A Fault Detection Method for Uncertain Systems with Unmodelled Dynamics, Linearization Errors and Noisy Inputs

Abstract

This paper deals with the fault detection problem in uncertain linear/nonlinear systems having both undennodelling and noisy inputs. A fault detection method is presented which accounts forthe effects of noise, model mismatch and nonlinearities. The basic idea is to embed the umnodelled dynamics in a stochastic process and to use the nominal model with a predetemlined fixed denominator. This allows the system parameters to be estimated using ordinary least squares and also facilitates the quantification of the effect of model mismatch and linearization on parameter estinlation. Comparisons are made with alternative fault detection methods which do not account for model mismatch or linearization errors. The new method is shown 10 have vastly superiorperfonnance on a number of simulated systems. This improvement is a consequence of the fact that the new method explicitly accounts for the effects ofundennodelling and linearization errors in nonlinear systems.