Improved Measurement Method for Locating Winding Faults in On-board Transformers in High-Speed Trains

Frequency response analysis (FRA) is a widely used fault diagnosis approach for transformer windings. However, its application to on-board transformers in high-speed trains faces challenges because the four groups of high-voltage (HV) windings are often connected in parallel, making it impossible to measure them separately outside the oil tank. Additionally, locating winding faults becomes more difficult because the condition of the windings changes over time, and prior knowledge about the winding status is not always readily available. This paper proposes an improved measurement method for locating winding faults without reference data. The method establishes multiple signal transmission paths between the HV windings and traction windings through a superimposed series–parallel connection to the winding terminals. The responses of the traction windings are measured simultaneously when excitation is imposed on the HV windings, and the faulty windings are detected based on the symmetry of the responses between the terminals. The correlation coefficient and the sum of resonant peak amplitudes are proposed to evaluate the winding status. First, a finite element method model and an experimental transformer are established to evaluate the efficacy of the proposed method. Experimental studies are then conducted. Test results show that the shift direction, correlation coefficient, and the sum of resonant peak amplitudes are effective in identifying faulty windings. For the FRA signatures of the faulty winding, the correlation coefficients remain around 5 in the low-frequency band, while they all drop below 1 in the mid-to-high frequency bands. The shift direction of the FRA signatures are closely correlated with the type of winding fault. Regarding resonant peak amplitudes, their sum remains stable in the low-frequency band. However, it increases in the medium-frequency band before decreasing in the high-frequency band. This finding confirms that the proposed measurement method can successfully capture fault diagnostic information and locate the faults without the need for reference data. The accuracy of the fault location of the on-board transformer is significantly improved.