EXAMINATION OF THE EFFECTS OF BOTTLENECKS AND PRODUCTION CONTROL RULES AT ASSEMBLY STATIONS

EXAMINATION OF THE EFFECTS OF BOTTLENECKS AND PRODUCTION CONTROL RULES AT ASSEMBLY STATIONS By TIMOTHY M. ELFTMAN A THESIS PRESENTED TO THE GRADUATE SCHOOL OF THE UNIVERSITY OF FLORIDA IN PARTIAL FULFILLMENT OF THE REQUIREMENTS FOR THE DEGREE OF MASTER OF SCIENCE UNIVERSITY OF FLORIDA 1999

Copyright 1999 by Timothy M. Elftman

ACKNOWLEDGMENTS During the last few years, I have received an incredible amount of opportunities and was given many chances to broaden my experiences. I do not believe I could acknowledge everyone who has helped me during this time. But I will highlight specific people whom I believe deserve more than this simple recognition. First and foremost, I would like to thank Sam and Charlene Scaggs for their emotional support over the last few years. I would also like to thank June Cheng for her support and understanding. I know it has been difficult. Finally, I thank Dr Tufekci for his insight and support in this paper s development. iii

TABLE OF CONTENTS page ACKNOWLEDGMENTS...iii LIST OF TABLES...vi LIST OF FIGURES...vii ACRONYMS...ix ABSTRACT...x 1 INTRODUCTION...1 1.1 Motivation...1 1. Fundamentals of Manufacturing Control Systems...8 1.3 Manufacturing Systems Philosophies...11 1.4 Manufacturing System Control Methodologies...1 1.4.1 MRP and MRPII Systems...1 1.4. DBR System...15 1.4.3 Kanban System...16 1.4.4 CONWIP...18 1.4.5 Pull-Push Systems...19 1.4.6 Comparison of Production Methodologies...1 MRP / MRPII and DBR...1 MRP / MRPII and Kanban... Kanban and CONWIP... CONWIP and Push...3 JIT and TOC...3 1.4.7 Comparison Summary...4 1.5 Preliminaries of Queuing Theory...4 1.6 Statistical Hypothesis...5 SIMULATION MODELING AND THE ASSEMBLE SYSTEM MODEL...9.1 Simulation Modeling...9.1.1 Emulated Flexible Manufacturing Laboratory Software...9 Factory Setup Object...30 Machine Object...31 iv

Dispatch and Raw Material Object...3.1. Comparison of EFML to Traditional Simulation Programs...3. Simulation Model...33..1 Experimental Conditions...36.. Calculations...37 Cycle Time...37 Final Inventory Status...37 3 FEEDER LINE ANALYSIS...38 3.1 MRP Feeder Lines...38 3. Kanban Feeder Lines...44 3.3 CONWIP Feeder Line...50 3.4 Feeder Line Production Control Systems Summary and Conclusions of Findings...55 4 ASSEMBLY SYSTEM ANALYSIS...58 4.1 Pure Push Assembly Systems Using a Synchronization Process...60 4. Pure Push Assembly Systems with No Synchronization Process...63 4..1 Process Analysis of Unmatched Feeder Line Inventory...68 4.. Verification of Analysis...69 4.3 Pull-Push Assembly Systems...70 4.4 Hybrid Pull/Push-Push Assembly Systems...74 4.5 Assembly System Summary and Conclusions of Findings...79 5 ASSEMBLY SYSTEM COMPARISON ANALYSIS...8 5.1 Push and Pull-Push Assembly Systems...84 5. Push and Hybrid Pull/Push-Push Assembly Systems...85 5.3 Hybrid and Pull-Push Assembly Systems...85 5.4 Assembly System Comparison Summary of Findings...86 6 CONCLUSIONS...88 GLOSSARY OF TERMS...9 LIST OF REFERENCES...94 BIOGRAPHICAL SKETCH...96 v

LIST OF TABLES Table page 1.1: Hypothesis Test on Variance...6 1.: Hypothesis Test on Means of Large Samples...6 1.3: Hypothesis Test on Means of Small Samples...7 4.1: Assembly System Types...59 4.: Series Comparison of Actual Feeder Line Inventory to the Predicted Value...70 vi

LIST OF FIGURES Figure page 1.1: Generalized Assembly Process...5 1.: Synchronization Process...6 1.3: General Push System...9 1.4: General Pull System...10 1.5: MRP Process...14 1.6: Single Card Kanban Process...17 1.7: CONWIP Process...19 1.8: Pull Push Process...0.1: Feeder Line Production Process...33.: Assembly System Production Process...34.3: Modified Simulation Model Configuration...35.4: EFML Simulation Model...36 3.1: Analyst Procedure...39 3.: Throughput Analysis of 3 Machine MRP Lines Interarrival Time Constant...39 3.3: Throughput Analysis of 3 Machine MRP Lines Bottleneck Position Constant...40 3.4: WIP Analysis of 3 Machine MRP Lines Interarrival Time Constant...41 3.5: WIP Analysis of 3 Machine MRP Lines Bottleneck Position Constant...4 3.6: Cycle Time Analysis of 3 Machine MRP Lines Interarrival Time Constant...43 3.7: Cycle Time Analysis of 3 Machine MRP Lines Bottleneck Position Constant...44 3.8: Throughput Analysis of 3 Machine Kanban Lines Card Allocation Constant...45 3.9: Throughput Analysis of 3 Machine Kanban Lines Bottleneck Position Constant...46 3.10: WIP Analysis of 3 Machine Kanban Lines Card Allocation Constant...47 3.11: WIP Analysis of 3 Machine Kanban Lines Bottleneck Position Constant...48 3.1: Cycle Time Analysis of 3 Machine Kanban Lines Card Allocation Constant...49 3.13: Cycle Time Analysis of 3 Machine Kanban Lines Bottleneck Position Constant...50 3.14: Throughput Analysis of 3 Machine CONWIP Lines Cards Allocated Constant...51 3.15: Throughput Analysis of 3 Machine CONWIP Lines Bottleneck Position Constant5 3.16: Cycle Time Analysis of 3 Machine CONWIP Lines Cards Allocated Constant...54 3.17: Cycle Time Analysis of 3 Machine CONWIP Lines Bottleneck Position Constant55 4.1: Throughput Analysis of Base Push Assembly System...60 4.: WIP Analysis of Base Push Assembly System...61 4.3: Cycle Time Analysis of Base Push Assembly System...63 4.4: Unmatched Inventory after 4000 Batches - Bottleneck Feeder Line...65 4.5: Unmatched Inventory after 4000 Batches - Dual Nonbottleneck Feeder Lines...65 4.6: Unmatched Inventory after 4000 Batches - Assembly Systems with One Bottleneck Feeder Line...66 vii

4.7: Unmatched Inventory after 4000 Batches - Dual Bottleneck Feeder Lines...66 4.8: Unmatched Inventory after 4000 Batches - Nonbottleneck Feeder Line...67 4.9: Unmatched Inventory after 4000 Batches - Assembly Systems with Two Bottleneck Feeder Lines...67 4.10: Assembly Station Raw Material Combination...69 4.11: Throughput Analysis of Pull-Push Assembly System...71 4.1: WIP Analysis of Pull-Push Assembly System...7 4.13: Cycle Time Analysis of Pull-Push Assembly System...73 4.14: Throughput Analysis of Hybrid Pull/Push-Push System...75 4.15: WIP Analysis of Hybrid Pull/Push-Push System...77 4.16: Cycle Time Analysis of Hybrid Pull/Push-Push System...78 5.1: Throughput Comparison Analysis of Assembly Systems...83 5.: WIP Comparison Analysis of Assembly Systems...83 5.3: Cycle Time Comparison Analysis of Assembly Systems...83 viii

ACRONYMS This thesis uses a variety of acronyms that the reader may not be aware of or that differ from literature source to literature source. These acronyms are defined when first used, but are supplied here to aid the reader. Acronym BOM CONWIP CT DBR EFML MPS MRP MRPII TOC TH WIP Definition Bill of Materials Constant Work in Process Cycle Time Drum Buffer Rope Emulated Flexible Manufacturing Laboratory Master Production Schedule Material Requirement Planning Manufacturing Resource Planning Theory of Constraints Throughput Work in Process ix

Abstract of Thesis Presented to the Graduate School of the University of Florida in Partial Fulfillment of the Requirements for the Degree of Master of Science EXAMINATION OF THE EFFECTS OF BOTTLENECKS AND PRODUCTION CONTROL RULES AT ASSEMBLY STATIONS By Timothy M. Elftman May 1999 Chairman: Dr. Suleyman Tufekci Major Department: Industrial and Systems Engineering In manufacturing centers products manufactured at different locations are often joined together at assembly stations. If not managed properly this common event can lead to orphaned products, lost throughput, and increased WIP. All of which will result in lost capital for the manufacturing center. This study analyzes an assembly line, which is fed by three parallel independent feeder lines, to determine characteristics unique to assembly systems. The assembly systems are managed by MRP / MRPII, Pull-Push, and Pull/Push-Push Hybrid production control methods. The study focuses on the effects of bottlenecks, batch synchronization, and production control methods on the assembly systems throughputs, WIPs, and cycle times. The study will show that assembly systems have several unique characteristics. The first characteristic occurs in assembly systems that use push control techniques to x

manage the assembly station. In these systems if the bottleneck feeder lines are not synchronized with the nonbottleneck feeder lines, instability results. This instability is the accumulation of the nonbottleneck feeder lines product at the assembly station. If the assembly system is left unchanged the orphaned products can grow infinite in number. The second characteristic occurs in assembly systems with two bottleneck feeder lines. In these systems the probability of both bottleneck feeder lines finishing a product simultaneously is zero. The result is a delay in processing at the assembly station that can decrease the system s throughput. Although all of the assembly systems with two bottleneck feeder lines experience this delay only the systems that manage the bottleneck feeder lines with pull techniques show a significant reduction in throughput. The third characteristic concerns assembly systems that are controlled entirely by push techniques. In these systems if the nonbottleneck feeder lines are controlled by pull techniques, the assembly system will experience decreased WIP with no change in throughput. The fourth and final characteristic is that assembly systems, which manage the bottleneck feeder lines with pull techniques, can outperform systems that manage the bottleneck feeder lines with push techniques. xi

CHAPTER 1 INTRODUCTION 1.1 Motivation Manufacturing centers are a conglomerate of workstations, assembly stations, bottleneck stations, dispatch stations, buffers, inventories, forklifts, hand trucks, and personnel. The center s dependence on the conglomerate s performance is similar to a human s dependence on his muscles, tissues, eyes, hands, legs, organs, skin, and brain. As we are more than the sum of our parts so are manufacturing centers. As succinctly put by Eliyahu Goldratt [7] the primary goal of manufacturing is to make profit in the present and in the future. Accomplishing this goal has been a daunting challenge for manufacturing managers since the dawn of time. There are many obstacles in a manufacturing enterprise that prevents management from accomplishing this goal. Goldratt calls these obstacles constraints or bottlenecks. The Theory of Constraints (TOC) is a management philosophy proposed by Goldratt that deals with managing system constraints or bottlenecks. The five-step methodology focuses on identifying the system constraints, exploiting the constraint, subordinating the rest of the system to the needs of the constraint, improving the constraint, and repeating the process continually. In a factory the bottlenecks are usually those machines or processes which control the throughput of the system. Managing the bottlenecks effectively and efficiently yields higher system throughput. Many production control systems have been proposed to 1

improve throughput in the past. Among them are the Materials Requirement Planning (MRP), Just-in-Time (JIT), Kanban, Constant Work in Process (CONWIP), and Drum- Buffer-Rope (DBR) systems. In this thesis an analysis of bottlenecks and their impact on throughput, work-in-process (WIP), and cycle time in manufacturing systems where three parallel production lines feed components into an assembly line is carried out. Successful manufacturing centers are required to identify and manage their system s throughput, WIP, and cycle time. Here, throughput is the number of final products produced per unit time by the system, WIP is the material within the system undergoing transformation into a final product, and cycle time is the average amount of time required for raw material to be transformed into a final product. Insufficient throughput leads to unmet demand. Excessive WIP requires tying up excessive capital. Excessive cycle time leads to the loss of customer orders. In short, if any of these parameters are not managed properly, then the manufacturing center loses money. These parameters are influenced by process variability, process time, process reliability, system bottlenecks, and the production control system used. Recent work has investigated how bottlenecks affect a system s throughput, WIP, and cycle time in relation to different control methodologies. The goal of these works is to determine optimal settings of control parameters within the production control systems, and selecting the appropriate production control systems for different manufacturing environments. The current manufacturing control systems may be classified into three categories. The first is MRP and its successor Manufacturing Resource Planning (MRPII). These control systems push materials into the production facility based on forecasted demand, and are thus known as push systems. In the second category of

3 control systems, known as pull systems, the material is released into the production facility only when the demand for the end product triggers it. Since the material is released into the system only when it is needed, these control system are also called JIT systems. The two popular implementations of JIT control systems are Kanban (card) control systems and CONWIP control systems. In all JIT systems the WIP is controlled by the number of authorization cards assigned to the individual workstations or system of stations. The third category of control systems is mixed control systems. In these systems, the pull and push control systems are used to manage certain segments of the production line. Examples of mixed control systems are DBR, pull-push and push-pull control systems. These systems will be further defined later in this section. There is a great amount of literature evaluating the performance of these systems. Cook demonstrates that serial production systems using DBR results in greater average throughput and lower levels of WIP variance than when the same system is managed by kanban [4]. Guide in the analysis of a re-manufacturing facility proves that DBR results in a reduction in WIP and throughput variance compared to MRP [8]. Bonvik et al. in the analysis of CONWIP, kanban, and pull-push production control systems demonstrates that the pull-push systems carries the lowest WIP at any particular throughput level with the kanban system generally carrying the highest WIP [3]. Altug also demonstrates the superiority of the pull-push control system over pure MRP, kanban, and CONWIP [1]. The above analyses were mainly conducted with serial systems, or flow lines, as the case of most manufacturing studies, but serial systems only represent a portion of production systems in manufacturing.

4 The main work done with assembly systems (systems containing parallel production lines feeding an assembly line) is modeled as a fork-join with blocking type of queuing system. The fork operation is when a product is decomposed into smaller units with each unit following a separate production process. The join operation, which occurs at the assembly station, is the synchronization of operations over a set of units. In other words, multiple parallel production lines or workstations feed an assembly station different components. The assembly station then joins the components. Blocking is the limiting factor of the number of units in the model. All fork-join articles provide description of how mathematical modeling techniques can be applied to manufacturing control methodologies, and assembly systems. Yves Dallery et al. derived methods of modeling kanban and assembly operations in production facilities [5]. Rao et al. reviews the use of queuing theory in flexible manufacturing systems, computer-integrated manufacturing, and kanban systems [13]. Agnetis et al. studies the effects that push, pull, and synchronization procedures have on assembly stations by using simulation []. His results indicate that push systems lead to increased WIP compared to pull. Furthermore, the results show that pull systems have increased throughput compared to synchronized push systems. The assembly system of Agnetis study consists of a main production line with assembly stations located within it. The assembly stations are fed from the main line and other feeder lines. In the assembly system it is detrimental for the main line to be balked. By balking the main line the throughput of the system is reduced. If the feeder line s product is late in arriving to the assembly station, or tardy, the productivity suffers. Therefore Agnetis study used the system s throughput and WIP as well as the number of tardy jobs as performance indicators.

5 In this study the performance indicators are system throughput, cycle time, and WIP. The assembly system, as indicated in figure 1.1, will be studied to determine general characteristics of the push, pull, and synchronization strategies at the assembly station. Line 1 Line Line 3 Machine 1 Machine Machine 3 Machine 4 Machine 5 Machine 6 Machine 7 Machine 8 Machine 9 Line 4 Machine 10 Machine 11 Machine 1 Figure 1.1: Generalized Assembly Process In literature, assembly stations are stations where the act of joining components is carried out. The components to be joined are not necessarily produced in that production system. By evaluating a process where the synchronization of sub-components is of valued importance the use of the assembly station term holds special significance. The assembly station is the station which two or more workstations feed with different products. The assembly station combines these items into its own unique product. A workstation is a station that does not combine products from two or more workstations. A workstation has one input and one output per product type. A sequential series of workstations is referred to as a serial production line or flow line. Parallel production lines feeding an assembly station are feeder lines. An assembly station followed by a

6 sequential series of workstations is an assembly line. A system of two or more production lines feeding an assembly line is an assembly system. The feeder lines that have the lowest average throughput are the bottleneck feeder lines, and the other feeder lines are the nonbottleneck feeder lines. Raw Material A0 A0 is transformed to A1 B0 is transformed to B1 A1 and B1 are combined into C1 Raw Material B0 Figure 1.: Synchronization Process A synchronization process occurs when each of the components used in the assembly station s product is introduced into the assembly system simultaneously to specifically combine with one another. For example the product C1, in figure 1., is manufactured from items that were introduced to the assembly system simultaneously. If product A1 is present at the assembly station and product B1 is not present, then A1 is regarded as unmatched. A consequence of this synchronization process is that the maximum amount of unmatched product A1 waiting at the assembly station is the total sum of unprocessed and processing B0 items. The primary purpose of this work is to study the effects of bottlenecks on an assembly system under different production rules. In particular the location of the

7 bottlenecks relative to the assembly station and the resulting effects on selected performance indicators of throughput, WIP, and cycle time will be studied. The goal of this thesis will be achieved in the following manner. 1. Decompose the manufacturing system into its component production lines.. Analyze these subsystems with one and two bottlenecks at differing locations, following differing production rules, and using differing control parameters. 3. Use the information generated in Step, to design a push system with a synchronization procedure, a push system without a synchronization procedure, a pull-push system, and a hybrid pull/push-push system. 4. Analyze these systems with two bottlenecks in a single feeder line. 5. Analyze these systems with one bottleneck in a feeder line, and one bottleneck in the assembly line. 6. Analyze these systems with the bottlenecks in two separate feeder lines. 7. Compare all control systems in regard to throughput, WIP, and cycle time. All statistical analyses will be conducted using hypothesis tests at an α level of 0.05. The throughputs and the WIPs will be compared by using t-statistics, and the cycle times will be compared using by z-statistics to determine the relation the parameters have to one another. The experimental data samples were generated from the following two simulation software programs: Emulated Flexible Manufacturing Laboratory (EFML), and Arena. The remainder of the paper is organized in the following manner. The latter part of this chapter will provide background information in regard to manufacturing control

8 systems, statistical hypothesis testing, and queuing theory. Chapter provides information on the EFML software and the simulation model. Chapter 3 is the in-depth analysis of the feeder lines under MRP, kanban, and CONWIP production rules. Chapter 4 is the in-depth analysis of the four assembly system types. Chapter 5 is the comparison of the assembly systems. Chapter 6 is the summary of the results and recommendations. 1. Fundamentals of Manufacturing Control Systems There are two primary manufacturing control systems: push and pull. All other control systems are either combinations or derivatives of these two systems. This section will describe the theories and philosophies associated with manufacturing production control and the following sections will be used to define different techniques that have been developed to implement these philosophies. Many theories have been proposed in managing manufacturing. In a flow shop environment each product follows a fixed routing. In a job shop environment the routing depends on the shop and the job being processed. At each station, buffers or finite storage spaces exist for receiving incoming material and storing completed units. The buffers act as a safety net to guard against line starvation and blockage caused by random events. Manufacturing control systems manage how products are passed on, how buffers are utilized, and when raw material enters the system. Push control systems utilize forecasted demand to determine a production schedule. The production schedule sets when raw material is delivered to the appropriate workstations. Each workstation provides the necessary processing to the units waiting in its buffer prior to releasing it to be transferred to the next downstream station. This cycle

9 of receiving, processing and releasing of material is carried out until the end product is complete. In a push control system shipping of goods downstream is independent of the downstream stations condition. This independence can cause problems if the downstream station is offline. If the downstream station is offline, the WIP in the system escalates until the station is online again. The WIP may or may not decrease at this time. A push system is represented in figure 1.3. The arrows in the diagram refer to the movement of the product through the system. Since the production schedule represents demand information, the quantity of moving products represents the movement of information in the system. Raw Material Process 1 Process Process 3 Process 4 Figure 1.3: General Push System End Product Pull systems rely on the status of the system to determine production. In this type of system, inventory is controlled through a system of cards. The number of cards available determines the maximum allowable inventory for a particular workstation or system of workstations. In such a system the production rate is determined by how the finished goods of the final workstation is demanded by the customer. When the finished goods are removed from the system the cards associated with these units are released.

10 The released cards enable the final station to procure additional material from the upstream station to process. Upon procurement of raw material from the upstream station and release of the associated cards, the upstream station is able to obtain its own raw material from its upstream station. This process of card release and material procurement is repeated throughout the system until the raw material of the first station is obtained. Since product movement is dependent on the condition of the next station, the entire production line may stop due to the breakdown of an upstream station. A pull system is represented in figure 1.4. The solid and dashed arrows in the diagram respectively refer to the movement of the products and information. Since the cards represent demand information, the movement of cards represents the movement of information in the system. Unlike the push system, demand information originates in the final station and proceeds to the initial station. Raw Material The solid line is product being pulled to the next station. The dashed line is the release of cards or information transfer. Process 1 Process Process 3 Process 4 Figure 1.4: General Pull System End Product

11 MRP and its successor MRPII are push systems; kanban and CONWIP are pull systems. Pull systems follow the Just-In-Time (JIT) philosophy, and DBR and some pullpush systems follow the Theory of Constraints (TOC) philosophy. 1.3 Manufacturing Systems Philosophies Philosophies in manufacturing systems define goals to be accomplished by control techniques. The JIT philosophy s goal is to have raw material of a process delivered justin-time for processing. The TOC philosophy s goal is to maximize profit. Both of these philosophies ascribe to process improvement. The improvement reduces variability caused by breakdowns and raw material shortages. JIT focuses on minimizing the waste in a system by striving for no buffers, no defects, and no variation. This is accomplished by: designing products for optimal quality, cost, and manufacturing ease, minimizing the amount of resources used to design and produce the product, designing the product to meet the customer s needs, obtaining and maintaining good relationships with suppliers and vendors, and, developing a commitment to improving the manufacturing system [1]. When JIT is implemented, its purpose is to set a production rhythm that exploits the available capacity of the system to fully meet the customer s demand. Since JIT is a pull-oriented system, the demand of the customer directly sets the production rhythm. TOC focuses on maximizing profit now and in the future by maximizing the flow in a system. This is accomplished by:

1 identifying the bottleneck in the system, scheduling jobs and operations to ensure the complete utilization of the bottleneck, determining the appropriate bottleneck buffer size to guard against upstream station variability, improving bottleneck performance, and, then repeating the process [7, 8]. TOC is a profit oriented manufacturing control system. When using the TOC system throughput is defined as the rate at which money is generate from sales, and inventory is defined as the amount of money captured in the system [7]. By defining manufacturing in this manner a bottleneck may be located off the production floor, such as poor product sales. When TOC is implemented, its purpose is to set a production rhythm that exploits the bottleneck of the system. The bottleneck is the constraint that hinders greater throughput. By scheduling the bottleneck of the system, the WIP is reduced while maintaining high throughput. Since the bottleneck determines the capacity of the system by improving bottleneck performance, the system s capacity is improved. [7, 8] 1.4 Manufacturing System Control Methodologies 1.4.1 MRP and MRPII Systems MRP is the oldest push-type manufacturing control system. Its major components are the bill of materials (BOM), the master production schedule (MPS), and the materials requirement planning system. The BOM is a chart that shows the required components at

13 each stage of production starting from the final product, preceding with the intermediate products, and then ending with the raw material for the initial processes. Each stage of the BOM lists the quantities and the types of components required in producing that stage s product. The MPS contains information such as the time required at each stage of production, the outstanding order status, the inventory status, and the demand for the final product. The production time at each stage is regarded as fixed and the demand is forecasted. The MRP system, as shown below, determines the production schedule. 1. Determine net requirements by subtracting on-hand inventory and scheduled receipts from demanded requirements.. Determine the job lot sizes. 3. Offset the due dates of the individual jobs with the production times of the product to arrive at the start times. 4. Using the start times, the lot sizes, and the BOM determine the demanded requirements for the material used in the production of that stages product. 5. Starting with the final product repeat this process until all production stages have been processed. MRPII adds capacity analysis to MRP by incorporating information such as setup time, resource requirements, and labor requirements into the MRP system. Through the use of this information the MRP system provides a more realistic production schedule. Figure 1.5 demonstrates how the MRP system generates planned order releases or a production schedule for the manufacturing facility.

14 MPS BOM Inventory MRP Process Net Requirements Job Lot Sizes Offsetting Explosion Purchase Orders Production Schedule Figure 1.5: MRP Process Although extensively used in the United States, MRP and its successor MRPII have many shortcomings. Some of them are listed below. The MRP / MRPII systems do not consider fluctuations in production time due to worker illness, machine breakdown, demand change, and availability of raw material. To accommodate for these uncertainties safety stocks and safety lead times are often used, but the inclusion of the safety stocks and safety lead times increases inventory levels and production cycle times. MRP systems assume fixed cycle times or lead times regardless of the inventory level. A consequence of this assumption coupled with a large WIP leads to a large throughput. Regretfully throughput is limited by the production rate of the bottleneck station. Once the throughput is maximized, any additional inventory in the system results only in increased cycle time [14].

15 The purpose of MRP systems is to meet the projected demand as provided by the MPS. No effort is specifically expended to improve production. 1.4. DBR System DBR is a newer system of production control that follows the TOC philosophy. In doing so, it concentrates on managing the flow of products to meet the bottleneck constraint s needs. Since the bottleneck acts as a valve controlling the system s throughput, managing the bottleneck s throughput manages the system s throughput. To maximize the system s throughput, the bottleneck must utilize all of its available capacity. Similar to the MRP / MRPII systems, the DBR system uses a scheduled release of products to control the production rate, and a safety stock or buffer at the bottleneck to guard against stoppages from the upstream workstations [8]. Although the DBR control system provides improvement over the MRP / MRPII systems, it is not immune from shortcomings. Some of them are listed below. Failure to locate the bottleneck of the system will result in lost throughput, or increased WIP and cycle time depending on the false bottlenecks location relative to the real bottleneck. The use of fixed lead times to schedule the bottleneck can lead to increased WIP much as in MRP / MRPII systems. Incorrect bottleneck buffer size can result in bottleneck starvation; thus system throughput is lost [15].

16 1.4.3 Kanban System The kanban system was developed by Japanese manufacturers to implement the JIT philosophy. In this system the kanban acts as an inventory control mechanism and information relay device. It controls the inventory by requiring every batch in production to be assigned to a kanban. The number of kanbans in the system thus determines the maximum inventory possible. The kanbans transmit demand information from the downstream stations to the upstream stations through the number of kanbans available and how often the kanbans become available. The single card kanban system, figure 1.6, allocates a set amount of kanban cards to each workstation in the system. A kanban card is initially attached to a batch to be processed by that workstation. The kanban card stays attached to the batch until a downstream workstation has a kanban card available. When this occurs the attached kanban card is freed from the workstation s product and the previously freed kanban card from the upstream workstation becomes assigned to that batch. Thus a free kanban card allows a workstation to obtain material from the previous station when the material is available [14].

17 Raw Material The solid line is product being pulled to the next station. The dashed line is the release of cards or information transfer. Process 1 Process Process 3 Process 4 Figure 1.6: Single Card Kanban Process End Product Successful implementation of the kanban system requires large production runs, minimal defects, steady demand, reliable workstations, few product types, and reliable vendors. Determining the number of kanban cards to allocate to each workstation is of significant importance. One manner of determining the initial amount of kanban cards uses the formulation below. N > D * L (1 + α) / a Here, N equals the number of kanban cards allocated to the workstation, D is the demanded throughput, L is the cycle time, α is the safety factor, and a is the batch size. N is first estimated by choosing a high α value in the range [0, 1]. Once the number of kanban cards is determined the system is operated for a set length of time. Depending on the production system dynamics, N is then adjusted on each separate workstation in an effort to reduce WIP but maintain throughput [3].

18 For all of kanban s improvements to the production system, it also has its shortcomings. Some of them are listed below. Kanban systems are not suited for manufacturing environments with short production runs, highly variable product demand, poor quality products, and a multitude of product types [11]. A breakdown in the kanban system can result in the entire line shutting down. The throughput of a kanban system is not managed but is instead a result of controlled WIP and known cycle times. 1.4.4 CONWIP The CONWIP system is a generalized form of the kanban system. Like kanban, CONWIP uses cards to limit the WIP of a system; unlike kanban the cards are allocated to a system of workstations instead of just one. This difference allows CONWIP to be applied in production environments that are detrimental to the kanban system [10]. In a CONWIP system the cards get attached to batches only at the first station. The card remains affixed to the batch until the batch has finished processing on the final workstation of the CONWIP system and the batch is used to satisfy a customer s demand. The released card is then returned to the initial workstation of the CONWIP line and to authorize the entry of a new batch into the system. Under a CONWIP system enough cards should be allocated to ensure the bottleneck is fully utilized. If the number of cards is insufficient the bottleneck starves and thereby reduces the system s throughput. Figure 1.7 shows the CONWIP process.

19 Raw Material The solid line is product being pushed to the next station. The dashed line is the release of cards or information transfer. The double line is product being pulled into the system. Process 1 Process Process 3 Process 4 Figure 1.7: CONWIP Process End Product Even though CONWIP generally provides improvement over kanban and MRP / MRPII, it does have its share of shortcomings. Some of them are listed below. Kanban systems can achieve higher throughput with lower WIP in some situations over CONWIP systems. CONWIP systems cannot be successfully implemented in a job shop environment. Incorrect card determination can lead to increased WIP or lost throughput for the system. Machine breakdown can bring the entire CONWIP system to a halt. 1.4.5 Pull-Push Systems Spearman et al. proposed a generalization of DBR that is modeled by using CONWIP and a push system on a flow line [14]. By using a CONWIP system that encapsulates all stations between and including the initial and the bottleneck stations, and using a push system following the bottleneck station, the DBR system is approximated.

0 The DBR system determines a production schedule to ensure bottleneck of the system is completely utilized. In the DBR system, the bottleneck is protected against variation from the upstream stations via a buffer prior to the bottleneck. In the pull-push system enough cards are allocated to the CONWIP segment to ensure the bottleneck is completely utilized. Prior to the bottleneck a buffer will naturally develop and will be limited in quantity by the CONWIP cards. Following the bottleneck in both systems the batches are pushed through as fast as possible. The pull-push process is represented in figure 1.8, where process 4 is the bottleneck process. Raw Material The solid line is product being pushed to the next station. The dashed line is the release of cards or information transfer. The double line is product being pulled into the system. Process 4 is the bottleneck. Process 1 Process Process 3 Process 4 Process 8 Process 7 Process 6 Process 5 End Product Figure 1.8: Pull Push Process

1 Although the pull-push system provides improvements to pull systems and DBR it still has some shortcomings. Some of them are listed below. Failure to locate the bottleneck of the system will result in lost throughput, or increased WIP and cycle time depending on the false bottlenecks location relative to the real bottleneck. Incorrect card determination can lead to increased WIP or lost throughput for the system. Machine breakdown can bring the entire pull-push system to a halt. Increased complexity over pure systems. 1.4.6 Comparison of Production Methodologies There is extensive literature proving that pull systems tend to have lower WIP and cycle time mean and variance compared to push systems. Pull systems are also easy to control since WIP can be controlled directly whereas push systems manage throughput. On the other hand, push systems can be implemented in many environments [4, 9, 14]. MRP / MRPII and DBR MRP / MRPII and DBR are very similar systems, the difference lies in the focus of production and the manner in which it is carried out. MRP / MRPII focuses on maximizing the capacity of the production system. DBR focuses on maximizing the flow of the production system. As a result DBR experiences reduced WIP levels and are more capable of adjusting to fluctuations in the production environment [4].

MRP / MRPII and Kanban MRP / MRPII and kanban systems differ in philosophies, environmental settings, and control methods. MRP / MRPII systems operate under the philosophy of maximizing throughput, can be applied in most manufacturing environments, and place production control in the production schedule. Kanban systems use a philosophy of improvement, require stable environments with large production runs, small setup times, minimal defects, consistent demand, and places production control on the factory floor. Once the production environment is achieved, the kanban system can achieve high throughput with lower amounts of WIP compared to the MRP / MRPII systems [4, 11]. Kanban and CONWIP Kanban and CONWIP systems only differ in that kanban fixes the inventory on a per station basis, whereas CONWIP fixes the inventory on a per system basis. This difference in implementation results in the following performance differences [9, 14]. CONWIP can be implemented in production environments that have variable demand and have a multitude of products, whereas kanban requires stable environments and few product types. CONWIP does not attempt to control the location of the WIP in the production system. CONWIP generally results in lower WIP levels than kanban given the same throughput levels. Kanban in some situations can outperform CONWIP by optimally placing cards in some systems [6].

3 CONWIP and Push CONWIP is superior to push systems when the production systems run under the highest possible throughput rate. In this situation at equivalent throughput rates, the push system experience higher WIP and cycle time compared to the CONWIP system [4, 9, 14]. JIT and TOC TOC and JIT, through different approaches in managing a production environment, achieve similar results. Both systems strive to improve and reduce variation in the production system. TOC concentrates on improving the bottleneck station, and JIT improves each station in the system. Both systems experience a reduction in WIP compared to MRP / MRPII systems at equivalent throughputs. TOC accomplishes this by scheduling the bottleneck to its fullest, and JIT does this by allocating kanban cards to keep the bottleneck working. These systems differ in that TOC generally provides better throughput than JIT with only slightly higher levels of WIP and greater cycle time [4].

4 1.4.7 Comparison Summary From the previous descriptions of the above production systems the following can be inferred. Scheduled releases of raw material can lead to increased system WIP, depending on the variability of the system. Complete utilization of bottleneck maximizes system throughput. If the system conditions determine the release of raw material, the WIP of the system is controlled. Pull systems are much more susceptible to system variation than push systems. Improving the production system can decrease system variability. 1.5 Preliminaries of Queuing Theory Queuing theory studies how people, messages, and items flow through a system. Practically all models and formulas developed in queuing theory are for systems in steady state or equilibrium conditions. For a queuing system to be in steady-state the average capacity of the system (C) must always be greater than the average arrival of entities to the system (R). The above statement, C > R, is true for a single server system as well as for networks of queues, as captured by Little s Law. In a steady state queuing system the following fundamental relationship N = λt always holds true. Little s Law states that the average number of customers in a queuing system (N) is equal to the average arrival rate of customers to that system (λ), times the average time spent in that system (T). Furthermore, it does

5 not depend upon any specific assumptions regarding the arrival rate or the service time distribution; nor does it depend on the particular queuing discipline within the system. [10, page 17] In manufacturing, production lines can be viewed as queuing networks. Ergo, the above law may be adapted as follows: Work-In-Process (WIP) is equal to throughput (TH) times cycle time (CT). WIP = TH* CT 1.6 Statistical Hypothesis A statistical hypothesis is a formal statement concerning the parameters of a probabilistic distribution. In order to check a parameter s relationship to a specific value hypothesis testing procedures are used. In testing a hypothesis a random sample is obtained from the population under study and used to generate a test statistic. The value of the test statistic determines whether to reject or fail to reject the null hypothesis, Ho. The results obtained from the simulation programs often yielded similar results. In order to test if a difference existed or if one parameter was greater than the other statistical hypothesis tests were used. Based on the number of runs carried out, different test statistics were used to validate the null hypothesis. The test results are available in chapters 3, 4, and 5. Tables 1.1, 1., and 1.3 illustrate the types of tests used in the analysis of the simulation results.

6 Table 1.1: Hypothesis Test on Variance Hypothesis Test Statistic Criteria for Rejection H : σ = σ 0 1 S1 F H : σ σ 0 = S 1 1 F0 > Fα, n1 1, n 1 F0 < F α, n1 1, n 1 1 H : σ = σ 0 1 H : σ > σ 1 1 S F 0 = F0 > Fα, n1 1, n 1 S 1 H : σ = σ 0 1 S1 F F0 < F1 α, n 1, n1 1 H : σ < σ 0 = S 1 1 Table 1.: Hypothesis Test on Means of Large Samples Hypothesis Test Statistic Criteria for Rejection H : µ = µ 1 0 1 Z 0 = H : µ µ S1 S 1 1 + n1 n x H : µ = µ 1 0 1 Z 0 = H : µ > µ S1 S 1 1 + n1 n x x x x H : µ = µ 1 0 1 Z 0 = H : µ < µ S1 S 1 1 + n1 n x Z 0 > Zα Z 0 < Z α Z 0 > Z α

7 Table 1.3: Hypothesis Test on Means of Small Samples Hypothesis Test Statistic Criteria for Rejection H H H H : µ = µ 0 1 : µ µ 1 1 0 : µ 1 = µ : µ < µ 1 1 σ = t 0 1 σ = S x 1 1 n 1 v = n n x p + 1 + 1 n t 0 > tα, v t 0 < tα, v H H 0 : µ 1 = µ : µ > µ 1 1 S p ( n = 1 1) * S1 + ( n 1) * S n + n 1 t 0 > tα, v H H H H H H 0 : µ 1 = µ : µ µ 1 1 : µ = µ 0 1 : µ < µ 1 1 : µ = µ 0 1 : µ > µ 1 1 σ 1 σ t 0 = S1 n1 v = S 1 n1 n + 1 x 1 S n 1 1 x S + n S + n S n + n + 1 1 t > t 0 α, v t 0 < tα, v t 0 > tα, v The use of hypothesis tests requires a certain degree of normality for the sample data. The central limit theorem (CLT) ensures this for the larger samples. The CLT implies that the sum of n independently distributed random variables regardless of

8 distribution followed is approximately normal. For the smaller samples normality is assumed.

CHAPTER SIMULATION MODELING AND THE ASSEMBLE SYSTEM MODEL.1 Simulation Modeling Simulation is a tool that allows actual or hypothetical facilities or processes to be studied. Similar to other methods of analysis an accurate model is essential for the analysis to be meaningful. The simulation models advantage over mathematical models is that it can be used to study large and complex systems quickly. The simulation program s speed and its ability to collect system data enables the analyst to study alterations to the system easily. But unlike mathematical models, simulation cannot place theoretical limits on a system and it cannot prove innate characteristics of the system..1.1 Emulated Flexible Manufacturing Laboratory Software The Emulated Flexible Manufacturing Laboratory (EFML) Software is a real-time simulation program developed in the Industrial and Systems Engineering Department at the University of Florida. It has been designed using Borland s Delphi 4.0 and runs on the Windows NT, 95, and 98 platforms with TCP/IP network communication protocol. The software creates an interactive environment that allows users to experience the basics in operation management, and production control systems. 9

30 EFML is written in an object oriented programming language. Each object in EFML emulates an actual object in a manufacturing plant and is represented in an object window. The main objects are factory setup, machine, dispatch and raw material, transport, repair, and finished goods. These objects communicate with each other using a message passing protocol over the internet thus allowing large facility and supply chain management emulation. A small percentage of the software s capabilities were utilized during this study. Of all the available objects only the factory setup, machine, and the dispatch and raw material objects were used. Factory Setup Object The factory setup object is the main object of EFML. It controls the factory setup, the emulation model loading, the starting and stopping of the emulation, and the gathering of the emulation statistics into a usable report. Additionally this window allows the user to view and adjust remote system components and parameters. Once EFML is started, a preset factory can be loaded by selecting Autoload from the Global menu, or a factory can be built with the objects available to the user. If a preset factory is loaded, selecting objects and changing their parameters can alter the factory. If the user desires to save this change for future factory runs, it can be accomplished using the Save option from the File menu. When the setup is accomplished manually, the user chooses the needed objects from the Components bar and places them in the active setup window. The setup window serves as a background and moderator for the other objects (windows). The overall control of the emulation resides in the setup object. In this object, the user adjusts

31 the speed of the emulation, the stop condition, where the report file is saved, and can manually start and stop the emulation. Machine Object The machine object serves a dual purpose. It can be used to represent a process center or it can emulate an assembly station. During emulation each machine is in one of four states: idle, running, setup, and breakdown. In the running mode the machine follows MRP, CONWIP, or kanban production rules. Each of these states and modes of the machine object are color coded for ease of viewing. The color codes and states are listed below. Idle status and any production process is white. Running status and MRP is blue. Running status and CONWIP is green. Running status and kanban is red. Setup status and setup is the color of the production process (green, blue, or red). Breakdown status and any production process is yellow. When setting up a machine object, the user is asked to specify the incoming and outgoing batch names and sizes, the production rule to follow, the process distribution and parameters, as well as the failure rate of the machine. While running, the machine objects compute the statistics on the mean and variance of the machine s processing time, the mean and variance of the machine s cycle time, the machine s throughput, the batch s start and end times, and the number of batches processed.

3 Batches move through the system as directed by the machine and dispatch objects and follow the implemented production control system. While at a machine the batch is in one of three states: waiting in the in-queue buffer, being processed, or waiting in the outqueue buffer. Following processing the batches are transported to the next machine object or to the finished goods object. Dispatch and Raw Material Object The functions of the dispatch and raw material object include the storage of the generated BOM, storage of the product routing, managing raw material inventory, and dispatching of inventory to the appropriate workstations. The dispatching can either be performed automatically or manually. In addition this object may be used to track inventory cost and record transaction data..1. Comparison of EFML to Traditional Simulation Programs In traditional simulation programs, model events are determined to occur at a specific time. The simulation program finds the earliest occurring event, sets its simulation clock to that time, carries out any required functions, and then looks for the next event to repeat this process. EFML, unlike the traditional simulation programs, advances through time and at each clock tick determines if an event has occurred. If an event occurred, EFML carries out any required functions before advancing to the next clock tick to repeat the process. In this manner EFML runs in real time and allows individuals to see virtual processes being accomplished.

33. Simulation Model The primary system under analysis is an assembly system composed of three feeder lines, and a three-station assembly line. The three-workstation feeder lines products are the raw material for the assembly line. The assembly system is depicted in figure.1. Using throughput, WIP, and cycle time as performance indicators, the goal of this study is to ascertain general effects of push systems with and without batch synchronization procedures implemented, pull-push systems, and hybrid pull/push-push systems, in the presence of bottlenecks. In particular the location of the bottlenecks relative the assembly station and the resulting effects on the performance indicators of the assembly systems will be studied. The feeder lines, figure.4, were studied using one and two bottlenecks under differing bottleneck locations, differing production rules, and differing control settings. Following the feeder line study the full assembly system is studied extensively. In all our experiments the processing times are assumed to be independently and identically distributed random variables. Single Line Machine 1 Machine Machine 3 Figure.1: Feeder Line Production Process

34 The assembly system in this study is assumed to contain two separate bottlenecks with at least one of the bottlenecks located prior to the assembly station. A description of the four assembly systems follows. The base push system is a pure push system using a batch synchronization procedure. Batch synchronization is accomplished by releasing the raw material to all the feeder lines simultaneously. The pure push systems without a batch synchronization procedure use different interarrival times for the raw material of the bottleneck and nonbottleneck feeder lines. The pull-push system uses CONWIP control rules to manage all of the feeder lines. The assembly line is managed by a push system. The hybrid pull/push-push system uses a pull process to manage the nonbottleneck feeder lines and a push process elsewhere. The entire manufacturing assembly system studied in this paper is shown in figure.. Line 1 Line Line 3 Machine 1 Machine Machine 3 Machine 4 Machine 5 Machine 6 Machine 7 Machine 8 Machine 9 Line 4 Machine 10 Machine 11 Machine 1 Figure.: Assembly System Production Process

35 EFML though a very dynamic and powerful software package was not originally designed to handle feeder lines / assembly stations models for all control systems. In order to create the experiments a dummy workstation with a constant zero processing time is used to match the required batches in front of the assembly line. Figure.3 is the new simulation model configuration and figure.4 is the modified assembly system with the dummy or matching station noted. Line 1 Line Line 3 Machine 1 Machine Machine 3 Machine 4 Machine 5 Machine 6 Machine 7 Machine 8 Machine 9 Dummy Matching Station Line 4 Machine 10 Machine 11 Machine 1 Figure.3: Modified Simulation Model Configuration

36 Matching Station Dispatch Object Machine Object Setup Object Figure.4: EFML Simulation Model..1 Experimental Conditions The following control parameter settings were used throughout the experiments. The processing times on each nonbottleneck machine follows an exponential distribution with a mean of 5 seconds per item. The processing times on each bottleneck machine follows an exponential distribution with a mean of 30 seconds per item. The batch size is 10 items. In order to reach equilibrium conditions, each experiment was run until 4000 batches were produced.