Takuya Ishioka

Knowledge-based modeling of physical processes

Abstract We propose a knowledge-based modeling method to represent process dynamics for deriving control procedures and diagnosis rules. Our modeling method treats process characteristics as interactions of autonomous objects, corresponding to sub-components or units that form some local characteristic, and evaluates process dynamics as a sequence of equilibrium states of objects. To represent the characteristics of each autonomous object, we use structured frame-type data to represent state variables for each object, and rule descriptions to represent the numerical relations of autonomous objects with a combination of logical and numerical operations. We also show a method for constructing a control algorithm from evaluated data which uses fuzzy inference rules to interpolate the sequence of points around a fixed point. With fuzzy inference, complex characteristics with multiple inputs and multiple outputs can be easily presented in the form of simple IF-THEN rules

Trace visualization and analysis tool for supervisory control systems

Much time has been taken to examine supervisory control systems in their software engineering processes. Trace analysis has been one of the effective methods in proceeding with this examination since traces represent the actual flow of events in the systems. We have developed a trace analysis tool to help us with system understanding. The tool analyzes traces and extracts the kind of event, when it completed and which tasks it involved. It also visualizes the flow of events and the interaction that had taken place between tasks on the screen. The typical feature of the tool is to identify frequent event patterns such as periodic patterns by extending the algorithm for text data compression. We applied the tool to the engineering process of rail traffic control systems, and found the tool was effective in reducing the time for simplifying understanding of the dynamic behaviours of the system and finding problem areas.