UNIVERSITY OF GRONINGEN
FACULTY OF ECONOMICS AND BUSINESS Master Thesis – Technology & Operations Management
Simulating the influence of the structure of unit-based pull production control systems on the workload balancing capability
Author: Supervisor:
Christian Rodriguez Mallo - S1973916 N. Ziengs, Msc
Co-assessor:
Dr. J. Riezebos
20 June 2016
Academic Year 2015/2016
Abstract
Unit-based pull production control systems limit the amount of work that is allowed onto the production floor. As a result, a better control over the throughput time of orders is achieved.
Balancing the workload in production systems is possible if orders can follow several routes and the control is route specific. This is important since workload balancing can improve the throughput times of orders. Pull production control systems vary in terms of structure and configuration. This paper shows how the structure and configuration affect the workload balancing capability of unit-based pull production control systems in a divergent multi-stage production line.
The results show that the exchange of route-specific information downstream is crucial to successfully balance the workload. It is shown that this works best with an m-CONWIP structure.
Furthermore the decision concerning the configuration entails a trade-off between reducing the throughput time and possible other performance indicators.
Keywords: Pull production control, unit-based, generic model, simulation, structure,
configuration, discrete event simulation
Inhoud
Abstract ... 2
Preface... 6
1. Introduction ... 7
2. Background ... 9
2.1. Production planning and control ... 9
2.2. Pull production control systems ... 10
2.3. Push production control systems ... 11
2.4. Pull systems structure ... 12
2.5. Workload Balancing ... 15
3. Methodology ... 18
3.1. Research Method ... 18
3.2. Model Design ... 20
3.3. Experimental Design ... 21
3.4 Simulation Validity ... 23
3.4.1 Warm-up Period ... 23
3.4.2 Number or runs ... 24
4. Results ... 25
4.1 Outcomes – Experiment series 1... 25
4.1.1 Total Throughput Time ... 26
4.1.2 Shop floor Throughput Time ... 26
4.1.3 Order pool time ... 26
4.1.4 Configuration ... 27
4.2. Outcomes – Experiment series 2... 30
4.2.1 Total Throughput Time ... 30
4.2.2 Shop floor Throughput Time ... 30
4.2.3 Configuration ... 31
5. Discussion ... 32
5.1 CONWIP ... 32
5.2 Kanban ... 33
5.3 POLCA ... 34
5.4 Base Stock Policy ... 36
5.5 m-CONWIP ... 37
5.6 Limitations ... 38
5.7 Theoretical Implications ... 39
5.8 Managerial Implications ... 39
6. Conclusion ... 40
6.1 Future Research ... 42
References ... 44
Appendix A ... 48
Appendix B ... 49
Appendix C ... 50
Appendix D ... 51
Preface
Over the course of the last six months I have enjoyed working on this thesis although
acknowledging the well-known and expected ups and downs. As expected I too have not only
relied on myself during the creation of this work. I would therefore first like to thank the
supervisor of this work, Nick Ziengs, for his close involvement and ever so fruitful discussions
whenever doubts and problems came about. Thanks to his extensive feedback and flexible
attitude I was able to surpass most of the hurdles encountered during the elaboration of my
thesis. Furthermore, I would like to mention my pleasant collaboration with Fiodor Bodnar who
has been a great source of motivation and for our extensive discussions on our thoughts and
work. I would also like to thank the co-assessor of this work Dr. Jan Riezebos for his time and
effort in evaluating this work. Finally I would like to thank my parents, sister and girlfriend for
their unconditional support and good care all through the last six months of occasional stress and
doubts.
1. Introduction
The throughput time performance of a production line is largely determined by controlling the work in process that is allowed onto the shop floor (Ziengs, Riezebos, & Germs, 2012). Hopp
& Spearman (2004) define such mechanisms that limit work in process as pull systems. There is a large variety of pull-type systems, for an extensive review of all the systems we refer to González- R, Framinan, & Pierreval (2012). The commonly known systems are Kanban, CONWIP, m- CONWIP and POLCA. These systems distinguish themselves by their structure. Which means, at which place or places in a production line the WIP is limited. By how much the WIP it is limited, is referred to as the configuration. In literature the performance of these systems has been related to their workload balancing capabilities (Germs & Riezebos 2009). However, the influence of the difference in structure and configuration on the workload balancing capability of pull production systems is not yet thoroughly discussed. However, Ziengs et al. (2012) argue that workload balancing can help companies gain a competitive advantage. The focus of this thesis is therefore on determining how the structure and configuration of pull production control systems influence the workload balancing capability. To achieve this a flexible model is developed which can simulate pull production systems for a divergent production line.
The workload balancing capability of varying structures and configurations of pull
production systems in divergent production lines is a relevant problem since previous research has
focused mainly on straight production lines (Liberopoulos & Dallery 2000; Gaury et al. 2000). By
looking at a divergent production line we introduce additional possibilities of variability, namely,
variability in the routing of orders.
An extensive search for a generic control system design has shown that there is little done in this aspect. González-R & Framinan (2009) and Gaury et al. (2000) have designed a generic model to make a selection between KANBAN, CONWIP and a hybrid, although this was for a straight line. However as we have learned from Germs & Riezebos (2009), the most effective workload balancing control systems for divergent production lines are POLCA and m-CONWIP.
Germs & Riezebos (2009) found that route-specific control systems perform better in terms of workload balancing capability. Germs & Riezebos (2009) included several types of possible variability in their simulations and have only looked at CONWIP, POLCA and m-CONWIP. In order to test whether route-specific information is indeed crucial, it is important to isolate route variability as the only influential factor. Furthermore structures that do not necessarily allow for route-specific information to be available at the order release should also be included. We therefore aim to show how the difference in structure and configuration actually influences the workload balancing mechanism. Hereby focusing on how these system characteristics cope with routing- variability. This is executed with a unit-based release mechanism which releases order based on the number of orders on the shop floor. This release mechanism is chosen since it is more commonly used in practice, due to its relative simplicity compared to a load-based control. Which releases orders based on the capacity in terms of the time necessary to process an order.
In conclusion, this research will provide insights into the influence of varying
characteristics which come along with the different structures and configurations of unit-based
pull production systems, on the workload balancing capability of pull production systems in a
divergent production line.
2. Background 2.1. Production planning and control
Production systems can be defined as “a set of interrelated elements that are designed to act in a manner that generates final products whose commercial value exceeds the cost of generating them” (Maccarthy & Fernandes, 2000, p.485). Production systems tend to be complex systems and controlling them therefore, poses a challenge. In order to gain some control over the performance of production systems, production planning and control (PPC) systems have been developed. The main purposes of a PPC system are that: 1) they help determine the amount to be produced in order to fulfill customer demand, 2) they are used to plan the order of raw materials, 3) they balance the available resources (capacity) and 4) they control and plan order release to the system (Zäpfel & Missbauer, 1993). Gaury, Kleijnen and Pierreval (2001) state that the influence of PPC systems is major in terms of inventories, production delays and make span, and thus influencing the overall competitiveness of a company. Although the major influence of PPC is acknowledged in literature, the questions remains which PPC best copes with the often changing state of companies and their production systems. (Banerjee 1996, p.58), for example, states that
“Experience has shown that while millions have been spent on manufacturing systems [...] no real solution to the need of greater responsiveness and flexibility has been found”.
This should further demonstrate to production companies the importance of thoroughly investigating which PPC system to use. In order to choose the right PPC system one must have an understanding of the characteristics of the different options.
The general distinction within production control can be made between push and pull
control systems. The main difference between push and pull systems is that in pull systems a
production process at a certain stage of the system is started to fill the gap left by a part that has gone on to the next stage, while push systems produce to fulfill demand without considering the status of the system (Akturk & Erhun 1999; Hopp & Spearman 2004). To get a better understanding of the individual control systems they are discussed at length in the next sections.
2.2. Pull production control systems
Pull production control systems are used to control throughput times of orders by limiting the amount of work in process (WIP) in a system or parts thereof (Hopp & Spearman 2004). The interest for pull systems was sparked by the introduction of the just-in-time (JIT) philosophy on production in the late seventies which aims to balance service level with a minimal WIP level (González-R, Framinan, & Pierreval, 2012). The service level is defined by Karaesmen & Dallery (2000) as the fill rate which represents the proportion of demand that can be satisfied from inventory at the moment of order arrival.
The foundation of JIT production and therefore pull systems lays in Japan. In the seventies,
MRP and other computerized control systems were gaining traction in American and European
industries. Meanwhile in Japan, partly due to the lack of computerization of their industry, a more
resource conservative control system was emerging (Hopp & Spearman 2004). Under the watchful
eye of Taiichi Ohno, Toyota started to implement the “Toyota Production System”. This system
focused on making as much goods as possible, but in a continuous flow (Ōno 1988). When the
times of abundant demand started to diminish all over the world, and efficiency started to become
important in order to maintain the business results, American and European industries began to
grasp the value of JIT production.
The Toyota Production System includes a card-based pull system known as Kanban (Liberopoulos & Dallery 2000). In card-based pull systems the WIP is controlled by control loops in the system. The amount of cards within such control loops determines the amount to which WIP is limited. Whenever a downstream production stage is available a card is released to the upstream station, signaling that work can be released (Ziengs et al., 2012). All card-based pull production systems work in this same general fashion, nevertheless there are several ways to implement this basic mechanism. The implementation is defined by the characteristics of the systems, such as the structure and configuration.
Pull production systems can also be characterized by the way they release orders. Pull systems can either be route-specific or non-route-specific and additionally product-anonymous or product-specific (Germs & Riezebos 2009). Route-specific control systems do not look at the product type but rather have cards for product routes. This means that to release an order at the right time it needs to be aware of the capacity at all the workstations that lay within the (partial) route of the arrived order. Non-route-specific control however disregards information concerning the capacity within the route. Product-anonymous control disregards all the information concerning the order and just releases an order based on the availability of a card. A visualization representation of these different types adopted from (Ziengs et al. 2012, p.4360) can be found in appendix A.
2.3. Push production control systems
The opposite of pull systems, as the name implies, are push systems. In general, push
systems do not produce depending on the status of the system such as pull systems but rather work
towards a predetermined target. Work is then released onto the work floor either constantly or following a pre-determined schedule; the so-called Master Production Schedule (MPS) (González- R et al., 2012). Another, more concise, definition of the difference between push and pull is given by Bonney et al. (1999). They state that the flows of information and materials in push system go in the same direction, while in pull systems the flows move in contrary directions. This distinction is rather important for this thesis considering that we are focusing on the transfer of route-specific information within the system.
2.4. Pull systems structure
As mentioned in section 2.2 card-based pull systems distinguish themselves by their structure and configuration (Gaury et al. 2001). The structure is determined by the placement and size of control loops, whereas the configuration entails the WIP limit that is established within such a control loop. A control loop can contain one or multiple workstations, depending on the chosen control system (González-R et al. 2012). We will now consider the structure of KANBAN, CONWIP, m-CONWIP and POLCA in more detail since these pull systems have previously been used in similar studies (Germs & Riezebos 2009; Ziengs et al. 2012) that addressed workload balancing capabilities of unit-based pull production control systems. Furthermore these are the systems that are simulated in this thesis.
Kanban uses cards to control and limit the amount of jobs that are released between workstations and thus place a control loop containing only one workstation. A Kanban control loop can be placed at every workstation of a production system or at parts thereof. (Sharma &
Agrawal 2009). A schematic view of the basic structure of KANBAN is shown in figure 1. Because
the control loops only contain a single workstation the only information that is transferred between control loops is the availability of a card.
Figure 1. Schematic representation of the KANBAN structure
CONWIP does not allow for placement of control loops at specific workstations. Instead
CONWIP is based on limiting the WIP of the whole line. The order release is triggered by a
finished product at the end of the line. Again leaving information concerning the capacity of
individual workstations out of consideration. A schematic view of a simple CONWIP structure is
shown in figure 2.
Figure 2. Schematic representation of CONWIP structure
The control loop in an m-CONWIP system contains all the workstation of a certain route within the system. This means that the information flow accounts for the capacity of a whole route when releasing a card to the queue. A schematic of this structure can be seen in figure 3.
Figure 3. Schematic representation of m-CONWIP structure
POLCA is a pull production system where the control loops overlap each other. This means
that whenever an order is finished within a loop the system will wait with releasing the card to be
attached to a new order until a card of the successive loop has become available. Hence, the finished order can continue. A schematic of this structure is shown in figure 4.
Figure 4. Schematic representation of POLCA structure
2.5. Workload Balancing
Workload balancing is the ability of a production system to stabilize the queues before the
workstation based on a certain order release strategy (Land & Gaalman 1996). Pull control systems
are such control strategies. In order to balance workload the control strategy has to be able to
prioritize orders based on the available capacity upstream. This implies that if more queues exist
in a system there are more points at which workload can be balanced. Ziengs et al. (2012) however,
found that it isn’t necessarily better to balance the workload at as much points as possible. They
found that the structure and limit of control loops within a system have a great influence on the
workload balancing capabilities of that system (Ziengs et al., 2012).
This raises the question of how the structure and configuration actually influence the workload balancing capability. To address this question we will discuss how and where each of the systems described in section 2.4 would actually balance workload.
As can be seen in figure 1-4 some systems create more queues than others. Kanban being the system where the most queues are created (7) and CONWIP the system with the least queues, namely one. In theory a Kanban system therefore contains more chances to balance the workload.
However, we have already established that in a divergent topology where routing variability exists
the benefits of workload balancing can be annihilated if orders are in the queue for too long due to
lacking information on upstream capacity. We therefore expect that systems with a structure that
allow for (partial) route-specific information to reach the point of order release will balance the
workload better. If we look at the structures as described in section 2.4 we identify two systems
which allow for route-specific information to travel between loops, namely m-CONWIP and
POLCA. In an m-CONWIP system a loop contains all the workstations of a certain route and
workload is balanced at arrival. Whereas for POLCA, the workload is balanced at the queue that
is created at the workstation where two control loops overlap. So the information concerning a part
of the route is made available to the whole previous control loop at the point where workload is
balanced. However, if we look at CONWIP and KANBAN we see that it is non-route-specific.
Based on the background as described in this chapter 1 we expect that the workload balancing capabilities of unit based pull production control systems may be closely related to its ability to transfer route-specific information downstream. We have seen that the structure does influence the capability of passing this information downstream due to how and which workstations are placed within a control loop. To thoroughly investigate the influence of the structure and configuration a flexible model should be developed that allows the simulation of systems that are route-specific as well as non-route-specific. In this case the systems of interest are m-CONWIP, CONWIP, Kanban and POLCA . Although previous research has already proven that some of these systems have workload balancing capabilities, we can better evaluate the influence of the structure and configuration by isolating the route variability from other possible causes that may influence the performance of these systems. Furthermore, we expect that the configuration that is chosen for these systems is of importance since we want to find the optimal performance of the structures. Which, as we have established, not only depends on the reduction of throughput times but also obtaining a certain service level.
In the end the following research question and sub-questions are to be answered:
How does the structure of a unit-based pull system affect its workload balancing capability?
SQ1: Where in the production line should workload be limited to achieve workload balancing?
SQ2: Which control system configurations work best for which structure?
1