Logarithmic reordering kurtogram: An automatic resonance demodulation method for bearing fault diagnosis

Abstract

This paper proposes an automatic resonance demodulation method, the logarithmic reordering kurtogram (LRKurtogram), based on frequency band division and band evaluation for bearing fault diagnosis. Additional frequency band division is provided by the subband-reconstructed, rearranged, and ensemble dual-tree complex wavelet packet transform (SRRE-DTCWPT). Excellent performance of the dual-tree complex wavelet packet transform (DTCWPT) is thus achieved while mitigating the disorder problem in frequency subbands and band-leakage issues. Meanwhile, a robust band evaluation index based on the log-envelope, namely logarithmic reordering kurtosis, is designed to address the kurtosis index’s vulnerability to intense isolated impulse noise and its ineffectiveness. The effectiveness and improvements of the LRKurtogram are confirmed by analyzing several simulated and real signals and comparing with many advanced methods in the field. In particular, the LRKurtogram demonstrates remarkable effectiveness and robustness in extracting bearing fault characteristics from a series of signals with a wide range of signal-to-impulse noise ratios, highlighting its potential for bearing fault diagnosis.