An improved cepstrum analysis method to diagnose faults in bearings

Cepstrum analysis is suitable for decomposing and identifying fault features of different types. However, it is tough to find the weak fault frequencies from running mechanical systems in heavy background noises. In order to diagnose faults accurately, an improved version of cepstrum analysis is proposed. First, the cepstrum analysis is employed to separate and obtain the periodic parts of dense modulated signals. Second, the cepstrum line in associate with the lower periodic components is removed from cepstrum by carefully selecting the thresholding of cepstrum value. The signal-to-noise ratio (SNR) is enormously modified by filtering out the low-frequency terms interfered by the rotational speed and its harmonics. Finally, the traditional Hilbert envelope analysis is carried out for the reconstructed time-domain signal to extract bearing fault features. The experiment investigations of bearings with inner race faults are performed. The comparison results show that the present improved version of cepstrum analysis has higher performance to detect faults.