Multi-bearing weak defect detection for wayside acoustic diagnosis based on a time-varying spatial filtering rearrangement

The wayside Acoustic Defective Bearing Detector (ADBD) system plays an important role in ensuring the safety of railway transportation. However, Doppler distortion and multi-bearing source aliasing in the acquired acoustic bearing signals significantly decrease the accuracy of bearing diagnosis. Traditional multisource separation schemes using time-frequency filters constructed by a single microphone signal always show poor performance on weak signal separation. Inspired by an assumption that the spatial location of different sources is different, this paper proposes a novel time-varying spatial filtering rearrangement (TSFR) scheme based on a microphone array to overcome current difficulties. In the scheme, a zero-angle spatial filter and peak searching are proposed to obtain the time-centers of corresponding sources. Based on these time-centers, several time-varying spatial filters are designed to extract different source signals. Then interpolation and rearrangement are used to correct the Doppler distortion and reconstruct the corresponding separated signals. Finally, the train bearing fault diagnosis is implemented by analyzing the envelope spectrum of the corrected signals. Because the time-varying spatial filter construction is only dependent on the source location and has little relationship with the signal energy, the proposed TSFR scheme has significant advantages in weak signal separation and diagnosis in comparison with traditional ones. With the verifications by both simulation and experiment cases, the proposed array-based TSFR scheme shows a good performance on multiple fault source separation and is expected to be used in the ADBD system.