Objective The intelligent upgrade of railway electrical maintenance system has raised new demands for the fault diagnosis technology of turnout switch machines. Therefore, it is necessary to systematically evaluate the application and development paths of various methods in this field.

Method A multi-dimensional technical analysis of relevant methods and achievements is conducted to reveal the technical characteristics and bottlenecks of current machine learning-driven fault diagnosis systems. Addressing the current research status and taking the ZD6 and S700K switch machines as examples, common fault types of both DC （direct current） and AC （alternating current） switch machines are summarized, and the shortcomings of existing case databases regarding data comprehensiveness and fault category coverage are pointed out. In terms of feature extraction, targeting the data characteristics of switch machine action curves, the methods and challenges of fault feature extraction are reviewed, and the effectiveness of feature processing strategies across different dimensions is discussed. Furthermore, the primary applied algorithms in fault diagnosis are detailed, including typical application cases of support vector machines, neural networks, and cutting-edge algorithms, with a focus on analyzing the performance and applicable scenarios of each algorithm.

Result & Conclusion  Deep learning methods have strongly facilitated the improvement of fault diagnosis capabilities for turnout switch machines, leading to significant enhancements in both accuracy and diagnostic efficiency. Based on certain deficiencies in existing research, potential future research directions are proposed from the data, methodology, and engineering application levels. This provides a theoretical basis and technical pathway for the engineering deployment of intelligent diagnostic systems, thereby promoting the intellectualization and practical upgrade of fault diagnosis for turnout switch machines.