Abstract: Aiming at the problem of strong tunnel reflection noise interference in the process of metro vehicle axle box bearing fault diagnosis, a fault intelligent detection system based on wavelet packet decomposition and big data neural network algorithm is proposed. Through wayside non-contact acoustic sensor array, the sound signal acquisition of axle box bearing in vehicle running part is realized. Wavelet packet decomposition combined with kurtosis calculation method is used for feature extraction. Fault analysis and identification is carried out based on big data neural network algorithm, and intelligent diagnosis is completed. The reliability of the promoted system for metro vehicle axle box bearing fault detection is verified by field measurement.