USF Libraries
USF Digital Collections

Performance evaluation of multi-product Kanban-like control systems

MISSING IMAGE

Material Information

Title:
Performance evaluation of multi-product Kanban-like control systems
Physical Description:
Book
Language:
English
Creator:
Deokar, Sachin S
Publisher:
University of South Florida
Place of Publication:
Tampa, Fla.
Publication Date:

Subjects

Subjects / Keywords:
dedicated cards
shared cards
Just-in-time
simulation
breakdowns
Dissertations, Academic -- Industrial Engineering -- Masters -- USF   ( lcsh )
Genre:
government publication (state, provincial, terriorial, dependent)   ( marcgt )
bibliography   ( marcgt )
theses   ( marcgt )
non-fiction   ( marcgt )

Notes

Summary:
ABSTRACT: Over the years, much attention has been given to the analysis of the pull type ordering system to reduce in-process inventory and to improve product quality. Kanban Control Systems are widely used to control the release of parts in multi-stage manufacturing systems operating under a pull mechanism. Considerable research has been done to study the individual manufacturing systems for multi stage and single product. However, not much research has been done to compare different pull control policies for multi product manufacturing systems. Most of the research done in multi-product system assumes that a kanban card is dedicated to a part type. The aim of this research is to compare the Kanban Control System (KCS), Generalized Kanban Control System (GKCS) and Extended Kanban Control System (EKCS) in the context of multi-product manufacturing systems where the kanban cards are either dedicated to a single part type or shared among the different part types.In this study, we analyze the performance of various control policies for a multi-product multi-stage manufacturing system. The manufacturing system considered in this research use a single-card kanban system, where the transportation of materials between the different work-centers is controlled by production kanbans. Demands that arrive to the system are satisfied from the finished goods inventory whenever possible and are backordered otherwise. Performance measures are number of backorders, average waiting time of backordered demand and average work in process. Our results show that Shared GKCS has lower number of backorders when the variability in the processing time is low, while Shared EKCS performs better when variability in the processing time is high. Trade off analysis was performed on average WIP and time to satisfy backorders. The Shared EKCS makes a better service-inventory compromise than traditional KCS.The Shared GKCS results in lower average waiting time to satisfy the backordered demand indicating responsiveness of this control system. The overall results indicate GKCS and EKCS with dedicated or shared kanbans outclass kanban control policy. The shared kanban-like control systems outperform dedicated control systems for all performance measures considered in this research.
Thesis:
Thesis (M.S.I.E.)--University of South Florida, 2004.
Bibliography:
Includes bibliographical references.
System Details:
System requirements: World Wide Web browser and PDF reader.
System Details:
Mode of access: World Wide Web.
Statement of Responsibility:
by Sachin S. Deokar.
General Note:
Title from PDF of title page.
General Note:
Document formatted into pages; contains 71 pages.

Record Information

Source Institution:
University of South Florida Library
Holding Location:
University of South Florida
Rights Management:
All applicable rights reserved by the source institution and holding location.
Resource Identifier:
aleph - 001498275
oclc - 57722962
notis - AJU6880
usfldc doi - E14-SFE0000558
usfldc handle - e14.558
System ID:
SFS0025249:00001


This item is only available as the following downloads:


Full Text

PAGE 1

Performance Evaluation of Multi-product Kanban-like Control Systems by Sachin S. Deokar A thesis submitted in partial fulfillment of the requirements for the degree of Master of Science in Industrial Engineering Department of Industrial and Management Systems Engineering College of Engineering University of South Florida Major Professor: Suresh K. Khator, Ph.D. Geoffrey Okogbaa Ph.D. James Stock, Ph.D. Date of Approval: November 10, 2004 Keywords: just-in-time, dedicated cards, shared cards, breakdowns, simulation Copyright 2004, Sachin Deokar

PAGE 2

Dedication To my late sister Neelima

PAGE 3

Acknowledgements I would like to express my gratitude and appreciation to my major professor Dr. Suresh Khator for his guidance, advice and comments on this research. I would also like to thank Dr. Geoffrey Okogbaa and Dr. James Stock for their thoughtful comments. I appreciate the financ ial support offer ed by Educational Outreach during my graduate studies at USF. I am indebted to my parents, espe cially my mom fo r her blessings, encouragement and patience throughout my education.

PAGE 4

i Table of Contents List of T ables .....................................................................................................iv List of Figures....................................................................................................v Abstract ............................................................................................................vii Chapter One Manufacturi ng Control S ystems....................................................1 1.1 Introduction to Push-pu ll Control S ystems..................................1 1.2 Kanban Ca rds................................................................................3 1.2.1 Single Card System......................................................3 1.2.2 Dual Card System........................................................4 1.3 Production Contro l Systems...........................................................5 1.3.1 Base Stock Cont rol System .........................................6 1.3.2 Kanban Contro l System ..............................................6 1.3.3 Generalized Kanban Control Syst em.......................... 8 1.3.4 Extended Kanban Control S ystem....................................10 1.4 Overvi ew......................................................................................11 Chapter Two Multi-product Manufacturing S ystems.........................................13 2.1 Dedicated and Shar ed Kanban Cards..........................................13 2.2 Dedicated Production Control Syst ems........................................15 2.2.1 Dedicated Kanban Control S ystem...................................15 2.2.2 Dedicated Generalized Kanban Control System...............16 2.2.3 Shared Extended Kan ban Control S ystem........................17 2.3 Shared Production Co ntrol Syst ems............................................17 2.3.1 Shared Kanban C ontrol Syst em........................................18

PAGE 5

ii 2.3.2 Shared Generalized K anban Control System....................19 2.3.3 Shared Extended Kan ban Control S ystem........................19 2.4 Summa ry......................................................................................20 Chapter Three Problem Defini tion....................................................................21 3.1 Problem Description and Object ives............................................21 3.2 Modeling A ssumptions .................................................................22 3.3 System Parameters......................................................................23 3.3.1 Demand A rrival..................................................................23 3.3.2 Processing Times..............................................................24 3.3.3 Time Betw een Failur es......................................................25 3.3.4 Service Levels ...................................................................26 3.3.5 Number of Kanban Cards..................................................26 3.4 Simulati on Model ..........................................................................27 3.4.1 Length of Si mulation ..........................................................28 3.4.2 Number of Replicatio ns.....................................................29 3.5 Design of Experiment ...................................................................30 3.6 Summa ry......................................................................................31 Chapter Four Result s and Analysi s .................................................................32 4.1 Simulation Results .......................................................................32 4.2 Analysis of Result s.......................................................................34 4.2.1 Analysis of Result s for Backorders at 95% Service Le vel.....................................................................34 4.2.2 Analysis of Results Fa ctor Interactions at 95% Service Le vel.....................................................................35 4.2.3 Analysis of Result s for Backorders at 98% Service Le vel.....................................................................39 4.2.4 Analysis of Results Fa ctor Interactions at 98% Service Le vel.....................................................................42

PAGE 6

iii 4.2.5 Analysis of Results for Average Work In Process.............45 4.2.6 Analysis of Results for Average Waiting Time for Backordered Demand .......................................................48 4.3 Summa ry......................................................................................51 Chapter Five Summary and Conclu sions..........................................................52 5.1 Summa ry......................................................................................52 5.2 Conclusions.................................................................................53 5.3 Applicat ions..................................................................................54 5.4 Future Research..........................................................................57 Referenc es ........................................................................................................59

PAGE 7

iv List of Tables Table 1 Processi ng Times...........................................................................24 Table 2 Time Between Failur es and Repair Times.....................................25 Table 3 Number Of Kanban Cards and Target Invent ory Level..................27 Table 4 The Factors And Co nditions Of Ex periments.................................31 Table 5 Number of Backorders for 95 % Service Level..................................33 Table 6 Number of Backorders for 98 % Service Level..................................40 Table 7 Average Work-in-process at 95 % Serv ice Level ...........................46 Table 8 Average Work-in-process at 98 % Serv ice Level ...........................47 Table 9 Average Backorder Time In System (BTIS) at 95 % Service Le vel..................................................................................49 Table 10 Average Backorder Time In System (BTIS) at 98 % Service Le vel..................................................................................50

PAGE 8

v List of Figures Figure 1 Single Card Pu ll Control Syst em.................................................. 4 Figure 2 Dual Card Pull Control Syst em..................................................... 5 Figure 3 Base Stock C ontrol System ......................................................... 6 Figure 4 Kanban Cont rol System ............................................................... 7 Figure 5 Generalized Kanban Control Syst em........................................... 9 Figure 6 Extended Kanba n Control S ystem............................................... 11 Figure 7 Multi-product M anufacturing S ystem............................................ 13 Figure 8 Dedicated Kanban Cards............................................................. 14 Figure 9 Shared Kanban Card s.................................................................. 15 Figure 10 Dedicated K anban Control System .............................................. 15 Figure 11 Dedicated Generaliz ed Kanban Contro l System .......................... 16 Figure 12 Dedicated Exten ded Kanban Control System .............................. 17 Figure 13 Shared Kanban Control S ystem................................................... 18 Figure 14 Shared Generalized Kanban Control System.............................. 19 Figure 15 Shared Extended K anban Control System ................................... 19 Figure 16 Within Run WIP Plots fo r KCS..................................................... 29 Figure 17 Analysis of Variance for Backorders for 95 % Service Level........ 34 Figure 18 Interaction Plot for Ba ckorders (System vs. Coefficient of Variation) at 95% Service Lev el................................................... 36

PAGE 9

vi Figure 19 Interaction Plot for Backorders (System vs. Number of Stages) at 95% Service Lev el...................................................... 37 Figure 20 Analysis of Variance fo r Backorders for 98 % Service Level 41 Figure 21 Interaction Plot for Ba ckorders (System vs. Coefficient of Variation) at 98% Service Lev el................................................... 42 Figure 22 Interaction Plot for Backorders (System vs. Number of Stages) at 98% Service Lev el...................................................... 43

PAGE 10

vii Performance Evaluation of Multiproduct Kanban-like Control Systems Sachin Deokar ABSTRACT Over the years, much attention has been given to the analysis of the pull type ordering system to reduce in-process inventory and to improve product quality. Kanban Control Systems are widely used to control the release of parts in multi-stage manufacturing systems operating under a pull mechanism. Considerable research has been done to study the individual manufacturing systems for multi stage and single product. However, not much research has been done to compare different pull c ontrol policies for multi product manufacturing systems. Most of the research done in multiproduct system assumes that a kanban card is dedicated to a part type. The aim of this research is to compare the Kanban Control System (K CS), Generalized Kanban C ontrol System (GKCS) and Extended Kanban Control System (EKCS) in the context of multi-product manufacturing systems where the kanban card s are either dedicated to a single part type or shared among the different par t types. In this study, we analyze the performance of various control polic ies for a multi-product multi-stage manufacturing system. The manufacturing syst em considered in this research

PAGE 11

viii use a single-card kanban system, where the transportation of materials between the different work-centers is controlled by production kanbans. Demands that arrive to the system are satisfied from the finished goods inventory whenever possible and are backordered otherwise. Performance measures are number of backorders, average waitin g time of backordered demand and average work in process. Our results show that Shared GKCS has lower number of backorders when the variability in the processing time is low, while Shared EKCS performs better when variability in the processing time is high. Trade off analysis was performed on average WIP and time to satisfy backorders. The Shared EKCS makes a better service-inventory compromi se than traditional KCS. The Shared GKCS results in lower average waiting ti me to satisfy the backordered demand indicating responsiveness of this control system. The overall results indicate GKCS and EKCS with dedicated or shared kanbans outclass kanban control policy. The shared kanban-like control systems outperform dedicated control systems for a ll performance measur es considered in this research.

PAGE 12

1 Chapter One Manufacturing Control Systems For years, manufacturing organizations have shown interest in the study and analysis of production control mechani sms for manufacturing systems that reduce work in process and lead times. According to the flow of the material in the manufacturing system, production contro l mechanisms are classified into push and pull type control systems. The aim of a pull system is to produce as much as needed, while that of push system is to produce as much as possible. 1.1 Introduction to Push-Pull Control Systems In push control systems, the production schedule triggers work release in the manufacturing system. Orders arrive at the first stage based on demand forecasts or production orders for future consumption. As soon as the work is completed at a workstation the part is pushed to the downstream work station. Production schedule in a push control system is based on the demand forecast to control the flow of material from an upstream workstation to a downstream workstation. Demand forecasts in a push control system are made for inventory levels or work-in-process at each produc tion stage. To avoid incorrect demand forecasts and keep a satisfactory safety stock, the in-process inventories are often kept at high level, which can result in unnecessary holding costs and

PAGE 13

2 production-related problems. The major drawback of a push control system is high work in process inventory and fore casting error can result in excess inventory and longer lead times. Much attention has been given to t he analysis of the pull type ordering system to reduce the in-process inventory and improve product quality. In order to control the flow of materials in a manufacturing system, pull type control mechanisms work on the basis of actual occurrences of dema nd rather than the demand forecasts (Gershwin et al. 1993). Pull control systems avoid excessive inventory levels between the production stages and reduce lead times. In pull system the production is init iated when a demand arrives at the last stage. The demand from a downstream work station is signaled to the upstream workstation based on actual downstream consumption of the product. Thus, in pull control systems, work release is triggered by the actual demand and the upstream workstation produces just in time to meet the demand needed by the downstream work station, which ultimate ly is controlled by the demand for the final product. Just-in-time manufacturing is a pulltype system that ideally depends on customer demand to trigger production. According to Monden (1983), the idea of producing the necessary units in the nece ssary quantities at the necessary time is described by the short te rm just-in-time. It is defined as a repetitive production system producing the necessary units in the necessary quantities at the necessary point of time.

PAGE 14

3 1.2 Kanban Cards Kanban, a Japanese word meaning card, is used for transferring information from downstream work center to the upstream work center to control the movement of material in a manuf acturing system. Kanbans serve as the production order for the pull control syst em. The number of kanbans in the manufacturing system determines throughput rates and the amount of work in process inventory in the system. Kanban systems can be either dual-card or single-card. Determination of the number of kanban ca rds at every stage is crucial for the performance of the syst em. It determines the pr oduction quantities at each stage, work in process inventory and throughput of the syst em. Number of kanban cards in the manufacturing system depends on the coefficient of variation in processing times, machi ne utilization, and the autocorrelation of processing times. Feryal et. al. (2003) proposed an analytical model to determine the kanban sizes and number of kanbans simultaneously in a multi-item, multi-stage kanban system. 1.2.1 Single Card System Single Card Kanban System is generally used to convey the movement of material in the system. Single card systems work very effectively in situations where work stations are close to each other and there is an excess inventory in the system available for pickup (Schni ederjans 1993). W henever, a customer demand occurs, it tries to fu lfill it from the finished goods inventory. Kanban card

PAGE 15

is detached from the produced item and it sends a signal to the upstream work center to produce the respective item. Each work center has one buffer stock available. Single card pull system shown in figure 1 is simple to implement and works best for serial production systems. Figure 1. Single Card Pull Control System 1.2.2 Dual Card System Production kanban and withdrawal kanban are the two main types of kanban cards used by Dual-Card Kanban System. Withdrawl kanban card signals the need to deliver more parts, thus defines the quantity that the succeeding stage should withdraw from the preceding stage. Production kanban cards signal the need to produce more parts, thus defines the quantity of the specific part that the producing stage should manufacture in order to replace those which have been removed (Groenvelt 1993). Erik & Bohez (2004) presented a generic black token timed Petri net model to determining the optimal work in process, the number of kanban cards for a dual card KCS. Dual card kanbans are used in manufacturing systems where processes are physically separated, e.g. in different plants. Dual card system has an input and output 4

PAGE 16

buffer at every stage. The original Toyota kanban system is an example of a dual card kanban system. A typical Dual Card system is shown in figure 2. Figure 2. Dual Card Pull Control System 1.3 Production Control Systems In recent years, considerable interest has been shown in research to analyze the performance of a just in time manufacturing system which uses the pull control philosophy and various analytical models have been proposed (Uzsoy and Martin Vega 1990). Kanban Control System is the commonly used just in time manufacturing system also referred as Toyota production system was introduced at Toyota (Monden 1983). Manufacturing systems operate according to a set of production control policies and these policies determine when to start and stop production and when to switch from one product to another for each stage in a multi product manufacturing system. Base Stock and Kanban are two of the better known pull control mechanisms. 5

PAGE 17

1.3.1 Base Stock Control System Figure 3. Base Stock Control System Base-stock Control System depends on a single parameter per stage which corresponds to safety stocks. The safety stock determines the maximum number of finished parts. Base stock control system is considered reactive as the demand signal is transmitted to all production stages when an external demand arrives as shown in figure 3. The finished products are stored in the finished goods inventory until they are used to satisfy the customer demand. If there are no finished products in the finished goods inventory when the demand arrives, then the demand is backordered. 1.3.2 Kanban Control System Kanban mechanism depends on a single parameter per each stage which corresponds to the production authorization cards. Kanban Control System (KCS) coordinates production by using a finite number of production authorization cards. These production authorization cards known as kanban cards transmit demand requests from a downstream work center to the upstream 6

PAGE 18

work center and serve as production authorizations for the corresponding stage. The demand in KCS is sequentially transmitted to the preceding stages. Figure 4. Kanban Control System A typical Kanban Control System (KCS) having three stages in series is shown in figure 4. Each stage of a multi stage kanban control system is associated with fixed number of kanbans (k). Since there is fixed number of kanban cards associated with each stage (k), this number is an upper bound on the number of parts in each stage, either in the manufacturing process or in the output buffer. The behavior of the KCS depends on the initial condition that is before any demand has arrived at the system, the output buffer of each stage contains k number of finished parts, each part having a kanban card attached to it and all other queues are empty. Infinite supply of raw materials assumed at the first stage. When a demand occurs at the last stage it is satisfied by the part in finished goods inventory and if there is no part in the finished goods inventory the demand is backordered. When a demand is satisfied by a part in finished goods inventory, a signal is transmitted to the preceding stage. This signal is the 7

PAGE 19

8 authorization for the preceding stage to pr oduce a part and acts as a production card. Production occurs at a workstation onl y if raw material is available and the material has a production authorizing card. The main advantage of KCS is that it is simple to understand and implement, but unfortunately it also pl aces significant restrictions on its behaviour. Also the control policies, base stock and kanban control system do not always achieve a good trade off bet ween inventory costs and customer service levels. To achieve a better trade off several variations of the basic kanban production system have been proposed and much work has been carried out in the analysis and performance eval uation of such al ternative systems (Baynat et al. 2002). Another important feature of the KCS is that the demand and kanban cards are simultaneously tr ansferred from a given stage to the upstream stage at the same time a part is consumed by the downstream stage. 1.3.3 Generalized Kanban Control System Generalized Kanban Control System was simultaneously introduced by Buzacott (1989) and Zipkin (1989). It comb ines the respective advantages of kanban system which achieves better co-o rdination and control of work in process, and base stock that reacts rapi dly to demand. The Generalized Kanban Control System depends on two parameters per stage, the number of kanbans and the base stock level. The first param eter controls the work in process inventory at that stage and the base sto ck determines the number of parts that

PAGE 20

must be produced to be stored at the output of the stage to maintain a level of inventory. Frein and Dallery (1995) investigated the influence of these design parameters on the efficiency of generalized kanban control system. Figure 5. Generalized Kanban Control System Figure 5 describes the behavior of a single product Generalized Kanban Control System (GKCS) where each stage of the manufacturing system consists of a manufacturing process and an output buffer. Kanban cards are used as production authorization cards, to transfer parts to the downstream stages. The maximum number of parts in a buffer at each stage is determined by the target inventory level (s). When an external demand arrives, it is transmitted from downstream stages to upstream stages giving rise to a demand for production of a new part at every stage, which in turn will give rise to the release of a part at every manufacturing process. Lack of kanban cards at certain stages delays the transfer of demand while the release of parts at some stages may be delayed because of the lack of finished parts at the previous stages. Generalized Kanban Control System differs from the kanban control system in the way the demand and kanban cards are transferred independently of each other in the manufacturing system, whereas in the KCS they are done 9

PAGE 21

10 simultaneously. In Kanban Contro l System demand requests cannot be transferred upstream if there are no fini shed parts at certain stage. In GKCS, even if there are no finished parts availabl e at that stage the additional number of kanban cards transfer the demand to the upstream from a stage. 1.3.4 Extended Kanban Control System Extended Kanban Contro l Systems combines the advantages of base stock and Kanban control mechanisms and is also defined by two parameters per stage and include both the kanban and base stock systems as special cases (Baynat et al. 2002). The total work in process is determined by the number of kanban cards in the manufacturing syst em. The demand is immediately transferred to all the stages unlike in t he kanban control system and is the main advantage of extended kanban control system. Simplicit y and limitation of the work in process in each stage are import ant features of the extended kanban control system (Baynat et al. 2002).

PAGE 22

Figure 6. Extended Kanban Control System Figure 6 describes Extended Kanban Control System having two stages in series. Each stage consists of a manufacturing process with an output buffer. The Extended kanban control system is the combination of both base stock and kanban control system. The EKCS, like the GKCS, depends on two parameters per stage, number of kanban cards and the target inventory level. The extended kanban control system has the condition that the number of kanban cards at each stage must be greater than the target inventory level of that stage (Karaesmen and Dallery 2000). When a demand arrives to the system, it is transmitted to all the stages immediately, so that all the stages have demand signal as soon as it arrives. EKCS is equivalent to KCS if the number of kanban cards is equal to the target inventory level for each stage (Baynat et al. 2002). 1.4 Overview There has been much work done on the individual production control systems but relatively few comparison studies have been done. Comparisons have been made to analyze the performance and behaviour between different 11

PAGE 23

12 control mechanisms. Berkley (1992) compared push and pull control systems with amplification orde ring quantity and inventory level variance as the performance measures. Variance amplif ication was used as a performance measure to compare the kanban contro l system with push systems by Kimura and Tereda (1981). Kanban control mechani sm is not flawless despite having several advantages over other push control systems. Various alternative systems to the kanban control system have been proposed and considerable research has been done comparing these alternat ive control systems with the kanban control system. Duri et al. (2000) compared kanb an, base stock and a generalized kanban control system. Accordi ng to Karaesmen and Dallery (2000) generalized control system does not necessarily perform better than the base stock or kanban control system. In some cases the base stock or kanban control system may perform poorly. However, the in ventory carrying cost of a GKCS is at least equal to, or less than the base stock or kanban control system. The question arises which pull control system performs best. In this chapter introduction to various pull control systems has been presented. In the following chapter pr oduction control policies for multiple product type manufacturing systems have been discussed. Chapter three outlines the proposed research and vari ous performance measures considered.

PAGE 24

Chapter Two Multi-product Manufacturing Systems In recent years, much of the research has been focused on modeling simple control systems with single product type and various methods have been proposed to evaluate their performance. Most of the literature on multiple product kanban systems has been focused on planning and scheduling issues. Figure 7 shows a multiple product manufacturing system with two stages in series and two different product types. Each product type has a fixed number of kanban cards at every stage which are used as production authorization cards. Figure 7. Multi-product Manufacturing System 2.1 Dedicated and Shared Kanban Cards Dedicated and shared kanbans are two alternative ways of specifying Kanban-like Control Systems for multiple-product systems. In dedicated kanban systems, for each stage, there is a fixed number of kanbans associated with 13

PAGE 25

each type of product as shown in figure 8. Each product will have its own dedicated kanbans and that the number of kanbans for one product is determined independently of the choice of the number of kanbans for other products. Figure 8. Dedicated Kanban Cards In shared kanban systems, kanbans are shared between different part types. The difference between shared and dedicated kanbans is that in dedicated systems the maximum total number of kanbans in the stage is limited, while in the shared system the maximum total number of kanbans in the stage is the sum of the number of kanbans for each product type. A shared kanban in figure 9 can be used to trigger the production of any part type in a given stage and when it becomes free, the pull control system determines the type of the part with which it is going to be associated. 14

PAGE 26

Figure 9. Shared Kanban Cards 2.2 Dedicated Production Control Systems The main objective of the research is to study different multiproduct manufacturing control systems. As described earlier, dedicated and shared kanbans are the two methods used to distribute kanban cards for multiproduct and multistage manufacturing systems. Multiproduct kanban control systems have fixed number of kanban cards dedicated to each product type at every stage and hence the name dedicated kanbans. 2.2.1 Dedicated Kanban Control System Figure 10. Dedicated Kanban Control System 15

PAGE 27

Figure 10 gives the queuing model for the multiproduct dedicated kanban control system. The manufacturing control system shown above has two stages in series producing two parts. At each stage there are fixed numbers of kanban cards associated with each part type. Each stage has a manufacturing process and an output buffer which stores parts to be processed with kanban cards attached to it. 2.2.2 Dedicated Generalized Kanban Control System Figure 11. Dedicated Generalized Kanban Control System Figure 11 shows the queuing model for dedicated generalized kanban control system. Again, the manufacturing system has two stages in series producing two part types with fixed number of kanban cards dedicated to each part type at every stage. 16

PAGE 28

2.2.3 Dedicated Extended Kanban Control System Figure 12. Dedicated Extended Kanban Control System The above figure shows the queuing network model for dedicated extended kanban control system with two stages in series producing two product types and fixed number of kanban cards dedicated to each part type at every stage. As shown in the figure, demand and authorization for production for each part type is independently transferred to the upstream stages unlike in kanban control system. Since the demand signal is concurrently transferred to all the stages in series, it reduces blocking of stations and bottlenecks if any from the manufacturing system. 2.3 Shared Production Control Systems Shared production control systems have fixed number of global kanban cards known as shared kanbans at every stage. According to Baynat et al (2002) shared kanbans are non dedicated production authorization cards that can be used to trigger production of any part type in the stage. The release of 17

PAGE 29

shared kanban cards for a part type depends on the control mechanism. Shared kanbans control the amount of work in process inventory of the manufacturing system and considerable reduces the in-process inventory cost. 2.3.1 Shared Kanban Control System Figure 13. Shared Kanban Control System The queuing network model for multiproduct shared kanban control system with two stages in series is shown in figure 13. Each stage of the manufacturing system has global number of shared kanban cards. Total number of kanban cards determines the work in process inventory of the system. 18

PAGE 30

2.3.2 Shared Generalized Kanban Control System Figure 14. Shared Generalized Kanban Control System Above figure 14 illustrates the two stage shared generalized kanban control system producing two part types. Similar to single stage generalized kanban control system, each stage of the multi product manufacturing system is associated with a global number of kanban cards and base stock level. Production authorization cards and the demand signal are independently transferred to the upstream stage. 2.3.3 Shared Extended Kanban Control System Figure 15. Shared Extended Kanban Control System 19

PAGE 31

20 Multi-product Shared kanban control syst em with two stages in series manufacturing two products is shown in figure 15. Every stage in multi product manufacturing system is also associated with global number of kanban cards and base stock level. Unlike kanban control syst em, the transfer of kanban cards to the upstream stages does not depend on the demand arrival which is concurrently transmitted to all the stages in the manufacturing system. 2.4 Summary Considerable research has been done to study the individual manufacturing systems for multi stage and single product. However, little research has been done to compare the di fferent pull control policies for multi stage manufacturing systems. Mo st of the research do ne in multi product system assumes that each kanban card is dedi cated to each part type. Baynat et al. (2001) introduced Shared Kanbans, an altern ate way to specify kanban cards for a multiple product system. In this study we analyze the performance of various multi product control policies for a two stage manufacturi ng system producing three products. The manufacturing system considered in the research will use a single-card kanban system, where the tr ansportation of materials between the different work-centers is controlled by production kanbans. The next chapter covers the problem definiti on, and the tools used for des igning and verification of proposed research systems.

PAGE 32

21 Chapter Three Problem Definition In the previous chapter different production control systems for multiproduct manufacturing system were analyzed. Recent research has been focused on pull control systems for mult i-product manufacturing systems, but comparative study of thes e systems does not exist at this time. This current chapter will cover the probl em statement and tools used to solve the problem. 3.1 Problem Description and Objectives The aim of this research is to compare the perfo rmance of three production control mechanisms for multi-product manufacturing systems. Production control mechanisms considered are Kanban Control System (KCS), Generalized Kanban Control System (GKCS) and Extended Kanban Control system (EKCS). Shared Kanbans and Dedi cated Kanbans are the two cases considered. The following performance measures are considered in the system Number of backordered demands Average waiting time of backordered demands Average work in process.

PAGE 33

22 Waiting time of each backordered item, that is, period from the time when the demand is backordered to the time when the demand is satisfied by the production system determines the aver age waiting time of backordered demands. If the inventoried product does not run out, the arrived demand will be satisfied just on time and hence there will be no backorders and the waiting time will be zero. It is used as a performance m easure to introduce a delay in filling orders. Giving a delay in filling orders is equivalent to authorizing some demands to wait. However at the end of the delay, the demand must be satisfied if possible. Average waiting time of eac h backordered item is the measure to evaluate service level of the production system. 3.2 Modeling Assumptions A Single Card manufacturing system having stages in series and producing three types of products is cons idered in the research. For each control system studied simulation models are dev eloped for three and six stages having dedicated or shared kanbans. Each type of part must be processed by each stage. The following are the most import ant assumptions made for this system. Infinite supply of raw material is available at the first stage. Setup time for each part type at ev ery stage and transfer time of parts from one stage to another is assumed to be negligible. Each stage consists of a manuf acturing process and an output buffer.

PAGE 34

23 If there is a demand for a part type and is not satisfied from the Finished Goods Inventory then the demand is backordered. Machines are assumed to be failure-prone. 3.3 System Parameters Following are the definitions of di fferent parameters used while model formulation. 3.3.1 Demand Arrival Part type 1 was assumed to have high demand with dem and rate 20 parts per hour. Demand rate for part type 2 and part type 3 is 15 and 10 parts per hour representing medium and low demand levels respectively. The inter-arrival time of product demand is assumed to follow an ex ponential distribution. In case that the distribution of product demand is defined by that of the inter-arrival time, the exponential is the most commonly used condi tion. In the expone ntial distribution, the standard deviation is equal to the mean, that is, the coefficient of variation is equal to one. Typically, Poisson process is used for modeling demand which implies that the time between arrival is modeled using exponential distribution. The inter-arrival time between demands was therefore, assumed to be exponentially distributed with a mean 3, 4 and 6 minutes for part type 1, 2 and 3 respectively.

PAGE 35

24 3.3.2 Processing Times The processing time of parts determine the workload on the system. The processing time in this re search follows the gamma distribution. Krishnamurthy et al. (1992) proposed the gamma distribution since it specifically meets the requirements for describing pr ocessing times in the JIT environment and is computationally efficient. Unlike the research reported in literat ure, the processing times for three part types have been varied for the different processing stages as shown in table 1. By assuming different processing time s we can study different levels of machine utilizations in the shop (high 90%, low 80% and medium 85%). To study the effect of variability in proc essing time for the control systems, the coefficient of variation was changed fr om 0.2 to 0.4 and 0. 6. The number of kanban cards and target inventory level for a given number of stages were kept same. Table 1. Processing Times Part Type 1 Part Type 2 Part Type 3 Stage 1 0.75 1.5 1.25 Stage 2 1.5 1.0 0.75 Stage 3 0.9 1.2 1.10 In this research we study the e ffect of number of stages on the performance of manufacturing control system and in order to have a consistent production environment with three stage manuf acturing control system, the mean

PAGE 36

25 processing times for stages 4, 5 and 6 were assumed to be same as for stages 1, 2 and 3 respectively. 3.3.3 Time Between Failures We consider a failure-prone manufacturi ng system to study the effect of machine breakdown and its impact on system performance. Time between failures as well as the repair times is exponentially distributed with a mean value as shown in table 2. These values were determined based on the assumption that the machine utiliz ation at each stage should not exceed 90%. The time between failures and repair times were chosen to represent varying frequency of failures. As shown in table 2, stage 1 is prone to infrequent breakdowns with longer repair time, while stage 3 has more frequent breakdowns with shorter repair times. Table 2. Time Between Failures and Repair Times Stage 1 Stage 2 Stage 3 Time between failures (hours) 6 5 4 Repair Times (min) 9 7.5 6 For the system wit h six stages, then time betw een failures and repair times for the last three stages were considered to be the same as the first three stages.

PAGE 37

26 3.3.4 Service Levels The objective of the manufacturing control systems considered in this research is to achieve a certain pre-spec ified service level. In this study, the service level is defined as the fraction of demand satisfied from inventory. A base level of 95% was set for all three manuf acturing systems at two different stages using dedicated and shared kanbans. In order to investigate whether the difference in the service level has a signi ficant influence on WIP inventories and the waiting time of backordered demand for the different production control systems considered, experiments are carri ed out for 98% service level, where the maximum number of backorder does not exceed 2% of the demand. 3.3.5 Number of Kanban Cards An optimal level of kanban cards at each stage for three stage KCS is determined by systematically varying the number of cards until 95% service level is reached i.e. the number of backorders for each part type is less than 5% of the demand for the corresponding part type. The target inventory level for GKCS and EKCS is equal to the number of k anban cards in the KCS for optimal configuration. The number of kanban ca rds for GKCS and EKCS are varied to satisfy the 95% service level condit ion. The number of kanban cards and the target inventory level for KCS, GKCS and EKCS was kept same for the six stage system to be able to compare it with the three stage system.

PAGE 38

27 The following table shows the optimal configuration for the production control systems considered in this research, where k is the number of kanban cards and s is the target inventory level. Table 3. Number of Kanban Cards and Target Inventory Level Control Systems Service Level KCS GKCS EKCS 95 % k (7, 6, 4) k (11, 10, 8), s (7, 6, 4) k (11, 10, 8), s (7, 6, 4) 98 % k (9, 8, 6) k (13, 12, 10), s (9, 8, 6) k (13, 12, 10), s (9, 8, 6) 3.4 Simulation Model Simulation can be defined as a proce ss of designing a model of a real system and conducting experiments wit h this model for the purpose of understanding the behavior of the system and/or evaluating various strategies for the operation of the system. Simulation is a very powerful tool that can be used to analyze the performance of different m anufacturing systems. In recent years, simulation is being widely used as a tool to discover the benefits and risks associated with implementing Just in Time (JIT) manufacturing techniques. Assumptions like machines will never break down and fixed daily production, reflects the ideal characteristics of JI T manufacturing system, but contradicts real production environments. The manufacturing process at each stage consists of part type currently being processed or waiting to be process ed. Such part types are referred to as Work in Process for the given stage. The output buffer of the la st stage referred

PAGE 39

28 to as Finished Good Inventory (FGI) consists of part types that have finished processing. When a demand arrives for a giv en part type, it is satisfied from the FGI. The kanban card is detached and sent to the preceding stage. If there is no part in the FGI the demand is assumed to be backordered. 3.4.1 Length of Simula tion and Warm-up Period In order to analyze the results of a system based on a simulation run, it is important to decide several items. Wa rm-up period, run length and number of replications are the thr ee primary ones that need to be carefully considered. Since all the simulation runs begin with all the stages idle and buffer at each stage equal to the preset number of kanban cards, two approaches are used to minimize the bias due to these in itial conditions. First, a warm-up period will be used to clear the statistics collected during initial time period, second, system will be run long enough to dilute the im pact of initial conditions. A plot of WIP for a three stage dedicated KCS with a run length of 5 days for 5 replications (superimposed) is shown in figure 16. From the plot one can see that the WIP is building up initially during the tr ansient state. This build up is over at 1 day period which will be used as warm-up period. It was further assumed that a 15 day period for collecting statistics will be suffici ent to get a steady-state behavior of the system performance and this was used as run length in all replication.

PAGE 40

Figure 16. Within Run WIP Plots for KCS 3.4.2 Number of Replications Initially 25 independent replications were made for KCS to satisfy 98% service level. The total numbers of backorders for KCS were 325 62. The half width for the given number of replications was almost 20% of the average number of backorders. It was decided that 90% confidence interval with a half width equal to 10% of the mean value will be desired. The number of replications was calculated using the following approximation. 29

PAGE 41

20hhnno Where n o = number of initial replications, h o = half width and n = desired number of replications 935.3262252n One hundred independent replications were performed to determine the variability in the performance measures 3.5 Design of Experiment Design of experiment is a systematic approach to investigation of a system or process. A series of structured tests are designed in which planned changes are made to the input variables of a process or system. The effects of these changes on a pre-defined output are then assessed. The experiments investigated in this research are classified into the following two parts. Part I investigates the production control systems for 95% service level, and Part II investigates the production control systems for 98% service level. As shown in table 4 each experiment consists of three input factors namely, the type of control system, the variability of processing time and the number of stages. The levels of experimental factors investigated in this research are summarized below. 30

PAGE 42

31 Table 4. The Factors and Conditions of Experiments Production Time (min) Control Systems Mean Variance Number of Stages KCS GKCS EKCS Shared GKCS Shared EKCS 0.75,1.50, 0.90 1.5, 1.0, 1.2 1.25, 0.75, 1.10 0.2 0.4 0.6 3 6 3.6 Summary In this chapter we stated the res earch problem and modeling approach to solve it. We also defined the input fact ors and their level and created a design of experimental model. In the next chapter we will discuss the results from the simulation model and analysis of those re sults to find significant factors and factor interactions. Analysis of variance (ANOVA) will be used to see the significant differences between differ ent factors and their interaction and appropriate conclusion will be drawn. ANOVA experiments are conducted using SAS software and results are given in the next chapter.

PAGE 43

32 Chapter Four Results and Analysis In the last chapter we presented the problem statement, assumptions made to solve the problem, simulation model, different system parameters, approach to solve the problem and the des ign of experiments wit h its factors and their levels. In this chapter the resu lts will be analyzed using an Analysis of Variance model. 4.1 Simulation Results In order to analyze the factors dete rmined in the design of experiment, we chose three different response variables: number of backorders, backorder time in system and work-in-process. The results obtained for different factor combinations are tabulated in table 5 and 6. Table 5(a) shows number of backorders for a three stage system at 95% service level. Table 5(b) shows number of backorders for a six stage syst em at 95% service level. We can observe from these tables that as the coefficient of variation increases the number of backorders also incr ease. The effect is very significant at coefficient of variation of 0.6. Similarly, we c an see as the number of stages is increased from 3 to 6, the number of backorders increased by about 10%.

PAGE 44

33 Table 5. Number of Backorders for 95 % Service Level (a) Three Stage Control Systems Coefficient of Variation Control System 0.2 0.4 0.6 KCS 788 45 917 53 1268 81 GKCS 639 31 779 40 1090 56 EKCS 716 55 853 58 998 69 Shared GKCS 504 26 615 33 912 43 Shared EKCS 578 31 694 38 854 27 (b) Six Stage Control Systems Coefficient of Variation Control System 0.2 0.4 0.6 KCS 860 58 1095 67 1436 70 GKCS 725 47 929 65 1270 63 EKCS 767 38 943 64 1141 79 Shared GKCS 552 31 658 32 995 24 Shared EKCS 620 33 754 44 879 48

PAGE 45

34 4.2 Analysis of Results The analysis of variance commonly referred to as ANOVA is used for studying the effects of different factors separately (thei r main effects) and their interaction effects. Analysis of vari ance was conducted using SAS software at 95% confidence interval. The table lists s ource of variation, degrees of freedom, sum of squares, mean squares, F-values and the p-values. P-value determines which factor is significant. If the p-value is less than the level of significance then the factor is said to be significant. 4.2.1 Analysis of Results for Backorders at 95% Service Level Figure 17 refers to the ANOVA for number of backorders at 95% service level and consists of three factors namely number of stages (2 levels), type of control system (5 levels) and coefficient of va riation of processing time (3 levels). Class Levels Values stage 2 3 6 system 5 KCS DGKCS DEKCS SGKCS SEKCS cov 3 0.2 0.4 0.6 Source DF Type I SS Mean Square F Value Pr > F stage 1 7139050.57 7139050.57 111.83 <.0001 system 4 49252392.17 12313098.04 192.88 <.0001 cov 2 87456963.85 43728481.92 684.99 <.0001 stage*system 4 1414582.39 353645.60 5.54 0.0002 stage*cov 2 347429.83 173714.92 2.72 0.0660 system*cov 8 6465916.70 808239.59 12.66 <.0001 stage*system*cov 8 288046.34 36005.79 0.56 0.8081 Figure 17. Analysis of Variance fo r Backorders for 95 % Service Level

PAGE 46

35 From figure 17, the following can be observed All the factors namely number of stages, types of control system and the variability in the processing time prove to be significant factors. All two-factor interactions are signifi cant except for the interaction between number of stages and variability in the processing time, which is insignificant. Also three factor interactions of num ber of stages, types of control system and processing time variability prove to be insignificant. 4.2.2 Analysis of Factor Inte ractions at 95% Service Level Figure 18 shows the interaction bet ween type of control system and processing time variability for 95% service level where x-axis corresponds to three levels of coefficient of variat ion and y-axis corresponds to the average number of backorders. The interaction between types of system and number of stages is shown in figure 19 with average number of backorders on x-axis and number of stages on y-axis. The main objective of this research was to study the performance comparison of different types of control systems. Therefore, only interactions with the cont rol systems are analyzed.

PAGE 47

Figure 18. Interaction Plot for Backorders (Type of Control System vs. Processing Time Variability) at 95% Service Level 36

PAGE 48

Figure 19. Interaction Plot for Backorders (Type of Control System vs. Number of Stages) at 95% Service Level 37

PAGE 49

38 For the control system and processing time variability interaction, at lower and medium processing time variability, Shared GKCS has lower number of backorders as compared to the other cont rol systems, while at higher variability Shared EKCS performs better then other types of control systems. Production control systems using shared kanbans have more number of free kanbans at each stage as compared to the same control system using dedicated kanban cards. Thus the extra free kanban card s available at each stage help the production control system perform bette r than the dedicat ed control systems when the variability in the processing time in creases. It is very clear from figure 18 that the Shared EKCS deals with larger processing time variability in more effective manner compared to the other control systems, though the values for Shared GKCS are very close. Overall the Shared GKCS has lower number of backorders. The reason for this is that in GKCS a kanban card is released as soon as the finished part enters the output buffer of the stage and hence the demand is transferred faster to the previous stages as compared to the other control systems. In case of KCS and EKCS the kanban card is released when part is taken out from the buffer and hence the variability in the processing time introduces a delay in transferring kanban author ization to the previous stages. For system and number of stages interaction as shown in figure 19 the number of backorders increases when t he number of stages are varied. The reason for this is that the additional number of stages introduces delay in transferring the kanban authorizations and hence the increased number of backorders. Shared GKCS and Shared EKCS have lower number of backorders

PAGE 50

39 as compared to the other control syst em, though shared GKCS marginally performs better then the shared EKCS. 4.2.3 Analysis of Results for Backorders at 98% Service Level In order to study the effect of servic e level on the performance of different production control systems, service leve l was increased to 98%. The number of kanban cards and target inventory level at each stage of the production control system increases when the service level is changed from 95% to 98%. Tables 6(a) and 6(b) show the number of backorder s for 98% service levels. Again, as in case of 95% service level, increase in backorders with the increase in coefficient of variation and the number of stages for corresponding control systems can be observed.

PAGE 51

40 Table 6. Number of Backorders for 98 % Service Level (a) Three Stage Control Systems Coefficient of Variation Control System 0.2 0.4 0.6 KCS 307 27 368 23 575 44 GKCS 199 15 279 25 388 32 EKCS 238 28 309 32 382 38 Shared GKCS 165 13 223 16 317 21 Shared EKCS 191 11 211 14 234 24 (b) Six Stage Control Systems Coefficient of Variation Control System 0.2 0.4 0.6 KCS 372 23 437 34 569 56 GKCS 235 25 339 27 464 36 EKCS 258 21 322 26 402 31 Shared GKCS 207 11 281 14 376 30 Shared EKCS 247 19 262 22 314 25

PAGE 52

41 Figure 20 refers to the ANOVA for num ber of backorders at 98% service level. From the figure, following can be observed: All main factors namely number of stages, types of control system and the variability in the processing time prove to be significant factors. All two-factor interactions are signifi cant except for the interaction between number of stages and variability in the processing time, which are insignificant. Also three factor interacti ons prove to be insignificant. These observations are consistent with 95% service level. Class Levels Values stage 2 3 6 system 5 KCS DGKCS DEKCS SGKCS SEKCS cov 3 0.2 0.4 0.6 Source DF Type I SS Mean Square F Value Pr > F stage 1 1639125.13 1639125.13 84.92 <.0001 system 4 13916159.80 3479039.95 180.24 <.0001 cov 2 12993754.44 6496877.22 336.58 <.0001 stage*system 4 189834.98 47458.75 2.46 0.0435 stage* cov 2 3986.95 1993.48 0.10 0.9019 system* cov 8 2084098.21 260512.28 13.50 <.0001 stage*system* cov 8 245386.10 30673.26 1.59 0.1227 Figure 20. Analysis of Variance fo r Backorders for 98 % Service Level

PAGE 53

4.2.4 Analysis of Factor Interactions for 98% Service Level Figure 21 shows the interaction between types of control system and processing time variability for 98% service level. The interaction between types of control system and number of stages is shown in figure 22. Figure 21. Interaction Plot for Backorders (Type of Control System vs. Processing Time Variability) at 98% Service Level 42

PAGE 54

Figure 22. Interaction Plot for Backorders (Type of Control System vs. Number of Stages) at 98% Service Level In order to investigate the effect of service level on performance of the different flow control strategies experiments were carried out for 98% service level. Increasing the service level increases the number of kanban cards and target inventory level of the flow control strategies and hence the work in process. As in the case of 95% service level, it can be seen from figure 21 that the difference for number of backorders between KCS and Shared EKCS increases with the increase in the variability in the processing time. The reason 43

PAGE 55

44 for this is that in Shared EKCS when a demand arrives at the system, it is immediately broadcasted to every stage in the system and also the number of kanban cards are shared. This implies that each stage in the system knows immediately the need for production of a new part in order to replenish the finished-product buffer. While in case of KCS the demand is sequentially transmitted to the preceding stages c ausing the delay in authorization for production. At lower processing time variability, however, Shared GKCS performs better, the reason being, that the kanban cards are tr ansmitted to the preceding stage before it ent ers the output buffer, but at medium and high variability in the processing time, it re sults in more number of backorders as compared to Shared EKCS, since the demand signal and kanban cards are partially coupled. For type of system and number of stages interaction as shown in figure 23, Shared EKCS has less number of ba ckorders as compared to the other control system. The difference between t he number of backorders for KCS and shared EKCS is almost 30%. Table 6(b) shows that for six stage control systems with low processing time variability Shared GKCS and Dedicated GKCS perform better than Shared EKCS, but as the processing time variability increases Shared GKCS results in less number of bac korders. The results for 98% service level are consistent with the 95% service level exc ept that the Shared EKCS performs better than the Shared GKCS for medium processing time variability.

PAGE 56

45 4.2.5 Analysis of Results for Average Work in Process The average work in process considered in this research is the sum of actual work in process and finished goods inventory. The average work in at 95% and 98% service level is shown in table 7 and table 8 respectively. Shared EKCS has 10% lower average WIP compared to KC S. It is due to the fact that in Shared EKCS each stage of t he system consists of a gl obal number of cards that can be used to trigger the production of a par t type in that stage. Also, in Shared EKCS when a demand arrives, it is immedi ately transferred to every stage of the system and hence starving of stages can be av oided. The individual factors such as the type of system, number of stages and the proces sing time variability have a significant effect on average work in process in a manufacturing system. Every system has a different level of work in process, with KCS having the maximum and Shared EKCS having the minimum. Table 8 shows the corresponding result s for three and six stage control systems at 98% service level. The Shared EKCS makes a better serviceinventory compromise than kanban at both 95% and 98% service levels. The WIP for other control policies fall bet ween the KCS and Shared EKCS. One of the trade-off associated with increasing the se rvice level from 95 to 98 percent is the increase in WIP of the control system which can contribute to overhead cost for the manufacturing system. As the servic e level is increased from 95% to 98% the WIP increases by 20 to 25%.

PAGE 57

46 Table 7. Average Work in Process at 95 % Service Level (a) Three Stage Control Systems Coefficient of Variation Control System 0.2 0.4 0.6 KCS 39.2 43.5 47.3 GKCS 37.4 41.9 46.2 EKCS 38.9 42.6 47.1 Shared GKCS 36.4 39.8 42.3 Shared EKCS 33.6 35.4 39.6 (b) Six Stage Control Systems Coefficient of Variation Control System 0.2 0.4 0.6 KCS 92.4 96.8 99.6 GKCS 90.3 94.6 98.7 EKCS 91.5 95.2 97.6 Shared GKCS 86.4 90.1 93.2 Shared EKCS 82.6 85.9 88.6

PAGE 58

47 Table 8. Average Work in Process at 98 % Service Level (a) Three Stage Control Systems Coefficient of Variation Control System 0.2 0.4 0.6 KCS 56.4 59.6 63.1 GKCS 53.2 55.9 59.6 EKCS 54.9 57.1 61.2 Shared GKCS 48.7 52.1 53.6 Shared EKCS 46.5 49.3 50.2 (b) Six Stage Control Systems Coefficient of Variation Control System 0.2 0.4 0.6 KCS 122.4 127.6 132.3 GKCS 118.7 121.9 130.5 EKCS 120.9 125.6 129.7 Shared GKCS 114.2 119.7 124.6 Shared EKCS 109.6 116.4 121.3

PAGE 59

48 4.2.6 Analysis of Results for Average Waiting Time of Backordered Demand Number of backorders does indicate the performance of a manufacturing system. However, the time required to sa tisfy the items backordered may be an important criteri on when comparing two systems. Therefore, an additional performance measure was defined to captur e the expected delay in satisfying the backorders. The period from the time when the demand is backordered to the time when the demand is satisfied by the production system determines the average waiting time of backordered demands. The average waiting time of backordered demands associated with th ree and six stage c ontrol systems at 95% service level is shown in table 9. The individual factors such as the type of control system and the processing time variability have a significant effect on average waiting time of backordered dem and. We observe that Shared GKCS performs better than the other control syst ems. As expected, the average waiting time for backordered demand increase with the increase in processing time variability and the number of manufacturing stages. Table 10 shows the corresponding result s for three and six stage control systems at 98% service level. If the objecti ve is to minimize the average waiting time of backordered demands, Shared GKCS is a better choice compared to the KCS. It is very clear that Shared GKCS has a significantly lower average waiting time of backordered demands than KCS. The reasons for Shared GKCS having lower average waiting time for backorders compared to the other control policies is that the demand moves upstream separately from the kanban cards

PAGE 60

49 Table 9. Average Waiting Time of Backordered Demand at 95 % Service Level (a) Three Stage Control Systems Coefficient of Variation Control System 0.2 0.4 0.6 KCS 12.9 14.6 17.9 GKCS 10.5 12.9 15.4 EKCS 11.6 13.1 14.9 Shared GKCS 8.3 9.7 11.1 Shared EKCS 9.4 10.9 12.6 (b) Six Stage Control Systems Coefficient of Variation Control System 0.2 0.4 0.6 KCS 14.8 16.5 19.1 GKCS 11.3 13.4 17.6 EKCS 12.6 14.8 18.3 Shared GKCS 9.6 11.2 13.1 Shared EKCS 10.8 12.9 15.9

PAGE 61

50 Table 10. Average Waiting Time of Backo rdered Demand at 98 % Service Level (a) Three Stage Control Systems Coefficient of Variation Control System 0.2 0.4 0.6 KCS 15.3 17.2 21.3 GKCS 14.6 16.7 19.8 EKCS 15.1 16.2 18.9 Shared GKCS 10.1 12.9 14.6 Shared EKCS 11.7 13.1 17.5 (b) Six Stage Control Systems Coefficient of Variation Control System 0.2 0.4 0.6 KCS 17.1 19.4 24.8 GKCS 16.6 20.4 23.9 EKCS 17.9 19.8 22.2 Shared GKCS 12.4 14.5 16.8 Shared EKCS 13.9 16.7 21.8

PAGE 62

51 The average waiting time of backo rdered demands for Shared GKCS is almost 30% less then KCS for both service levels. Shared GKCS and Shared EKCS have similar results, though Shared GKCS is marginally better and has 10% less average waiting time for backordered demand when compared to Shared EKCS at lower processing time variability. However, as the processing time variability increases this difference in creases to 20 to 30% (at coefficient of variation of 0.6). As a final observation, the average waiting time of backordered demand increases with the increasing service level. The impact of service level is more evident than the processing time variabi lity and the number of stages on this performance measure. 4.3 Summary In this chapter we studied the per formance of KCS, Dedicated GKCS, Dedicated EKCS, Shared GKCS and Shared EKCS in a simulation of multi-stage and multi-product manufacturing systems. Results of the simulation runs for different types of control system for 95% and 98% service level were presented. Also, we discussed factor interactions by carrying out ANOVA using the SAS software. The research undertaken will be c oncluded in the next chapter and recommendations for further research will be made.

PAGE 63

52 Chapter Five Summary and Conclusions 5.1 Summary Just in time (JIT) manufacturing systems were originally designed for deterministic production environment s with a smooth and stable demand and constant processing times; their perform ance is optimum in that environment. Once implemented, however, JIT systems face the uncerta inties inherent in any manufacturing system, including variati ons in processing time and demand, as well as equipment malfunctions. The overall goal of the JIT production philosophy is to reduce or eliminate t he variations that can lead to these problems. The aim of this research was to compare the performance of Kanban Control System (KCS), G eneralized Kanban Control System (GKCS), Extended Kanban Control System ( EKCS), Shared Generalized Kanban Control System (Shared GKCS) and Shared Extended Kanban Control System (Shared EKCS). Simulation models for different types of multi-product and multi-stage control systems were developed using Arena softw are package. Each model was run with appropriate warm up period, run length and number of replications. Different performance measures such as number of backorders, average waiting time of backorders and work in process were considered. Analysis of variance was

PAGE 64

53 carried out for using a full factorial design and graphs were plotted for comparison. The significant and insignific ant factor interactions with control systems were explained. In this research we discussed t he application of Kanban-like control systems in the context of multi product manufacturing systems where the kanban cards associated with each stage are eit her dedicated to a given part type or shared among different part types. The extension of multi-product control systems in case of dedicated kanban cards is fairly straightforward since the kanban control applied to each part type is identical to that used for single-part type systems, while in case of shared kanban cards it is much more involved. 5.2 Conclusions Overall the shared control systems outperform dedicated control systems regardless of the processing time variabili ty, number of stages and service levels. The behavior of a Shared KCS is equivalent to that of corresponding Dedicated KCS. It implies that when deal ing with KCS, there is real no way actually to share a number of kanban cards among the differ ent part types. On the other hand, in case of GKCS and EKCS, the use of shared kanban cards improves the performance compared to the de dicated kanban-like systems. We observed that Shared EKCS deals with larger time variability in more effective manner compared to the other control systems, though the values for Shared GKCS are very close. However, for low and medium variability shared GKCS outperforms shared EKCS. The number of backorders for the optimal

PAGE 65

54 control system was 30 to 40% lower than for traditional KCS, a significant reduction. In order to investigate the effect of service level on performance of the different flow control strategies experiments were carried out at 95% and 98% service levels. As the service level incr eased from 95% to 98% the lower number of backorders was achieved with a signi ficant increase in average work-inprocess. The Shared EKCS makes a better service-inventory compromise than traditional KCS. The Shared GKCS result s in lower average waiting time to satisfy the backordered demand indicating responsiveness of this control system. The overall results indicate GKCS and EKCS with dedicated or shared kanbans outclass kanban control policy. The shared kanban-like control systems result in lower number of backorders, low average waiting time for backorders and lower work-in-process inventory. 5.3 Applications Kanban, a technique for work and inventory release, is a major component of Just in Time and Lean Manufacturing phi losophy. It was originally developed at Toyota in the 1950s as a way of managi ng material flow on the assembly line. Kanban Control System was firmly in pl ace in numerous Japanese plants by the early 1970's and began to be adopted in the U.S. in the 1980's. Over the past three decades the Kanban process, a highly efficient and effective factory production system, has developed into an optimum manufacturing environment leading to global competit iveness. JIT systems were originally designed for

PAGE 66

55 deterministic production environments with a smooth and stable demand and constant processing times; their perform ance is optimum in that environment. Once implemented, however, JIT systems face the uncerta inties inherent in any manufacturing system, including variati ons in processing time and demand, equipment malfunctions, as well as known or planned interruptions such as preventive maintenance. The overall goal of the JIT production philosophy is to reduce or eliminate the variations that can lead to these problems. Dramatic changes away from hi gh product throughput and high capacity loads towards the new idea of lower production times and WIP have lead to the idea of incorporating Kanban Systems in manufacturing industries (most notably in automotive industries). These systems are most commonly used to implement the pull-type control in production syst ems with aims at reducing costs by minimizing the WIP inventory. This allows an organization the ability to adapt to changes in demand, and therefore production more quickly. The essence of the Kanban concept is that a supplier, the warehouse or manufacturing should only deliver components as and when they ar e needed, so that there is no excess Inventory. A pull-type production line is a sequenc e of production stages performing various process steps on parts where each stage consists of several work stations in tandem. Within this system, workstations located along production lines only produce/deliver desired componen ts when they receive a card and an empty container, indicating that more par ts will be needed in production. In case of line interruptions, each workstati on will only produce enough components to fill

PAGE 67

56 the container and then stop. In addition, Kanban limits the amount of inventory in the process by acting as an authorization to produce more inventories. Since Kanban is a chain process in which order s flow from one process to another, the production or delivery of co mponents are pulled to the pr oduction line, in contrast to the traditional forecast oriented met hod where parts are pushed to the line. The production kanban is removed and both this and the container label are scanned. The production kanban is then pl aced near the production line as an authorization to produce another container of parts. When the production line has finished producing a full container of par ts, the production kanban is then placed in this container and moved to finished g oods storage. If preferred the production Kanban can update the production line finished goods inventory and a Kanban card is used to move the stock to the finished goods storage. Following are the examples of manufacturing systems t hat implement kanban policy but use alternate ways to transmit kanban si gnals for production authorization. Mechanical assembly clients have oft en used tote bins as their kanban signals. Each bin has information attached about the product and quantity, source location, and user location. As t he bins are emptied, they are cycled back to the producing department for refill. The total number of empty kanban bins waiting for refilling is closely controlle d and when the upper limit is approached, a signal is sent to request addi tional production help, or to plan overtime. In many assembly operations, the workbench s pace has kanban locations marked on the work surface between operations. When the kanbans are full, the preceding operation stops producing until a kanban space is once again available.

PAGE 68

57 Another example of a kanban system for a Vendor Managed Inventory item utilized simple painted lines on the sides of the item's pallet rack location. The lines represented the kanban level, or lowest level of inventory, that would trigger replenishment. For example, when stock of a specific cardboard box fell below the line, the supplier would see t he signal during daily delivery, and drop off the needed boxes the next day. As indicated in this section, applications of JIT based systems have resulted into significant improvements in system performance in terms of its responsiveness and work-in-process invent ory. This research was aimed at further improving the behavior of the tr aditional Kanban Control System for a multi-products multi-stage manufacturing environment. 5.4 Future Research This research could be extended in some areas that are not considered in this study. The most significant are outlined below. This research assumed setup times for each product type to be negligible. The effect of setup time and setup ru les should be investigated. Also, lotsizing has not been considered in this research. If the resources are not flexible enough and require significant se t-up times when going from one part to another, the production of parts has to be grouped in lots, each lot consisting parts of same type. It is t hen important to decide how the lot-sizing issue is addressed in pull-c ontrolled manufacturing systems.

PAGE 69

58 In this research, the kanbans consi dered can either be dedicated to a single part type or shared among all part types. Ho wever, it could be of interest to consider intermediate situations wher e the kanban cards are dedicated to a subset of part types but shared am ong the part types belonging to each subset. It was assumed that the transfer time to move a part from one stage to another is negligible. It maybe of interest to see the effect of transfer time of parts between stages on the different contro l policies studied in this research. It is expected that the performanc e of GKCS and EKCS will be further improve when transfer times are cons idered due to delay involved in transmitting the kanban cards in KCS system. Machine breakdown were modeled using exponentially distributed time between failures as well as time to repair. An extension of this work could be to consider the impact of preventiv e maintenance policy in reducing the impact of breakdowns. One approach to model preventive maintenance will be to include planned shutdowns. Howe ver, these will reduce the frequency and durations of breakdown failures.

PAGE 70

59 References Bard, J. F. & Golony, B. (1991). Determining the number of kanbans in a multiproduct, mult istage production system International Journal of Production Research, 29, 881-895. Baynat, B., Dallery, Y., Di Mascolo, M. & Frein, Y. (2001). A multi-class approximation technique for the analysis of kanban-like control systems. International Journal of Produc tion Research, 39(2), 307-328. Berkley, B. (1992 ). A review of the kanban production c ontrol research literature. Production and Operat ions Management, 1(4), 393-411. Buzacott, J. A. (1989). Queueing models of kanban and MRP. Engineering Cost and Production Economics, 17. Duri, C., Frein, Y. & Di Mascolo, M. (2000). Comparison among three pull control policies: kanban, base st ock and generalized kanban. Annals of Operations Research, 93, 41. Erik, L. & Bohez, J. (2004). A new generic timed Petri net model for design and performance analysis of a dual kanban FMS International Journal of Production Research, 42, 719 740. Feryal, E., Akturk, M. & Turkcan A. (2003) Interaction of design and operational parameters in periodic review kanban systems International Journal of Production Research, 41, 3315 3338. Gershwin, S., Garrett R., S heldon, X. & Lou, C. (1993). Production control for a tandem two-machine system. IIE Transactions, 25(5), pp(16). Groenvelt, H. (1993). The Just-in-Time system Handbooks in OR & MS 4, S.C Graves et al. Eds., Elsevier Science Publishers B. V., Am sterdam, 629-671. Karaesmen, F. & Dallery, Y. (2000). A performance comparison of pull type control mechanisms for multi-stage manufacturing systems International Journal of Production Economics, 68, 59-71.

PAGE 71

60 Kimura, O. & Terada, H. (1981). Design and analysis of pull system: a method of multistage production control. International Journal of Production Research, 19, 241-253. Krishnamurthy, M., Swenseth, S. & Wilson, R. (1992). Describing processing time when simulating JIT environments. International Journal of Production Research, 30(1), 1-11. Monden,Y. (1983). Toyota Production System: Practical Approach to Production Management Industrial Engineering and Management Press. Monden, Y. (1984). A simulation analysis of the J apanese just-in-time technique (with kanbans) for a multiline, multistage production system: a comment Decision Sciences, 15, 445-447. Schniederjans, M. J. (1993). Topics in Just-I n-Time management Allyn & Bacon, Boston. Uzsoy, R. & Martin-Vega, L. A. (1990). Modeling Kanban-Based Demand-Pull Systems: A Survey and Critique. Proceedings, Manufacturi ng International '90 Atlanta, GA. Zipkin, P., (1989). A kanban-like production control system: analysis of simple models. Technical report, Graduate School of Business, Columbia University, New York.


xml version 1.0 encoding UTF-8 standalone no
record xmlns http:www.loc.govMARC21slim xmlns:xsi http:www.w3.org2001XMLSchema-instance xsi:schemaLocation http:www.loc.govstandardsmarcxmlschemaMARC21slim.xsd
leader nam Ka
controlfield tag 001 001498275
003 fts
006 m||||e|||d||||||||
007 cr mnu|||uuuuu
008 041209s2004 flua sbm s000|0 eng d
datafield ind1 8 ind2 024
subfield code a E14-SFE0000558
035
(OCoLC)57722962
9
AJU6880
b SE
SFE0000558
040
FHM
c FHM
090
T56 (ONLINE)
1 100
Deokar, Sachin S.
0 245
Performance evaluation of multi-product Kanban-like control systems
h [electronic resource] /
by Sachin S. Deokar.
260
[Tampa, Fla.] :
University of South Florida,
2004.
502
Thesis (M.S.I.E.)--University of South Florida, 2004.
504
Includes bibliographical references.
516
Text (Electronic thesis) in PDF format.
538
System requirements: World Wide Web browser and PDF reader.
Mode of access: World Wide Web.
500
Title from PDF of title page.
Document formatted into pages; contains 71 pages.
520
ABSTRACT: Over the years, much attention has been given to the analysis of the pull type ordering system to reduce in-process inventory and to improve product quality. Kanban Control Systems are widely used to control the release of parts in multi-stage manufacturing systems operating under a pull mechanism. Considerable research has been done to study the individual manufacturing systems for multi stage and single product. However, not much research has been done to compare different pull control policies for multi product manufacturing systems. Most of the research done in multi-product system assumes that a kanban card is dedicated to a part type. The aim of this research is to compare the Kanban Control System (KCS), Generalized Kanban Control System (GKCS) and Extended Kanban Control System (EKCS) in the context of multi-product manufacturing systems where the kanban cards are either dedicated to a single part type or shared among the different part types.In this study, we analyze the performance of various control policies for a multi-product multi-stage manufacturing system. The manufacturing system considered in this research use a single-card kanban system, where the transportation of materials between the different work-centers is controlled by production kanbans. Demands that arrive to the system are satisfied from the finished goods inventory whenever possible and are backordered otherwise. Performance measures are number of backorders, average waiting time of backordered demand and average work in process. Our results show that Shared GKCS has lower number of backorders when the variability in the processing time is low, while Shared EKCS performs better when variability in the processing time is high. Trade off analysis was performed on average WIP and time to satisfy backorders. The Shared EKCS makes a better service-inventory compromise than traditional KCS.The Shared GKCS results in lower average waiting time to satisfy the backordered demand indicating responsiveness of this control system. The overall results indicate GKCS and EKCS with dedicated or shared kanbans outclass kanban control policy. The shared kanban-like control systems outperform dedicated control systems for all performance measures considered in this research.
590
Adviser: Khator, Suresh K.
653
dedicated cards.
shared cards.
Just-in-time.
simulation.
breakdowns.
690
Dissertations, Academic
z USF
x Industrial Engineering
Masters.
773
t USF Electronic Theses and Dissertations.
4 856
u http://digital.lib.usf.edu/?e14.558