Results of numerous studies and experiments show that cepstrum analysis has the ability of simplifying the equally spaced sideband feature in the spectrum and highlights the signal components of defects. However, for most cases of early gear failure, the periodic phenomenon is always buried in strong background noises and the interference of the rotating frequency with its harmonics. Moreover, the features would be further weakened by the average effect of Fourier transform after cepstrum processing. In this paper, an improved cepstrum method named local cepstrum is proposed. The detection principle of local cepstrum is mainly based on the part of spectrum information to enhance the capability of extracting periodical features of detected signals. Besides, the autocorrelation and extended Shannon Entropy Function are also involved enhancing the periodic impulsive features. In the end, only several distinct lines with larger magnitudes would be left in the local cepstrum, which is very effective for gear fault detection and identification. Both simulation and experimental analysis show that the proposed method, which is more sensitive to the gear failure compared with conventional cepstrum analysis, could partially eliminate the interference of background noise and extract the periodical features of premature failure signals effectively.
Gear Fault Feature Extraction Based on Fuzzy Function and Improved Hu Invariant Moments