Workload control in job shops, grasping the tap. Martin Land

Transcription

1 Workload control in job shops, grasping the tap Martin Land

2 Published by: Labyrint Publications PO box AX Ridderkerk The Netherlands Tel: +31 (0) Printed by: Offsetdrukkerij Ridderprint B.V., Ridderkerk ISBN , M.J. Land Alle rechten voorbehouden. Niets uit deze uitgave mag worden verveelvoudigd, opgeslagen in een geautomatiseerd gegevensbestand, of openbaar gemaakt, in enige vorm of op enige wijze, hetzij electronisch, mechanisch, door fotokopieën, opnamen, of enig andere manier, zonder voorafgaande schriftelijke toestemming van de uitgever. All rights reserved. No part of this publication may be reproduced, stored in a retrieval system of any nature, or transmitted in any form or by any means, electronic, mechanical, now known or hereafter invented, including photocopying or recording, without prior written permission of the publisher.

3 RIJKSUNIVERSITEIT GRONINGEN Workload Control in job shops, grasping the tap Proefschrift ter verkrijging van het doctoraat in de Bedrijfskunde aan de Rijksuniversiteit Groningen op gezag van de Rector Magnificus, dr. F. Zwarts, in het openbaar te verdedigen op donderdag 29 april 2004 om uur door Martin Jaap Land geboren op 18 mei 1968 te Groningen

4 Promotor : Prof. dr. ir. G.J.C. Gaalman Copromotor : Dr. W.M. Nawijn Beoordelingscommissie : Prof. dr. J. Wijngaard Prof. dr. ir. J.W.M. Bertrand Prof. dr. M. Perona

5 Preface Writing this thesis was akin to a long journey. I got a great deal of pleasure out of it, but as I was approaching the end of it, I was looking forward to coming home and to making plans for the next voyage. On my journey I had the best company one could ever wish for, the impressive Operations Management group in Groningen. Indeed, so many people showed me the way that it is impossible to name them all. My supervisor, Gerard Gaalman, aroused my interest in the field of Operations Management while I studied Econometrics and Operations Research at the University of Groningen. As an instructor he made a lasting impression. He still serves as an example to me, both as an individual and as a scientist. Wim Nawijn from the University of Twente joined the supervision team later. He always made time whenever I contacted him and has been a great help. During my journey I had the pleasure of sharing ideas with many European colleagues in the field of Workload Control. Our meeting point has often been the inspiring environment of Igls in Austria. I have very fond memories of the meetings with Linda Hendry, Mark Stevenson, Alberto Portioli-Staudacher, Donatella Corti, Henny van Ooijen, Günther Zäpfel, Hubert Missbauer, Manfred Gronalt, and, of course, Jan-Wilhelm Breithaupt and Peter Nyhuis with whom I wrote one of the articles in this thesis. Peter Henrich joined our Workload Control crew in Groningen and makes enthusiastic contributions to our research. Brian Kingsman from Lancaster University has been extremely stimulating to me. His untimely death, which occurred just before the completion of my journey, made me feel that I had arrived too late. Our plans for future co-operation will now never go beyond the stage of dreams. I will miss him at my thesis defence but the memory of Brian will always be with me. It is a great honour to have Marco Perona and Wil Bertrand, two other highly respected researchers in the field of Workload Control, on the Assessment Committee. Among all his other research interests, Marco Perona keeps our European collaboration in Workload Control alive. His energy is inexhaustible and contagious. Wil Bertrand wrote his seminal thesis on Workload Control more than twenty years ago, and it remains one of the richest sources for research in that field. Jacob Wijngaard is one of the committee members and has been a great boss. His creativeness, pragmatism, drive and humour are unparalleled. Under Jacob s direction a dynamic Production Management department has been developed. With my current colleagues Gera, Jan, Leonieke, Manda, Marjan, Marjo, Renny, Taco, the words colleagues and friends have become synonymous. I also owe thanks to many other (ex)colleagues, and in particular Manon and Klaske who created a haven for me at WSN 817 for many years.

6 My paranimfen Jan Riezebos and Bas Oosterman mean a lot to me. Jan introduced me to the field of Operations Management. I continue to learn from him and I really enjoy working with him. Bas contributed to chapter 7 of this thesis. He is not only a great friend, but he was also a jovial colleague who brightened up my WSN days for many years. My parents have always been there for me, offering their unconditional love. Last but not least, I turn my attention to Antheia. She is a wonderful source for support and inspiration. This journey would not have been complete without her.

7

8 Contents Chapter 1 Introduction Research backgrounds Objectives and research questions Thesis outline... 6 Chapter 2 Job shop control Performance in job shops Utilisation/throughput Inventories Lead times Delivery reliability Job acceptance and due date assignment Job release Infinite loading methods Finite loading methods Dispatching Conserving flow Improving throughput Reducing the dispersion of lateness Combined rules Job shop control in this thesis Chapter 3 The workload control (WLC) concept Workloads and throughput times The relationship visualised Little s result The role of processing time units A decomposition of workloads and throughput times The WLC decision framework Entry level Release level Dispatching level The classical WLC release methods The basic release procedure Workload measures Workload norms and other parameters... 42

9 3.4 Extensions and adjustments of the classical WLC methods Inclusion of downstream loads to reduce feedback requirements Intermediate release Pool restrictions Analysis and considerations Timing qualities Balancing qualities: realising the norm level Balancing: direct load indicating qualities of the workload measures Detailed research questions Appendix chapter Chapter 4 Performance of the WLC release methods Experimental design Job and shop characteristics Entry Release Priority dispatching Performance measurement Results Performance overview Load balancing Timing Sensitivity Utilisation level Planned station throughput time Length of the release period Time limit Summary and assessment of results Appendix chapter 4: simulation parameters Chapter 5 Redesign of WLC release methods WLC without norms: SLAR Design principles SLAR SLAR in chapter Norm improvement: corrected aggregate loads Design principles Corrected aggregate loads Corrected aggregate loads in chapter Appendix chapter

10 Chapter 6 WLC without norms (article) Introduction Job release within workload control (WLC) concepts Two classical approaches of workload control The release procedure A simulation of the WLC release methods Modelled release procedure Experimental design Simulation results Discussion of results Opportunities for improvement An alternative release approach for workload control Release method based on the two release principles Simulation results Discussion of results Conclusions Chapter 7 Norm improvement (article) Introduction The workload control (WLC) concept and job release Three approaches of workload control Expected influences of shop characteristics Experimental design Release methods Shop configurations Workload norms and performance measurement Simulation results Discussion of results The influence of variable station positions and routing lengths The influence of a directed flow Conclusions Chapter 8 Conclusions General conclusions Timing and balancing performance Influence of norm types and norm levels Sensitivity to other factors WLC without norms Norm improvement Final remarks

11 Appendix A (article) A.1 Introduction A.2 The workload control paradigm A.3 Existing WLC concepts A.3.1 Bechte s WLC concept A.3.2 Bertrand s WLC concept A.3.3 Tatsiopoulos WLC concept A.3.4 Other methods of controlled release A.4 Workload definitions and shop floor stationarity A.4.1 Bechte: Queue only A.4.2 Bertrand: Queue and upstream included A.4.3 Tatsiopoulos: Queue, upstream and downstream included A.5 The timing/balancing conflict and pool stationarity A.6 Conclusions and suggestions for further research Appendix B (article) B.1 Introduction B.2 The workload control (WLC) concept B.2.1 General principles B.2.2 LOOR (Load-Oriented Order Release) B.3 Strengths and weaknesses B.3.1 Use of norms B.3.2 Reducing and balancing workloads B.3.3 LOOR-specific aspects B.4 LOOR extensions based on experiences from practice B.4.1 Determination of appropriate parameter values B.4.2 Uncertainties during the load conversion procedure B.4.3 Rejection of orders caused by overloaded downstream centres B.5 Conclusions Summary Samenvatting References

12 Input Output The bathtub model [after Plossl & Wight 1971]

13

14 Chapter 1 Introduction 1.1 Research backgrounds Workload control in job shops, grasping the tap is the title of this dissertation. In the early nineteenseventies Plossl and Wight [1971] used the picture of a bathtub to illustrate the basics of controlling the workload (the water in the tub). The tap controlling the flow into the bathtub depicts the decision of releasing orders for production. Like its Dutch synonym vatten, the verb to grasp has a double meaning: (1) to take something and hold it firmly, and (2) to understand something. In this thesis we will grasp (1) the tap, with the aim of grasping (2) the tap. The reader might question what should be grasped with respect to something as simple as a tap. But seemingly meaningless, the tap releasing orders is an essential element in controlling the workload in the bathtub and with consequences difficult to understand, especially since our bathtub is a production environment indicated as Job Shop and our tap is delivered by the concept of Workload Control (WLC). The tap of the WLC concept contains a smart sifting mechanism to release the right order at the right time. The term job shop is used to indicate a type of manufacturing situation where a large number of different products are produced according to customer specification with highly variable routings and processing times. We often observe this type of situation for small-subcontractor firms producing machine parts. Individual partproducing departments of machine-building companies may have similar characteristics. The production orders are generally indicated as jobs. Job flows through job shops cause one of the more difficult control problems, i.e. it is nearly impossible to predict which future states of the job shop a job will encounter, even when its routing and processing times could be determined accurately upon its arrival. Research on job shop control has been performed from different perspectives. The static scheduling perspective supposes a certain set of jobs to be available, creating a schedule to optimise a certain performance objective. The resulting problem is known to be a so-called NP-hard problem, requiring long computing times for any reasonable problem size. Another perspective supposes the more realistic dynamic situation where new jobs arrive continuously. In that case, scheduling exact start and end times for the operations of a job becomes less realistic. New arrivals combined with the uncertainty and disturbances inherent to job shops Introduction 1

15 would create schedule instability and put high demand on feedback from the shop floor. A large body of research on dynamic job shops has focused on priority rules, specifying sequencing criteria to be followed at each workstation. This approach acknowledges that it is nearly inevitable that jobs have to wait for the availability of workstations in a job shop, resulting in queues of jobs. The queues act as buffers that guarantee sufficient utilisation of the workstations. Control of these queues is one of the main principles of the workload control (WLC) concept, one of the few comprehensive control concepts developed particularly for job shops [Hendry & Kingsman 1989, Henrich et al. 2003a]. The research in this thesis is confined to this concept. The first publications on WLC as a comprehensive concept appeared in the early nineteeneighties, but ingredients have been developed before. In general, research on job shop control finds its roots early in the twentieth century. Combinatorial and queuing theory approaches had already been developed, when computers first made it possible to simulate job shops in the nineteenfifties. Probably the first textbook that extensively dealt with job shop control was written by Conway, Maxwell and Miller [1967]. It gives a broad overview of the basic knowledge resulting from the different research approaches. The dissertation of Bakker [1965] on lead times in job shops makes notice of a concept with the Dutch name of doorstroomplanning (Eng. throughput planning), which can be seen as one of the parents of the workload control (WLC) concept. The basic idea behind doorstroomplanning is the planning of release dates for jobs based on processing and waiting times. Bakker already noticed that this approach requires controlling the quantity of work per workstation because of the relationship between work-in-process and average waiting times. When doorstroomplanning is identified as one parent of the WLC concept, Input/Output control must genetically be the other. Doorstroomplanning concentrated on determining planned release dates, Input/Output control on controlling the underlying planned throughput times by monitoring backlogs of work. Input/Output control was introduced by Wight [1970] as a prerequisite for material requirements planning (mrp) which became popular at that time. In the early nineteeneighties three dissertations appeared [Bechte 1980, Bertrand & Wortmann 1981, and Tatsiopoulos 1983] which independently elaborated the above ideas into roughly comparable control approaches for job shop production. These approaches restrict the release of work to the shop floor by workload norms for each workstation - in terms of processing time units - in order to control throughput times. The controlled throughput times are used in turn to determine accurate planned release dates for each job. A set of jobs is released periodically, after being checked - in order of planned release date - on workload consequences. Once on the shop floor, simple priority dispatching rules are used locally at the workstations to control the 2 Chapter 1

16 progress of jobs. The main difference between the approaches suggested in each of the dissertations can be found in the type of workload norms that is used. Land and Gaalman [1996a] were the first to compare these approaches in detail. They indicated the common backbone of these approaches as the workload control (WLC) concept. [Land and Gaalman 1996a] is included as appendix A of this thesis. The WLC concept provides a specific approach to controlled release, though numerous other approaches to controlled release have been researched. For reviews of the complete spectrum of release methods we refer to Wisner [1995] and Bergamaschi et al. [1997]. Since the development of the WLC concept, paradoxical research results have evoked confusion on the contribution of controlled release methods in job shops. Practitioners have widely acclaimed the importance of controlled release, while simulation studies show adverse effects on performance. A debate on this paradox has been raised by Melnyk and Ragatz [1989]. It should be recognised that the claims of practitioners are hardly supported by evidence from good empirical research. But the question why controlled release has not contributed to performance in most simulation studies has puzzled many academics. Simulation studies modelling job shops as (semi-)open queuing systems report increasing total lead times, when controlled release methods replace direct release. That is, the average waiting time before release exceeds the reduction of average waiting times on the shop floor. Often explanations have been sought in modelling factors, external to the release method, more realistically. As such, Kingsman and Hendry [2002] have investigated output control, and Melnyk et al. [1991, 1992] and Van Ooijen [1996] have investigated the influences of load smoothing at higher planning levels on the performance of controlled release. In addition, Van Ooijen has modelled a relationship between workload and worker productivity - elaborated in [Bertrand & Van Ooijen 2002, Van Ooijen & Bertrand 2003] - which clearly influences the effectiveness of controlled release. More generally, Land and Gaalman [1996a] have raised the question whether the dynamics of job shop environments can even be captured in steady state simulations. Another explanation for the paradox can be found in not modelling the positive side effects that may result from the increased transparency of the shop floor situation and from postponement of release to the shop. Irastorza and Deane [1974] were probably the first to mention the indirect influence on logistic performance. For instance, with the postponement of release, orders might be cancelled before release. The resulting reduction of utilisation may in turn reduce the lead times. Nevertheless, it will be practically impossible to model all kinds of indirect relations, neither will it be wise, since it decreases the transparency of the simulations and clouds the direct Introduction 3

17 influences of the release methods on logistic performance. Still, the effects mentioned are important in practical implementations of controlled release. A final explanation could be that the simulated release methods are simply not good enough. It must be noticed that most simulation studies have tested release methods that do not have the properties of those used within the WLC concepts. Most simulated methods mainly restrict the quantity of work on the shop floor. Within the WLC concept, the procedure of fitting jobs into the workload norms for multiple stations will not only restrict but also balance the loads. This means that a more regular arrival pattern of work is created at the workstations. As we suggested in the beginning of this section, the tap of WLC contains a smart sifting mechanism to regulate the release of jobs, rather than being an ordinary tap that just opens and closes during certain period. The research in this thesis is triggered by the limited knowledge on how the tap of WLC affects performance. Moreover, past research has not fully answered the intriguing question whether a release method can control workloads and reduce them significantly without jeopardising other aspects of logistic performance and without being helped by any side effects. Although practitioners have apparently embraced the ideas of WLC, a better understanding may improve the practical use of this concept. 1.2 Objectives and research questions The preceding section showed an interesting research domain. Though several studies have researched the influence of controlled release methods in job shops, there is a clear need for research on the more particular release methods of the WLC concept. The approach behind these methods uses a control loop not found in any other method. That is, the release method is used to control throughput times and in turn these throughput times are used to determine accurate planned release dates for this method. As we consider the release methods of the WLC concept as the more sophisticated within the field, knowledge of these methods may provide a good starting point for developing improved methods. Considering this, we will pursue two objectives within this thesis: 1) To deepen knowledge on the functioning of release methods within the WLC concept; 2) To improve release methods within the WLC concept. The first objective can be seen as a prerequisite for the second. It is also an objective in itself, from which this research derives its theoretical relevance. This 4 Chapter 1

18 thesis must open the black box of mechanisms between WLC release methods and their final influences on logistic performance. The first objective reflects the aim of grasping the tap. The second objective addresses practical relevance in the first place. Besides, improvement of methods will be based on hypotheses regarding the functioning of job release. By evaluating the hypotheses behind the improvement, the second objective is in turn supportive to the first. Three central questions will guide this research: 1) What is the influence of release methods within the WLC concept on logistic performance? 2) How do the methods affect performance? 3) How can the methods be improved? The questions depict the line of measuring performance, understanding it, and finally improving it. Notice the subtle difference between the first two questions. Many studies have shown what the effects of methods on performance are. Of course, this must be investigated for WLC release methods as well. But more important for scientific progress is the how-question. We aim at understanding the influences, not just measuring them. Only this can lead to constructive design of new methods to answer the third question. Improving will relate both to the direct influences of the methods on logistic performance and to the robustness of the methods, i.e. making them less sensitive to both internal and external factors. The research will be performed from what could be called a mechanics perspective. Though many intermediate and often less tangible factors may affect logistic performance in practice, we aim at understanding the direct influences of the release methods. This means that models will be kept as lean as possible to keep influences transparent and separated from side effects. The most important modelling decisions are summarised in figure 1.1. Small job shop consisting of a single department with a number of workstations No assembly operations Capacity of each workstation constrained by a single resource No flexibility of capacity, nor breakdowns or absence No transport and set-up times, unless included in processing times Material available when needed Process plans fixed and known upon job entry Figure 1.1: modelling decisions Introduction 5

19 1.3 Thesis outline The next chapter will briefly review the relevant job shop control issues and related literature. The chapter discusses the production control decisions following the flow of jobs through the system. Particular attention will be paid to the influence of decisions on performance objectives. Chapter 3 provides a detailed assessment of the WLC concept and a comparison of the release approaches within this concept. It details and extends the research published in [Land & Gaalman 1996a] and [Breithaupt et al. 2002] (included as appendix A and B, respectively) The resulting considerations trigger the detailed research questions (section 3.6) that drive the simulation study in chapter 4. Chapter 4 presents a simulation study to analyse the release methods of the WLC concept. It unravels the performance to provide the basic ideas for redesign. These ideas have been elaborated in two completely different redesign approaches presented in chapter 5. The redesigned methods have been published in two articles, included as chapter 6 and 7, respectively. These articles include compact earlier versions of the analysis in chapter 4 and a simulation study testing the performance of the redesigned method. Finally, chapter 8 evaluates both methods and presents the final conclusions. 6 Chapter 1

20 Chapter 2 Job shop control Chapter 2 discusses the elementary decisions in job shop control and reviews related research. As extensive surveys have been presented in the literature, the focus will be on those aspects that position this thesis within the field of job shop research. The previous chapter briefly typified job shop production. The jobs differ with respect to the set of operations to be performed, with respect to the sequence in which the operations must be performed, and with respect to processing times. As a consequence of this variability, a number of different jobs may compete for the capacity of a workstation at any time. Matching these time-phased capacity requirements with the available capacities is the crucial task of job shop control. Although not considered in this thesis, the flexibility of available capacities, could also be used to match the requirements. The time at which each job will require the capacity of particular workstation can be influenced in different stages of job progress. We will follow the flow of a job in the stylised job shop with three stations of figure 2.1. This figure shows three important moments in the flow: entry, release, and dispatching. entry release dispatching shop floor Figure 2.1: Decision moments in the flow of a job An order quotation process will often precede the entry of a job. At its entry, the job is accepted and its due date is specified. The acceptance of a job results in a quantity of work to be done. The specification of the due date determines the amount of slack in doing the job. This slack is important in time-phasing the capacity requirements. It specifies the allowance for production planning and control to find a match between capacity requirements and available capacities, as the time a job can Job shop control 7

21 wait for the availability of capacity increases with a looser due date. Research on job acceptance and due date assignment is reviewed in section 2.2. Job release determines when a job enters the shop floor. The time between job entry and job release allows for preparation activities, such as process planning and checking the availability of material. Until release, a job will be just paperwork with no material attached to it. Should changes in job specifications be required, they can be made without wasting material and without disturbing the shop floor. We will suppose that, once released, a job will remain on the floor until all its operations have been completed. Therefore, job release is the last moment to create balanced capacity requirements on the floor. The release decision may withhold jobs from the floor to postpone the start of their first operation, and thus avoid excessive work-in-process. Accurately timing the release of jobs facilitates the final decision to select the right job for processing. Release research is briefly reviewed in section 2.3. Chapter 3 will deal more extensively with the role of release within the workload control concept. We will suppose that the final decision to select the next job for processing is made locally and on line at each workstation, which is called dispatching. The dispatching decision is generally based on some priority rule. The priority dispatching approach differs from the static scheduling approach, which centrally schedules the operations for all workstations, determining exact start and completion times beforehand. For reviews of job shop scheduling research, the reader is referred to [Blazewicz et al. 1996] and [Jain & Meeran 1999]. The dispatching decision finally fulfils the capacity requirement of one operation of a job. But, the influence of dispatching goes further. Selecting one job means that fulfilling the requirement of other waiting jobs is postponed. As each job has its own routing, the dispatching decision influences the availability of jobs at other stations. In the course of time hundreds of priority rules have been tested. Section 2.4 reviews the most important findings of this research area. Each decision will be made with the perspective to influence performance. To measure the performance, different criteria can be used. Traditional criteria are resource utilisation, inventory levels, lead times, and delivery reliability. The next section starts our review of job shop control with a concise discussion of these performance criteria in job shops. 2.1 Performance in job shops The production control decisions discussed in the introduction of this chapter traditionally pursue objectives such as high utilisation, low inventories, short lead times and high delivery reliability. These basic objectives have not changed during the last decades. However, the relative importance of the objectives has been subject 8 Chapter 2

22 to change [Wiendahl 1995] and new insights have led to the development of new performance indicators for each of them (see also [Land & Gaalman 1994]). The role of these objectives, see figure 2.2, and their interrelations in job shop production will be considered in the next subsections. These considerations have direct consequences for further modelling choices within this thesis, as each of the following subsections will show. high utilisation/throughput low inventories short lead times high delivery reliability Figure 2.2: Performance objectives Utilisation/throughput Originally, the objective of high resource utilisation was mainly driven by recovering investment costs. With the appearance of The Goal [Goldratt & Cox 1984] the drum-buffer-rope concept brought a new awareness of utilisation. It was emphasised that only bottlenecks should be exploited to increase the throughput and non-bottlenecks should be scheduled instrumentally. However, job shop production has to deal with shifting bottlenecks: several stations in a job shop can be potential bottlenecks. Idleness at any such station may cause backlogs to increase, which in turn will have its repercussions on future lead times. In most job shop models, also in this thesis, this is reflected in the use of equal utilisation levels for all stations. Some job shop studies [e.g. Park & Salegna 1995, Salegna & Park 1996, Enns & Costa 2002] have looked at the influences of relaxing this modelling assumption. Many job shops have moved from a situation where machine capacity is restrictive to a situation where worker capacity is restrictive. So, often the focus will be on worker utilisation rather than on machine utilisation. From a production control point of view there will be hardly any difference as long as a single resource restricts the capacity of each workstation. Dual resource constrained (DRC) shops receive increasing attention in literature as they give interesting opportunities for production control when multi-skilled workers can move between workstations. Bertrand and Wortmann [1981] already faced dual resource constraints in their seminal work on workload control and more recently Riezebos et al. [2003] showed interesting opportunities for workload control in a practical DRC situation. But since the basic Job shop control 9

23 theoretical development of workload control has not been crystallised that far, this thesis is confined to shops where workstations are constrained by a single resource. Within job shops utilisation levels below 80% for constraining resources are not uncommon in practice. It is generally impossible to reach utilisation levels close to 100%. Early queuing models of jobs shops already showed that lead times increase excessively at high utilisation levels as the consequence of irregular job arrival patterns, and of high routing and processing time variety. In job shop research, and more particularly simulation research, two approaches towards modelling utilisation/throughput can be distinguished. The first approach, typically modelling the job shop as a closed queuing network, uses the throughput of the system as an objective, while generally Work-In-Process (WIP) is kept at some fixed level. This approach requires the unrestricted availability of jobs to enter the system. The second approach, typically modelling the job shop as an open or semiopen queuing system, deals with the job arrival process as an exogenous variable. Since all of the arriving jobs have to be processed, utilisation levels can be determined beforehand, when routings, processing times and capacities are known. Where a logical objective for the closed system is to aim at high throughput, the objectives in the open system will include high throughput speed in terms of low throughput times. In the closed system WIP and related throughput times (see and chapter 3) are controlled, while in the open system the utilisation levels are the controlled variables. The latter approach is followed in this thesis Inventories Different types of inventory can be distinguished in production environments. Work-In-Process (WIP) inventories are the main concern in job shop production. Since production takes place on customer order, final good inventories are restricted to jobs that are completed ahead of their due date. It has been argued from Just-In-Time perspectives that early completions of jobs should be penalised in job shop research. Kanet and Christy [1984] were among the first to investigate the influence of penalised early completions on job shop control. Rohleder and Scudder [1993] found that particularly the release method gains importance when early completions are penalised. Inventories of raw material do exist in job shops, but these will generally not be very valuable. Often, basic raw material is used in many different products. In some situations the customers themselves might even supply job specific material. The objective of low WIP can be seen from the viewpoint of cost minimisation. Capital is tied up in inventories. Also here, the last decades have put other perspectives on the role of inventories, with Japanese production philosophies 10 Chapter 2

24 pointing at the fact that inventories may hide the obstacles to effective production control. WIP inventories typically reduce the shop floor transparency and as a consequence critical jobs may get lost in inventories, tracing specific jobs may require large efforts, etc. Job shops tend to have considerable levels of WIP. The high WIP levels result from the jobs that have to queue for each operation, waiting for the availability of the workstation. Due to the complexity to match the capacity requirements with available capacities, the queues of jobs tend to pile up on the floor. It may therefore be clear that WIP levels and lead times have a strong relationship. This relationship plays an important role within workload control concepts, so chapter 3 will discuss this relationship in more detail. Job shop research has paid only little attention to the modelling of WIP inventories. Often, the number of jobs on the floor is used as a simple indicator of WIP. But many simulation studies do not even measure WIP. In studies where jobs are immediately released to the floor, WIP levels may be derived from the average lead time using Little s result. Chapter 3 will elaborate on the measurement of workin-process in this thesis, since it plays an important role within the WLC concept Lead times There is a lot of confusion on the meaning of terms related to lead time, as Plossl [1988] already concluded. The term lead time itself sometimes refers to the planned time to complete a job and sometimes to the realised time to complete (part of) a job. We will generally speak of planned and realised throughput times. We define the throughput time of a job for a certain system as the time between its arrival at the system and the time it leaves the system. The (sub-)systems within the job shop will be distinguished in section 3.1. For instance, the shop floor throughput time of a job is defined as the time from its release to the shop floor until the completion of its final operation. The throughput time of a job can in turn be decomposed into different elements: processing time (including set-up and run time), transport time and waiting time. As said before, throughput times relate strongly to work-in-process levels. With the high work-in-process levels in job shops, shop floor throughput times tend to be high as well. When release to the shop floor is controlled, waiting time on the shop floor can be replaced by waiting time before release. Still, total throughput times will be relatively high. Often more than 90% of a throughput time consists of waiting time. This indicates the complexity to match the time phased capacity requirements with the available capacity, which is one of the main reasons for the occurrence of Job shop control 11

25 waiting times. Chapter 3 will elaborate on the measurement of throughput times in this thesis and their relation to WIP Delivery reliability The issue of delivery reliability strongly differs between make-to-stock and make-to-order environments. In make-to-stock environments finished good inventories must guarantee delivery reliability, while delivery reliability in make-toorder environments relates to the match between promised delivery time and the realised throughput times. The difference is indicated as the due date deviation or lateness of a job. In job shops the delivery reliability performance must result from a combination of (1) well estimating the throughput times for a job when promising a delivery time to the customer, and (2) controlling job progress such that the promised delivery time is met. Since both are complex tasks considering the high variability of arrivals, routings and processing times, job shops are typically not characterised by high delivery reliability. Within job shop research delivery reliability is always indicated by some function of the due date deviations of delivered jobs. The percentage tardy, i.e. the percentage of jobs with a positive lateness is one of the more traditional indicators. However, this indicator gives no impression of the amounts of lateness. To give a more detailed picture of delivery reliability, a combination of first and second moments (average and variability indicators) of the distribution of lateness among jobs can be included. In this research the average lateness and standard deviation of lateness will be measured. Correspondingly, two general approaches can be discerned [Baker 1974], which contribute to delivery reliability after the due dates have been fixed: (a) speeding up throughput, reducing the average throughput time and thus the average lateness of jobs, and (b) keeping individual jobs on schedule, reducing the dispersion, i.e. the standard deviation, of lateness across jobs. Both approaches may reduce the percentage tardy, as can be seen in figure Chapter 2

26 freq. freq. freq. 0 lateness speeding up througphut 0 lateness reducing the dispersion of lateness 0 lateness Figure 2.3: Two approaches to reduce the percentage tardy 2.2 Job acceptance and due date assignment At its entry, the job is accepted and its due date is specified. As suggested in the introduction of this chapter, the acceptance of a job is generally not simply a matter of saying yes or no, but it will be preceded by an order quotation process. As such, it will be the result of a bidding strategy and the consequential reaction of the customer. A structured review of bidding research is given by Easton and Moodie [1999]. Within the concept of WLC, job acceptance provides the first and most powerful opportunity to control the input of work in a job shop, so acceptance methods have been studied specifically [e.g. Hendry & Kingsman 1993]. For research on release methods, Melnyk et al. [1991, 1992, 1994] have suggested to model the result of the Job shop control 13

27 acceptance decision by some kind of load smoothing, while others [e.g. Philipoom and Fry, 1992] have modelled a situation where specific jobs are accepted and others refused based on the internal shop situation. Within this thesis the result of the acceptance decision will be simply modelled as a stationary arrival process of accepted jobs, considering the fact that many job shops in practice generally say yes to each job, and because of the focus on the release decision. The due date j of job j can be written as its entry time t E j plus an allowance j, that is j = t E j + j. Generally, the allowance results from a negotiation process with the customer and becomes fixed at the entry time. Job shop research commonly uses one of two extremes: due date allowances are either externally imposed or internally set. In the former case the due date allowance j will be modelled as an exogenous variable, in the latter case the assignment of a value is part of the production control policy and as such modelled as an endogenous variable. Table 2.1 overviews the possible components of the due date allowance, which are briefly discussed below. Baker and Bertrand [1981], Baker [1984], Ragatz and Mabert [1984], and Cheng and Gupta [1989] give more extensive reviews of due date assignment research. Category Exogenous Endogenous, job related Endogenous, shop congestion related Component (related simple due date rule) Constant (CON) Random (RAN) Processing times (TWK) Routing (NOP) Number of jobs in the shop (JIS) Nr of jobs in queues on routing (JIQ) Processing times on routing Table 2.1: Due date allowance components Typical examples of exogenous allowances are constant allowances (CON) and random allowances (RAN). The first case may represents a policy where uniform delivery times are quoted by management, the second case may model the situation where the customer establishes the due date. Internally set allowances can be seen as throughput time estimates [Vig and Dooley 1993]. These estimates may include information on job characteristics and/or on shop congestion at the time of a job s entry. The Total Work (TWK) rule estimates the throughput time of a job as a multiple of its total processing time, the Number Of Operations (NOP) rule estimates it as a multiple of the number of operations of the job. 14 Chapter 2

28 Estimates of shop congestion can be based on workload information at different levels of aggregation. In its most basic form the estimate can be a multiple of the number of jobs in the shop (JIS) at the time of arrival. More detailed estimates use the routing of the job and the number of jobs in the queues (JIQ) to be visited. Loads can be further detailed by measuring them in terms of processing times [e.g. Bertrand 1983, Enns 1992, Enns 1994]. All kinds of combinations of the above elements can be used in an estimate of a job s throughput time. Additionally, the estimated throughput time may include some constant or static component based on the idea that a shop returns to a steady state [Vig & Dooley 1993]. Due date assignment models may show strong interactions with other production control policies: due dates based on processing times positively interact with SPT sequencing while the NOP rule positively interacts with FCFS sequencing. Comparative studies that have the objective to compare different production control policies should therefore carefully choose a due date assignment method. In this thesis, due dates are modelled as exogenous variables, determined independent from job or shop characteristics, in order to isolate the influence of the release decision. 2.3 Job release This section will give a brief overview of job release methods. A more extensive discussion of release methods within the WLC concept is given in chapter 3. The release decision, sometimes referred to as order review and release (ORR) [Melnyk & Ragatz 1988,1989], is the instrument to control the input of jobs to the shop floor. Since Wight [1970] recognised the importance of input/output control, job shop research has increasingly paid attention to the release method. Wisner [1995] and Bergamaschi et al. [1997] present extensive reviews of release research. Wisner gives a classification based on research characteristics, Bergamaschi et al. focus on the inherent characteristics of release methods. Infinite loading methods Uncontrolled (no particular influence) - immediate/periodic release Controlled (reducing lateness dispersion): - backward infinite loading (BIL) - modified infinite loading (MIL) Finite loading methods Load limiting (reducing WIP): - Maximum number of jobs on floor Load limiting and balancing (reducing WIP + controlling throughput): - forward finite loading (FFL) - classical WLC release methods Table 2.2: Release methods categorised with some typical examples Job shop control 15

29 Table 2.2 distinguishes between release methods based on finite loading and methods based on infinite loading. Finite loading methods use restrictions on the quantity of work that can be released, whereas infinite loading methods do not and thus they assume infinite capacity. Within each of these classes we further categorise release methods according to their intended influence (between brackets) on logistic performance and give some typical examples Infinite loading methods The first subclass of uncontrolled infinite loading methods hardly deserve the indication method. Uncontrolled release can take place on a continuous or periodic basis, but leaves control of the jobs to the dispatching decisions. Controlled release methods based on infinite loading typically estimate the required shop floor throughput time to complete the job, and try to release each job at the right time relative to its due date. Thus, they aim at reducing the dispersion of lateness. Contrary to due date assignment methods the release methods take the due date j of a job j as given and subtract a shop floor throughput time allowance β j to determine the release time t R j for the job: t R j = j - β j. The allowance β j may have the same type of components as discussed in section 2.2 for the due date allowance j (see table 2.1). It can be based on job characteristics and on the shop floor situation at the time of estimating. For instance, the Backward Infinite Loading Method (BIL) uses job characteristics to determine the release date; the Modified Infinite Loading method (MIL) additionally includes estimates of loads. Accurately estimated release times for jobs avoid that early jobs compete for the capacity of workstations with late jobs. The study of Rohleder and Scudder [1993] shows that controlled release methods based on infinite loading may particularly contribute to performance when early completions are penalised. Though controlled methods of infinite loading may consider the loads on the floor, they do not limit these loads. Such methods may cause vicious cycles when increasing loads on the floor lead to earlier releases, which further increases the loads on the floor. With infinite loading the jobs still compete for capacity on the floor. With finite loading methods, shop management is forced to consider capacity conflicts before jobs are released to the floor, as loads on the floor are strictly limited Finite loading methods Within the class of finite loading methods we distinguish two main subclasses: load-limiting methods, mainly contributing to reduced work-in-process, and methods that additionally balance the loads both across stations and over time (Shimoyashiro 16 Chapter 2

30 et al. [1984]) in order to improve or maintain throughput. However, there is not always a rigid line between limiting and balancing loads. The simplest pure load limiting release method just restricts the number of jobs on the shop floor to a certain maximum. In semi-open queuing models of a job shop, Kanet [1988] indicated that such methods would increase the average throughput time as waiting times before release would normally exceed the reduction of waiting times on the shop floor. This article has been one of the triggers of the discussion on the paradox mentioned in section 1.1. A method with strong load balancing properties will sequence jobs for release in such a manner that those jobs filling gaps in the workload of a station are prioritised. The resulting more regular arrival process at each station reduces average throughput times, comparable with the functioning of work-in-next-queue dispatching, which will be discussed in the next section. Irastorza and Deane [1974] where probably the first to develop a typical balancing method. Another strong focus on balancing can be found in the theoretical methods developed by Shimoyashiro et al. [1984] and Wein and Chevalier [1992]. The latter method is based on research in semi-conductor industries. A complete review of release methods (generally load balancing methods) developed for the specific characteristics of this industry can be found in [Fowler et al. 2002]. Commonly used finite loading methods fitting jobs in a time-phased projection of capacity, such a Forward Finite Loading (FFL), can also be categorised as load limiting and balancing. An important group of release methods in this class have been developed as part of what we indicated as the WLC concept. These methods periodically release a set of orders and try to keep workload within certain norm levels. Workloads are measured in units of processing time, and norms can be specified for each station. The WLC methods balance the workload across time as will be explained in Chapter 3. Besides, the WLC methods sequence the jobs for release according to planned release times, which should contribute to a reduction of lateness variance. Comprehensive descriptions of the classical WLC methods can be found in [Bechte 1988, Hendry & Kingsman 1991]. Recent extensions and improvements with a focus on improved balancing have been developed and tested by Cigolini, Perona and Portioli [Perona & Portioli 1996, Cigolini & Portioli 2002]. Also this thesis has the aim to extend and improve the classical WLC methods. Chapter 3 will discuss the WLC release methods in detail. Job shop control 17