Bearing Fault Feature Extraction Method Based on Center-Frequency Adaptive Filter Assisted Sparse

Abstract

When bearings, a key component of rotating machinery, fail during the operation of the equipment, the failure features are relatively weak. Without timely monitoring and diagnosis, it is easy to cause downtime and even major safety accidents. Therefore, an algorithm based on center frequency adaptive filter assisted sparse (CFAFAS) is proposed, which can be used to extract the fault features in the signal. First, a center frequency-based unilateral attenuation filter bank is constructed by unilateral attenuation wavelets. The filter atoms matching the signal are obtained by adaptive selection of the filter bank. Then, the obtained filter atoms are used for convolutional sparse coding to reduce the redundant components of the signal, and the sparse coefficients characterizing the impact components of the signal are obtained. Finally, the sparse coefficients are analyzed by envelope spectrum to determine the fault types. Simulation experiments with test bench-bearing fault signals verify the effectiveness of the CFAFAS algorithm. Meanwhile, through the comparison experiment of the traditional fast spectral kurtosis method and GMC sparse enhancement diagnostic method, the comparison method cannot clearly extract the fault feature frequency, which verifies that the CFAFAS algorithm has a better feature extraction effect.