Simulation Based Decision Support for Manufacturing ... - CiteSeerX

Published in Journal of Advanced Manufacturing Systems, Volume 3 Number 2, December 2004, pp 115-128

Simulation Based Decision Support for Manufacturing System Life Cycle Management Leo J. De Vin, Amos H.C. Ng and Jan Oscarsson Centre for Intelligent Automation, University of Skövde, Skövde, Sweden Box 408, 541 28 Skövde, Sweden. E-mail leo/amos/[email protected] Abstract Previous research has highlighted the role of virtual engineering tools in the development of manufacturing machinery systems. Simulation models created for this purpose can potentially be used to provide support for other tasks, such as operational planning and service and maintenance. This requires that the simulation models can be fed with historic data as well as with snapshot data. Furthermore, the models must be able to communicate with other business software. The paper describes how simulation models can be used for operational production planning and for service & maintenance support. Benefits include a better possibility to verify production plans and the possibility to monitor and service manufacturing machinery from remote locations. Furthermore, the expanded and continuously updated models provide a good tool to study the effect of for instance planned new product introduction in existing manufacturing systems. The paper also presents directions for future research. One ambition is to add AI tools to the system so as to develop a semi-autonomous system for decision support. Keywords: Simulation, Decision making, Manufacturing

1 INTRODUCTION Key factors to remain competitive in the manufacturing business are the effective and efficient operation of installed manufacturing systems, as well as the ability to quickly and readily reconfigure these systems in the case of for instance new product introduction. Manufacturing systems usually consist of a number of people, machines and handling devices. These systems are sufficiently complex and sophisticated that (re)design, operational production planning (scheduling) and machine service & maintenance are intricate specialist tasks. As a result, these tasks increasingly require adequate decision support systems. These systems need to be able to support the decision maker with accurate information, based on historical data, actual status and expected trends/events. In this area, a number of issues can be identified, such as: (i) the need to provide manufacturing system lifecycle decision support, (ii) the need to collate, manage, and make available information and data throughout the system’s lifecycle, (iii) the need for remote monitoring/service capabilities due to globalisation or centralisation of functions within a plant. Apart from (in some cases) a physical distance between the decision maker and the system at hand, the availability and management of reliable information is a critical issue whereas on the one hand, there may be information overload whereas other information may be hard to find/retrieve. Furthermore, the type of information needed may vary, depending on the type of decision that has to be made. Previous and ongoing research at the Centre for Intelligent Automation has highlighted the potential use of simulation as a decision support tool in different manufacturing system life-cycle phases. This paper describes the ambition and first efforts to integrate tools and methods developed so far with each other. By adding a self-learning capability, the overall future solution is expected to be a semi autonomous decision support system.


The research aims at developing a semi-autonomous decision support system for life-cycle management of manufacturing systems, and this system is planned to be an integral part of the recently defined research programme ‘information fusion from databases, sensors and simulations’ at the University of Skövde. This research programme aims at developing decision support systems for a relatively wide scope of applications and with fusion of information from a variety of source types. Other research in this area is often either limited in scope (e.g., related to defence tasks only) or limited in information sources (e.g., from sensors only). A key element in the proposed solution is the fusion of information regarding the system’s history and previous operation (stored in for instance databases), the actual status (obtained through monitoring), with information regarding the future (such as expected trends/events or information explored through so-called virtual scenarios during simulation runs). 2 INFORMATION AVAILABILITY AND FUSION Information Fusion (IF) encompasses the theory, techniques, and tools conceived and employed for exploiting the synergy in the information acquired from multiple sources (sensors, databases, information gathered by human, etc.) such that the resulting decision or action is in some sense better (qualitatively or quantitatively, in terms of accuracy, robustness, etc.) than would be possible, if these sources were used individually without such synergy exploitation 1. An example of IF in manufacturing is the fusion of information from multiple sensors 2. In this paper, with IF we mean the fusion of information from the past operation of a manufacturing system (e.g., stored in databases), from the present (e.g., sensor signals, machine status), and from the future (in particular, predictions obtained through simulations). However, in order to be able to execute IF, the information must be present and accessible. Furthermore, the user or system that needs that information also needs to know that it is available, as well as where, in which format, and how to access it. In many cases, information is missing or not readily available (for instance, tacit knowledge about a particular machine, stored in the head of the operator). In principle, three situations in decision making for product realisation can be distinguished 3, 4: 1. Information about the environment and the objectives is complete 2. Information about the objectives is complete, but information about the environment is incomplete. 3. Information about the objectives and about the environment is incomplete. Situations 2 and 3 give rise to the need for emergent synthesis 17, however one should always try to strive after arriving in situation 1 3. 3 VIRTUAL MANUFACTURING (VM) Virtual Manufacturing (VM) has been around since about 1990 5, 6. The use of VM, or manufacturing simulation, in industry has increased ever since and even dedicated study programmes have emerged to meet the industrial demand for qualified simulation engineers 7. Modern simulation tools offer good visualisation techniques. This contributes to their potential role as communication tools 8. The simulation technologies considered in this paper are Discrete Event Simulation (DES, or DEVS, Figure 1) and Computer Aided Robotics (CAR, Figure 2). DES tools are normally used to study production flow, although their use can be extended to other areas than manufacturing 9. CAR tools are normally used to study movements of industrial robots and so on, with collision avoidance and off-line programming as important applications. Ergonomic simulation, for instance posture analysis for workplace design, is a technology related to CAR.


Figure 1: Snapshot from DES.

Figure 2: Snapshot from CAR. 4 VM AND INFORMATION SUPPORT IN BUSINESS SYSTEMS Typically, information sources for conducting various VM activities are not one single specific source, but rather all the different technological and business information systems that are used in a company. The integration of these sources is seldom an out-of-the-box-solution but most often highly customized solutions, engineered by specialists. Information that is valuable in VM has the same problem as any other piece of information; it is often bound to the structure used by the software in which it was created. The lack of possibility to use a common format for all information, not depending on any CAE tool, poses problems. There are three typical scenarios: (i) The information exists and it can be used directly or converted to a format that can be used, (ii) The information exists but it cannot be translated into a format that can be read by the VM tool and therefore the information has to be recreated in a new format, (iii) The information is not available and has to be created or collected. The lesson learned from this is that if the information exists, it would almost be foolish not to have it presented in a format that is more or less neutral. The integration with information sources e.g. ERP systems, PDM systems and MRP systems, is a necessity. The integration builds on three pillars 10. One of them is structure, which is discussed above, the other two being context and semantics, see Figure 3. As in every type of information exchange, all three of them have to be considered and adequately managed. It is shown that information is only valuable if the context is presented correctly, the semantics are understood, and the structure can be handled by the receiving system. STEP AP 209 (ISO 10303) for instance offers a standard for importing CAD data into simulation models.


Figure 3: Integration of VM into business systems. 5 VM FOR MANUFACTURING SYSTEM DESIGN The VIR-ENG project 11, 12 has highlighted the potential role of virtual engineering in machine system design. The main objective of the project was to develop highly integrated design, simulation and distributed control environments for building agile modular manufacturing machine systems which offer the inherent capacity to allow rapid response to for instance product model changes. In the project, a component based paradigm was adopted for both hardware and software development. In essence, machine systems including their control system are developed in a virtual environment and subsequently implemented as a physical system. Software tools, solutions and technologies include the simulations tools QUEST (DES) and IGRIP (CAR). Furthermore, OLE for Process Control (OPC) has been used. OPC is an open interoperability industrial standard for sharing manufacturing information in an enterprise-wide manner. It is based on the Microsoft technologies of OLE (Object Linking and Embedding), COM (Component Object Model) and DCOM (Distributed Component Object Model). It provides ‘plugn-play’ connectivity and interoperability between disparate automation devices, systems and software, from the shop floor to enterprise wide systems. In VIR-ENG, the IEC 1131-3 was adopted. The main environments developed in the VIR-ENG project as shown in Figure 4 are the highly integrated Modular Machine Design Environment (MMDE), Control System Design Environment (CSDE) and Distributed Run-time Environment (DRE). An Infrastructure and Integration Services (IIS) environment based on the component object-based computing platform provides all the ‘pipes and plumbing’ for the information integration within the VIR-ENG environment. The integration services are applied not only to the VIR-ENG environment but also support enterprise wide information integration.

Figure 4: VIR-ENG environments


6 VM FOR SERVICE AND MAINTENANCE Service and maintenance is an intricate specialist task and machine builders often have to provide service at short notice. Machine builders would benefit enormously from extended possibilities to monitor and diagnose equipment operating at distant locations – both for condition-based preventive maintenance and for diagnostic purposes before flying in qualified maintenance personnel and spare parts. Even within a single plant, the possibility to monitor manufacturing equipment from a central service and maintenance office has obvious benefits. Most condition monitoring applications in manufacturing are limited to sensor information fusion for applications such as tool-wear monitoring 13, 14 or spindle bearing monitoring 2, 15. Examples of sensor information are vibrations and oil contamination 16, temperatures and electric currents. Whilst this approach is extremely useful for preventive service and maintenance, it is limited to some features of individual machines rather than being applicable to machines as part of a machine system. The approach proposed in this paper builds upon the tools and techniques from the VIR-ENG project and continued research. A key element in this approach is the seamless integration between simulation model (which is a hybrid DES/CAR model) and physical equipment. This integration makes it possible for instance to: • Study failure modes and their effects during machine system design through simulations in which certain disturbances/faults are emulated. • Monitor the operation of the machinery system on-line, which facilitates both supervision by humans and data-acquisition. • In the case of breakdowns, retrieve control code execution, sensor information from a temporal database and carry out a ‘replay’ of the machinery systems recent history in the simulation model. • Develop, test and upload temporary control code in the case of temporary reconfiguration due to machine service activities (either preventive or corrective maintenance). An example of networked connections between the real process and the simulation model is shown in Figure 5.

Figure 5: System architecture for a machine service support system


Within the MASSIVE project, the above-said concepts are being realised through the design and implementation of an integrated software environment called MSSS (Machine Service Support System), as an extended part of the machine design and control environments developed in VIRENG. A system architecture that defines various components of MSSS and their interactions has been preliminarily designed and is illustrated in Figure 5. MSSS is essentially a remote data acquisition and analysis system. Therefore, it is obvious to see that an advanced data acquisition, pre-processing and management framework is the foundation for all other functions. The data acquisition system can be remotely configured so that specified parameters, machine process variables, discrete-event signals can be acquired in prescribed time intervals and sampling rates. Configurations for routine periodic data logging can also be selected for day-to-day monitoring. All configurations to the data acquisition components are done through the Web methods provided by the XML Web services using the user interface functions provided by Scenario Manager. For continuous visual monitoring or in the case of a machine failure (breakdown), MSSS users can use the historical data saved in the database to carry out “playbacks” to investigate the recent history of the machine system and current status using the corresponding simulation models. In these cases, animations are driven by the historical data acquired, but simultaneously, the reference process models are used to generate the nominal dynamic response of the system with the input data from the historical data. The output data generated by the simulator and from the collected historical data can be visualised and compared using various data analysis and residual analysis techniques. The data visualisation features accomplish the 3-D animation for presenting useful “non-animated” data like electric current and voltage produced both from the simulator and the collected data as an additional means for assisting any monitoring and diagnostic tasks. Fault alarms can be generated by the diagnostic agents, for instance, if a residual signal is evaluated to exceed a certain threshold; but more advanced fault detection algorithms can be easily incorporated into MSSS. Control system verification is a desirable feature that the simulation models can be used to verify the control programs for testing and verification during the machine system design, development, commissioning or re-configuration stage. While this functionality has been explored in-depth during the VIR-ENG project, the focus of MASSIVE is to extend the research outcomes from VIR-ENG to support verification of control programs that are developed/modified to cope with maintenance service tasks. 7 VM FOR OPERATIONAL PLANNING Operational production planning is another specialist task. Typically, the planning horizon is about a week but the planning activity must be executed several times per week or even daily, due to disturbances, changes in production orders and so on. Although most ERP/MRP systems incorporate planning algorithms, the possibility to address dynamic effects due to disturbances, rush orders and so on is often limited. Usually, operational production planning is a task for experienced engineers. With their experience, they can arrive at a feasible planning solution, but usually, there is no time to compare alternative solutions. Hence, usually no optimisation takes place. Another problem is the lack of accurate information regarding the actual status of the production facility. Furthermore, parameters such as mean time between failures (MTBF) or mean time to repair (MTTR) are often treated as static parameters, whereas in reality, they are dynamic. This means that the production plan for the next period is often drawn up with poor indata or uncertainties. For this task, a simulation-based approach is being developed. Key elements are the online monitoring of the status of the production facility, such as buffer levels, machine status and so on, and information exchange with ERP/MRP systems (Figure 7). By carrying out the planning activity with the help of a DES tool, the following can be achieved: • The starting point of the planning is the actual status of the production facility, based on socalled ‘snapshot information’ obtained through online monitoring. The availability of snapshot


information also means that rescheduling in case of breakdowns or rush orders can be carried out swiftly and readily. • The snapshot information can be fused with information about the previous operation of the production facility. This means that historical information is used, and that parameters are dynamic, with a certain statistical distribution, instead of static. Furthermore, expected repair times can be based on the type of disturbance instead of just on the average repair time of a certain machine. • Different planning solutions can be tested and compared during simulation, furthermore different scenarios (e.g., order forecasts or technical planning solutions) can be simulated. This allows for optimisation, or for selecting a robust planning solution (e.g., a solution that is less sensitive for variations such as order variation or disturbances). Furthermore, most DES tools have powerful visualisation possibilities, making them very suitable for training purposes, with the online connection providing realistic training conditions. Figure 6 shows the system architecture of the Integrated Simulation Based Planning and Scheduling (ISBP) environment, currently under development within the SimPlan project at the University of Skövde 17, 18. The system design contains the following components: a model of the production system implemented in the Quest simulation package, a database, a middleware developed to act both as the actual configuration tool, the user interface, and as the gateway between the simulation package, the production monitoring system and the ERP-systems. The middleware is implemented in C# and thus utilizing the Microsoft .NET technology. Furthermore, the middleware features a component design in order to be generic and flexible. This generic design ensures the possibility to replace various components such as for instance the simulation package or the database. This design makes it adaptable and the actual implementation is currently being tested under laboratory conditions and later on, two test cases will be conducted in different production contexts. The first test case will be performed at a dairy plant and the second at a major domestic appliances manufacturer. The communication from the middleware to the external systems is handled via standard interfaces. The integration with Quest is handled via a socket server that is running as a component in the middleware. Data is passed to the server via a socket data stream and then further processed in the middleware. The ERP-integration is handled via ODBC and via SOAP/XML. Finally, the communication with the production system is handled via OPC, which is the de facto standard when interacting with Windows applications. OPC communication is usually done via an OPC-client or by a programming API. The SimPlan implementation uses a C# API, supplied by OPC-labs. This design choice further sustains the concept of the architecture being generic and flexible.

Figure 6: System architecture for simulation-based operational planning


8 VM FOR LIFE CYCLE DECISION SUPPORT Whilst the topics discussed in Sections 5 through 7 are interesting areas for simulation and information fusion in their own right, there is a need to address these issues in an integrated manner. For instance, machine maintenance must be taken into account during planning, and regular failure to meet planning objectives may prompt service or reconfiguration. Thus, there is a need to take a machine system life cycle perspective with a focus on design and operation as the main phases within the GERA model 19, Figure 7.

Figure 7: GERA life cycle phases Figure 8 shows the simulation based parallel development of products and the manufacturing system. Ideally, the development of the manufacturing system starts during product development, so as to tailor product and production system requirements and properties to each other in a concurrent engineering fashion 20. The backbone of such parallel processes is the common information platform, which provides information to the two processes and stores information generated. It also provides information to the simulation tools and stores simulation results. The operation phase includes continuous improvement activities. Reconfiguration activities can be necessary in the case of new product introduction (NPI), normally meaning the introduction of a new variant within a product family.

Figure 8: Parallel development processes. When the simulation model is to be used in the runtime phases, it must be updated continuously. This is an advantage in case of reconfiguration of the manufacturing system, as it means that an updated model is readily available to build upon. Current practice is unfortunately that one often builds a new model, as updating the old model (if it can be retrieved at all, including all relevant documentation) can be more time-consuming than building a new model from scratch. In terms of the different situations that can occur regarding the completeness of information about the environment and objectives as discussed in Section 2, the exploration of future scenarios can be seen as an effort to move from a situation 2 to a situation 1 regarding the future environment. Although such exploration of future scenarios never can give a complete picture of the future environment, at least it means that the most likely conditions have been explored and that the decision maker can readily address the most likely scenarios.


9 FUTURE WORK Whilst VM, through the extended use of simulation models, offers a powerful tool for decision support in the runtime phase of a manufacturing system, decisions regarding which simulations to run, identification of abnormal conditions and so on, remains a human task. However, through the use of soft computing techniques such as genetic algorithms and artificial intelligence, these tasks can be supported as well. 9.1 ARTIFICIAL INTELLIGENCE IN MANUFACTURING Examples of artificial intelligence (AI) are artificial neural networks or neurofuzzy computation. Although its potential use in manufacturing has been identified long ago (e.g., by Monostori 21), industry has only recently started to embrace these techniques for use in their manufacturing systems 13. In the context of VM, AI can be particularly useful to: • Achieve a better correspondence between VM model and real system. For instance, simulated cycle times tend to deviate from real cycle times. If the simulation model ‘learns’ from the real system so that its behaviour resembles normal operation better, then it becomes easier to detect abnormal behaviour of the real system. • Let the VM model ‘learn’ which scheduling rules tend to be more successful when exploring different planning solutions under different conditions. This means that depending on the conditions, certain promising solutions are simulated first. Instead of manually suggesting alternative production schedules, genetic algorithms can be used to generate these alternatives. • Let the VM model ‘learn’ how good various planning solutions are in practice by comparing simulation results for a selected solution with the results as they occur in practice. Furthermore, AI can be used to capture tacit knowledge by observing the actions/decisions of experts, however this is not specific for VM but rather a more traditional and more general application of AI. 9.2 TOWARDS A SEMI-AUTONOMOUS DECISION SUPPORT SYSTEM The long-term goal for the research presented above is to develop a semi-autonomous decision support system. In most cases, decision making is a ‘man-in-the-loop’ activity requiring adequate support 3. The decision support system may suggest simulation experiments to be carried out and assist in the automatic gathering of input data and information. It should be noted that design of simulation experiments is a specialist task, and that the information need to carry out a simulation experiment may vary. For instance, regular scheduling, which could have minimisation of costs as an objective, requires different information than emergency rescheduling with minimisation of job lateness as objective. In some cases, the system may carry out some actions by itself. This can be the case for relatively straightforward actions such as correction of normal machine drifts. Another example is a case in which there is no time for a ‘man-in-the-loop’ decision. In military jargon, this corresponds with an OODA (observe-orient-decide-act) loop that needs to be completed very quickly. An example can be a (impending) tool or bearing failure requiring an immediate stop of a machine. As a first step, industry based studies are being carried out to identify industry’s prioritised need regarding support in decision making in the operation phase of manufacturing systems. This is carried out in parallel with technical development in this area and refinement of tools and techniques developed for the design phase. The result of these studies will be a ‘handbook for simulation based decision support’, which will be implemented in a demonstrator after verification in industry.


10 CONCLUSIONS Simulation models as used in virtual engineering during development of manufacturing systems can be used during the operation phase of manufacturing systems as well. In order to fully benefit from this, the simulation model must be connected to the physical system and other business software. In this way, information regarding past operation and current status can be fused with information regarding possible future operation, explored through virtual scenarios. The overall result can be used for decision support in for instance operational planning or service and maintenance. In this way, simulation serves as a tool for arriving at a situation in which the future environment is perhaps not completely known, but in which one can readily address the most likely scenarios in an adequate manner. Artificial intelligence can play a role in virtual manufacturing by improving simulation models or by offering better decision support. Extending the use of simulation models from the design phase to the operation phase also has advantages when new products are to be introduced or the manufacturing system needs to be reconfigured. 11 ACKNOWLEDGEMENTS This paper was prepared under the sponsorship from the Swedish Agency for Innovation Systems and from The Knowledge Foundation in Sweden. The authors would like to thank Thomas Karlsson, Michel Nilsson and Petter Solding for their contributions to this paper, and Professor emiritus Gunnar Sohlenius (RIT, Stockholm) for valuable suggestions regarding the preparation of the manuscript. 12 REFERENCES 1. Dasarathy, B.V., 2001, Information Fusion – What, Where, Why, When, and How?, Information Fusion, 2:75-76. 2. Salvan, S.M.E., Parkin, R.M., Coy, J., Jackson, M.R., Li, W., 2002, Condition Monitoring and Location of Multiple Roller Bearings Using Three Sensors, Proc. Mechatronics 2002, Univ. Twente, NL, 998-1007. 3. Sohlenius, G., Fagerström, J. and Kjellberg, A., 2002, The Innovation Process and the Principle Importance of Axiomatic Design, 2nd ICAD, MIT, Cambridge, US. 4. Ueda, K., Markus, A., Monostori, L., Kals, H.J.J. and Arai, T., 2001, Emergent Synthesis Methodologies for Manufacturing, Annals of the CIRP, 50/2, 535-551 5. Onosato, M., Iwata, K., 1993, Development of a Virtual Manufacturing System by Integrating Product Models and Factory Models, Annals of the CIRP, 42/1:475-478. 6. Mills, J.J., 1989, Virtual Manufacturing: A New Concept in Automated Design Aids, 3rd Intnl Conf on CAD/CAM robotics and factories of the future, Southfield, Michigan US, 115-118. 7. Oscarsson, J., Hermansson, A., Ujvari, S., De Vin, L.J., 2002, A Curriculum in Simulation Engineering: The Skövde Experience, CIRP International Manufacturing Education Conference CIMEC2002, Univ. Twente, NL, 373-382. 8. Iwata, K., Onosato, M., Teramoto, K., Osaki, S., 1997, Virtual Manufacturing Systems as Advanced Information Infrastructure for Integrating Manufacturing Resources and Activities, Annals of the CIRP, Vol 46/1:335-338. 9. Urenda Moris, M., Eriksson, P.T., De Vin, L.J., 2004, Introducing Discrete Event Simulation for Decision Support in the Swedish Health Care System, WMC2004, San Diego, US, 48-53. 10. Karlsson, T., Rogstrand, V. and De Vin, L.J., 2004, Verifying Manufacturing Requirements Using Tools for Digital Plant Technology, 37th CIRP International Seminar on Manufacturing Systems, Budapest, Hungary, 291-296. 11. Adolfsson, J.K., Ng, A.H.C., Moore, P.R., 2000, Modular Machine System Design Using Graphical Simulation, 33rd CIRP Int Sem on Mfg Sys, KTH, Stockholm, Sweden, 335-340.


12. Olofsgård, P., Ng, A., Moore, P., Pu, J., Wong, C.B. and De Vin, L.J., 2002, Distributed Virtual Manufacturing for Development of Modular Machine Systems, Journal of Advanced Manufacturing Systems, Vol 1 No 2, 141-158 13. Dasarathy, B.V., 2003, Information Fusion as a Tool in Condition Monitoring, Information Fusion, 4:71-73. 14. Kandilli, I., Ertucc, H.M., Cakir, B., 2002, Real-Time Tool Wear Monitoring Using Neural Networks, Proc. Mechatronics 2002, Univ. Twente, NL, 1018-1027. 15. Li, B., Chow, M.-Y., Tipsuwan, Y., Hung, J.C., 2000, Neural-Network Based Motor Rolling Bearing Fault Diagnosis, IEEE Transact. on Ind. Electr., 47/5:1060-1069. 16. Carnero, M.C., 200x, Selection Of Diagnostic Techniques and Instrumentation in a Predictive Maintenance Program. A Case Study, Decision Support Systems, in press. 17. Nilsson, M, Solding, P. and De Vin, L.J., 2004, A System Architecture for Integrated Simulation-Based Production Planning and Scheduling, Mechatronics 2004, Ankara, Turkey, 815-824. 18. Solding, P., Nilsson, M., Eriksson, P. and De Vin, L.J., 2004, Structure for Data Management in Simulation Based Planning Activities, submitted to 37th CIRP International Seminar on Manufacturing Systems, Budapest, Hungary, 193-198 19. IFIP-IPAC, Task Force, 1998, GERAM Generalised Enterprise Reference Architecture and Methodology, Version 1.6.1. 20. Sohlenius, G., 1992, Concurrent Engineering, Annals of the CIRP, 41/2:645-655. 21. Monostori, L., 1993, A Step Towards Intelligent Manufacturing: Modelling and Monitoring of Manufacturing Processes with Artificial Neural Networks, Annals of the CIRP, 42/1:485-488.