Unsupervised fault diagnosis framework for underwater thruster system using estimated torques and multi-head convolutional autoencoder

Abstract

The thruster system is one of the most critical parts of an autonomous underwater vehicle. Feedback, propeller, motor and driver faults are four common field faults of thruster systems. This paper investigates an unsupervised framework based on torques and multi-head convolutional autoencoders for the mentioned faults. The proposed method utilizes an extended state observer to estimate the motor load. The propeller torque is estimated using a physics-guided neural network. Meanwhile, the relationship between the patterns change among torques (motor load and propeller torque) and different faults are analyzed. Then, to avoid feature aliasing, two independent encoders of the multi-head convolutional autoencoder extract the pattern features between the torques automatically. Finally, faults are detected and classified based on the various patterns of the extracted torque features. The sea trial data verified the effectiveness of the scheme. The results show that the method can accurately detect and classify five specific faults: current feedback fault, speed feedback fault, propeller loss, propeller winding and open-circuit fault.