Abstract: In order to make up for the shortcoming of the existing methods for compensation capacitor fault detection in both timeliness and cost aspects, a simulation model for induced voltage envelope of cab signal was proposed based on the transmission-line theory. On this basis, the influence law of compensation capacitor failure on the induced voltage envelope was analyzed, and a frequency-based equivalent regression model for the induced voltage envelope was built and verified. Based on this regression model, a compensation capacitor fault detection method was proposed based on the AOK-TFR (adaptive optimal kernel time-frequency representation) and the fact that the frequency of the induced voltage envelope changes ahead and behind the location of the defected capacitor. In this method, the signal to be processed, directly from a cab signal recorder, is denoised by using the B-spline discrete binary wavelet transform (DBWT). Then its time-frequency distribution information is extracted by the AOK-TFR, and the fault compensation capacitor is lastly located in light of the change of the frequency extracted. The experimental result of actual data from 100 segments selected randomly shows that the accuracy rate is about 92.58% to indicate that the proposed method can accurately detect the location of a defected compensation capacitor under the condition that other capacitors are in good condition.