Ilse Bosklopper
Student Number:
1968645
MSc. Program:
Technology & Operations Management
Institution:
University of Groningen
Company:
Trelleborg
Supervisor:
Dr. J. Riezebos
Co-assesor:
Dr. N.D. van Foreest
1
ABSTRACT
Purpose – The goal of study is to examine whether the planners choice of aggregation level,
within an MRP system, influences the relationship between routing uncertainty and lead times.
Design/methodology/approach – The data supporting this case study is obtained from
interviews, observations and simulation experiments.
Findings – According to this study high-level MRP is better able to cope with routing
uncertainty, and should be the preferred method of planning in an MTO/ETO environment.
Furthermore, the participant of the experiment appreciated the reduced complexity of his tasks in
the high-level MRP.
Practical implications – Especially in MTO/ETO environments planning is important yet
extremely difficult. The variability, which is inevitable in these environments, can obstruct the
usability of MRP systems. Increasing the aggregation level can enhance the implementation
ability of MRP systems in MTO/ETO environments without having to develop more complex
planning algorithms.
Originality/value – This paper uses experiments to evaluate the performance results of different
aggregation levels. However, this study also observed and interviewed the planners during the
experiments to assess the impact of aggregation level on their tasks.
Keywords – MRP, ETO/MTO environments, Lead times, Planning Aggregation level,
simulation
2
PREFACE
As the closing chapter of my master in Technology and Operations Management at the
University of Groningen, writing this thesis has challenged me more than any other part of the
master. However, I am satisfied with the results and overexcited to hand in the final results of my
Master Thesis.
This results would never been accomplished without the help of others. First, I would like to
thank Jan Riezebos for giving me the chance to do this research. His feedback, positivity and
especially his patience have helped me a lot to keep challenging myself.
The work presented in this thesis was carried out at Trelleborg Ridderkerk. I am very grateful for
being able to perform my research project there. It gave me the chance to connect theory with
practice. I would very much like to thank my supervisor at Trelleborg, Eric-Jan Dregmans for his
enthusiasm, the in-depth discussions and his eagerness to help me. Moreover, I would like to
thank all my colleagues at Trelleborg. Without their time and information contributions it would
not have been possible to finish this paper.
Finally, I would like to thank my family and friends for heir support, feedback but especially for
their encouraging words.
3
INDEX
Abstract
1
Preface
2
Index
3
1 Introduction
4
2 Theoretical background
7
2.1 MRP Systems 72.2 Routing uncertainty in MTO/ETO environments 9
2.3 MRP adjustments to cope with routing uncertainty 12
2.4 Human influence 14
2.5 Research questions 15
3 Methodology
16
3.1 Research design 16
3.2 Measurements 16
3.3 Case organization selection and description 17
4 Simulation design
19
4.1 General design 19
4.2 MRP Tool 19
4.3 Production simulation 20
4.4 Experimental Design 22
4.5 Verification and validation 23
4.6 Experimental settings: 24
5 Findings
26
5.1 Framework for specifying planned lead-‐times 26
5.2 Influence of uncertainty on lead-‐time 29
5.3 Influence of aggregation level on lead-‐time performance 31
5.4 Human-‐system interaction 32
6 Discussion
34
6.1 Planned lead-‐times 34
6.2 Lead-‐time performance 34
6.3 Implications for human scheduler 35
7 Conclusion
36
7.1 Theoretical implication 36
7.2 Practical implications 36
8 Limitations and Further research
37
9 References
38
Appendix A: Production simulation model
42
Appendix B: Workstation settings
43
Appendix C: Welch’s Method
49
Appendix D: Confidence interval method
49
Appendix E: Normallity tests
50
Appendix F: Influence of PLT method
51
Appendix G: Influence of uncertainty on lead-‐times
51
Appendix H: Average difference in Lead-‐time
52
4
1 INTRODUCTION
The current market demand for customised products is argued to be greater than ever before. This has led to a large growth in the number of MTO and ETO companies, which produce non-‐repetitive, high-‐ variety and bespoke products. Resulting in an increase in competition among them (Aslan et al. 2012; Van Nieuwenhuyse et al. 2011; Stevenson, L. C. Hendry, et al. 2005). With this increased competition between MTO/ETO companies, the response time to customer orders has become more important for obtaining competitive advantages. In MTO/ETO environments the response time consists of order processing time (i.e. engineering, designing) and lead-‐time. Lead-‐time is defined as the time between authorization of production to the completion of processing, at which point the material is ready to fill a customer order (Yücesan & de Groote 2000). For MTO/ETO companies, shorter lead-‐times means faster customer response, less cost due to work-‐in-‐process (WIP) and higher efficiency (Pahl et al. 2007; Wedel & Lumsden 1995; Suri 2010).
Not only the physical flow influences the lead-‐time of an order, but also the planning plays an important role (Wedel & Lumsden 1995). Organizations often use an MRP system to support the scheduler in making the production planning, and determining when orders are released to the shop floor (Jonsson & Mattsson 2006; Mabert 2007; Pahl et al. 2007). MRP assumes infinite capacity and static bill-‐of-‐materials (BOM) with known product routings. It treats lead-‐times as static input data, called planned lead-‐times (PLTs), representing the amount of time allowed for orders to flow through the task/facility (Ioannou & Dimitriou 2012; Jodlbauer & Reitner 2012; Ioannou & Dimitriou; Ho & Chang 2001a; Bertrand & Muntslag 1993). PLTs play an important role in the actual lead-‐time performance of the system, as they influence the order release moment. Setting PLTs too high causes orders to be released too early, which increases the level of WIP and results in a self-‐fulfilling extension of lead-‐time (Karmarkar 1989; Pahl et al. 2007; Selçuk et al. 2006; Wedel & Lumsden 1995). On the contrary, if PLTs are too tight and the orders are released to the shop floor too late, it is not possible to meet the due date (Ho & Chang 2001b). These relationships show the importance of having accurate PLTs for attaining short actual lead-‐ times.
5
al. 2011). However, actual lead-‐times are not purely the effect of capacity loading, but also of several other factors often highly present in MTO/ETO companies. While MRP assumes a static-‐BOM with known product routing, MTO/ETO companies are often characterized by high flexibility and variety in product routing. This routing uncertainty can result in a gap between how the MRP system models the production processing of the product, and how the product is processed in reality. MRP systems are vulnerable to uncertainty, and research indicates that uncertainty has a damaging effect on the accuracy of PLTs. Driven by the still increasing power of computers, research focused on adjusting MRP systems towards MTO/ETO environments has been concentrated on extending the MRP logic with complex algorithms to capture the processes and its variables more precisely and correctly. However, as actual lead-‐times are influenced by many factors, including all of these factors in an algorithm will be a very complex task and the presence of uncertainty will make it imposable to capture the actual process perfectly. Moreover, complex algorithms are often not well understood by the scheduler (Pandey et al. 2000), which makes it hard for the planner to keep overview and to have meaningful interaction with the system. Therefore, in this paper we propose a different method for adjusting MRP towards MTO/ETO, and preserve one of the most important strengths of MRP logic; its simplicity.
6
companies by limiting our attention on the aggregation level of an MRP system, and do not revise any of the other elements of MRP calculations. While several researchers have suggested a more aggregated MRP system, little research is done about the effects it has on specifying lead-‐time and lead-‐time performance. Therefore, we have chosen to perform an exploratory case study at a company producing a mix of MTO and ETO products. The goal of this exploratory case study is to create more insight in the ability of aggregation to cope with routing uncertainty, and the reflection this has on lead-‐times. The research question guiding this case study is:
How does the planner’s choice of the aggregation level of MRP influence the
relationship between routing uncertainty and lead-‐times?
7
2 THEORETICAL BACKGROUND
This theoretical framework will give an overview of applicable literature regarding MRP production planning. We will start with describing how MRP works. Then, we will discuss the routing uncertainty that is present in MTO/ETO companies and how this influences both the planning and the performance. Then, we will discuss how scholars have tried to cope with the characteristics of MTO/ETO companies and we will discuss an alternative solution, we will look at how these methods have affected the human scheduler. In the last section the research questions we will answer in this paper are presented.
2.1 MRP Systems
MRP is developed as a solution to the problem of how the right component parts can be received in the right quantity, at the right time (Ioannou & Dimitriou 2012; Murthy & Ma 1991; Ho & Chang 2001b). The scope and usage of MRP systems has grown since the 1970’s (Orlicky 1975; Wight 1984; AMR 1995), but in this research MRP refers to the content and processes in software programs used to make a production planning. MRP systems assume a known BOM, predetermined fixed product routings, and infinite capacity (Jodlbauer & Reitner 2012; Ioannou & Dimitriou; Ho & Chang 2001a; Bertrand & Muntslag 1993).
The BOM shows the relationship between end items and their constituent parts (Hopp & Spearman 2011). In MRP systems, PLTs are specified for each level of the BOM. PLTs represent the amount of time allowed for orders to flow through the specific task(s). PLTs determine when an order is released to the shop floor, by subtracting the total PLT of the due date of the product, after which the material is pushed through all subsequent work centres. An MRP system is often complemented by dispatching rules, which arrange the queues in front of the workstations (Vandaele et al. 2008). Examples of these dispatching rules are: First-‐in-‐First-‐out (FIFO), Last-‐in-‐First-‐out (LIFO) and Earliest-‐Due-‐Date (EDD). The order release policy of an MRP system can be seen as an input control mechanism, as it releases jobs to the shop floor without taking into account the system status (Fernandes & do Carmo-‐Silva 2006).
8
be used as protection against demand quantity uncertainty, and safety lead times should be used to cover completion time uncertainty. Their research does not mention safety capacity, which also applies to most research about buffering in MRP systems. On the contrary, extensive research has been done about safety stock, and a modest amount of research has focused on safety lead-‐times (Dolgui & Prodhon 2007).
PLTs are fixed input parameters that need to be specified by the planning department (Suri 2010). Although it is long known that actual lead-‐times are heavily influence by PLTs, prescriptive ways of setting either have not been adequately developed (Enns 2001). According to Enns (2001) PLTs should be based on actual lead-‐times, yet he recognizes the complexity of doing this due to the stochastic and dynamic capacity constrained production characteristics. Hoyt (1978) argues that planned lead-‐times should be set on the basis of the average flow times being observed. This method seems not appropriate for the stochastic real world. It can lead to a high deviation between the due date and the production completion date, as processing requirements can vastly differ between products. This results in both a high amount of products waiting for shipment, and a low service level. If, for example the lead-‐times are normally distributed, half of the products will be too late and the other half will be too early. This is made visible in figure 2.1.
FIGURE 2.1 NORMALLY DISTRIBUTED LEAD-TIME
9
MTO/ETO job shops are complex dynamic systems, for which future conditions cannot be anticipated by analysing only current performance (Fabrycky & Onur 1987). Another approach is used by Dolgiu & Prodhon (2007), who state that the PLT is the sum of the theoretical lead-‐time and the safety lead-‐times. They refer to Melnyk & Piper (1981), who have proposed that the safety lead-‐time should be determined by k times the standard deviation of the lead-‐times. There are a lot of different opinions about how PLTs should be set, and no consensus has been reached on the best method. However, it is known that the PLT should incorporate the estimated processing time, the waiting time and an appropriate buffer. Even though it is clear that a deep understanding on the effects of PLTs on lead-‐time performance is needed, literature is lacking in a clear guidance on to how to specify accurate PLTs. A good system must result in acceptable due date performance, without incurring excessive inventory overall (Enns 2001). Important relations are:
• Increasing planned lead times result in higher WIP inventory due to queues (Enns 2001)
• Lead-‐times increase non-‐linear long before resource utilization reaches 100% (Pahl et al. 2007; Ioannou & Dimitriou 2012)
• Several amplifiers (variability, uncertainty, capacity and demand dynamics, heterogeneity of product mix) negatively influence lead-‐times (Pahl et al. 2007; Ioannou & Dimitriou 2012). • Lot sizing is about balancing the desire to reduce inventory (by using smaller lots) and increasing
capacity (by using larger lots to avoid setups) and can have severe effects on lead-‐time performance (Enns 2001; Hopp & Spearman 2011).
In MTO/ETO companies the product mix is very dynamic, this results in high variation in machine utilization, regularly updating of PLTs is thus necessary. In the next section we will more specifically address uncertainty within MTO/ETO environments, and how to buffer against it. Moreover, the particular challenges for implementing and using an MRP system are discussed.
2.2 Routing uncertainty in MTO/ETO environments
10
(Bertrand & Muntslag 1993; Hong & Kim 2002). To understand the problems associated with using an MRP system in an MTO/ETO environment, it is necessary to explore the characteristics of these environments.
In MTO/ETO companies the goods flow consists of both a non-‐physical (order processing) and a physical stage (Bertrand & Muntslag 1993). However, in this research we are only concerned with the characteristics of the physical flow and in factors influencing this physical flow. This is appropriate since this paper is concerned with the planning and lead-‐times of the physical stage of the flow. Important characteristics of MTO/ETO companies are: the important role of the customer order, the customer specific product specifications, and product and production variability and uncertainty (Bertrand & Muntslag 1993; Ioannou & Dimitriou 2012). These characteristics have their reflection on routing uncertainty, but also on the methods that can be used to buffer against this uncertainty.
11
ETO/MTO products it is impossible to keep an end-‐item inventory, as the specifications are unknown prior to the order.
Not only does the role of the customer orders effect the uncertainty in product routing, the shop floor configuration adds up to that routing uncertainty. Many MTO/ETO environments can still be classified as a job shop due to the flexible nature of this configuration (Hendry & Muda 2003; Stevenson, L. Hendry, et al. 2005). In this configuration the routing is often somewhat flexible and can be changed by operators if that fits the current shop floor conditions, or if it fits the product requirements. If for example, a job is planned on machine A, but there is a long queue in front of this machine an operator can, if possible, decide to use machine B for certain jobs. While this flexibility is one of the selling points of this configuration, it has consequences for the production planning as variability and uncertainty often causes control problems (Soepenberg et al. 2012). The consequences are especially prevalent with long routings and even when both processing times and routings are known beforehand, predicting the future state of an order is nearly impossible. Only a small disruption of an order at a station or a deviation of the routing can have consequences for the progress of the order itself and for many other orders (Soepenberg et al. 2012). Another variable adding to the routing uncertainty is the possibility to outsource the production or part of the production while the product was already released to the shop floor. Outsourcing can have various reasons, regulation of capacity through outsourcing can be one of the reasons (Riezebos, 2001), The actual lead-‐times can be both negatively and positively affected by outsourcing.
In an MRP system without buffers, whenever a routing is changed which negatively affects the lead-‐time, due dates are not met and the lead-‐time will be expanded. On the contrary, in the same MRP system when a routing is changed that positively affects the lead-‐time, this will only result in finished products waiting at the Finished Goods Inventory (FGI) till its due date. Especially in capital-‐intensive MTO/ETO companies this is considered a problem. Safety lead-‐time buffering can be used to avoid missing the due date; however it can lead to high FGI. These two should thus be balanced, depending on the context. However, in order to attain the desired service level, it is often necessary to buffer against uncertainty (Koh & Saad 2003).
12
nervousness of the system or because no feedback mechanisms are present. Companies thus settle with out-‐dated routing information, and base new order releases on old information.
2.3 MRP adjustments to cope with routing uncertainty
The robustness of a production plan relies heavily on the possibility of modifying the routing of a product with no penalty in terms of lead-‐time performance in the companies’ objectives (Alfieri et al. 2012). Due to the high product routing uncertainty in MTO/ETO companies, it is clear that a production planning should be able to incorporate a certain degree of anticipation to these uncertain events, while providing a robust schedule for the execution of activities and utilization of resources.
In the battle to make MRP systems more veracious, most papers focus on the relationship between capacity loading and actual lead-‐times, and propose a variable PLT that is based on the shop floor condition. Examples are dynamic lead time estimation (Jodlbauer & Reitner 2012; Ioannou & Dimitriou 2012), Advanced Resource Planning (Vandaele & De Boeck 2003) and Workload dependent lead times (Pahl et al. 2007). These examples all base the PLTs on the system’s actual workload. The main advantage of these approaches is that they effectively take into account the congestion that is caused by the interference of different products in the shop floor. These authors however, do not include other factors that influence actual lead-‐times in their model. It can even be argued that introducing these extensions of MRP systems can make the influence of uncertainty and flexibility in product routings more severe as the planned lead-‐time of an order will be based on the position of orders in production according to the production planning. Due to the uncertainty and variability in product routing this information can be incorrect. Besides, the robustness of the production plan will probably be low when routings are modified on a regular basis, as the PLT will fluctuate even more heavily than in a standard MRP system.
13
(2011) propose the use of a heuristic to deal with re-‐routing, to support planning decisions However, in MTO/ETO environments implementing a heuristic like this requires a lot of information to be available in time. While this could be done by implementing and integrating manufacturing execution systems, this is often expensive and difficult to accomplish in job shop environments (Saenz de Ugarte et al. 2009). Another approach suggested in literature to make MRP more suitable for MTO/ETO companies is high-‐ level MRP (HL/MRP). The idea behind this approach is reflected in for example Period Batch Control and QRM (Suri 2010; Riezebos 2001). It suggests that the planning system does not need to prescribe who will work at the various tasks and when they have to start within a period, it suffices to know that there will be enough capacity at the planning level to accomplish all tasks that are scheduled within this period (Burbidge 1996; Riezebos 2013). In HL/MRP the amount of BOM levels is reduced by combining tasks into one level. In a detailed MRP system the planning department has to specify PLTs per task, in a more aggregated MRP this has to be done per subset of tasks. The rest of the logic of MRP remains unchanged in HL/MRP.
In detailed MRP systems, every small change in the routing should be adjusted in the MRP system if one wants to prevent a gap between the real process and the modelled process. These changes often lead to nervous behaviour of the system, which influences the performance of the plant negatively. By reducing the level of detail, and specifying PLTs for sets of operations that are performed within a department or team, small changes will not effect the planning. This is illustrated with the following example: Station 1 to 4 forms a group within the HL/MRP and the group has a PLT of 5 hours. Product A is planned to flow through station 1 and station 2, with both a processing time of 2 hours. In a HL/MRP system the planning does not have to be updated when the routing has been changed to station 3 and 4, with differing process times, because the new routing belongs to the same group. A change within the routing of the department or team does not influence the planning within HL/MRP. Reducing the level of detail, using subsets of resources is comparable to the time-‐bucket approach as for example discussed by Taal & Wortmann (1997). They state that if aggregated information is used, nervousness of a plant can be greatly reduced.
14
time on one workstation can be compensated by a shorter lead-‐time in another workstation (Vandaele & De Boeck 2003). Pooling variability tends to dampen the overall variability by making it less likely that a single occurrence will dominate performance (Hopp & Spearman 2011). Due to the variability pooling, HL/MRP needs lower buffers.
We propose that by reducing the level of detail, the uncertainty of the product routing will be reduced at the planning level. Small deviations from the planned routing will no longer affect the product routing at the planning level. This will positively influence the robustness of the planning. Moreover, buffering against routing uncertainties is done at the group level, which will reduce the buffer that is needed due to variability pooling. We propose that the reduction in uncertainty and the centralized buffers will positively influence lead times.
2.4 Human influence
The influences of changes to the MRP system are hardly or not discussed in relation to their effects on the human scheduler who will work with it. In complex manufacturing organizations, planning and scheduling still requires significant human support to ensure effective performance. Planning should thus not be considered as a mere technical problem. The scheduler is and will stay a critical factor in the planning process (MacCarthy et al. 2001; Taal & Wortmann 1997). The absence of discussing how the proposed extension influences the scheduler can be seen as a flaw in previous research extending or changing MRP systems.
Most extensions of MRP systems are designed from a mathematical perspective and focus on finding a mathematically optimal planning. Mathematical optimality does not always correspond to ‘real world’ optimality. It is the task of the scheduler to create a feasible and reasonable planning; the main function of the planning system is supporting the planner in the planning process (Taal & Wortmann 1997). The interaction between human-‐system should not be underestimated while researching revised MRP systems.
15
simple logic of MRP is preserved, which probably leads to a better understanding than complex algorithmic adjustments.
2.5 Research questions
The main question of this study, ‘How does the aggregation level of MRP influence the relationship
between routing uncertainty and lead-‐times?’ will be answered in the proceedings of this thesis. The sub
questions addressed are:
1. How does implementing HL/MRP influence the planned lead-‐times?
2. How is lead-‐time performance influenced when routing uncertainty is present?
3. How does the level of MRP aggregation affect lead-‐time performance when uncertainty is present?
16
3 METHODOLOGY
The purpose of this study is to broaden the insights into how the aggregation level of planning can influence the relationship between routing uncertainty and lead-‐times. The core goals are to identify how the aggregation level influences the lead-‐time performance, especially related to planned lead-‐ times, uncertainty and system-‐scheduler interaction.
3.1 Research design
After performing a literature study, empirical research is conducted through an exploratory case study (Voss et al. 2002; Karlsson 2009). The objective of this study is to answer “how” questions, and the focus is on a phenomenon within a real life context, which makes case study most appropriate (Yin, 2009). A single case study is chosen to increase the depth of the analysis. Within the settings of the case company, a broad range of methods is used to gather both quantitative and qualitative results. In order to answer the (sub-‐) questions guiding this research, a combination of interviews, behavioural observations and two interrelated simulation models have been used.
Within the empirical settings of the case organization, we have carefully developed two models; one simulates the production system and the other simulates the planning system. Robinson (2004) defines simulation as experimentation with a simplified imitation of an operations system as it progresses through time, for the purpose of better understanding and/or improving the system. The use of simulation enables us to test multiple scenarios in a relatively short amount of time, which is necessary for answering our research questions. The simulation models are not only used to gather quantitative data about lead-‐time performance, but the models were also used in a simulation game in which the interaction with the human scheduler was assessed with both behavioural observations as interviews.
3.2 Measurements
17
before or on the due date. FGI exists because of deviations between the completion time of an order and the due date agreed upon with a customer, and will be measured by the average time a product spends in FGI. In short the variables under consideration are:
• Lead-‐time (hours)
• Customer service (% on-‐time)
• Average time in FGI (due to deviation between due-‐date and lead-‐time) (hours)
Furthermore, planned lead-‐times are measured, to evaluate the accurateness of the PLT and to evaluate the difference between the two aggregation levels. The observations of the participant and the interviews conducted are used to assess the scheduler-‐system interaction.
3.3 Case organization selection and description
We have selected a MTO/ETO company in which the production process could be categorized as a job shop with varying, flexible routing. This company desires to reach shorter lead-‐times and the planners were used to working with an MRP system. The case company, Trelleborg Ridderkerk, is a global supplier of engineered rubber solutions in e.g. civil engineering, dredging and energy, and is located in the south of the Netherlands. Trelleborg distributes products to more than 40 countries in the world. Although, part of a larger group, the organization can be considered as a medium-‐sized enterprise (MSE) with around 160 employees. It produces a mix of MTO and ETO products with a high level of customization and variability, which make it appropriate for our research. Figure 3.1 gives a clear picture of their annual results in terms of products and product mix.
The organization is currently involved in a Quick Response Manufacturing (QRM) transformation. QRM pursues the reduction of lead-‐time in all aspects of an organizations operations, both internally and externally (Suri 2010). The transformation has until now mainly focused on the office and engineering practices of the case organization (Q-‐ROCs, redesigning processes). While the office is designed in QRM cells, the shop floor can be classified as a job shop and is therefore suitable for this research. The level of uncertainty in product routing is high due to the existence of late design changes, changes due to shop floor conditions and the possibility of outsourcing.
18
FIGURE 3.1 TRELLEBORG PRODUCTION ORDERS19
4 SIMULATION DESIGN
This section will discuss the design, validation and settings of the simulation models that are used. It also provides insights into how the experiments were conducted. This section is split up into four parts. First, we explain the general design of the study, and show the interrelationship between the simulation models. Second, a detailed description of the two simulation models and the in-‐ and outputs of the models are discussed. Third, the experimental design is discussed. Fourth, the verification, validation and experimental settings of these two models are discussed.
4.1 General design
To examine the influence of the aggregation level of MRP, we have simulated the MRP system. Two configurations have to be possible within this simulation, a detailed planning and a high-‐level planning. Furthermore, the simulation model is designed in such a manner that it allows for analysing the interaction with the human-‐scheduler. In order analyse the influence of the aggregation level of the MRP system on the lead-‐time performance, a simulation model of the production process is made. Input data for both models is obtained from the ERP system, observations of the real system and interviews with the production planner, the plant manager and several operators.
4.2 MRP Tool
The frame of the MRP model and its boundaries result from the scope of this dissertation: planning the physical flow of the product. Therefore, the model needs to span the entire production operation. It excludes process steps like engineering, designing and tendering that are likely to occur within MTO/ETO companies (Hicks & McGovern 2009).
The model is built using Microsoft Excel. This software is mainly chosen because of the familiar interface. As the scheduler has to interact with this model, it is suitable for the purpose of this study. The study requires two configurations of the model: an MRP at task level and an MRP at team/department level. The input data obtained for the purpose of this model is:
20
• Planned routings of orders
• Expected processing times per task • Expected setup time per task
• Capacity of the departments/machines/people
• Buffer time per task/department to accommodate uncertainty
The output of the system is, in both configurations, an order release list for the coming period specifying the day an order has to start. Next to that, a capacity requirement overview is presented in time buckets of one week.
4.3 Production simulation
Again, the frame of this simulation model and its boundaries are a result of the scope of this dissertation: planning the physical flow of the product. The aim of this simulation model is to assess the ability of HL/MRP to cope with routing uncertainty in the physical flow, and the model should thus be a representation of the whole job shop through which the products flow. Within the boundaries of the simulation, there were almost limitless options for the routing. With the restriction of two stations, the exit strategy was not limited and all other stations could be their successor, of course depending on the characteristics of the product under consideration. The study requires two configurations of the model: a production system with low routing uncertainty and one with high routing uncertainty. This is modelled by increasing the deviation between the planned product routing and the actual product routing due to late design changes, decisions on the shop floor and outsourcing of tasks.
The model is built using the simulation software Tecnomatrix Plant Simulation, which is developed by Siemens with the purpose to model, simulate, analyse, visualize and optimize production systems, material flows and logistic operations in an efficient way (Bangsow 2010). The ability of Plant Simulation models to represent the variability, interconnectedness and complexity of a system makes it appropriate software for modelling a job shop production environment. To give an indication of the content of this simulation model, a small overview of some content is given in appendix A.
21
behaviour of the real production system (Robinson 2004). To assure data-‐validity, the source all data has been checked for inconsistencies and unusual patterns and outliers. Knowing the distribution of a data set matters, as using differences in distribution deliver great differences when implemented into a simulation. The distribution of the process and the distribution of the setup time are determined for all 31 workstations within the boundaries of the production simulation. Evaluating the distributions of the workstations has been performed with the use of DataFit, an Add-‐in of Plant Simulation 12. A significance level of 95% is used for determining the distribution type. The results are summarized in appendix B.
Input Type Dataset/parameter
Production datasets
Order release list
Planned Product Routing Actual Product Routing
Process parameters
Work-‐hours per machine/employee Process time parameters per process step Setup time parameters per process step Batch size parameter per process step
TABLE 4.1: INPUT DATA
In order to assess the lead-‐time performance of the system, all performance measures defined in the methodology are tracked per order and are averaged over the run-‐length to make it possible to make comparisons between the difference experimental settings. Descriptive statistics such as minima, maxima and standard deviation are also calculated for the lead-‐time. Moreover, the simulation model provides the scheduler with information for determining the PLTs. For this purpose, the simulation model also tracks the average waiting times of each buffer, the average processing time together with the capacity loading of the past period.
22
2004). The simplifications made are based upon analysis of case company’s data and information gathered longitudinally. The simplifications that have been made are:
• Machines never fail
• Input of external supplies are always present at the release moment specified by the MRP model • Human capacity is not affected by illness or other factors other than lunch breaks
• Transport from one station to another does not require time
• Product Quality Check Type Three (QC3) does not require any time, and is done during ‘normal’ processing time
• Aggregation of one type of machines (caldron) to one machine
4.4 Experimental Design
This study will compare different configurations of both the MRP tool and the simulation model. This section will provide an overview of the scenarios that are tested. However, we will start with outlining the planning procedure that is used. In collaboration with the scheduler of the case company, in all configurations of the MRP tool, order release lists are made on a weekly basis, for a period of five weeks, indicating which order should be released to the shop floor on which day. Both determining and adjusting the planned lead times, and readjusting the planning due to capacity limits was the responsibility of the scheduler. This made it possible to study the behaviour of the planner, and study the impact the level of aggregation has on the planner.
With these order release lists, several scenarios have been tested. The focus of the scenarios is on the lead-‐time performance of these order release lists under routing uncertainty. The scenarios tested are presented in table 4.2.
Scenario Settings
Detailed MRP
HL/MRP
HL/MRP-‐PM
No routing uncertainty
Scenario 1D
Scenario 1H
Scenario 1H-‐PM
Routing uncertainty
Scenario 2D
Scenario 2H
Scenario 2H-‐PM
TABLE 4.2 SCENARIO SETTINGS
23
To give a clear insight into the several scenarios, the scenarios are discussed below.
• Scenario 1D: In this scenario all planned products are produced in exactly the routing as planned, the planning is made with the use of the detailed MRP tool. Planned lead-‐times are intuitively set based on the experience of the scheduler
• Scenario 2D: This scenario is similar to scenario 1D, but with routing uncertainty due to design changes introduced.
• Scenario 1H: In this scenario all planned products are produced in exactly the routing as planned, the planning is made with the use of the HL/MRP tool. Planned lead-‐times are intuitively set based on the experience of the scheduler
• Scenario 2H: This scenario is similar to scenario 1H but with routing uncertainty due to design changes introduced.
• Scenario 1H-‐PM: In this scenario all planned products are produced in exactly the routing as planned, the planning is made with the use of the HL/MRP tool. Planned lead-‐times are set according to the method proposed in the next chapter.
• Scenario 2H-‐PM: This scenario is similar to scenario 1H-‐PM but with routing uncertainty due to design changes introduced.
4.5 Verification and validation
24
Simulation 12. We have tried to find the cause of outliers and whenever found appropriate the outliers are removed from the data.
Doing several tests, and watching the models at a low running speed verified the models. This was done to check whether the products are processed correctly and whether the methods do what they are intended to do. Moreover, a semi-‐experienced user of Plant Simulation was asked to test the production simulation model. Beforehand, he was informed about the goal of the model and a quick screening through the model was done in collaboration.
By comparing the capacity loading of individual workstations in the commercial ERP system with the MRP simulation, white-‐box validation of the MRP simulation model was done. At first high deviations in capacity requirement were found for one particular workstation, however it appeared that the capacity profile of this workstation (Bouwwikkel) was out-‐dated in the commercial ERP system as the merger of two-‐production sites added extra capacity. When all settings within both models were the same, no deviations were found in both the order release list as the capacity requirements. As the HL/MRP is just an adjustment in configurations of the detailed MRP simulation, and the logic remains the some, not specific white-‐box testing has been performed. We have verified the model by setting the PLTs to zero in both configurations; this should result in exactly the same capacity requirements and order release lists. White-‐box validation of the production simulation was done by inspecting the output reports for individual stations, and discussing them with the production planner, the shop floor manager an operator. Within these sessions in-‐ and output of the model were discussed, and important variables were discussed like utilization and actual lead-‐time. Some minor adjustments were made, especially in the hours a day a workstation worked. Moreover, breaks were reduced from one-‐hour to a half-‐hour per shift. This as a rotating system is used, which implies that while every worker goes on a shift of an hour, machinery is only standing still for a half hour.
After this, black-‐box validation is used to check the overall behaviour of the model (Robinson 2004). For both models, extensive validation-‐sessions with the production planner and the shop floor manager were conducted. Black-‐box validation of the MRP simulation model is done by comparing the order release dates of the detailed MRP simulation with the commercial ERP system the company is currently using. The issue of experimentation validity is discussed in the next section.
25
In the simulation game, the production simulation model used is a terminating process, as a one-‐week production schedule in the end-‐point of the simulation. Therefore, the run length will be a week of operations (Monday to Sunday). While often terminating process, returning to the empty setting after each period is not a realistic starting point for this model. From the second week onwards, the situation at the end of the previous week is the starting point for the week after. However, because the system starts in an empty state, it has to be filled with products before a representative state is achieved for the first week. The weekly order release lists from the previously described experiments were combined in the simulation game so that the simulation model can be described as non-‐terminating.
In this case it is appropriate to use a warm-‐up period before obtaining results. The warm-‐up time is defined as the period the model needs to reach a representative state (Robinson 2004). An appropriate warm-‐up period is determined with the use of the Welch’s method. The Welch method is applied on the average lead-‐time obtained from the first scenario (1D). After testing several window sizes, it is concluded that a window size of 5 is best for this data as it smoothens the data best. After processing 43 products, a representative state is reached. This is equal to a warm-‐up period of 1 day. See Appendix C for a complete overview of the Welch method. Furthermore, a run length of 5 weeks (35 days) will be used, thus the end-‐time of the simulation will be set to 36 days (run length + warm-‐up).