A Sparsity-Assisted Fault Diagnosis Method Based on Nonconvex Sparse Regularization

Abstract

Sparse representation theory can be adopted for fault feature extraction and classification. Inspired by these two capabilities of sparse representation theory, this paper proposes a novel collaborative sparsity-assisted fault diagnosis (CSFD) method. Specifically, due to the repeatability and sparsity of fault feature signal in the whole signal, the feature extraction capability of sparse representation is utilized to extract fault features and construct a feature matrix. Subsequently, owing to the sparsity of fault classification problem itself, the classification capability of sparse representation is employed to achieve fault classification. Moreover, fault feature extraction is a key issue for fault diagnosis. Therefore, in order to improve the effectiveness of fault feature extraction, a RC-adjustment strategy with the ability to adaptively adjust regularization parameter is designed to assist the generalized minimax-concave (GMC) penalty for feature extraction. The advantage of the GMC penalty is that it can establish a nonconvex sparse regularization model with convex optimization solution. Finally, the proposed CSFD method is tested and verified by Case Western Reserve University (CWRU) bearing dataset and actual engineering dataset. Especially, the diagnostic accuracy of this method can reach 99.92%, which is nearly 7.68% higher than the traditional method based on L1 norm penalty. Enormous experiment results have thoroughly demonstrated the effectiveness of the proposed CSFD method for fault diagnosis.

Adaptive feature selection method with FF-FC-MIC for the detection of mutual faults in rotating machinery

“…For example, literature [8] proposed a latent component decomposition method based on sparse representation to realize the effective identification of weak fault characteristics of bearings and gears under strong noise background. Literature [9] Proposed a novel collaborative sparsity-assisted Fault Diagnosis (CSFD) method improves feature extraction capability and fault classification performance of rotor systems with strong noise. In the literature [10], the periodic weighted kurtosis sparse denoising is used and combined with the periodic filtering method to extract the repetitive pulses in the compound fault, so as to achieve the purpose of denoising and fault separation of the bearing compound fault signal.…”