To address the issues of low noise resistance and complex engineering implementation in traditional methods for diagnosing wheel tread faults in high-speed trains, a hierarchical diagnostic method based on envelope demodulation is proposed. This method collects axle box vibration signals and synchronized speed signals, calculates the tread fault characteristic frequency based on wheel diameter, and, after signal preprocessing and Hilbert envelope demodulation, locates the maximum amplitude at the 1x frequency corresponding to the characteristic frequency in the envelope spectrum. After conversion to dB, fault level is determined by using fixed three-level thresholds of 13 dB (preliminary warning), 18 dB (alarm), and 23 dB (alarm). The results of the simulation and field experiment show that this method can accurately extract fault features even in a strong noise environment with a signal-to-noise ratio as low as -3 dB, and the diagnosis accuracy rate reaches 93.3%. Additionally, the process is simple, computationally light, and easy to implement in engineering, providing an efficient and reliable solution for online monitoring of wheel tread faults in high-speed trains.