The influence of card setting and card controlling on workload balancing
- A simulation study
Sjoerd Louis Meijs S1901028 University of Groningen Faculty of Economics and Business
Master Thesis - Technology & Operations Management
31 March 2017
Abstract
One of the objectives of pull production control systems is to improve the throughput times which can be achieved by balancing the workload. The balancing of workload exists for specific unit-based pull production control systems and this effect strongly depends on the number of cards and the processing times. Processing time variability has a major negative effect on the workload balancing capabilities. However, balancing the workload with dynamic card based systems is quite new and current research is lacking this issue. This study compares the throughput times of CONWIP and m-CONWIP in how the workload balancing capability copes with processing time variability when using card based systems. A discrete event simulation is simulated with the software Tecnomatix Plant Simulation and the results show that card controlling configurations with constant processing times outperforms push systems, but it does not mitigate the negative effect of processing time variability on workload balancing capability.
Table of Contents
Abstract ... 2
1. Introduction ... 4
2. Background ... 5
2.1 Production control systems ... 6
2.2 Pull systems ... 7
2.3 Workload balancing ... 10
2.4 Card based policies ... 12
3. Methodology ... 14
3.1 Research method ... 14
3.2 Model design ... 16
3.3 Experimental design... 20
4. Results and Discussion ... 26
1. Introduction
Since the introduction of Kanban there has been an increasing interest in card-based systems (González-R et al., 2012). One of the objectives of pull production control systems is to improve the throughput times which can be achieved by balancing the workload (Germs & Riezebos). However, there are very few papers that have been able to show the effective workload balancing capability of unit-based pull production control systems (Germs & Riezebos, 2010). Germs & Riezebos (2010) found that the balancing of workload exists for specific unit-based pull production control systems and that this effect strongly depends on the number of cards and the processing times.
The number of cards is the configuration of the pull system and can be either determined by card setting, which is static or card controlling, which is dynamic (Framinan et al., 2003). Card setting has a predetermined amount of cards and does not change during the process, while card controlling adds and subtracts cards during production processes. Therefore, card controlling configurations are much more flexible than card setting configurations. As for card controlling you can add and subtract cards during the process, it is not strange to think that card controlling might outperform card setting configurations. Knowing when control systems should use either card setting or card controlling policies when using constant or random processing times is therefore worth identifying.
enable workload balancing when processing time variability is present. The research question that follows is in what degree specific pull production control systems, with constant or random processing times, balances the workload for either card setting or card controlling policies.
In order to determine whether this could be the case, a discrete event simulation is simulated with the Tecnomatix Plant Simulation software (Siemens PLM Software, 2014). 142 experiments will be executed with specific structures and configurations that have either processing time variability or not and have card setting or card controlling policies. These experiments will be compared and results might show a significant performance improvement for workload balancing capabilities regarding card controlling policies.
The paper is structured as follows. First the current literature regarding card based pull systems will be elaborated. Then specific card setting and card controlling mechanisms will be identified. The methodology explains how the experiments are conducted and how data has been gathered. The output of the experiments will be discussed in detail in the results and discussion section.
2. Background
2.1 Production control systems
Since the introduction of the Just In Time philosophy, most organizations have placed an increasing attention to the way they control production (González-R et al., 2012). Production control has a major effect on the performance of a production organization which has a crucial impact on the company's competitiveness (Gaury et al., 2001). Production control is an
optimization problem that typically addresses the question of when and how much to produce in order to achieve a satisfactory customer service level (Liberopoulus & Dallery, 2000). This makes production control difficult. These difficulties that may be encountered when using production systems have led to the development of production control systems such as Kanban and CONWIP (Riezebos et al., 2003).
production control system (Ziengs et al., 2012). Figure 1 shows a graphical overview of where jobs are accepted, released and dispatched (Germs & Riezebos, 2010).
Figure 1. The acceptance, release and completion of jobs at a production control system.
2.2 Pull systems
Pull production control systems have two defining characteristics; structure and configuration (Gaury et al., 2001). The structure of a production system refers to the arrangement of control loops used to limit the work in progress on the shop floor (Gaury et al., 2001). The control loops that are used are specified by the structure of the pull system. The structure can be seen as a long term strategic decision while the configuration can be seen as a short term operational decision as it can change quickly. The most commonly known pull production control systems are Kanban and CONWIP. These structures will be shortly described.
Kanban has been widely described because it represented the first system to be termed a pull system (Hopp & Spearman, 2004). Kanban has quickly become very popular as it is a simple pull system that allows a job to be processed only if there is a demand for it, which is shown in figure 2 (González-R et al., 2012). Kanban provides multiple benefits for a production line, for example a reduced work in process, cycle time, smoother production flow, improved quality and reduced costs (Hopp & Spearman, 2004).
Spearman and Zazanis (1992) proposed a hybrid push/pull system known as the CONWIP (constant work in process) that possesses the benefits of Kanban, but can be applied to a more general manufacturing setting (Hopp and Spearman, 2004). In the CONWIP system, a card is attached to a job at the beginning of the manufacturing line. When a job is finished, the attached card is released and sent to the beginning of the line, where it can be attached to another job, which is shown in figure 3 (González-R et al., 2012).
The objective of CONWIP is to combine the limited work in process of Kanban with the high throughput of push systems (Gaury et al., 2001). Implementing CONWIP is much easier than Kanban because there are fewer control loops. A CONWIP system is actually a Kanban system that controls the whole production line (Gaury et al., 2001).
This limits the work in process that is in the production system but does not balance the workload across the routings in the production system. For example, we can use as many CONWIP loops as there are as much routings in the production system. We denote such a system as “m-CONWIP”, where the m stands for multiple CONWIP loops (Germs & Riezebos, 2010). Germs & Riezebos (2010) found that the balancing of workload exists for m-CONWIP but not for CONWIP and that this effect strongly depends on certain parameters, which will be elaborated later on.
The configuration indicates to the number of cards (Framinan et al., 2003). Pull production systems consist out of control loops that represents a part of the shop floor in which cards control the work in process. The number of cards determines the work in process limit that is controlled by the loop. An available card signals that downstream capacity is or soon will become available (Ziengs et al., 2012). When the card is attached, the job can continue to the next workstation where production can start. In essence, work in progress limits enable downstream information related to the availability of capacity which is signaled upstream and used in releasing or dispatching work (Ziengs et al., 2012). In pull systems, workstations may be blocked because there are no parts in the buffer so it is starving. Or because it is not allowed to produce due to the fact all cards are attached to products and they will be released only if the downstream workstation needs one of these products (Gaury et al., 2001).
2.3 Workload balancing
systems lack the capability to balance this workload. However, Land & Gaalman (1998), Breithaupt et al. (2002) and Land (2004) have been able to show the existence of the workload balancing capability in certain types of control systems and they found a significant effect on system performance. Additionally, Germs & Riezebos (2010) found that the balancing of workload exists for m-CONWIP and that this effect strongly depends on the number of cards and the processing time variability of the resources.
The workload is said to be balanced effectively when the total throughput time reduces in the system by restricting orders (Ziengs et al., 2012). This workload balancing capability of a pull system is shown in figure 4. Due to the fact average queue lengths will become lower, the total throughput time will reduce. Additionally, having a better workload balance between the routings causes an improved control of the arrival of jobs at the workstations on the shop floor. Pull systems in which the work in process is hard to locate workload balance capabilities are low (Germs & Riezebos, 2010). When these workload balance capabilities are low in the production system a reduction in shop floor throughput time, which is caused by an increase in order pool time, will cause an increase in total throughput time (Land, 2004). Therefore, it is crucial to choose a control system that offers opportunities to balance the workload which improves overall system performance by diminishing both shop floor throughput time and total throughput time.
processing times (Germs & Riezebos, 2010). Card based systems can possibly react on these different processing times. These card based systems will be discussed in the next section.
Figure 4. The workload balancing capability of a pull system.
2.4 Card based policies
In order to accomplish a balanced workload, cards are needed that are used to send information to the workstation upstream. There are two types of card based policies; card setting and card controlling (Framinan et al., 2003). Card setting is defined as a procedure to set the number of cards that makes the performance of the system acceptable (Framinan et al., 2006). The number of cards is fixed during the decision interval.
allowable amount of work in process will change as well. For example, when the production starts with a card count of five and there are two available cards left, there must be three orders in the system. Therefore, establishing the right amount of cards is an important decision and greatly affects the performance of the system (Framinan et al., 2006). In general, if the card count is too high the pull system will start to resemble a push system, but if the card count is too low orders will not be released, and starving will occur more frequently.
Germs & Riezebos (2010) found that the balancing of workload exists for specific structures and that this effect strongly depends on the number of cards and the processing time variability of the resources. Higher processing time variability gives less control over the system. While card controlling is known for its responsiveness on the systems’ state it might be a
solution to have a better grip on the system with high processing time variability because the card controlling mechanism can react on varieties that occur during the process, which is not the case for card setting. Framinan et al., (2006) state that card controlling could also achieve a throughput rate between two integer values of cards, but it is still something that needs to be investigated. Because the configuration of the card controlling system keeps changing over time, average card counts of for example 5,56 can be achieved. This gives the opportunity to achieve a more accurate desired average card count.
The literature on card controlling is rather limited and there are only a few methods. For example, Gupta & Al-Turki (1997) proposed a card controlling procedure for a Kanban system with two types of cards. Given an initial number of Kanbans, the procedure calculates the demand that is necessary and determines whether a card should be added or subtracted.
and Maaseidvaag (2001) developed a procedure for card controlling that is applied to CONWIP systems for a MTS environment. In the procedure, if the inventory level is under a certain threshold an ‘extra card’ is added to the system. There are a maximum number of extra cards that can be used.
Finally, Framinan et al., (2006) proposed a procedure with the objective to obtain a throughput rate for an MTO environment. Just like Tardif and Maaseidvaag (2001), they use a number of extra cards that will be added or subtracted to the system. The process of adding or subtracting cards depends on the throughput level of the system. If the throughput level is below a predefined threshold an extra card is added to the system. This will result in an increase of the production rate (Framinan et al., 2006). Vice versa, when the throughput level is above the threshold, the work in process can be reduced by subtracting an extra card from the system. This study will focus on the card controlling mechanism of Framinan et al., (2006). This mechanism is chosen because of its simplicity and it is also simulated for an MTO environment. The actual card controlling mechanism will be elaborated in the experimental design section.
3. Methodology 3.1 Research method
Table 1
A comparison of simulation models (Robinson, 2004).
Feature Spreadsheet Programming
language
Specialist simulation software
Range of application Low High Medium
Modeling flexibility Low High Medium
Duration of model build Medium Long Short
Ease of use Medium Low High
Ease of model validation Medium Low High
Run speed Low High Medium
Time to obtain software Short Low Medium
Price Low Low High
3.2 Model design
Figure 5. The three staged production system based on Germs & Riezebos, 2010.
Germs & Riezebos (2010) found that the m-CONWIP has workload balancing capability and that this is not the case for CONWIP. Therefore, using CONWIP as a benchmark, in order to see how workload balancing reacts on card controlling in combination with high processing time variability is an excellent strategy.
Figure 7 shows a graphical representation of the control loops that are used in the m-CONWIP. Additionally, table 2 shows the control loops that the CONWIP and m-CONWIP uses.
Figure 7. The four control loops of the m-CONWIP.
Table 2
Control loop placement for the experiments.
Structure Loop 1 Loop 2 Loop 3 Loop 4
CONWIP A:B:C:D:E:F:G n/a n/a n/a
m-CONWIP A:B:D A:B:E A:C:F A:C:G
days were used. The output of the two models were quite similar, as the model of Germs and Riezebos (2010) found for m-CONWIP a workload balancing capability; a TTT reduction of 5,87 percent and a STT reduction of 53,26 percent. In the simulated model similar results were found; a TTT reduction of 6,79 percent and a STT reduction of 48,93 percent. The difference can be because of the use of different simulation software or different programming skills. Additionally, for CONWIP no workload balancing capabilities was found. Furthermore, the simulated model showed that just like the model of Germs and Riezebos (2010) that high processing time variability has a major negative effect on workload balancing capability. Now that the model was validated, the experimental design is elaborated in the next section.
3.3 Experimental design
Table 3 Experimental factors
Factor: Experimental levels:
Structure CONWIP, m-CONWIP
Configuration Static, dynamic
Batch size 1 Card count: static dynamic 1-25, ∞ n+1, ∞
Inter arrival time Constant
Utilization 90%
Processing time Constant, random Erlang-2
The processing times for the different experiments are either constant or random Erlang-2 distributed. The Erlang-2 processing times are calculated and shown in table 4. The Erlang-2 distribution for processing times of 1 hour is shown in Appendix A.
Table 4
Erlang -2 distribution processing times
Workstation(s) Mean Processing time Standard Deviation Standard Deviation (t)
Workstation A 1 hour 0,707107 42 minutes 26 seconds
Workstation B and C 2 hours 1,414214 1 hour 24 minutes 51 seconds Workstation D, E, F and G 4 hours 2,828427 2 hours 49 minutes 42 seconds
starting card count will be set to twenty for CONWIP and five for m-CONWIP. These starting configurations for the card controlling are less important because the controlling mechanism will find a stable state. For conducting the experiments with the specific control loops and its specific card counts, the other card counts of the control loops need to be set to infinite. 142 series of experiments were executed. An overview of the series of experiments is shown in table 5.
Table 5
Experiments performed.
Card setting (static) Card controlling (dynamic) Constant processing times
CONWIP CONWIP 5-25, ∞ CONWIP 20
m-CONWIP m-CONWIP 1-25 m-CONWIP 5
Random processing times
CONWIP CONWIP 5-25, ∞ CONWIP 20
m-CONWIP m-CONWIP 1-25 m-CONWIP 5
In order to reduce the variance across experiments pseudo-random numbers will be used. In order to determine whether the warm-up period is sufficient the Welch’s procedure, detailed by Law and Kelton (2000), was executed. This method is used to reduce variations and determine when the experiments are stabilized in order to have trustworthy experiments. For the card setting experiments the warm-up period was set to fifteen days as it was derived from the Welch method which can be seen in Appendix B. The card setting experiments have a run length of 150 days as it should be ten times as much as the warm up time. The number of runs is twenty, which is sufficient according to Robinson (2004). For the card controlling experiments a warm-up time of 30 days and a run length of 300 days was simulated. This is because the fluctuation of the card count needed more time to stabilize and reduce the variation.
the order pool. The order release method checks whether a card is available for the specific control loop. When a card is available the card will be attached to the order and the order will move from the order pool to the buffer of the first workstation. When a card is attached the number of cards is reduced by one for that specific control loop. The order will move towards the consecutive workstation depending on its route. When the order is completed the card will be released again. When a card is released, the number of cards is increased by one for that specific control loop. This is the case for card setting experiments, as for each card attached an equivalent count card will be released. The card setting mechanism implemented in Tecnomatix Plant Simulation is shown in Appendix D.
Figure 8. Flowchart of the card controlling mechanism.
The throughput rate that is used as the center of the threshold is determined by the optimal throughput rate of the card setting experiments, which can be found in table 8 and 9 in the results section. With constant processing times m-CONWIP3 has the best throughput times and with random processing times the push system has the best throughput times.
at least one because else the system would starve due to the fact that there are no cards in the system left.
Table 6
Threshold levels in seconds for constant processing and random processing times Threshold level
(T)
Constant processing time (Lower Bound – Upper Bound)
Random processing time (Lower Bound – Upper Bound)
0% 10% 20% 30% 40% 50% 60% 70% 80% 90% 99% 73470 66123 58776 51429 44082 36735 29388 22041 14694 7347 735 73470 80817 88164 95511 102858 110205 117552 124899 132246 139594 146206 138089 124280 110471 96662 82853 69045 55236 41427 27618 13809 1381 138089 151898 165707 179516 193325 207134 220942 234751 248560 262369 274797
In order to draw conclusions from results, output needs to be collected from the different experiments. Table 7 describes the output values. Now that the experiments are described in detail the results will be discussed in the next section.
Table 7
Output values of the experiments.
Output Values: Description:
root.Summary["OrderpoolTime",1] Average Orderpool Time
4. Results and Discussion 4.1 Card setting
For the card setting experiment with constant processing times an optimal throughput rate of 73470,27 seconds was found for m-CONWIP with a card count of three (Table 8). For the card setting experiment with random processing times an optimal throughput rate of 138089,04 seconds was found for m-CONWIP with a card count of 23 (Table 9). These optimal throughput rates are used as the center of the threshold levels.
Table 8
Throughput rates for different card setting card counts with constant processing times.
Experiment STT TTT mCONWIP1 mCONWIP2 mCONWIP3 mCONWIP4 mCONWIP5 mCONWIP6 25200,00 30528,64 40251,74 49152,29 56080,17 61336,29 2758338,00 125940,09 73470,27 74306,57 74920,31 75427,99 Push system 78819,82 78819,82 Table 9
Throughput rates for different card setting card counts with random processing times.
As expected for the card setting experiments the simulated model found a workload balance capability for the m-CONWIP (Figure 9). The optimal throughput rate with a card count of three has a TTT reduction capability of 6,79 percent and a STT reduction of 48,93 percent.
Additionally, for CONWIP no workload balancing capability was found as the push system is the optimal set up (Figure 10). Furthermore, the simulated model showed as expected that high processing time variability a major negative effect has on the workload balancing capability (Figure 11). The results show a TTT reduction of 0,02 percent and a STT reduction of 0,63 percent.
Figure 10. Card setting CONWIP found no workload balancing capability
4.2 Card controlling
As mentioned in the experimental design section, the card controlling starting configurations CONWIP20 and m-CONWIP5 is used. This is due to the fact that the starting amount is not really important because the mechanism searches for the optimal state. Additionally, differences in starting card counts only showed negligible differences in throughput times. The results of the card controlling for constant processing times show that the CONWIP can hardly benefit from other configurations compared to the push system. The total throughput time can only be reduced by 0,14 percent at a threshold level of zero percent (table 10). Furthermore results show that for the m-CONWIP with constant processing times, the system can benefit from a total throughput reduction of 4,86 percent compared to the push system at a threshold level of 98 percent (table 11). However, the card controlling mechanism with constant processing times does not improve the performance compared to the card setting mechanism. As the card setting mechanism had a total throughput reduction of 6,79 percent compared to the push system.
Table 10
Card controlling CONWIP20 results with constant processing times.
Table 11
Card controlling m-CONWIP5 results with constant processing times.
Threshold (T) m-CONWIP5 TTT reduction STT reduction
0% 10% 20% 30% 40% 50% 60% 70% 80% 90% 97% 98% 99% 78623,08 78687,50 78645,13 78570,52 78412,00 78295,82 78163,37 77243,88 76879,30 76772,20 75537,92 74989,03 74996,77 0,25% 0,17% 0,22% 0,32% 0,52% 0,66% 0,83% 2,00% 2,46% 2,60% 4,16% 4,86% 4,85% 0,25% 0,17% 0,22% 0,32% 0,52% 0,66% 0,83% 2,00% 2,46% 2,60% 4,16% 4,86% 4,85%
Figure 12. Sensitivity of the card controlling for CONWIP and m-CONWIP with constant processing times.
Table 12
Average card counts of the CONWIP and m-CONWIP Threshold (T) m-CONWIP5 Average
The results of the card controlling for random processing times show that the CONWIP cannot benefit from other configurations compared to the push system. The total throughput time cannot be reduced as shown in (table 13). Unlike the m-CONWIP experiments with constant processing times, for random processing times there is no improvement possible compared to the push system. For both CONWIP as m-CONWIP with random processing times, order pool times dramatically increase at a threshold level at around 70 percent (figure 13). Due to the fact that the total throughput times do not get below the 70 percent threshold level cards will get subtracted from the system .Meaning cards do not get released and the order pool times increases.
Another aspect that is noteworthy is the difference of the threshold levels of constant and random processing times when the total throughput times increases. Probably because the total throughput times for random processing times are higher and the variability is higher. This makes it easier for the throughput times to reach below the lower bound of the threshold level and still add cards to the system.
Table 13
Card controlling CONWIP20 results with random processing times.
Threshold (T) CONWIP20 TTT reduction STT reduction
Table 14
Card controlling m-CONWIP5 results with random processing times.
Threshold (T) m-CONWIP5 TTT reduction STT reduction
0% 10% 20% 30% 40% 50% 60% 70% 80% 90% 99% 142020,96 141678,48 141394,56 141263,25 142227,15 142496,48 143084,29 144944,01 178097,34 268547,64 299349,42 -2,91% -2,66% -2,46% -2,36% -3,06% -3,25% -3,68% -5,03% -29,05% -94,59% -116,91% 1,25% 1,31% 1,39% 1,62% 2,07% 3,21% 4,22% 6,19% 13,94% 39,44% 44,33%
5. Conclusion
The results showed that card controlling is not able to cope and mitigate the random processing times better than the card setting mechanism. CONWIP and m-CONWIP with random
processing times shows total throughput times which are higher than the push system. However, just like card setting it does outperform the push system with constant processing times. The card controlling mechanism is easier to implement than the card setting policy as you only need to know the average throughput times and the system will find the optimal state. While for card setting you have to experiment on which system state performs best.
Earlier research found that processing time variability makes workload balancing capability impossible and in this study card controlling is not a solution that copes with the processing time variability. However, further research that provides a more accurate card controlling mechanism might outperform the push system. For example, the card controlling mechanism has its limitations as it only focuses on the system state at the moment when the order is completed and based on that it changes the card count of the system. But it does not take into account the orders that are still in the system that also influences the systems’ performance. Meaning, you can say that the reaction of the card controlling is lagging. More accurate card controlling mechanisms that take the actual work in progress into account might have even lower throughput time outcomes as it better predicts and reacts on the systems’ state.
As the optimal system state for m-CONWIP lies around a card count of three, a card controlling mechanism that also adds maximum and minimum card counts might be even more precise because it can moderate variations in the system.
References
Breithaupt, J.W., Land, M.J., & Nyhuis, P. (2002). The workload control concept: theory and practical extensions of Load Oriented Order Release. Production Planning and Control, 13(7), 625-638.
Framinan, J.M., González, P.L., Ruiz-Usano, R., (2003). The CONWIP production control system: Review and research issues. Production Planning & Control 14(3), 255– 265. Gaury, E. G. A., Kleijnen, J. P. C., & Pierreval, H. (2001). A Methodology to Customize Pull
Control Systems. The Journal of the Operational Research Society, 52(7), 789–799. Germs, R., & Riezebos, J. (2010). Workload balancing capability of pull systems in MTO
production. International Journal of Production Research, 48(8), 2345–2360.
González-R, P. L., Framinan, J. M., & Pierreval, H. (2012). Token-based pull production control systems: An introductory overview. Journal of Intelligent Manufacturing, 23(1), 5–22. Hopp, W.J., & Spearman, M.L., (1996). Factory Physics: Foundations of Manufacturing
Management. Irwin, Chicago.
Hopp, W. J., & Spearman, M. L. (2004). To Pull or Not to Pull: What Is the Question? Manufacturing & Service Operations Management, 6(2), 133–148.
Kimura, 0., & Terada, H. (1981). Design and analysis of pull systems, a method of multistage production control. International Journal of Production Research, 19(3), 241-253. Land, M.J., & Gaalman, G.J.C. (1998). The performance of workload control concepts in job
shops: improving the release method. International Journal of Production Economics, 56-57, 347-364.
Law, A.M. & Kelton, W.D., (2000). Simulation modeling and analysis. Boston: McGraw-Hill. Liberopoulos, G., & Dallery, Y. (2000). A unified framework for pull control mechanisms in
multi-stage manufacturing systems, Operation Research, 93, 325–355.
Rafiei, H., Rabbani, M., & Alimardani, M. (2013). Novel bi-level hierarchical production planning in hybrid MTS/MTO production contexts, Internation Journal of Production Research, 51(5), 1331-1346.
Riezebos, J., (2010). Design of POLCA material control systems, International Journal of
Production Research, 48(5), 1455–1477.
Riezebos, J., Korte, G.J., & Land, M. J. (2003). Improving a practical DBR buffering approach using Workload Control, International Journal of Production Research, 41(4), 699-712. Robinson, S., (2004). Simulation: the practice of model development and use. Palgrave
Macmillan.
Siemens PLM Software. (2014). Tecnomatix Plant Simulation [Computer program]. https://www.plm.automation.siemens.com/en/products/tecnomatix/manufacturing-simulation/material-flow/plant-simulation.shtml
Spearman, M. L., Woodruff, D. L., & Hopp, W. J. (1990), CONWIP - A pull alternative to Kanban, International Journal of Production Research, 28(5), 879–894.
Spearman, M. L., & Zazanis, M. A. (1992). Push and pull production systems: Issues and comparisons. Operation Research, 40, 521–532.
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 of Production Research, 43(5), 869-898.
Ziengs, N., Riezebos, J., & Germs, R. (2012). Placement of Effective Work-In-Progress Limits in Route-Specific Unit-Based Pull Systems. International Journal of Production
Appendix A
Figure A.1 Erlang distribution at mean=1 and k =2
Appendix B
Appendix C
Appendix D
Figure D.1 Card setting mechanism in Tecnomatix Plant Simulation.
Appendix E
