Data-driven Method for Fault Diagnosis in Braking System of Urban Rail Vehicle Based on Reduced Kernel Principal Component Analysis

Condition monitoring and fault diagnosis in braking system of urban rail vehicles have important practical value, and can directly improve the stable operation of equipment and passenger comfort. In this study, a data integration system was established to consistently acquire accurate online information, and then the operating conditions were extracted from a massive historical dataset. Kernel principal component analysis (KPCA) is widely used for fault detection and diagnosis, however, the approach is limited by the high computational cost and memory requirements of large-scale training datasets. Herein, the reduced KPCA (RKPCA) approach is developed. The proposed method reduced redundant data in the training dataset by 32.84%. Furthermore, fault detection performance of RKPCA is superior to those of principal component analysis and KPCA.