Abstract: Bearings are essential components in rotating machinery, and their health status directly determines the production efficiency and operational safety of industrial equipment. Traditional bearing fault diagnosis methods typically rely on the extensive practical experience of experts for judgment or depend on sophisticated signal processing technologies to process the collected data. These conventional approaches are struggling to cope with complex operating conditions in modern industrial scenarios, which are characterized by variable working parameters, strong noise interference, non-stationary fault signals and compound fault occurrences. In recent years, with the rapid development and in-depth application of artificial intelligence technology, neural network technologies within the field of deep learning have emerged as a prominent research hotspot in bearing fault diagnosis. Endowed with powerful adaptive feature learning, nonlinear mapping and end-to-end pattern recognition capabilities, neural networks can automatically extract deep and high-dimensional fault features from raw monitoring data with minimal manual intervention, effectively making up for the deficiencies of traditional methods in dealing with complex industrial environments and thus providing a new and effective solution for accurate and intelligent bearing fault diagnosis.This review aims to systematically sort out and summarize the latest research progress of bearing fault diagnosis methods based on neural networks, providing a comprehensive reference for subsequent in-depth research and engineering application in this field. Firstly, the