An identification method for mechanical fault diagnosis based on generalized matrix norm sparse filtering

Acoustic signals have attracted increasing attention in mechanical fault diagnosis due to the advantage of non-invasive measurement. However, the acoustic signal has low signal-to-noise ratio (SNR) and weak fault characteristics, which brings difficulty for fault feature extraction. To address the above deficiencies, a novel sparse filtering method based on generalized matrix norm (GMNSF) is proposed in this paper, which uses the matrix norm to determine the optimal sparse feature distribution. Specifically, principal component analysis (PCA) is employed on the overlapping segments of the acquired sound signal first. Then, the GMNSF model is trained by principal component matrix and sparse features are mapped from the trained weight matrix. Finally, softmax regression is used as a classifier to categorize different fault types. The diagnostic performance of the proposed method is verified by the bearing and planetary gear datasets. Results show that the GMNSF model has good feature extraction performance than other traditional methods that can be used for mechanical fault diagnosis under acoustic signals.