Aiming at the problem of large background noise and difficulty in extracting effective fault features in the acoustic fault diagnosis of subway wheel bearings, a method for fault feature extraction under strong noise background was proposed. Perform short-time Fourier transform (STFT) on the sound signal to obtain a time-frequency diagram, and the stripes in the time-frequency diagram are the fault features; add the signal intensities of each point of the image along the direction of the stripes to obtain the time-frequency diagram The signal intensity corresponding to the graph is superimposed on a line graph to show the fault characteristics; and an adaptive cyclic noise reduction algorithm based on the peak height is proposed to reduce the noise of the signal intensity superposition line graph, and the evaluation index of the graph is the number of effective peaks; finally An adaptive sliding window detection method is proposed to intercept the fringe distribution area in the time-frequency graph, so as to obtain the optimal fault feature display effect. Experimental results show that the proposed method can extract obvious fault features from the collected audio signals.