Crack detection in rotating shafts using wavelet analysis, shannon entropy and multi-class SVM

Abstract

Incipient fault diagnosis is essential to detect potential abnormalities and failures in industrial processes which contributes to the implementation of fault-tolerant operations for minimizing performance degradation. In this paper, an innovative method named Self-adaptive Entropy Wavelet (SEW) is proposed to detect incipient transverse crack faults on rotating shafts. Continuous Wavelet Transform (CWT) is applied to obtain optimized wavelet function using impulse modelling and decompose a signal into multi-scale wavelet coefficients. Dominant features are then extracted from those vectors using Shannon entropy, which can be used to discriminate fault information in different conditions of shafts. Support Vector Machine (SVM) is carried out to classify fault categories which identifies the severity of crack faults. After that, the effectiveness of this proposed approach is investigated in testing phrase by checking the consistency between testing samples with obtained model, the result of which has proved that this proposed approach can be effectively adopted for fault diagnosis of the occurrence of incipient crack failures on shafts in rotating machinery.