An improved local cepstrum analysis method

The traditional cepstrum analysis is difficult to realize fault detection from low SNR signal. An enhanced local cepstrum analysis method is proposed on the basis of analysis of local cepstrum analysis method, the corresponding principles and algorithm are given. The theory of interference reduction of aperiodic components in original signal of time domain is expounded emphatically by introducing the related de-noising processing in time domain. The rules of interference reduction of aperiodic components of spectral peaks in frequency domain and the realization of convergence of periodic fault features near the zero frequency by adopting the related de-noising processing in frequency domain are also explained. The frequency band selection difficulty in the calculation of local cepstrum analysis method is solved, and the limitation of highly localized distribution of periodic fault features in spectrum is eliminated by frequency domain related de-noising processing. The proposed method is employed to analyze both simulated signal and vibration signal of a fault gear. Compared with the traditional cepstrum and local cepstrum methods, the proposed method has strong ability to resist noise and is able to provide 8.05 bit and 1.11 bit more information respectively for fault diagnosis in experiment.