull in many process industry firms, such as firms that manufacture chemical products, because such processes frequently involve long product sequence dependent changeover times. As reported by Leschke (1995) changeover times represent a sizeable proportion of plant capacity, e.g. 47-92%. and are not easily reduced through conventional changeover time reduction methods. One means of supporting changeover time and delivery competitiveness in the high variety environment faced by such companies is through the development of improved master production scheduling. In this paper we focus on improving plant performance by revising the master production schedule (MPS) in a make-to-stock process industry environment. The MPS specifies the timing and size of production quantities for each product. The MPS links the firm's broad strategies, as expressed in the aggregate production plan, to more specific tactical plans that will enable the firm to balance the product demands of customers with the supply of products made available by plant schedules and inventory. Process industry firms generally perform master production schedulRevision received June 2002. tOwen Graduate School o\' Management. Vanderbilt University. Nashville. TN 37203. USA. J Fisher College of Btisiness. The Ohio State University. Columnbus. OH 43210, USA *To whom correspondence should be addressed, e-maii: james.hill((( owen.vanderbilt.edu liuenunioiiiil Jmirtiut oj Protluciioii Rt'.u-tmh ISSN 1)020 7543 prinllSSN 1366 5HKX online i; 2003 Taylor & hrancis Lid http:,''www.landr.Lo.uk'j(.n]rnals DOI; IO.I0K0/0020754O3IOO0I23BH6 2022 J. A. HilUt al. ing differently from job shop industries. These firms tend to schedLile capacity first belbre raw materials because of resource constraints within the chemical blending and mixing process. If planning for capacity at the chemical mixing and blending operations indicates potential infeasibilities. the master production schedule shouid be revised to accommodate these capacity limitations. Most of the previous work on master production scheduling does not consider process industry firms where changeovers are sequence dependent {Lin and Krajewski 1992. Sridharan et al. 1987. and Zhao and Lee 1993). In this paper we address the problem of revising a rolling horizon MPS in process industries with sequence dependent changeovers. This paper has two objectives: (I) to present computationally efficient methods to revise the MPS to improve plant performance in sequence dependent processes, and (2) to determine the sensitivity of these methods to frequently encountered differences in plant operating conditions. We begin by discussing previous literature relative to the master production scheduling problem. In the next section, two heuristics are presented to revise the MPS under rolling horizons. Next, we present the research questions to be examined and the research design used in the simulation experiments. Finally, computational results are discussed that demonstrate the improvements in master production scheduling performance. 2. Literature review Very little work has been reported on master production scheduling methods that consider changeover time and product sequence restrictions often encountered in the process industries. Product sequence restrictions and changeover costs were incorporated in a process industry model reported by Oliff and Burch (1985). They present a mixed integer programming model that determines lot sizes, line assignments and inventory levels for 28 MPS products. However, because of the problem size and computation requirements, a linear programming approximation that generates near-optimal solutions is limited to small problem sizes. De Matta and Guignard (1995) study lot-sizing and changeover decisions in production schedules that are implemented on a rolling-horizon basis. Their results show that as the planning horizon lengthens, the changeover and holding costs converge. These authors do not include re-planning intervals within the planning horizon. Hill et al. (2000) introduce a two-level approach to master production scheduling. The authors show that master production scheduling should be done at the bottleneck process (level two) not the end item (level one) in order to improve plant performance. Some research has been done on re-planning the MPS under a rolling horizon. Lin and Krajewski (1992) present a mathematical model for designing a MPS system in an uncertain environment. The model can be used to estimate the expected cost per period for any combination of the re-planning interval, the frozen interval and the forecast window. Venkataraman (1996) addresses the problem of re-planning frequency for a rolling horizon MPS in a process industry environment. The author uses actual data from a paint company to determine the appropriate re-planning frequency. Kern and Wei (1996) evaluate the relative effectiveness of rescheduling policies in capacity-constrained, just-in-time production environments. The authors found that higher demand variation, large forecast errors, and tight capacity led to the degradation of system performance. Yang and Jacobs (1999) examine the use of both frozen and re-planning intervals for planning the MPS for a capacityconstrained job shop. The results show that forecast error, demand lumpiness, set-up time, and planned lead time have a greater impact on total inventory and mean total Revising the master production schedule 2023 backlog than the frozen and re-pkinning intervals. None of these aulhors have focused on master production scheduling in environments where changeovers are sequence dependent. In this paper, we examine master production scheduling performance where heuristics are used to revise the MPS within the planning horizon. We also determine operating conditions that effect MPS performance in a process industry environment where changeover times are sequence dependent. In the next section, we demonstrate two approaches to revising the MPS in a process industry environment where changeovers are sequence dependent. Two scheduling heuristics. SWAP and 30PT can be used to revise the MPS for maketo-stock sequence dependent products. Since the planning horizon may be sufficiently long to include multiple orders for an individual product, these heuristics can be used to sequence a series of orders that includes both different products and repeat orders for the same product. These computationally efficient heuristics, which represent different approaches to revising the MPS. can be applied to large problems. 3. Scheduling heuristics The SWAP heuristic is a simple local search heuristic that has been tested extensively in the scheduling literature (Wilkerson and Irwin 1971. Hill ct al. 2000). It is an improvement heuristic, which uses a neighbourhood search approach similar to that used by Gupta and Darrow (1986). The heuristic begins with an initial MPS set of orders and reduces the changeover time in the MPS through swaps of adjacent orders until no further improvement is possible. The local switching of orders is thought to be advantageous in master production scheduling because orders are not moved far from their original due date generated by the initial MPS. thus allowing for improvement in both process changeover time and delivery performance, 30PT is an adaptation of Lin and Kernighan's (1973) Three-Opt routine, which seeks to improve a given MPS by examining interchanges of three orders at a time. As an illustration, consider a sequence often orders in a master production schedule (numbered from I to 10) 1.2,3,4,5,6,7,8.9.10. A possible three-way interchange would be to take the orders and changeovers represented by 1.2; 5,6; and 8, 9, and create a new sequence where I is followed by 6; 8 is followed by 2; and 5 is followed by 9. All other changeovers in the original sequence remain the same. Specifically, the resulting sequence would be: 1,6,7,8,2,3,4,5,9,10. If the second sequence results in a smaller total changeover time, then it is retained in place of the first. There are a number of different ways to perform a three-way interchange. We have chosen for simplicity to implement solely the swapping approach used in the illustration above, which maintains the sequence of all orders that were not swapped. This method provides a counterpoint to the SWAP heuristic. 4. Research design In this section, we present the research design for evaluating master production scheduling performance under process industry environments. The research questions are provided first. A discussion of the environmental factors that impact 2024 J. A. performance in process industries is then provided. Next, the experimental design and performance criteria are described. Finally, the simulation model is outlined. 4.1. Research questions Previous research on replanning the master production schedule has not addressed processes with sequence dependent changeovers. Nor has this research addressed processes with multiple products. Our research attempts to address some important issues relative to manufacturing plants that have multiple products with sequence dependent changeovers. Three questions are examined in this research to improve master production scheduling performance in process industries with sequence dependent changeovers. These are: (1) To what extent can performance he improved hy revising the rolling horizon MPS with the SWAP and 3OPT scheduling heuristics in sequence dependent processes? (2) How is performance of the revised master production schedule affected by changes in the replanning frequency? (3) Is the relative performance of the revised master production schedule dependent on the interaction of the independent variables? 4.2. Experimental factors 4.2.1. Coefficient of variation of changeover times The proportion of plant capacity explained by changeover time can be very large. When such a large portion of the process capacity is devoted to changeovers. a relatively small percentage reduction in changeover time can represent an important improvement in operating performance. Given the sensitivity of many companies in process industries to improving margins in a high product variety environment, the design of the master production schedule to incorporate scheduling heuristics that reduce changeover time can be very important in improving plant performance. Furthermore, while previous research indicates that scheduling performance is quite sensitive to changes in the coetticient of variation in changeover times, relatively small values have been used for this factor in evaluating sequence dependent scheduling heuristics. Previous work by Gavett (1965) and Guinet (1993) report results covering a range of 0.17-0.57 for the magnitude of changeover time coefficient of variation. While very little has been published concerning the actual magnitude of the changeover time coefficient of variation in plants, our field research indicates values as large as 0.97. or nearly twice the maximum value studied in previous research. Understanding plant performance at much higher levels for the coefficient of variation in changeover times is important in understanding the implications of this factor in master production scheduling. This experimental factor brings a new dimension to master production scheduling research because, in previous research, changeovers are considered to be either negligible or sequence independent. To understand how the revised MPS performs under different operating conditions, we varied the changeover matrix to produce two levels of coefficiem of variation (high = base level of 0.97; low = 0.32 one-third of the hase). 4.2.2. TBO {time between orders) When scheduling a set of orders, several heuristics have been proposed in previous research to minimize changeover time in sequence dependent scheduling (Clark Revising the master production seliedide 2025 and Ciark 2000. Haase and Kinims 2000). These authors provide efficient algorithms 10 reduce the amount of CPU lime. However, the rules are tested on relatively small problem sizes and not under operating conditions outlined in this paper. With the exception of Guinet (1993). much of the previous research on sequence dependent single machine processes has been conducted using relatively small problem sizes, involving 20 orders or fewer. In master production scheduling the time between orders (TBO) is commonly used to generate diflerent problem sizes (Sridharan et at. 1987, Sridharan and LaForge 1994). The TBO is the ratio between the economic order quantity and the average period demand. If the number of product items is fixed, a plant having make-to-slock products with predominately small TBO values reflects the frequent production of small lot sizes and relatively large MPS problem sizes. Likewise, a plant with products having predominately large TBO values reflects a situation where products are produced infrequently in large lot sizes, yielding relatively small MPS problem sizes. In our research, we have chosen levels for TBO (high = 20; low = 4) which generate order sizes ranging from 20 to 100. 4.2.3. MFS method The experimental factor master production scheduling method was set at three levels. The first setting represents the results of the initial MPS without using replanning heuristics. The MPS is still updated and rolled forward, although without the use of scheduling heuristics. From this point further we will refer to the MPS without scheduling heuristics as the unrevised MPS. The second and third settings represent results of the revised MPS using SWAP and 30PT respectively. 4.2.4. Demand uncertainty Past research (Sridharan and Berry 1990. Lin and Krajewski 1992, Zhao and Lee 1993. Kern and Wei 1996. Yang and Jacobs 1999) has shown that demand uncertainty is an important factor affecting master production scheduling performance. Demand uncertainty was modelled by generating actual demand for multiple products. The forecast error was modelled using a normal distribution with a mean of zero and a standard deviation with two experimental settings: I5yo and 30% of mean demand. The forecast requirements used in determining the MPS were computed by adding the forecast error to the actual demand each period over the simulation run. A critical assumption is thai the forecast is unbiased. Previous research has shown that under conditions of demand uncertainty there is no clear choice of lot-sizing rules (Wemmerlov and Whybark 1984). This .study uses the Fixed Periodic Requirement (FPR) rule. Wemmerlov (1979) reported that most companies prefer a FPR lol-sizing rule. The use of the FPR is also supported by Yang and Jacobs (I999J. 4.2.5. Re-planning frequency The main objective of this study is to examine the affect of revising the MPS in a process industry environment. Sridharan and Berry (1990) have shown that the length of the planning horizon has the least influence on system cost and customer service. Therefore, it is not our intention to focus on the length of the planning horizon. We consequently set the re-planning frequency (R) to be a fraction of the fixed planning horizon (A'). The values of R for our experiments are 0.25, 0.50. 0.75 2026 / A. ///7/et al. and 1.0. When R is equal to 0.25, revising is done every 0.25A' periods. These are the same levels used by Zhao and Lee (1993). 4.3. Performance criteria Customer pressure for improved delivery performance makes total shortages an important measure. Delivery performance is defined as the degree to which manufacturing cannot meet the MPS due dates, i.e. order tardiness. Order tardiness is measured in terms of the number of scheduling periods each MPS order is delivered late. This measure is converted into total shortages by multiplying order tardiness by the average sales forecast per scheduling period. Total inventory is similar to that for total shortages, and reflects performance frequently reported in process industry firms. Changeover time is an important measurement because the MPS is often developed in an uncapacitated environment considering other criteria, such as customer service inventory investment and instability, leaving changeover time to be considered in shop floor scheduling. Also, additional capacity can be used as a buffer against demand variability. As a result, this measure reflects the potential improvement in capacity obtained by revising the MPS. Based on our above discussion, we use three performance criteria to evaluate the MPS performance; (I) the total amount of shortages in the MPS planning hori/.on. (2) the total amount of finished goods inventory in the MPS planning horizon, and (3) ihe total amount of changeover time in the planning horizon. 4.4. Experimental design Simulation experiments were conducted to examine the research questions presented earlier. A simulation model was developed using FORTRAN and run on a Dell OptiPlex GXI desktop system. The SPSS statistical package was used to analyse the data. A full factorial Analysis of Variance design with 40 replications was used to examine the research questions. The replications in these experiments were 40 different product sets. Each product set includes ten products, each having the following data; average period sales forecast for the next 30 periods (in units), the MPS batch size (in units), the MPS initial inventory and the run time processing rate (in units/hour). The changeover time matrix data represent actual operating data collected from a chemical processing firm. For all 40 replications, the following procedure was used to randomly generate Ihe product structure data for each of the ten products. The parameter values listed in this paragraph are representative of actual operating data. First, an average period demand was randomly generated from a uniform distribution with a mean of 12000, a lower limit of 4000 and an upper limit of 20000. Second, the experimental factor setting for the time between orders (TBO) value was used as the mean of a TBO distribution. A TBO value for each product was randomly generated from a uniform distribution having upper and lower limits of ±50% of the mean value setting. Third, the formula product batch size was computed to be the average period demand multiplied by the lime between orders value for each product. Fourth, the changeover to proce.ssing time ratio value for each product was randomly generated from a uniform distribution having upper and lower limits set at ±50% of ihe mean value of 0.25. This setting implies that the average proportion of process capacity explained by the changeover time is 25%. After the ratio value was set, the run time Revising the master production schedule 2027 processing rate (in units/hour) was calculated for each product using the fixed value for the ratio and the average changeover lime from the changeover time matrix. Finally, the changeover time matrix having the appropriate level for the changeover time coefficient of variation was used in the MPS. Next, an MPS record v^as established for each product containing a forecast, projected inventory, and MPS rows. These records were processed to develop a 30 period MPS. At the end of each replanning cycle, the MPS is revised using the scheduling heuristic and rolled forward through time. This procedure was repeated until all 30 periods of the requirements were processed. To eliminate the effect of transient conditions upon the operating performance, the first five periods of operating performance data in each simulation run are discarded. The five-period initialization interval was selected after visual inspection of the inventory level data. In these experiments, the initial inventory conditions were identical for all product sets. Therefore, the differences in performance between the replications resulted from differences in the randomly generated average period forecast, the product ratio value, TBO values between product structures, and differences in the changeover time coefficient of variation value. 5. P.xperimental results and discussion It is clear from the results in tables 1. 2 and 3 that the MPS scheduling method and the replanning frequency have a large influence on performance in comparison with the other parameters studied. In fact, their F-statistic ranks near the top in all three performance measurements. Therefore the next two sections are focused around these two parameters. The results showing the effect of experimental factors F- Source MPS CV TBO Demand Replanning freq CV X TBO CV X MPS TBO X MPS Repl. X MPS Repl, X CV Repl. X TBO 60.2 50.8 108.9 41.3 57.5 31.7 22.9 28.9 20.8 48.1 27.0 Sig. 0.00* 0.00* 0.00* 0.00* 0.00* 0.00* 0.00* 0.00* 0.00* 0.00* 0.00* * Signilicant al the 0.05 level Table I. (a) ANOVA results for total shortages. Unrevised MPS MPS-SWAP MPS-3OPT Unrevised MPS MPS-SWAP MPS-30PT 1.0 0.00* 0.00* 1.0 0.00* 1.0 iU al ihf l).()^ level Table 1. (b) Tiikey multiple comparisons. 2028 / . A. /////et al. Source F Sig. MPS CV TBO 187.3 6.3 33.0 46.1 94.2 0.084 0.00* 0.02* 0.00* 0.00* 0.00* 0.77 0.01* 0.00* 0.07 Demand Replaniiing frcq CV X TBO CV X MPS TBO X MPS Repl. X MPS Repl. X CV Repl. X TBO 7.1 30.7 3.2 10.4 37.2 n.o2* 0.00* * Sij![iLlic:int al the 0,05 level Table 2. (a) ANOVA results for total inventory. Unrevised MPS MPS-SWAP MPS-3OPT Unrevised MPS MPS-SWAP MPS-3OPT 1.0 0.931 0.00* 1.0 0.00* 1.0 * Significant al [he 0.05 level Table 2. (bl Tukev mul Source F Sig. MPS CV TBO Demand Replanning freq CV X TBO CV X MPS TBO X MPS Repi. X MPS Repl X CV Repl X TBO 49.3 0.594 3787.5 0.827 1478 0.295 10.4 23.7 18.5 1.8 34.2 0.00* 0.44 0.00* 0.29 0.00* 0.58 0.00* 0.00* 0.00* 0.13 0.00* * Sijiniticant at the 0.05 level Table 3. (a) ANOVA results for total changeover time. Unrevised MPS MPS-SWAP MPS-30PT Unrevised MPS MPS-SWAP MPS-3OPT 1.0 0.00* 0.00* I.O 0.04* 1.0 * Significanl at ihc 0,1)5 level Table 3. (b) Tukey multiple comparisons. Revising the master production schedule 2029 on manufacturing performance are presented in three parts. Part I addresses research question 1, ihc statistical dilTerence in performance between ihe MPS without using scheduling heuristics and the revised MPS using SWAP and 3OPT respectively. Part II addresses research question 2. changes in the re-planning frequency and the affect on performance. Part III addresses research question 3. two-way interactions between the experimental factors. 5. i. Master production scheduling methods The results in table 4 show the performance of the MPS without scheduling heuristics and the two revised MPS solutions with scheduling heuristics for all criteria (total shortages, total inventory and total changeover time). These results are averaged across all operating environments. The ANOVA results shown in table I indicate the main effect MPS method is significant for total shortages. Further, the Tukey results in table l(a) show there is a significant difference between the unrevised MPS, revising the MPS with SWAP and revising the MPS with 30PT. The results for total shortages shown in table 4 indicate the local SWAP method has lower total shortages averaged across all operating environments than both the global 30PT method and the revised MPS without scheduling heuristics. Thus, we can conclude that revising the MPS with the local SWAP heuristic provides the lowest total shortages when operating in an environment with sequence dependent changeovers. The ANOVA results shown in table 2 indicate the main effect of the MPS method is significant for total inventory. The Tukey results in table 2(b) show there is a significant difference between 30PT with both SWAP and the unrevised MPS. However, there is no significant difference between SWAP and ihe unrevised MPS TBO CV Shortage Inventory Changeover CVL 362 8 2151 CVH 987 5 2389 CVL 107 6 420 CVH 292 4 473 CVL 296 316 1675 CVH 573 423 1848 CV[ 156 115 334 CVH 200 204 375 CV, 165 13 1805 CVH 253 13 1972 CVL 80 11 364 CVH 149 TBOL Unrevised MPS TBOH TBO MPS 30PT TBOL MPS SWAP TBOH Table 4. Average performance at eiich setting. 39? 2030 J. A. /////et al. MPS. Because 30PT is a global heuristic, orders are potentially moved forward in the planning horizon causing inventory to build. This result shows that the design of the scheduling heuristic used to revise the MPS can lead to degradation in MPS performance when operating under sequence-dependent changeovers. The result provides partial support to previous research (Yano and Carlson 1987, Sridharan and Berry 1990) that frequent revising of the MPS can be undesirable under certain operating environments. However, it is important to note that their measure of performance was cost while ours is total inventory, and our research is multi products with sequence dependent changeovers. The ANOVA results shown in table 3 indicate the main effect MPS method is significant for changeover time. The Tukey results in table 3(b) show there is a significant difference between the unrevised MPS versus using SWAP or 3OPT to revise the MPS- There is also a significant difference between 30PT and SWAP. Thus., we can conclude that when changeover time is the chosen performance measurement it is beneficial to revise the MFS to improve performance. We can also conclude that 30PT outperforms SWAP in reducing changeover time across the planning horizon. 3OPT"s performance advantage is due lo its ability to hatch orders together across the planning horizon versus the local SWAP heuristic. Now that we have concluded that revising the MPS will improve plant performance we turn our attention to how often the MPS should be revised. 5.2. Changes in the re-planning frequency The Tukey procedure was used to test the differences in mean performance between different MPS replanning frequencies for both heuristics and for all three performance measurements. It is clear from the Tukey results in tables 5 and 6 that there is a significant difference in performance between the different levels of replanning frequency. The effect of the replanning frequency on performance is dependent on the method chosen to revise the MPS and the performance measurement chosen. The results show that there is a significant difference in changeover time between all levels of replanning frequency for both 30PT and SWAP. Frequent revision of the MPS is beneficial when reducing changeover time is the chosen performance measurement. This result is different from previous research where frequent revision of the MPS causes degradation in cost and MPS stability (Sridharan and Berry 1990, Zhao and Lee 1993. and Yang and Jacobs 1999). Frequent replanning of the MPS leads to more orders open for rescheduling. With more orders open for rescheduling, more changeover time can be reduced. The results also indicate that there is a significant difference in total shortages when the replanning frequency is greater than 0.50 of the planning horizon tor the SWAP heuristic. As replanning becomes more frequent, more orders in the planning horizon are open for rescheduling. This leads to an increased probability that more orders will be moved farther away from their original due date causing an increase in shortages. In contrast, the 3<3PT heuristic only improves shortage performance when the replanning frequency is set at 0.75. In this case, only 25% of the planning horizon is open for rescheduling. Frequent revision of the MPS using the 3OPT heuristic can cause degradation in total shortages. This result occurs because 30PT is a more global heuristic ihan SWAP and, as more of the planning horizon is open for rescheduling. 30PT moves orders further away from iheir original due date thus increasing the likelihood of increased shortages. As far as inventory is concerned, there is a significant difference in performance between all levels of the replanning frequency for 3OPT. More Revising the master production schedule 2031 SWAP Changeover Time Replanning frequency 0. 25 0.50 0.25 0.50 0.75 1.0 LO 0.03* 0.00* 0.00* LO 0.02* 0.00* 0.75 LO 0.00* Short ages LO 0.25 1.0 1.0 0.40 0.01* 0.00* 0.50 1.0 0.08 0.02* 0.75 LO 0.03* Inventory LO 0.25 0.5 1.0 0.34 1.0 0.06 0.02* 0.70 0.16 0.75 LO 1.0 0.99 1.0 * Sigiiificanl at p •= 0.1)5 Table 5. Tukey results for replanning frequency. 30PT Replanning frequency 0.25 0.50 0.75 LO Changeover Time 0.25 0.50 0.75 1.0 0.00* 1.0 0.00* 0.00* LO 0.00* 0.00* 0.00* Shortages LO 0.25 0.50 0.75 1.0 0.12 LO 0.04* 0.13 1.0 LO 0.00* 0.00* 0.04* Inventory 1.0 0.25 0.5 0.75 LO 0.00* LO 0.00* 0.00 1.0 1.0 0.00* 0.00* 0.00* 1.0 LO * Signilkanl al p = 0,05 Table 6. Tukey resuit.s for replanning frequency. changeover time is taken out of the schedule as the frequency of replanning increases. As more changeover time is taken out of the schedule, inventory starts to build. For the SWAP heuristic there is no significant difference in inventory until frequent replanning occurs at 0.25 of the planning horizon. The Tukey results for the MPS where scheduling heuristics arc not used were nol conclusive. In fact, the results of a full factorial ANOVA indicated that the re-planning frequency is not as significant a factor in comparison with other environmental factors. This is consistent with previous research (Sridharan and Berry 1990, Zhao and Lee 1993, and Yang and Jacobs 1999). 5.3. Two way interactions Many of the two-way interaction effects in tables 1, 2 and 3 are significant. The results indicate the MPS x CV interaction is significant for all three performance measurements. This result indicates Ihe effect of CV on total shortages is not the same for the unrevised MPS and the two revised master production schedules. The interaction is shown in figure I. Total shortages degrades in the unrevised MPS and in ihe revised MPS using 3OPT as CV increases versus the revised MPS using SWAP, where total shortages is relatively constant across both levels of CV. The MPS X CV interaction can also be seen under total changeover time in figure 2. An increase in CV causes more changeover time in the unrevised MPS versus the revised MPS solutions. The results in tables 1, 2 and 3 also indicate that the MPS x TBO interaction is significant for all performance measurements. J. A. Hlllet al. 2032 700 600 g 500 • mps/3opt Io 400 mps/swap « 300 n •un-revised mps o 200 100 0 low high CV Figure L MPS method versus 1600 J a, 1400 -- I 1200 I 1000 - • mps/3opt 800 mps/swap 600 - •un-revised mps 400 200 0 low high ev Figure 2. MFS method versus cv. Many of the two-way interactions involving the replanning frequency with other operating factors are significant in tables I. 2 and 3. A summary of the replanning frequency x CV interaction on shortages, inventory and changeover time is shown in table 7 for the 30PT heuristic. The data indicate that the replanning frequency produces a greater eft'ect on inventory and shortages when the coefficient of variation is high and the 3OPT heuristic is used to revise the MPS. A similar eftect is also 3OPT Changeover Time Reduction Total Shortages Total Inventory Replanning frequency CV (low) CV (high) CV (low) CV (high) CV (low) CV (high) 0.25 0.50 0.75 LO 1649 1043 716 60S 1907 1091 802 646 2% 241 183 214 522 401 277 346 328 250 183 101 421 Table 7. Results of replanning frequency interaction with CV (in units). 379 300 LS4 Revisini; the master production schedule 2033 observed when TBO is low. Therefore, the conditions under which the decision is made as to when replanning the MPS is beneficial can have an important influence on performance. 6. Summary These experiments show how master production scheduling performance in process industries with sequence dependent changeovers can be significantly improved through revising the MPS. These experiments show manufacturing performance can be improved by adjusting the timing of production orders in sequence dependent processes through the use of scheduling heuristics such as 3OPT and SWAP. This is especially important in plants having a high coefficient of variation of changeover times. 6.1. Improvement against MPS without revision Using the results in table 4 it is apparent that the two heuristics used to revise the MPS, 3OPT and SWAP are superior to the unrevised MPS when considering changeover time and shortage performance. Total changeover time was reduced by 15% over all experimental settings using SWAP, and by 21% using 3OPT to revise the MPS. In the case of total shortages, SWAP has a 50.7% reduction against the unrevised MPS versus 11.5% reduction for 30PT. It is apparent that revising the MPS can be critical in improving both changeover time and total shortages in sequetice dependent processes. From these results, we can conclude that iho local SWAP heuristic provides lower shortages while 30PT provides lower changeover time. The design of the local SWAP heuristic is more advantageous in reducing shortages because the MPS is already in the earliest due date sequence and the local SWAP heuristic improves performance by adjusting orders to decrease changeover time without moving orders too far away from their original due date. It is also important for managers to consider the trade-off of changeover time and inventory when revising the MPS. This is especially true with 30PT. The more often the MPS is revised the larger the trade-off between inventory and changeover time reduction. Managers should pay close attention to the frequency of the replanning cycles, as this will have an etTect on the size of the trade-off. 6.2. Scheduling environment impact The experimental results indicate the nature of the scheduling environment (measured in terms of TBO, changeover time coefficient of variation, replanning frequency and the choice of scheduling heuristics) can affect plant performatice. Replanning frequency and TBO were significant under all criteria studied. These results indicate plants having a large number of MPS orders can expect improvement in performance measured in terms of total changeover time, and total shortages by revising the MPS. It is also true that frequent revision of the MPS is beneficial for changeover time reduction. However, it is also critical to revise only a portion of the MPS when TBO is low because frequent revision will cause too many open orders to be moved away from their original due date, which will cause degradation in shortage performance. It is clear that CV plays an important role in plant performance. An increase in CV does not alTect changeover time as it does shortages. An increase in CV leads to more shortages in the planning horizon, although total changeover time remains relatively constant. An increase in CV allows more chances for the scheduling 2034 J. A. Millet al. heuristic to reduce changeover time, which could lead to more orders being moved far away from their due date. Under low CV changeover time can be reduced with far less movement of orders, thus reducing the effect on shortages. Thus process industry environments characterized by high CV and low TBO indicate potential for significant improvement in changeover time and shortage performance through revising the MPS. The frequency of revision is dependent on the scheduling heuristic used to revise the MPS. Therefore, to improve overall plant performance, managers should look to decrease the variability in changeovers between products through improvements in product and process design. If further improvement cannot be made in this area, then using scheduling heuristics to revise the MPS is another alternative. 6.3. Conclusions These experiments demonstrate the importance of revising the MPS in process industry environments with sequence dependent changeovers. These results represent significant improvements against the unrevised MPS. The results are especially significant since they were achieved under operating conditions not considered in master production scheduling literature where changeovers were assumed to be either negligible or sequence independent. Previous authors (Sridharan and Berry 1990, Zhao and Lee 1993, and Yang and Jacobs 1999) have found replanning not to be as significant a factor as we have and their results were not as conclusive. This study is an exploratory examination of revising the MPS in process industries with sequence-dependent changeovers. The paper also builds on previous research by explicitly studying the combined effects of revising the MPS with processes that have multiple products with sequence dependent changeovers. Further work should investigate the design of the master production schedule in process industries with sequence dependent changeovers, and the sensitivity of scheduling methods to changes in operating conditions. References CLARK. A. and CLARK. S.. 2000. Rolling horizon lol-sizing when sel-up times are sequence dependent. International Jounuil of Production Research. 38(10), 2287-2307. D E MATTA, R . and GUIGNARD, M., 1995, The performance of rolling production schedules in a process industry. HE Transactions. 27. 564- 573. GAVHIT, J. W . , 1965, Three heuristic rules for sequencing jobs lo a single production facility. Management Science. 1I(H). 166 176. GuiNET. A.. 1993. Scheduling sequence-dependent Jobs on identical parallel machines to minimize completion time. International Journal of Production Research. 31(7). 15791594. GUPTA. J. N . D . and DARROW. W. P.. 1986. The two-machine sequence dependent llowshop scheduling problem. European Journal oJ Operational Research. 24. 439 446. HAASE, K . and KIMMS, A., 2000. Lot sizing and scheduling with sequence-dependent setup costs and times and efficient rescheduling opportunities. Internationa! Journ,il of Production Economics. 66, 159-169. HILL, J. A. BERRY, W . L . , LEONG. G . K . and SCHILLING, D . A., 2000. Master production scheduling in capacitated sequence-dependent process industries. International Journal of Production Re.wareh. 38( I S). 4743 4761. KERN. G . and Wi-.i, J., 1996, Master production rescheduling policy in capacity-constrained just-in-time makc-to-stock environments. 27(2). 365 387. LESCHKE. J. P.. 1995, Plastech, Inc.—the importance of matching production capabilities and market requirements. Production and Inventory Management Journal. 3, 11-15. Revisitig the master production schedule 2035 LIN, S. and KERNIGHAN. B., 1973, An cITeclive heiinstic algorithm for the traveling salesman problem. Operations Resecireh .Journal. 21, 498-516. LIN. N . and KRAJrwsKi. L.. 1992, A model for master production scheduling in uncertain environments. Di'iision Science Journal.. 23(4). 839-S61. OLLIFF, M . D . and BURCH, E. E., 1985. Miiltiproduct production scheduling ut OwensCorning Fiberglas. Interfaces. 15. 25-34. SRIDHARAN. S. and BKRRY. W . L., 1990, Freezing the master production schedule under demand uncertainty. Decision Sciences.. 21(1), 97-120. SRIDHARAN, S. V. and LAFORGL, R. L., 1994. Freezing the master production schedule: implications for fill rale, Deei.sion Sciences. 25(3), 461 469, SRIDHARAN, S.V.,BhRRY,W.L. and UDAYABHANL', V., 1987. Freezing the master production schedule under rolling planning horizon. Management Science. 33(9), 1137-1149. VENKATARAMAN, R,, 1996, Frequency of replanning in a rolling horizon master production schedule for a process industry environment: a case study. Production and Operations Management. 5(3), 255 265, WEMMERLOV, U . , 1979, Design factors in MRP systems: a limited survey. Production and Inventory Management. 20(4). 15-35. WEMMERLOV. U , and WHYBARK, D , C , 1984, Lot-sizing under demand uncertainty in a rolling schedule environment. International Journal of Production Re.search. 22(3), 367 384, WILKERSON, J. and IRWIN, J.. 1971, An algorithm for scheduling independent tasks, .4!IE Transactions. 3, 239 245, YANG, K . and JACOBS. F,, 1999, Replanning the master production schedule for a capacity constrained job shop. Decision Sciences. 30(3), 719-748. YANO, C . A . and CARLSON. R. C , 1987, Interaction between frequency of rescheduling and the role of safety stock in material requirements planning systems. International Journal of Production Research. 25(2). 221-232. ZHAO, X. and LEE, T . S.. 1993, Freezing the master production schedule for material requirements planning systems under demand uncertainty. Journal of Operations Management., 11(2), 185 205,