Fault Diagnosis Method of Autonomous Underwater Vehicle Based on Deep Learning

Abstract

In order to solve the questions that autonomous underwater vehicle(AUV) can't accurately predict thrusters' fault only depend on sensor's information, which is caused by the effect of closed-loop control system, and shallow neural network can't well fit the complex nonlinear system, this paper proposes a new method for the fault diagnosis of AUV's thruster, based on Deep Neural Network(DNN) and Denoising Autoencoder(DAE).In this proposed method the difference between AUV's theoretical state value and measured state value are used as input signal. Considering the disturbance of AUV's working environment, when using DAE to pre-training the DNN, Gaussian noise was added in the input signal to simulate environment. The trained DNN network was finally used to detect the AUV's fault propeller components. The results show that the proposed method is more effective, accurately and robust than other traditional methods.