Neural-network-based fault-tolerant control of unknown nonlinear systems

A neural-network (NN) based fault-tolerant control for unknown nonlinear systems is proposed. The faultless system is controlled by an NN-based one-step-ahead controller, which is designed using a modified-gradient approach. Using the residual signal generated from the fault-detection path, an extra NN-based fault-compensation loop is introduced. This neural network consists of two-layer perceptrons and the weights are again updated by the modified-gradient approach. In this context, a fault-tolerant control scheme is obtained. The stability of the closed loop is discussed. It has been shown that the closed-loop system so formed is locally asymptotically stable for the nonlinear case, and is globally asymptotically stable when the system is linear. The simulated results have shown that the faulty system can be well compensated