Fault Detection and Identification of Underwater Thruster Without Fault Sample Based on Adversarial Variational Autoencoder

This article proposes a new underwater thruster fault detection and identification method based on adversarial variational autoencoder (AdvVAE). Adversarial training and variational autoencoders are combined, without any fault samples, only part control signals and their corresponding current signals and speed signals in the normal state of the thruster are needed for training, solve the problem that fault samples are difficult to obtain in large quantities. The reconstruction score obtained by the network is used to predict the degree of thruster fault, which realizes an end-to-end design. A real thruster pool experiment is constructed. The experimental results show that the proposed method is far superior to other algorithms in detection accuracy and efficiency, and the fault degree predicted by the model is highly consistent with the actual fault degree.