Design and implementation of a physics-guided PSD filter for bearing vibration signal processing

Abstract Vibration analysis is commonly used for monitoring the condition of bearings. However, the collected vibration signals are often contaminated by structural, measurement, and environmental noises. Therefore, filtering the signals is essential before conducting further diagnostic issues. While various filters have been developed, most are designed for general signals and do not account for the specific characteristics of acceleration responses in faulty bearings. To address the gap, this paper proposes a physics-guided power spectral density (PSD) filter. By examining the distribution patterns of fault characteristic frequency (FCF), modulated sidebands and high-order harmonics in the spectrum of bearings with different faults, we can determine the threshold for the PSD filter with physics-guided principles. Furthermore, the proposed method introduces a mode interface that allows for tuning the filtering performance, particularly when considering varying levels of FCF and sideband peaks. Experimental results from both the CWRU and XJTU-SY datasets demonstrate the superior performance of the proposed method.