Trackside acoustic diagnosis of axle box bearing based on kurtosis-optimization wavelet denoising

As one of the key components of railway vehicles, the operation condition of the axle box bearing has a significant effect on traffic safety. The acoustic diagnosis is chosen which is more suitable than vibration diagnosis for trackside monitoring. The acoustic signal generated by the train axle box bearing is a amplitude modulation and frequency modulation signal with complex train running noise. Although empirical mode decomposition (EMD) and some improved timefrequency algorithms are proved to be useful in bearing vibration signal processing[1-2], it is hard to extract the bearing fault signal from serious trackside acoustic background noises by using those algorithms. Therefore, a wavelet packet denoising algorithm based on the kurtosis optimization (KWP) is proposed, as the kurtosis is the key indicator of bearing fault signal in time domain. Firstly, the geometry based Doppler correction is applied to signals of each sensor, and with the signal superposition of multiple sensors, random noises and impulse noises (which is the interference of the kurtosis indicator) are suppressed. Then, the KWP is conducted. At last, the EMD and Hilbert transform (HT) is applied to extract the fault feature. Experiment results indicate that the proposed method consisting of KWP and EMD is superior to the EMD.