WORKLOAD CONTROL AND ORDER RELEASE IN COMBINED MTO-MTS PRODUCTION

Size: px
Start display at page:

Download "WORKLOAD CONTROL AND ORDER RELEASE IN COMBINED MTO-MTS PRODUCTION"

Transcription

1 WORKLOAD CONTROL AND ORDER RELEASE IN COMBINED MTO-MTS PRODUCTION N.O. Fernandes 1, M. Gomes 1, S. Carmo-Silva 2 1 Polytechnic Institute of Castelo Branco, School of Technology Castelo Branco, Av. do Empresário , Portugal nogf@ipcb.pt; miguelgomes18@hotmail.com 2 Department of Production and Systems, University of Minho Braga, Campus de Gualtar , Portugal scarmo@dps.uminho.pt Abstract - Workload Control (WLC) has been developed as a production planning and control system for make-to-order manufacturing. But, in practice, it has become increasingly common for companies to offer both make-to-stock (MTS) and MTO items. This manufacturing strategy is frequent because it allows for production based on customer specifications while keeping short response times. In this paper we study a two-stage hybrid production system, with an intermediate buffer, that makes multiple types of semi-final items, some of which are made to order while others are made to stock. Then semi-final items are subject to a customization process at the second production stage. Production of semi-final items to stock follows a policy of one-for-one replenishment, i.e. are treated as closed orders, whereas MTO items are open orders. This means that items from open orders follow to the second production stage, while items from closed orders are kept on stock until customer demands for customization arrive. Evaluating the performance of WLC release methods in this context is an important step towards improving the alignment between WLC theory andpractice. Simulation results are used to illustrate the general behaviour of the dynamic hybrid two-stage system and to compare the performance of state-of-the-art release methods. Keywords: Workload control; hybrid MTS-MTO; Simulation. 1. Introduction A major challenge in the today s manufacturing industry is how to increase product variety and at the same time decrease delivery times. One of the proposed solutions for competing in this environment is to adopt a combined or hybrid MTS-MTO production strategy. Such hybrid strategy has become increasingly common in manufacturing industries. The pharmaceutical industry, the food industry and aluminium profiles manufacturing industry provides good examples of coping with both types of demand in a hybrid system. Most of the operations management literature categorises manufacturing systems as either make-toorder (MTO) or make-to-stock (MTS). In MTO, the manufacturing systems produces according to customer requests and no finished goods inventory is kept[1]. Performance measures are order focussed, e.g. short delivery time and high delivery reliability. In MTS, the system produces according to a forecast of customer demand, and completed products enter a finished goods inventory, which in turn serves customer demand [1].Performance measures are product focussed, e.g. high fill rate and low work-in-process. Combined MTS- MTO allows for production based on customer requirements with reduced inventory costs and short delivery times. While there is a large body of literature on MTO and MTS production control, hybrid MTO-MTS has been neglected, with notable exceptions, see e.g. [2], [3] and [4]. In the context of MTO, Workload Control (WLC) is advocated asone of themost promising production planning and control systems. WLC aims at firmly controlling orders flow times through the production system by means of input/output control decisions towards meeting the promised delivery dates. This is usually attempted at three major decision levels: order entry, order release and priority dispatching. The main instrument of control within WLC is the order release, which leads to a pre-shop pool of orders.whereas the release decision is responsible for the control of workloads on the shop floor, acceptance and delivery date decisions should control the load and waiting times in the pre-shop pool. This paper is focuses on the control of workloads on the shop floor and, consequently, on the release decision and applies WLC to hybrid MTO-MTS production. To the best of our knowledge there is only one paper concerned with this issue: Eivazy et al. [5]. They developed a production control and scheduling model based on WLC for controlling the combined production of MTO and MTS products in the semiconductor manufacturing. However, state-of-the-art release The Romanian Review Precision Mechanics, Optics & Mechatronics, 13, No

2 methods are not considered in the study and there have been no attempts to compare release methods in twostage production systems. Therefore, the literature does not provide answers to questions regarding which release method to apply in hybrid MTS-MTO two-stage production system. Moreover, research has not investigated whenand where MTO order release should be controlled in this context. In response, this study uses simulation to assess the performance of three state-of-the-art release methods, identified from the literature, in a hybrid MTS-MTO two-stage production system. In addition the questions of where to control order release, i.e. at only one stage or at both stages, and whencontrol should be exercised, i.e. for the two stages simultaneously or for next stage only, are addressed. The remaining of the paper is organized as follows: the second section presents a model for production control in hybrid MTS-MTO production; then, in the third section, we describe the research methodology adopted, including the simulation model, the experimental set-up and the performance measures; the fourth section presents and discuss the results of the simulation study, the last section of the paper will be devoted to the presentation of the main conclusions of the work. 2. Simulation model for hybrid mts-mto two-stage production system We consider a two-stage production system with an intermediate buffer of semi-final products between the two stages. Stage 1 produces both standard items to stock and customized items to order, and stage 2 is only a customization stage, based on customer demand requirements. MTS orders follow a base stock policy, i.e. demand is filled from the intermediate buffer of semi-final products and each item pulled from this buffer triggers a replenishment order to restore the base stock level of product type kto the desired base stock level, see Figure 1. Demands that cannot be met due to insufficient inventory of semi-final products result in a MTO order for stage 1. MTO products are open orders, whereas MTS products are closed orders. We assume that each order requires just one unit of the semi-final product and the customization processing time for these semi-final products is not a function of the product type. Replenishment MTS orders MTO orders Pool Stage 1 Stock buffer Stage 2 Order delivery Figure 1. Two-stage production system for combined MTS-MTO. The system reaction upon the arrival of an MTO order can be summarized as follows: If the semi-final product buffer is not empty and a product of same type is already in stock, the order is filled from the intermediate buffer and enters the poolwhere waits for its release to processing stage 2 (this refers to orders for customization of items on inventory). This demand arrival will also trigger the release of a MTS order of a semi-final product of the same type to processing stage 1 to replenish the consumed semi-final product. If the buffer is empty, then the order enters the pool and waits for its release to processing stage 1. Each processing stage is capacitated, with stage 1 having only one machine and stage 2 having six machines organised in a general flow shop (see[6]). Orders are processes at machines accordingly to its planned starting times (PST). These are determined as follows: (1) where T w is the planned throughput time at machinew, S jv is the set of machines in the remaining routing of j including machinev and d j is the due-date of j. Both types of orders, MTO and MTS, are not released immediately to the shop floor, but wait in a pre-shop pool for a release decision. Orders are only released if the resulting workload at machines (of stage 1 or/and of stage 2, see the next section of the paper) does not exceed some predefined workload norms or limits. In our study, the machines workloads are calculated by the corrected aggregate load approach(see [6]). This means that only part of the processing times of an order being released are included in the workload of the machines. This part results from multiplying the processing time by a fraction given by the ratio between the planned throughput time at the considered machine and the sum of the planned throughput times at all machines in the routing of the order up to the considered machine. The orders in the pool are considered for release according to its planned release time, which is determined in the same way as the planned starting times, i.e. the planned release time of j is equal to the planned starting time of the first operation of j. In the pool MTO orders takes precedence over MTS orders. The model assumptions can be summarized in the following points: 34 The Romanian Review Precision Mechanics, Optics & Mechatronics, 13, No. 43

3 It is assumed that all orders are accepted and materials are available; Orders inter-arrival times follow an exponential distribution; Six types of products are considered, each of which with an equal probability of being assigned to an arriving order; Only one of the six products types (i.e. 16.7%) are produced to the stock of semi-final products. Due dates are set using the TWK rule (TNOW+c.TWK), where TNOW is the current time, c is a parameter and TWK is the total work content of the order.the value of c was set so that about 10% of the orders were late when orders were immediately releasedand the fill rate equals 90%; Machines capacities remain constant over time and are continuously available; The number of operations per (MTO) order is drawn from a discrete uniform distribution with a minimum of 2 and a maximum of 7; Operation processing times follow a truncated 2- Erlang distribution with a mean value of 1 time unit and maximum of 4 time units; Setup times are assumed to be sequenceindependent and included in the operation processing times; When a MTS order finishes processing at stage 1, it is transferred immediately to the intermediate buffer of semi-final products; Distances and transportation times between machines are assumed to be negligible. These assumptions allow the simulation model to be kept simple to avoid interactions that might inhibit the full understanding of the effects of the experimental factors, which improves the practical applicability of the findings. Table 1 summarises the relevant characteristics of the simulation model. Table 1. Characteristics of the simulation model. Shop type Stage 1 (single machine); stage two(general flow shop) Types of products, k Discrete uniformly distributed [1, 6] No. of machines One at stage 1 and six at stage 2 Machines utilization All equal (90%) Inter-arrival times Exponentially distributed, mean= time units Due date TNOW+c.TWK; c= 9.4 time units Routing Length Discrete uniformly distributed [2, 7] operations Operation Processing Times 2-Erlang, truncated at 4; mean=1 time unit Machines capacities Constant over time Setup times Sequence independent, included in the processing time Experimental Design The research question that this model addresses is: which release method to apply in the context of hybrid MTS-MTO two-stage production system? In order to answer this research question an experimental design was set-up. The experimental factors and simulated levels are summarised in Table 2. The release method is tested at three levels (PR periodic release, PPR periodic with intermediate pull release and CR continuous release), the control policy is testes at two levels (policies III and IV), fill rate is tested at three levels (0%, 90% and 99%) and workload norms are tested at 5 levels (4, 6, 8, 10 and infinity). This leads to a full factorial design with 90 (3x5x2x3) combinations of settings. Table 2. Summary of experimental factors and levels. Experimental factor levels Release methods PR PPR CR Workload norms norm ={4, 6, 8, 10, infinity} Control policy III IV Order fill rate 0% 90% 99% WLC release methods may be classified as periodic, continuous and hybrid. The first method, periodic release (PR), is based on periodic observations of workload on the shop floor. At fixed periods of time, the workloads on the machines of the shop floor are computed and the decision to release one or more orders is taken until no further releases would allow the workload to remain below the norms. This method is specified in detail by e.g. [6] and requires setting a release period length. The second method, continuous release (CR), is based on the continuous monitoring of workload on the shop floor. Releases are triggered when the loads of all machines in the routing of an order would stay below their norm after the release of the order. The third method, periodic with intermediate pull release (PPR), is specified in [7] and combines both periodic and continuous release. A push release is carried out at periodic time intervals and a pull release is carried out each time the workload of any machine falls to zero. In this case, the method only considers those orders within the pool that have a first operation at one of the starving machines and the selected order is not subjected to workload norms. There are two main factors that influence the level of control exercised at order release in a two-stage The Romanian Review Precision Mechanics, Optics & Mechatronics, 13, No

4 production system: (1) where control is exercised, i.e. whether release control is exercised at only one stage (i.e. before stage 1 or stage 2) or at both stages (i.e. before stage 1 and 2); and when control is exercised, i.e. whether orders are considered for release for the two stages simultaneously or for only the next stage.to reflect this, four different control policies are identified, as follows: Control policy I: Release control is only exercised at the stage 1. An MTO order enters the stage 2 immediately once its processing at stage 1 is complete. Control policy II: Release control is only exercised at the stage 2. MTO orders are immediately released to stage 1 and after processing at this stage enter the pool to be considered for release to the stage 2. Control policy III: Release control is exercised at both stages.mto orders are considered for release to stages 1 and 2 simultaneously. Control policy IV: Release control is exercised at both stages.an MTO order is considered for release to the stage 1, and after processing at this stage enters the pool to be considered for release to the stage 2. In this simulation study we restrict ourselves to control policies that exercised at two stage, i.e. policies III and IV. It is common practice in WLC simulation studies to define workload norms as an experimental factor.workload norms are tested at 5 levels by tightening the norm level stepwise down from infinity. Order fill rate, defined as the percentage of orders of product type k that are filled from the semi-final products buffer, is tested at three levels: 99%, 90% and 0%. The fill rate of type k products is expected to approach 100% as the type k base-stock level increases. Performance Measures We use two types of criteria to evaluate the system s performance: (1) the ability to provide short delivery times, and (2) the ability to deliver orders on time. To measure performance with regard to the former, the shop floor throughput time and the total throughput time are used. To measure the performance with regard to the latter, the percentage of tardy orders and the standard deviation of lateness are recorded. Varying workload norms will result in different levels of workload on the shop floor, influencing the average shop floor throughput time. Therefore, all performance measures will be shown in relation to the average shop floor throughput time.these are order focussed performance measures typical of the MTO production. Product focused performance measures (typical of the MTS production) are not used. Instead we are focused on comparing release methods and control policies for different fill rates. 3. Simulation Results and Analysis This section discusses the results of the simulation study described in the previous section. During simulation experiments, data were collected under steady state conditions. The average values of 100 independent replications are presented as the result of one experiment. The length of each replication is 10,000 time units including a warm-up period of 3,000 time units. Common random numbers are used as a variance reduction technique. Table 3. Performance results. CP 1 RM 2 norm STT 3 0% fill rate 90% fill rate 99% fill rate TTT L StD P tard STT TTT L StD P tard STT TTT L StD P tard III PR PPR CR Inf Inf Inf Inf The Romanian Review Precision Mechanics, Optics & Mechatronics, 13, No. 43

5 IV PR PPR CR Inf Inf Control policy; 2 Release method; 3 Shop floor throughput time; 4 Total thoughput time; 5 Standard dviation of lateness; 6 Percent tardy; Table 3 summarises the performance results for the two control policies (III and IV) and three release methods (PR, PPR and CR) for three levels of the fill rate (0%, 90% and 99%). An overview of the release methods performance under these control policies is also presented in Figures 2, 3 and 4. In these figuresthe percentage of tardy orders, the total throughput time and the standard deviation of latenessare plotted as a function of the average shop floor throughput time. Every curve represents the performance of a release method for a givenfill rate. A marker on a curve is the result of simulating a release method at a specific workload norm level. Five norm levels have been simulated, including infinity, which means the unrestricted release of orders to processing stages and refers to the right hand mark on each curve. Frist, it can be observed that increasing the fill rate allows for a lower percentage of tardy orders, independently of the release method or the control policy applied. As could be expected higher fill rates result in lower throughput times and thus in lower percentages of tardy orders. This happens because orders have a high probability of being filled from the intermediate buffer. Second, control policy IV allows for lower total throughput times and lower percentage of tardy orders than control policy III. However, control policy III performs better in terms of the shop floor throughput times. Controlling the release of orders to stages 1 and 2 in two different steps (policy IV) does seem to be able to red uce the shop floor throughput time. In fact, as can be observed from figure 2(b), when the workload norms start being restricted the shop floor throughput time starts to be reduced. However for lower levels of workload norms, as restriction continues, the shop floor throughput starts to increase again. This happens because orders that are released to the stage 1 cannot proceed immediately to stage 2 after processing, i.e. they must wait in the pool for a release decision. For low workload norms, new orders that enter the pool find fewer opportunities to be released to stage 2, being, therefore, delayed between stages. However, in spite of these orders being delayed between the two stages a lower average total throughput time is achieved by control policy IV, since orders have higher opportunities to be release to stage 1 (because they need to fit only the workload norms of one machine). Finally, independently of the control policy or the fill rate, continuous release clearly outperforms the other two methods for the percentage of tardy orders. The second best is the periodic release method. The periodic with intermediate pull release shows the worst performance on this performance measure. This is particularly interesting since this method is referred in the WLC literature as best performing (see, e.g. [7]). 16 (a) 14 P 16 (b) 14 P Tardy Orders (%) Tardy Orders (%) Figure 2. Percentage of tardy orders under control policies (a) III and (b) IV. The Romanian Review Precision Mechanics, Optics & Mechatronics, 13, No

6 (a) (b) P P Figure 3. Total throughput time under control policies (a) III and (b) IV. (a) (b) P P Figure 4. Standard deviation of the lateness under control policies (a) III and (b) IV. 4. Conclusions and Managerial Implications Order release is the main control element within Workload Control (WLC).In this regard, we evaluate the performance of three state-of-the-art WLC release methods in the context of hybrid MTS-MTO production. It has focused on a two-stage production system with an intermediate buffer of semi-final items, some of which are made to order while others are made to stock. Standard items that are made to stock at a first stage arethencustomized at the second processing stage, according to the customer requests. We selected this environment because we observed this reality in several settings in industry. Research questions regarding where to control order release, i.e. at only one stage or at both stages, and when control should be exercised, i.e. for the two stages at once or for next stage only, have been addressed.the following conclusions and managerial implications may be derived from the simulation results: Workload control can be successful applied to control the production of both standard items to stock and customized items to orderif combined with a base-stock policy for the MTS products. The decision to release the orders to both stages should preferably by carried out in two phases: first to stage 1 and then, after processing, to stage two. The release of orders simultaneously to both stages even if it allows for short shop floor throughput times it results in a higher percentage of tardy orders. The release method adopted is critical to the system performance. Continuous release outperforms periodic and periodic with intermediate pull releases in the hybrid MTS- MTO two-stage production system studied. These last two managerial implications however need to be further tested with additional system configurations to be possible reach an acceptable level of generalization of such implications. This will be subject to future work. Moreover, in this simulation study we restrict ourselves to a situation where release control is exercised at both stages1 and 2. Future research work should investigate the performance of the WLC release mechanisms where control is exercised at only one of the stages. 5. References [1] Wein, L., Dynamic Scheduling of a Multiclass Make to Stock Queue. Operations Research. (1999); 40(4) [2] Soman, C., van Donk, P. Gaalman, G., Combined make-to-order and make-to-stock in a food production system. International Journal of Production Economics. (04); 90 (2) [3] Soman, C., van Donk, P. Gaalman, G., Capacitated planning and scheduling for combined make-to-order and make-to-stock production in the food industry: An illustrative case study. International Journal of Production Economics. (07); 108 (1 2) The Romanian Review Precision Mechanics, Optics & Mechatronics, 13, No. 43

7 [4] Zhang. Z.W., Kim, I., Springer, M., Cai G., Yu, Y., Dynamic pooling of make-to-stock and make-to-order operations. International Journal of Production Economics. (13); In Press. [5] Eivazy, H. Rabbani, M. and Ebadian M., A developed production control and scheduling model in the semiconductor manufacturing systems with hybrid make-to-stock/make-to-order products. International Journal of Advanced Manufacturing Technology. (09); 45: [6] Oosterman, B.J., Land, M.J. and Gaalman, G.J.C., The influence of shop characteristics on workload control. International Journal of Production Economics. (00); 68 (1), [7] Thürer, M., Stevenson, M., Silva, C, Land, M.J., and Fredendall, L.D., Workload control (WLC) and order release: a lean solution for make-to-order companies. Production and Operations Management. (12); 21 (5), The Romanian Review Precision Mechanics, Optics & Mechatronics, 13, No