Using simplified drum-buffer-rope for re-entrant flow shop scheduling in a random environment

African Journal of Business Management Vol. 5(26), pp. 10796-10810, 28 October, 2011 Available online at http://www.academicjournals.org/ajbm DOI: 10.5897/AJBM11.723 ISSN 1993-8233 2011 Academic Journals Full Length Research Paper Using simplified drum-buffer-rope for re-entrant flow shop scheduling in a random environment Yung-Chia Chang* and Wen-Tso Huang Department of Industrial Engineering and Management, National Chiao Tung University, Hsinchu 300, Taiwan, R.O.C. Accepted 17 May, 2011 The traditional simplified drum buffer rope () does not consider the application of re-entrant flow shop (RFS) in a random environment which might involve variable processing times and machine breakdowns. This paper proposed a weighted layer production buffer and weighted production buffer to monitor the status of the buffer deviation when applying in a RFS with non-deterministic parameters. The buffer status deviation used the overall urgency and actual urgency to estimate the influence of overall accumulated machine downtime and the influence of actual machine downtime of capacity-constrained-resource (C) at different layers. A dispatching rule, called _D Reentry, was applied to decide the priority of all work orders by the buffer status deviation of each C machine with the consideration of the re-entry feature. A simulated RFS was designed and four different methods (including ours) were applied in order to demonstrate the effectiveness of our approach. The experimental results show that when compared to six performance indexes related to the due date, our approach has better performance than the other three methods when the product mix with a large proportion of multi-reentrant orders and when the C utilization increases from 60 to 90%. Key words: Drum-buffer-rope, simplified drum-buffer-rope, re-entrant flow shop, scheduling, random environments. INTRODUCTION The basic characteristic of a re-entrant flow shop (RFS) is that all jobs have the same production route over the machines of the shop and require the same amount of processing time at each machine. In such a shop, each job may return to the same machine one or more times before completion (Graves et al., 1983). For example, in semiconductor manufacturing, a wafer needs to visit some certain machines several times for processing (Vargas-Villamil and Rivera, 2001). Furthermore, a production system is typically operating under a random environment. Sources of randomness include, for example, machine breakdowns, unexpected releases of high-priority jobs, the processing times that are not precisely known in advance, and the like (Pinedo, 2002). Nevertheless, there was not a complete and clear method for delineating a production system with a RFS in *Corresponding author. E-mail: smythambit.iem96g@g2.nctu.edu.tw. Tel: +886-3-5731815. a random environment. For example, many papers have studied the RFS scheduling problems. Hwang and Sun (1998) applied a dynamic programming to solve a twomachine flow shop problem with re-entrant work flows and sequence dependent setup times. Their objective was to minimize the makespan. Park et al. (2000) proposed an approximate method based on the mean value analysis for estimating the average performance of RFS with single job machines and batch machines. Chen et al. (2007) introduced a hybrid tabu search technique to solve the scheduling problem of minimizing the makespan in a RFS. Chen et al. (2008) proposed a hybrid genetic algorithm to minimize the makespan for the RFS scheduling problems. Another line of work applied simplified drum buffer rope (), a standard production planning method using the theory of constraints (TOC), in a non-rfs environment. Schragenheim (2006) described the make-to-order aspects of implementing the due date setting method and the release policy of in a non-rfs environment. Lee et al. (2010) extended the work of Schragenheim

Chang and Huang 10797 Figure 1. The buffer management of an system (Schragenheim, 2006). (2006) by dealing with whether the capacity-constrained resource (C) is located in the middle of the routing or not. Schragenheim and Dettmer (2001) stated that the advantage of is that it can be easily applied in most manufacturing environments when other methods seem to be complicated in practice. There is literature discusses the implementation of in a RFS environment. Chang and Huang (2011) proposed a set of shop-floor control mechanisms (named as Reentry) in a RFS and showed that it performs better when the product portfolio has a large proportion of multi-reentrant orders or when the bottleneck machine of the system is full loaded. However, Chang and Huang (2011) did not implement in a random environment. This paper extended the work done by Chang and Huang (2011) and proposed a set of control mechanisms and dispatch rules, named as _D Reentry, for a random RFS. In this random RFS, the job process times are random and the machines are subject to random breakdowns. To show the performance of _D Reentry, we applied _D Reentry to dispatch jobs with random process times in the simulated RFS based on a case company described in Chang and Huang (2011) but with random machine breakdowns. In additional to _D Reentry, three other types of dispatching rules, including the original, the approach used by the case company, and the critical ratio approach, were also adopted. The simulated results were compared over the six due-date related performance indexes. THE CORE OF THE Schragenheim and Dettmer (2001) stated that assumes that market demand (or the drum) is always a constraint and that the internal resources often have excess capacity. Therefore, no detailed schedule is created for the C under, but only monitors the C with total planned load (PL) to ensure that there is enough capacity to meet all due dates. The PL intends to convert each work order into the number of capacity hours of C that is needed in the future. For example, suppose work order 1 and work order 2 both require one hour of work on the C, the PL is then simply 2 h. Since the market is always a constraint, the buffer in is simplified into one single production buffer (PB) that represents the time of a work order, from the release date to the due date, and is equal to the cycle time. estimates the PB of each product using the touch time multiplied by a constant based on practical experience. The touch time represents the shortest processing time of a work order, including the setup time. The total touch time of a work order can be calculated by the summation of the shortest processing time on each machine where a work order passes through to obtain the estimated PB. Since each work order has an accumulated point in the C, the release point (or the rope) is obtained by moving each PL point backward by subtracting half of a PB. On the other hand, safe delivery is obtained by adding half of a PB. Usually, the promised due date can be estimated as additional buffer plus safe delivery. Schragenheim (2006) and Schragenheim and Burkhard (2007) introduced buffer management (BM), which is a mechanism that adopts the buffer status (BS) to decide the urgency of a work order. Figure 1 presents an example of buffer management. PB is divided into three equal areas (green, yellow and red). The penetration of a specific work order is obtained by calculating the time difference between its release date and when it reaches the C. In other words, penetration means the consumption of PB on a specific work order in front of the C, which can be simplified to the ratio of the accumulated flow time on the shop floor to the PB, that is, the BS, by equation (1). The actual sequence should be determined based on the overall urgency of each work order accumulating in front of the C by prioritizing their BS from high to low to obtain a complete real-time dispatching schedule for the C. Buffer status (BS) = Flow time Production buffer(pb) An illustrative example of the dispatching rule: Lee et al. (2010) further defined BS as the dispatching rule of C by Equation (2). (1)

10798 Afr. J. Bus. Manage. Table 1. An illustrative example of buffer status. Order no. Touch time PBa Releasing date Due date Remain days to due date BS Priority 1 9 18 1 19 10 (18-10)/18=44% 2 12 24 1 25 10 (24-10)/24=58% 3 9 18 1 19 9 (18-9)/18=50% 4 12 24 1 25 6 (24-6)/24=75% * a. PB is two times of touch time. (Production Buffer - Remain Days To Due Date) Buffer status (BS) = Production Buffer (2) The numerator in equation (2) represents the penetration. A high penetration rate implies that the increasing consumption of PB is leading to the possibility of delayed delivery. In other words, an order with a high BS value results in a higher priority to be processed by the C. Table 1 provides an illustrative example of four orders, in which the fourth work order has a high BS value and a higher priority of the C.Chang and Huang (2011) provided an example to illustrate the dispatching results of using the and Reentry, as shown in Table 2. It was assumed that the PB is twice as much as the touch time for each product, and that all operations to be performed by a machine require the same processing time. The resulting outcomes provided in Table 2 show that under three situations, the performance of traditional would be limited when it is applied to a re-entrant manufacturing environment. By contrast, Reentry, which adopts the deviation rate BS as the dispatching rule in a RFS, will choose work orders that might result in possible delay. In fact, neither the dispatching rule BS introduced by Lee et al. (2010) nor the dispatching rule BS illustrated by Chang and Huang (2011) considers machine failure in a RFS. _DReentry MODEL The Reentry introduced by Chang and Huang (2011) has better performance in a RFS when the product portfolio has a large proportion of multi-reentrant orders, or when the C machine of the system is close to full loaded. However, Reentry does not consider variable processing times and random machine failures even though in practice these two production variances are a common occurrence (De et al., 1993; Arzi and Herbon, 2000; Mohebbi, 2008). In order to improve the applicability of for RFS in a random environment, this paper proposed the _DReentry method by extending the layered control of the reentrant layers proposed by Chang and Huang (2011). The layered control diagram of the re-entrant layers in Reentry is shown in Figure 2. In _D Reentry, the PB of each product in a random RFS is divided into the layer production buffer (LPB) based on the number of Cs. The estimated value of each LPB includes the influence of variable processing times and machine breakdowns. The difference between Reentry and _DReentry is that Reentry only calculates the deviations of the overall buffer status (BS) and the layered buffer status (LBS), while _DReentry discusses every work order at different C layers as well as the influence of overall accumulated machine downtime on the BS and the influence of actual machine downtime at different C layers on the LBS. The deviation between BS and LBS, denoted as BS, is utilized as the dispatching rule for the C. The actual sequence on the shop floor should be determined based on the BS of each work order accumulating in front of the C. A higher value of BS implies a higher priority. Ranking BS from high to low can obtain a complete real-time dispatching schedule for the C layers. The notations used in this paper are given as: i : Index of product; where i = 1, 2,, m. j : Index of order; where j = 1, 2,, n i. k: Index of C machine; where k = 1, 2,, o. l : Index of re-entrant layer number; where l = 1, 2,, p i. M: Number of orders. n i: Number of orders of product I; where i = 1, 2,, m. o : Number of C machine. P i: Number of re-entrant layers of product I; where i = 1, 2,, m. BS ij : The buffer status ratio for order j of product i; where i = 1, 2,, m, j = 1, 2,, n i. PB ij : The production buffer time for order j of product i from the release date to the due date; where i = 1, 2,, m, j = 1, 2,, n i. LPB ijl : Layer Production Buffer, the production buffer time for order j of product i at the l th layer; where i = 1, 2,, m, j = 1, 2,, n i, l = 1, 2,, p i. LPTi jl: Layer Processing Time, the actual processing time for order j of product i at the l th layer; where i = 1, 2,, m, j = 1, 2,, n i, l = 1, 2,, pi. LFT ijl: Layer Flow Time, the flow time for order j of product i at the l th layer; where i = 1, 2,, m, j = 1, 2,, ni, l = 1, 2,, pi. LBS ijl: Layer Buffer Status, the buffer status ratio for order j of product i at the l th layer; where i = 1, 2,, m, j = 1, 2,, n i, l = 1, 2,, pi. LDT ijl: Layer Down Time, the down time for order j of product i at the lth layer; where i = 1, 2,, m, j = 1, 2,, n i, l = 1, 2,, pi. TFT ij: Total Flow Time, the total flow time for order j of product i on the shop floor; where i = 1, 2,, m, j = 1, 2,, n i. MTBF k: The mean time between failures of C machine k; where k = 1, 2,, o. MTTR k: The mean time to repair of C machine k; where k = 1, 2,, o. MULT: Constant buffer size multiplier. Q ijlt: =1, if order j of product i is queued in front of the C at the lth layer at time t; =0, otherwise; where i = 1, 2,, m, j = 1, 2,, n i, l = 1, 2,, p i. U ij: Indicate whether order j of product i is completed by its due date. That is, if DD ij> CD ij, then U ij = 1; otherwise, U ij = 0; where i = 1, 2,, m, j = 1, 2,, n i. CD ij : The completed date for order j of product i, for i = 1, 2,, m, j = 1, 2,, n i.

Chang and Huang 10799 Table 2. An illustrative example of and Reentry dispatching results (Chang and Huang, 2011). Situation Order no. PBa C Layer Total C layer Total flow time Layer flow time LPBb Reentry BSc (%) Priority LBSd (%) BSe (%) Priority a 1 18 2 2 16 5 8 89 * 63 26 2 24 2 3 16 3 8 67 38 29 * b 3 18 1 2 12 1 4 67 * 25 42 4 24 2 3 16 1 8 67 * 13 54 * c 5 24 3 3 16 1 6 67 * 17 50 6 24 2 3 16 1 8 67 * 13 54 * a. PB is assumed to be twice as much the touch times for each product; b. LPB equals to the weight of different layers multiplied by the production buffer; c. BS = Total Flow Time/ PB; d. LBS = Layer Flow Time/ LPB; e. BS =BS(%)- LBS(%). Figure 2. Layered control diagram of the re-entrant layers in Reentry (Chang and Huang, 2011). DD ij : The due date for order j of product I; where i = 1, 2,, m, j = 1, 2,, n i. OV ij : The value of the orders for order j of product I; where i = 1, 2,, m, j = 1, 2,, n i. RD ij : The release date for order j of product I; where i = 1, 2,, m, j = 1, 2,, n i. WV ij: The value of the WIP for order j of product I; where i = 1, 2,, m, j = 1, 2,, n i. The core concepts of _D Reentry are stated thus: first, the layer touch time used in Reentry was replaced by the actual layer processing time (LPT) at each C layer. Therefore, the LPB of each work order is calculated by multiplying LPT by the weight of the C machine downtime occurring at different layers and a constant multiplier (MULT), as shown in Equation (3): MTTRk MTTRk + MTBFk LPBijl = LPTijl (1 + ) MULT,

10800 Afr. J. Bus. Manage. where i = 1, 2,, m, j = 1, 2,, ni, k = 1, 2,, o, l = 1, 2,, pi.(3) The PB of each work order is then represented by the sum of LPB, as shown in Equation (4): p i LPB l = 1 PBij = ijl, where i = 1, 2,, m, j = 1, 2,, ni. (4) BS is used to estimate the overall urgency of the due date for a specific work order. It is the ratio of the accumulated LFT at each C layer on the shop floor to the PB, calculated by Equation (5): BSij = TFT PB ij ij * 100%= p i l = 1 PB LFT ij ijl * 100%, where i = 1, 2,, m, j = 1, 2,, ni. (5) Furthermore, LBS is used to estimate the actual urgency of a specific work order at each C layer. LBS is calculated by equation (6). It is the ratio of the LFT at each C layer on the shop floor to the corresponding LPB. LBSijl = LFT ijl LPB ijl 100% where i = 1, 2,, m, j = 1, 2,, ni, l = 1, 2,, pi. (6) When the C needs to determine the priority of work orders accumulating in front of it, BS is calculated by using Equation (7). BS is essentially the difference between the BS and LBS after considering the influence of the accumulated machine downtime on each work order and the influence of C machine downtime of each work order at a specific layer. BSijl = BSij * (1- p i l = 1 PB LDT ij ijl LDT ijl LPB ijl ) - LBSijl * (1- ), where i = 1, 2,, m, j = 1, 2,, ni, l = 1, 2,, pi. (7) A higher BS implies a higher priority. Ranking the BS from high to low would then create a complete real-time dispatching schedule for C. Non-C machines are dispatched following the First-In- First-Out (FIFO) rule. Procedures of _DReentry This summarizes the _DReentry procedures into 7 steps as follows: Step 1: Calculate the LPB of each work order by equation (3). Step 2: Calculate the PB of each work order by equation (4). Step 3: Calculate the overall urgency BS of each work order by equation (5). Step 4: Calculate the actual urgency LBS among the layers for each work order by equation (6). Step 5: When the C needs to determine the priority of work orders accumulating in front of it, calculate the BS by using equation (7). Step 6: Ranking the BS from high to low can obtain a complete real-time dispatching schedule for C. Step 7: Non-C machines are dispatched following the FIFO rule. Example: We use an example to illustrate the procedure of _D Reentry. The inputs and the resulting outcomes are listed in Table 3. In Table 3, columns 2 to 10 are given as the settings of this example and columns 11 to 15 are calculated by using equations (3) to (7). In Table 3, the accumulated down time (%) shown in column 7 is the ratio between ΣLDT and PB, which represents the overall accumulated C machine downtime of each work order. The LDT (%) shown in column 8 is the ratio between LDT and LPB, which represents the downtime of a specific C machine of each work order at a specific layer. The BS (%) shown in column 11 is obtained from equation (5), which represents the dispatching rule of. The LBS (%) shown in column 13 is obtained by using equation (6), which represents the ratio of the influence of specific C machine downtime of each work order at a specific layer. The BS (%) shown in column 14 is obtained from equation (7), which denotes the dispatching rule of _D Reentry. In Table 3, orders 1 and 2 in situation (a) involve different products but have the same urgency. As the influence of machine breakdown is not taken into consideration (that is, using ), the BS dispatching priority in recommends to choose either one arbitrarily. However, _D Reentry recommends choosing the work order 2, which generates a risk of delay after measuring the deviation rate BS of the influence of the overall accumulated C machine downtime on the planned value (BS) and the influence of actual C machine downtime at different C layers on the actual value (LBS). On the other hand, orders 3 and 4 in situation (b) are of the same product and have the same urgency, but are at different C layers. Per the method, since the BS values are the same for both orders, either one can be chosen arbitrarily. However, it appears that work order 4 in layer 2 has a higher risk of delay than work order 3 in layer 3. When choosing a specific work order by using BS, _DReentry will choose work order 4, the one that has higher risk of delay. From this example, we can see that the influence of C machine downtime for work orders that generate a risk of delay could be distinguished when _DReentry is applied to two or more work orders with the same urgency rate in a random environment. Computational experiment In order to compare the performance of _D Reentry in a random RFS environment, this paper adopts a simulation plant designed by Chang and Huang (2011) based on the manufacturing process in Company K. In 2007, Company K implemented the approach in job dispatching and scheduling (Chang and Huang, 2011). Company K is interested in knowning if can be applied to a more realistic situation in which processing times are varied and machines randomly break down. To solve this problem, we propose a set of control mechanisms and dispatching rules in applying the for RFS in a simulated random environment. Profile of the simulation environment To illustrate the proposed method, we adopt the data stated in Chang and Huang (2011). The simulated company, says company K, produces three types of products: A, B, and C respectively. Each of them has 0, 1, and 2 C reentries processes respectively. Table 4 lists the product routing in the simulated plant. The simulated plant has six types of machines: M1, 2, 3, 4, 5, and 6. In addition, the numbers of M1, 2, 3, 4, 5, and 6 are 3, 2, 3, 2, 2, and 2 respectively. In this paper, the C can be calculated by the approached developed by Russell and Taylor (2005). They indicate that the C is equal to sum over total processing time of each product for each machine. Also, the average load for each machine would be equal to the sum of total processing time divided by the number of machines. Table 5 shows that the machine, M2, with the

Chang and Huang 10801 Table 3. An illustrative example of _D Reentry dispatching results in a random environment. (1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12) (13) (14) (15) Situation Order no. Total C layer C layer no. TFT LFT Accumulated down time (%) LDT (%) LPB PB _DReentry BS (%) Priority LBS (%) BS (%) a 1 2 2 15 4 20 10 8 18 83 * 50 22 2 3 2 20 4 20 30 8 24 83 * 50 32 * b 3 3 3 20 4 20 10 6 24 83 * 67 7 4 3 2 20 4 20 30 8 24 83 * 50 32 * Priority Table 4. Product routing in the simulated plant (Chang and Huang, 2011). Product Step 1 2 3 4 5 6 7 8 9 10 11 12 A M1 M2 M3 M1 M4 M5 M6 B M1 M2 M3 M1 M4 M2 M3 M5 M6 C M1 M2 M3 M1 M4 M2 M3 M5 M2 M3 M5 M6 Table 5. Calculation results of the average load for machines in the simulated plant (Chang and Huang, 2011). Product Machine M1 M2a M3 M4 M5 M6 A 2 1 1 1 1 1 B 2 2 2 1 1 1 C 2 3 3 1 2 1 Sum 6 6 6 3 4 3 Average load 2 3 2 1.5 2 1.5 ac machine which is identified by summing the processing times of all operations to be performed at a machine. highest average load, can be the C in the simulated plant. Therefore, Product A, B, and C require 0, 1, and 2 re-entrant layers, respectively. We assume that PB is three times of touch time. In addition, we assume that there is only one type of WIP provided for all three types of products. The values of the WIP of three different products are 1, 2, and 3 units, respectively. Also, the values of the orders are 2, 4, and 6 units. We fix that one work order represents one lot, processing lot and transfer lot.

10802 Afr. J. Bus. Manage. Table 6. Machine operation data in the simulated plant. Machine Quantity λa MTBF MTTR System availability M1 3 0.025 40 b 2 0.95 c M2 2 0.025 40 2 0.95 M3 3 0.025 40 2 0.95 M4 2 0.025 40 2 0.95 M5 2 0.025 40 2 0.95 M6 2 0.025 40 2 0.95 a. where λ is the failure rate of the machine; b. MTBF = 1/λ = 1/0.025 = 40; c. System availability = MTBF/(MTBF+MTTR)=40/(40+2) = 0.95. This paper considers that each variable processing time of the non C machines (M1, 3, 4, 5, and 6) follow a uniform distribution with a range from 3 to 12 h. Also, the C machines (M2) follows a uniform distribution ranging from 10 to 20 h. Furthermore, we assume that the time required to repair a machine follows an exponential distribution with a mean of 2 h. Once the repair of a machine is completed, the time until the next breakdown of the machine has an exponential distribution with a mean of 40 h. The data are derived from the actual manufacturing environment of the company K as shown in Table 6. This paper intends to verify the following 4 approaches involving the dispatching and release rules. That is, the _D Reentry, the, the case-study method (we refer the method as the method K), and the critical ratio (). This study adopts six performance indexes related to the due date. The throughput dollar day (TDD) which is the summation of the value of the orders, multiplied by the number of days by which their delivery is late (Ho and Li, 2004), as shown in Equation (8): TDD = * M m ni 1 { ( )} OV * Max 0, CD - DD n ij ij ij i=1 j=1 i Equation 8 indicates that the manager can focus on the increase of the throughput by the TDD. The inventory dollar day (IDD) is the summation of the dollar value of the WIP multiplied by the time since the WIP entered the plant (Ho and Li, 2004), as shown in Equation (9): IDD = * M m ni 1 WV ij * ( CD ij - RDij ) n i=1 j=1 i Equation 9 is used to guide managers to focus on reducing the plant s actual WIP and production cycle time. The due date slack time (DDST) is represented by the number of days that the order is completed ahead of schedule, which depicts the ability to protect the due date, as shown in equation (10): DDST = * M m ni 1 ( DD - CD ) ij n i=1 j=1 i ij (10) The due date performance (DDP) is tracked to guide managers to focus on ensuring there is no delay for each order, as shown in Equation (11): (8) (9) 1 DDP = * M m n i Uij ( ) n i=1 j=1 i (11) The average queue length in front of the C (Q_C) is the total average of queued work orders in front of the C, as shown in Equation (12): Q_C t m ni pi 1 Qijlt = * ( ) M n i= 1 j = 1 l = 1 i The flow time is calculated by Equation (13): n i m { ij ij 1} Flow Time = (CD - RD ) + i=1 j=1 (12) (13) Test problem design: This study considered variable processing times and machine breakdown situations in applying the for RFS in a simulated random environment. Each variable processing time for both non-c and C machine came from a uniform distribution with intervals ranging from 3 to 12 h and from 10 to 20 h, respectively. The machine breakdown for every machine in the simulation plant is illustrated in Table 6. The test problem in the simulation plant was considered using two factors in practice: i) Different utilization ratios of the C: The basic assumptions behind are that the market is always a constraint and that internal resources often have excess capacity. Hence, this paper considered four non-full load C utilization ratios: 60%, 70%, 80%, and 90%, respectively. ii) Different layer mixes of the C: Due to the complexity of the re-entrant environment being exacerbated by re-entrant processing (Yan et al., 2000), the performance of _D Reentry might be limited when it is applied to a random environment. This paper took into account three combinational ratios of three layer mixes, which were simulated as M-M-M (1 : 1 : 1), L-H-L (1 : 6 : 3), and L-L-H (3 : 1 : 6). M represented Medium, L represented Low and H represented High. For example, L-L-H (3 : 1 : 6) represented a ration of A : B : C = 3 : 1 : 6. Table 7 shows the twelve test scenarios of the two factors. Description of, critical ratio method and method K: To show _D Reentry for RFS in a random environment, this paper

Chang and Huang 10803 Table 7. The design of the test problems. Layer mix Proportion Scenario Utilization (%) A a B C A B C 1 M b M M 1 1 1 2 60 L H L 1 6 3 3 L L H 3 1 6 4 M M M 1 1 1 5 70 L H L 1 6 3 6 L L H 3 1 6 7 M M M 1 1 1 8 80 L H L 1 6 3 9 L L H 3 1 6 10 M M M 1 1 1 11 90 L H L 1 6 3 12 L L H 3 1 6 a. The simulation plant produces three types of products: A, B, and C. b. The triplet can be interpreted as follows: M represented Medium, L represented Low, and H represented High. selected,, and method K to evaluate its efficiency, which was proposed by Chang and Huang (2011). The detailed steps are described thus: Procedures of Step 1. Calculate the touch time of each product and estimate the PB: Calculate the touch time of each product at the initial stage, then estimate PB of each product by the touch time multiplying the constant times on practical experience. Step 2. Calculate the PL of C: When receiving a new order, the managers will convert each work order into the number of capacity hours of C that is needed in the future. Combining the resulting outcome with the original C, we can derive the planned load. Ensure that it will not exceed the due date after adding the half of a PB. Step 3. Decide the release schedule: The work order is released into the shop floor according to the flow time it will work on C minus the half of a PB to get the release date, and prioritize the release date from now to future to obtain the complete release schedule. Step 4. Decide the dispatching schedule: Calculate the overall urgency BSij of each work order accumulates in front of the C, and rank the BSij from high to low to obtain the complete real-time dispatching schedule for C. Non-C machines are dispatched under the FIFO rule. Procedures of Step 1. Decide the C dispatching schedule: Calculate the respective of each work order waiting before C according to equation (14) (Russell and Taylor, 2005), and rank the ratio from low to high to obtain the real-time dispatching schedule of C. Time Remaining = Work Remaining (Due Date - Today' Date ) = 100% Remaining Processing Time (14) Step 2. Decide the non-c dispatching schedule: Non-C dispatching schedule based on the principle of FIFO rule. Procedures of method K Step 1. Decide the release schedule: The release point is obtained by moving backward each due date was assigned by the client subtracting one average quoted lead time (QLT) of the market, and rank them from the present time to the future to obtain the integrated release schedule. Step 2. Decide the dispatching schedule: Calculate the Modified Critical Ratio (M) of each work order waiting before C according to equation (15). The difference between and M is the remaining processing time. That is, the part of denominator is the major difference, and rank the M ratio from low to high to obtain the complete C real-time dispatching schedule. (Due Date - Today' Date) M = 100% (3 * Remaining Touch Time) (15) Step 3. Decide the non-c dispatching schedule: Decide the dispatching schedule of non-c under the FIFO principle. EXPERIMENTAL RESULTS The simulation experiments began with an empty plant. Machines were at either up or in idle states at the beginning. Each machine only processed one work order each time, while operating twenty-four hours per day, seven days per week. If a machine breakdown occurred during processing, the work order would start again after the machine's maintenance was completed. To provide a consistent basis for comparison between the four dispatching rules, the due date and release date were kept constant as three of touch times for each product. The four dispatching rules were respectively executed ten

10804 Afr. J. Bus. Manage. 2800 TDD($) 2600 2400 -D-reentry 2200 TDD ($) 2000 1800 1600 1400 1200 1000 800 60 70 80 90 Utilization(%) Utilization (%) Figure 3. TDD under different utilizations. times under twelve scenarios. The resulting simulation outcomes were as follows. Analysis of performance under different C utilization ratios Figures 3 and 4 show the TDD and IDD of the four methods under different C utilization ratios. When the PB was three times the touch time and the layer mix was kept constant, the TDD and IDD obtained from _D Reentry was the smallest under the four C utilization ratios. Moreover, it had the largest gap with the other methods if the C utilization ratio was 90%. When the C utilization ratios increased from 60% to 90%, the gap between _D Reentry and also increased. This study inferred that it was driven by the deviation rate BSijl of the urgency of the influence on machine breakdown added into _D Reentry. Figures 5, 6 and 7 show the DDST, the Q_C and the flow time obtained from the four methods, respectively, when the PB was three times the touch time and the layer mix was kept constant under different C utilization ratios. It can be seen from Figure 5 that the DDST obtained from _D Reentry was the largest under all four C utilization ratios, which represented a better protection of the due date under _D Reentry. Figure 6 shows the Q_C obtained from _D Reentry. The Q_C was the smallest under the different C utilization ratios and had the biggest gap in comparison with, when the C utilization ratio increased from 60% to 90%. This resulted from the deviation rate BSijl of the urgency of the influence on machine breakdowns for each re-entrant layer in _D Reentry. Figure 7 shows that the flow time obtained from _D Reentry was the smallest in comparison with other methods under the four C utilization ratios. The simulation results showed that the DDP of the four methods was 0%. Therefore, no further discussion was required. Analysis of performance under different C layer mixes Figures 8 to 12 show the changes of the C resulting from the five indexes. A change of the C showed that the layer mix had changed, but the rest of the factors were the same as before. In Figures 8, 9, 11 and 12, the TDD, the IDD, the Q_C and the flow time obtained from _D Reentry were the smallest under the three layer mixes. _D Reentry had the biggest gap in comparison with the other methods when work orders with two C reentries were in the majority, which indicated a better performance if the product portfolio

Chang and Huang 10805 2100 2000 1900 IDD($) -D-reentry 1800 1700 IDD ($) 1600 1500 1400 1300 1200 1100 60 70 80 90 Utilization(%) Figure 4. IDD under different utilizations. Utilization (%) -150 Slack time(hours) -D-reentry -200-250 -300 Slack time (h) -350-400 -450-500 -550-600 60 70 80 90 Utilization(%) Figure 5. DDST under different utilizations. Utilization (%) had more multi-reentrant work orders. Figure 10 demonstrates that _ D Reentry had the largest gap with the DDST in other methods when work orders (L-L-H) with two C reentries were in the

10806 Afr. J. Bus. Manage. 70 Queue length 65 60 55 Queue length 50 45 40 35 30 25 20 -D-reentry 60 70 80 90 Utilization(%) Figure 6. Q_C under different utilizations. Utilization (%) Flow time(hours) 20000 -D-reentry 18000 16000 14000 Flow time (h) 12000 10000 8000 6000 60 70 80 90 Utilization(%) Utilization (%) Figure 7. Flow time under different utilizations. majority. _D Reentry had the largest gap with when work orders with one C reentry were in the majority, indicating a better protection of the due date. The above

Chang and Huang 10807 2600 TDD($) 2400 -D-reentry 2200 TDD ($) 2000 1800 1600 1400 M-M-M L-H-L L-L-H Layer-Mix Figure 8. TDD under different layer mixes. Layer-mix 2000 IDD($) 1900 -D-reentry 1800 IDD ($) 1700 1600 1500 1400 1300 M-M-M L-H-L L-L-H Layer-Mix Figure 9. IDD under different layer mixes. Layer-mix results showed that the deviation rate BSijl of the urgency of the influence on machine breakdowns added

10808 Afr. J. Bus. Manage. -320 Slack time(hours) -D-reentry -340-360 Slack time (h) -380-400 -420-440 -460-480 -500 M-M-M L-H-L L-L-H Layer-Mix Layer-mix Figure 10. DDST under different layer mixes. 55 -D-reentry 50 Queue length 45 40 35 30 25 20 M-M-M L-H-L L-L-H Layer-Mix Layer-mix Figure 11. Q_C under different layer mixes. into _D Reentry had a positive impact on the original. DISCUSSION As mentioned earlier, this paper attempted to illustrate why _D Reentry could perform better than for RFS in a random environment. In the example, utilized the BS to determine the priority of all work orders, while _D Reentry utilized the BS for RFS in a random environment (Table 3). Note that the BS depicted the planned value of the urgency for a specific work order with re-entrant flows, and did not consider machine failure

Chang and Huang 10809 14500 Flow time(hours) -D-reentry 14000 13500 Flow time (h) 13000 12500 12000 11500 11000 M-M-M L-H-L L-L-H Layer-Mix Layer-mix Figure 12. Flow time under different layer mixes. situations for RFS in a random environment. By contrast, _D Reentry adopted the weighted LBS to judge the actual influence of machine failures on each work order among layers. Finally, after the deviation rate BS was measured, _D Reentry could choose work orders that would result in the influence of machine failures causing delay for a random RFS. Most previous studies implemented the in a non-random environment, but this paper pioneered the application of for a random RFS. For example, Schragenheim (2006) and Schragenheim and Burkhard (2007) introduced buffer management, which is a mechanism that adopts the buffer status to decide the priority of a work order in front of the C, but does not consider the influence of machine failures. By contrast, _D Reentry determines the priority of a work order with a consideration of the influence of machine failures. In addition, _D Reentry emphasizes the concept of a C machine that reduces the complexity of scheduling problems from all the machines to the C machine. CONCLUSIONS AND FUTURE RESEARCH The main contribution of this study was to propose _DReentry, which considers the features of variable processing times and machine breakdowns of and adopts them for RFS in a random environment. This paper calculated a machine breakdown ratio through the overall accumulated and current C layers for each work order, then made a decision on the priority of all work orders based on the deviation rate BS for the buffer status of each re-entrant layer. This study compared six performance indexes related to the due date by applying four methods, as mentioned earlier. The results showed that the proposed modified method has better performance in a random environment with RFS characters when the product portfolio has a large proportion of multi-reentrant orders, or when the C utilization ratio increases from 60% to 90%. The TDD, IDD, due date slack time, Q_C and flow time obtained from _D Reentry also has better performance than the other methods. As this study did not consider the two machine bottleneck situations, future research should investigate this. In addition, further studies could perform bottleneck shiftiness on a system. REFERENCES Arzi Y, Herbon A (2000). Machine learning based adaptive production control for a multi-cell flexible manufacturing system operating in a random environment. Int. J. Prod. Res., 38(1): 161 185. Chang YC, Huang WT (2011). A modification of simplified drumbuffer-rope for re-entrant flow shop scheduling. Int. Technol. J., 10(1): 40 50. Chen JS, Pan JCH, Wu CK (2007). Minimizing makespan in reentrant flow-shops using hybrid tabu search. Int. J. Adv. Manuf. Technol., 34(3): 353 361. Chen JS, Pan JCH, Lin CM (2008). A hybrid genetic algorithm for the re-entrant flow-shop scheduling problem. Expert. Syst. Appl., 34:

10810 Afr. J. Bus. Manage. 570 577. De E, Ghosh JB, Wells CE (1993). Job selection and sequencing on a single machine in a random environment. Eur. J. Oper. Res., 70(3): 425-431. Graves SC, Meal HC, Stefek D, Zeghmi AH (1983). Scheduling of reentrant flow shops. J. Oper. Manag., 3: 197 207. Ho TF, Li RK (2004). Heuristic dispatching rule to maximize TDD and IDD performance. Int. J. Prod. Res., 42(24): 5133 5147. Hwang H, Sun JU (1998). Production sequencing problem with reentrant work flows and sequence dependent setup times. Int. J. Prod. Res., 36(9): 2435 2450. Lee JH, Chang JG., Tsai CH, Li RK (2010). Research on enhancement of TOC simplified drum-buffer-rope system using novel generic procedures. Expert. Syst. Appl., 37: 3747 3754. Mohebbi E (2008). A note on a production control model for a facility with limited storage capacity in a random environment. Eur. J. Oper. Res., 190(2): 562 570. Park Y, Kim S, Jun CH (2000). Performance analysis of re-entrant flow shop with single-job and batch machines using mean value analysis. Prod. Plan. Control, 11(6): 537 546. Pinedo M (2002). Scheduling: Theory, Algorithms, and Systems. Prentice-Hall: Englewood Cliffs, NJ. Russell RS, Taylor IIIBW (2005). Operations Management: Quality and Competitiveness in a Global Environment. John Wiley and Sons. Schragenheim E, Dettmer HW (2001). Manufacturing at Warp Speed: Optimizing Supply Chain Financial Performance. St. Lucie Press, APICS, Boca Raton, FL. Schragenheim E (2006). Using in rapid response projects. (http://www.scienceofbusiness.com/free-stuff/free-toc-vvarticles.aspx). Schragenheim E, Burkhard R (2007). Drum Buffer Rope and buffer management in a make-to-stock environment. (http://www.scienceofbusiness.com/free-stuff/free-toc-vvarticles.aspx). Vargas-Villamil FD, Rivera DE (2001). A model predictive control approach for real-time optimization of reentrant manufacturing lines. Comput. Ind., 45(1): 45 57. Yan H, Lou S, Sethi SP (2000). Robustness of various production control policies in semiconductor manufacturing. Prod. Oper. Manage., 9(2): 171 182.