Fault diagnosis for nonlinear systems via neural networks and parameter estimation

In this paper we present a novel approach for the fault detection and diagnosis of nonlinear systems described by NARMA models. Firstly a known nonlinear system is considered, where an adaptive diagnostic model incorporating the estimate of fault is constructed. This has led to a new filtering design that either minimizes the residual entropy or controls the shape of the probability density function (PDF) of the residual. The diagnostic algorithm is then developed which produces the estimate of the fault so that the error between the system output and that of the model is minimized. Unknown nonlinear systems are then studied using a feedforward neural network trained to estimate the system under healthy conditions. Taking the trained neural network as the neuro-model of the system, similar detection and diagnostic algorithms to that of known systems are obtained.