Abstract: Aiming at the problem of low accuracy and universality of fault diagnosis models in the scenario of imbalanced sample size for mechanical faults, a twin weighted imbalanced matrix classifier (TWIMC) is designed, drawing on the idea of obtaining a strong supervised model using fuzzy attribute theory. TWIMC adjusts the weight of each sample using a fuzzy membership function based on the degree of sample imbalance, thereby enhancing the focus on minority class samples and balancing the model's tendency towards all types of samples. Meanwhile, TWIMC utilizes prior knowledge to assign weights to the singular values of nuclear norm, preserving the strongly correlated low-rank information of matrix samples by filtering out smaller singular values with a larger threshold. Finally, the proposed method is validated using roller bearing and gear fault datasets. The experimental results showed that TWIMC performed outstandingly under different imbalance ratios, demonstrating excellent mechanical fault diagnosis and classification performance.

Key words: twin weighted imbalanced matrix classifier, support matrix machine, fuzzy membership function, imbalanced sample, fault diagnosis