A KNOWLEDGE BASED SYSTEM FOR PROCESS CONTROL

This global representation is able to incorporate all the relevant information for monitoring and supervising the manufacturing. In addition it should provide the operator interface with all the information required by the opera-tor. Finally, such an object-based architecture will constitute an open representational system, which is able to represent additional information and knowledge when found useful or necessary. ACKNOWLEDGEMENTThe author would like to thank Prof. Dr.-Ing. U. Rembold at the University of Karlsruhe for his advice and sup-port of this project.The work has been carried out at the Forschungszentrum Informatik (FZI), Dept. of Technical Expert Systems and Robotics (Director: Prof. Dr.-Ing. U. Rembold). It is part of the ongoing PROMEAT project (Quality Improve-ment in Meat Transformation Processes by Vision Sensors and Knowledge-Based Techniques), partly founded by the EU under BRITE/EURAM (BREU 4265).REFERENCES[Allen 83]J. F. AllenMaintaining Knowledge about Temporal IntervalsCommunications of the ACM 26, 832-843, 1983[Erman et al. 80]L. D. Erman, F. Hayes-Roth, V. R. Lesser, D. R. ReddyThe Hearsay-II speech-understanding system: Integrating knowledge to resolveuncertaintyComput. Surv. 12, 213-253, 1980[Forbus 84]K. D. ForbusQualitative Process TheoryArtificial Intelligence 24, 85-168, 1984[Hayes-Roth 85] B. Hayes-RothA Blackboard Architecture for ControlArtificial Intelligence 26, 251-321, 1985[Hayes-Roth 90] B. Hayes-RothArchitectural Foundations for Real-Time Performance in Intelligent AgentsReal Time Systems, Vol. 2, 1990[Jagannathan et al. 89]V. Jagannathan, R. Dodhiawala, L.S. BaumBlackboard Architectures and ApplicationsAcademic Press, 1989[Janson & Sutschet 86] A. Janson, G. SutschetA Process oriented Approach for Qualitative Modelling and AnalysisTEX-B Memo 07-86, 1986(2)Solution elements are generated and recorded on the blackboard by independent processes called knowl-edge sources.In order to develop a blackboard application suitable criteria for the sectioning of the overall knowledge into distinct knowledge sources has to be devised. It is possible to split the knowledge into two dimen-sions:•a logical dimension reflecting the various manufacturing phases, and•a functional dimension reflecting the basic tasks that must be carried out for a certain phase (qualityevaluation, control actions, etc.).This sectioning will lead to small knowledge sources that are easily implemented and maintained. More-over, additional reasoning features can be added in a straightforward way. A priority is assigned to each knowledge source so that the most important activities are performed first.(3)On each problem-solving cycle, a scheduling mechanism chooses a single knowledge source for execu-tion.The blackboard approach offers the possibility of applying the available knowledge in the absence of a known problem-solving model. An opportunistic scheduler decides on each problem solving cycle whose potential knowledge source is the most promising. The decisions made depend on the actual situ-ation (the actually performed process step) and the priority of the unsolved tasks.The process-model of the blackboard contains processes that have to be executed periodically and also those that have to be executed just the once. For the execution of periodic knowledge sources the sched-uler needs a timer. Real time operating system techniques are used to activate, deactivate, and suspend knowledge sources.The process control system brought with it several types of knowledge (heuristics, structural knowledge, etc.) however no method for integrating them. The blackboard architecture was designed to incorporate diverse knowl-edge into a single problem-solving system. Furthermore it offers the possibility to compensate for unreliability in the available knowledge and guarantee real time problem solving.The process control system is coupled to a powerful user interface, which is able to display (on-line) all relevant information such as process state and product quality to the operator. This interface seems to be essential for the user of the system. Moreover, the adopted knowledge bases should be modifiable by a user interface incorporating the required editors and parsers.3.1KNOWLEDGE REPRESENTATION OF THE PROCESS-MODELThis section describes the knowledge representation of the process-model where all the real processes involved in manufacturing are modelled. The behaviour of a dynamical system is described in a set of asynchronous pro-cesses that correspond through the change of parameters[Janson & Sutschet 86] [Forbus 84]. The modelling is characterized by the following features:•Description of processes in different levels of abstractionProcesses are aggregated from sub-processes.•Representation of temporal relationshipsEach process is given an interval in which the process is active. Temporal relations are defined between the intervals to specify whether a process is performed in sequence with another, in parallel, overlapping, or anything else[Allen 83].•Reduction of the domain of continuously varying parameters to discrete domains.The main steps involved in the manufacturing are described by objects providing a representation of the major processes to be carried out at the manufacturing plant. Each process can be further divided into its components corresponding to the various operations that have to be executed to complete the process; each component will be described by an object.•Storage of status and parameters of all processes in the plantEach process can be characterized by a set of parameters, their domains, and by the status of the process (active, inactive, suspended, etc.). The frequency and the method of data acquisition are of course deter-mined by the type of the parameter.•Parameter supervision and alarm handlingThe supervision system has to warn the operator when a parameter reaches a critical level. The operator has to solve the problem with the help of the system.•Quality evaluationUsually at the end of the process-steps quality data acquired by sensors and the inspection personnel has to be interpreted in order to detect quality problems. The process parameters have to be related to the quality of the products.•Fault diagnosisA diagnosis has to be done in order to determine the causes of observed quality problems. When a defectis found, corrective actions have to be proposed by the system.•Operator inputThe operator evaluation of the progress of the work has to be done continuously during processing. In addition the observations related to the various products have to be recorded at the end of each process step. The process control system has to give a complete image of all processes and all machines in the factory in order to support the operator.The analysis of these tasks shows the basic design goals of the system architecture:a)It should be as flexible as possible and be suited to a wide range of applications.b)It should be efficient enough to operate in real time.c)It should incorporate different types of knowledge and compensate for unreliability in the availableknowledge. Multiple reasoning methods contribute in finding a solution.d)It should deal with a variety of input data and has to integrate diverse information (multiple sensor inter-pretation). Noisy and unreliable data have to be compensated for.e)It should apply an opportunistic reasoning strategy that comes up with an evolutionary solution.The problem-solving model of blackboard systems provides a conceptual framework for organizing knowledge and a strategy for applying that knowledge appropriately. The following sections describe the design of a black-board application for process control.3.ARCHITECTURE OF THE PROCESS CONTROL SYSTEMThe architecture of the process control system extends and elaborates on the standard blackboard architecture [Erman et al. 80] as it has been seen as one of the best alternatives to match the properties of real time control sys-tems [Hayes-Roth 90] [Jagannathan et al. 89]. It consists of three main components [Hayes-Roth 85]:(1)All solution elements generated during problem-solving are recorded in a structured, global databasecalled blackboard.The blackboard includes an appropriate hierarchical representation (i.e. an image) of all the processes and sub-processes involved in manufacturing. This process-model is based on objects with appropriate attributes, describing the relevant features of each process.Furthermore, an appropriate object-oriented representation of the history and the current state of the manufactured products is included in the blackboard (product-model).In addition all sensors and machines in the plant are represented in the plant-model of the blackboard.These models maintain a complete image of all processes and states in the plant.A KNOWLEDGE BASED SYSTEM FOR PROCESS CONTROLThomas CordFZI – Forschungszentrum InformatikDept. Technical Expert Systems and RoboticsHaid- und Neustr. 10-14, D–76131 Karlsruhe, GermanyE-mail: cord@fzi.de1.INTRODUCTIONModern manufacturing is increasingly automated and specialized. Flexibility and control in manufacturing opera-tions are growing in the same extend as plant automation and instrumentation improve. Because of increasing automation and flexibility there are more opportunities for tracking and optimizing manufacturing operations for throughput, consumption of the materials used, machine time, product quality, etc. There is no general model of a manufacturing plant available because of increasing specialization. Different manufacturing approaches require different models (e.g. batch and continous manufacturing).On the one hand, the need to adjust operations in a timely fashion and the limits imposed on the control of manu-facturing processes present both, challenges and opportunities for AI approaches to model manufacturing opera-tions. On the other hand, real-time expert systems get an increasing amount of attention, at the moment. In real-time systems reasoning methods must be reactive to external events and have to abide to stringent timing require-ments. The application of knowledge based systems in process control is useful but difficult for a number of rea-sons. First, we are confronted with the fundamental problem of continuous reasoning over time-varying situations. Second, we have to face practical difficulties in achieving adequate speed of reasoning and response time given by the time constraints imposed by the embedded nature of the applications.This paper describes the blackboard architecture of a supervisory control and data-acquisition (SCAD A) system. It works as an operator support tool by diagnosing malfunctions of a production process and inspecting the prod-uct quality. The SCADA system provides the operator in a plant with relevant (filtered) process data and meets tough real-time requirements. It is shown how intelligent alarm processing is integrated into a large supervisory control and data acquisition system. Techniques which cover on-line processing of real-time data and intelligent alarm processing are discussed.The system is based on a blackboard architecture which integrates a sequence of problem solving methods and exhibits a characteristic incremental and opportunistic problem solving style. Multiple cooperating knowledge sources are available to perform the required functions. The activities of the knowledge sources are controlled by a scheduler. The knowledge about the manufacturing process and the parameters is stored in a global database which is called blackboard. A main objective was to produce an open architecture that can be easily tailored to suit different field requirements and is able to integrate different kinds of knowledge representations and reason-ing methods, too.2.TASKS OF THE SYSTEMIn most industrial applications supervisory control forms an important bridge between local control and plant-wide control. One can distinguish between a variety of tasks:•Supervision of local controllersThe parameters and set-points of traditional controllers have to be estimated and their performance has to be supervised.。