INT. J. PROD, RES., 1986, VOL. 24, NO. 2, 359-374 Capacity control policies for MRP systems KURT M. GUTZMANN¢ and RICHARD A. W KE ‘Two types of heuristic capacity-control policies are evaluated using a simulation model of a materials-requirements-planning production system. ‘The control- policy decisions are based solely on the size of the queue, as measured in standard hours of work, at each work centre in the production system. Several classes of product mixes and product structures and several levels of the control parameters: ofeach policy, are investigated. The results indicate that each control policy gives, rise to a unique population of weekly labour, work in process and intermediate inventory level. Product structure is identified as a major variable in materials- requirements-planning systems performance. Sensitivity analysis of the cost functions for each policy indicates the conditions under which it will minimize the sum of labour costs, work-in-process holding costs and inventory holding costs. Introduction Material-requirements-planning (MRP) systems operate under the implicit assumption of sufficient productive capacity to meet the production plans generated by them (Orlicky 1975). An MRP production-planning system will not, therefore, realize its potential effectiveness unless it is coupled with some means of controlling capacity. In the short run, capacity is fixed by the capital equipment on hand, so that changing demands for capacity are usually met by varying the length of the working week through such means as overtime and multiple shifts. ‘The efficacy of a capacity-control policy may be measured by the production costs incurred from its use. In this paper, the performances of two heuristic eapacity control policies under varying production environments are considered and the major factors influencing the selection and implementation of capacity-control policies in MRP-based production systems identified. Huang et al. (1982) used a simple system modelled with Q-GERT to investigate capacity- and material-requirements problems. A simulation approach is also used here, with a more general SLAM (Pritsker and Pegden 1979) discrete-event model (Gutzmann 1983). The heuristic capacity-control policies investigated, and the design of the simulation experiment used to bench-mark the control policies, will be described. The experiment: objectives and product structure ‘The objectives of this investigation were: (1) to develop an MRP/production system simulation model, and (2) to use the model to evaluate the effects of: (a) two output control policies on production-system operating costs in an MRP production shop, Received October 1984. Planning Research Corporation, 1500 Planning Research Drive, MeLean, Virginia 22102 } The Pennsylvania State Copyright © 2001. All Rights Reserved.K. M. Gutzmann and R.A. Wysk (b) product structure on the operating characteristics of those policies, and (c) varying the control-policy parameters. Frequent reference will be made to various mixes of product structure. The terms used to describe product structures are defined in Table 1 Product structure Definition Flat Have one or two levels of parts and tend to spread horizontally Tall Have a relatively large number of levels and do not spread horizontally but extend vertically Mixed Have tallness and flatness characteristics within one structure Tall and flat product mix Several product structures each of which is either flat or tall Mixed- product mix Several product structures each of which may be classified as @ mixed-product structure Table 1. Product structure terminology Output control policies The objective of the experiment was to evaluate the effects of two types of control policies and their associated parameters on the cost of production operations. The ‘two control policies investigated were: (1) a simple production-switching control policy (Eilon 1975), and (2) a backlog-reduction policy. The production-switching (PS) policy operates as follows: (1) Two levels of production, high and low, are specified; low may correspond to one shift per week and high to two (2) Acontrol limit (CL) is specified for the number of standard hours in the work- centre queue at the beginning of the work week (3) Ifthe number of standard hours in the work-centre queue at the beginning of the week exceeds the CL, the high production level is implemented for that week: otherwise, the low production level is implemented The parameters of this policy are the CL and the low and high production levels (work-week lengths). The backlog-reduction (BR) policy operates as follows: (1) A control limit is specified as for the produetion-switching policy (2) A reduction factor greater than zero is specified (3) A nominal (minimum) work-week length is specified. (4) A maximum work-week length is specified. (5) If the number of standard hours in the queue at the beginning of the work week is less than or equal to the CZ, the nominal work-week length is implemented. (6) If the number of standard hours in the queue at the beginning of the work week exceeds the CL, the work-week length is determined from week length = nominal length +reduetion factor x (queue length—CZ). Ifthis calculated week length exceeds the maximum, the week length is set to the maximum. Copyright © 2001. All Rights Reserved.Capacity control policies for MRP systems 361 The parameters of the baeklog-reduetion policy are the CL, the reduction factor and the work-week lengths, nominal and maximum. Production costs ‘Two types of costs were deemed pertinent to this investigation—labour and inventory-holding costs. These costs were assessed as unit costs to permit sensitivity analysis of policy performance based on the ratio of labour to holding costs (R).. Inventory costs were identified as the sum of the component inventory levels (finished components) and the work-in-process level. Uniform holding costs for all parts was assumed. The average weekly total cost function (A WTC) was computed as follows: AWTC=C,(AWL)+C,(AI + AWIP) a) where AWTC Average weekly total cost ($) AWL average weekly labour (h) AI average inventory level (parts), inventory being defined as the stocks of components and subassemblies that have completed all processing on their routings. AWIP average work-in-process level (parts), work in process (WIP) being defined as all orders either being processed at a work centre or in a work. centre queue waiting to be processed. ©, hourly labour and burden rate ($/h) C, weekly holding cost for inventory and WIP ($/part per week) By dividing eqn. (1) by ©,, AWC can be evaluated as a function of the ratio of labour to holding costs. This permits sensitivity analysis of the results to determine ranges of the ratio where one policy would dominate the others. The resulting cost, function in terms of equivalent labour hours (ELH) per week is then: ELH=AWL+(C2/C,\(AI +A WIP) 2) ELH =AWL+R(AI+AWIP) (3) where ELH is the equivalent labour hours per week and R=C,/0, Eqn. (3) yields the average weekly cost of production in terms of equivalent labour hours, that is the WIP and inventory-holding costs have been converted to labour hour equivalent units. AWTC for a specife value of C, is found by multiplying ELH by C. Simulation model event SLAM (Pritsker and Pegden 1979) simulation model was constructed to evaluate the control policies. Simulation was selected because: (1) detailed analysis of a complex MRP system is relatively easy and (2) simulation provides flexibility in experimental design. A discret Copyright © 2001. All Rights Reserved.362 K. M. Gutemann and R.A. Wysk Logic of model ‘The simulation model implements standard MRP logic (Fig. 1). A more detailed description has been given previously (Gutzmann 1983). Although a simulation model, it operates essentially deterministically. Since product structures and routings are fixed, it processes entities in a known manner and its only variation is from the master schedule. Using the same sequence of end-item demands for each alternative reduces the variance of the estimators, Master schedule Pritsker and Pegden (1979) note that using trace-driven models for comparing alternatives reduces variance when the time series used affects both alternatives similarly. The master schedule for the model is derived by exponentially smoothing a Poisson sequence of weekly demands for each end item. This is the only souree of variation in the model, since, in an MRP system, only end items are of independent demand. Smoothing is applied to the time series to reflect the management. practice of production levelling. TNFETALIZE ‘EKPLODE MASTER [scuepuLe ReguIRenEwrs| arcs ae tz | [v0 connor ADJUST capacity PROCESS ORDERS PER ROUTINGS ROVE. FRGI ADVANCE TIME Figure 1. Logic of simulation model Copyright © 2001. All Rights Reserved.Capacity control policies for MRP systems 363 Routings Routings were generated randomly for each set of parts associated with a product structure mix, the number of operations being uniformly distributed between 2 and 6. Operation times (in hours) were normally distributed with a mean of 0-10 and a standard deviation of 0-05 (negative values rejected). Set-up times (in hours) were fixed at a constant 0-01. Quantity multipliers were set at 1 for all parts, since the quantity multiplier and the operation time are directly related. Master schedule and demand rates For each set of product structures, the same master schedule drove the system This is a ‘trace-driven’ design and reduces the variance of the estimators. Demand rates were set to load the shop at 105% of nominal capacity on average. The smoothing factor was set at 0-15, The demand rates for individual items within a product mix were adjusted relative to each other to balance the work-load at each work centre, Lead times In industry, a common practice for determining lead times is to allow one week per operation. This was tried in preliminary experiments, but certain parts with longer operation times or more operations blocked their parent assemblies. Lead times for these parts were then increased until the blockage was either eliminated or reduced to a very low level. The planning-horizon length was set to the maximum cumulative lead of the end items produced. This is necessary in MRP systems to ensure that all requirements are captured within the planning horizon. Lot sizing Lot-for-lot ordering was used for all the experiments. Orlicky (1975) notes that such ordering minimizes inventory-holding costs. Using an ordering policy that minimizes holding costs for all alternatives provides a common basis for comparing the effect of the output-control policies on inventory. When analysing MRP lot sizing procedures, the planned lead times are assumed to equal the actual manufacturing or delivery lead times but in actual production plants, this is not usually true. If planned lead times are much longer than actual lead times, then the average inventory level will be higher than if the actual and planned lead times were approximately equal. Since lead times were maintained at constant values through out @ product structure mix, the difference between actual and planned lead times would be due solely to the control po Work centres Four work centres each with two machines were used for all the experiments. The nominal week length was 40 hours and the maximum 80 hours. Since capacity is the week length x the number of machines, the nominal capacity was 80 hours per week and the maximum 160 for each work centre. Buffa and Zaubert (1972) note that the number of machines in a job-shop simulation model has never been shown to be a. significant variable. The production system used in this analysis is shown in Fig. 2 and typical (although not all) product structures in Figs. 3 to 6 Copyright © 2001. All Rights Reserved.364 K. M. Gutzmann and R.A. Wysk INVENTORY wep (RAW MATERIAL 8 SUBASSEMBUES) 7 O00) bro samo Ln | OOO! (MACH 2 ‘QUEUE FOR WK CTR & SUBASSEMBLIES RETURN TO RENTORY Figure 2. Experimental production system: TEEN | (Sian o o Pane Figure 3. Flat product structures, Copyright © 2001. All Rights Reserved. work ceNrne «Capacity control policies for MRP systems 365 Oe @” [Om ‘ Figure 4. Tall product structures, Sample size Reliable estimates of three quantities were desired: the average number of hours worked per week, the average total inventory level and the average work-in-process level. Preliminary runs were made for all product mixes, these estimators being plotted as a function of the number of weeks of simulated production. These plots indicated that all the estimators varied by less than 1% from one observation to the next after 550 simulated weeks. Statistics were cleared after 50 weeks. This gave a sample size of 500 weeks for all runs. All the experiments were replicated 11 times to give independent samples of AWL, AI, and AWTP. Presentation of results ‘The information presented is divided into two major sections: (1) Definition of the statistical test used and the cost sensitivity analysis and (2) means and standard deviations for weekly labour, total inventory and work in process. Statistical and cost analysis procedures Hotelling T? test The Hotelling 7? test is used to test the hypothesis that two samples were from a multivariate normal population. The null hypothesis then is that the mean vectors of Copyright © 2001. All Rights Reserved.366 KM. Gutzmann and R.A. Wysk Figure 5. Mixed product structures, each sample are equal. This test is invariant under linear transformations of the data (Morrison 1967). Significant differences in weekly labour, inventory level and work. in-process level may, therefore, be detected without applying cost coefficients to the results. Within the major subdivision of product structure, this test is applied pairwise to the six policy alternatives, giving 15 combinations tested for cach product structure. Cost-sensitivity analysis ‘The cost function given in eqn. (3) is parametric and based on the ratio of the weekly holding cost per part to the cost per hour of labour and burden. Since the estimators of labour are given in hours per week, the cost coefficients so defined are compatible in terms of units. The maximum-likelihood estimators for weekly labour, inventory level and work-in-process level are their means, which are also unbiased, and may therefore be used to determine the weekly equivalent labour hours (ELH), A typical range of labour and burden costs is $10 to $100 per hour. A typical range of the holding costs is based on the time value of money (interest rate) and the value of a part. Other cost factors contribute to the actual holding cost a firm exp such as warehouse costs, insurance and security. The holding-eost range ca however, be reasonably estimated from the opportunity cost. If the interest is set at 10% annually and part values range from $0-00 to $500000, the weekly holding costs ences, Copyright © 2001. All Rights Reserved.Capacity control policies for MRP systems 367 sunny @ (Om: Figure 6. ‘Tall and flat product structures. will range from 0-0 to 961'53 §/week per part. This range would apply to most manufacturing operations today. Combining the low and high values of labour and holding costs, the range of their ratios is then (+1 to 16-0255. Within this range, the cost function for each policy is evaluated to determine the range of values where one poliey dominates the others in terms of minimizing the weekly cost. The information generated by this analysis is a range of the cost ratios where one policy dominates all the others in terms of minimizing the weekly equivalent labour hours (ELH). Discussion of simulation results Statistical T? test results Table 2 gives the nomenclature and layout of the experimental design, and ‘Table 3 the means and standard deviations for weekly labour, work-in-process level and inventory level for each of the 24 experiments. Results of the pairwise tests of mean vectors with the Hotelling 7? test for each product structure (Table 4) indicate that each policy gives rise to a unique population of labour hours and inventory and work: ss levels. This is supported where the significance level for any policy pair is greater than 1 —0-008. Copyright © 2001. All Rights Reserved.368 KM. Gutzmann and R.A, Wysk Product Reduction Control structure factort limit Run (PS) Policy RE OL 1A flat HL 80 1B flat HL 120 Ic flat BR 80 1D flat BR 120 1E flat BR 80 Fr flat BR 120 2A tall = HL 80 2B tall = HL 120 2c tall = BR 80 2b tall BR 120 2E tall BR 80 oF tall = BR 120 3A flat & tall = HL 80 3B fiat & tall = HL 120 3C flat & tall BR. 80 3b fiat & tall BR. 120 3E flat & tall = BR. 80 3F flat & tall = BR. 120 4A mixed = HL 80 4B mixed = HL 120 4c mixed BR 80 4D mixed BR 120 4E, mixed BR 80 4aF mixed = BR 120 +HL: High/Low production switching. BR: Backlog Redu $NA: Not applicable, ‘Table 2. Experimental design, Observations from cost-sensitivity analysis Several observations about the various control policies can be made from the results of the cost-sensitivity analy (ay (2): (3): is (Table 5 and Figs. 7 to 10) Policy A was the preferred policy for all four product mixes when the cost ratio R was relatively large. Policy D was the preferred poli for all product structure mixes except the tall/flat, when the cost ratio R was relatively small. The production switching (PS) policies (A and B) were preferred for the tall flat product structure mix, and the backlog reduction (BR) policies were not preferred on any interval of the cost ratio for this produet structure mix (Fig. 9). Policy B was preferred for the lower values of Rand policy A for the relatively larger. : As the cost ratio R increased the order in which the policies were preferred was D-C-E-A for the flat and the tall product structure mixes (Figs. 7 and 8) and D-F-C-E-A for the mixed product structure mix (Fig. 10) The weekly labour for each policy (except A which is labour-intensive) did not vary significantly amongst the product structure mixes (Table 5), their values being within three standard deviations of each other These efficient operational ranges are summarized in Table 5, Copyright © 2001. All Rights Reserved.Capacity control policies for MRP systems Work in process Labour (hjwk) Inventory (partst) (partst) Product standard standard standard structure Experiment mean deviation mean deviation mean _ deviation Flat 1A 181 14538 558 1B 129 1961-1 1381 1c 130 17679 836 1D 134 27813 1481 1E 136 16250 838 IF 150 26681 7 Tall 2A 307 105841 1967.7 369 2B 327 10084 32282 925 20 3279925, 30670 818 2b 332 11081 41072 1207 2E 340 0848. 26215 540 oF 322 10741 4250 1265, ‘Tall and 16133393 (2245-9 234 flat 21800 aT3 38 618 719 248 1163 Mixed 4a 134 317 4B 110 1337 4c 132 472 4D 11925478 1029 4E 14516764 434 4F 132 2716 47 + Time-integrated averages. ‘Table 3. Simulation results. (Sample size: 11 replications, 550 weeks simulation, week 60.) Conclusions and recommendations Observations 3 and 4 indicate that homogeneity of product structure within a product mix leads to a greater choice of alternative control policies as the cost ratio R varies. Conversely, non-homogeneity of product structure within a product mix apparently limits the number of policies that may be considered to minimize operating costs. ‘The order in which policies are preferred as R increases, indicates that the control limit (CL) has a larger impact on operating costs than other policy parameters. Logically, the CL directly controls the amount of WIP, whereas the reduction factor (RF) for the BR policies influences labour more than WIP. Only two values of each of these parameters were observed in this investigation, however, so this may not be a, general conclusion Copyright © 2001. All Rights Reserved.Wysk Gutemann and R.A M K 370 (481 €) a 405 wopaars jo saastop “11 ois aldieg) synsor [wonsHeIS Jo SMUNG, ¥ ae, "809 Jo [Aj aouwogTEE e-0L08 POZLL 20000 ponporg Copyright © 2001. All Rights Reserved.Capacity control policies for MRP systems 371 On a practical level, the results permit the following recommendations to be made: () ntensive control pol labour costs is high. (2) Labour-conservative control policies should be used when the ratio of holding to labour costs is low. (3) Control policies that balance labour and holding eosts to minimize operating costs are indicated when neither labour nor holding cost completely dominates the cost function. -s should be used when the ratio of holding to ‘The logie underlying these recommendations is that labour-intensive policies (low CL) tend to work off any backlogs, thus keeping inventory and WIP levels (and costs) low. Likewise, where labour is the more costly commodity, large backlogs do not impose significant penalties. ‘The first-line production manager can now control his labour costs simply and effectively, simply because only knowledge of the backlog at the start of a work week is required to make a decision. Detailed knowledge of the planned work-load for the Product Lower Upper structures Poliey bound BLHt bound ELHt Flat 1D 0-000000E 00 33564 0-307098E-03 338-63 1c 0-307098E-03. 33863 0-7648031-03 34148, 1E 0-764803E-03 34148, 0:5727918-02 368-66 1A 0572791 8-02 36866 0-160255E 02 6826206 1F Not optimal over any interval 1B Not optimal over any interval Tall 2D 0-000000E 00 33490 026769602 350-47 2c 0-267696E-02 35047 0-4854268-02 35931 2E 0-4854268-02 35931 018103901 40709 24 0:181039E-01 40709 —0-160255E 02 4868520 oF "t optimal over any interval 2B ‘Not optimal aver any interval ‘Tall and 3F 0-000000E 00 40 0-000000E 00 353-40 flat 3B 0-000000E 00 40 0393748E-02 37680 3A (039374802 37680 O-160255E 026210880 3C Not optimal over any interval 3D Not optimal over any interval 3B, Not optimal over any interval Mixed 4D 90000008 00 34380 0356692E-03 34668 aE 0356692103, (08831458-03, 4c 0-883145E-03 35036 012306202 1E 0-1230628-0: 30219 01431 90E-01 4a 0-143190E-01 4 0:160255E 02 Not optimal over any interval + ELH=AWL+R(AI + AWIP), where ELH is equivalent labour hours (cost), AWL average weekly labour, & ratio of weekly inventory holding cost to hourly labour-burden rate, AT average ntory and AWTP average work in process. Table 5. Ranges of cost ratio R over which policies achieve minimum cost Copyright © 2001. All Rights Reserved.Lira) $unon wots) anstearn03s 372 KM. Gutzmann and R.A. Wysk coming weeks is not explicitly required as in other methods of capacity control, such as load capacity analysis and finite-loading. How much of the backlog will be worked off in the coming week is determined by the control policy parameters which, in ‘actual practice, will probably be determined by higher management after an analy based on their particular product structures, product mixes, labour costs, holding LINE TYPE pouiey Figure 7. Cost functions for flat product structures, Figure 8. Cost functions for tall product structures, Copyright © 2001. All Rights Reserved.Capacity control policies for MRP systems 73 LINE TYPE Pouicr Figure 9. Cost functions for tall and flat product structures. ELH = UL + ReRTeAUTPD E Line Tye poLtey - (oma) sunow soReT ANE WANES 008 “a8 "eis cont ratio ® Figure 10. Cost funetions for mixed product structures, costs and shop configuration. This analysis is possible with the simulation model developed for the current research. ‘This study has indicated that product structures have the greatest impact on production costs and control-policy alternatives. The MRP simulator developed ‘enables management to develop optimal policies for capacity control Deux types de politiques de contrdle heuristique de la capacité sont évalués a aide d’un modéle de simulation d’un systéme de production avee planification des besoins en matiére premiére. Les décisions de politique de contréle sont basées, uniquement sur la taille de la queue, mesurée en heures standards de travail, & chaque poste de travail dans le systéme de production. Différentes classes de Copyright © 2001. All Rights Reserved.374, Capacity control poticies for MRP systems mélanges de produits ot de structures de produits sont examinées, ainsi que plusieurs niveaux des paramétres de controle de chaque politique. Les résultats indiquent que chaque politique de contrdle donnerait lieu & une population unique de main d'oeuvre a la semaine, de travaux en cours, et de niveaux Cinventaires intermédiaires, La structure du produit est identifige comme une variable importante dans la performance des systémes A planification des besoins en matiére premiére, Des analyses de sensibilité des fonctions du coat pour chaque politique indiquent les conditions selon lesquelles celle-ci minimisera la somme des colts de main d’oeuvre, de maintien des travaux en cours et de maintien de Pinventaire. Zwei unterschiedliche, heuristische Kapazititskontrollverfahren werden mit Hilfe eines Modells zur Simulation von Herstellungssystemen mit Material bedarisplanung untersucht. Die Entscheidungen der Kontroliverfahren beruhen einzig auf der Linge der Warteschlangen, ide in Standardarbeitsstunden an jeder Bearbeitungsstation im Herstellungssystem gemessen wird. Mehrere Acten von Produktkombinationen und -strukturen werden ebenso untersucht wie verschiedene Werte fiir die Kontrollparameter jedes Verfahrens. Die Engebnisse zeigen, daB jedes Kontrollverfahren eine einzigartige Kombination von wachentlicher Arbeit, augenblicklich stattfindender Arbeit und mittlerer Lagerbestandsgrife bewirkt. Die Produktstruktur wird als eine Hauptvariable der Leistung von Materialbedarfsplanungssystemen herausgestellt. Empfindlieh keitsanalysen der Kostenfunktionen fiir jedes Verfahren deuten auf die Bedingungen hin, unter denen es die Summe der Lohn- und der Lagerungskosten und der Kosten zur Gewahrung kontinuierlicher Arbeit auf ein Minimum reduriert. References Burra, Euwoon, and Taunzer, Wri1aM, 1972, Production-Inventory Systems/Planning and Control (Homewood, Ilinois: Richard D. Irwin, Ine.) 1975, Five approaches to aggregate production planning, American Institute of Industrial Engineers, Transactions, 7, 18 Gvramanw, Kur, 1983, Capacity control policies in an MRP production environment, ‘Master's thesis, Virginia Polytechnic Institute and State University, Blacksburg, VA (unpublished). Hvano, P. Y., CLaytox, KE. R., and Moore, L. J., 1982, Analysis of material and capacity requirements with Q-GERT, International Journal of Production Research, 20, 701 Moraisox, D. F., 1967, Multi-variate Statistical Methods (New York: MeGraw-Hill Book Co). Onticky, JosaPH, 1975, Material Requirements Planning (New York: McGraw-Hill Book €o.) Prirskr, A. A, B., and Peapen, C. D., 1979, Introduction to Simulation and SLAM (New York: John Wiley & Sons). Eno! Copyright © 2001. All Rights Reserved.