Fault feature extraction of rolling element bearings based on short-time processing

Abstract

Fault diagnosis of bearings is a crucial part of the maintenance process of the rotary machinery. Extracting the cyclic characteristics of the impact force is of significant importance for the bearing diagnosis. To highlight the fault features from signals combined with heavy background noise, a novel approach for bearing fault diagnosis based on the short-time processing is proposed. Fault signals are regarded as periodic impulse response signals. Firstly, a vibration signal is band-pass filtered with a subsequent spectral analysis. Then we integrate the energy of the filtered signal with a constant length, and the natural logarithm is considered to obtain the energy curve. The energy curve is a straight decaying curve, and its spectral energy is more concentrated on the fault characteristic frequency compared with envelope. Finally, the fault characteristic frequency of the bearing is found by the spectral analysis of the energy curve. The effectiveness of the proposed method is verified by simulation and experiments. The harmonics and sidebands in logarithmic energy spectrum are suppressed well, and the fault characteristic frequency is highlighted. Comparison of the proposed method with Hilbert envelope method shows that the proposed method can highlight the fault characteristic frequency.