Neural Network Processing of Observer Generated Residuals in Fault Diagnosis of Time Delay Systems

Abstract

Listen

Translate article

The technique of observer-based fault detection is extended for systems with significant delays and distributed parameters by means of functional models using the so-called anisochronic state space. This approach: reduces considerably the number of state variables to be monitored and major part of these variables may be selected as measurable system outputs. As a special tool for the observer design Ackermann formula is extended for a functional synthesis of observer feedback design compensating the system delays. The observer-generated residuals are viewed as signals processed by neural network predictor. Dynamical application of this predictor enables fault diagnosis on the basis of prediction criterion. General design procedure is demonstrated on an application in monitoring a laboratory-scale thermal system with heat exchangers.