A Robust Data-driven Fault Detection Framework for Traction Dual Rectifiers

A framework to analyze the impact of feature selection, threshold level, and applied data window on fault detection accuracy is proposed in this study. It aims to bring a step closer to state-of-the-art fault detection research to practical applications. A relatively comprehensive database is established by emulating traction dual rectifiers for railway applications under diverse normal and fault operations. Factors such as grid-side fluctuations and harmonics, load changes and parameter variances, and a broad range of traction operation modes are considered in the database preparation. The proposed framework makes it possible to evaluate the accuracy with different design variables in the fault detection implementation. The presented study is exemplified by a case study on open-circuit faults in power devices of the traction dual rectifiers.