A fault diagnosis approach for roller bearing based on boundary smooth support matrix machine

Support matrix machine (SMM), as a typical matrix classification method, is commonly used in the field of mechanical fault diagnosis due to its ability to fully utilize the strong correlation information between rows or columns in the matrix. However, the constraint terms of SMM have the property of local non-differentiability, which affects computational efficiency and accuracy. To address these limitations, a boundary smooth SMM (BSSMM) is proposed in this paper. In BSSMM, the squared hinge loss function is utilized to construct the loss term, which gives the model good generalization performance and robustness. Meanwhile, the square hinge loss function is smooth, which can achieve rapid convergence and avoid falling into the local optimal solution problem. Experimental verification is performed using vibration signals of two types of roller bearings, and the analysis results show that the proposed BSSMM method has superior classification performance compared to SMM and its improved methods.