An Adaptive Envelope Orthogonal Sparsity Matching Method for Rolling Bearing Transient Feature Enhancement

The vibration signal features of rolling bearings often exhibit weakness and transience under noise interference. Enhancing these features is of crucial importance for the effectiveness of fault diagnosis. In industrial production, inaccurate fault diagnosis resulting from the ineffective extraction of fault features may lead to equipment damage, production disruptions, and substantial economic losses. Traditional sparse learning methods generate a dictionary through the construction method based on analytical expressions, which cannot completely represent the real fault impulses. Superadded by the complexity of sparse solution, this kind of method fails to achieve satisfactory results to fully recover the fault information. This article proposed a novel adaptive envelope orthogonal sparsity matching (AEOSM) method for rolling bearing transient feature enhancement. To improve the adaptability of the sparse dictionary, the adaptive variational mode decomposition (AVMD) method is first introduced to adaptively extract the intrinsic features. Then a size-fixed window is applied to pick the impulse atoms and combine them with the K-singular value decomposition (K-SVD) algorithm to construct the learning dictionary. The learned atoms are further evaluated with maximum SVD co-kurtosis (MSCK) assessment to filter the final atom and its corresponding dictionary. In addition, the orthogonal matching sparsity is solved with the envelope of intrinsic feature and learned envelope dictionary. Finally, through a series of inverse transforms, the enhanced waveform and corresponding spectrum can be obtained for precise fault diagnosis. Two cases of experiments and discussion verified the efficiency of the proposed AEOSM method.