A Review of Deep Learning-Based Fault Diagnosis Techniques for Axlebox Bearings in Rail Vehicles

As a core component of the bogie of high-speed trains, the performance of axlebox bearings is critical to ensuring train operational safety and stability. However, under the continuous high pressure and complex dynamic loads, this component fails frequently, and traditional methods often fail to satisfy the demands for both high precision and real-time monitoring. In recent years, the breakthrough of deep learning technology provides a new solution for intelligent fault diagnosis of axlebox bearings. This paper systematically reviews the research progress of axlebox bearing fault diagnosis of rail vehicles based on deep learning: firstly, it introduces the importance of bearing fault diagnosis, and then analyzes the characteristics of the bearing fault diagnosis methods on the railroad and the research overview of domestic and foreign scholars. Then the fault diagnosis methods based on deep self-encoder, deep confidence network, convolutional neural network, recurrent neural network, and multi-deep model hybrid are elaborated in detail, and the characteristics of different methods are compared. Finally, the limitations of the current research are discussed, including the lack of data diversity, computational complexity and interpretability, etc. Future research in deep learning should focus on the use of fault datasets from real scenarios, fusion of information from multiple sources, and interpretable deep diagnostic models, in order to promote the wide application of deep learning in the field of axlebox bearing fault diagnosis and the enhancement of the level of intelligence.