Threshold Adaptive Intelligent Health Monitoring Method for Axle Box Bearings of High-Speed Trains Under Multiple Operating Condition

To overcome the fluctuations in health monitoring indicators of axle box bearings in high-speed trains due to variations in operating lines and speeds, and to address the difficulties of traditional threshold methods in monitoring bearing health based on a unified standard, a threshold adaptive intelligent health monitoring method for axle box bearings under multiple operating conditions is proposed. First, by analyzing the Spearman correlation coefficients between features and rotational speed, the health indicators of the bearings at multiple operating speeds are normalized to eliminate fluctuations caused by changes in train speed. This enables the automatic adaptation of monitoring thresholds to the train’s operating conditions. Next, an improved Isolation Forest algorithm based on random feature subsets and principal component direction segmentation (SPCD-iForest) is established. This algorithm uses the collaborative information provided by multidimensional features to classify the normal and faulty states of the bearings along the principal component direction of the data, enhancing the computational efficiency of anomaly detection while maintaining monitoring accuracy. Finally, the proposed intelligent health monitoring method is validated using data from line tests of train axle box bearings. Results show that the proposed method eliminates the impact of changes in train operating conditions on the health indicators of axle box bearings. The output anomaly factor is a dimensionless indicator ranging from 0 to 1. effectively reflecting changes in the health status of the axle box bearings. Compared to the TADS trackside acoustic detection method, it can detect bearing faults more than 10 days in advance and is adaptable to monitoring and diagnostic needs across different train speeds and lines. This method is significant for ensuring the safe operation of high-speed trains and enabling predictive maintenance for axle box bearings.