Research on roller bearing fault diagnosis based on robust smooth constrained matrix machine under imbalanced data

In mechanical health management, the issue of data imbalance is frequently encountered, particularly in cases where there is an extreme excess or scarcity of data. Although support matrix machine (SMM) can effectively transmit and classify structured information through matrix transmission, its effectiveness remains limited when faced with the challenge of data imbalance. Additionally, the SMM model is sensitive to noise and exhibits slow training convergence, both of which further undermine the classification performance of the model. To address the above problems, a method for the diagnosis of roller bearings, termed robust smooth constrained matrix machine (RSCMM), is proposed in this paper. Firstly, a dynamic adjustment factor is designed in the RSCMM model, which can dynamically adjust the coefficients of the loss term according to the sample imbalance ratio, and improves the performance of recognizing imbalanced data. Then, the RoBoSS loss term is employed in the model, mitigate the impact of noise and improve the convergence speed and robustness. Finally, the effectiveness of the proposed method is verified by two roller bearing experimental datasets. The experimental results show that RSCMM performs well under different imbalance ratios, which demonstrates that RSCMM has excellent performance in roller bearing fault diagnosis.