Tailoring Automata for Fault Diagnosability

Abstract

Fault diagnosis in plants quite commonly utilizes measurements on limits of measurements being crossed. The fault-detection system attached to these limit detectors is thus clearly a discrete-event dynamic system. The plant, which operates in continuous time, a control system, which usually operates in discrete time and the supervisory control part with the fault detection part, which responds to the detected discrete-events, form together a hybrid system. The analysis shows clearly the ability and limitations of the approach. Any diagnostic tool will only get as input the detected limit and the direction of its crossing. Diagnostic systems must thus be tailored to make optimal use of the gradient components and their sign changes. Limits that can be used for fault diagnosis must thus be placed into domains where the signs of the gradient component are sensitive to particular faults or groups of faults. Domains with such properties are computed and serve as the basis for tailoring specific diagnostic systems.