Motor fault diagnosis is a core aspect of predictive maintenance in industrial equipment. To address the limitations of traditional data-driven methods in decoupling multi-band features, this paper proposes a motor fault diagnosis method based on Transformer-ResNet feature fusion. The method uses discrete wavelet transform (DWT) to achieve multi-band signal decomposition, employs a residual network (ResNet) to extract low-frequency features, utilizes a Transformer to mine high-frequency features, and then fuses the two types of features via a feature fusion mechanism to improve the model’s feature extraction capability. Experiments demonstrate that the proposed method achieves an average accuracy of 95.8% in motor fault diagnosis, which is 12.3% higher than that of single-model methods, validating the effectiveness of the Transformer-ResNet feature fusion approach.

motor fault diagnosis; Transformer; ResNet; multi-band feature decoupling; feature fusion