Faults diagnosis of rolling bearings based on shift invariant K-singular value decomposition with sensitive atom nonlocal means enhancement

Abstract K-singular value decomposition (K-SVD) dictionary learning method has been widely used for the fault diagnosis of rolling bearings. However, the periodic impulses extracted by traditional K-SVD method are limited with strong noise as the learned atoms in dictionary are often redundant while the fault information is incipient in the atoms. To address this issue, this paper proposed an NLM-KSVD method for fault diagnosis of rolling bearings based on shift-invariant K-SVD with Nonlocal Means (NLM) sensitive atom enhancement. In this method, the shift invariant K-SVD method is proposed to extract the incipient periodic impulsive features of rolling bearings. As the periodic impulses at different locations with the same characteristic can be represented by feature atom through shift operation, an autocorrelation impulses harmonic to noise ratio (AIHN) index is proposed to select the sensitive atom from all the possible atoms in the learned dictionary. Considering that the periodic impulses of sensitive atom are incipient under strong noise, the Nonlocal Means algorithm is employed to enhance the sensitive atom. The estimated fault feature signal is reconstructed based on the optimized atom. Evaluation results show that NLM-KSVD method can accurately estimate the fault of rolling bearings and is robust to random slippage and random noise. Besides, simulation and experimental results demonstrate that the proposed method produce better detection results than the SIDL-KSVD method and Fast Kurtogram.