A Fault Diagnostic Approach for Underwater Thrusters Based on Generative Adversarial Network

Underwater vehicles are effective and irreplaceable vehicles for exploring and developing the underwater world, and their level of development has become an important symbol for evaluating a country’s maritime strength and comprehensive national power. Underwater thrusters are the main power components of underwater vehicles. The performance of fault diagnosis for underwater thrusters directly affects the safety and reliability of underwater vehicles. In this article, a fault diagnostic approach for thrusters combining data augmentation by generative adversarial network (GAN) and fault identification by improved temporal convolutional network (ITCN) is proposed. In order to solve the problem that real samples of propeller entanglement and transmission anomaly faults are difficult to be collected completely, a data augmentation theory relying on controllable generation of thruster signals is proposed according to the prior knowledge of these two fault modes. And a GAN for underwater thruster signals (GAN-UTS) is designed as an implementation of the proposed data augmentation theory. Considering the slightly insufficient accuracy of the original TCN in identifying some fault modes, a fault classifier based on ITCN is proposed to further improve the overall accuracy of fault identification. The experiments show that the diagnostic performance is significantly improved by jointly using the GAN-UTS-based data augmentation and ITCN-based fault identification. Compared with the original TCN-based fault classifier without data augmentation, the average F1 score of the proposed approach is raised by about 8.590%, fully proving the effectiveness of the approach.