Abstract: A fault diagnosis method based on the AFF-Stablenet model is proposed to address the issues of low diagnostic accuracy and weak generalization of rolling bearings under small sample conditions.Initially,the samples are decomposed into sub-signals of multiple frequencies using EMD. The cross-correlation coefficients between the sub-signals and the original signal are computed. The top three sub-signals with higher coefficients are selected.These sub-signals are transformed into time-frequency representations using CWT. Through attention-based feature fusion,the time-frequency features are integrated.Finally,the fused features are input into the Stablenet model for training and prediction. To validate the effectiveness of the proposed model,comparative experiments are conducted using the Case Western Reserve University bearing dataset and verified using the Politecnico di Torino bearing dataset. Experimental results demonstrate that the AFF-Stablenet model exhibits superior generalization and robustness under small sample conditions compared to other models,affirming the superiority of the proposed approach.

Key words: pay attention to feature fusion, deep and stable learning, rolling bearing, small samples, fault diagnosis