Fault feature extraction, feature fusion, and severity identification approaches for AUVs with weak thruster faults

Abstract

Fault feature extraction, feature fusion and severity identification approaches for autonomous underwater vehicles with weak thruster faults are studied in the article. The traditional method uses the modified Bayes algorithm for fault feature extraction from different signals, then the fault features are fused through the Dempster-Shafer evidence theory, and finally, the severities of the faults are obtained by the grey relation analysis method through the fused features. But for weak thruster faults, in the stage of feature extraction, it exists the problem that the ratios of fault eigenvalues to noise eigenvalues of the extracted features are low. In the stages of feature fusion and severity identification, it exists the problem that the errors of the identification results obtained from the fused fault features are not satisfactory. Aiming at the above problems, the smoothed pseudo Wigner-Ville distribution together with the modified Bayes method is presented for feature extraction for weak faults. The feature-level fusion together with the decision-level fusion method is presented for feature fusion and severity identification for weak faults. The experimental prototype pool experiments verify the effectiveness of the approaches presented in this article.