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.
SIMULATION IN DESIGNING AND SCHEDULING MANUFACTURING SYSTEMS F. HANK GRANT ABSTRACT As manufacturing companies strive to achieve increased efficiencies, they must make effective use of technology. Simulation is an important tool in accom- plishing this. The use of simulation for scheduling and control of production systems is a natural outgrowth of its application for the design of systems. Simulation, when used for produc- tion scheduling, is a useful vehicle for providing the discipline necessary for effective production management of the factory floor. This paper discusses the applications that provide and support the use of simulation for the design and operation of integrated manufacturing systems. A discussion is given of the new technology Mat makes simulation available for production scheduling. The differing objectives of the production system scheduler and the production system de- signer are discussed. Important research topics in simulation are also identified and discussed. INTRODUCTION Manufacturing companies have a press- ing need to understand new technology and the potential for its use in a rapidly chang- ing environment. Although simulation has been used traditionally to help explore the ramifications of new technology for manu- facturing systems, many companies now re- quire that simulation studies of proposed manufacturing facilities be performed be- fore a final decision is made on implemen- tation of either new or current technology. Simulation can provide insight into issues that are not apparent or are counterintui- tive. Many software tools now available provide engineering support for system de- sign projects. The use of simulation for scheduling and control of production systems is a natural outgrowth of its application in design. The 134 objective in designing a manufacturing fa- cility is to obtain knowledge sufficient to make capital commitment decisions that satisfy production objectives, and these same characteristics are also important in controlling production. Simulation, when used for production scheduling, provides a vehicle for achieving the discipline neces- sary for effective production management on the factory floor, thereby helping to achieve productivity goals through effi- cient, effective use of new, integrated tech- nology. New computer technology, such as powerful workstations, coupled with ad- vances in easy-to-use software systems, makes simulation a useful tool for the pro- duction scheduler. Simulation is an analysis tool that can be applied effectively to a variety of shop floor design and real-time shop floor control problems. Simulation can support longer-
SIMULATION IN DESIGNING AND SCHEDULING MANUFACTURING SYSTEMS 135 term system design evaluations of resource requirements, equipment needs, inventory buffer sizes and availabilities, and sensitiv- ity analysis of a variety of product demand and equipment performance probabilities. Simulation can also support shorter-term decisions involving equipment scheduling, shop order release, and work order sched- uling. Historically, simulation techniques have been highly successful and used extensively for the planning and analysis of current op- erations or proposed designs. Although on- line simulation analyses are feasible and can be cost-effective, applications of simulation in this mode have been few. On-line appli- cations of simulation for real-time shop floor control purposes can be applied effectively only if the data supplied to the simulation models are accurate, organized, and timely. In the past these constraints have proved difficult to overcome. More recently, major advances have been made in data-base technologies. These im- provements make the access and manipu- lation of data for real-time factory control a reality. A modeler can now construct a simulation model of an operating unit and supply this model with accurate and timely data describing the performance of ma- chines and the operators and the expected demands on the manufacturing system. Other data processing improvements that facilitate the use of simulation for on-line analysis relate to computational capability and the graphic display of results. Mana~- ers have historically disliked having to wait a long time for analysis. With the comput- ing capabilities and graphic constructs that are now available, managers not only can get quick response for analysis of work or- der scheduling or work order release, they can also receive the information in more easily understood form. This paper discusses how these latest de- velopments will support the use of simula- tion for designing and controlling produc- tion systems. It also identifies and discusses the important research areas and the chal- lenges in effectively applying this technol- ogy. SIMULATION TECHNOLOGY FOR DESIGN Simulation is the process of creating a representation or model of the operation of a system on a digital computer. This model is created by providing to the computer a description of the physical components of the system and the logic associated with the operation of the system. The model of the system under study is put together using a simulation language. This language gives structure to the model-building process by providing special constructs that relate to the system under study. For example, many languages provide constructs to represent components like queues and servers. Other more system-specific languages provide constructs to represent machines, parts, and process plans. Many different simulation languages are available. These include general-purpose languages, such as SLAM II, SIMSCRIPT II.5, and GPSS, and more specialized lan- guages oriented toward manufacturing, such as MAP/ 1, SPEED, and MAST (Haider and Banks, 1986; Miner and Rol- ston, 1983; Pritsker, 1986~. Once built, the simulation model serves as a laboratory in which various design al- ternatives can be tested and compared. By running the model on a computer, the ac- tions of the proposed system are represented in detail, permitting inferences to be drawn about overall system performance. These inferences are made on the basis of numer- ous performance measures provided by the simulation, such as machine utilization, in- process inventories, part waiting times, and throughput. Through this process of exper- imentation, the best overall system design is selected.
136 Several features of simulation make it particularly advantageous for the design of manufacturing systems (Musselman, 1984~. These features include physical and control system balance, system flexibility, random system behavior, and animation. Physical and Control System Balance Proper balance among the physical and the logical and control components of a manufacturing system must be maintained during system design. Should the design of one component be emphasized dispropor- tionately, overall system performance could suffer. Designing a system with propor- tional importance given to each component is possible through the total systems per- spective that simulation provides. For example, in modeling a flexible man- ufacturing system (FMS), the system can be viewed as being composed of two interre- lated subsystems: a physical subsystem and a control subsystem. The physical subsys- tem, which transports, stores, and processes parts, consists of programmable machines, material handling equipment, and in- process storage facilities. The control sub- system, which selects, sets priorities, routes parts, and controls traffic flow, consists of situation-dependent logic to coordinate part interchange in the physical subsystem. These two subsystems must work in con- cert. It is important, therefore, to consider the interaction between these two subsys- tems during the design. Isolating, for anal- ysis purposes, one subsystem from the other or giving one preferential treatment will most likely result in misleading conclusions. A more complete evaluation, with ap- propriate consideration being given to the interplay between these two subsystems, is possible with simulation. In contrast to strictly analytical approaches, simulation includes the system's operating and control strategies as an integral part of the model. As parts move through the various opera- tions, the model processes them according F. HANK GRANT to each machine's operating characteristics and routes them according to the system's situation-dependent control logic. Simula- tion provides a rich environment for the design of physical as well as logical and control systems. System Flexibility The increasing popularity of simulation is due, in part, to its ability to represent various levels of detail. Analytical formula- tions, while offering closed-form solutions, tend to be restrictive, since details must of- ten be neglected in order to accommodate the formulation. Simulation, on the other hand, can provide as much or as little detail as the analyst wants. All relevant system characteristics can be taken into account, such as processing time variability, equip- ment reliability, fixture restrictions, in- process storage requirements, complex routing decisions, operating policies, and scheduling constraints. Simplifying as- sumptions, such as reduced decision mod- els, are not required. The result is a flexible experimental setting in which to test alter- native design strategies where the analyst has control over the details and assumptions that are included. Random System Behavior The operational behavior of manufactur- ing systems can be quite dynamic over time. That is, the interaction of various system components can trigger unexpected behav- ior that may occur only infrequently. The system can exhibit, for example, significant variation in demand for resources, depend- ing on the particular interactions of its com- ponents. Although this dynamic behavior tends to increase the complexity of the de- sign, and often the model, knowledge and understanding of this behavior can lead to significant design efficiencies. Simulation provides a means of understanding system dynamics and learning about components
SIMULATION IN DESIGNING AND SCHEDULING MANUFACTURING SYSTEMS 137 of the system that may interact and behave in a counterintuitive manner. Proper accounting of random variability of system parameters helps ensure that a proposed design will exhibit stable perfor- mance in a variety of configurations. Ironi- cally, instability can actually be promoted when certain sections of a system are selec- tively overdesigned to guard against pro- cessing variations. As a result, there can be unforeseen weak links in the system. Through systematic experimentation with a model, weak links can be revealed and actions taken to correct them. An example of this use of simulation is in the study of fixture load stations. To exam- ine the effectiveness of a particular confi~- uration for a fixture load station, the part arrival sequences and their demand on the station must be understood. Unfortunately, given the variety of part types that could be simultaneously processed and the process- ing variations that could occur for each part type, the number of different part arrival sequences is extremely large. Examining all possibilities in great depth is unreasonable. However, rerunning a representative, ran- domly selected set of part mixtures in the simulation model can provide insight into the station's performance. If the design of this station were found to be deficient, it could be modified and then tested against a broader range of part arrival sequences. This ability to deal with random behavior with the models and to create representa- tive samples that stress the system's capabil- ity is a major benefit of simulation. Animation the operation of the control subsystem can be studied in action. This provides a detailed understanding of the implications associ- ated with the system's control policies. Sub- tle errors can be identified and corrected before the system becomes operational. SIMULATION TECHNOLOGY FOR SCHEDULING AND CONTROL The same simulation technology that has been used traditionally in system design can be used as the kernel for day-to-day produc- tion scheduling applications. As an intro- duction to the application of simulation to detailed production scheduling, traditional tools, the limitations of these tools, and the ways that simulation technology can ad- dress these limitations are discussed. Traditional Scheduling Methods and Their Limitations Shop floor scheduling is an important task in managing a production system. It entails complex decisions that affect such objectives as meeting delivery due dates and maintaining a desired level of inventory. Although it is not possible to consider all of the variables that determine the effective- ness of a particular schedule, major produc- tivity improvements can be realized by making the production scheduling process more effective. The quality of a production schedule in- volves many, sometimes conflicting, objec- tives. Whereas maximizing throughput is certainly one important consideration, an ideal schedule will also have the following characteristics: One aspect of simulation technology that has made considerable progress in recent years is graphical animation. Besides the usual plots of system performance, it is now possible to animate a manufacturing sys- tem operation in great detail (Grant and Weiner, 1986~. By bringing the model of the physical subsystem to life graphically, · Delivery due dates will be met. · Inventory costs will be maintained at acceptable levels. · Equipment, personnel, and other lim- ited resources will be well utilized and have balanced workloads. · Adaptations can be made quickly in
138 the case of an unexpected event, such as equipment failure or raw material short- age. Since it is difficult to "optimize" a sched- ule over all these characteristics, one or two characteristics are often chosen, depending on current production objectives. Gener- ally, trade-offs must be made to reach a balance between these more limited objec- tives. Production scheduling is done in many ways in industry. Probably the most com- mon methods of scheduling are purely manual techniques. In the most straightfor- ward form, an expert such as the depart- ment foreman or the machine operator se- lects the next job from those waiting in front of the machine. The criteria used in this circumstance often reflect the measures by which the scheduler is evaluated and may not reflect overall business objectives. lob status control boards are also used to lay out schedules visually. A more analytical approach to schedul- ing is sequencing by dispatching rules. This method uses rules that set priorities for the jobs waiting for processing. Research has demonstrated that rules such as the weighted shortest processing time rule can generate reasonable schedules. The effec- tiveness of the schedule may vary widely, depending on the particular rule selected, the type of production facility, and the mix of jobs to be produced. It is impossible to predict which dispatching rules will work best in most manufacturing systems by tra- ditional methods. They are also limited in the scope of what they consider and are often hard to implement on the shop floor in a cost-effective manner. Material requirements planning (MRP) was one of the earliest computerized tech- niques for factory management. In its ear- liest form it generated unconstrained pro- duction schedules, which were based on a bill of materials and estimated production time requirements. Manufacturing re- F. HANK GRANT source planning (MAP-II) expanded the scope of MRP to consider many other facets of production facility management. In ad- dition to sophisticated factory accounting capabilities, modules for capacity planning and shop floor data collection were also provided. These techniques are effective for longer-term scheduling and order launch- ing, but they lack the detail necessary for effective day-to-day production scheduling. Recently, there has been emphasis on more sophisticated definitive capacity plan- ning tools as a means of generating produc- tion schedules. These methods determine the expected critical resource and then schedule forward and backward around that resource. At present, these techniques do well on a global level but are inadequate when highly interactive components are present and frequent changes are needed. The computational time required to gener- ate schedules is usually large, with interac- tive execution being impractical. Also, the critical, or bottleneck, resources tend to change, based on production demand. As an alternative to these approaches, simulation-based scheduling can provide an effective tool for shop floor scheduling while requiring few assumptions. The schedules generated are based on an accurate, realis- tic model of the production facility. Simulation-Based Scheduling Simulation practitioners are familiar with the ability of simulation models to predict system behavior in great detail. It is a natural extension to attempt to apply sim- ulation on a day-to-day basis to predict schedule performance and resolve problems before they occur. Simulation is well suited to this type of analysis and, with proper support, can be used successfully to create and evaluate production schedules. The simulation model provides a computer rep- lica of the department in the factory. The model plays through the schedule and pro- vides performance information. Event trace it.
SIMULATION IN DESIGNING AND SCHEDULING MANUFACTURING SYSTEMS 139 information can provide the details of a fea- sible schedule, given the constraints speci- °fied in the model. The detailed interaction of various production limitations can be in- cluded at any level of detail. The dynamic interactions between re- sources can be captured and analyzed with simulation. For example, a material han- dling vehicle may deliver product to several machining stations. Even though this vehi- cle may have an expected total workload requiring only 50 percent of its available time, two stations that need material at the same time will cause the system to perform differently than it would if there were no conflict for the vehicle's services. Simula- tion can be used to evaluate the effect of these conditions on schedules and to resolve the conflict. In another case, the scheduler may have many options for selecting the transfer batch size in a production order. The entire order may be processed at each operation, or the order may be split into two or more separate loads and allowed to move inde- pendently through the operation. Simula- tion would allow the scheduler to contrast alternatives and select the one that provides the best performance. Because this analysis is performed on a computer, many alternatives can be ex- plored with relatively little expense or risk. Simulation models can represent a large va- riety of the factors critical to a manufactur- ing system's performance in great detail. The influence of tooling, personnel, and other resources can be evaluated. Capacity changes, such as a machine breakdown or scheduled maintenance, may also be in- cluded in the model. Most simulation languages have the modeling features necessary to represent production systems. But there are many ad- ditional needs in scheduling that are not typically provided. To be effectively used for scheduling, simulation models must provide reports that can be readily under- stood and used by the production sched- uler. Preferably, reports should be provided that serve as actual job performance sched- ules for equipment and personnel and are consistent in format. Also, a strong user in- terface is needed to support the scheduler and to allow schedules and other reports to be created easily. The scheduler should not have to interact with the details of a simu- lation model but still should be able to use it easily. Most simulation languages do not provide these features. The requirements for simulation in a scheduling and control mode are discussed in more detail in a later sec- tion. Traditionally, large simulation models of production facilities have been too expen- sive to build and too cumbersome for use on a daily basis. The modeling process has required a highly trained and experienced analyst with a solid understanding of both the simulation language and the system un- der study. Even after the model is com- pleted, execution of detailed models could take too long to be useful in production scheduling. This problem has been ad- dressed in part by the advance of computer technology. Workstations that provide mainframe-like capabilities are moving rapidly to the factory floor. With the de- crease in price of local area networks for factories, and their increasing use, comput- ers that focus on scheduling will proliferate. Also, more advanced simulation languages permit faster execution and more efficient model development. These languages are tailored to the simulation of manufacturing systems. The following overview of an ex- isting tool illustrates how simulation can be used in this mode. One Available Scheduling Tool Currently there are few commercially available products that provide tools for ap- plying simulation to production scheduling. One product that addresses scheduling is called FACTOR (Grant, 1986, 1987; Prits- ker et al., 1986~. It is designed to be highly
140 interactive and used by shop floor schedul- · ng supervisors. FACTOR supports two types of users in providing a shop floor scheduling system. The primary user is the factory floor pro- duction supervisor, who uses it to generate production schedules. The second user is the individual responsible for modeling the factory and developing the data that char- acterize the manufacturing system and the products that are produced in it. The model development activity typically occurs only once in the installation of FACTOR=. Modification of the FACTOR model is easily accomplished to incorporate changes in the production department after it has been installed. FACTOR can be integrated with the production management software and plant data to provide a useful tool for the produc- tion scheduler. One important feature that FACTORY provides is an interface to the MRP system to extract information related to a master schedule. It also interfaces with the shop floor status system to acquire pro- duction data. In FACTOR=, this is typi- cally a straightforward interface to the par- ticular MRP system that is in use. This interface collects the current list of orders to be processed at the production depart- ment and automatically loads them into the FACTOR system. A similar interface is provided to collect information regarding the current location and status of released orders from the shop floor data collection system. This information is also loaded into FACTORY to initialize the scheduling sys- tem automatically to start at current shop floor status. These data are stored in a local data base that may be accessed and edited by the production supervisor. Once the scheduling supervisor has gath- ered the current production requirements, FACTOR is used to simulate production and identify potential problems. The super- visor can select from a large variety of rank- ing priority codes to sequence jobs through the production system. These codes include F. HANK GRANT familiar priority measures such as due date and shortest processing time. In addition, more complex ranking priorities are pro- vided, such as estimated remaining process- ing time based on past performance as well as a critical or tactical ratio. The critical ratio measure is the ratio of the scheduled remaining production time to the estimated production time. The tactical ratio is the inverse of the critical ratio. Additional pri- ority schemes are available based on pro- duction costs and other system status vari- ables. Tools such as FACTOR make many contributions, including implementation of technology necessary to schedule the fac- tory floor. Another is the development of the framework necessary for further devel- opments in scheduling technology. The next section describes a set of re- quirements to support the application of simulation in production scheduling and compare those with requirements in design application. THE FOCUS IN SCHEDULING AND CONTROL VERSUS DESIGN There are significant differences when applying simulation for production system design as opposed to production scheduling. Table 1 summarizes these differences. Effectiveness of the User Interface An issue of primary importance is the effectiveness of the user interface. Simula- tion for the design environment uses a lan- guage-based interface, and the end user typically builds the model. Model manage- ment tools, such as the TESS system supple- ment (Standridge and Pritsker, 1987), have the ability to provide model-building tools that interface through a graphics device. Their focus is primarily on building the model and managing the simulation analy- sis. Scheduling and control applications have a need for significantly more powerful
SIMULATION IN DESIGNING AND SCHEDULING MANUFACTURING SYSTEMS 141 TABLE 1 Issues and Requirements for Simulation Technology Applied to Scheduling and Control versus Design Scheduling and Control Design 1. Strong user interface is needed to define the production replica and to generate production schedules. 2. Implemented set of algorithms for sequencing production orders is used. 3. Execution of the system is interactive. 4. An interface is provided to external data sources to integrate the simulation-based scheduling tool into base management systems. 5. Production schedule reports are explicitly produced for the factory floor. 6. All input and output is stored in a data base that can be interfaced to other systems. 7. Internal design of the software is oriented toward fast execution to respond to the needs of the production scheduler. 8. Orientation of the model is toward production planning and setting production objectives. 9. Technology is an application generator that supports two users: the modeler and the scheduler. 1. Language-based interface is used to build the design model and generate performance reports for various design alternatives. 2. User-designed and-coded algorithms associated with queue-ranking procedures are used. 3. Execution of the system is typically in a batch mode. 4. Any interfaced external data must be explicitly created by the user, and most design-mode simulation languages are not created to support an interface to external data. 5. Standard performance reports are provided for system performance measures, but any specific reports must be user-coded. 6. Data-driven design-level tools typically use data stored in a flat file, but input data and output data are sometimes stored in a data base, as is provided by a simulation study management system. 7. A general-purpose language addressing a wide variety of system and analysis objectives will require moderate execution times. 8. Orientation of the model is toward broader design issues, including randomness as a standard component. 9. Technology is a modeling tool and focuses on design models for the modeler-analyst. tools to support the production scheduler. Such tools are needed since production schedulers tend to be experts in their partic- ular manufacturing operations but not in simulation or other computer technology. An easy-to-use interface must be provided to support the generation of production schedules as well as the need to manage the implementation of those schedules on the factory floor. The user interface is also an excellent application for expert system tech- nology. Expert system tools can be used to analyze the need to reschedule based on progress toward expected performance. They can also identify problems in produc- tion schedules developed for alternative scenarios and present them in an easy-to- interpret format. They can extend the scheduler's experience and greatly ease the data analysis requirements. Support Features for Production Scheduling Models The second issue is concerned with sup- port features required for the production scheduling models. In the design mode, the user typically develops and codes the vari- ous algorithms used in the model. In the scheduling and control mode, the user re- quires an implemented set of algorithms that can be easily accessed to support sched-
142 uling applications. Further, the scheduler will typically access more than a single al- gorithm in any scheduling exercise. From these alternative scenarios, the alternative generating the best performance will be se- lected for implementation. Execution of the System In the design mode, execution of the model is typically done in a batch process without much interaction. The exception to this procedure occurs when animation is used to view the simulation where the ca- pability exists to interrupt the animation and review and modify certain system pa- rameters. Execution of the model in the scheduling and control mode must be inter- active. The scheduler must have the ability to interact with both the management sys- tem and the scheduling tools to review the data provided and make production man- agement decisions. Data Needs for Production Models Extensive external data requirements ex- ist in scheduling and control applications. It is necessary to provide an interface that allows the scheduling supervisor to inte- grate the scheduling control sv~tem easily into the other management tools that are used in the production facility. The inter- face must be able to download the data easily whenever required into the schedul- ing model and to develop the necessary schedules. In the design mode, the user is focused on specific design tasks and has lit- tle need for a day-to-day interface with other production data. The data require- ments are addressed by gathering perfor- mance data to assess the impact of the de- sign on productivity goals. The interface to external data for scheduling and control ap- plications must be very general as well. Each factory has different data storage characteristics and capabilities, and to be O , ~ , F. HANK GRANT effective in this area, a tool must be able to address a variety of applications. Reporting Requirements Reporting needs in the design mode are focused on system performance measures and are usually tied to simulation con- structs. Special-purpose languages are alle- viating this issue in providing reports in the context of system-specific issues, such as machine performance or operator utiliza- tion. In the scheduling and control mode, reports must be provided in a form explic- itly applicable to the factory floor situation. This includes not only performance reports but also detailed production schedules that the operators can follow and execute. A re- port generator is also required to provide additional presentation flexibility. Storage of Data in a Data Base Although many design tools have adopted the philosophy that input and output data should be stored in a data base, several oth- ers provide input data in the form of a flat file. They also store output data in a flat file. In a schedule and control application, all input and output data must be stored in a data base. The user has the need to query the output data base interactively to evalu- ate various performance issues as well as to compare alternative scenarios. That same need exists in the design mode, but it is even more critical in production scheduling be- cause of the time constraints imposed on the production scheduler. Ability to Address a Variety of System and Analytic Objectives The requirement to address a wide vari- ety of objectives typically imposes con- straints on execution speed. Given the breadth of application design, there are bounds on the execution efficiencies that can be achieved in these tools if they are to -
SIMULATION IN DESIGNING AND SCHEDULING MANUFACTURING SYSTEMS ]43 execute quickly. In a scheduling and con- trol mode, the application is focused on a specific objective to generate achievable production plans in the shortest possible time. Therefore, the internal design of the simulation software kernel can be oriented toward fast execution in response to the needs of the production scheduler. Issues such as stochastic variables and interpretive languages, which are important in a design mode, may be eliminated entirely in the production scheduling applications. Breadth of Demands on the Models Design issues are typically relatively broad. The user is focused on exploring al- ternative designs in the production system and evaluating the response of the system to various stresses in production demands. In scheduling and control, the focus is to develop a replica of what exists on the fac- tory floor in order to plan production. The approach is not to run the simulation and see what happens, which tends to be the objective of design, but rather to run the model and produce an achievable plan that the factory floor can execute. This yields a significantly different model from that in the design mode. These models may be more focused and often require less detail and scope than those required for design. They often require, however, significant detail in the representation of the decision processes that occur on the factory floor. These processes tend to require very explicit rules that emphasize the characterization of existing procedures. Users of Simulation Simulation technology for scheduling production systems typically addresses a different group of users than simulation technology applied in the design mode. When used in the scheduling and control mode, the technology is actually an appli- cation generator that the production mod- eler uses to create a scheduling system spe- cific to the production environment for use by production schedulers. The modeler can be an engineering group or a data process- ing group that is closely tied to the manu- facturing environment. The scheduler, of course, is someone with production sched- uling authority on the factory floor. In the design mode, the technology tends to focus on the needs of the system designer, who is typically someone from an engineering group. The user's objectives are to explore various alternatives for meeting production goals and objectives and to justify the cost- effectiveness of the proposed system. RESEARCH OPPORTUNITIES Simulation has been used for the design of industrial systems for many years. Its use as a scheduling and control tool is more recent. Significant opportunities for further research exist in both applications. A1- though the design mode is certainly better developed, significant problems still exist. The application of simulation technology in scheduling and control is an extremely fer- tile area. The following sections describe the more important research to be ad- dressed and provide additional details on the opportunities and challenges of each re- search area. Use and Integration of Artificial Intelligence Tools in Simulation Artificial intelligence is an extremely broad area and has many possible applica- tions in simulation. The most attractive is expert system technology. Research has demonstrated the feasibility of developing simulation in an expert system framework and using standard expert system technol- ogy to generate the simulation executive, models, and outputs (Ready and Fox, 1982~. A major difficulty with this ap- proach has been the execution speed of the programs.
144 A primary opportunity for the applica- tion of expert system tools is in support of output analysis. Simulation requires signif- icant effort on the part of the user to extract and condense the large amount of infor- mation that is produced. Expert system technology could drastically condense the output data and produce a knowledge base with which the user could interact to assess the effectiveness of a design. Expert system tools should also be able to support the design task as the model is de- veloped. Tools are required for developing the model, characterizing components of the model, and relating them to system ob- jectives and analysis output. A related tech- nology is the application of expert system tools to computer programming. Significant research opportunities also exist for the application of expert system technology in scheduling and control. First, expert system technology can be applied to capture expertise on the factory floor and integrate that information with the model. This application would include rules for de- cision making regarding characteristics of production sequencing. Such rules could become part of the scheduling system. The integration of expert system technology with simulation models should be relatively straightforward where the technical chal- lenges will be to implement the integration in such a way that simulation execution performance is not degraded. A second application area is concerned with output analysis. To make decisions, production schedulers must digest the vast quantities of data produced by a scheduling system. Expert systems could perform data reductions as well as analyze the output. They could make decisions to generate al- ternative scenarios and select the best. This application of expert system technology would also be of use in automated environ- ments, where the user of the scheduling module might be a cell controller, or an- other CPU. In this circumstance, the need is to be able to reschedule and evaluate the F. HANK GRANT results of the simulation scheduling appli- cation without human intervention. If the analysis procedures could be captured in an expert system, the decisions could be made automatically. The third area for the application of ex- pert system technology is concerned with analyzing performance to plan and deter- mine the need to reschedule the system. Production systems will always experience perturbations around a plan. The support needs center on determining the signifi- cance of those perturbations, defining the need to reschedule, and determining differ- ent methods of responding to a reschedul- · ~ ng requirement. Real-Time Data Collection and Me Interface win Models Factory floor data collection systems have existed for several years and are available in a variety of forms. They collect a wide variety of data, from production order tracking to detailed machine status. They have had varying degrees of success, pri- marily because of a lack of significant ap- plication for the data collected other than simple status reporting. Typically, by the time the information was collected and re- ported, it was too old to be of use. New developments in manufacturing, particu- larly in automated production systems, can provide the means for maintaining current data about the system. As the price/per- formance ratio continues to improve in computer technology, the use of effective data collection systems will increase in manual operations as well as automated systems. The primary application for factory status data is in scheduling and control ap- plications of simulation. Design activities are typically much longer term and, al- though some of the performance data col- lected would be of use, a real-time interface is not necessary. Real-time data collection and its inter-
SIMULATION IN DESIGNING AND SCHEDULING MANUFACTURING SYSTEMS 145 face with scheduling and control applica- tions offer interesting research opportuni- ties in many areas. The scheduling and control systems obviously require good status information to generate production schedules effectively. A real-time interface to the data collection system would permit the scheduling system to access those data whenever necessary. In fact, the simulation model could be run in parallel with the physical system to determine the resched- uling needs dynamically. Real-time collec- tion of information on performance could also drive certain parameters in the simu- lation model to increase its realism while still providing effective projections for con- trol of the production system. Data between the physical system and the scheduling models also flow to the phys- ical system. Simulation models can gener- ate a detailed production plan, and that plan can be implemented through the flow of data from the model to the production system. This can be implemented at a vari- ety of levels of detail, starting with an agenda prescribing specific manufacturing operations for the various components of the production cell. More detailed schedul- ing models could actually drive physical equipment and serve as an important part of the cell control software. Data Integration and Distribution Data integration is a significant issue in the application of simulation in production systems. The sources for data that drive the production system are typically distributed throughout the manufacturing control sys- tem and may require significant time and effort to organize and use effectively. A re- search opportunity exists in developing ef- fective methods for describing the integra- tion of data across all manufacturing operations and also for providing tools for effectively transferring the data from com- ponent to component. In a design mode, data are needed to characterize existing production policies and performance char- acteristics and integrate those into the model to accurately predict changes in pro- duction characteristics and evaluate the de- s~gn. When used in a scheduling and control mode, data integration from distributed data sources is critical. The data needed to support effective simulation models for scheduling and control range from the list of orders from the MRP system down through shop floor status information. Ma- chine characteristics, machine status, pro- cess plans, part descriptions, etc., are also required. In most corporate systems, this information lies in various sets of data bases. Current applications use a standard ASCII file to transfer data. Additional re- search is needed to define how the data may be more effectively distributed and ac- cessed. It is clear that the success of addi- tional production management tools using simulation technology depends on their ability to be easily integrated into existing components. Interaction and Integration of Simulation with Automated Systems Automated systems are being imple- mented throughout industry. Research op- portunities for implementation of simula- tion in automated systems exist primarily in scheduling and control applications. Cell control software development efforts can be greatly reduced if the technology to sched- ule the cell and react to various problems that occur in production are included as a part of the control software development. It has been shown that this function can be successfully implemented as a separate module. This allows the cell control soft- ware to be relatively simple and focus on hardware control issues as opposed to scheduling issues. Implementation of sched- uling within the cell controllers also tends to create a significant overhead burden on the control hardware, since it requires
146 continuous iteration to regenerate the schedule. Production scheduling systems, such as the one described earlier in this paper, can be interfaced with automated systems to provide scheduling tools to generate de- tailed production schedules effectively and provide a significant analysis capability as well. Cell control software can invoke the production scheduling system to run several scenarios and generate a "best schedule" for implementation in the automated cell. This can often be done without human interven- tion when various scenarios for analysis have been set up ahead of time. Several significant research questions re- main. One of the most important is con- cerned with data integration in the auto- mated system. Data integration requires solving the problem of effectively interfac- ing the automated system with the sched- uling and control models to generate de- tailed production schedules efficiently. This process may require frequent access to data by the scheduling systems as production is monitored and the need to reschedule arises. Another issue is concerned with the analysis requirements that must be addressed in providing scheduling systems for auto- mated cells. These analytic requirements may demand the implementation of expert system technology to automate decision making. The analysis capabilities must be sufficiently robust not only to address se- quencing of product but also to evaluate the need to reschedule based on comparison of actual and planned performance of the system. Animation as a Formal Modeling Tool Animation of industrial systems driven by simulation models has been available for some time (Pritsker, 1986~. It has recently reached a high level of sophistication and availability to modelers because of the de- creased hardware costs and increased ca- pabilities of the hardware. High-resolution F. HANK GRANT animation models can now easily be cre- ated with a variety of simulation software tools and integrated with the simulation model. Significant research opportunities still ex- ist in characterizing animation as a formal modeling tool. These opportunities exist primarily in the application of simulation for system design. Animation of a physical system is basically another way of modeling that system. When building a design model that will include animation, significantly more effort must be expended, for the ca- pability needed to drive the animation is much greater than is needed to carry out analysis. More research is needed to characterize the content of an animation relative to de- sign objectives and also in interfacing that animation model efficiently with the anal- ysis model. Research is also needed to min- imize the added model development bur- den imposed by animation. A formal modeling procedure is needed for anima- tion, including procedures for integration with the design or analysis model. Additional research is also needed in de- fining the human interface requirements between the animation and the animation model builder. Most animation systems are somewhat limited in the amount of infor- mation they can display because of the size of the graphic screen available. Currently available features such as windows and di- vided screens help, but easier-to-use tools are required. Animation is also a useful tool in describ- ing the system's status in a real-time mode and triggering responses by the user to problems that exist in the system. It can be important in scheduling and control models to support the user in analyzing the dynam- ics of the proposed production schedule and identifying problems that might occur. However, production schedulers typically do not have the time required to review the variety of animations necessary to make ef- fective use of this tool. Perhaps additional
SIMULATION IN DESIGNING AND SCHEDULING MANUFACTURING SYSTEMS 147 work to reduce the data displayed and the review time required might make anima- tion an effective tool for production sched- uling environments. CONCLUSIONS This paper discusses the effective use of simulation both for design applications by engineering groups and for scheduling and control applications by production manag- ers. Simulation is a technology that Is intu- itively appealing to a wide audience anti that can provide much insight in predicting the performance of industrial systems. Its growth in the future will depend on the effective use of both hardware and software technology and the integration of the tech- nology with simulation software. New tools such as expert systems will help solve many of the problems. Cheaper and more pow- erful computer hardware will expand the availability of simulation technology to a wicler group of users. Simulation remains a fertile area for development and provides many challenges for researchers. REFERENCES Grant, F. H. 1986. Production scheduling using sim- ulation technology. Pp. 129-138 in Proceedings of the Second International Conference on Simulation and Manufacturing. Bedford, England: IFS Con- ferences, Ltd. Grant, F. H. 1987. Scheduling and loading tech- niques. In Production and Inventory Control Handbook, 2d ea., F. H. Grant and J. H. Green, eds. Falls Church, Va.: American Production and Inventory Control Society. Grant, J. W., and S. A. Weiner. 1986. Simulation Series, Part 4: Factors to consider in choosing a graphically animated simulation system. Industrial Engineering 18~8~:37-38 and 65-68. Haider, S. W., and J. Banks. 1986. Simulation Series, Part 3: Simulation software products for analyzing manufacturing systems. Industrial Engineering Errata 18~9~:87. Miner, R. J., and L. J. Rolston. 1983. MAP/1 User's Manual. West Lafayette, Ind.: Pritsker & Associ- ates, Inc. Musselman, K. J. 1984. Simulation: A design tool for EMS. Manufacturing Engineering 93~3~:117-120. Pritsker, A. A. B. 1986. Introduction to Simulation and SLAM II, ad ed. New York: Halsted Press and West Lafayette, Ind.: Systems Publishing Corpo- ration. Pritsker, A. A. B., F. H. Grant, and S. D. Duket. 1986. Simulation in real-time factory control. Pre- sented at a conference on Real-Time Factory Con- trol, May 13-14, 1986. Dearborn, Mich.: Society of Manufacturing Engineers. Reddy, Y. V., and M. S. Fox. 1982. KBS: An Artifi- cial Intelligence Approach to Flexible Simulation. CMU-RI-TR-82-1. Robotics Institute, Carnegie Mellon University. Standridge, C. R., and A. A. B. Pritsker. 1987. TESS: The Extended Simulation Support System. West Lafayette, Ind.: Pritsker & Associates, Inc.
