Thruster fault feature extraction for autonomous underwater vehicle in time-varying ocean currents based on single-channel blind source separation

Abstract

This article presents a novel thruster fault feature extraction method for autonomous underwater vehicle in time-varying ocean currents. The novel method is based on the combination of ensemble empirical mode decomposition and independent component analysis, which is a universal single-channel blind source separation method. When using ensemble empirical mode decomposition to decompose the autonomous underwater vehicle surge speed signal into a set of intrinsic mode functions, the original number of intrinsic mode functions is big, making the independent components that are separated and reconstructed from the intrinsic mode functions based on independent component analysis contain many false components. To reduce the original number of intrinsic mode functions, a method combining wavelet decomposition with empirical mode decomposition is proposed. When extracting fault feature value from independent components based on modified Bayes’ classification algorithm, the results show that the difference of fault feature value and noise feature value is small, and the ratio of fault feature value and noise feature value is small as well, which make it difficult to distinguish fault feature value from noise feature value. To increase the difference and the ratio, a wavelet detail component–assisted feature extraction method is proposed. The method combining wavelet decomposition with empirical mode decomposition and the wavelet detail component–assisted feature extraction method together constitute a novel thruster fault feature extraction approach which is suitable for autonomous underwater vehicle in time-varying ocean currents. The effectiveness of the proposed methods is verified by the pool experiments of the experimental prototype.