Incipient Fault Monitoring for Traction Systems of High-Speed Trains Based on Multivariate Collaborative Analysis

Aiming at the difficulty in real-time monitoring of incipient minor faults in high-speed train traction systems due to their weak amplitudes and susceptibility to noise masking, this paper presents an incipient fault monitoring approach based on Multivariate Collaborative Analysis (MCA). Firstly, canonical correlation analysis is employed to explore the dynamic correlations between input and output variables of the traction system, followed by construction of augmented vectors, solving covariance matrices, and obtaining projection matrices via singular value decomposition to separate the principal component and residual subspaces. Secondly, statistical indices are established by leveraging the high sensitivity of the Bhattacharyya distance (BD) to differences in probability distributions. Finally, the monitoring threshold under a specific confidence level is determined through offline training, and online fault decision-making is achieved by comparing real-time calculated statistical indices with the preset threshold. The proposed method is validated using operational data of four typical faults on a high-speed train Traction Drive Control System (TDCS) test platform. Experimental results demonstrate that the proposed method features high sensitivity to incipient faults and strong robustness against noise, with computational efficiency satisfying real-time monitoring requirements, providing effective technical support for the safe operation of high-speed train traction systems.