Suitable Dispatching Rules for Increasing Delivery Reliability in a CONWIP-Controlled Make-to-Order Job Shop: a Simulation study.

(1)

Suitable Dispatching Rules for Increasing

Delivery Reliability in a CONWIP-Controlled

Make-to-Order Job Shop: a Simulation study.

by O.J. Heerink

Student number: 2369664

University of Groningen

MSc. Technology and Operations Management

Supervisor: dr. N.D. van Foreest

Co-assessor: N. Ziengs, MSc.

(2)

2

Abstract

In order to retain and attract new customers, customer satisfaction has become one of the most important factors for many companies. Fulfilling promises by delivering the right products before the right date is crucial for sustaining customer satisfaction. In order to achieve this on-time-delivery, the decision of when and what to produce is essential. One of the methods that determine when an order is being produced is Constant-Work-In-Process (CONWIP). This is a widely used method that ensures a stable workload on the shop floor by allowing a constant level of Work-in-Process (WIP). Although the implementation of CONWIP results in advantages that can be highly favorable for a company, literature shows that CONWIP seems to have some issues with respect to finishing products in time. Hence, this research focuses on minimizing tardy jobs at a CONWIP system through the identification of suitable dispatching rules. The performance of CONWIP namely highly depends on the dispatching rules that are implemented at the order pool and on the shop floor.

The results of this paper show that CONWIP can minimize the amount of tardy jobs by combining it with either EDD or EODD. Herein, the CONWIP level determines which of the two dispatching rules should be implemented to result in a maximum delivery reliability. This research provides first insights into the synergetic effects of the CONWIP release method and dispatching rules. These insights can help companies to determine their CONWIP level and apply the right dispatching rules to maximize their delivery reliability and, hence, their customer satisfaction.

(3)

3

1 PREFACE

During this research and the writing of my thesis I received a lot of support from different people. I would like to use this opportunity to thank them for their help.

First of all, I would like to thank my parents and brother, from whom I always felt a very close involvement and interest during my studies. They let me believe in myself and this supported me in the writing of this thesis as well. Besides, I would like to thank my cousin, Marc. The discussions that we had about my thesis thoughts and progression functioned as great input.

Subsequently, I would like to thank the other members of my thesis group, Pernella and Jeff. We had a lot of fruitful discussions with our group and they provided me with helpful feedback on my research proposal and thesis draft. The project was supervised by dr. N.D. van Foreest. I want to thank him for all the support, feedback, and guidance that he provided. Furthermore, he was always available to answer questions and provide relevant insights at any time throughout the entire project.

Last but not least, I want to thank Neways Leeuwarden. They ensured that I received the right data in-time and provided all the information that I needed to complete my thesis. This

(6)

6

2 INTRODUCTION

In today’s era, customers are able to explore, research, and share every purchase decision they make. As a consequence, companies lose and attract customers more rapidly than ever before. In order to retain and attract new customers, customer satisfaction has become one of the most important factors for many companies. Fulfilling promises by delivering the right products before the right date is crucial for sustaining customer satisfaction (Patil, 2010). This includes that the company which is able to provide the desired quality in the shortest time is likely to attract most customers. But how does a company pursue this strategy effectively? Prior research has shown that delivery speed and reliability are, in part, affected by dispatching rules (Barman and Laforge, 1998). Dispatching rules are a workload control concept, which are specifically developed for high-variety contexts, such as make-to-order (MTO) companies with a job shop configuration (Zäpfel and Missbauer, 1993; Thürer, Qu, et al., 2014; Thürer, Stevenson, et al., 2014).

A dispatching rule is a scheduling method, which determines the next job in queue that is being ‘dispatched’ for processing (Brown, Dimitrov and Barlatt, 2015). Over the years, many different dispatching rules have been examined. Since the performance of a dispatching rule depends on many factors, (e.g. the scenario, system conditions, and the criterion that is intended to be improved (Heger et al., 2016)), there is no optimal dispatching rule to be designated. In addition, Thürer et al. (2012) show that a dispatching rule’s performance is influenced by the moment the next job is being forwarded from the order pool. This moment is determined by the release method. Hence, Thürer et al. (2012) state that a dispatching rule and a release method should complement each other, in order to serve the performance objective.

As with dispatching rules, there are also many release methods to identify. However, especially in high-variety environments, the constant work-in-process (CONWIP) release method seems to have many advantages over other release methods. Firstly, CONWIP tends to collect jobs at the bottleneck. Therefore, CONWIP has a tendency to produce relative high utilization, which leads to a higher throughput (Spearman, Woodruff and Hopp, 1990; Huang

et al., 2015). Second, through the constant amount of work-in-process (WIP), a stable

(7)

7 and shifting bottlenecks, this means CONWIP can be favorable in increasing delivery reliability at an MTO job shops.

Since CONWIP has several advantages in complex environments, it is highly relevant to identify dispatching rules that complement CONWIP for an MTO job shop. MTO job shops namely are considered to have a high-variety environment. Besides, customer satisfaction at MTO job shops is majorly impacted by delivery speed and reliability (Soepenberg, Land and Gaalman, 2012). Hence, the research question that guides this research is: ‘Which

dispatching rule(s) in combination with CONWIP is/are suitable for delivery reliability maximization in a make-to-order job-shop?’ First, a literature study is performed to identify

feasible dispatching rules for the context of the research. Subsequently, a simulation study is executed to measure the performance for each dispatching rule that is identified. The results of this study will contribute to literature and practice by providing dispatching rules, which maximize delivery reliability in CONWIP MTO job shops. This is of great relevance, since delivery reliability has become a major aspect in MTO companies (Soepenberg, Land and Gaalman, 2012). It helps companies to deal with the increased competitive pressure that originates from the growing importance of supply chain effectiveness and efficiency (Sridharan and Li, 2008).

(8)

8

3 THEORETICAL BACKGROUND

3.1 Release method

The core purpose of release methods is to control the workload on the shop floor. Release methods can be classified by their determination of the release decision moment. This can be either continuously (i.e., at any moment in time, commonly started by an event on the shop floor) or periodically (i.e., at fixed intervals) (Thürer, Qu, et al., 2014). The CONWIP release method can be grouped among the ‘continuous’ release methods, since it triggers an order release when the WIP becomes less than a predetermined level (Thürer, Qu, et al., 2014). Hereby, the jobs are ‘push-produced’ once they have entered the line, although it is typically positioned as a pull system (Framinan, Ruiz-Usano and Leisten, 2000). Prior research indicates that CONWIP has the same benefits as Kanban (e.g. shorter flow times and inventory level decrease) (Spearman, Woodruff and Hopp, 1990). Similar to Kanban, CONWIP is a pull system and makes use of cards to control the WIP. However, the cards are attached to a job at the beginning of production. When the job is finished, its card is attached to the next job to be released. It is not allowed to start the production of a job without a card (Framinan, Ruiz-Usano and Leisten, 2000). The essence of the CONWIP release method is shown in Figure 3.1. Station 1 Station 2 Station 3 Station 4 Material flow Order signal Order done? Start new order

Figure 3.1 – CONWIP release method

(9)

9 lower WIP levels than Kanban with the same throughput in most cases. In short, if Kanban is good, CONWIP is better (Huang, Wang and Ip, 1998).

3.2 Dispatching rules

Dispatching rules, also known as priority rules, are a commonly used scheduling method, of which most are human implementable. In a job-shop, the production consists of jobs (or orders) that need to be completed by one or more specific machines, before a specific moment. This moment is called the ‘due date’. If a job exceeds the due date, the job is labeled ‘tardy’. At each machine, jobs in the queue are waiting to be processed. A dispatching rule is used to assign a job in queue to be the next to process after a job is finished (Brown, Dimitrov and Barlatt, 2015). Hundreds of dispatching rules are already composed throughout the years (e.g. Panwalkar and Iskander, 1977; Blackstone, Phillips and Hogg, 1982), differing from simple rules (e.g. first-come-first-served (FCFS)) to refined rules (e.g. due date oriented rules) (Land, Stevenson and Thürer, 2013). Nonetheless, through the literature study, only several dispatching rules will be defined for simulation purposes. Only those rules that are presumed to lead to an increase of delivery reliability together with CONWIP will be selected.

3.3 Delivery performance

Improving performance with respect to delivery reliability is becoming more and more essential for MTO companies (Soepenberg, Land and Gaalman, 2012). Hence, short and reliable throughput times are an important competitive advantage for MTO companies (Ziengs, Riezebos and Germs, 2012). In this research, delivery reliability is defined as the percentage of orders that are finished in time. To determine this reliability, the percentage of orders that are delivered too late (tardy) are observed. However, also the delivery speed is taken into account in the determination of delivery reliability. In this research, delivery speed is measured by the average tardiness. Tardiness is defined in this research as the difference between the promised delivery time and the realized throughput time of an order (Soepenberg, Land and Gaalman, 2012).

Hence, our research question can be divided into two separate questions, being: ‘Which

(10)

10

3.4 Assessment of the literature

CONWIP studies have been executed for more than twenty years. This research is performed in performance evaluation, modelling, parameters solving, and model improvement (Huang et

al., 2015). In order to be able to identify suitable dispatching rules, the scope of the literature

review does not restrict to CONWIP. This provides a broad basis for the identification of potential dispatching rules for CONWIP. Provided that dispatching rules and release methods affect each other’s performance and outcome, the question is how dispatching rules perform in co-existence with CONWIP.

Many dispatching rules have been used throughout the years, differing from simple dispatching rules to very complex ones. Examples of dispatching rules are: First In First Out (FIFO), First In Last Out (FILO), Shortest Processing Time first (SPT), Longest Processing Time first (LPT), Earliest Completion Time first (ECT), Weighted Shortest Processing Time first (WSPT), With biggest weight (WI), Work In Next Queue (WINQ), Earliest Release Date first (ERD), Earliest Due Date first (EDD), Earliest Operational Due Date first (EODD) and Least Slack first (LS) (Jayamohan and Rajendran, 2000; Kacem and Chu, 2008; Xiong et al., 2017). All these dispatching rules have different characteristics and perform in another way. Some, for example, aim to align the workload (Fernandes, Land and Carmo-Silva, 2016); others perform well with minimizing maximum completion time (Li, Freiheit and Miao, 2016). Hence, the different dispatching rules can be considered within four different categories, being: ‘Process-time-based rules’, ‘Due-date-based rules’, ‘Combination rules’ and rules that are neither process nor due date based (Holthaus and Rajendran, 1997b). This research only focusses on the dispatching rules that, according to literature, have the tendency to positively influence the delivery reliability. This can be done through the minimization of tardy jobs.

(11)

11 but they concluded that the ‘Earliest Operational Due Date’ (EODD) performed best. The EODD rule prioritizes jobs that need to be finished first at the specific operation (Patil, 2010; Thürer, Stevenson, et al., 2014; Beemsterboer et al., 2017).

However, what should be noticed is that Holthaus and Rajendran (1997) draw their conclusion by only taking into account the criterion that is intended to be improved: minimization of tardy jobs. Brown, Dimitrov and Barlatt (2015) amplify on this by taking scenarios into consideration: routings. Since a job shop consists of different routings, the findings of Brown, Dimitrov and Barlatt (2015) are more interesting.

Just as EODD, another due-date-based rule is the ‘Earliest due date’ (EDD). This rule also has the tendency to improve delivery reliability (Patil, 2010). The difference between the EODD and EDD rule is that EODD takes into account the due date and processing times at the specific workstation, where EDD only sequences on an ascending due date. However, both Holthaus and Rajendran 1997 and Brown, Dimitrov and Barlatt (2015) did not elaborate on the release method that was used.

In comparison to Holthaus and Rajendran 1997 and Brown, Dimitrov and Barlatt (2015), Thürer et al. (2012) do take into account the release methods involved. What’s more, they even analyze the CONWIP release method in combination with dispatching. Their results show that CONWIP is outperformed by e.g. immediate and periodic release methods when aiming for a minimization of late-delivered jobs, i.e. tardy jobs. Subsequently, they conclude that CONWIP seems to be least suitable for increasing delivery reliability. However, Thürer

et al. (2012) analyzed the CONWIP release method in combination with only one dispatching

(12)

12 Framinan, Ruiz-Usano and Leisten (2000) also studied the influence of different dispatching rules in a CONWIP system. In the end, they conclude that CONWIP in combination with SPT results in the least amount of tardy jobs in a flow shop environment. However, the big difference of the article of Framinan, Ruiz-Usano and Leisten (2000), in contrast to this study, is that they based their findings on solely a make-to-stock (MTS) environment. Since MTS production is very different from MTO production, the performance of dispatching rules between the two production strategies differs (Beemsterboer et al., 2017). This means that the conclusion of Framinan, Ruiz-Usano and Leisten (2000) cannot be generalized to a job shop that produces both MTS and MTO. Nevertheless it places even greater emphasis on the implementation of the SPT rule in the simulation since it complements the conclusion of Holthaus and Rajendran (1997).

Besides dispatching rules, companies can also choose to make use of algorithms for their production scheduling. Dispatching rules are generally outperformed by these algorithms (e.g. the shifting bottle neck algorithm or constraint programming methods) (Brown, Dimitrov and Barlatt, 2015). Those algorithms namely take into account future activities and resources, resulting in all-round production schedules. However, as mentioned before, MTO job shops are subject to a high-variety environment, which forces them to cope with rapid and unpredictable changes (McKay, Safayeni and Buzacott, 1988; Brown, Dimitrov and Barlatt, 2015). Hence, MTO job shops are not stable enough to benefit from algorithms. In contrast to algorithms, dispatching rules do not create complete schedules, but solely determine which job a machine or resource is determined to process (Brown, Dimitrov and Barlatt, 2015). Because of this, MTO job shops are more flexible and can more easily consort with changes. Likewise, this is seen as an important feature by Tung Dang (2013) and Heger et al. (2016).

3.5 Suitable Dispatching rules for CONWIP

(13)

13 The operational due dates for the use of the EODD rule can be calculated at specific operations through substracting the processingtime that is left from the final due date.

Eventually, also the ‘First-Come-First-Serve’ (FCFS) and the ‘Least Slack’ (LS) rule are simulated, in order to be able to compare all dispatching rules to respectively the most fundamental one and the one that our industry partner is currently using for their order sequencing. The FCFS-rule gives priority to the job which arrived at the station first (Panwalkar and Iskander, 1977; Land, Stevenson and Thürer, 2013). The LS-rule, on the other hand, provides priority to the jobs that have the least amount of slack. Slack is calculated by subtracting the current time and processing time of a specific job from the due date (Raghu and Rajendran, 1993).

3.5.1 EXAMPLE SITUATION

In this section, further explanation is provided about the analyzed dispatching rules. To give a clear view of the different dispatching rules, an example situation is discussed. This example consists of four different orders and three different operations. The processing times for orders 1, 2, 3 and 4 are provided in Table 3.1.

Order Processing time at Operation 1

Processing time at Operation 2

Processing time at Operation 3

1 2 days 1 day 1 day

2 1 day 3 days 4 days

3 1 day 1 day 1 day

4 2 days 4 days 2 days

Table 3.1 – Example: Processing times FCFS

The FCFS rule gives priority to the order that arrives first and, hence, is the longest in queue. The sequencing according to the FCFS rule does not take into account any order-specific characteristics. This is the case at the SPT, EDD, EODD and LS rules.

SPT

The SPT rule gives priority to the rule that has the shortest total processing time. The total processing time is calculated by taking the sum of the processing times at all operations.

(14)

14

Order Total processing time SPT sequence 1 4 days 2 2 8 days 3 3 3 days 1 4 8 days 3

Table 3.2 – Example: SPT rule EDD and EODD

The EDD rule prioritizes jobs according to an ascending final due date. Which job needs to be finished first, is processed first. The EODD rule prioritizes jobs that have the earliest operation due date. This operation due date is the due date of the job’s current operation. This means that the operation due date of a job will change as operations are processed. This rule proceeds like EDD, so the job with the earliest operation due date is selected for processing. The operation due dates are calculated recursively from the final due date, by subtracting the processing times of the remaining operations. This means that the due date at operation 1 is calculated by subtracting the processing times of operation 3 and operation 2 from the final due date. The due date at operation 2 is calculated by only subtracting the processing time of operation 3 from the final due date. Finally, the due date of operation 3 is equal to the final due date, since this is the last operation of the process. This results in the operation due dates and corresponding sequences shown in Table 3.3.

Order Due date Op. 1 Due date Op. 2 Due date Op. 3 Final due date EDD sequence EODD sequence Op. 1 Op. 2 Op. 3 1 12-07-2017 13-07-2017 14-07-2017 14-07-2017 2 4 3 2 2 09-07-2017 12-07-2017 16-07-2017 16-07-2017 3 1 2 3 3 10-07-2017 11-07-2017 12-07-2017 12-07-2017 1 2 1 1 4 11-07-2017 15-07-2017 17-07-2017 17-07-2017 4 3 4 4

Table 3.3 – Example: EDD and EODD rules

(15)

15

LS

The LS rule prioritizes orders according to an ascending timespan between the current time and the order-specific due date. Herein, the remaining processing time of an order is also taken into account. Hence, the slack of an order is calculated by: Final due date – Current time – Processing time. To give a good representation of the least slack rule, the slack for each order is calculated at the same time for the different operations. The slack at operation 1, 2 and 3 is calculated respectively on, 07-07-2017, 09-07-2017 and 13-07-2017. If the slack of an order is equal or more than zero, the order is in time. If the slack of a job is less than zero, the order will be finished too late (tardy). The prioritization of orders according to the LS rule is shown in Table 3.4.

Order Slack Op. 1

(07-07-2017) Slack Op. 2 (09-07-2017) Slack Op. 3 (13-07-2017) Final due date Slack sequence

Op. 1 Op. 2 Op. 3

1 3 days 3 days 0 days 14-07-2017 3 4 3

2 1 day 0 days -1 day 16-07-2017 1 1 2

3 2 days 1 day -2 days 12-07-2017 2 2 1

4 2 days 2 days 2 days 17-07-2017 2 3 4

Table 3.4 – Example: LS rule

(16)

16

4 METHODOLOGY

4.1 Model of the situation

The research is performed at Neways Electronics, Leeuwarden. Neways mainly develops, produces, and assembles printed circuit boards and (micro)-electronics. To perform a simulation study, an effort is made to represent the situation of Neways in a model. Setup times should be taken into account, since it is expected that setup times influence the performance. Furthermore, not all orders at Neways will be processed the same day that they arrive. The fact that a production day starts with orders in queue from previous days is expected to also influence the situation.

4.1.1 SHOP FLOOR

The model that is simulated is based on three different routings. These routings are determined by analyzing the order flow on the shop floor at our industry partner. In Figure

4.1, the simulated shop floor is shown.

SMD

Tinbath

Post-assembly Testing

Defense

Orderpool Coating Expedition

Figure 4.1 - Shop floor

4.1.2 ROUTINGS

The model in figure 2 represents all main routings at the shop floor of our industry partner. The routings that are implemented are shown in Figures 4.2, 4.3 and 4.4.

SMD

Tinbath

Defense Orderpool

(17)

17

Tinbath assemb

Post-ly

Testing

Figure 4.3 - Routing 2: Tandem

Testing

Defense

Coating Expediti on

Figure 4.4 - Routing 3: Convergent

The simulation will consist of experiments in which dispatching rules are applied in front of every routing. In this way, the performance of specific dispatching rules can be linked to the specific routings the shop floor consists of. This means that there is a total of three dispatching control points, namely: from the order pool (1), in front of the tinbath (2) and in front of testing and defense (3).

4.2 Simulation study

4.2.1 SOFTWARE

The simulation study is performed using Technomatix Plant Simulation 13. This is a computer application developed by Siemens PLM Software for modeling, simulating, analyzing, visualizing and optimizing production systems and processes, the flow of materials and logistic operations. Through the use of this application, a CONWIP environment can be accurately simulated with the different dispatching rules. This allows us to analyze the process in a reliable way, conduct results and, eventually, draw conclusions. Screenshots of the simulation model can be found in Appendix 1.

4.2.2 ADVANTAGES AND DISADVANTAGES

(18)

slow-18 down the simulated process, which supports the analysis. Third, clear insight can be achieved about interaction between the dispatching rules and CONWIP. This leads to a better understanding of the real-world system and provides a good indication of the actual performance. On the other hand, a disadvantage of the simulation is the interpretation of the simulation results. Results of simulation can namely be hard to interpret in the right way. Besides, simulation modeling and analysis can be very time consuming.

4.2.3 RUN LENGTH

The run length of each simulation equals the total cycle time of the 2000 orders from the order pool. The order amount is set on 2000, since this provides a reliable basis for the results analysis. The CONWIP level and dispatching rules influence the throughput times of these orders. Hence, the run length differs per experiment between approximately 75 and 100 days.

4.2.4 WARMUP AND COOLING DOWN TIME

The warm up and cooling down time of the simulation are based on the throughput of orders on the shop floor (see graph in Appendix 2). Since throughput times become stable after approximately six days, a warm up time of 8 days is taken to include a safety margin. There seems to be no significant decrease or increase in throughput times at the end of the simulation. Besides, the differences regarding the throughput times of the last orders are likely to only affect the results of the SPT-rule in a minimal way. Hence, no cooling down time is defined. This point is also discussed in section 7.2.

4.2.5 NUMBER OF RUNS

The number of runs is determined by analyzing changes in the outcomes of the KPI’s, being number of tardy jobs and average tardiness. The relative changes compared to doing only one run are shown (see Appendix 3), as well as the relative changes compared to doing one run less. From the analysis it becomes clear that outcomes are stable at approximately twenty replications. In order to be sure that outcomes are reliable, a safety margin of five runs is applied. Hence, the number of replications is determined to be 25.

4.2.6 PROCESSING TIMES AND DUE DATES

(19)

19

Processing times

The processing times of all orders are distributed in such a way that a clear distinction can be made between orders with a short processing time and orders with a long processing time. This is essential for the performance measurement of dispatching rules that are based on the processing time of orders; the LS and SPT rule in this research. All orders were assigned a processing time in between three and six hours, with an interval of five minutes.

At our industry partner, the processing time of all orders is based on two factors: the batch size and the amount of components a circuit board needs. The processing time of orders at the different workstations depends on if the workstation’s processing time is affected by one or both of these factors. Through this, the average portion an order spends at that workstation can be calculated. The time an order spends at the SMD-line, post-assembly, and testing is all based on both factors. The time an order spends at the tinbath and the coating department is based on only one factor: the batch size. Subsequently, the defense department processes all steps of tinbath, post-assembly, and testing for the defense orders. Hence, defense is assumed to have an equal processing time of the three stations combined, being 5. Finally, expedition is only a small part of the process. Hence, the factor that is assigned to this department is 0.5. This results in a total of 8.5 factors per routing. The processing time at the SMD-line, for example, therefore is 2 / 8.5 times the processing time of the order. Hereby, the processing time proportions for the different routings approach the real situation of our industry partner.

Due dates

Besides processing times, all 2000 orders also need a due date. To measure the KPI’s in a clear way, the due dates should be somewhat challenging, taking into account the processing times. Hence, it is determined that the orders need to be finished in between 1 week and 2 months. These due dates are randomly assigned to all orders as well.

All order characteristics are summarized in Appendix 4. An overview of the processing times and due dates that are randomly assigned to all orders is shown in Appendix 5.

4.2.7 SIMULATION OF DISPATCHING RULES WITH CONWIP

(20)

20 The different dispatching rules that will be simulated are summarized in Table 4.1.

Abbreviation Full name Brief description

LS Least slack The job which has the minimum amount of slack in between the due date and its processing time is chosen from the queue.

SPT Shortest

processing time

The job which has the shortest processing time is chosen from the queue.

EDD Earliest due

date

The job which has to be finished at the earliest moment is chosen from the queue.

EODD Earliest operational due date

The job which has to be finished at the earliest moment at the specific operation is chosen from the queue.

FCFS First come first serve

The job which arrived first is chosen from the queue.

Table 4.1 – Identified dispatching rules for simulation

These dispatching rules are simulated for different CONWIP levels. The range in which CONWIP is analyzed embraces levels between 20 and 200, with an interval of 20. Since our simulation model approaches the maximum utilization of the bottleneck and throughput at CONWIP level 120, the range of CONWIP 120-200 is referred to as high and the range of CONWIP 20-120 is referred to as low.

4.2.8 ASSUMPTIONS

During the simulation study, assumptions were made. These assumptions are listed below. - It is assumed that the orderflow from SMD to its two successors is equally divided. - We assume an operation cannot be preempted. This means that, once processing

begins on an operation, it cannot be stopped until complete. - Each job can be processed on only one machine at a time. - Each machine can process only one operation at a time. - All stations do not have any setup time.

- Transportation time of jobs between machines is negligible. Hence, it is not implemented in the simulation study.

- The breakdowns of machines is not taken into account.

(21)

21

4.2.9 SIMULATION KPI’S

At the simulation study, the following outcomes serve as key performance indicators of delivery reliability:

- The throughput times - The amount of tardy jobs - The average tardiness

Firstly, the throughput (TH) for every scenario is measured. The throughput, or throughput rate, embraces the average output of a production process per time unit (Hopp and Spearman, 2008, p. 229). Since throughput depends on the level of work in process, the throughput is assumed to be positively correlated with the CONWIP level. Besides, the throughput can also differ for every dispatching rule combination, because of the convergent routing. The analysis of TH enables us to identify which dispatching rule scenario ensures the maximization of average production speed. Hereafter, the amount of tardy jobs is measured, since delivery reliability is directly determined through this. This KPI will, in the end, be the major indicator for the identification of the dispatching rules that lead to maximization of delivery reliability in combination with CONWIP. However, in order to get more insight in the details of the tardy jobs, the delivery speed is also taken into account. The delivery speed will be analyzed by the throughput times and the average tardiness (time by which tardy jobs exceeded the due date). Solely the amount of tardy jobs does namely not clearly display the distribution of the produced jobs. Besides, for a comprehensive analysis of delivery reliability, the average number of tardy jobs and average lateness of orders, should be complemented by an analysis of the variance (Soepenberg, Land and Gaalman, 2012).

4.2.10 VALIDATION OF THE MODEL

In order to come up with reliable results and conclusions, the model needed to be validated. First, the behavior of the dispatching rules is analyzed. This is done by observing the sequence in which orders flow through the routings. It seemed to be the case that all dispatching rules behaved as they should; e.g. the SPT rule sequenced orders on an ascending processing time and the EDD rule sequenced orders on an ascending due date.

Second, the flow of orders was analyzed. Since an equal amount of orders is assumed to flow through the tinbath routing as through the defense routing, approximately 50 percent should also do that. This is verified by checking the amount of jobs for every routing.

(22)

22 of items to the system, λ, multiplied by the average waiting time of an item in the system, W (Little, 2011). Thus, L = λW. Since we implemented a CONWIP system, little’s law should be approximately equal to the specified CONWIP level of each experiment. This was analyzed for every CONWIP level between 10 and 200 for an interval of 20. Little’s law showed approximately 1 to 5% difference with the CONWIP level at each experiment (see

(23)

23

5 RESULTS AND INTERPRETATIONS

In this section, the results of all simulated experiments are shown. The results embrace the analysis of the two key performance indicators, being the number of tardy jobs and the

average tardiness. These performance indicators are analyzed as a function of throughput

realization. This is done, because CONWIP levels influence the throughput of the system. CONWIP levels can be very different for each company. In one company a CONWIP level of e.g. 50 ensures a maximum throughput of the bottleneck, where at another company a CONWIP level of 500 is needed to ensure this. Hence, it is hard to interpret the results as a function of the CONWIP level for a wider context. However, when the corresponding throughput of the CONWIP level is shown, the results will be good to interpret for differing scenarios.

All results are compared with the implementation of the LS-rule, since this is the dispatching rule that our industry partner currently uses. The simulated experiments are separated in a clear and distinct way. Initially, the implementation of one single rule for both order pool and shop floor is simulated. Hereafter, different combinations of two rules are addressed. All different combinations are explained in section 5.2. All these combinations are analyzed with respect to the averages and their variance.

For the identification of the most suitable rules for improving delivery reliability, the variability of the averages is also taken into account. This means that, despite the fact that the average results of a dispatching rule can be promising, this does not necessarily mean that it is suitable. If the results of a rule have a relatively high level of variance, its instability makes it harder to control the production process and therefore the delivery reliability. The decision if a dispatching rule is suitable is also influenced by another factor: the easiness of implementation. Hence, the implementation of one single rule at the order pool and shop floor is more favorable than the implementation of a combination of rules at different routings.

5.1 Throughput

(24)

24 higher CONWIP levels than 140. This indicates that this amount of cards enables a maximum utilization of the bottleneck.

Figure 5.1 – Throughput per hour for CONWIP levels 20 -200.

The throughput per hour is analyzed for the implementation of a single dispatching rule. The graph in Figure 5.1 shows that the throughput of orders per hour is approximately the same in all four situations. The throughput related to all discussed combinations of dispatching rules show the same results. All single dispatching rules and combinations start at a throughput of approximately 0,86 orders per hour at a low CONWIP level, and approach 0,95 orders per hour at a high CONWIP level.

5.2 Performed experiments

This section discusses all experiments that are performed. These experiments consist of four combinations that are clearly distinguishable, being: single rules, SPT-rule combinations, FCFS-rule combinations and due date-rule combinations. All experiments consist of a maximum of two different dispatching rules. For the analysis of the implementation of a single rule in the order pool and on the shop floor, the scenarios in Table 5.1 are simulated.

Scenario Order pool

SF1 SF2

1 LS LS LS

2 EDD EDD EDD

3 EODD EODD EODD

4 SPT SPT SPT

5 FCFS FCFS FCFS

(25)

25 For the analysis of the implementation of a due date rule in the order pool and the SPT-rule or FCFS-rule on the shop floor, the scenarios in Table 5.2 and Table 5.3 are simulated.

Scenario Order pool SF1 SF2 1 LS SPT SPT 2 LS SPT LS 3 LS LS SPT 4 EDD SPT SPT 5 EDD SPT EDD 6 EDD EDD SPT 7 EODD SPT SPT 8 EODD SPT EODD 9 EODD EODD SPT

Table 5.2 – SPT-rule combinations

Scenario Order pool SF1 SF2 1 LS FCFS FCFS 2 LS FCFS LS 3 LS LS FCFS 4 EDD FCFS FCFS 5 EDD EDD FCFS 6 EODD FCFS EODD 7 EODD EODD FCFS

Table 5.3 – FCFS-rule combinations

Since FCFS does not change anything to the sequence of orders in which they arrive at a station, the combination of EDD/FCFS/EDD is left out. The EDD/FCFS/EDD rule namely shows the same results as the implementation of solely EDD, since orders would arrive at the tandem routing already sequenced according to an ascending due date. This also applies to the EDD/FCFS/FCFS combination, showing the same results as the EDD/EDD/FCFS combination. For the analysis of the implementation of a due date rule in the order pool and a due date rule on the shop floor, the scenarios in Table 5.4 are simulated.

Scenario Order pool

SF1 SF2 Scenario Order

pool

SF1 SF2

1 LS EDD EDD 10 EDD LS LS

2 LS LS EDD 11 EDD EDD LS

3 LS EDD LS 12 EDD LS EDD

4 LS EODD EODD 13 EODD LS LS

5 LS LS EODD 14 EODD EODD LS

6 LS EODD LS 15 EODD LS EODD

7 EDD EODD EODD 16 EODD EDD EDD

8 EDD EDD EODD 17 EODD EODD EDD

9 EDD EODD EDD 18 EODD EDD EODD

(26)

26

5.3 Number of tardy jobs

The results of the implementation of one single rule regarding the average number of tardy jobs are shown in Figure 5.2. The results show that there is a lot of difference between the performances of the various dispatching rule categories. Both FCFS and SPT seem to perform a lot worse than the due date dispatching rules regarding number of tardy jobs. However, what is striking is that this is not the case for very low levels of CONWIP. The due date rules show a very low and stable number of tardy jobs for CONWIP levels above 40, but they perform worse at lower levels than the rules that do not take into account due dates. Both the FCFS and SPT rule have significantly lower averages at the small CONWIP levels, with SPT performing better than FCFS. However, the number of tardy jobs does not decrease that much for both rules throughout the CONWIP levels.

Figure 5.2 – Number of tardy jobs at single rule implementation

To get a better insight in the performance of the due date rules, Figure 5.3 shows a close up.

(27)

27

Figure 5.3 - Number of tardy jobs at single rule implementation (Close up)

The close up of the due date rules shows that the EDD-rule performs somewhat better than the LS-rule and the EODD-rule for relatively low CONWIP levels. For the higher CONWIP levels, the performance of EDD and LS is approximately equal, with LS performing slightly better. The EODD-rule, however, shows great results regarding the relatively high levels of CONWIP. It seems to reduce the number of tardy jobs with approximately 50%.

The rules which do not take into account the due date of orders, perform relatively bad at all CONWIP levels regarding the number of tardy jobs. This can be related to the implementation of these dispatching rules in the order pool. Through the interpretation of the results above, and logical reasoning, the rule that sequences order in the order pool should be due date related to be suitable for delivery reliability maximization. Without a due date-related dispatching rule in the order pool, there will simply be a lack of coordination with respect to due date fulfillment. Hence, the SPT and FCFS rules are only implemented on the shop floor. Next, the analysis of the rule combinations is performed. The results regarding the SPT-rule combinations show that the EDD/SPT/SPT performs relatively well. This combination sequences orders in the order pool according to an ascending due date. In this sequence, orders also flow through the divergent routing. In front of the tandem routing, orders are sequenced according to ascending processing times. This also applies to the sequencing at the convergent routing. Implementing the EDD rule at the convergent routing instead of the SPT rule seems to perform slightly better at a relatively low CONWIP level. However, this difference is negligible and does only apply for a very small interval of CONWIP levels.

(28)

28 Subsequently, the results regarding the FCFS-rule combinations show that the EDD/EDD/FCFS combination is performing relatively well for the lower CONWIP levels. This rule sequences orders in the order pool according to an ascending due date. In this sequence, orders also flow through the divergent routing and the tandem routing. However, the improvement that this combination provides seems to be negligible and only applies to a small interval of CONWIP levels. Besides, it performs very badly with respect to high CONWIP levels. Other FCFS-rule combinations do not show any improvement for low or high CONWIP levels at all.

Eventually, the due date-rule combinations are analyzed. The results show that the EDD/EDD/EODD is obviously performing best for all CONWIP levels. This combination sequences the orders in the order pool according to an ascending due date. In this sequence, orders also flow through the divergent routing. At the tandem routing, orders are again processed according to an ascending due date. However, at the convergent routing the orders are not processed according to an ascending due date, but according to an ascending operation due date. It is remarkable to see the relatively high difference with the other combinations of due date rules. The results of all analyzed rule combinations with respect to the number of tardy jobs can be found in Appendix 6.1.

5.3.1 COMPARED TO LS-RULE

In Figure 5.4, the relative differences regarding least amount of tardy jobs are shown between the rules that were identified to be best performing and the LS-rule.

(29)

29 When the best performing rules are compared to the LS-rule, one of the rules that really stands out at the minimization of tardy jobs is the EODD rule. This single rule shows a decrease of approximately 40% in amount of tardy jobs when CONWIP levels enable a maximum throughput. Besides, the results of the EODD-rule also show less variance than the results of the LS-rule (Appendix 7). A second rule that stands out is the combination of EDD/EDD/EODD. This combination shows a little increase in amount of tardy jobs compared to the LS-rule for the lower CONWIP levels, but shows great results for medium CONWIP levels that approach maximization of throughput. It decreases the number of tardy jobs with more than 40% on its optimum. The EDD rule seems to perform better than the LS-rule for some levels of CONWIP, but has more variation in its averages and therefore shows very instable behavior. The EDD/SPT/SPT and the EDD/EDD/EODD combinations show a better overall performance than the EDD rule, although these rules also show a lot of instability and are outperformed by the EODD-rule at relatively high CONWIP levels. To gain more insight in the variance of the results, the averages of the identified rules that minimize amount of tardy jobs including the standard deviations are shown in the next figures.

Figure 5.5 – Standard deviation EDD Figure 5.6 - Standard deviation EDD /SPT/SPT

(30)

30

Figure 5.7 shows that the EDD/EDD/EODD combination has a relatively high variance for all

CONWIP levels. As already discussed, the averages also fluctuate much for different CONWIP levels. These fluctuating averages and high levels of variance make the EDD/EDD/EODD combination very unpredictable and uncontrollable. The instable behavior of the EDD rule is also partly represented in its standard deviation shown in Figure 5.5. The EDD/SPT/SPT and EODD rules, on the other hand, show much less variance for the higher CONWIP levels. These rules also perform a lot better regarding minimization of tardy jobs compared to the LS-rule. The averages of the amount of tardy jobs including standard deviations for the LS-rule can be found in Appendix 7.

5.4 Average tardiness

Where the results in the previous section showed that the FCFCS and SPT rule performed better than the due date rules regarding the very small levels of CONWIP, they do not anymore with respect to the average tardiness. The average tardiness for both FCFS and SPT rules are very high for all CONWIP levels compared to the due date rules. When we compare the FCFS and SPT rule, the SPT rule, again, seems to perform slightly better than the FCFS for all CONWIP levels. To get a better insight in the performance of the due date rules, a closeup of these rules is shown in the graph below. The full graph including the FCFS and SPT rule can be found in Appendix 6.1.

Figure 5.9 – Average tardiness at single rule implementation (Close up)

When we look at the closeup of the performance of the due date rules (Figure 5.9), the results seem comparable with the results of the average number of tardy jobs. Again, EDD performs better than the LS-rule for smaller CONWIP levels, where EODD performs best at higher

(31)

31 CONWIP levels. The results of the implementation of the SPT and FCFS rules on the shop floor are shown in the next sections.

Next, the SPT-rule combinations are analyzed regarding the minimization of the average tardiness. The EODD/EODD/SPT combination seems to perform best for nearly all CONWIP levels regarding the minimization of the average tardiness. Only the EDD/SPT/EDD combination seems to perform slightly better for a small interval of CONWIP rules. However, this difference is negligible.

Subsequently, the results regarding the FCFS-rule combinations show that the EDD/EDD/FCFS rule seems to perform best for the relatively low and medium levels of CONWIP. For the higher levels of CONWIP, the EDD/FCFS/EDD rule performs best. However, just as any other FCFS-rule combination, they are outperformed for every CONWIP level by other combinations. Hence, it can be stated that the FCFS-rule combinations are not favorable to implement regarding the minimization of average tardiness. Eventually, the performance of the due date-rule combinations is observed. Hereby, it seems that the EDD/EDD/EODD combination and the EODD/EODD/EDD combination perform best. For the lower CONWIP levels, the EDD/EDD/EODD shows optimal results in terms of tardiness minimization. However, this only applies to a relatively small amount of CONWIP levels and does not provide major improvement compared to the LS-rule. For medium and high CONWIP levels, the EODD/EODD/EDD combination obviously performs way better than all other combinations.

The results of all analyzed rule combinations regarding the average tardiness can be found in

Appendix 6.2.

5.4.1 COMPARED TO LS-RULE

(32)

32

Figure 5.10 – Average tardiness compared to LS

At the minimization of average tardiness, the results show that the EDD rule performs better than the LS-rule for the lower CONWIP levels. With respect to the higher CONWIP levels, the EODD rule performs best, decreasing the average tardiness with more than 15%. The combination of EODD/EODD/EDD shows the best performance for some medium CONWIP levels. However, it should be noticed that this combination of rules does have a relatively high variance, as shown in Figure 5.10, and only performs best for a limited amount of CONWIP levels. Since EDD shows less variance for these CONWIP levels, it is assessed to be suitable for low and medium CONWIP levels. EODD, on the other hand, is assessed to be suitable for high CONWIP levels. The variance of EODD at these levels is also very low, resulting in a more predictable and controllable delivery reliability.

(33)

33

Figure 5.13 - Standard deviation EDODD

To present the performance of the best performing rules and the current rule of our industry partner, the delivery reliability and the average tardiness are shown for varying CONWIP levels in Figure 5.14. The dots on the lines represent a CONWIP level interval of twenty, with a maximum CONWIP level of 200.

Figure 5.14 – Delivery reliability and Ave rage tardiness of best performing rules

The results discussed throughout this section already revealed that the EODD, the EDD and the, LS rule perform best throughout the different CONWIP levels. Figure 5.14 shows that the EODD rule tends towards a delivery reliability of 97.5% and an average tardiness of 0.14 days. EDD and LS, on the other hand, seem to result in approximately the same delivery reliability and average tardiness, tending towards a delivery reliability of 95.5% and an average tardiness of 0.16 days. However, EDD and LS seem to perform better in terms of delivery reliability and average tardiness at lower CONWIP levels. The full graph, showing the results for all CONWIP levels, can be found in Appendix 6.3.

-0,5 0 0,5 1 1,5 2 2,5 3 3,5 4 4,5 20 40 60 80 100 120 140 160 180 200 Day s CONWIP level 85% 86% 87% 88% 89% 90% 91% 92% 93% 94% 95% 96% 97% 98% 99% 100% -0,4 -0,35 -0,3 -0,25 -0,2 -0,15 -0,1 -0,05 0 De liv ery re liab ili ty

Average tardiness in days

(34)

34

6 DISCUSSION

In this section, the results of the previous section are discussed. Subsequently, the implications of the results are mentioned, as well as the limitations of the research. As already mentioned in the previous section, the throughput of orders per hour increases relatively fast for the lowest CONWIP levels. When CONWIP levels become higher, the throughput increases less, finding its maximum between CONWIP level 120 and 140. This stability seems to go hand in hand with the performance of dispatching rules. Through the analysis of the different combinations of dispatching rules, it namely becomes clear that the dispatching rules’ performance also becomes stable when the throughput realized by the CONWIP level gets to 100%. Figures 5.3 and 5.9 show this in a clear way. Between 90% and 99,9% throughput realization (according to Figure 5.1), the performance of dispatching rules fluctuates a lot.

(35)

35 namely limit the WIP in a way that it limits the throughput and hence the delivery speed. Applying the EODD rule in such a situation has the consequence that the system tries to get every order finished in time, but ends up delivering most orders too late. This tendency can also be observed for EDD at CONWIP level 20 and less. Here, FCFS and SPT perform better than EDD (Figure 5.2). EDD is namely constantly overtaken by events at these lowest CONWIP levels.

When we compare the results of this research with the conclusions of the research of Brown, Dimitrov and Barlatt (2015), it can be stated that there are some similarities. Brown, Dimitrov and Barlatt (2015) state that the EODD rule performs best with respect to the maximization of delivery reliability. However, they show that the combination of EODD in the order pool and SPT on the shop floor performs slightly better. According to this research, the combination of EODD and SPT seems to perform fine with respect to delivery reliability maximization in combination with CONWIP as well. Yet, within a CONWIP system, combining SPT on the shop floor with the EDD rule in the order pool seems to perform even better. However, it can be derived from the results that the implementation of neither SPT nor FCFS seems to lead to great improvements, compared to the implementation of a single rule.

One of the implications according to the results is that the implementation of a single rule is most favorable. However, there is no single rule that is the most suitable with respect to the minimization of tardy jobs and average tardiness for all CONWIP levels. The results of the EDD-rule and LS-rule seem to be pretty comparable for the lower CONWIP levels. Their averages show that the rules alternate each other in their performance and the variance of the averages is also comparable. The existence of a trade-off seems to be the case. On the one hand, a company can choose to deliver a maximum amount of orders in time, allowing a somewhat higher tardiness at the jobs that are tardy. On the other hand, it can choose to minimize the tardiness of jobs, allowing somewhat more jobs to be tardy. However, the EDD rule shows more stability when CONWIP levels get somewhat higher. Moreover, implementing EDD shows the highest probability to minimizes the average tardiness for nearly all low CONWIP levels. For the CONWIP levels that enable maximum throughput, this trade-off is not an issue at all. EODD to be the most suitable regarding both number of tardy jobs and average tardiness for these levels.

(36)

36 into the suitable dispatching rules if delivery reliability is exposed to high pressure. This can be the case when e.g. machine failure occurs or orders need to be re-produced due to a mistake in production. These circumstances can have the consequence that the due date of orders will be even more challenging to meet. In these situations, the dispatching rules identified to result in maximum delivery reliability at low throughput rates can be very helpful.

6.1 Limitations

A limitation of this research is that the results are achieved by a simulation study. This simulation study aims to approach reality. Since it is an approach, results should not be interpreted as equal to the real world scenarios. Factors that are not included in this research are e.g. less common routings, setup times, transportation times and inter-arrival times of orders. Besides, environments of different job shops can vary a lot, and this research only focusses on an environment as it is explained in section 3.1. Hence, the conclusions of this research cannot be generalized for every job shop.

(37)

37

7 CONCLUSION

In this section, the conclusions that can be drawn on the results of the analysis are presented. These conclusions provide answers on the questions that are stated in section 3.3. Hereafter, the main conclusion that answers the research question is provided. Eventually, the recommendation for our industry partner is discussed and suggestions for future research are provided.

In this research, we aimed to identify suitable dispatching rules that complement CONWIP in the maximization of delivery reliability. It seems that the performance of dispatching rules differs much among the CONWIP levels. When we compare all combinations of rules to the LS-rule, most of the rules perform worse regarding delivery reliability. However, some rules were identified to show better results than the LS-rule.

Through the analysis of all different combinations of dispatching rules it became clear that, most of the times, the implementation of a single rule performs best. Single rules show, in general, less variation and therefore are a lot more predictable. Besides, the single rules also show a clear pattern. This is particularly the case at the average tardiness, where EDD obviously performs best at CONWIP levels that limit the throughput and EODD performs best at CONWIP levels that enable maximization of throughput. The low level of variance and the clear pattern both ensure a more controllable production. Hence, a single rule is more favorable to implement.

The research shows that the most suitable dispatching rules differ throughout the different CONWIP levels. Regarding the minimization of the number of tardy jobs for high CONWIP levels, the EODD rule is assessed to be the most suitable dispatching rule. This rule also provides a lot of stability, showing no big differences in averages of tardy jobs and little variance. For the lower CONWIP levels, the results are less clear. Since EDD results in low numbers of tardy jobs and has relatively low variance, it is assessed to be most suitable for the lower CONWIP levels regarding minimization of tardy jobs.

The rules discussed in the previous paragraph also perform best for the corresponding CONWIP levels with respect to the average tardiness. This performance is very clear (Figure

5.10). Hence, the answer on the research question: ‘Which dispatching rule(s) in combination

(38)

38 of delivery reliability for CONWIP levels that limit the throughput of a system. EODD, on the other hand, results in the maximization of delivery reliability for CONWIP levels that maximize the throughput of a system.’

7.1 Recommendation to Neways Leeuwarden

The recommendation to Neways Leeuwarden, based on this research, is to apply CONWIP as the order release method in a clear and concise way. This can be achieved through the

determination and implementation of the CONWIP level that maximizes their throughput and hence, their delivery reliability. Together with CONWIP, Neways should solely apply one and the same dispatching rule in the order pool and on the shop floor. This dispatching rule should be the EODD rule. Figure 5.14 clearly shows that the EODD rule results in the maximum delivery reliability at a high CONWIP level.

7.2 Future research

A suggestion for future research is to analyze the dispatching rules’ performance for specific routings, e.g. on the aspect of minimization of throughput times. This research does take into account routings, but does not perform in-depth analysis on the dispatching rules’ performance for a specific routing. Future research should also take into account some cooling down time of the simulation. At the SPT rule, jobs can overtake each other according to their processing time. The last jobs of the simulation are less likely to be overtaken, since there are no jobs left. Hence, the last jobs could be facing a more beneficial situation, which decreases their throughput time.

(39)

39

8 REFERENCES

Barman, S. and Laforge, R. L. (1998) ‘The impact of simple priority rule combinations on delivery speed and delivery reliability in a hybrid shop’, Production and Operations

Management, 7(4), pp. 402–416.

Beemsterboer, B. et al. (2017) ‘Integrating make-to-order and make-to-stock in job shop control’, International Journal of Production Economics. Elsevier, 185(December 2016), pp. 1–10. doi: http://dx.doi.org/10.1016/j.ijpe.2016.12.015.

Blackstone, J. H., Phillips, D. T. and Hogg, G. L. (1982) ‘A state-of-the-art survey of

dispatching rules for manufacturing job shop operations’, International Journal of Production

Research, 20(1), pp. 27–45.

Brown, A., Dimitrov, S. and Barlatt, A. Y. (2015) ‘Routing distributions and their impact on dispatch rules’, Computers & Industrial Engineering. Elsevier Ltd, 88, pp. 293–306. doi: 10.1016/j.cie.2015.07.014.

Fernandes, N. O., Land, M. J. and Carmo-Silva, S. (2016) ‘Aligning workload control theory and practice: lot splitting and operation overlapping issues’, International Journal of

Production Research, 7543(February), pp. 1–11. doi: 10.1080/00207543.2016.1143134.

Framinan, J. M., Ruiz-Usano, R. and Leisten, R. (2000) ‘Input control and dispatching rules in a dynamic CONWIP flow-shop’, International Journal of Production Research, 38(18), pp. 4589–4598. doi: 10.1080/00207540050205523.

Heger, J. et al. (2016) ‘Dynamic adjustment of dispatching rule parameters in flow shops with sequence- dependent set-up times’, International Journal of Production Research. Taylor & Francis, 7543(May), pp. 0–13. doi: 10.1080/00207543.2016.1178406.

Holthaus, O. and Rajendran, C. (1997a) ‘Efficient dispatching rules for scheduling in a job shop’, International Journal of Production Economics, 48(1), pp. 87–105. doi:

10.1016/S0925-5273(96)00068-0.

Holthaus, O. and Rajendran, C. (1997b) ‘New dispatching rules for scheduling in a job shop --- An experimental study’, The International Journal of Advanced Manufacturing Technology, 13(2), pp. 148–153.

Hopp, W. J. and Spearman, M. L. (2008) Factory Physics. 3rd edn. McGraw Hill.

Huang, G. et al. (2015) ‘A simulation study of CONWIP assembly with multi-loop in mass production, multi-products and low volume and OKP environments’, International Journal of

Production Research, 53(14), pp. 4160–4175. doi: 10.1080/00207543.2014.980458.

Huang, M., Wang, D. and Ip, W. . (1998) ‘Simulation study of CONWIP for a cold rolling plant’, International Journal of Production Economics, 54(3), pp. 257–266. doi:

10.1016/S0925-5273(97)00152-7.

(40)

40 10.1080/002075400189301.

Kacem, I. and Chu, C. (2008) ‘Worst-case analysis of the WSPT and MWSPT rules for single machine scheduling with one planned setup period’, European Journal of Operational

Research, 187(3), pp. 1080–1089. doi: 10.1016/j.ejor.2006.06.062.

Land, M. J. (2004) Workload Control in Job Shops: Grasping the Tap., PhD thesis. University of Groningen.

Land, M. J. and Gaalman, G. J. C. (2009) ‘Production planning and control in SMEs: time for change’, Production Planning & Control, 20(7), pp. 548–558. doi:

10.1080/09537280903034230.

Land, M. J., Stevenson, M. and Thürer, M. (2013) ‘Integrating load-based order release and priority dispatching’, International Journal of Production Research, 52(4), pp. 1059–1073. doi: 10.1080/00207543.2013.836614.

Li, W., Freiheit, T. and Miao, E. (2016) ‘A lever concept integrated with simple rules for flow shop scheduling’, International Journal of Production Research. Taylor & Francis,

7543(June), pp. 1–16. doi: 10.1080/00207543.2016.1246762.

Little, J. D. C. (2011) ‘OR FORUM—Little’s Law as Viewed on Its 50th Anniversary’,

Operations Research, 59(3), pp. 536–549. doi: 10.1287/opre.1110.0940.

McKay, K. N., Safayeni, F. R. and Buzacott, J. A. (1988) ‘Job-Shop Scheduling Theory: What Is Relevant?’, Interfaces, 18(4), pp. 84–90. doi: 10.1287/inte.18.4.84.

Panwalkar, S. S. and Iskander, W. (1977) ‘A Survey of Scheduling Rules’, Operations

Research, 25(1), pp. 45–61. doi: 10.1287/opre.25.1.45.

Patil, R. J. (2010) ‘Due date management to improve customer satisfaction and profitability’,

International Journal of Logistics-Research and Applications, 13(4), pp. 273–289. doi:

10.1080/13675561003747423.

Raghu, T. S. and Rajendran, C. (1993) ‘An efficient dynamic dispatching rule for scheduling in a job shop’, International Journal of Production Economics, 32(3), pp. 301–313. doi: 10.1016/0925-5273(93)90044-L.

Soepenberg, G. D., Land, M. J. and Gaalman, G. J. C. (2012) ‘A framework for diagnosing the delivery reliability performance of make-to-order companies’, International Journal of

Production Research, 50(19), pp. 5491–5507. doi: 10.1080/00207543.2011.643251.

Spearman, M. L., Woodruff, D. L. and Hopp, W. J. (1990) ‘CONWIP, a pull alternative to kanban’, International Journal of Production Research, pp. 879–894. doi:

10.1080/00207549008942761.

Sridharan, V. and Li, X. (2008) ‘Improving delivery reliability by a new due-date setting rule’, European Journal of Operational Research, 186(3), pp. 1201–1211. doi:

10.1016/j.ejor.2007.03.037.

(41)

make-to-41 order companies’, Production and Operations Management, 21(5), pp. 939–953. doi:

10.1111/j.1937-5956.2011.01307.x.

Thürer, M., Qu, T., et al. (2014) ‘Continuous workload control order release revisited: an assessment by simulation’, International Journal of Production Research, 52(22), pp. 6664– 6680. doi: 10.1080/00207543.2014.907514.

Thürer, M., Stevenson, M., et al. (2014) ‘Lean control for make-to-order companies: Integrating customer enquiry management and order release’, Production and Operations

Management, 23(3), pp. 463–476. doi: 10.1111/poms.12058.

Tung Dang, T. (2013) ‘Combination of dispatching rules and prediction for solving multi-objective scheduling problems’, International Journal of Production Research, 51(17), pp. 5180–5194. doi: 10.1080/00207543.2013.793857.

Xiong, H. et al. (2017) ‘A simulation-based study of dispatching rules in a dynamic job shop scheduling problem with batch release and extended technical precedence constraints’,

European Journal of Operational Research, 257(1), pp. 13–24. doi:

10.1016/j.ejor.2016.07.030.

Zäpfel, G. and Missbauer, H. (1993) ‘New concepts for production planning and control’,

European Journal of Operational Research, 67(3), pp. 297–320. doi:

10.1016/0377-2217(93)90287-W.

(42)

42

9 APPENDICES

9.1 Appendix 1 – Screenshots of the simulation model

Shop floor

Suitable Dispatching Rules for Increasing Delivery Reliability in a CONWIP-Controlled Make-to-Order Job Shop: a Simulation study.