Abstarct: The fault is an inseparable part of the industrial system. In this thesis, a robust actuator fault diagnosis method baxsed on neural observer for a class of nonlinear systems is presented. Unlike most existing work that assume system states are available, the proposed method does not depend on the presence or absence of system states. Unknown functions and system states are estimated by using a neural observer, Neural network update rules calculated and the stability of the observer is proved using Lyapunov method. Finally, simulations on a single-lixnk robot are performed and the results show the effectiveness of the proposed observer.