Design of the neural-network-based adaptive sliding mode observer for fault reconstruction

A novel actuator fault reconstruction scheme based on neural network and adaptive sliding mode observer is presented for a class of nonlinear systems with uncertainty. To improve the accuracy of actuator fault reconstruction, the radial basis function neural network is applied to approximate the uncertainty on line. An adaptive algorithm is designed to modify the gain of the controller by considering the condition of unknown upper bound of fault. According to the output of the designed sliding mode observer, the equivalent output error injection method is employed to reconstruct actuator fault using the characteristics of the sliding mode observer. Finally, the feasibilities and effectiveness of the proposed approach is illustrated by the simulations on the three-axis stability spacecraft attitude control system.