The impact of order prioritization on the workload balancing capability of POLCA systems: A discrete-event simulation study

(1)

1

The impact of order prioritization on the workload balancing

capability of POLCA systems:

A discrete-event simulation study

Vivian Turnbull S2357062

Master thesis, MSc Technology & Operations Management University of Groningen, Faculty of Economics and Business

June 2017

(2)

2

Abstract

Purpose - POLCA is a pull production system that aims to help MTO firms shorten

throughput times by balancing the workload. However, research has shown that unit-based POLCA systems have modest or no effective workload balancing capability when high variability in processing times is present. A plausible explanation is that unit-based POLCA systems do not take into account orders’ processing times. Potential solutions for this problem are the implementation of a load-based POLCA system or a unit-based POLCA system with a Shortest Processing Time (SPT) priority rule. The three considered POLCA systems all prioritize orders in a different way, and therefore, the aim of this study is to find the impact of order prioritization on the workload balancing capability of POLCA systems under various levels of processing time variability.

Methodology - A discrete event simulation is performed in a divergent topology with three

stages. In the POLCA systems, the same number of cards is used in each control loop. Various coefficients of variations are used for the processing time variability.

Findings – It was found that under processing time variability, order prioritization based on

the SPT rule results in a much more effective workload balancing capability than order prioritization based on the unit-based or load-based POLCA mechanisms. Further, the results show that load-based POLCA systems achieve a slightly better workload balancing capability than unit-based POLCA systems.

Originality/value – This study addresses the knowledge gap on how order prioritization in

POLCA systems affects the workload balancing capability. For practice, the results of this study suggest that POLCA might actually not be the most suitable system for MTO firms when significant processing time variability exists.

Keywords – POLCA, unit-based, load-based, Shortest Processing Time priority rule, order

(3)

3

Abbreviations

CONWIP Constant Work In Progress FCFS First Come First Serve m-CONWIP Multiple-CONWIP MTO Make-To-Order OPT Order Pool Time PC Production Control

POLCA Paired-cell Overlapping Loops of Cards with Authorization PPC Production Planning and Control

(4)

4

1. Introduction

Ensuring short throughput times is important for Make-To-Order (MTO) firms in order to obtain a competitive advantage. POLCA is a production control system that aims to help MTO firms shorten throughput times (Riezebos, 2010). POLCA stands for paired-cell overlapping loops of cards with authorization and is capable of controlling the material flow in shop floor configurations with many different routings between the cells or work stations (Suri, 1998). POLCA limits the Work-In-Progress (WIP) by using cards and is used to regulate the order progress between cells of workstations (Riezebos, 2010). Within POLCA systems, a distinction is made between unit-based and load-based systems (Germs & Riezebos, 2010). Unit-based systems limit the workload on the shop floor by controlling the number of orders, while load-based systems do this by controlling the work content (processing time) of orders.

The throughput time performance of POLCA in an MTO environment can be attributed to the workload balancing capability of the POLCA system (Germs & Riezebos, 2010). Germs and Riezebos (2010) and Ziengs, Riezebos, & Germs (2012) showed that unit-based POLCA systems in a divergent topology have modest or no effective workload balancing capability, and thus do not perform well in terms of total throughput time, when a significant amount of processing time variability is present. Yet, processing time variability is present in MTO firms, since these firms are characterised by customization and by a large variety in products (Stevenson, Hendry, & Kingsman, 2005).

The lack of workload balancing capability under high variability in processing times can be caused by the fact that unit-based POLCA systems control the workload based on the number of orders and do not consider the processing times of orders (Germs & Riezebos, 2010). The variability in processing times then leads to variability in workload at each workstation. As a consequence, the workload will not be evenly distributed among the workstations, which results in no or modest reduction in total throughput time.

(7)

7

improved workload balancing capabilities compared to unit-based m-CONWIP systems. In the study of Bodnar (2016), this was only true for small long/short order ratios. No research has been conducted yet on the effect of the implementation of a load-based POLCA system on the workload balancing capability. However, since m-CONWIP and POLCA are both pull production systems, similar results as Bodnar (2016) and Wilz (2017) could be envisioned. The mechanism of load-based POLCA systems can prioritize smaller orders when there is limited capacity available at the upcoming workstations. Still, orders are placed in a limited number of groups to indicate their order size. Thus, not all differences in processing times are taken into account.

The Shortest Processing Time (SPT) priority rule, on the other hand, does take into account all differences in orders’ processing times. This rule prioritizes orders in the queue that have the shortest processing time. The SPT rule is another possible solution to the workload balancing problem of unit-based POLCA systems under high processing time variability. It has already been shown that the SPT rule effectively decreases the average order throughput time in job shops (Land, 2004). However, no research has been conducted yet on the performance of POLCA systems combined with an SPT rule.

In this study, the workload balancing capability of unit-based POLCA systems, load-based POLCA systems and unit-based POLCA systems with an SPT rule are analysed under different levels of processing time variability. These three systems all prioritize orders in a different way, and therefore, the aim of this study is to find the impact of order prioritization on the workload balancing capability of POLCA systems under various levels of processing time variability. The corresponding research question is:

What is the impact of order prioritization on the workload balancing capability of POLCA systems under various levels of processing time variability?

To address the aim and research question, a discrete event simulation is performed. The problem is addressed in a divergent topology with three consecutive stages. Five main systems are implemented in the model: a push system, a unit-based POLCA system, a push system with an SPT rule, a unit-based POLCA system with an SPT rule and a load-based POLCA system. Various coefficients of variation in processing time variability are used.

(8)

8

time performance of POLCA systems. Moreover, it investigates the applicability of POLCA systems in MTO firms when significant processing time variability is present.

For practice, this understanding will help firms to find a more appropriate POLCA design than the unit-based POLCA systems. The implementation of it can then lead to shorter throughput times and a competitive advantage. The results of this study can also make firms aware that POLCA might not be the most suitable system for MTO firms when significant processing time variability is present.

(9)

9

2. Background

2.1 Production Planning and Control Systems

Production Planning and Control (PPC) systems are important tools for MTO firms, since these systems aim to minimize shop floor throughput times and lead times, reduce WIP and improve responsiveness to changes in demand (Stevenson, Hendry, & Kingsman, 2005). Zäpfel & Missbauer (1993) define main functions of a PPC system, namely; scheduling and capacity planning (order acceptance), order release, and order and capacity control (order dispatching).

Zäpfel and Missbauer (1993) distinguish PPC systems and Production Control (PC) systems in that the latter only consider the order release and order dispatching. The order release controls the input of orders to the shop floor (Land, 2004). Order dispatching addresses which order should be selected to be processed next at a workstation.

Within the set of developed PPC and PC systems, a distinction can be made between push and pull production systems. Hopp & Spearman (2004) give a definition for pull and push production systems: “A pull production system is one that explicitly limits the amount of work in process that can be in the system. By default, this implies that a push production system is one that has no explicit limit on the amount of work in process that can be in the system” (p. 142). Examples of push production systems are Material Requirements Planning (MRP) systems and Manufacturing Resources Planning (MRP II) systems (Vollmann, Berry, & Whybark, 1992). Kanban (Sugimori, Kusunoki, Cho, & Uchikawa, 1977), Constant WIP (CONWIP) (Spearman, Woodruff, & Hopp, 1990) and POLCA systems (Suri, 1998) are examples of pull production systems.

Whenever there are disruptions in the production system, the WIP limit in pull production systems ensures that the WIP will not grow further than the determined limit, because the release and dispatching will halt (Hopp & Spearman, 2011). Push systems, on the contrary, schedule order releases in the system and there is no limit on WIP. Therefore, when disruptions occur orders can keep being released, which can cause a large amount of WIP to accumulate on the shop floor.

(10)

10

only control the release and dispatching of orders. In the next section, the order release and order dispatching mechanisms of POLCA will be explained.

2.2 POLCA

POLCA is designed specifically for MTO firms and aims to achieve short and stable throughput times (Riezebos, 2010). Riezebos (2010) thorhoughly discusses the design of POLCA systems. POLCA uses route-specific control on the shop floor. This means that information on the availability of capacity in the routing of orders is taken into account in decisions concerning the order release and progress on the shop floor. The route-specific control is implemented by visual signals (cards) and overlapping control loops. Figure 1 shows the control loops of a POLCA system.

Figure 1: POLCA control loops.

(11)

11

routing, such an order can then already be selected and processed. That way, the order for workstation B will be bypassed. This provides important implications for balancing the workload of the system and shortening throughput times.

2.3 Workload balancing capability

Since the shop floor configurations of MTO firms often have a high routing variety, the throughput time performance of a POLCA system largely depends on its capability to effectively balance the workload (Ziengs, Riezebos, & Germs, 2012). The workload balancing capability of a pull system, such as POLCA, indicates the capability of the system to evenly distribute the workload over the workstations on the shop floor (Land & Gaalman, 1998). The workload balancing capability of POLCA is considered to be effective if both the average shop floor throughput time (STT) and the average total throughput time (TTT) are reduced compared to a pure push system (Germs & Riezebos, 2010). The STT of an order is the time between the release of an order to the shop floor and the completion of the order. The time between the arrival and the completion of an order is the TTT of an order. Figure 2 provides a visual representation of these throughput time measures. Since POLCA limits the WIP on the shop floor, shop floor throughput time can be reduced following Little’s law (Hopp & Spearman, 2011). However, this will lead to an increase in the order pool time (OPT), since the orders now will queue there. Thus, only when the STT is decreased sufficiently such that it compensates the increased OPT, the TTT can be reduced and the workload is considered to be effectively balanced.

Figure 2: Throughput time measures in MTO production systems.

2.4 Unit-based POLCA systems

(12)

12

can be caused by the fact that unit-based POLCA systems control the number of orders on the shop floor (Germs & Riezebos, 2010); one card represents one order. Processing times of orders are not taken into account when dispatching and releasing orders. When there is variability in processing times, the signal represented by the cards becomes less effective and a large variability in workload can occur at each workstation. Consequently, the workload is not evenly distributed among the workstation and the workload balance is reduced. Appendix A contains an example that illustrates how the mechanisms of a unit-based POLCA system can result in an unbalanced workload in the system.

2.5 Load-based POLCA systems

Load-based POLCA systems do consider the processing time of orders when releasing and dispatching orders, and are therefore expected to provide a better workload balancing capability under processing time variability than unit-based POLCA systems (Germs & Riezebos, 2010). Vandaele, Van Nieuwenhuyse, Claerhout & Cremmery (2008) proposed a load-based POLCA system in combination with an advanced resources planning (ARP) system. During the implementation of this load-based POLCA system in a metal shop positive preliminary results were obtained. However, the paper does not explicitly discuss the workload balancing capability or throughput time performance of the system. Further, Bodnar (2016) and Wilz (2017) showed that load-based m-CONWIP systems had improved workload balancing capabilities compared to unit-based m-CONWIP systems. In the study of Bodnar (2016) this was only the case for small long/short order ratios. Further, Wilz (2017) made an alternation to the load-based system of Bodnar (2016), which resulted in a load-based system that achieved a better workload balancing capability than the load-based system of Bodnar. For POLCA systems, no research has been done yet on the effect on the workload balancing capability of the system when a load-based system is implemented. Since m-CONWIP and POLCA are both pull production systems, it is expected that also in POLCA systems more effective workload balancing can be found when a load-based system is implemented.

(13)

13

The main difference between unit-based and load-based POLCA systems lies in the use of the cards. While in the unit-based POLCA system one card represents one order, in the load-based POLCA system a card represents a certain amount of processing time. Each load-based card indicates a certain maximum amount of processing time. If an order requires less processing time than this threshold, then it requires one card to be attached to it in order to proceed. If an order requires more processing time than the threshold, it requires two cards. Thus, actually two bins are created; one bin of orders’ processing times that require one card and one bin of orders’ processing times that require two cards. In this study, only two bins are used, since the study of Bodnar (2016) showed that no differences in TTT performance were found in a load-based m-CONWIP system when three instead of two bins were used.

(14)

14

Figure 3: Decision tree of the load-based POLCA system.

Because the load-based POLCA system takes into account processing times of orders, it is expected that it achieves a better workload balancing capability under variability in processing times than unit-based POLCA systems. This would also be in line with the studies of Bodnar (2016) and Wilz (2017). Appendix B contains an example that illustrates how a load-based POLCA system can result in a better workload balancing capability than a unit-based POLCA system. Moreover, a second example is given in appendix B to visualize the prioritization of small orders in load-based POLCA systems.

2.6 Shortest Processing Time priority rule

Bodnar (2016) and Wilz (2017) underlined the positive effect that the prioritization of small orders in load-based m-CONWIP systems has on the workload balancing capability. This positive effect is also expected in load-based POLCA systems. The SPT priority rule is a more extensive implementation of the prioritization of small orders and therefore might provide an even better workload balancing capability. The SPT rule sequences orders based on their processing time at the upcoming workstation and prioritizes orders with a shorter processing time (Haupt, 1989). The rule takes into account all differences in orders’ processing times.

Haupt (1989) discusses several other priority rules that also prioritize orders based on their processing time, such as the Least Work Remaining rule, which uses the cumulative processing time of the remaining operations. Haupt (1989) further mentions priority rules that consider orders’ due dates, such as the Earliest Due Date rule. However, this rule is outside the scope of this study, because the aim of the study is to analyse systems that consider the processing times of orders.

(15)

15

priority rule that combines load-oriented and time-oriented sequencing. However, compared to such rules, the SPT rule is much more straightforward to implement. Moreover, the aim of this study is to analyse the systems’ workload balancing capability, not the due date performance.

A priority rule such as the SPT rule aids in the release and dispatching decisions of a Production Control system. The SPT rule is implemented in each queue of the unit-based POLCA system. Then, the orders get sequenced in the queue based on their required processing time at their upcoming workstation. When a card becomes available, the order with the shortest processing time will be selected from the set of orders that follow this card’s routing. Moreover, when a workstation becomes available, the order with the shortest processing time will be selected from the queue in front of it. If orders have the same required processing time, they will be sequenced on a FCFS basis.

The SPT rule is one of the most applied and most efficient priority rules (Haupt, 1989). Amongst others, it has shown to be effective in decreasing the throughput time of orders in job shops (Land, 2004). The studies of Bodnar (2016) and Wilz (2017) also touched upon the performance benefits of prioritizing small orders in m-CONWIP systems. It is expected that the implementation of an SPT rule in a unit-based POLCA system will also lead to an improved workload balancing capability and throughput time performance. In appendix C, an example is provided to illustrate the benefits of the SPT rule in a unit-based POLCA system.

2.7 Order prioritization mechanisms

(16)

16

A distinction between the load-based POLCA system and the unit-based POLCA system with an SPT rule is that the load-based POLCA system divides the orders into two groups: small and large orders. The unit-based POLCA system with an SPT rule, on the other hand, sequences the orders from small to large in each queue, thus taking all differences in processing time into account. As a consequence, the truly shortest orders will be processed first. Therefore, it might be expected that the unit-based POLCA system with an SPT rule will have a better flow of orders and workload balancing capability compared to the load-based POLCA system. However, Bodnar (2016) found that adding more distinct processing time bins in the load-based m-CONWIP system did not change the workload balancing capability and this might indicate that a larger distinction between orders’ processing times will not lead to a better workload balancing capability. Further, the largest orders are literarily placed at the end of the queue each time in the unit-based POLCA system with an SPT rule, which might result in larger negative consequences for the throughput time performance of the largest orders than in the load-based POLCA system. In the latter system, the same issue might arise, but it will be less severe because of the inaccuracy in taking all processing times into account. Within the processing time bin of large orders, the largest orders will not have to wait longer than the other orders.

Since the three discussed POLCA systems all have different order prioritization mechanisms, this study will investigate the research question:

(17)

17

3. Methodology

3.1 Research method

To answer the research question, discrete-event simulation is used. Discrete-event simulation is used for modelling queueing systems (Robinson, 2014). In the MTO queuing system of this study, several sources of variability are present, such as variability in the order routings, in the order inter-arrival times and in the processing times. Furthermore, the workstations are interconnected and influence each other. Since simulation models are able to represent complex systems which include variability and interdependencies, simulation is the most applicable method to answer the research question (Robinson, 2014). Siemens’ Tecnomatix Plant Simulation 12 is used to build the simulation model and to run the experiments (Tecnomatix Plant Simulation, 2014). This software is chosen, because it provides tools with sufficient ease of use that allow to model and analyse complex, stochastic systems in an efficient way.

3.2 Model design

The simulation model consists of a divergent MTO system. The layout of the workstations, which is called the shop floor topology, can be seen in figure 4. In the divergent topology there are three stages and the number of workstations doubles at each stage. In total there are seven workstations and four routings are possible. The divergent topology with different routings enables that the workload can be balanced throughout the system. Therefore, it is an appropriate topology to analyse the workload balancing capability and throughput time performance of the systems.

(18)

18

The four different routings occur with the same probability and the routing of an order is known when the order arrives in the order pool. One order arrives at a time, thus the batch size is one. Because the system is an MTO system, it has the characteristics that the production cannot start until the order has arrived and that there is no inventory of finished goods. Furthermore, each workstation processes a single order at a time and has a constant capacity. The choice for the inter-arrival time variability and the utilization levels in the simulations is based upon the study of Germs and Riezebos (2010). First of all, the results of Germs and Riezebos (2010) showed that the considered POLCA systems had a better workload balancing capability when variability in order inter-arrival times was present compared to constant inter-arrival times. This can be explained by the fact that increased inter-arrival time variability leads to larger average queue lengths. Therefore, there will be more choice of orders in the queues, which enables better workload balancing. Exponentially distributed inter-arrival times are considered in the simulations of this study, since more effective workload balancing can then take place. Second, similar as in the study of Germs and Riezebos (2010), the arrival rate in the simulations is set such that the workstations have an average utilization level of 80% or 90%. The results of Germs and Riezebos (2010) showed that workload balancing is more effective in POLCA systems with higher levels of utilization.

The level of utilization is a first experimental variable. The second experimental variable is the variability in processing times. The considered processing times are random with an Erlang-2, Erlang-4 and Erlang-6 distribution. The corresponding squared coefficients of variation for processing time variability respectively are 0,71, 0,50 and 0,41. Also the case for constant processing times is investigated for model validation. Under constant processing times, the unit-based POLCA systems with an SPT rule and the load-based POLCA system are the same as a unit-based POLCA system. The push system with an SPT rule then is the same as a FCFS push system.

(19)

19

have the same average utilization level. The required processing times of an order are known when it arrives in the order pool.

A third experimental variable is the configuration of the POLCA system. Figure 4 shows the control loops of the POLCA system. This structure is the same for each POLCA system. However, the number of cards is varied in the experiments. The same number of cards is used for each control loop. First, the number of cards is set to a very large number, such that the release of orders will not be restricted. This represents the system without POLCA (a push system). Then, the number of cards is restricted and therefore the POLCA system becomes constrained. The number of cards is an experimental variable and an exhaustive search is performed to find the optimal configuration of the POLCA systems. The optimal configuration is the configuration that achieves the largest reduction in TTT. The exhaustive search is performed within the range of 1 till approximately 70 cards. The optimal configurations of each system are then compared to each other, given the restriction that all control loops have the same number of cards.

Further, a unit-based POLCA system is implemented, which means that each order always requires one card. However, also the load-based POLCA system that was explained in chapter 2 is implemented. The load-based POLCA system has a certain small to large order ratio, which indicates the ratio of small (require a single card) against large orders (require two cards). For example, when the order ratio is 80/20, 80% of the orders are considered small and 20% are considered large. Bodnar (2016) and Wilz (2017) found that load-based m-CONWIP systems with different small to large order ratios resulted in different TTT performances and workload balancing capabilities. Therefore, the ratio of small to large orders also is an experimental variable in this study. Then the load-based POLCA systems with the most effective workload balancing capabilities can be compared to the other systems. The small to large order ratios are taken relatively large, since Bodnar (2016) and Wilz (2017) found that load-based m-CONWIP systems showed a better TTT performance when the small to large order ratio was large.

(20)

20

total processing time at their first and second workstation. For a small to large order ratio of 80/20, the threshold is then taken to be the processing time of the order that is on the border of the largest 20% orders. This is then set as the threshold for the cards in the control loops between the first and second stage in the divergent topology. In a similar manner the threshold is determined for the control loops between the second and third stage. Now the orders are sorted based on their total processing time at their second and third workstation. The processing time thresholds of the control loops are determined for each small to large order ratio and for each coefficient of variation in processing time variability. The last experimental variable is the adaptation of an SPT priority rule in each queue of the system.

3.3 Experimental design

An overview of the experimental variables and their possible settings can be found in table 1. The experimental variable of the number of cards per loop only applies to POLCA systems and the order ratio only applies to load-based POLCA systems. Further, the SPT rule is not used in combination with the load-based POLCA system. This experimental design leads to a set of 68 different experiments, excluding the exhaustive search for the optimal card counts. Including the exhaustive search, roughly 2500 simulations are then performed.

Variable Experimental setting

Utilization level 90%, 80%

Processing time variability None, Erlang-2, Erlang-4, Erlang-6 POLCA implementation None, unit-based, load-based Number of cards per loop 1-70, ∞

Order ratio 80/20, 90/10, 95/5, 97,5/2,5, 98,5/1,5, 99/1

SPT priority rule Yes, No

Table 1: Overview of experiments.

(21)

21

determined by taking the total processing time of the orders that are situated on the border of the 20%, 40%, 60%, 80% and 100% smallest orders. In the performed experiments, an order can then fall into one of the five order size ‘buckets’. An overview of the groups can be seen in table 2.

Group Percentage of smallest orders Bucket 1 0 – 20 %

Bucket 2 20 – 40 % Bucket 3 40 – 60 % Bucket 4 60 – 80 % Bucket 5 80 – 100%

Table 2: Overview of the five main order size groups.

For the simulations, a warm-up period of 30 days and a run length of 450 days were chosen. The Welch’s procedure (Heidelberger & Welch, 1983) confirmed that the warm-up period was sufficient and the graphical convergence run length method of Robinson (2014) confirmed that the run length was sufficient. Further, the graphical approach of Robinson (2014) was used to select the number of replications and it was found that 25 replications would provide accurate results. The warm-up period, run length and number of replications have been conservatively selected and more details of the selection procedure can be found in appendix D.

3.4 Model validation

Several steps were taken to validate the simulation model. First of all, the model was validated visually. Second, calculations with queueing theory were performed and confirmed the results of the push system in which the orders subsequently had the routings 1, 3, 2 and 4. Third, for all experiments it was checked whether the average utilization level of the workstations indeed was 90% and 80%.

(22)

22

4. Results

In this section, the results of the experiments are presented. First, the workload balancing capability of the unit-based POLCA system, unit-based POLCA system with an SPT rule, push system with an SPT rule and load-based POLCA system will be given for all considered levels of utilization, levels of processing time variability and small to large order ratios. Next, the performance per order size is outlined to indicate potential negative effects of the systems.

4.1 Workload balancing capability

First of all, the percentages reduction in TTT and STT compared to a pure push system were determined for each considered system to analyse its workload balancing capability. Table 3 shows the results under various utilization levels and coefficients of variation in processing time variability. The results for the load-based POLCA system are for a small to large order ratio of 97,5/2,5. The optimal configurations of the POLCA systems are also stated in the table.

90% utilization 80% utilization

CV System conf. %TTT %STT conf. %TTT %STT

0,71 Unit-based [∞,∞] 0,00% 0,00% [26,26] 0,001% 0,02% Unit-based + SPT [∞,∞] 47,94% 47,94% [16,16] 32,27% 32,27% Push + SPT NA 47,94% 47,94% NA 32,27% 32,27% Load-based 97,5/2,5 [28,28] 0,11% 1,36% [22,22] 0,008% 0,107% 0,50 Unit-based [34,34] 0,08% 0,27% [21,21] 0,01% 0,06% Unit-based + SPT [27,27] 36,92% 36,93% [16,16] 23,03% 23,03% Push + SPT NA 36,91% 36,91% NA 23,02% 23,02% Load-based 97,5/2,5 [19,19] 0,54% 4,07% [11,11] 0,29% 1,65% 0,41 Unit-based [21,21] 0,16% 0,96% [14,14] 0,03% 0,26% Unit-based + SPT [16,16] 30,28% 30,47% [15,15] 18,41% 18,42% Push + SPT NA 30,25% 30,25% NA 18,40% 18,40% Load-based 97,5/2,5 [15,15] 0,74% 4,27% [9,9] 0,24% 2,47%

Table 3: Reduction in TTT and STT compared to a pure push system.

(23)

23

of the unit-based POLCA systems under the various coefficients of variation and a utilization level of 90%. The figure shows plots of the percentage STT and TTT compared to a push system for various card counts for the unit-based POLCA systems. Thus, a push system has a %STT and %TTT of 100%. If the unit-based POLCA systems have a %STT and %TTT smaller than this, a reduction in STT and TTT is achieved and the workload is considered to be balanced. From the left to the right in the graph, the card count becomes larger. For the push system, there is no card count and therefore, the push systems are respresented by one point. It can be seen that the unit-based POLCA system only achieves small TTT reductions under a coefficient of variation of 0,50 and 0,41.

Figure 5: Percentage of STT and TTT of unit-based POLCA systems compared to a push system.

The results of the experiments show that the systems with an SPT rule as well as the load-based POLCA systems achieve a larger TTT and STT reduction than the unit-based POLCA systems. This indicates that these systems have a better workload balancing capability than unit-based POLCA systems.

The systems with an SPT rule show a much larger TTT reduction than the unit-based and load-based POLCA systems. As can be seen in table 3, under a coefficient of variation of 0,71 and a 90% utilization level the systems with an SPT rule achieve a TTT reduction of approximately 48% compared to a pure push system. Under a coefficient of variation of 0,50 and 0,41 the TTT reductions of the systems with an SPT rule are approximately 37% and 30%. For the 80% utilization level, the systems with an SPT rule achieve TTT reductions between the 18% and 32,3%.

99,60% 99,80% 100,00% 100,20% 100,40% 100,60% 100,80% 101,00% 97,00% 97,50% 98,00% 98,50% 99,00% 99,50% 100,00% % TTT o f p u sh sy ste m % STT of push system

(24)

24

Another finding was that under both utilization levels the unit-based POLCA system with an SPT rule only result in a very small additional TTT reduction compared to a push system with an SPT rule. From table 3 it can be seen that under a coefficient of variation of 0,50 and 0,41 the additional TTT reductions were only 0,01% or 0,03%. When the coefficient of variation was 0,71, the unit-based POLCA system with an SPT rule did not even show any additional TTT reduction at all.

Figure 6 shows a plot of the percentage STT and TTT compared to a push system for various card counts for the unit-based POLCA systems with and without an SPT rule. The coefficient of variation is 0,71 and the utilization level is 90%. The percentage TTT and STT compared to a pure push system are also indicated for the push system with an SPT rule. First of all, the figure illustrates that the unit-based POLCA system does not show any TTT reduction compared to the push system and thus does not show an effective workload balancing capability. Second, the figure depicts the TTT reduction that is achieved by the push system with an SPT rule. Lastly, it is visualised that the unit-based POLCA system with an SPT rule also achieves a TTT reduction compared to the pure push system, but does not show any additional TTT reductions compared to the push system with an SPT rule.

Figure 6: Percentage of STT and TTT compared to a push system for three different systems under a coefficient of variation of 0,71.

Beside the systems with an SPT rule, also the load-based POLCA systems show larger TTT and STT reductions than the unit-based POLCA system, which indicates a more effective workload balancing capability. This is the case for all considered small to large order ratios, coefficients of variation and utililzation levels. Table 4 gives an overview of the TTT

40% 60% 80% 100% 120% 140% 160% 20% 30% 40% 50% 60% 70% 80% 90% 100% % TTT o f p u sh sy ste m % STT of push system

(25)

25

and STT reductions achieved by the different load-based POLCA systems. It can be seen that this time, the TTT reductions are still very small. Under a 90% utilization level and a coefficient of variation of 0,71 the largest TTT reduction a load-based POLCA system achieves is 0,19%. For a coefficient of variation of 0,50 and 0,41 the largest TTT reductions a load-based POLCA system achieves are 0,54% and 0,74%. Under a 80% utilization level, the TTT reductions are even smaller.

90% utilization 80% utilization

CV System conf. %TTT %STT conf. %TTT %STT

0,71 Load-based 80/20 [59,59] 0,002% 0,022% [28,28] 0,007% 0,075% Load-based 90/10 [48,48] 0,01% 0,09% [27,27] 0,005% 0,048% Load-based 95/5 [35,35] 0,06% 0,56% [25,25] 0,003% 0,052% Load-based 97,5/2,5 [28,28] 0,11% 1,36% [22,22] 0,008% 0,107% Load-based 98,5/1,5 [29,29] 0,19% 1,20% [22,22] 0,013% 0,104% Load-based 99/1 [30,30] 0,12% 0,97% [22,22] 0,006% 0,096% 0,50 Load-based 80/20 [35,35] 0,15% 0,68% [14,14] 0,09% 1,15% Load-based 90/10 [25,25] 0,19% 1,94% [13,13] 0,16% 1,11% Load-based 95/5 [24,24] 0,41% 2,03% [10,10] 0,32% 2,24% Load-based 97,5/2,5 [19,19] 0,54% 4,07% [11,11] 0,29% 1,65% Load-based 98,5/1,5 [19,19] 0,47% 3,91% [11,11] 0,22% 1,54% Load-based 99/1 [23,23] 0,44% 2,08% [12,12] 0,14% 1,00% 0,41 Load-based 80/20 [24,24] 0,26% 1,20% [11,11] 0,11% 2,11% Load-based 90/10 [17,17] 0,44% 3,38% [10,10] 0,30% 2,69% Load-based 95/5 [16,16] 0,53% 3,56% [8,8] 0,35% 4,19% Load-based 97,5/2,5 [15,15] 0,74% 4,27% [9,9] 0,24% 2,47% Load-based 98,5/1,5 [15,15] 0,70% 4,10% [10,10] 0,20% 1,55% Load-based 99/1 [16,16] 0,54% 3,16% [11,11] 0,14% 0,98%

Table 4: Reduction in TTT and STT compared to a pure push system.

(26)

26

Figure 7: Percentage TTT reduction for different order ratios in the load-based POLCA systems.

Figure 8: Percentage TTT reduction for different order ratios in the load-based POLCA systems.

In figure 9, the workload balancing capability of the load-based POLCA system with an order ratio of 98,5/1,5 is visually depicted. The coefficient of variation is 0,71 and the utilization level is 90%. The figure shows a plot of the percentage STT and TTT compared to a push system for various card counts for this load-based POLCA system. The reductions in TTT and STT can be seen in the graph, which indicate the effective workload balancing capability of the system.

0,000% 0,100% 0,200% 0,300% 0,400% 0,500% 0,600% 0,700% 0,800% TTT r e d u ction Order ratio

Load-based POLCA systems, 90% utilization

CV = 0,71 CV = 0,50 CV = 0,41 0,000% 0,100% 0,200% 0,300% 0,400% 0,500% 0,600% 0,700% 0,800% TTT r e d u ction

Load-based POLCA systems, 80% utilization

(27)

27

Figure 9: Percentage of STT and TTT compared to a push system under a coefficient of variation of 0,71 and a 90% utilization level.

Another observation that is done from the results is that the unit-based POLCA systems and load-based POLCA systems perform slightly better in TTT and STT under a lower level of processing time variability. On the contrary, the systems with an SPT rule achieve much smaller reductions in TTT for a lower level of processing time variability. Figure 10 shows the slighly increasing TTT reduction for the unit-based and load-based POLCA system when the coefficient of variation decreases and the decreasing TTT reduction for the systems with an SPT rule. The utilization level is 90%. The systems with an SPT rule are represented by one line, since the TTT reductions were almost exactly the same.

Figure 10: TTT reductions for a decreasing coefficient of variation and a 90% utilization level.

Lastly, it can be seen in table 3 and table 4 that the optimal card count of the systems decreases when the utilization level goes from 90% to 80%. Also, in general the TTT and

(28)

28

STT reductions decrease then. These observations are both in line with the results from the studies of Germs and Riezebos (2010) and Ziengs, Riezebos and Germs (2012). Only under a coefficient of variation of 0,71 the unit-based POLCA system and the load-based POLCA system with an order ratio of 80/20 achieve slightly larger reductions in TTT under a utilization level of 80%. However, the reductions are very small.

4.2 Performance per order size

Besides the workload balancing capabilities of the main systems, other interesting measures to consider are the average TTT, STT and waiting time of several groups of order sizes. These measures are of interest, because they help to investigate whether large differences exist between the performance of different order sizes and to identify possible negative effects of the systems. Five buckets of order sizes were created: buckets containing orders with 0-20%, 20-40%, 40-60%, 60-80% and 80-100% of the smallest total processing time.

It was found that for a 90% utilization level, under all considered levels of processing time variability the average TTT and STT increase for larger order sizes. However, the differences in average TTT and STT between the order size groups are much larger in the systems with an SPT rule than in the other systems. Further, it was found that there are also large differences in the average waiting time per order size in the systems with an SPT rule. Large orders have a much longer average waiting time than small orders. In the push system, the unit-based POLCA system and the load-based POLCA system, the average waiting times do not differ much for the different order size groups.

(29)

29

Figure 11: Average TTT of different order size groups under a coefficient of variation of 0,50 and a 90% utilization level.

Figure 12: Average STT of different order size groups under a coefficient of variation of 0,50 and a 90% utilization level.

Figure 13: Average waiting time of different order sizes under a coefficient of variation of 0,50 and a 90% utilization level.

0,0 10,0 20,0 30,0 40,0 50,0 60,0 70,0

Push Unit-based Push + SPT Unit-based + SPT Load-based 97,5/2,5 H o u rs

Total Throughput Time

Bucket 1: 0 - 20 % Bucket 2: 20 - 40 % Bucket 3: 40 - 60 % Bucket 4: 60 - 80 % Bucket 5: 80 - 100 % 0,0 10,0 20,0 30,0 40,0 50,0 60,0 70,0

Push Unit-based Push + SPT Unit-based + SPT Load-based 97,5/2,5 H o u rs

Shop Floor Throughput Time

Bucket 1: 0 - 20 % Bucket 2: 20 - 40 % Bucket 3: 40 - 60 % Bucket 4: 60 - 80 % Bucket 5: 80 - 100 % 0,0 10,0 20,0 30,0 40,0 50,0 60,0 70,0

(30)

30

Concerning the average waiting time, it was also found that in the load-based POLCA systems, the group of smallest orders often does not have the shortest average waiting time. This can for example be seen in the load-based POLCA system in figure 13. Also, in several cases of the load-based POLCA systems, the group of largest orders has the shortest average waiting time. For example, under a coefficient of variation of 0,71 and a 90% utilization level this is the case for all considered load-based POLCA systems. Figure 14 shows this graphically.

Figure 14: Average waiting time of different order size groups under a coefficient of variation of 0,71.

More data on the performance per order size can be found in appendix E. This appendix gives an overview of the average TTT, STT and waiting time per order size group for each considered system and coefficient of variation under a utilization level of 90%.

Lastly, a different way of defining the order size groups in load-based POLCA systems was considered. The orders are then divided into groups based on their size in their encountered control loops. For example, an order can be considered small in the first control loop it encounters (requires a single card) and large in the second loop it encounters (requires two cards). Again it was found that in load-based POLCA systems the group of smallest orders often do not have the shortest average waiting time and the group of largest orders have the shortest waiting time in several cases. Appendix F gives a detailed explanation of the groups that were formed and provides a full overview of the results for these groups.

(31)

31

Table 5 provides a short summary of the main results of this chapter. The summary covers all considered levels of utilization and processing time variability.

Unit-based POLCA Unit-based POLCA + SPT Push + SPT Load-based POLCA Workload balancing capability No/Modest Effective Effective Modest TTT and STT reductions … for higher

levels of processing time variability Decrease Increase Increase Decrease TTT and STT differences per order

size Small Large Large Small

(32)

32

5. Discussion

After presenting the results in chapter 4, the results will now be discussed. All results are for a system with a three-stage divergent topology and random-inter arrival times.

5.1 Interpretation

5.1.1 Unit-based POLCA systems

First of all, the results show that under a coefficient of variation of 0,71 in processing time variability a unit-based POLCA system does not achieve any (or a very small) TTT reduction compared to a pure push system. Thus, a unit-based POLCA system does not have an effective workload balancing capability under these circumstances. Under a coefficient of variation of 0,50 and 0,41 a unit-based POLCA systems does show small TTT reductions, indicating a modest workload balancing capability.

One of the possible reasons that a unit-based POLCA system shows no or modest effective workload balancing capability under processing time variability is that it does not take the processing times of orders into account. Unit-based POLCA systems with an SPT rule and load-based POLCA systems do take into account the processing times of orders and the results show that these systems reach a larger reduction in TTT and STT compared to a unit-based POLCA system for all considered coefficients of variation and utilization levels. This indicates that under processing time variability these systems indeed have a better workload balancing capability than the unit-based POLCA system.

5.1.2 Shortest Processing Time rule

The systems with an SPT rule show large TTT reductions compared to a pure push system. The unit-based POLCA systems with an SPT rule achieve large TTT reductions under processing time variability, but the additional TTT reductions compared to the push systems with an SPT rule are zero or minimal. These results indicate that the SPT rule just performs very well in TTT reduction on its own and is the major contributor to the TTT reductions in unit-based POLCA systems with an SPT rule.

5.1.3 Load-based POLCA systems

(33)

33

5.1.4 Differences SPT rule and load-based POLCA systems

The TTT and STT reductions are thus much smaller in the load-based POLCA systems than in the systems with an SPT rule. This indicates that the workload balancing capability of load-based POLCA systems is less effective than that of the systems with an SPT rule. To understand why this large difference occurs, the main mechanisms of the different types of order prioritizations must be analysed. First of all, one of the main differences in the prioritization of the systems is that the load-based POLCA system divides the orders into two groups for the prioritization: small and large orders. The SPT rule, on the other hand, sequences the orders from small to large in each queue, thus taking all differences in processing time into account.

The second main difference is that the load-based POLCA system’s main principle is to better reflect the capacity that is free in the control loops. This can then result in a prioritization of small orders. However, it was found that in the load-based POLCA systems the small orders are not always prioritized. For some small to large order ratios, the group of smallest orders even has the largest average waiting time and the group of largest orders has the shortest waiting time. On the contrary, the SPT rule’s main principle is different: it always prioritizes small orders, independent of the available capacity. The results of this study show that the prioritization of the SPT rule achieves a much larger TTT and STT reduction than the load-based POLCA system’s prioritization.

Lastly, the load-based POLCA systems are found to perform best in TTT when the small to large order ratio is 98,5/1,5, 97,5/2,5 or 95/5. This means that 98,5%, 97,5% or 95% of the orders require one card. These load-based POLCA systems thus lies rather close to a unit-based POLCA system. Therefore, it is not surprising that the TTT performance of the load-based POLCA systems also lies close to that of a unit-based POLCA system.

5.1.5 Small to large order ratio

(34)

34

cards in the control loops. Apparently, when only a small amount of orders is considered to be large in the control loops and require two cards, this will positively influence the average TTT.

A possible explanation for this positive influence on the average TTT could be that there are then less orders that take up two cards and hold them for a long time, because these orders’ processing time is relatively large. These orders block the orders that are still waiting for cards. Another factor that could promote the larger order ratio is that then an order can almost always proceed when one card becomes available. If there would be more orders that needed two cards, there could be a higher chance that one card becomes available and there are only orders in the queue that require two cards. This card, which indicates free capacity, will then not be used. Nevertheless, when the small to large order ratio becomes too small, almost all orders will require one card in the control loop and the load-based POLCA system will resemble a unit-based POLCA system. Thus, the TTT reduction will then decrease again.

5.1.6 Level of processing time variability

It was found that when the level of processing time variability increases, the unit-based POLCA system and the load-based POLCA systems in general show slightly smaller TTT reductions. This indicates that the effective workload balancing capability of these systems decreases for a higher level of processing time variability. For the unit-based POLCA system, this result is expected, because this system does not take into account the processing time of orders. Also, it is in line with the studies of Germs and Riezebos (2010) and Ziengs, Riezebos and Germs (2012). The load-based POLCA system does take into account the processing time of orders, but apparantly still cannot handle an increasing level of processing time variability effectively. The systems with an SPT rule, on the contrary, show larger TTT reductions when the level of processing time variability increases. This means that the effective workload balancing capability of these systems increases for a higher level of processing time variability. When more variability in processing time exists, the benefits of the SPT rule become greater.

5.1.7 Performance per order size

(35)

35

be explained by the fact that every time a smaller order enters the queue, it will be put in front of the large order. The larger the order is, the higher the chance is that a smaller order will enter the queue and the large order consequently has to wait. This finding is in line with the statement of Haupt (1989) that a disadvantage of the SPT rule is that individual large orders can have large TTTs. This can have implications for the due date performance of the system.

In load-based POLCA systems, the differences in the average TTT and STT per order size are much smaller than in systems with an SPT rule. Moreover, the average waiting time does not differ that much per order size. Still, it was found that in some systems the group of largest orders had the shortest average waiting time, while in other systems the group of smallest orders had the shortest average waiting time. This means that the implementation of a load-based POLCA system does not necessarily lead to the prioritization of small orders.

An explanation can be found in the card count of the load-based POLCA systems. It was found that in the load-based POLCA systems in which the group of largest orders had the shortest waiting time the optimal card count was rather large. This means that most of the time two cards are available when large orders need them. As a consequence, the average waiting time of the large orders stays relatively small. Besides that, the large orders block the route for the small orders by taking up the two cards and being processed by the workstation for a long time. This causes that the smaller orders have larger waiting times. The load-based POLCA systems in which the group of smallest orders had the shortest average waiting time had a smaller optimal card count than the other considered load-based POLCA systems. Because there are less cards available then, there is a higher chance that there are not two cards available for a large order and the order has to wait. When one card is available, the small orders now do get prioritized. It seems like the load-based POLCA systems in which this happens show a better TTT performance than the load-based POLCA systems in which the small orders do not get prioritized.

5.2 Theoretical implications

(36)

36

A second theoretical implication from this study is that it is shown that also in POLCA systems, load-based systems have a better workload balancing capability than unit-based systems. This finding is in line with the studies of Bodnar (2016) and Wilz (2017), who already found that load-based m-CONWIP systems had an improved workload balancing capability compared to unit-based m-CONWIP systems. Still, the load-based POLCA systems also only show small improvements in TTT reduction compared to the unit-based POLCA systems.

Third, it was found that both unit-based and load-based POLCA system in general have a less effective workload balancing capability when the level of processing time variability increases. This indicates that also the load-based POLCA system does not cope well with an increasing variability in processing times. This was not studied before and the theoretical implication that comes forth from this is that it is questionable whether unit-based or load-based POLCA systems are applicable in MTO firms, in which a lot of variability in processing time is present. Since POLCA systems were specifically designed for MTO firms, this is an important consideration.

Further, it was found that the implementation of an SPT rule leads to a large reduction in TTT. This is in line with previous studies that already showed that the SPT rule decreases the TTT of orders in job shops (Land, 2004). This study now shows that this is also the case in a three-stage divergent MTO system. Moreover, this study found that a push system with an SPT rule has a much better workload balancing capability than a unit-based or load-based POLCA system. When a unit-based POLCA system is implemented in combination with an SPT rule also large reductions in TTT are found, although this reduction can almost completely be attributed to the SPT rule. However, this still implies that the SPT rule then helps the unit-based POLCA system to achieve a much more effective workload balancing capability.

(37)

37

of bins should then not be increased from 2 to 3, as in the study of Bodnar (2016), but to 10 or 15.

5.3 Managerial implications

An important managerial implication that can be taken from this study is first of all that implementing a simple SPT rule in a push system results in a very effective workload balancing capability and brings about a large TTT reduction in a divergent shop floor topology. It can be interesting to implement such a push system with an SPT rule, since the TTT reduction it can achieve is much larger than the TTT reductions that can be achieved by a unit-based or load-based POLCA system. Moreover, the SPT rule is very easy to implement, since workers can just select the smallest order. The disadvantage of systems with an SPT rule, however, is that large orders have a very large TTT. This can have implications for the system’s due date performance.

Further, from a managerial point of view it can be interesting to consider implementing a unit-based POLCA system with an SPT rule. Then, the TTT reduction is the largest and at the same time the WIP on the shop floor is limited. Having a limit on the WIP provides several benefits such as having a better overview of the production process and having more flexibility to cope with required changes in orders. It must be noted that under a coefficient of variation of 0,71 in processing time variability the unit-based POLCA systems with an SPT rule show an optimal card count of infinity. This means that when a unit-based POLCA system with a smaller card count is implemented together with the SPT rule, the TTT reduction is likely to decrease.

Lastly, when an SPT rule is not a desired option, it can be interesting for firms to implement a load-based POLCA system, since it shows a better workload balancing capability and TTT performance than unit-based POLCA systems.

5.4 Limitations

(38)

38

(39)

39

6. Conclusion

POLCA is a pull production system that aims to help MTO firms shorten throughput times. Different than unit-based POLCA systems, load-based POLCA systems and the SPT priority rule take the processing times of orders into account and provide a possible solution to the workload balancing problem of unit-based POLCA systems under a high level of processing time variability. Therefore, in this study unit-based POLCA systems, load-based POLCA systems and unit-based POLCA systems with an SPT rule were investigated. These systems all prioritize orders in a different way, hence the following research question was formulated:

What is the impact of order prioritization on the workload balancing capability of POLCA systems under various levels of processing time variability?

The results of this study show that the order prioritization of the load-based POLCA systems and the systems with an SPT rule result in larger TTT and STT reductions compared to the unit-based POLCA systems, under all considered levels of utilization and processing time variability. Thus, it can be said that these systems have a better workload balancing capability than the unit-based POLCA systems. Further, it was found that the TTT and STT reductions achieved by the load-based POLCA systems remain small, while the systems with an SPT rule achieve large reductions in TTT and STT. It seems that the system’s workload balancing capability becomes more effective when the order prioritization based on orders’ processing times is more accurate.

Moreover, it was found that the systems with an SPT rule can balance the workload more effectively when the level of processing time variability increases. On the contrary, the unit-based and load-based POLCA systems have more trouble balancing the workload when the level of processing time variability increases.

(40)

40

While the systems with an SPT rule show a very effective workload balancing capability, they do have a disadvantage: The TTT performance per order size differs greatly in these systems. The largest orders have a much larger TTT than the smallest orders. This can have implications for the due date performance of the system. Unit-based and load-based POLCA systems, on the other hand, show small differences in TTT per order size.

A first suggestion for future research would thus first of all be to include due date measures in the simulation experiments to analyse the due date performance of the studied systems. Then, a second suggestion for future research would be to study push systems and unit-based POLCA systems in which more complicated priority rules are used that include both processing times and due dates of orders. For example, a priority rule such as the one Thürer, Land, Stevenson, Fredendall, & Godinho Filho (2015) developed could be implemented to analyse how it affects the workload balancing capability of the systems. This more complex priority rule combines load-oriented and time-oriented sequencing and can address the SPT rule’s weak point of having large TTTs for individual orders.

Third, it would be interesting for future research to investigate the workload balancing capability of load-based POLCA systems with a much larger number of bins. Even a load-based POLCA system that can use fractions of cards could be analysed. Then, the system can take into account all different processing times of orders and does not have to make a distinction only between several groups.

(41)

41

References

Bodnar, F. (2016). Card based production planning and control: Investigating workload balancing capability of load-based pull systems. University of Groningen, Faculty of Economics and Business.

Germs, R., & Riezebos, J. (2010). Workload balancing capability of pull systems in MTO production. International Journal of Production Research, 48 (8), 2345-2360.

Haupt, R. (1989). A survey of priority rule-based scheduling. Operations-Research-Spektrum, 11 (1), 3-16.

Heidelberger, P., & Welch, P. (1983). Simulation run length control in the presence of an initial transient. Operations Research, 31 (6), 1109-1144.

Hopp, W. J., & Spearman, M. L. (2004). To Pull or Not to Pull: What is the Question? Manufacturing & Service Operations Management, 6 (2), 133-148.

Hopp, W., & Spearman, M. (2011). Factory Physics, Third Edition. Waveland Press, Inc. Land, M. J. (2004). Workload control in job shops, grasping the tap. s.n.

Land, M. J., & Gaalman, G. J. (1998). The performance of workload control concepts in job shops: Improving the release method. International Journal of Production Economics, 56-57, 347-364.

Riezebos, J. (2010). Design of POLCA material control systems. International Journal of Production Research, 48 (5), 1455-1477.

Robinson, S. (2014). Simulation: The Practice of Model Development and Use. Palgrave Macmillan.

Spearman, M., Woodruff, D., & Hopp, W. (1990). CONWIP: a pull alternative to kanban. International Journal of Production Research, 28 (5), 879.

Stevenson, M., Hendry, L. C., & Kingsman, B. G. (2005). A review of production planning and control: the applicability of key concepts to the make-to-order industry.

(42)

42

Sugimori, Y., Kusunoki, K., Cho, F., & Uchikawa, S. (1977). Toyota production system and Kanban system Materialization of just-in-time and respect-for-human system.

International Journal of Production Research, 15 (6), 553-564.

Suri, R. (1998). Quick response manufacturing: a companywide approach to reducing lead times. Portland: OR: Productivity Press.

Tecnomatix Plant Simulation. (2014). Siemens PLM Software. Retrieved from

https://www.plm.automation.siemens.com/nl_nl/products/tecnomatix/manufacturing-simulation/material-flow/plant-simulation.shtml

Thürer, M., Land, M., Stevenson, M., Fredendall, L., & Godinho Filho, M. (2015). Concerning Workload Control and Order Release: The Pre-Shop Pool Sequencing Decision. Production and Operations Management, 24 (7), 1179-1192.

Vandaele, N., Van Nieuwenhuyse, I., Claerhout, D., & Cremmery, C. (2008). Load-Based POLCA: An Integrated Material Control System for Multiproduct, Multimachine Job Shops. Manufacturing & Service Operations Management, 10 (2), 181-197.

Vollmann, T., Berry, W., & Whybark, D. (1992). Manufacturing Planning and Control Systems. 3rd ed. Irwin: Chicago etc.

Wilz, H. (2017). Workload balancing capabilities of load based pull production control systems - An evaluation of various load based mechanisms using discrete event simulation. University of Groningen.

Zäpfel, G., & Missbauer, H. (1993). Production planning and control (PPC) systems including load-oriented order release - problems and researh perspectives. International Journal of Production Economics, 30 - 31, 107-122.

Ziengs, N., Riezebos, J., & Germs, R. (2012). Placement of effective work-in-progress limits in route-specific unit-based pull systems. International Journal of Production

(43)

43

Appendix A: Example of mechanism unit-based POLCA system

Figure 15 and 16 serve as an example to illustrate how the mechanism of a unit-based POLCA system can lead to an unbalanced workload. In these figures, a simple unit-based POLCA system can be seen. The dashed loops indicate the POLCA control loops. In the example, each control loop has four cards and each card represents one order. The large boxes represent orders with a long processing time and the small boxes represent orders with a short processing time. The colour of the box indicates the final workstation that the order has to go to. First, in figure 15, it can already be seen that several orders are waiting for the large blue order to be processed at workstation B. This causes workstations C and D to be idle. Then, figure 16 indicates what happens when the orders proceed. Since the orders in the queue are handled on a FCFS basis and there are cards available for loop BC, the blue orders keep being processed at workstation B and proceed to workstation C. However, there they have to queue until the large blue order is finished processing. In the meanwhile, workstation D does not have many red orders to process. This illustrates the workload unbalance that can be present in a unit-based POLCA system. Workstation C has a much larger workload than workstation D in figure 16.

(44)

44

The impact of order prioritization on the workload balancing capability of POLCA systems: A discrete-event simulation study