Sensor soft fault detection method of autonomous underwater vehicle

Operating in complex ocean environment, the condition monitoring and fault diagnosis of sensors have great impact on the safety of autonomous underwater vehicle (AUV). When the sensor soft fault of AUV is detected by the traditional method of observer based on the means of close-loop control and close-loop detection, the sensor measured value with fault information is fed into the input of observer, which will affect the output of observer and make it track with the measured value of sensor. Meanwhile, as the sensor information with fault is fed into the controller, the fault of sensor will be compensated by the adjusting function of controller. To solve the problem that it is difficult to detect the soft fault of sensors from the state data of AUV and the output of observer, the paper presents a novel diagnosis method to detect the sensor soft fault. Based on the means of close-loop control and open-loop detection, it constructed the open-loop state observer model using RBF neural network and take the observer residual and actual residual as the judicative residual and the basis residual respectively. The sensor condition is adjudged according to the different trend of the two kinds of residual mentioned above. In the process of training of RBF neural network state observer, the selective method of initial center is improved and the repeated selection of initial center is avoided. The experiment results show that the improved RBF learning algorithm has faster convergence speed and better training effect. The pool experimental results prove that the method of sensor soft fault detection is feasible and effective for the autonomous underwater vehicle.