Neural Network Models of Hammerstein Systems and Their Application to Fault Detection and Isolation

摘要

Detection of abrupt changes of both a static nonlinear function f and a linear dynamic subsystem via generation of residuals is considered. It is assumed that a neural network model of unfaulty Hammerstein system is known. A new fault detectionand isolation method is presented. In this method, sequences of residuals are processed to estimate parameters of a residual generator model and compute all changes of system parameters. Two different residual generator models are discussed. The first ofthem is a linear in parameters one and it can be estimated using the linear regression methods. The other model is of a neural network type. This model can also be used for systems with static nonlinearity that can not be modeled as a polynomial of finite and known order.