Improved local cepstrum and its applications for gearbox and rolling bearing fault detection

Cepstrum is a kind of powerful and widely used method in the fields of speech signal processing, echo signal detection, building acoustics, seismic analysis, condition monitoring and fault diagnosis. But in some circumstances, such as signals with localized distribution of spectrum or low signal-to-noise ratio (SNR) which are heavily drowned by noise, cepstrum fails to provide effective results. An improved local cepstrum is proposed on the basis of local cepstrum. The principle and algorithm of the proposed method are given. By introducing the autocorrelation denoising processing in the time domain and in the frequency domain individually, the background noise and nonharmonic components in vibration signal and power spectrum are reduced. In the mean time, the highly localized fault features in power spectrum are enhanced and converged to zero frequency in the power spectral autocorrelation function (PSAF). The difficulty of frequency band selection in local cepstrum analysis is solved with the help of spectral negentropy. At the step of transformation from the frequency domain to the quefrency domain, autoregressive (AR) spectrum estimation is adopted instead of the fast Fourier transform (FFT) for improving fault feature extraction effectiveness. The proposed method is employed to analyze simulated signal and experimental vibration signals of gearbox and rolling element bearing. In contrast to conventional cepstrum and local cepstrum, the proposed method has a strong ability to resist noise and is effective to detect the early fault.