Cyclical production planning without losing flexibility

Transcription

1 Cyclical production planning without losing flexibility Simons, Paul H.W.; Fransoo, J.C.; VanDijk, Jan Published: 01/01/1994 Document Version Publisher s PDF, also known as Version of Record (includes final page, issue and volume numbers) Please check the document version of this publication: A submitted manuscript is the author's version of the article upon submission and before peer-review. There can be important differences between the submitted version and the official published version of record. People interested in the research are advised to contact the author for the final version of the publication, or visit the DO to the publisher's website. The final author version and the galley proof are versions of the publication after peer review. The final published version features the final layout of the paper including the volume, issue and page numbers. Link to publication Citation for published version (APA): Simons, P. H. W., Fransoo, J. C., & VanDijk, J. (1994). Cyclical production planning without losing flexibility. (TU Eindhoven. Fac. TBDK, Vakgroep LBS : working paper series; Vol. 9401). Eindhoven: Eindhoven University of Technology. General rights Copyright and moral rights for the publications made accessible in the public portal are retained by the authors and/or other copyright owners and it is a condition of accessing publications that users recognise and abide by the legal requirements associated with these rights. Users may download and print one copy of any publication from the public portal for the purpose of private study or research. You may not further distribute the material or use it for any profit-making activity or commercial gain You may freely distribute the URL identifying the publication in the public portal? Take down policy f you believe that this document breaches copyright please contact us providing details, and we will remove access to the work immediately and investigate your claim. Download date: 15. Nov. 2018

2 DepaNment of Operations Planning and Control -- Working Paper Series Cyclical Production Planning without Losing Flexibility Paul H.W. Simons, Jan C. Fransoo, Jan VanDijk Research Report TUE/BDK/LBS/94-01 January 1994 Graduate School of ndustrial Engineering and Management Science Eindhoven University of Technology P.O. Box 513, Paviljoen F16 NL-5600 MB Eindhoven The Netherlands Phone: Fax: This paper should not be quoted or referred to without the prior written permission of the author(s). Please address all correspondence to Dr. Fransoo. The research of Dr. Fransoo has been made possible by a fellowship of the Royal Netherlands Academy of Arts and Sciences.

3 Cyclical Production Planning without wsing Flexibility * Paul H.W. Simons, Jan C. Fransoo, Jan VanDijk Eindhoven University of Technology Eindhoven The Netherlands ABSTRACT The plant considered in this study manufactures synthetic building panels which are used for exterior and interior applications. The plant is a subsidiary of a major European chemical company. A few thousand different varieties (in three families) are delivered to customers in Europe. The manufacturing process is concentrated around the presses. Traditional problem characteristics in semi-continuous process industries may be found here, such as high khongeover ~i.rqe~ and ~~ensive installations. ~_o.f..jh.e. pradu~t _ar.e"prqguc~dj! ~or-'hng.to c~q!!!~,.,9!~,~. Production control problems at the plant included the coordination between manufacturing and sales, capacity management in the press department (including batching and sequencing), and long lead times. We developed a hierarchical production planning concept. This concept focuses on the one hand on an efficient use of capacity in the pressing department (using cycle time control) and on the other hand on a well-balanced customer order acceptance procedure. n this paper, we focus on the various hierarchical relationships and on the methodology by which the planner can maintain the production cycles without giving up flexibility to the customer. The company has implemented the concept and now uses it with the support of a standard MRP software package. NTRODUCTON The hierarchical production control policy in this paper has been developed for a manufacturer of building panels for exterior and interior applications. This manufacturer is a subsidiary of a major European chemical company. A few thousand different varieties (in three families) are delivered to customers in various European countries. The manufacturing process is concentrated around the presses. There, the thermosetting resins, reinforced with cellulose fibers, are pressed to form the panels. n the press department, traditional semi-continuous process industries problem characteristics may be found, such as high changeover times and expensive installations [1]. Before pressing, the various fiber layers are assembled from feedstock material. All building panels are pressed to order, i.e. generally a panel is. manufactured only, if a customer order is This paper is based on the un ublished Master's thesis of Paul Simons (Eindhoven University of Technology, Graduate School 0 n ustrial Engineering and Management Science). All correspondence should be addressed to Dr. Fransoo. 2

4 present. The reason for this is the large variety in final products (a few thousand), while the variety in raw materials is limited (a few hundred). After pressing, the products are finished and distributed to the customers. Production capacity is sufficient in the assembly, finishing, and distribution operation, but it is restricted in the press department. Therefore, in this paper, we will f~s on a s!~ press and~~ the other operations. AdilitiOilaily, the Situation (fuscrbed in this paper simplifies the -actuarstuatlon at-me company. This has been done to improve the clarity of exposition. The production control problem in make-to-order situations in semi-continuous process industries is focused on an efficient use of the bottleneck capacity. The available capacity for production is determined by the amount of capacity spent on changing over, assuming the technical availability of the system is given. Any capacity spent on changing over cannot be used for production. Since products may only be manufactured if a customer order is present, batching the orders into economic lots may lead to long and varying lead times. The hierarchical approach presented in this paper keeps ~!!:21 2YG the ra~q_ otce.~~~qy'el~},q~p.j()d\lctjy~_c~p~ itt_<!nqleaqs,!5> a f~~d all(la..~~~ptable!'~'!cl time of customer orders _- fn~tilenert section, we will briefly review some approaches to similar problems presented in the literature. After having outlined the structure and basic ideas of the hierarchical control approach, we will present the results of a series of simulation experiments. These were used to evaluate the policy concerning delivery reliability and several other performance measures before actual implementation at the company. n the discussion section, we will address some additional issues regarding implementation in practice. LTERATURE REVEW Lot sizing and scheduling on a single machine has been a prominent research topic since the 1960's. Most of the research however is limited to stationary deterministic problems (such as the economic lot scheduling problem [2]). n 1978, Vergin and Lee [3] demonstrated that the use of deterministic lot sizing and sequencing rules leads to a bad performance in stochastic environments. Consequently, several articles have been published addressing the lot sizing question under uncertainty [3,4] or under dynamic deterministic conditions [5]. Campbell and Mabert~]in~~s_ti~a~e t~e costofus.i~~~~lical schedules under dynamic demand and uncertainty. They conclude that the extra cost of cyclical schedules is very J1linor as compared to the well-known advantages of simplicity and ease of control. All these papers are limited to make-to-stock situations. n [7], Bertrand et al. introduce a production control concept that provides for cyclical scheduling under uncertainty in make-to-order production situations. The idea is to accept customer orders against a capacity load curve that checks the aggregate available capacity against the aggregate capacity demanded by the accepted customer orders. All customer orders are allotted a fixed lead time that equals the length of the production cycle. f a product i is set up for production (exactly once per production 3

5 , cycle), all customer orders of product i that have been accepted since the last production run of i, are manufactured. Since the aggregate required capacity is controlled by the order acceptance procedure, the length of the total production cycle is supposed to be almost constant. The present application is an extension of the work presented by Bertrand et al. in several ways. FirsttJhe si~uatiq!l!ljl(t'!l(;on~i~~ration is, more c2!!l~!<?~j~e._eause the CY~l~UiD~,s"_Qf,~llPJo<:lllGtsJ\'~" not.identi<;aj. St'!c0!-~, we have inve~t_ig~te~.j_~_~,~~havior _ of the proce(hlr(,! jn~~~n~iy~_ ~id!1.1ajj()1l~xp~rlments t().investigat~ both.th~d~li"ery PeJfQlnl~Jt~,a1l4_tiiestabili!Y,_QLthe pr()d\l~!i()llcycles.j'hird, the company has ~ompletely implemented the procedure(!nd included it in the MR:P software S~!~l!l' THE HERARCHCAL CYCLE TME MODEL As mentioned above, the capacity of the plant is determined, to a large extent, by the capacity of the press. The capacity of the press is either used for production or for changing over. The changeover times between the three product families are relatively long, while the changeover times between the various products within a single family are very short. To retain an efficient use of the capacity, the number of changeovers between product families needs to be limited. n other words, the production runs of each product family need to be sufficiently long to ensure capacity feasibility. The parameter, which determines the use of capacity, is the cycle time. The cycle time is defined as the length of time between two consecutive starts of production runs of the same product. Th~)~~~tti.mt'!Jhat can be agreec:lllp_q~ L~tthll:t_ef!!~!Qm~rj~.9Jr:eftly ~ dependent upon the cycle time (if the lead time Parameter Setting for the preceding and succeeding operations is added), since each product family is set up for production exactly once during its cycle time. f "- ~ capac!ty\lti!i?-~!!q.l!..j~_.,y~!y_lljgl1~.._(l pohcy with stable cycle times leads to a higher service level _and control of capacity than Jlpolicy with variabl~ lead till_~s [8]. This is caused by the fact that Order Acceptance any production capacity spent on setting up due to a decrease in cycle time, cannot be regained ~ l-'" later - unless at the expense of an increase in lead time. Lot Sizing Sequencing The control of cycle times is structured in a!hree-tiered hierarchical model (see Figure 1). Figure 1. Three-tzered hzerarchical. model The top level of the model sets the parameters of the production system. n this way it determines the lead time of the building panels and the throughput of the plant. The middle level of the model is the order acceptance procedure. The objective of this procedure is to ensure that the capacity required by the '4

6 accepted customer orders does not exceed the plant's planned available capacity. Finally, at the bottom level of the model, the sequencing and scheduling decisions are made. We will now consecutively discuss the three hierarchical levels. Top Level: Parameter Setting f we assume that the aggregate demand is stable, the following balance equation may be formulated. where ; n d Pi c T C N n {d. c.}.e...!. +...: s en ;1 Pi Ti product family index number of product families demand (to accept) for product family; (units/year) production rate of product family i (units!hour) changeover time to product family i (hours) cycle time of product family i (years) net capacity (hours/year) (1) The net capacity C N is the capacity available for production (productive capacity Cp ) or changeover (changeover capacity C c )' t therefore does not include time allotted to maintenance and other down time. C N is considered as a given. n the parameter setting decision a tactical tradeoff needs to be made between the cycle time (lead time) and the throughput (demand to accept). f the throughput is increased, so is the lead time. Therefore a careful balance needs to be found, taken the strict manufacturing balance equation as a basis in this marketing decision. Middle Level: Customer Order Acceptance The purpose of the middle level decision is to ensure that the aggregate capacity required by the accepted customer orders does not exceed the available productive capacity C p. However, the exact moment in time when a particular customer order will be pressed is unknown at the time of the order acceptance decision. At that moment we only know the interval in which the customer order will be pressed. This press interval is the part of the standard lead time reserved for pressing. Figure 2 displays four customer orders from the same family, each with its own due date and planned lead time. Every customer order has one opportunity to be pressed in its press interval. On average, however, the customer order will be pressed in the middle of its press interval. This moment is indicated as the average press moment. The start of the press interval is the earliest start for pressing this customer order. The end of the press interval is the latest 5

7 finish for pressing this customer order. The average press moment is used in the order acceptance procedure. The order acceptance procedure decision is based on the cumulative capacity load diagram (Figure 3). When a d customer order is received, the average press moment and the order work load in the press department are determined. Using the capacity load diagram, it is checked whether sufficient capacity is available to press the order. f this is the case, the order can be accepted and the standard lead time can be realized. Upon acceptance, the order may be released to the press department. Bottom Level: Lot Sizing and Sequencing orders, preceding press interval succeeding a operations ; operations pr~ing press interval: preceding ~ interval su~ing b operations : operations succet;ding C operations operations preceding press interval succeedoper_a~np Hope~~~t~~s4~!--~ ~~ ~~~ run start Figure 2. Position of the press interval in the lead time f time new load profile old load profile average press moment Figure 3. Order acceptance using load profiles n the press department, a set of accepted customer orders is available which need to be sequenced. The objective of the press interval is to enable the production planner to combine customer orders into production lots. f a production run of a family is started, all customer orders need to be produced, for which expression (2) holds: time where lfx - latest finish of order x Px expected processing time of order x to start of next production run Tf(x} - cycle time of the family of order x This can be visualized by a time window (see Figure 4). n the time window, the customer orders are represented by rectangles. These rectangles are positioned at their latest finish moment and their length is the expected ddration of the press operation. n the example of figure 4, starting now a run of family A would include two customer 6

8 orders, family B zero, and family C again two. All orders that are located outside the time window are not included in this production run. They will be included in the next run of this family, since after Tf(x) a new production run of this B ,., family is expected to be set up again for production. Note that the f time window gr , number of orders that are grouped C 'into a lot depends on the specific! moment in time. Once a family is to, selected for production, the time Figure 4. Lot sizing using time windows l A time ~~~~-~-~~-~~------~ window is moved, since to will then be the end of the production run of this family. n short, at the end of production of \ family, n orders of this family have been processed and to is reset: to : to + l: {Pi + Cj } 11 n (3) The selection of the next family to produce (sequencing) is a separate decision. Product families are selected such that the cumulative potential tardiness is minimized. The potential tardiness of an order is 0 if the order is completed before its latest finish, otherwise it is the difference between the completion time and the latest finish. The cumulative potential tardiness is the sum of the potential tardinesses of all orders. When the sequencing decision needs to be made, the cumulative potential tardiness of all possible family sequences is calculated. From the sequence with the least potential tardiness, the first family is selected as the next family to be produced. The cumulative potential tardiness is recalculated at the next decision moment, i.e. when the production run of the current family is completed. SMULATON EXPERMENTS n the previous section we have outlined the basic principles of control in the proposed hierarchical structure. The structure has been designed in order to reach a number of objectives: reduction of the lead time increase of the delivery reliability more efficient use of available resources (people, machines and money) The three objectives are interrelated. The approach by which these objectives will be reached is based on the assumption that the cycle time may be controlled with the 7

9 hierarchical approach. n a more or less stable situation, this seems probable. However, if the situation becomes more dynamic, this may not be so clear. Additionally, it is not clear from the analysis whether complicating factors influence the performance. Examples of complicating factors are: an uneven distribution of customer demand over the week, irregular aggregate demand over the year (seasonal pattern), etc. t is also important that some decision freedom for the planner remains after the lot sizing and sequencing decision has been made. Within the family production run, the planner needs some freedom to decide on the sequence of individual customer orders without infringing on the order due dates. Since these aspects are difficult to capture in an analytical model, a simulation model of a press has been built in SMULA. n this section, we will describe the experimental questions, the setting of the experiments and the results of the experiments.. Experimental Questions The basic question in the simulation experiments involves the robustness of the approach. How sensitive is the performance of the model for complicating (practical) factors, as mentioned above? The following research questions have been defined: 1. What is the performance of the proposed approach with respect to: mix variations between the product families irregular order arrivals over the week 2. What is the remaining decision freedom for the planner? We have defined five performance measures: the percentage of late orders that is late the average tardiness of late orders the mean and standard deviation of the order lead time the actual cycle times the planner's decision freedom The decision freedom of the planner is the ratio of the time span in which the order can be planned (between the start according to EDD-sequencing and its latest finish) and the time span between the start and the finish of the family production run. The exact definition can be found in the appendix. Experimental Setting n the simulation model, a single press is included. The press has three queues, one for each product family. All customer orders for a week are generated at the same time. For two families (B and C), they are generated using a normal distribution with 8

10 .u15 orders and a10 orders. The remaining number of orders is then fil1ed to 100 by family A orders. At the company, family A products are produced to adjust for fluctuations in aggregate demand. These products have little variety in demand and large volume, so they can be traded later easily without much risk. All generated customer orders have the same size. We are not interested in the behavior under varying order sizes for two reasons. First, in practice the variations are very limited and only a small group of orders has a significantly different order size. Second, the limited number of orders that are considerably larger in practice must be cut into smaller production orders because of the press capacity. Therefore, they can be considered as a group of smaller orders and are captured in the variation coefficient of demand. The generated customer orders are distributed over the days of the week, according to the pattern that is experienced at the company. Customer orders only arrive on week days, while processing takes place in the weekends as well. When a customer order arrives, it is checked whether sufficierit capacity is available according to the capacity check explained in the previous section. f capacity is sufficient, the order is accepted and the standard lead time is attached to the order. f capacity is insufficient, the order is not accepted and is given a different lead time, to be examined for acceptance later. The press time is assumed to be deterministic. Lot sizing and sequencing is done according to the procedure described in the previous section. f the potential tardiness for all groups equals zero, the family with the largest run length is selected. n the experiments, the length of the start run has been determined by a graphical steady state analysis. The start run length has been set at 1000 to 2500 days depending upon the utilization rate. The length of the subruns has been set at 1000 days, using the Von Neumann-ratio. Experimental Results The major results with a utilization rate of 100% are presented in Table 1. 9

11 Table 1. Major simulation results (utilization rate 100%) Simulation nr.: target cycle time family A target cycle time family B tar~et ~cle time family C percent of orders late average tardiness of late orders average lead time family A orders average lead time family Borders average lead time family Corders actual average cycle time family A actual average cycle time family B actual average cycle time family C decision freedom family A decision freedom family B decision freedom famil~ C Note: cycle time, lead time, and tardiness are given in days. "Late" has been defined as more than 0.2 days late. The 100% utilization rate situation is the most realistic one for the practical situation, since internal (inventory) orders are generated for family A up to 100% utilization to ensure a high usage of the press. t is interesting to see that the lead time is still controlled and the percent of late orders is very small. Additionally, the average tardiness of late orders is negligible. Less than half a day late means no lateness in practice, since production takes place during days and nights, while products are distributed during the day only. t is probable that no orders will arrive late at the distribution points if the procedure is implemented in practice. Simulation results of lower utilization rates, which are not presented here, show a delivery reliability of over 98 percent at 99% utilization. The various cycle time variations have been tested to investigate their influence on the system's performance. Major criteria in determining them in practice are the capacity feasibility (equation 1) and the required lead time by the customers. Note that the actual cycle times which are realized while the system is operated, are very close to the target values of the cycle time parameters. The lead times are approximately a little more than half the cycle time for the families Band C. The lead times of family A are considerably larger. This is due to the fact that family A has a larger share in total demand than families Band C. Consequently, the production runs of family A are relatively large and orders of family A are produced relatively later after the average press moment than orders of family B and C. A more refined version of the order acceptance procedure would define the average press moment of order x as: lfx - es x es x + 2 +o.sxrlf(x) (4) where es x earliest start of order x 10

12 rlf(x) average run length of family f(x) Figure 5 illustrates the behavior of the lead time as a function of the utilization rate. For the cycle time a similar pattern can be found. This illustrates the close relation- lead time (days) 6r ~ -lead time family A 5-8-lead time family B -+- lead time family C ~... ~... ~... ~......:...:: 0 L ship between utilization rate (%) the lead time Figure 5. Lead time behavior under various utilization rates (target cycle times and the cycle for all families: 7 days) time as explained above. The striking bend for families Band C at high utilization rates is probably due to the large share in demand of family A. f the potential cumulative tardiness of all families is zero, the family with the longest production run is selected. n most cases, this will be family A. Consequently, at lower levels of utilization, the probabilities that families Band C are selected for production are considerably less than the selection probability of family A. However, at higher levels of utilization, a potential cumulative tardiness of 0 is less probable, and selection of families Band C for production becomes more likely. Consequently, the lead times of these two families improve at high levels of utilization. This anomaly in the lead time curve due to the discontinuity in the decision function is similar to the effects seen in allocation problems with the use of runout times in case of lost sales (see, e.g., [9]). DSCUSSON The simulation experiments have illustrated the basic behavior of the proposed procedure in a controlled and simplified environment. A number of complicating factors have been ignored in the experiments, but will be discussed briefly in this section. First, at the company, aggregate demand was not stationary, but has a seasonal pattern over the year. The company handled this variation by producing to stock a number of products, namely the products of family A This has been captured in the 11

13 simulation experiments by filling the capacity up to the required utilization by generating family A production orders. Second, the machines were not completely reliable as modeled in the simulation experiments. Breakdowns occurred during the week and the production rate was not always as high as planned. n the final implementation the average output was estimated and taken as a basis for the determination of the productive capacity. Since the production output was monitored using the MRP software system, data were available regarding actual production progress. These data were processed once a week to obtain an accurate cumulative net capacity line for the order acceptance procedure. t would have been possible, from an information technology point of view, to update the cumulative net capacity on line. However, this would result in a very hectic behavior of the order acceptance procedure. n addition to the results presented in this paper, it is worth to note that although the procedure based on arriving customer orders has been used as a sequencing procedure, a more or less stable sequence results when the proposed procedure is applied. Stability in cycles leads to stability in sequence, which has been acclaimed of high importance in the literature (see e.g. [6]). The proposed system has a number of disadvantages. The two most important ones will be discussed here. First, the system is conceptually difficult to understand for production planners. n the Netherlands, planners generally have an education which is comparable to two-year technical colleges in the U.S.A Most of the planners in this kind of production systems work with very detailed schedules, with planning horizons of up to one year. This provides the planner with a (fake) sense of security. This security is fake, since the actual value of a schedule which covers more than a week is very limited. As demonstrated above, a correct way of aggregation leads to a lack of necessity to constantly recalculate detailed schedules. The insight into the aggregation procedure is however hard to learn to production planners. n company courses, the Eindhoven University of Technology uses a simulation game to convey the ideas and opportunities of correct aggregation to the practitioners. Second, the system requires the availability and processing of a lot of data. Proper use of the system requires the support of an information system. While developing the current procedures, the company was implementing an MRP software system. The results of the simulation experiments convinced the management of the functionality of the proposed procedure. Consequently, it has been implemented in the MRP software system. SUMMARY n this paper, a procedure for hierarchical control of cycle times has been presented. The procedure has been designed for semi continuous process industries in a make-to-order environment. The purpose of the procedure is to keep control over the 12

14 share of capacity spent of setup, while allowing the system constant and reliable delivery times. Since the system has been designed for a make-to-order environment, it allows for flexibility on the short term. The system has been incorporated in a standard MRP software package and implemented at a plant of a major European chemical company. The main advantages of the hierarchical concept are twofold. Fist, the production system has become controllable. Lead times which are promised to the customer can be met. n the company where the concept has been implemented this has led, additionally, to a lead time reduction of about 50% in the press department, merely because the available flexibility (mix flexibility) has been used optimally. This is the second big advantage of the system: an intuitively inflexible system with high setup times still provides flexibility while cyclical schedules are used. ACKNOWLEDGEMENT The research of Dr. Fransoo has been made possible by a fellowship of ihe Royal Netherlands Academy of Arts and Sciences (KNA W). REFERENCES 1. Fransoo, J.c., and Rutten, W.G.M.M., "A Typology for Production Control Situations in Process ndustries", nternational Journal of Operations & Production Management, Vol. 14, No. 12, to appear in December Elmaghraby, S.E., ''The Economic Lot scheduling Problem: Review and Extensions", Management Science, Vol. 24, No.6, Vergin, R.C., and Lee, T.N., "Scheduling Rules for the Multiple Product Single Machine System with Stochastic Demand", NFOR, Vol. 16, No.1, Leachman, R.C., and Gascon, A, "A Heuristic Scheduling Policy for Multi-item Single-machine Production Systems with Time-varying, Stochastic Demands", Management Science", Vol. 34, No.3, Trigeiro, W.W., Thomas, L.J., and McClain, J.O., "Capacitated Lot Sizing with Setup Times", Management Science, Vol. 35, No.3, Campbell, G.M., and Mabert, V.A., "Cyclical Schedules for Capacitated Lot Sizing with Dynamic Demands", Management Science, Vol. 37, No.4, Bertrand, J.W.M., Wortmann, J.c., and Wijngaard, J., Production Control. A Structural and Design Oriented Approach, Elsevier Science Publishers, Amsterdam, Fransoo, J.C., "Demand Management and Production Control in Process ndustries", nternational Journal of Operations & Production Management, Vol. 12, No. 7/8, Van Donselaar, K.H., "Safety Stock Norms in Divergent Systems with Nonidentical Final Products", Research Report TUE/BDK/ORS/88-04, Eindhoven University of Technology,

15 APPENDX The definition of the planner's decision freedom is based on the sequencing freedom which remains after a family production run has been determined. The freedom consists of sequencing the individual customer order such that an even more efficient use of capacity is possible. Therefore the decision freedom's definition is based of the freedom of sequencing an individual order: (Ai) x 0, if llx S x,edd (A2) (A3) where x lfx x,edd rf trs - decision freedom for order x latest finish of order x start of order x in EDD sequence finish of the family production run start of the family production run by: '" f the expected production time is not the same for all orders, equation (Al) should be replaced x max ( lfx - Px - tx EDD ' 0) trf - rs where Px expected processing time of order x ' (Al') 14