An Approach for Overcoming Nesting and Bottleneck Set-up Complexities for Workload Control in General Flow Shops

(1)

An Approach for Overcoming Nesting and Bottleneck Set-up

Complexities for Workload Control in General Flow Shops

Faculty of Economics and Business

Technology & Operations Management

Master’s Thesis

25-01-2020

Student: 1st Supervisor:

Sander Rijnja Dr J.A.C. Bokhorst

s2782103 2nd Supervisor:

(2)

Abstract

(3)

List of Tables and Figures

Tables:

Table 3.1: The division of groups, flow and processing times. Table 6.1: Model characteristics.

Table 6.2: Experimental factors. Table 6.3: Performance measures.

Table 6.4: p-values difference in means percentage tardy at different nesting windows. Table 6.5: p-value difference in means percentage tardy with set-up dispatching. Table B1: Framework for the assessment of WLC appropriateness.

Figures:

Figure 2.1: The relationship between work centre workload and shop output in the presence of complexities.

Figure 2.2: The design cycle.

Figure 3.1: The elements of the current production system. Figure 3.2: Time slots for the scheduling of nests.

Figure 3.3: Distributions of lateness.

Figure 3.4: Distribution of lateness at Verheij.

Figure 3.5: Utilization per capacity group per month of 2019. Figure 3.6: Average manufacturing lead times.

Figure 4.1: Decision framework for determining the control level. Figure 5.1: Shift of nest formation date.

Figure 5.2: WLC Release with nesting.

Figure 6.1: Average Gross throughput time for different nesting windows. Figure 6.2: Percentage tardy for different nesting windows.

Figure 6.3: Material utilization for different nesting windows.

Figure 6.4: Average gross throughput time with set-up and ODD dispatching. Figure 6.5: Percentage tardy with set-up and ODD dispatching.

Figure 6.6: Material utilization with set-up and ODD dispatching. Figure A1: Process flow diagram of the conceptual model.

(5)

List of abbreviations

MTO Make-To-Order

PPC Production Planning & Control

WLC Workload Control

SME Small & Medium sized Enterprises

DSR Design Science Research

ERP Enterprise Resource Planning

ODD Operational Due Date

PRD Planned Release Date

MPS Master Production Schedule

WIP Work-In-Progress

SOPST Set-up Oriented Planned Operation Start Time

NW Nesting Window

(6)

Preface

I would like to take a moment to express my sincere appreciation and gratitude to all the people involved in the development of this thesis. Writing a thesis amidst the global COVID-19 pandemic, proved to be a challenging yet educational endeavour.

Especially, I would like to thank Dr J.A.C. Bokhorst for his guidance and input in the process of writing the thesis. Next to that, I would like to thank Gerlinde Oversluizen for her feedback, and Bastiaan Rijkse from Verheij Metaal B.V. for providing company data and useful insights into the metal production sector. Lastly, I would like to express my gratitude to Maaike Odijk, who has provided me with support and feedback on the writing.

Sander Rijnja

(7)

1. Introduction

The make to order (MTO) production strategy is riddled with challenges, requiring companies to implement appropriate production planning and control methods to maintain a competitive advantage. To keep up with globalization and the intense competitiveness of the current marketplace, MTO companies are forced to reduce flow times, improve quality and reduce the cost of their products (Slomp et al., 2009; Fernandes et al., 2016). MTO companies offer a great deal of customization, which can lead to high product variety and large variations in routings, processing times and set-up times (Stevenson et al., 2005), which complicates production planning and control (PPC) (Hendry & Kingsman, 1989). Though complicated, it is of the essence to implement the appropriate PPC concept, especially for small companies with limited financial resources (Stevenson et al., 2005).

Workload control (WLC) is considered the leading PPC solution for MTO high-variety job shops producing customised products (Stevenson, et al., 2005; Thürer et al., 2011). WLC is based on the concept of input/output control (Wight, 1970), where the input rate of work to the shop is controlled in line with the output rate. There are three input control levels: job (or order) entry, job release, and priority dispatching on the shop floor (Land, 2004). A pre-shop pool is used where the jobs wait for a release decision, and a release mechanism regulates access to the shop floor based on workcentre workload norms or limits (Hendry, Huang, and Stevenson 2013). The goal of the release mechanism is to ensure that orders are released in time to meet their due dates, whilst avoiding backlogs and utilising work centre capacities effectively (Fredendall, Ojha, and Patterson 2010). Therefore, at the time of order release, both workload balancing and timing functions are considered simultaneously (Land, 2004). However, these two functions often contradict, resulting in trade-off decisions between prioritising the most urgent orders and balancing the workload (Cransberg et al., 2016).

(8)

savings, combining workpieces can also decrease the number of set-ups necessary, thereby increasing capacity, which is relevant for enabling WLC (Poppinga, 2011). However, combining these workpieces based on raw material utilization may contradict the release decisions in WLC, considering timing and load balancing. Nesting imposes additional restrictions on release if long queues of orders are waiting for others without causing output losses due to the inefficient use of resources is to be avoided (Cransberg et al., 2016). Hence the trade-off becomes three-dimensional, as it concerns decisions between prioritising the most urgent orders (due date performance), balancing the workload (utilization and average flow times), and minimizing raw material waste (material utilization). In the case of nesting, material utilization is often prioritized over workload balancing, causing problems in the rest of the production system, especially when bottlenecks are present. Improper loading of a bottleneck can result in a major increase in lead time, making the combination of nesting and bottlenecks in a production system a pressing matter. Therefore, a PPC approach should incorporate all three dimensions of the trade-off.

Although prior research in this area has been performed, current literature still lacks a practical approach towards finding a combination between WLC and nesting in a production company where a bottleneck and set-ups are present. Verlinden et al. (2006) and Sakaguchi et al. (2018) tackle the problem of nesting workpieces for cutting operations and scheduling their succeeding operations heuristically. However, the concept of workload control is excluded from this research. Cransberg et al. (2016) derived a framework for handling complexities, such as nesting, in WLC. However, this remained a general framework providing directions for solutions, guiding as to where in the PPC system the complexity should be addressed, specific solutions on how to address these complexities is lacking. Therefore, offering a more specific solution for real-life cases could expand knowledge in the field of combining WLC and nesting in a production company where bottlenecks and set-ups are present. Subsequently, it contributes to the existing literature. This is supported by statements of Fernandes et al. (2016); Missbauer & Uzsoy (2020); Thürer, Stevenson, Land, & Fredendall (2018b) who stress the importance of including more empirical evidence from cases in practice.

(9)

during the employment of nesting and WLC concepts. Therefore, the research aim is to design and validate an approach for addressing nesting and bottleneck with set-up complexities in a WLC setting for MTO companies.

To achieve this, a Design Science Research (DSR) methodology is employed. Through this methodology, a new approach is designed, incorporating nesting, as well as bottleneck set-up complexities in a WLC setting. To make the research relevant to MTO companies, and give a more practical example of the application of WLC a case company is involved in the study. The case company linked to this research is SME Verheij Metaal BV. To derive the problem context, a thorough analysis of the current PPC elements in place is performed. After defining the problem context, literature relevant to this problem context is gathered, functioning as a solid base for the design of the approach. Finally, the approach is validated through an empirical experiment in the form of a simulation.

The output of this is research provides a novel approach for combining nesting and WLC, contributing to the context of both WLC and nesting literature. Furthermore, through the measurement of its performance, insights on the potential of this novel approach are obtained. For managers, the outcome of the research may influence the way they look at nesting in a WLC context, and perhaps this research persuades them to redesign the PPC of their operations.

(10)

2.

Research Objectives and Method

In this section, a background is provided on the problems that this thesis addresses within the WLC domain. From this context, the research objective is derived, which is then translated into several research questions. After stating the research objective and research questions, the methodology used in this research is elaborated on. The choice for WLC is justified in Appendix B, alongside a general description of WLC mechanisms.

2.1 Background

WLC research has been conducted throughout the past four decades (Thürer et al., 2011). However, since its emergence, theoretical attempts have been distant from practice (Hendry et al., 2013; Missbauer & Uzsoy, 2020; Thürer et al., 2011). A gap between theory and practice is a phenomenon which is quite common in the field of operations management (Holmström et al., 2009). This lack of familiarity in practice with WLC, causes implementation issues, hindering progress in the early stages of a WLC implementation project (Hendry et al. 2009, Stevenson and Silva 2008). Recently, researchers have started to focus more on narrowing this gap by identifying possible factors that hinder the mechanisms of WLC to improve applicability in real-life situations (Cransberg et al., 2016; Fernandes et al., 2016; Gran & Alfnes, 2019; Soepenberg, Land, & Gaalman, 2012c). As these factors complicate the applicability of WLC mechanisms in a rather complex manner, these are labelled complexities.

(11)

Figure 2.1: The relationship between work centre workload and shop output in the presence of complexities (Cransberg et al., 2016).

2.2 Research Objectives

An MTO company could significantly benefit from WLC mechanisms, however, real-life complexities are required to be addressed for it to be applicable and effective. Therefore, the objective of this research is to design and validate an approach that addresses nesting and bottleneck set-up complexities that arise in a WLC setting for an MTO company. To reach this objective, an analysis is performed on a real-life production system, from which complexities are identified. Relevant literature is used to aid in the design of an approach which addresses these complexities. Performance of the designed approach is then analysed to ensure its validity. Based on these three main research steps, the following research questions are formulated:

1. What are the characteristics of the current production system? 2. What are the important complexities that need to be addressed in the

development of a WLC approach in this context?

3. How can these complexities be addressed by the design of an approach? 4. What is the performance difference between the old approach and the designed

approach?

(12)

2.3 Method

The method applied in this research is that of design science research. In the core, this is a problem-solving method focused on the development and evaluation of artefacts, originating from the field of information systems (Hevner et al., 2004). The design science research method fits this research, for it enables the solution of practical problems and the generalization of knowledge from this solution (Wieringa, 2014). Concerning real-world problems, design science problems are restricted to the first three tasks of the engineering cycle of Wieringa (2014). These tasks are problem investigation, treatment design and treatment validation. The rest of this section will elaborate further on these tasks.

Figure 2.2: The design cycle, adapted from Wieringa (2014)

(13)

production dynamics in the future when the effects of the pandemic have worn off. Data obtained from observations were cross-validated during the interviews ensuring reliability and validity.

After a problem investigation, the next task is treatment design. This task is split into two different parts. The first part where relevant literature is reviewed, retrieving information regarding the problem from the existing knowledge base. This literature covers both existing solutions to the problems as well as contextual information which aids in the design of the new approach. The second part is the design of the approach itself, integrating the insights obtained from literature into one or multiple designs that apply to the production system under analysis.

Lastly, designed artefacts are validated. The goal of validation is to predict how artefacts will interact with its context, without actually observing an implemented artefact in a real-world context (Wieringa, 2014). Thorough testing of the proposed design generally provides pragmatic validity (van Aken et al., 2016). In this research, the artefacts will be validated through quantitative modelling in the form of a simulation study. A simulation model will be derived from a model of the artefact interacting with a model of the problem context. The assumption here is that objective models can be built that explain the behaviour of real-life operational processes (Karlsson, 2016). Simulations are useful for validation research, as they allow us to expose the model to controlled stimuli and analyse in detail which mechanisms are responsible for the responses (Wieringa, 2014).

(14)

(15)

3.

Case Description

In this chapter, the first part of the problem investigation is performed. The production system of the case company is analysed. The current setting within the case company is described alongside the problems that arise. This is done by combining multiple observations from previous studies performed at the case company by Wienholts (2020), van Veen (2020) and van Noort (2020) as well as analysis of ERP data obtained from the company. Some general information is provided, alongside a more detailed analysis of the current production system.

3.1 General Information

The case company for this research is Verheij Metaal BV. This company, located in the Netherlands, has about 45 employees and an estimated turnover of €5 Million (van Noort, 2020). Therefore, the company is classified as an SME. It produces customized metal components for logistics, automotive and appliance construction on an MTO basis. These components are sheet metal products made of aluminium, steel and stainless steel of up to 5mm thickness. In the production of these sheet metal products components go through four production stages: cutting, bending, machining/pressing and welding/treatment. With work generally travelling in one direction through these production stages, the company is labelled as a general flow shop (Stevenson et al., 2005).

(16)

3.2 The Production System

To describe the production system, the operations that are performed between the entry of an order and the finished product, as well as the production planning and controls that guide these operations, are analyzed. The operations are grouped into five different stages: nesting, cutting, bending, machining/pressing and welding/treatment.

The PPC used in the company is described using three decision points of Soepenberg et al. (2012a), which include: customer enquiry/order acceptance, release and (priority) dispatching. First comes the customer enquiry/order acceptance stage, which is followed by nesting. Hereafter, the orders are released to the cutting operation at the shop floor. Lastly, dispatching rules guide the order through the next stages of bending, machining/pressing and welding/treatment. The planning decision is decoupled to the extent that one department is responsible for planning nesting and cutting operations, while another is responsible for all subsequent operations. The aforementioned elements of the current production system are visualized in Figure 3.1.

Figure 3.1: The elements of the current production system.

(17)

subtracting 5 days of safety time and one day per operation from the due date requested by the customer.

Nests are formed from orders five days before their planned release date. If orders are not yet placed within another nest, a new nest is started with these orders. Orders with the same material type and thickness are combined on a sheet of raw material. However, only orders that fall within the nesting window of 5 working days are considered. The nesting window reduces the spread of planned release dates of orders within a nest. The setting of planned release date, planned nest formation date and the nesting window is illustrated in figure 3.2.

Figure 3.2: Time slots for the scheduling of nests.

(18)

After cutting, dispatching rules determine which job should be selected from the queue in front of the next workstation. Formally, the sequence of orders is determined by the operational due date (ODD) principle (van Noort, 2020). The operational due date of an order is determined by the ERP system. The order with the earliest operational due date is selected to be processed first. The planning department only intervenes in case orders risk extreme lateness or when orders of high priority come in, accelerating them at the cost of efficiency (van Noort, 2020).

The bending operation is the bottleneck of the production system. There exist two types of overload at the bending operation. One type is a machine overload, where too many products are released to the same group of bending machines, which causes a large buildup of work in front of the bending group. The second type is an overload of set-ups, as there are only two employees performing set-ups, releasing too many jobs to the entire bending operation causes too much workload for the set-up workers.

The bending operation consists of 13 press braking machines, divided into 4 groups based on their specifications. The ‘larger’ press braking machines are used with products that are large or require a large amount of force to be applied. The ‘small’ press braking machines handle smaller products and require less force. To get a better understanding of the grouping of the machines as well as the portion of flow and the ratio of processing times, these are displayed in the table below. The portion of flow is retrieved from the ERP data and indicates the percentage of jobs that move through each group. Besides that, the relative processing times indicate the ratio of which processing times differ from the processing times of jobs that are processed in group 1.

Group Number of machines Portion of flow Relative processing times

Group1 3 machines 0,32 1

Group2 5 machines 0,37 1,6

Group3 4 machines 0,28 1,8

Group4 1 machine 0,03 1,8

Table 3.1: The division of groups, flow and processing times

(19)

schedule and group multiple jobs of the same bend angle at the minimal required machine. Combining steps could potentially be a good way to increase production performance. However, in this case, the operators choose whenever they are ignoring the schedule, resulting in an occasional and inconsistent reduction or elimination of set-ups. It could be beneficial to look at a more consistent way of combining set-ups.

After bending, the product is either finished after bending or goes through the machining/pressing department, welding/treatment department or both. This is guided by the previously mentioned dispatching rules.

3.3 Performance Analysis

This section analyzes the current production system performance, followed by the revelation of the cause for its (poor) performance. The production performance is defined by the 4 main objectives of MTO companies as stated by Land (2004), which include high delivery reliability, short lead times, low inventory level/WIP, and high Utilization/Throughput.

As previously mentioned in Section 3.1., the company suffers from a poor delivery performance. Delivery reliability is often measured in a percentage of jobs that are delivered too late. Lateness is defined as the difference between the promised delivery date and the realised throughput time of an order (Soepenberg et al., 2012a). There exist two different types of lateness: positive lateness and negative lateness. On the one hand, positive lateness indicates that the order is delivered late. On the other hand, negative lateness indicates that an order is delivered early. When plotting the distribution of lateness, as visualized in Figure 3.3, a distinction is made between average lateness and the variance of lateness. Comparing the graphs of Figure 3.3 with the lateness distribution of the orders of the company, visualized in Figure 3.4, it is observed that there is a high variance in lateness.

(20)

Figure 3.4: Distribution of lateness at Verheij ( van Noort, 2020).

After discussing the lateness of the orders in the company, the utilization of a workstation is discussed. The utilization of a workstation refers to the percentage of time it is not idle due to lack of parts. Figure 3.5 represents the utilization per capacity group per month of 2019. As seen in Figure 3.5, the bending operation, denoted here as the press brake, has the highest utilization of averages 95%, while other workstations average only a 70-75% utilization. This is because this operation is the bottleneck of the production system (van Noort, 2020; van Veen 2020; Wienholts 2020). The utilization of a bottleneck machine is of significant importance to the performance as starving it will directly lengthen the lead time of orders, and overloading it significantly increases WIP. Therefore, it is desired to improve this by means of a designed new approach.

(21)

After defining the utilization of the different capacity groups, the lead times are discussed next. The lead time is the time allotted for the production of a product on that routing. Figure 3.6 depicts the manufacturing critical path. There are several relevant observations made. First, it is seen that products are nested a week prior to starting on the cutting operation. Second, the products spend a lot of time between the cutting and bending operations. Lastly, since a lot of products are finished early, there is a relatively long average waiting time between the last operation and shipping of the products. As mentioned in Section 3.1, long lead and unpredictable times are undesirable in this production environment. A newly designed approach should aim for shorter and more predictable lead times.

Figure 3.6: Average manufacturing lead times (Van Veen & Wienholts, 2020).

(22)

4.

Addressing WLC Complexities in Release and dispatching

After defining the production system and its performance in Chapter 3, literature that is linked to this specific problem context is explored in this chapter. A literature review is performed on WLC research, on nesting, and on bottlenecks as to guide the design of a suitable approach in the next chapter.

4.1 Nesting Complexities

The (material) efficiency gained by nesting cannot be overlooked and the nesting procedure, although conflicting with WLC goals, should be included in the release decision. Nesting encompasses the combination of multiple orders, based on material type and thickness, on a single sheet of raw material (Hermann & Delalio, 2001) to increase raw material utilization and the number of set-ups needed. Raw materials are usually a significant part of total expenses and improvements of a few per cent can result in large savings for manufacturing companies (Struckmeier & León, 2019).

Clustering of products with similar sheet thickness and material characteristics results in an output sequence of jobs based on material characteristics. This causes the products that require subsequent processing steps to arrive at the next station at a time-dependent on their material characteristics (Poppinga, 2011). This collides with the goals of WLC, which is to release orders based on load balancing and timing decisions (Cransberg et al., 2016; Land, 2004).

(23)

Figure 4.1: Decision framework for determining the control level (Cransberg et al., 2017)

In similar research by Poppinga (2011), it was reported that the negative influence of the nesting procedure may decrease through adjustment of the nesting window. The nesting window determines the spread of due dates of orders that are considered in forming a nest. Reducing this may avoid non-urgent orders being pulled forward too far to create batches at the expense of more urgent orders (Cransberg et al., 2017). Therefore, a shorter horizon is expected to increase delivery reliability, which would result in a decrease of lateness (Poppinga, 2011). However, it is noted that although he encountered significant performance increase in terms of delivery performance, this was most likely due to a reduced order input of which the cause is not mentioned. Therefore, actual performance effects are yet to be determined.

4.2 Bottleneck Complications

(24)

in the production system does not significantly affect production performance. Next, the complexity of bottleneck set-ups should be addressed.

Most WLC literature considers set-ups as part of the processing time or as non-existent (Fernandes, Carmo-Silva, 2014). However, due to the criticality of the set-ups, causing high utilization at the different capacity groups as well as the set-up workers involved, it is of high importance to address this complexity (Cransberg et al., 2017). It is possible to consider set-ups at both the release level, as the dispatching level (Fernandes, Carmo-Silva, 2011). However, the number of complexities that can be addressed in the release decision is limited (Cransberg et al., 2017). Also, Thürer et al. (2014) found that refining order release methods to consider set-up times does not have a significant positive effect on performance. Bottleneck set-ups will therefore be addressed at the dispatching level, due to the higher criticality of the nesting complexity as well as the presence of a buffer in front of the bottleneck. Nesting is deemed to have a higher criticality in terms of performance, as material utilization is a high priority in the sector of the specified case company. Next to that, the buffer in front of the bottleneck operation allows for the resequencing and selection of orders where the complexity of bottleneck set-ups can be addressed.

(25)

4.3 Requirements of a WLC approach

(26)

5.

Design of WLC Release & Dispatching

In the previous chapter, literature is used to outline specifications for the design of a suitable WLC method for the current production environment. Concerning the release, it has become clear that the release method needs to be adapted to include the formation of nests, as well as having a focus on loading the bottleneck. The method outlines the three-dimensional trade-off between material utilization, workload balancing and delivery performance.

5.1 Release

It has become clear that nesting needs to be included in the release decision. Normally, nesting would imply adding jobs to a nest based on physical constraints, i.e. jobs can only be combined if they fit on the same sheet of raw material. Combining nesting with WLC release means that, besides the physical constraints, a workload constraint is added. This means that while a job could be physically suitable for a nest, it is not added to the nest if its addition to the nest would violate workload norms. Besides that, the dimension of the nesting window limits the number of jobs that are considered for a nest. This affects the material utilization, while at the same time ensures that the jobs released are jobs with relatively higher urgency. In this research, the assumption is made that sheets of metal are always available. By making this assumption, it is possible to adapt the original planning presented in Chapter 3. As a result, nests are formed on the planned release date of the earliest job, changing the left-hand side of Figure 5.1, first presented in Chapter 3, to the right-hand side of Figure 5.1 presented below. In this case, it is possible to form a nest based on the current state of the production system in terms of workload.

(27)

The workload norms restrict the amount of work that is released to certain workcentres. In this case, special attention is given to the bottleneck. As the set-ups at the bottleneck are performed by a group of separate workers, it is defined as a capacity group of which workload norms are to be set upon. Next to that, due to high variability in processing times and flow towards the different bending groups, a workload norm is set per bending group. This is done to restrict the amount of workload in terms of processing time as well as set-up time to specific bending groups in an attempt to limit machine overload. The release process is depicted in Figure 5.2.

Figure 5.2: WLC Release with nesting

5.2 Dispatching

(28)

(29)

6.

Design validation

In this chapter, the model is first described, after which the experimental set-up is discussed and simulation results are visualized in graphs and discussed.

6.1 Model description

This section elaborates on the choices made in the creation of the simulation model, of which the process of deriving a conceptual model is shown in Appendix A. The conceptual model is derived from a part of the production system. This smaller system provides a better insight into the role of operating variables and, in practice, large systems can often be decomposed into several smaller systems (Bokhorst, Slomp, and Gaalman 2004). Further explanations in addition to the conceptual model are on the nesting procedure, due date setting, order structure, workload norms and the periodic release. The simulation is performed in Tecnomatix Plantsimulation 15.2.

First of all, in modelling the nesting procedure, a simplification is made. Usually, nesting is performed using optimization heuristics based on the cutting stock problem, i.e. optimizing material utilization by fitting jobs of different geometries on a single sheet as efficiently as possible. However, the software used does not include such heuristics and constructing one would not fit within the timeframe of the research. Instead, a simplified version is used, similar to Hermann & Delalio (2001).

The surface area of a job is used to determine whether a job would fit in a nest. The method used in the simulation adds jobs from the same product family to a nest until the sum of the surface areas of the jobs in the nest would exceed the surface area of a standard sheet of metal. The resulting material utilization could therefore be reduced. However, the other outputs of this method strongly resemble the usual manner of nesting as still, jobs with various due dates are added to a nest and some jobs are rejected based on their geometrical properties.

(30)

estimated based on the minimum shop floor throughput time, and the upper limit is determined based on about 5% of the jobs being tardy. The processing times and surface areas for jobs are individually determined.

The characteristics of the model are summarized in Table 6.1. The model will consist of 1 cutting station, which processes multiple jobs (nests) at the same time, and 13 bending stations, which is the same as in the real-life production environment. Furthermore, these 13 bending workstations are operated by a group of 12 workers, which perform the bending operation itself, and a group of 2 workers, which perform the set-ups. A buffer is located between the two operations. The warm-up period is determined in Appendix D and is set on 100 days with a simulation runtime of 1100 days.

Model characteristics

Resources 1 cutting machine, 13 Bending machines, 12 bending workers, 2 set-up workers

Cutting process time Batch: 10 min set-up + 25 min per job. Bending process time Normally distributed (avg =

380 min, StDev= 40 min)

Times 1,6 for group 2, times 1.8 for groups 3 and 4.

Bending Set-up time Constant: 126 min Material type Uniform (1-10)

Set-up type Uniform (1-10) Orders per day 4

Nr of jobs in an order Normal (avg=6, StDev=1) Due date Uniform(5-7)

Surface Area Weibull(1,7;200)

Table 6.1: Model characteristics.

(31)

6.2 Experimental Design

In this section, it is outlined which experiments are performed and which performance measures are used. The designed approach for combining WLC and nesting in the release is tested under different nesting windows. This yields results on both the performance of the approach as well as on the effect of the nesting window setting. In a separate set of experiments, the performance effect of switching between the currently used ODD dispatching rule, and a set-up oriented dispatching rule is then evaluated. These different experiments will each be performed under 11 different workload norms. Workload norms are loosened stepwise until the last name which is set at infinity. The unlimited workload release represents the current production PPC method. A summary of the experimental factors is shown in table 6.2.

Experimental factors

Nesting window 1; 2; 2,5; 3; 3,5; 4; 5 Dispatching at Bending ODD/SOODD

Workload norm 11 different levels of tightness

Table 6.2: Experimental factors.

Several performance measures will be used to evaluate the performance of the different experiments. First of all the average gross throughput time represents the total time an order spends in the system, also including the pre-shop pool. The percentage of tardy jobs, that is, jobs with a lateness that exceeds 0, is used as a performance measure to evaluate the delivery performance. Besides that, the percentage of material utilization is measured, which is the portion of the surface area of a standard sheet of raw material that is occupied by jobs. Next to that, the average throughput time is used as an intermediate variable. Two target workload levels in different methods are assumed to be equally tight if they result in the same average shop floor time. This approach is common in most studies on workload control (Kundu et al., 2020). A summary of the performance measures is presented in Table 6.3.

Performance measures Average gross throughput time % Tardy

% Material utilization Average throughput time

(32)

6.3 Simulation Results

This section is dedicated to the simulation results. First, the performance differences between different levels of workload norms tightness are shown, alongside the effect of different nesting windows. The different performance measures: average gross throughput time, percentage tardy and average material utilization are plotted against the average throughput time. The most right-hand point of each line in the graphs depicts the performance at an unlimited release, implying a workload norm of infinity. For all performance measures, a 95% confidence interval paired t-test statistical analysis is performed to determine the statistical difference in means between the results. Where differences are not significant, or differences are hard to read from the graphs, the results of the statistical analysis are included.

Figure 6.1: Average Gross throughput time for different nesting windows

(33)

Figure 6.2: Percentage tardy for different nesting windows.

Figure 6.2 visualizes the percentage of jobs that have a lateness that exceeds 0. The graph shows that with relatively loose workload norms, the percentage tardy stays somewhat constant. However, as workload norms are tightened the percentage tardy increases almost exponentially. Next to that, a nesting window of 1 day causes a larger portion of jobs to arrive late, as nesting windows are lengthened, the performance differences in terms of observed percentage tardy decrease. This is contrary to the expectations outlined when first discussing the potential outcome of decreasing the nesting window. Again, as the nesting window is increased, performance differences become smaller. Also, when looking at the p-values of the confidence interval, it is seen that between the higher nesting windows, performance differences do not have a 95% confidence of difference in means. Therefore, it is stated that these performance differences become insignificant. The p-values for the paired t-test are shown in Table 6.4.

NW=2 NW=2,5 NW=3 NW=3,5 NW=4 NW=5 NW=1 0,07 0,001 0 0 0 0 NW=2 0,158 0,098 0,022 0,021 0,017 NW=2,5 0,102 0,053 0,045 0,024 NW=3 0,095 0,087 0,078 NW=3,5 0,230 0,097 NW=4 0,121

(34)

Figure 6.3: Material utilization for different nesting windows.

Figure 6.3 visualizes material utilization in percentages. As workload norms are loosened, material utilization increases. As expected, shorter nesting windows result in decreased material utilization. This is because jobs are retained from nest formation as their workload would violate workload norms. The nesting window of 1 displays a similar curve as the average gross throughput time shown in Figure 6.1. As jobs are retained in the pool, more jobs are considered for release. This increase leads to higher material utilization as more jobs can be used to fill a sheet of raw material. For longer nesting windows, this effect decreases because it also causes more jobs to be considered for a nest. Besides that, the chance of finding a job that fits the workload norms increases.

To test the performance differences of the set-up oriented dispatching rule, a separate set of experiments was performed with the worst and best performing nesting windows (NW1 and NW5) with both ODD and SOODD priority dispatching rules. Again, the performance measures are plotted in graphs against the average throughput time.

(35)

Figure 6.4: Average gross throughput time with set-up and ODD dispatching.

The graph in Figure 6.5 visualized that applying a set-up oriented dispatching rule also decreases the portion of jobs that arrive late. The difference in delivery performance between nesting windows, although being significantly different, become marginal. The p-value for the paired t-test is shown in Table 6.5.

NW5/set-up

NW1/Set-up 0

Table 6.5: p-value difference in means percentage tardy with set-up dispatching.

(36)

The graph in Figure 6.6 visualized that for set-up oriented dispatching it also holds that a smaller nesting window decreases material utilization. Although employing a set-up oriented dispatching rule decreases the average throughput time, the maximum material utilization at unlimited release is the same.

(37)

7.

Discussion

In this chapter, the results of the simulation are evaluated and explained. The validation of the designed approach is discussed. As well as the effects of the different nesting windows and the choice of different dispatching rules are analysed. Next to that, the limitations of this study are discussed.

7.1 Interpretation of results

The foremost goal of the results is to validate the approach designed in chapter 5. The current situation is depicted by the most right-hand side of the graphs 6.1-6.6 where jobs are released without workload considerations, also known as immediate release. As workload norms are applied and tightened, the performance effects of workload control are observed. Tightening the norms can decrease the average throughput time while maintaining similar performance on the percentage of the tardy jobs. However, tightening workload norms decreases material utilization. These results are in line with Cransberg et al. (2017) who mentions that a three-dimensional tradeoff occurs when these types of complexities are addressed in workload control. Next to that, it was shown that a set-up oriented dispatching rule significantly increases performance. The approach meets the requirement outlined in Chapters 3 and 4. Therefore, the approach is deemed to be a valid method of addressing nesting and bottleneck set-up complexities in a WLC environment.

(38)

opportunities to find orders in the pool that both fit workload norms and physical constraints (i.e. fitting on the sheet of raw material). These increased opportunities may lead to a better load balancing (Fernandes & Carmo-Silva, 2011).

A post hoc analysis was performed on the product mix that is released each day. The average number of urgent jobs (jobs of which the PRD is the current day) and the standard deviation of these amounts were analysed. It showed that when applying a longer nesting window, the average number of urgent jobs decrease as well as having a lower standard deviation. This means that applying a short nesting window result in a larger chance of releasing a large number of jobs. This may explain the differences in delivery performance for the different nesting windows.

The result also shows that a shorter nesting window causes for lower material utilization. Again, this occurs because with a longer nesting window more jobs are considered for release, therefore there is a larger chance that a job is found which fits within the physical constraints of the nest. At a tighter workload norm, jobs spend more time in the pre-sop pool, as can be seen from the gross throughput time. As more jobs are retained in the pool, more jobs are eligible for release, increasing the chance that jobs are with the same material type that fit within the physical constraints of the nest. This explains the fact that at tighter norms, and with a short nesting window (NW=1), material utilization slightly decreases before it increases. This effect diminishes as the nesting window grows larger.

When looking at the performance of the different nesting windows it can be stated choosing a nesting window that is too short can significantly decrease performance. However for the delivery performance, as nesting windows are lengthened, performance differences become smaller, and in some cases even insignificant.

(39)

7.2 Limitations

The research contains several limitations. First of all, this research is limited by only considering a pure periodic release. There exist other relevant release methods which are not included in this research, such as LUMSCOR, which combines a periodic release with a continuous release. While in a similar setting, where a batch process preceded the rest of the operations, Thürer et al. (2020) used a method that would release jobs whenever a batch was completed or when new jobs arrived at the system. Performance differences of employing a different release in this particular context are therefore unknown. Besides that, the design of the release in this research is limited only to the case of a highly unbalanced system where workload norms were limited to the bottleneck operation. In more balanced systems, performance differences might differ due to bottleneck shiftiness. This could lead to different results in comparison to the results of this research.

Next to the limitations in terms of release, this research is limited to only the aspect of lateness as a measure of delivery performance. The release and set-up oriented dispatching rule may cause a lot of jobs to finished early. Land (2004) argues that from a just-in-time perspective, early completion of jobs should be penalized in job shop research. Penalizing earliness may cause for a different view on the performance of the designed approach. However, this is not included in this research.

As for the dispatching, the research was limited by considering only one type of set-up oriented dispatching rule. While it provides an indication of the performance effects a set-up oriented dispatching rule can have, it remains unknown whether other dispatching methods, such as non-exhaustive methods, perform better.

(40)

8.

Conclusion

This chapter draws conclusions based on the objective of the research and its results. Besides that, managerial implications are included.

This research aimed to design and validate an approach that addressed multiple complexities that arise in a WLC setting for an MTO company. Through analysis of the characteristics of the current production system of a case company, it was concluded that nesting and bottleneck set-up complexities were present. Guided by literature, an approach was designed in which nesting complexities were addressed in the release decision, and bottleneck set-ups were addressed through priority rule dispatching. Workload norms were set both on machine groups as well as on the amount of set-up workload that was released.

Results showed that tightened workload norms can positively influence the performance in terms of throughput time or WIP at an equal amount of percentage tardy as the current situation, in which no workload control took place (Immediate release). Even though the influence is positive for both throughput time or WIP at an equal amount of percentage tardy, the material utilization decreases under these circumstances. This shows the three-dimensional trade-off, which is made between load balancing, timing and material utilization.

Next to that, this research has shown that a shorter nesting window causes an overall decrease in performance. These results are in contrary to expectations for two reasons. The first reason entails that a larger nesting window causes a larger amount of jobs considered for release, which increases the chance of finding a job that fits both the workload norms and the physical constraints (fitting on the sheet of raw material). The second reason entails that the use of an ODD dispatching rule mitigated the negative effect of releasing non-urgent jobs early.

Besides that, results show that when the complexity of the bottleneck set-up is addressed by a set-up oriented dispatching rule, performance significantly increases in respect to the currently ODD dispatching rule.

(41)

(42)

References

Aken, J. V., Chandrasekaran, A., & Halman, J. (2016). Conducting and publishing design

science research. Journal of Operations Management, 47-48(1), 1-8.

doi:10.1016/j.jom.2016.06.004

Amaro, G., Hendry, L., & Kingsman, B. (1999). Competitive advantage, customisation and a new taxonomy for non make-to-stock companies. International Journal of Operations and

Production Management, 19(4), 349–371. doi:10.1108/01443579910254213

Cransberg, V., Land, M., Hicks, C., & Stevenson, M. (2016). Handling the complexities of real-life job shops when implementing workload control: A decision framework and case study.

International Journal of Production Research, 54(4), 1094-1109. doi:10.1080/00207543.2015.1047974

Fernandes, N., & Carmo-Silva, S. (2011). Order release in a workload controlled flow-shop with sequence-dependent set-up times. International Journal of Production Research, 49(8), 2443-2454. doi:10.1080/00207541003720376

Fernandes, N. O., Land, M. J., & Carmo-Silva, S. (2013). Workload control in unbalanced job

shops. International Journal of Production Research, 52(3), 679-690.

doi:10.1080/00207543.2013.827808

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

Production Research, 54(10), 2965-2975. doi:10.1080/00207543.2016.1143134

Fernandes, N. O., Thürer, M., Pinho, T., Torres, P., & Carmo-Silva, S. (2020). Workload control and optimised order release: An assessment by simulation. International Journal of

Production Research, 58(10), 3180-3193. doi:DOI: 10.1080/00207543.2019.1630769

Fredendall, L. D., Ojha, D., & Patterson, J. W. (2010). Concerning the theory of workload

control. European Journal of Operational Research, 201(1), 99-111.

(43)

Gran, E., & Alfnes, E. (2019). Practical Guidelines for Production Planning and Control in HVLV Production. IFIP Advances in Information and Communication Technology Advances

in Production Management Systems. Production Management for the Factory of the Future,

596-603. doi:10.1007/978-3-030-30000-5_73

Hendry, L. C. (2009). Product customisation: An empirical study of competitive advantage and repeat business. International Journal of Production Research, 48(13), 3845-3865. doi:10.1080/00207540902946579

Hendry, L., Huang, Y., & Stevenson, M. (2013). Workload control. International Journal of

Operations & Production Management, 33(1), 69-103. doi:10.1108/01443571311288057

Henrich, P., Land, M., & Gaalman, G. (2004). Exploring applicability of the workload control concept. International Journal of Production Economics, 90(2), 187-198. doi:10.1016/s0925-5273(03)00126-9

Holmström, J., Ketokivi, M., & Hameri, A. (2009). Bridging Practice and Theory: A Design Science Approach. Decision Sciences, 40(1), 65-87. doi:10.1111/j.1540-5915.2008.00221.x

Karlsson, C. (2016). Research methods for operations management. London: Taylor & Francis.

Kingsman, B., & Hendry, L. (2002). The relative contributions of input and output controls on the performance of a workload control system in Make-To-Order companies. Production

Planning & Control, 13(7), 579-590. doi:10.1080/0953728021000026285

Kundu, K., Land, M. J., Portioli-Staudacher, A., & Bokhorst, J. A.C.(2020). Order review and release in make‐to‐order fow shops: Analysis and design of new methods. Flexible Services

and Manufacturing Journal. doi:https://doi.org/10.1007/s10696-020-09392-6

Land, M. J. (2004). Workload control in job shops, grasping the tap. s.n.

(44)

Land, M., Stevenson, M., & Thürer, M. (2014). Integrating load-based order release and priority dispatching. International Journal of Production Research, 52(4), 1059-1073. doi:https://doi.org/10.1080/00207543.2013.836614

Land, M. J., Stevenson, M., Thürer, M., & Gaalman, G. J. (2015). Job shop control: In search of the key to delivery improvements. International Journal of Production Economics, 168, 257-266.

Lödding, H. (2012). A manufacturing control model. International journal of production

research, 50(22), 6311-6328.

Missbauer, H., & Uzsoy, R. (2020). Production planning with capacitated resources and

congestion. New York, NY, U.S.A.: Springer.

Robinson, S. (2008). Conceptual modelling for simulation Part II: a framework for conceptual modelling. Journal of the Operational Research Society, 59(3), 291-304.

Sakaguchi, T., Matsumoto, K., & Uchiyama, N. (2018). Nesting Scheduling in Sheet Metal Processing Based on Coevolutionary Genetic Algorithm in Different Environments.

International Journal of Automation Technology, 12(5), 730-738. doi:10.20965/ijat.2018.p0730

Slomp, J., Bokhorst, J. A. C., & Germs, R. (2009). A lean production control system for high-variety/low-volume environments: A case study implementation. Production Planning and Control, 20(7), 586–595. https://doi.org/10.1080/09537280903086164

Soepenberg, G., Land, M., & Gaalman, G. (2012b). Workload control dynamics in practice.

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. International Journal

(45)

Stevenson, M., & Silva, C. (2008). Theoretical development of a workload control methodology: Evidence from two case studies. International Journal of Production Research,

46(11), 3107-3131. doi:10.1080/00207540601039791

Stevenson, M., Huang, Y., Hendry, L. C., & Soepenberg, E. (2011). The theory and practice of workload control: A research agenda and implementation strategy. International Journal of

Production Economics, 131(2), 689-700. doi:10.1016/j.ijpe.2011.02.018

Thürer, M., Stevenson, M., & Silva, C. (2011). Three decades of workload control research: A systematic review of the literature. International Journal of Production Research, 49(23), 6905-6935. doi:10.1080/00207543.2010.519000

Thürer, M., Silva, C., Stevenson, M., & Land, M. (2012). Improving the applicability of workload control (WLC): The influence of sequence-dependent set-up times on workload controlled job shops. International Journal of Production Research, 50(22), 6419-6430. doi:10.1080/00207543.2011.648275

Thürer, M., Silva, C., Stevenson, M., & Land, M. (2014). Controlled order release: A performance assessment in job shops with sequence-dependent set-up times. Production

Planning & Control, 25(7), 603-615. doi:10.1080/09537287.2012.735801

Thürer, M., Stevenson, M., Silva, C., & Qu, T. (2017). Drum-buffer-rope and workload control in High-variety flow and job shops with bottlenecks: An assessment by simulation.

International Journal of Production Economics, 188, 116-127. doi:10.1016/j.ijpe.2017.03.025

Thürer, M., Stevenson, M., Land, M. J., & Fredendall, L. D. (2018a). On the combined effect of due date setting, order release, and output control: An assessment by simulation.

Thürer, M., Stevenson, M., & Renna, P. (2018b). Workload control in dual-resource constrained high-variety shops: An assessment by simulation. International Journal of

(46)

Thürer, M., Huang, Y., & Stevenson, M. (2020). Workload control in additive manufacturing shops where post-processing is a constraint: An assessment by simulation. International

Journal of Production Research. doi:10.1080/00207543.2020.1761038

Verlinden, B., Cattrysse, D., Duflou, J. R., & Oudheusden, D. V. (2006). Modeling Sheet Metal Integrated Production Planning for Laser Cutting and Air Bending. Sheet Metal 2007 Key

Engineering Materials, 913-920. doi:10.4028/0-87849-437-5.913

Wieringa, R. J. (2014). Design science methodology: For information systems and software

engineering. Heidelberg: Springer.

Wight, O. (1970). Input/output control a real handle on lead time. Production and Inventory

(47)

Appendix A: Creating the conceptual model

In the process of deriving the simulation model, a conceptual model is created. Conceptual modelling, the process of abstracting a model from a real or proposed system, is probably the most important part of a simulation study (Robinson, 2008). A conceptual model only partially describes the real-world system, while still incorporating sufficient elements to address the problem that is to be researched. The conceptual model consists of four main components: objectives, inputs (experimental factors), outputs (responses) and model content (Robinson, 2008).

Objectives, inputs and outputs

In determining the objective, a distinction is made between the modelling objectives and the general objectives. The modelling objective is directly related to the aim of the research, it identifies how development and use of the model contribute to these. In this case, the modelling objective is: to evaluate a PPC approach that incorporates nesting and bottleneck information in a WLC setting. In the newly developed approach, different variables of the nesting due date window and workload norm are used to research their effect on production performance and derive the optimal values that result in an overall best production performance. The general objectives are split between flexibility objectives ( the degree to which a model will be changed during and after the study), Run-speed objectives ( becomes more important when many experiments need to be performed), Visual display (The extent of graphical display, ranging between simple schematic and extensive 3D representation), and Ease-of-use (interaction should be appropriate for its intended user).

Modelling objectives:

• Compare the current situation with the new approach in terms of production performance. • Study performance effects of altering the nesting due date window.

• Study performance effects of different types of release General Objectives:

• Time scale: 45 days

• Flexibility: Limited, no extensive model changes expected • Run-speed: Reasonable,

• Visual display: Simple 2D graphics, as the focus is not on the graphical part, rather on the results obtained.

• Ease-of-use: The model is only for use by the modeller, therefore, simple interactive features will suffice.

(48)

• Percentage tardy (% of or orders with lateness that exceeds 0). • Material utilization.

• Average Throughput time

• Average gross throughput time (Time between entry in the system and delivery to the customer)

Model outputs (to determine reasons for failure to meet objectives) • Workstation utilization

• The standard deviation of lateness (with lateness being the difference between the time of completion and due date of an order)

• The standard deviation of throughput time

Model inputs (Experimental factors)

• Change of the nesting due date window

• Release method: WLC Nest Release at different workload norms • Dispatching: Set-up oriented, ODD

Model content: the model scope and the level of detail.

The model scope:

Component Include/exclude Justification

Entities

• Metal sheets Exclude Assume always available • Metal

components

Include Essential for model outputs

Activities

• Cutting Include Response: material utilization • Bending Include Bottleneck, therefore key influence. • Machining Exclude No significant impact on response • Welding Exclude No significant impact on response Queues

• Release buffer Include Important for response

• Pre-shop pool Include Determines the order which can be considered for nesting

Resources

(49)

Level of detail by component type

Component Detail Include/exclude Justification

Entities:

Metal components

Arrival pattern Include

Attributes: Material type, Thickness, Dimensions

Include Important attributes for nesting decisions also needed for response.

Routing include All components go through cutting and bending activities. And activities after bending are not part of the model scope. However, at bending components could either go through small or large press brakes.

Activities:

Cutting Quantity:1 Include It is 1 machine

Nature Include A batch of metal components comes out. It is a decomposition

Cycle time: varying Include Required for modelling throughput/WIP.

Breakdown/repair Exclude Breakdowns are considered not to have a significant impact.

Set-up/changeover Include Jobs require a set-up, which is performed manually by separate operators. Resources Exclude Shifts Exclude Routing Exclude Bending Quantity: 13 workstations

Include There are 13 workstations at the bending operation

Nature Exclude No assembly of entities

(50)

Breakdown/repair Exclude Breakdowns are considered not to have a significant impact.

Set-up/changeover Include Jobs require a set-up

Shifts Exclude No work takes place outside of on-shift time

Resources Include Workers are needed for the set-up and bending

Routing Exclude The routing is already determined by the component

Queues

Pre-shop pool

Quantity: 1 Include

Capacity: infinite Include This queue holds only the information of the order, not the tangible order itself Dwell time Exclude there is no minimum time the entities

should spend in the queue Queue discipline:

Nest release

Include Determines the sequence of entities out of the queue

Breakdown/repair Exclude Not possible

Routing excluded All entities move to cutting from here (fixed routing)

Pre-bending buffer

Quantity:4 Include Each group of bending machines shares the same queue

Capacity: fixed Included There is a limited amount of shop floor available for components in this queue. Dwell time Excluded There is no minimum time the entities

should spend in the queue Queue discipline:

ODD/Set-up oriented Dispatching

Included Experimental factor

Breakdown/repair Excluded Not possible

Routing include Jobs either go through small- or big press brakes

(51)

Bend Operators

Quantity: 12 Include These operators are considered the bottleneck and define the capacity of the bending operation

Where required: processing

Include These workers only perform the processing, the set-up is performed by other workers

Shifts Exclude No work takes place outside of on-shift time.

Other

Set-up Operators

Quantity: 2 Include These workers affect the throughput at the bending operation and are therefore included

Where required: bending machine set-up

Include These workers only perform the set-up, the processing is performed by other workers

Shifts Exclude No work takes place outside of on-shift time.

Other

Modelling assumptions

• Sheets of raw metal are always available

• Breakdowns are rare or have no significant impact on the system • No work takes place outside of on-shift times

Model simplifications

• Simplifying the nesting procedure

• The cutting operation is a single workstation

(52)

(53)

Appendix B: WLC

Instead of developing an entirely new PPC approach, it is argued that mechanisms of the ‘proven’ WLC approach should be retained, this is for multiple reasons. Firstly, the WLC has been identified as the leading concept in production planning and control for MTO companies (Henrich et al., 2004; Stevenson et al., 2005; Thürer et al., 2011). Next to that, a lot of MTO companies are SMEs, and due to the simplicity of WLC, it is easy and cost-efficient to implement in such environments (Land & Gaalman, 2009). Furthermore, the fit between the characteristics of the case company and WLC is assessed using the framework of Henrich et al. (2004) with results being promising, presented below. Therefore, in designing an approach, primarily WLC mechanisms are considered.

Table B1: Framework for the assessment of WLC appropriateness adapted from Hendry et al. (2013) and Henrich et al. (2004) by van Noort (2020).

Workload control

(54)

Figure B1: Decision moments translated into the hierarchical WLC framework (Land 2004).

At all levels, output control either lengthens lead time by adjusting due dates or reduces lead times by adjusting capacity (Kinsgman & Hendry, 2002). Capacity adjustments are conducted though working overtime or adding extra shifts (Lödding, 2012).

Input control at the entry-level improves delivery performance and reduces shop floor congestion by turning away some jobs, or by strategically setting due dates (Kingsman & Hendry, 2002; Thürer et al., 2018a). The due date is the date when the order is placed plus a lead time allowance (i.e. the time that a customer is willing to wait) (Thürer et al., 2018a). Two types of jobs can be distinguished: (i) those where the lead time allowance is specified or proposed by the company and, (ii those where lead time allowance is quoted by the customer, hence, relatively fixed (Ragatz and Mabert 1984; Cheng and Gupta 1989; Kingsman 2000).

(55)

norms and is released, it contributes to the workload at the workstations in its routing (Thürer et al., 2011). This procedure can be performed either at periodic time intervals (Oosterman et al., 2000) or continuously whenever a new order arrives at the shop or an operation is complete (Ferenandes et al., 2017). There exist many different order release methods in the WLC literature (Thürer et al., 2018a).

(56)

Appendix C: Determination of the warm-up period and simulation

length

The warm-up period has been derived from preliminary runs of the simulation. From each day, the average throughput time was noted. From the figure below it can be seen that the throughput time stabilizes at around 85 days. A safety margin of 20% is applied, resulting in a warm-up period of 100 days. The simulation time is set at 1000 days, resulting in an adequate amount of time from which data can be retrieved. Common random numbers will be used between different runs of experiments. At 20 runs per experiment, performance differences are deemed to be statistically significant.

An Approach for Overcoming Nesting and Bottleneck Set-up Complexities for Workload Control in General Flow Shops