An improved decomposition method for extracting unknown rolling bearing fault features from strong noise

Many decomposition methods have been developed and applied to find bearing fault in recent years, but it is quite difficult to effectively extract the bearing fault characteristics, especially under strong noise and variable speed conditions. Among them, empirical mode decomposition (EMD) is the most widely used. To improve the extraction effect of rolling bearing fault features, this paper proposes a bearing fault extraction algorithm based on fractional Fourier transform (FRFT). The collected vibration signal is first analysed by envelope demodulation and mean normalisation. Secondly, the EMD method is used to remove many noise interferences and retain the bearing fault characteristics. Finally, an effective FRFT filtering algorithm is applied to find fault characteristic signal and remove the residual noise. Both simulated and experimental analyses are conducted to illustrate the performance of the proposed method. The results indicate that this method can accurately and completely extract the unknown bearing fault features from raw signal, which contains noise and irrelevant vibration signals. The proposed algorithm may provide reference for the fault diagnosis of other machine elements, such as gears.