A Feature Vector Learning-Based Method for Diagnosing Main Circuit Ground Faults in Electrical Traction Drive Systems

Main circuit ground fault (MCGF) is an electrical fault that occurs in electrical traction drive systems (ETDS). Timely detection and processing are effective ways to avoid further deterioration and safety accidents. Aiming to improve maintenance efficiency and train operation safety, this article proposes a feature vector learning-based method for rapidly and accurately diagnosing MCGFs. The fault mechanism is first investigated for selecting related signals from the traction control unit. Second, a feature variable is constructed by fusing the selected signals using the analytical method. Third, for better representing MCGF of different types, a short-time Fourier transform is used to decompose the feature variable in the time-frequency domain, where four feature indicators are extracted and merged into a feature vector. Then, an MCGF diagnosis framework composed of fault detection and fault identification is proposed. A feature indicator is preliminarily chosen to monitor MCGF. Once it is detected as abnormal, an identification program is initiated, where extreme learning machine is employed to excavate nonlinear mapping between MCGFs and related feature vectors. Finally, experimental results show that the performance of the proposed diagnostic approach is superior to the existing ETDS grounding fault diagnosis methods.