Abstract: As one of the key components of a bogie, early fault detection of bearings is crucial. This paper proposes a method for early fault detection of key bogie components based on a relation network. A health status detection relation network model is designed, which can effectively extract bearing condition features and measure the nonlinear distance between features. In the offline modeling phase, normal samples from the bearing to be monitored are collected and used for training to learn the nonlinear distances between the health state samples. During the online monitoring phase, samples from the current operating state are used for detection, and the relation score is obtained as a health indicator of the bearing’s condition. The 3σ criterion is applied to determine the health threshold for detecting the bearing’s condition and identifying faults promptly. Experiments are conducted on the XJTU-SY rolling bearing full-lifecycle dataset. The results demonstrate that, compared with methods such as root mean square, kurtosis, and stacked autoencoders, the proposed method is more sensitive to early faults and exhibits better monotonicity and trend. Additionally, compared to methods like Isolation Forest, Support Vector Machine, and stacked autoencoders, the proposed method detects the first fault occurrence earlier, showing significant practical value.