Fault diagnosis using dynamic finite-state automaton models

Abstract

This paper continues the development of an algorithm for fault diagnosis using a Finite-state Automaton (FSA) model. The plant is represented by a set of Finite-state Automaton Tables (FATs) that can be dynamically generated for normal input and for fault input conditions. A fault detection and isolation algorithm compares actual plant events with that predicted by the FATs. This paper defines fault diagnosability of the FATs, identifies some conditions for nondiagnosability, discusses guidelines for choosing the set of boundaries that helps define the FATs and provides a method for dynamically computing the FATs. We illustrate the application of the approach on a batch system.