Toshihiko Niinomi

Refinements of approximation automata for synthesis of supervisory controllers for hybrid systems

The problem of supervisory controller synthesis is considered for a class of continuous-time hybrid systems (plants) in which the continuous-state dynamics are selected by a discrete-valued input signal, and output events are generated when the continuous output signal encounters specified thresholds. A discrete-state supervisor controls the hybrid plant by switching the discrete input when threshold events are observed. The objective is to synthesize a nonblocking supervisor (in the ω-language sense) such that the sequential behavior of the closed-loop system is contained within a specified set of admissible behaviors. This problem can be converted into a supervisor synthesis problem for a controDed discrete event system (DES) and solutions may be obtained by constructing finite-state generators that accept outer approximations to the exact DES model of the hybrid plant This paper presents the construction procedure for the approximating automata and a demonstration that increasing the order of the construction computations by appropriate amounts leads to refinements of the outer approximations.

An application of qualitative reasoning on flow topology: Reasoning without qualitative equations

Abstract There are several levels of abstraction for qualitative modelling and reasoning. After two years of effort in attempting to use current levels of qualitative reasoning for the process-fault diagnosis of an existing chemical-processing plant, we have found that the level of abstraction of the reasoning, and the granularity of the model, are too precise for our purpose. It is almost impossible to obtain a qualitative model of an existing process with consistent constraints, but it is also unnecessary to generate the dynamic causal path for such detailed variables. We have developed a more-abstract level of qualitative reasoning. The reasoning operates directly on the topology of the process flow, which is in turn obtained from a process-flow diagram. Qualitative equations used in previously-proposed qualitative reasoning systems are not needed. This level of qualitative reasoning seems more appropriate to the large-scale systems found in most existing process plants. In the reasoning, critical conditions such as set fixed over time or set free are made explicit for given flows. The inference engine for this flow-level reasoning is applied to generate an intermediate form of diagnostic knowledge, expressed as trees, where the given root event is developed for both cause and effect directions.