Bearing Fault Diagnosis with Multi-source Information Fusion and Deep Transfer Learning

In the fault diagnosis of bearings in hydropower station units, the information provided by a single sensor is limited, and the dynamic characteristics of the operating conditions vary under different load conditions. To address these issues, this paper proposes a bearing fault diagnosis method that integrates multi-source information with deep transfer learning. Firstly, multi-scale features of the input signals are extracted through a fusion convolution module to ensure the completeness and effectiveness of fault information. Subsequently, channel attention and spatial attention mechanisms are used to retain effective information with minimal data, enhancing the efficiency and accuracy of information fusion. Finally, the fault diagnosis method based on transfer learning effectively copes with detection under different loads. To more comprehensively mine and utilize the hidden features in fault signals, this paper combines vibration and temperature signals to build a bearing fault diagnosis experimental platform, collecting vibration and temperature signals at different speeds to generate a dual-channel dataset. Compared to common domain adaptation methods such as deep correlation alignment and domain adversarial neural networks, the proposed model's fault diagnosis performance under different working conditions is validated. Experimental results show that the proposed multi-source information fusion method performs with higher diagnostic accuracy both under the same and varying working conditions, significantly outperforming traditional methods.