A Fault-tolerant Controller for Unknown Nonlinear System Based on Neural Networks

A new neural-network-based fault-tolerant controller for nonlinear system is proposed. First a neural-network (NN) model is built for the unknown nonlinear system. The NN model is only trained during the initial period when the system is faultless. Then a fault-tolerant controller is designed. It consists of two parts, a main controller and a compensator. The main controller is designed for high performance. Using the residual signal generated from the differences between the nonlinear system and NN model, a NN-based compensation loop is constructed. Two controllers can be designed separately. The stability of the closed-loop is proved. The simulation results show the faulty system can be well compensated.