An Actuator Fault Diagnosis Combined Method Based on Intelligent Disturbance Observer

Abstract

For the fault diagnosis of attitude control system (ACS) of launch vehicle, an actuator faults detection and isolation method based on artificial neural network (ANN) disturbance observer is proposed. The diagnosis system can be divided into two parts, model-based observer and disturbance torque observer. The model-based observer consists of a set of unknown input observers (UIOs), which generate residual set to effectively isolate the faults. The disturbance torque observer introduces neural network which has good adaptability to estimate the trend of disturbance torque. The prediction results act as the disturbance compensation of the model-based observer, while guaranteeing the convergence of the observer and improving the robustness. The proposed method is tested in the ACS mathematical model of launch vehicle on MATLAB. The simulation results show that the diagnosis system with the combined method can effectively detect and isolate the actuator faults. Comparing with the sliding mode observer (SMO), the method based on disturbance observer gets higher accuracy and better real-time progressing performance, which helps to perceive the actuator faults at early stage.