Remaining Useful Life Prediction of Electric-Vehicle Traction-Motor Winding Insulation Using a Dominant-Frequency Impedance Feature and Gaussian Process Regression

The reliability of traction-motor winding insulation directly affects the safe operation of electric vehicles. This study proposes a remaining useful life (RUL) prediction method for traction-motor main insulation using bipolar-pulse response measurements and Gaussian process regression (GPR). Simulated winding modules were subjected to full-lifecycle accelerated thermal aging at 300 °C. During the diagnostic stage of each aging subcycle, synchronized insulation-voltage and insulation-current waveforms were acquired at 50 MSPS. A 2 μs segment following the positive-polarity rising edge was processed using a Hann window, a 10th-order Butterworth band-pass filter from 2 to 12 MHz, and a 100-point fast Fourier transform. The frequency bin with the maximum current-spectrum magnitude was identified, and a frequency-weighted impedance feature was obtained from the voltage-to-current spectral-magnitude ratio at the same frequency. One complete lifetime trajectory containing 119 diagnostic measurements was evaluated retrospectively. The experimentally observed EOL at 428.405 h was independently established when the 10 kHz winding-to-ground capacitance fell to 44.6% of its initial value. A trend-informed GPR model was updated after each new measurement, and RUL was calculated from the first forecast crossing of the fixed feature threshold. Across ten rolling origins after 80% of the observed lifetime, the mean bounded relative accuracy was 80.08%, with an MAE of 6.455 h and an RMSE of 8.923 h. However, at the final origin, where the actual RUL was 4.185 h, the bounded relative accuracy was 0% because the terminal feature increase occurred abruptly between two measurements. The results constitute a single-specimen retrospective proof of concept; independent multi-specimen validation is required before predictive performance can be generalized.