Remaining Useful Life Prediction of Bearings under Complex Operating Conditions: A DAENet-XLSTM Transfer Learning Model

In industrial applications, the working conditions of bearings are complex and varied, making it difficult for traditional prediction models to accurately estimate their remaining useful life (RUL). Therefore, this study developed a transfer learning model that combines an automatic denoising encoder with an improved extended long short-term memory network (XLSTM). By adopting a two-stage feature extraction and fusion mechanism, the DAENet module integrates features from noisy vibration signals, while the improved XLSTM focuses on capturing temporal degradation characteristics. Finally, output the RUL prediction value through the fully connected layer. The model was validated using the XJTU-SY dataset and the IEEE 2012 PHM Challenge dataset. The experimental results show that the error of this model is significantly lower than that of the four mainstream transfer learning methods, whether it is the same bearing under different operating conditions or different bearings under varying conditions. The results indicate that our approach not only enhances feature learning efficiency and model generalization capability, but also provides a reliable solution for industrial predictive maintenance systems.