Abstract: A fault detection method based on dynamic recurrent neural networks for uncertain nonlinear system is presented in this paper. A residual information is obtained by using adaptive neural networks observers, so the system faults can be detected rapidly. The uniformly ultimately bounded stability of closed-loop error system is guaranteed by Lyapunov stability theory. Finally simulation results of a fighter's control surface failure reveal the effectiveness of this method.

Key words: dynamic neural networks, adaptive observer, fault detection