Adaptive Neural-based Fault Tolerant Control for Nonlinear Systems

Abstract

A neural network based fault tolerant control for unknown nonlinear systems is proposed. The faultless system is controlled by a nonlinear Internal Model Controller (IMC), where both the direct and inverse models of the plant are carried out by neural networks. Using the residual signal generated from the fault detection path, an extra neural-network fault compensation loop is introduced. This neural network is a two layer perceptron and the weights and bias are updated on-line by the modified-gradient approach, which tries to minimize the control error induced by the fault. In this context, a fault tolerant control scheme is obtained. This scheme is tested in simulation in a pH plant with good results.