Bearing fault diagnosis based on improved cepstrum under variable speed condition

Abstract

The speed of rolling bearing often varies in actual operation process, the fault diagnosis of variable speed bearing is particularly important. In this paper, the definition of local maximum peak value of cepstrum is proposed, which can be used to extract variable speed features. And a variable speed bearing fault diagnosis method based on improved cepstrum is proposed. The proposed method could extract fault feature without tachometer and resampling. At first, the Local maximum peaks of the original cepstrum are extracted. Then, the corresponding thresholds of the peaks are calculated and determined one by one. Finally, the feature extraction is completed according to the thresholds. At the same time, this paper verified the feature extraction results under four different variable speed conditions, including acceleration, deceleration, acceleration before deceleration and deceleration before acceleration. The experiments results show that the proposed method has higher accuracy (99.08%) and less running time than the other spectral analysis methods, which demonstrates the accuracy, rapidity and effectiveness. This proposed method provides a novel idea for fault diagnosis under condition of variable speed.

A methodological integration of fisher score technique with intelligent machine learning methods for ball bearing fault investigation

“…Frequency-domain techniques, such as Fast Fourier Transform (FFT), Power Spectrum, Cepstrum analysis, and Time-frequency-domain Wavelet transform, as well as short-time Fourier transform (STFT) have also been reported by the researchers [14]. The suggested frequency methods, such as Fast Fourier transform [15], Empirical mode decomposition (EMD) [16], Hilbert transform [17], Hilbert-Huang transform [18], Spectral analysis [19], Stochastic resonance [20], Sparse decomposition [21], Wigner-vile distribution [12] and Cepstrum [22] have been discussed for feature extraction and fault classification. While these traditional investigation techniques have improved the accuracy of fault diagnosis methods, but they may not be suitable for handling complex and large datasets of recorded signals.…”