The detection of train safety is an important research topic for railway locomotive and vehicle engineering. Wheelsets are key components of high-speed train bogies. The traditional detection method can only diagnose macroscopic structural defects, and it is not suitable for the diagnosis of microscopic defects in the operation process, and dynamic observation of failures. In order to make up for the deficiencies of the traditional test methods, this study proposes a new dynamic detection system to meet the safety requirements of train driving. The system consists of three main parts, namely detection mode selection and data acquisition system, large-capacity data transmission system and signal noise reduction and feature extraction system. Experimental research shows that using metal magnetic memory technology to detect the early faults of wheelsets to ensure the safety of trains has important practical significance, and provides reference for the application of nondestructive testing.