SC-MambaFew: Few-shot learning based on Mamba and selective spatial-channel attention for bearing fault diagnosis

Bearings are crucial yet vulnerable components in electrical equipment, prone to rapid degradation and failure. Diagnosing bearing failures is both critical and highly informative, but constructing large datasets, especially in industrial settings, remains a significant challenge. To address this issue, we propose a novel end-to-end few-shot learning framework, SC-MambaFew(Selective Channel Mamba-based Few-shot learning), for bearing fault diagnosis, which achieves high performance even with limited training data. The proposed framework is built on the Mamba architecture incorporating spatial attention and a Global-Local Channel Attention (GLCA) module to extract both fine-grained local details and broader global patterns, enhancing model performance. The extracted features are further refined through aSelective Channel (SC) mechanism and passed into a covariance-based classification module. Extensive experiments on the HUST and Case Western Reserve University (CWRU) bearing datasets, evaluated using metrics such as Accuracy, Recall, and F1 score, as well as statistical significance testing with the Wilcoxon signed-rank test, demonstrates that the proposed SC-MambaFew outperforms existing deep learning methods. Both qualitative and quantitative results highlight its robustness and potential for real-world industrial applications, particularly in data-scarce scenarios. Our code is available at: https://github.com/giabao804/few-shot-mamba.