A Mechanism and Data Hybrid-Driven Method for Main Circuit Ground Fault Diagnosis in Electrical Traction System

Abstract

Ground faults can be caused by abnormal grounding of several different modules in the electrical traction drive system, leading to vulnerable train operation. Therefore, rapid and accurate diagnosis of ground faults is of great significance to improve safety and reliability of electric locomotives and electric multiple units (EMU). This paper proposes a mechanism and data hybrid driven method to diagnose the ground faults. Firstly, according to ground fault mechanism analysis, characteristic variables are constructed and several fault features are extracted by using signals selected from existing control system. Secondly, according to historical data, the probability membership function associated with fault features and typical ground fault types is established offline by fuzzy logic theory. Then, the probability information of fault features is fused using Dempster-Shafer (D-S) evidence theory and a decision-making method based on the basic probability number is applied to classify faults. Finally, field experiments are performed and the results show that compared with canonical correlation analysis (CCA) based method and on-board fault diagnosis method, the proposed method has better performance in the presence of large measurement noise and transient condition changes, and it is also promising for practical applications.