Fault detection and isolation in Wiener systems with inverse model of static nonlinear element

Fault detection and isolation in Wiener systems via generation and processing of residual sequences is considered. It is assumed that a model of unfaulty Wiener system is known, and a static nonlinear subsystem is invertitile and can be modeled as a polynomial of finite and known order. In detail, we discuss application of both non-recurrent and recurrent models for detection of abrupt changes of a static nonlinear function / and linear subsystem parameters. A new definition of residual is introduced and a new fault detection and isolation method is proposed. To isolate all changes of system parameters, parameters of a residual generator model are estimated using recursive least squares identification techniques. Due to its tracing properties, the method based on the recursive least squares with exponential forgetting seems to be particularly suitable for fault monitoring and detection and isolation of slowly developing faults. For Wiener systems with nonlinear subsystems such that their inverse models can not be modeled by polynomials of a finite and known order, a neural network model of residual generator is proposed. Our approach is addressed to parameter changes in a mathematical model of the Wiener system rather than directly to faults of its specified components.