A fast terminal sliding mode fault-tolerant control strategy based on adaptive neural network sliding mode observer was proposed for the possible failure of electromechanical servo system. The neural network was introduced into the adaptive sliding mode observer to estimate the fault, so as to improve the accuracy of state estimation and fault diagnosis. An active fault-tolerant controller was designed by using the state estimation value of the observer, combining the parameter adaptive technology and the fast terminal sliding mode control method. The parameter adaptive rate was designed to estimate the parameter uncertainty, and the feedforward compensation technology was used to compensate the fault and parameter uncertainty. A robust term with adaptive gain was designed to overcome the disturbance of unknown upper bound. Using Lyapunov theorem, it is proved that the proposed control method can achieve bounded stability of the system. A large number of simulation and experimental results verify that the controller has good fault tolerance, control accuracy and response speed in case of system failure.