An Optimal Band-pass Filter based on Adaptive Identification of Bearing Resonant Frequency Band

For a noisy vibration signal, the identification of the critical resonant frequency band (RFB) is critical for accurate fault diagnosis of rolling bearings. It is also a difficult task because of the existence of large noise and weak vibration caused by faulty bearing component(s). To solve this problem and make the identification of the RFB accurate and automatic, an optimal band-pass filter is proposed in this paper. Firstly, the whole frequency spectrum of the analyzed signal is equally divided into some small bands and then some windows are generated by automatically merging these small bands according to their energy changes, so that the energy concentration caused by the bearing defect can be enlarged and make the following identification of the RFB easier. After that, choosing the window with the maximal energy as the center, its adjacent windows are progressively involved in new filtering bands. Finally, optimal parameters of the band-pass filter are automatically determined when the change of kursosis of the filtered signal is small. The experimental results of bearings with single and multiple defects demonstrate that the proposed method can successfully search the location of RFB and extract the bearing signals under strong noise, so that weak faults cannot be missed for accurate bearing fault diagnosis.