Below is the uncorrected machine-read text of this chapter, intended to provide our own search engines and external engines with highly rich, chapter-representative searchable text of each book. Because it is UNCORRECTED material, please consider the following text as a useful but insufficient proxy for the authoritative book pages.
A NEW PERSPECTIVE ON MANUFACTURING SYSTEMS ANALYSIS RAJAN SURI ABSTRACT Most U.S. corporations are devoting tremendous efforts to revitalizing their manufacturing base. As a result of the enormous effort in planning, designing, evaluating, and operating this vast array of plants, manufacturing systems modeling has become a vital activity. Coincident with this major change in facilities, the manufactur- ing field has itself been undergoing radical changes. On the one hand are the high-tech developments enabling automation and integration, and on the other are philosophical developments such as just-in-time. Much of the "traditional" analysis activity has not been directly relevant to these changes, and the major recent thrusts in this area have come from outside the traditional community. The role of analysis in modern manufac- turing is discussed, research directions are proposed, and steps to be taken by the individ- ual and the research community as a whole are suggested. INTRODUCTION A tremendous effort is being devoted to restructuring the manufacturing base of most industrial corporations. In the wake of the resultant planning, design, evalua- tion, and operation of vast numbers of plants, the analysis and modeling of manu- facturing systems has become a vital activ- ity. But as this interest in manufacturing analysis has grown, those of us who have concentrated on "traditional" analysis ap- proaches must ask whether we are in the mainstream. This paper looks at some recent develop- ments in this field and contrasts them with traditional approaches. It presents the hy- pothesis that the recent principal contribu- tions in manufacturing analysis have come from outside the traditional communities. It proposes a framework for viewing the ~Z~ issues in manufacturing analysis and sug- gests some research opportunities. It pro- poses a novel concept, design for analysis, for treating this area. It concludes with re- search suggestions for individuals who wish to make useful contributions in this area. Throughout this paper, the term "analy- sis" is used generically to include both anal- ysis and modeling, and the term "tradi- tional" analysis methods means analysis and modeling approaches that were tradition- ally found in the publications of societies such as APICS, IEEE, IIE, ORSA, SME, and TIMS up through the early 1980s. Ex- amples of such publications are IIE Trans- actions, Management Science, and Opera- t?ons Research. In addition, the adjective "analytic" is used exclusively to describe techniques that rely on solving equations, while the term "analysis" is used to describe broadly any technique for systematically
A NEW PERSPECTI VE ON MANUFACTURING SYSTEMS ANALYSIS studying a problem. Thus, in this paper, simulation would be an analysis technique, not an analytic technique. "TRADITIONAL" MOTIVATION FOR MANUFACTURING ANALYSIS We begin by reviewing some of the tra- ditional arguments for conducting manu- facturing systems analysis. Manufacturing systems analysis and performance evalua- tion is not a new field, as any industrial engineer will testify. Throughout the life- time of a typical system, the organization responsible for it goes through many phases of decision making, from an analysis of ini- tial feasibility through design, operation, and finally obsolescence. Typical decisions that must be made in a manufacturing en- vironment and typical performance mea- sures used to evaluate these decisions are shown in Table 1. A modern manufacturing system, how- ever, can be quite different from the more traditional system. A modern manufactur- ing unit is most often a complex system, consisting of many interconnected compo- nents of hardware and software. Decision making, therefore, can become difficult be- cause of the greater complexity of the mod- ern system compared to, say, a conven- ~9 tional job shop. This complexity is due to several factors: · Highly interconnected components leading to a very large set of decisions that must be made simultaneously. In the mod- ern system, where both material and infor- mation move rapidly through the plant, a small change at one end of the plant can have, in a few minutes, a significant impact on an area at the other end of the plant. · Limited resources due to efficiency re- quirements. The sharing of resources as a means of reducing costs increases the com- plexity of managing that sharing and of predicting simultaneous demands for shared resources. · Little "slack" in the system. A com- monly quoted statistic for batch manufac- turing typical of job shops is that machine tools spend 5 percent of their time per- forming value-added tasks e.g., cutting metalso these conventional systems have a lot of "slack." In an automated system, such as a flexible manufacturing system (FMS), this number can be 10 times higher or more, leaving little slack (U.S. Congress, Office of Technology Assessment, 1984~. · Fewer "humans in the loop." Human operators can use common sense to correct or modify situations caused by unexpected changes in conditions. In a highly auto- TAB~E 1 Typical Performance-Related Decisions During the Design and Operation of a Manufacturing System Typical Decisions Typical Measures of Performance Number and types of machines Number of load/unload stations Part-types Alternative routings Tool allocation Number and types of fixtures Number of transporters and pallets System layout Buffer sizes Period of payback Return on investment Net present value Facility utilization Production rates Work in progress Part and material flow times System flexibility Queues at each resource Operating policies
120 mated environment, these unexpected per- turbations must have been anticipated and provided for or the control system will not be able to properly react. Disaster can re- sult from inadequate software controls. Because of all of these factors, even ex- perienced shop floor supervisors and man- agers have difficulty in perceiving all the consequences of any given action in the modern manufacturing context. Also, be- cause of such factors, we found that, in the words of an experienced FMS user, "People thought they were getting into FMS; in- stead they got into FM-Mess!" a. Schnur, personal communication, 1985~. These factors have also created a genuine dilemma for the designer of manufacturing systems. On the one hand we have complex systems involving high risk and high capi- tal, yet on the other we have the require- ment to design and operate these systems efficiently to achieve the corporate objec- tives of competing effectively. It is precisely to help resolve this dilemma that the tradi- tional analysis community entered the manufacturing arena. A typical reaction, including that of the author, was that the availability of tremendous computing power made it possible to build large mod- els to analyze the system and aid in the decision making. Researchers in this com- munity have been publishing increasingly complex analyses that had "application" to manufacturing. The journals are currently filled with articles that address complex modeling issues for such things as FMS, ro- botic cells, and automated storage and re- trieval systems (AS/RS). But what should be the priority afforded all of this research effort that is devoted to modeling and anal- ysis? It will be informative to see what the competition has been doing in these areas. THE "JAPANESE WAY" A DIFFERENT APPROACH The Japanese approach to manufactur- ing has fundamentally affected the way in- RAJAN SURI dustry views this field (Hall, 1983, Schon- berger, 1982~. There has been a correspond- ing effect on manufacturing analysis. It is helpful to contrast the approaches of most American manufacturers and the Japanese manufacturers to three common features of many systems. Complexity The American approach to dealing with large complex manufacturing operations has been to adopt material requirements planning (MRP) and manufacturing re- source planning (MRP-II). A little intro- spection leads us to realize that these are just large models, with various assumptions of lead times, lot sizes, demand, etc. The Japanese approach is to simplify the opera- tion rather than to attempt to deal with the complexity. Examples of this are the "flow line," which simplifies the product flow, and the kanban system of job tickets, which simplifies shop floor scheduling and control. Uncertainty The American approach to countering uncertainty in the manufacturing environ- ment has been to use buffer stocks and lead times with imbedded safety margins. These safety stocks or safety times have been de- rived by means of modeling techniques (e.g., economic order quantity, or EOQ). The Japanese way of tackling uncertainty is to get rid of it: the just-in-time (JIT) phi- losophy espoused in numerous recent arti- cles has, as one of its goals, the systematic reduction of uncertainty in all aspects of the operation. Constraints Manufacturing managers are constantly dealing with constraints constraints on capacity, on tooling, on precedence of op- erations, etc. The American way of dealing
A NEW PERSPECTIVE ON MANUFACTURING SYSTEMS ANALYSIS with these constraints has been to develop more and more sophisticated models that attempt to optimize schedules within nu- merous complex constraints. Once again, the Japanese have come forth with a radical approach to constraints break them! A1- though this may sound frivolous, there are many instances of success. Proponents of the Japanese approach are eager to cite the case of setups at large stamping presses in the automobile industry. Typically, Amer- ican manufacturers would take around 10 hours to change the setup for a different component. To minimize the number of tooling changes, large lot sizes were run. This constrained the schedules of preceding and following operations and also involved optimization to find the best inventory lev- els for the system as a whole. Finding the best lot sizes was a complex problem that had to be solved each month, to say nothing of the system-wide effects that resulted from any changes in schedule. The Japanese tackled the problem differ- ently. Studying the setup procedure itself, they found engineering and equipment so- lutions to shorten the procedure, eventually achieving tooling changeovers that con- sumed only a few minutes. Since the amount of time required to change a die was now negligible, it no longer mattered what sequence or lot sizes were needed and the setup and lot size constraint simply dis- appeared. The point of this example is to highlight the differences in the two approaches, as summarized in Table 2. Essentially, the American approach has involved taking 121 various problems as given and developing models to deal with those givers. The Jap- anese approach has been to try to eliminate the problems. In the light of the success of the Japanese techniques, and the rate at which American industry is attempting to adopt them, does this mean we can do away with traditional analysis of manufacturing systems? We will return to this question later. OTHER "NONTRADITIONAL ANALYSIS" SUCCESS STORIES Recently, there have been three other highly visible thrusts in the area of manu- facturing systems analysis. Artificial Intelligence and Expert Systems Artificial intelligence (AI)-based ap- proaches, particularly expert systems (Fox and Smith, 1984), have been receiving much enthusiastic attention from industry. Indeed, one gets the distinct impression that an AI-based project would be more likely to receive funding or contracts than one based on traditional approaches. Optimized Production Technology Optimized production technology (OPT) is a system that has quickly gained a lot of visibility in the area of manufacturing sys- tems analysis. OPT is both a software sys- tem and a philosophy (Lundrigan, 1986, Meleton, 1986~. The software system has TABLE 2 Two Cultures of Manufacturing Systems Analysis Approach Problem The American Way The Japanese Way Complexity ModelsMRP and MRP-II Simplify (e.g., flow lines) Uncertainty Models to derive lead times Get rid of it (e.g., JIT) and buffer stocks Constraints Models to optimize Break them (e.g., setups)
122 the ability to schedule very large factories. To use the software one has to adopt the OPT philosophy, much of which is reason- able and simple. Although one might de- bate some of the ideas in OPT, one fact seems clear in a short period of time, this system has shown that there exists a tremen- dous market opportunity for effective anal- ysis and scheduling software. A typical OPT installation may require investments ex- ceeding a million dollars. Animation Computer simulation with graphic ani- mation is one of the increasingly popular tools used in industry for analyzing manu- facturing systems. It is widely accepted that the animation component helps to convince both manufacturing engineers and senior management of the benefits of simulation. The popularity of this approach can once again be judged by the growing number of commercial software systems available for the task (Haider and Banks, 1986~. A1- though animation is having the positive ef- fect of getting simulation to be used, it is unfortunately also having a negative affect. It is becoming a substitute for simple anal- ysis. In fact, one sees conference presenta- tions where traditional analysis is replaced by poorly analyzed, but very pretty, graphic animation studies. One hears of major projects being funded on the basis of such studies, where the basic assumptions about the data and approach were flawed but the animation sold the project. Summing up the New Developments Four of the most recent and visible de- velopments in manufacturing analysis are the following: The Japanese approaches AI/expert systems · Optimized production technology · Graphic animation RAJAN SURI These are prominent recent approaches in the sense that they appear to be receiving more attention, publicity, and money than any other analysis techniques. The surpris- ing, indeed shocking, fact is that none of these approaches reflects developments from the traditional analytic community. One can raise many questions concerning this observation. Is this an indication of the inability of the traditional approaches to be implemented effectively? Is it an indication of an inability to sell concepts effectively? Or, should we, the traditional experts, give up working in this application area? A1- though specific answers to each of these questions will not be proposed, the balance of this paper attempts to identify new direc- tions for the traditional analyst that offer the possibility of reversing this trend. At least, it is hoped that raising some of these questions will lead to further introspection about where this community, the tradi- tional analysis experts, is going, where it needs to go, and how it should direct its efforts in getting there. ROLE OF ANALYSIS IN TODAY'S CONTEXT Recent Developments There is a role for traditional analysis techniques in the context of new develop- ments. As an example, let us consider the Japanese approach of just-in-time. · A point often lost in the publicity sur- rounding these methods is that the Japanese approach is an ideal. Although one must constantly strive to achieve it, we will never live in a perfect world such as one with no uncertainty. Real-world companies face constant change in demand, in technol- ogy, in people so one must continue to pursue improvements. · No corporation can implement the Japanese approach overnight they need to "get there from here." · It is recognized that the Japanese ap-
A NEW PERSPECTIVE ON MANUFACTURING SYSTEMS ANALYSIS proach may not be suitable for certain products and certain distribution of facili- ties and suppliers. Thus, some facilities will continue to be run in the more traditional "American style." Now reconsider these points, but in a log- ical order, starting from the final point above. Is the Japanese approach suitable for a given set of products? Traditional analysis methods can help answer this ques- tion. How do we get there from here, given that we would like to maintain a reasonable delivery schedule with our customers even as we implement JIT? Again, traditional analysis can help us anticipate some of the problems before they occur and can aid in maintaining deliveries. To use the familiar Japanese analogy, just as the JIT approach lowers the "water level" (the inventory) to expose the rocks (the problems), so analysis can act as a "sonar" to alert one to distant rocks. As the real world changes around us, various types of traditional models can serve as decision aids to understand the impact of alternative choices and varying changes in the environment, even while JIT is being used. In a similar way, traditional analysis methods can be used to complement or en- hance other recent analysis approaches such as expert systems. The challenge is to show that the traditional analysis community can successfully learn what is needed and pro- duce effective implementations and success stories similar to the four described in pre- · , . VlOUS SeC[lOllS. Keys to Effective Analysis A key to understanding the current trends in the general area of manufacturing is to recognize that manufacturing is now re- garded as a major weapon in a company's strategic arsenal (Hayes and Wheelwright, 1984; Skinner, 1985~. This must be con- stantly recognized during the analysis pro- cess. The manufacturing system analysis 123 process, as shown in Figure 1, must be driven by management's strategic objec- tives. Thus, one of the keys to effective analysis is a recognition of the drivers of this process. A second ltey to effective man- ufacturing system analysis can be appropri- ately stated by paraphrasing the just-in-time statement: Use the right model at the right time to answer the right q~st?~n. These two key points of effective manufacturing systems analysis have often been forgotten in the voluminous "traditional" publica- tions of industrial engineering, manage- ment science, and operations research. Even if one agrees, in principle, with the foregoing statements, the question arises as to how to identify the right questions and models. At a given point in time, what should one model? What decisions and pa- rameters should be included? What per- formance criteria should be evaluated? We propose that a systematic structure for answering these questions is obtained by looking at the "manufacturing system life cycle." LIFE CYCLE PHASES, ANALYSIS, AND RELATED RESEARCH ISSUES Manufacturing System Life Cycle The life cycle of an item is defined as the period from the initiation of the concept to its obsolescence. Both products and systems have life cycles. While the product life cycle is a widely stuclied concept, the life cycle of modern manufacturing systems is not. With the life cycle of modern products getting Drivers (Management strategy and objectives) l~lanagement Declalons Performance analyale process | ~ Recommendation FIGURE 1 Representation of the manufacturing system analysis process.
124 RAJAN SURI shorter e.g., two to three years for some · The identification of candidate equip- electronics products there is a major mo- ment choices for each alternative. tivation to create manufacturing systems with a life cycle that spans several product life cycles. Although many factors determine the life cycle of a manufacturing system, the em- phasis in the following discussion is on those aspects that are most amenable to analysis. Even though strategic analysis is a critical first step in the determination of business goals, markets, and products, this must be accomplished before any substantial con- ceptualization of the manufacturing sys- tem. We begin the following discussion with the feasibility analysis stage. Each of the following sections discusses some of the crit- ical stages in the manufacturing system life cycle, identifies appropriate analysis needs, and suggests selected research issues. Feasibility Analysis (Planning) Phase The objectives of the feasibility analysis phase are to establish the economic attrac- tiveness of various system alternatives that are candidates for the production of one or more of the stated products. It will be nec- essary to obtain initial estimates of various system measures, as described later, for each of these alternatives. Typical issues that must be addressed during this phase are as follows: · An understanding of the organiza- tional objectives that this system is intended to meet. · A determination of the geographic lo- cation of the manufacturing facility. · A determination of the products, sub- assemblies, and components that are to be made or bought. · An understanding of the impact of al- ternative product designs on the manufac- turing process and system configuration. · An understanding of the implications of alternative process approaches, such as "FMS" versus "flow line." · A decision concerning the degree of "flexibility" that the system should have. · An understanding of the constraints that will be imposed on the issues by differ- ent levels of capital investment. Several measures of performance must be applied to each of the system alternatives that are developed in this analysis. Typical measures of performance for the capital that will be invested are the return on the investment (ROI), the net present value (NPV), and the payback period. A measure of the strategic return will involve an assess- ment of the impact on quality, price, and responsiveness to the marketplace. The ability to respond to product changes, to demand surges in the marketplace, and to operational uncertainties must be esti- mated. Typical analysis techniques that are ap- propriate for this phase are forecasting, de- cision analysis, location analysis, economic analysis, mathematical programming (e. g., linear or integer programming and non- linear programming), facility layout, and group technology. Detailed references on these and a few other appropriate tech- niques are available (Surf, 1985~. Although it appears that traditional analysis methods are applicable for this stage of analysis, the literature and methodology are deficient in essential areas that are relevant to current manufacturing practice. Large, complex mathematical programs to study optimal location of a facility exist, but there is little available that can be used to characterize the impact of strategic alternatives on qual- ity and responsiveness. The following list of research issues is not exhaustive, but it is indicative of the prob- lems that remain. An additional recent per- spective on this topic can be found in Gershwin et al. (1986~. It will be obvious that the problems identified can be success- fully addressed only with new tools. It is
A NE W PERSPE C TI VE ON MANUFAC TURING SYS TEMS ANALYSIS the development of these tools that will form the basis for the research tasks. · Better decision models for evaluation of investments in rapidly evolving technol- ogy. Companies in the semiconductor in- dustry regularly confront this problem. They must make decisions concerning new factories that may not be in full production for four years, which means making deci- sions well before the time that the products or processes have been fully specified. · Development of integrated approaches to product, process, facility, and equip- ment decisions. As elaborated by Whitney et al. (1988, in this volume), the benefits of integrating these decisions can be substan- tial. The research challenge here is to do this more efficiently than is possible with large and complex computer models. · Improved quantitative understanding of the benefits of properties such as respon- siveness, product quality, and system flexi- bility and how these properties affect mar- ket share and profitability. · Improved frameworks for quantifying the benefits of investments in flexible facil- ities and for assessing the appropriateness of various types of flexibility. · Provide answers to questions such as 'Should this plant implement JIT?" and, if so, "Where should inventory be reduced, by how much, and when?" A gradual in- troduction of JIT is sometimes desirable. An understanding of alternative programs of implementation is needed (Surf and DeTreville, 1986~. Aggregate Analysis Phase The aggregate analysis phase arises after the feasibility analysis has shown that the project is worth pursuing and management has given the go-ahead for further design and analysis. The objectives of this phase are to further evaluate the systems selected in the previous phase, to design a rough configuration for each candidate system, to I25 reduce the number of alternative systems (typically to one or two), and to verify that the assumptions made in the feasibility analysis phase are still appropriate in the light of this additional analysis. Typical is- sues to be addressed during this phase are a determination of the processing capacities that will be needed at each stage of the operation; the effect of various parameters such as equipment reliability, process yield, and product volume and mix; the impact of alternative lot size choices on the processing parameters; the approximate requirements for material handling and storage, and the requirements for computer hardware and software. Various measures of performance are ap- propriate for this stage of analysis. These include refinement in the values of all the measures used in the earlier feasibility stage as well as measures of the product cycle times, the work in process (WIP) inventory, the size of queues, the level of equipment use that can be expected (including the ex- pected downtimes), the performance of the computer hardware and software and the expected production rates that can be achieved with the assumed production con- figuration. At this stage the analysis should be done at a "high" level, with many details still being approximated or aggregated. For example, it would be appropriate to con- sider material handling and storage equip- ment at an aggregate level and to delay any detailed scheduling and sequencing consid- erations. Various aspects of aggregate anal- ysis have been discussed elsewhere (Surf, 1985; Suri and DiehI, 1987~. The effective use of the aggregate analysis approach for a modern factory was described by Haider et al. (1986~. Typical analysis techniques that will be found useful in the aggregate analysis phase are strategic and economic analysis, such as decision-tree analysis and discounted cash flows, the use of queueing analysis and queueing network models, simulation mod- els that can treat aggregated systems (Surf
126 and Diehl, 1987), models for the reliability of machines and processes, models that pro- ject the yield from individual processes, and models that provide a measure for com- puter systems performance. Selected research issues that relate to ag- gregate analysis are these: · Development of improved analytical models e. g., queueing models and reli- ability models that can treat situations at an appropriate level of aggregation. The aim of such models should be to perform easy and quick high-level analysis of system alternatives. This approach implies that the models should be simple even at the expense of precision. This contrasts strongly with the direction of much of the research effort devoted to developing intricate refinements that make queueing models more accu- rate e.g., reducing the error from 15 per- cent to 2 percent. Given the role suggested here for such models, an accuracy of 15 percent is acceptable. The simplicity and improved speed of solution that can result from the less detailed models can be signif- icant advantages. Thus, in contrast to con- tinuing the search for further refinements, we pose the following key problems in this area: (a) the ability to model "blocking," such as occurs when limited buffers are available or when there is an attempt to simultaneously share a resource; (b) the de- velopment of a better understanding of the systems and conditions under which such models work well and the conditions under which they should not be used; (c) the de- velopment of improved analytical models for transient analysis e.g., start-up after a failure; and (d) the development of more validation studies based on the use of these models for real manufacturing systems. · Models that will provide an improved understanding of the aggregate dynamics of kanban systems. · An improved understanding of the overall impact on quality and yield through integrated models of equipment perfor- RAJAN SURI mance, system dynamics, and economics. · Improved algorithms for optimization of simulations. Detailed Analysis Phase Earlier analysis phases will have nar- rowed the candidate systems to one or two choices. The objectives of this phase are to establish, in as much detail as possible, how the candidate systems) will function, in- cluding how the system will operate in con- cert with the rest of the factory and the rest of the organization. It will be necessary to verify that assumptions made during the earlier analysis phases remain valid in the light of the increasing level of detail that is now available. The most desirable system must now be selected, along with decisions on all of the relevant parameters. Typical issues to be addressed in this phase are the detailed configuration for all equipment, including tooling and fixturing require- ments; the explicit details and characteris- tics of the material handling and storage system; an understanding of the "links" to the rest of the organization (e.g., suppliers, MRP system, shipping, and other logistics); the mechanism for translation of corporate requirements into individual tasks for equipment in this system; the determina- tion of effective planning and scheduling policies; the determination of the system response to short-term problems such as failures, blockages, and late arrival of ma- terial; the determination of the manage- ment structure and labor structure; and the definition of functional requirements for all computer hardware and software elements. Performance measures that are appropri- ate for this phase include those used in pre- vious phases plus the utilization levels and response times for the material handling and storage subsystem, the utilization levels of the tooling and fixturing, the frequency of blocking and resource contentions, the response times-for the computer hardware
A NEW PERSPECTIVE ON MANUFACTURING SYSTEMS ANALYSIS and software subsystems, and the reliability and responsiveness of the overall system to problems. Although the typical analysis techniques used in this phase include all those used in previous phases, one finds in practice that the technique most used here is discrete-event simulation. Additional techniques that are found to be useful, however, are production planning, lot siz- ing, scheduling and sequencing, real-time control, and structured system analysis and software design. The extensive literature on mathematical programming methods for lot sizing and scheduling is simply not appropriate in the current context of simplified cellular man- ufacturing and JIT operation. Although simulation is widely used by industry, few practical tools exist for the statistical analy- sis of simulation output, and almost none exist for the optimization of simulation re- sults. Hence, in the framework suggested in this paper, selected research issues relevant to the detail analysis phase are these: · Development of a theory of discrete event dynamic systems (DEDS). A much better qualitative and quantitative under- standing of the detailed operation of man- ufacturing systems should result from a good theoretical foundation for DEDS. There has been some progress in this area, such as developing a linear system represen- tation (Cohen et al., 1985), formalizing no- tions of observability and controllability (Ramadge and Wonham, 1982), and effi- cient use of system structure, such as is done with perturbation analysis (Ho, 1985; Suri, in press) or the likelihood ratio method (Glynn, 1986~. All of these developments have shown that combining a dynamic sys- tems view with the discrete event structure results in a fertile area for research (also see Ho, 1987~. · Development of more effective simu- lation tools. A manufacturing simulation model contains elements that include phys- ical systems, control systems, management 127 policies, external effects, etc. The use of interactive graphics and manufacturing terminology, from the start to the finish of a simulation task, can improve the effec- tiveness of simulation tools and make these techniques more usable. Building blocks are needed that will make such representations easy while retaining sufficient complete- ness. · Efficient optimization of simulations. In the context of deterministic system op- timization, a large number of software packages are available for system optimi- zation. In contrast, no generally usable package exists for optimization of Monte Carlo simulations. Although theoretical pa- pers abound, our experience is that the the- ory does not survive implementation! Much work needs to be done in this area, and there is room for substantial improvement over existing methods (Surf and Leung, 1987). · Scheduling. This critical area of re- search will be discussed in a later section. Implementation Phases Although this paper is concerned princi- pally with analysis, other important phases could properly be treated as a part of the manufacturing system life cycle, including procurement, installation, debugging and testing, and start-up. While each of these phases can benefit from analysis techniques such as project networks, Petri nets, soft- ware design, reliability, control theory, and many others, they are not as centrally af- fected by analysis as are the phases treated in more detail in this section. The bibliog- raphy in Suri (1985) and a relevant hand- book (Charles Stark Draper Laboratory, 1984) provide further discussion of these phases. Ongoing Operations Phase The objectives of the ongoing operations phase can be divided into three main cate-
128 Deal~n Phase Feasibillty Analysis Strategic Decisions Aggregate Analysis Tactical Decisions Detall Analysis ~ Operstlonal Decisions CORRESPONDENCE IN ACTIVITIES Ongoing Operations Phase FUGUE 2 Correspondence be - - n acti~ti~ in design phase and ongoing operation phase. gories: strategic, tactical, and operational. These three differ primarily in terms of the length of the time frame in which they are to be considered. Strategic objectives ex- tend over a period of years; tactical objec- tives, over several months; and operational objectives, from minutes to weeks. The principal aim for the strategic phase is capacity planning i.e., the develop- ment of long-range production plans and resource allocations and system modifica- tions i.e., the changes that accompany new products, processes, or equipment. For the tactical phase, the principal aim is to develop aggregate production plans and ag- gregate resource allocations. For the oper- ational phase, the main aims are the devel- opment of effective schedules, effective loading sequences, and job releases and the development of effective responses to disruptions due to breakdowns, non- availability of material, etc. Specific issues, performance measures, and analysis techniques for the ongoing op- erations phase are not enumerated here. In- stead we observe that there is a correspon- dence between activities in the three design phases (feasibility, aggregate, detail) and in the three ongoing operation phases (strate- gic, tactical, operational). That is to say, issues and techniques that apply to the fea- sibility stage during design also apply to the strategic decisions made during ongoing op- erations, and similarly for the other two sets of correspondences (see Figure 2~. Thus, in designing a manufacturing system, analysis tools developed during the design phase can continue to be useful in all aspects of the operational phase. More positively stated, RAJAN SURI since these tools can be important during the operation of a system, they should be developed during the design phase to gen- erate more effective system designs. Because of this correspondence, many of the research issues mentioned in previous sections apply here, too. We highlight a few critical ones: · Scheduling. A particularly critical re- search issue is that of scheduling. Current OR-based approaches are either too simple in their assumptions or too complex to be readily solved or implemented. As noted by Milton Smith, Texas Technological Univer- sity, at the 1984 ORSA/TIMS Conference on Flexible Manufacturing Systems, about 100 man-years had been expended by the OR community in solving minimum make- span scheduling problems, but he did not know of a single company that used mini- mum makespan to schedule shop floor op- erations (Gershwin et al., 1986~. We need to have scheduling approaches that incor- porate "real-world" considerations such as failures, shortages, schedule changes, and integration between corporate levels. In ad- dition, these approaches should be simple to implement and fast to execute. There is a vast opportunity for new approaches or new ways of tackling this area. We should also keep in mind some of the lessons pointed out in earlier sections of this paper. Perhaps the solution lies not in trying to incorporate all the requirements of failures, shortages, etc., but in finding good ways to minimize such effects and then in finding simple scheduling approaches that work well under these new conditions. There is definitely an opportunity for completely new methods in this area. · Real-time control of discrete event sys- tems. Given the proliferation of shop floor monitoring systems and shop floor comput- ers, it is natural that more will be expected of real-time control systems. Little is known about this area at the present time, except
A NEW PERSPECTIVE ON MANUFACTURING SYSTEMS ANALYSIS for a large body of heuristics. Recently there have been some successes using dynamic systems approaches (Akella et al., 1985~. There is also a potential for applying many of the new approaches to discrete event sys- tems mentioned earlier, in a real-time op- timization mode (Surf and Leung, 1987), and for coupling these with traditional control-theoretic ideas (Gershwin et al., 1986~. · AI-based approaches. This is also an area where AI-based approaches may do well (Thesen and Lei, 1986~. In fact, as mentioned earlier, AI-based methods, per- haps in combination with traditional ap- proaches, may be suitable for solving some of the difficult problems identified in all the phases that have been discussed. Obsolescence and Termination The last phase in the life cycle is that of obsolescence and termination. Because there is less current interest in this phase than in the others, it is not covered in this paper. Suffice it to say that the elements of flexibility, efficiency, and responsiveness also play an important role in this phase of the life cycle. DESIGN FOR ANALYSIS The concept of design for analysis is a broad research issue that covers the entire spectrum of manufacturing system analy- sis. Briefly stated, design for analysis in- volves creating and working with designs that will be simple and easy to analyze. This approach may appear backwards to most engineers, who would view analysis as a tool that must serve the needs of the de- signers. They might also think that this ap- proach would stifle the creativity and abil- ity of good designers. Nevertheless, there are some strong advantages of design for analysis. 129 Precedents There are precedents in the design field for changing the perceived user-resource re- lationships, and such changes can result in benefits. Up to the 1960s, it was taken for granted that a manufacturing designer knew best how to design products, and the job of the assembly engineer was to find the best way to assemble that product. The as- sembly engineer was considered to be a re- source serving the designers. In the 1970s, however, the concept of design for assem- bly was introduced. This concept recog- nized that designing a product, while keep- ing in mind how it will be assembled, offered significant benefits, including lower cost and higher quality. This concept has since been broadened to include what is called design for manufacturability, thereby including all manufacturing processes. More recently, this approach has been ex- panded (Whitney et al., 1988, in this vol- ume3 to suggest that it can encompass stra- tegic benefits as well. These developments have taken assembly and process planning and made them a part of the drivers that direct the user- the designer. What Is Design for Analysis? A central idea of design for analysis is that we should undertake the design of a product, or system, keeping in mind that the design will need to be analyzed. There- fore, if the design can be analyzed by using simpler and better-understood models, we may be able to model and analyze the con- cept faster, thereby allowing more time for examining various design alternatives, making sensitivity analyses, and asking "what-if" questions. The basic tenet is that this additional time will result in larger payoffs than the time spent in modeling and analyzing a more complex design. Design for analysis, as an overall meth- odology, may result in other less obvious,
130 but no less important, benefits. The best way to see these possibilities is through some examples. Example from Solid Modeling Voelcker (1988, in this volume) gives an example of the product and assembly de- scription language (PADL) solid modeling system being used to represent the compo- nents in a Xerox copier. The simplest PADL system is the level 1.0 system (in terms of the number of different primitives it can represent). It was found that PADL-1.0 could fully represent 30 percent of the com- ponents in the copier. However, it was found that another 30 percent of the com- ponents could be redesigned so that they could be modeled by PADL-1.0 (Samuel et al., 1976~. Voelcker now reports that a comparison of several of the redesigned components with their original counter- parts indicated that the redesigned compo- nents were, in fact, superior from several points of view. This illustrates an unusual situation. Placing constraints on a designer may actually lead to a better design. Example from Manufacturing System Design Consider a hypothetical situation in which two teams independently undertake the design of a group technology cell to meet certain corporate manufacturing ob- jectives. The design begins at the stage of selecting the products and equipment that should make up the cell. The design is to be done under typical time pressures. Team A wants to consider several different priority scheduling schemes, operator policies, and lot-splitting techniques to optimize the per- formance of the cell. The team decides to put together a simulation model to allow all of these parameters to be studied. Team B decides to consider simple performance policies and finds that a spreadsheet pack- age along with a simple analytic modeling RAJAN SURI tool (e.g., Suri and Diehl, 1987) are suffi- cient to study the alternatives. It is our claim, not as yet based on scientific re- search but only on personal observations of industrial operations, that Team B is likely to arrive at a strategically more effective design. As a result of the time pressure, Team A is likely to spend a good deal of time developing the complex model and not enough time on exploring the alternatives. Team B will start exploring alternatives early in the process and will quickly iden- tify some of the critical parameters to be examined. It will have an opportunity for several iterations of designs, and it will have explored a wide range of alternatives. In this hypothetical example, although Team B was constrained to look at a smaller class of solutions, with the analytic tools that it chose to use, it was actually able to explore more alternatives under the "real-world" pressure. So the design for analysis princi- ple would encourage the elimination of cer- tain designs that are too hard to analyze in favor of designs that are simpler to analyze. What Design for Analysis Is Not Since the concept of design for analysis may seem upsidedown to many engineers, it may be useful to further clarify it: · Design for analysis does not mean dis- torting reality to fit our models. It does mean asking, "Should we change what we have chosen to define as reality to fit our model?" and "What might be the benefits of this change?" It forces us to think about the impact of this change and not take re- ality to be determined by a given situation that must then be modeled. · Design for analysis does not mean we should stifle development of modeling tech- nology. We should definitely be pushing the frontiers of what we can do with analysis and modeling. It does, however, focus on using today's analysis technology most ef- fectively.
A NEW PERSPECTI VE ON MANUFACTURING SYSTEMS ANALYSIS · Design for analysis is not just a differ- ent way of saying "design for simplicity." Designing for simplicity is an important goal that many industries are recognizing today. Design for manufacturability is one step toward meeting that goal. Design for analysis is also a step toward that goal, in that it offers specific approaches and yard- sticks, through the available analysis tools, by which to learn whether simplicity is be- ing achieved. Research Potential There is more depth to this methodology than may be evident from the foregoing brief description, and there is scope for novel research on this concept. Design for analysis can lead to · Simplicity. In science this is usually coupled with elegance of solution and con- cepts. · Robustness. The resulting design works better over a wide range of conditions. · Responsiveness. Designs and manufac- tured products can be created faster and more accurately. · Simpler operations. Systems designed using this method will be easier to manage, operate, and change. · Strategic focus. Teams working with this method will find themselves forced to look at strategic issues and not just tactical responses. We do not claim that all of these results will follow in all situations, but we do be- lieve that they can follow in a number of nontrivial instances. The research issues that follow from this concept can be stated as follows: · Establish a number of instances, from different disciplines, where design for anal- ysis leads to recognizable benefits (e.g., the Voelcker case study described earlier) and at the same time identify counterexamples 131 i.e., instances where its use is counter- productive. · Develop an understanding, perhaps leading to a theory, of why this method works and when it can be expected to work, for example, why did it produce a better component for the copier? · Determine the building blocks that will make systems that are simple to design and analyze, and it is hoped as a consequence, simple to implement, operate, change, and manage. CONCLUSION There is a need for good "traditional" analysis approaches to be used in conjunc- tion with recent manufacturing methods JIT, AI, etc. and whole new areas for re- search are just opening up. However, it is essential that we work closely with real sys- tems and that we keep abreast of recent developments. On a more specific note, we offer the following concrete suggestions of what this research community can do. Individuals must make a commitment to get to know the area. They should work closely with industry. One could set a per- sonal goal to work a given number of days in a factory each year, or to get a certain amount of industrial funding, or to be in- volved in industrial projects. Individuals should also stay up to date on technologies, approaches, and philosophies in this field. Only after taking these basic steps should one look into the research areas in this field. The community professional societies, journals, and research establishments must also encourage developments in posi- tive directions. There should be an empha- sis on, and rewards for, good applications. There should be better mechanisms and in- centives for joint industry-university proj- ects. We must improve the exchange of ideas among theoreticians and practition- ers, perhaps through special workshops or publications. Finally, we must reexamine
132 our attitudes and journal acceptance pro- cesses so that emerging new ideas are better nourished, rather than simply publishing incremental contributions on safe, accepted topics. Our long-term health depends on continued innovation, not just refinement. The aim of this paper is not to present one person's view but to encourage the whole community to think more about these issues. If further introspection and dialogue increases the viability of the community, the resulting developments can be very beneficial. Acknowledgment This work was partially supported by the National Science Foundation under Grant No. DMC-8717093. REFERENCES Akella, R., Y. Choong, and S. B. Gershwin. 1985. Real time production scheduling of an automated cardline. Pp. 403~25 in Flexible Manufacturing Systems: Operations Research Models and Appli- cations, K. E. Stecke and R. Suri, eds. Basel, Switz- erland: J. C. Baltzer AG. Charles Stark Draper Laboratory. 1984. Flexible Manufacturing Systems Handbook. Park Ridge, N.J.: Noyes Publications. Cohen, G., D. Dubois, J. P. Quadrat, and M. Viot. 1985. A linear-system-theoretic view of discrete event processes and its use for performance evalua- tion in manufacturing. IEEE Transactions Auto- matic Control AC-30:210-220. Fox, M. S., and S. F. Smith. 1984. A knowledge- based system for factory scheduling. Expert Systems 1:25~9. Gershwin, S. B., R. R. Hildebrant, S. K. Mitter, and R. Suri. 1986. A control perspective on recent trends in manufacturing systems. Control Systems Maga- zine 6~2~:3-15. Glynn, P. W. 1986. Stochastic approximation for Monte Carlo optimization. Pp. 356-365 in Proceed- ings of the 1986 Winter Simulation Conference. New York: Institute of Electrical and Electronics Engineers. Haider, S. W., and J. Banks. 1986. Simulation soft- ware products for analyzing manufacturing sys- tems. Industrial Engineering 18:98-103. (Also see errata and corrected figures in Industrial Engineer- ing 18:86-87.) RAJAN SURI Haider, S. W., D. G. Noller, and T. B. Robey. 1986. Experiences with analytic and simulation modeling for a factory of the future project at IBM. Pp. 641- 648 in Proceedings of the Winter Simulation Con- ference. Hall, W. 1983. Zero Inventories. Homewood, Ill.: Dow Jones-Irwin. Hayes, R. H., and S. C. Wheelwright. 1984. Restor- ing Our Competitive Edge: Competing Through Manufacturing. New York: Wiley. Ho, Y. C. 1985. A survey of the perturbation analysis of discrete event dynamic systems. Pp. 393-402 in Flexible Manufacturing Systems: Operations Re- search Models and Applications, K. E. Stecke and R. Suri, eds. Basel, Switzerland: J. C. Baltzer AG. Ho, Y. C. 1987. Performance evaluation and pertur- bation analysis of discrete event dynamic systems: Perspectives and open problems. IEEE Transac- tions on Automatic Control AC-327):563-572. Lundrigan, R. 1986. What is this thing called OPT? Productivity and Inventory Management 27~2):2- 12. Meleton, M. P., Jr. 1986. OPTFantasy or break- through? Productivity and Inventory Management 27~2~:13-21. Ramadge, P. J., and W. M. Wonham. 1982. Super- vision of discrete event processes. Pp. 1228-1229 in Proceedings of the IEEE Conference on Decision and Control. Samuel, N. M., A. A. G. Requicha, and S. A. Elkind. 1976. Methodology and Results of an Industrial Part Survey. Report TM-21, Production Automation Project, College of Engineering, University of Rochester. Schonberger, R. J. 1982. Japanese Manufacturing Techniques. New York: The Free Press. Skinner, W. 1985. Manufacturing: The Formidable Competitive Weapon. New York: Wiley. Suri, R. 1985. Quantitative techniques for robotic sys- tems analysis. Pp. 605-638 in Handbook of Indus- trial Robotics, S. Y. Nof, ed. New York: Wiley. Suri, R. In press. Infinitesimal perturbation analysis for general discrete event systems. Journal of the ACM. Suri, R., and S. DeTreville. 1986. Getting from "just- in-case" to "just-in-time": Insights from a simple model. Journal of Operations Management 6~3~:295-304. Suri, R., and G. W. Diehl. 1987. Rough-cut model- ing: An alternative to simulation. CIM Review 3:25-32. Suri, R., and Y. T. Leung. 1987. Single Run Opti- mization of Discrete Event Simulations: An Empir- ical Study Using the M/M/1 Queue. Technical Re- port 87-3, Department of Industrial Engineering, University of WisconsinMadison. Thesen, A., and L. Lei. 1986. An expert system for
A NEW PERSPECTIVE ON MANUFACTURING SYSTEMS ANALYSIS scheduling robots in a flexible electroplating system with dynamically changing workloads. Pp. 555- 566 in Proceedings of the Second ORSA/TIMS Con- ference, K. E. Stecke and R. Suri, eds. Amsterdam: _. . ~lsevler. U.S. Congress, Office of Technology Assessment. 1984. Computerized Manufacturing Automation: Employment, Education, and the Workplace. Of- fice of Technology Assessment, OTA-CIT-235, Washington, D.C. 133 Voelcker, H. B. 1988. Modeling in the design process. In Design and Analysis of Integrated Manufactur- ing Systems, W. Dale Compton, ed. Washington, D.C.: National Academy Press. Whitney, D. E., J. L. Nevins, T. L. De Fazio, R. E. Gustavson, R. W. Metzinger, J. M. Rourke, and D. S. Seltzer. 1988. The strategic approach to prod- uct design. In Design and Analysis of Integrated Manufacturing Systems, W. Dale Compton, ed. Washington, D.C.: National Academy Press.
