Fault diagnosis research of UUV thruster based on sliding window and convolutional neural network

Abstract

In the face of complex ocean environment, underwater unmanned vehicle is prone to failure during navigation. With the application of UUV in many fields, the problem of fault diagnosis of UUV becomes more and more important. The traditional qualitative fault diagnosis method is short of manual experience and the analytical model-based fault diagnosis method is difficult to obtain high diagnostic accuracy. Data-driven fault diagnosis eliminates the dependence on dynamic model and it’s very suitable for UUV operating environment. The fault time series has great influence on the accuracy of fault diagnosis. A fault diagnosis method based on convolutional neural network and sliding window is proposed for time series problems. The data collected by the sensor are captured by sliding window, and then denoised by autoencoder. Finally, the convolutional neural network is used for diagnosis. The method is applied to the test of underwater unmanned vehicle “BYTUUV-I”. The test results show that the time of fault occurrence is taken into account in the process of fault occurrence. Compared with the traditional diagnosis model, the method is more suitable for practical application. The proposed method can solve the problem of fault detection and isolation in uncoupled fault mode, and the diagnosis accuracy is better than the traditional diagnosis algorithm.