Diagnosis technology of bearing with variable working conditions based on generalized sidelobe canceller and multi-domain acoustic fusion

Abstract The early stage of rolling bearing failure is often accompanied by detectable weak signals, which can be detected and recognized to a certain extent by non-contact acoustic emission (NAE) and acoustic emission (AE) detection. This paper presents a bearing diagnostic technique that combines generalized sidelobe canceller (GSC) and Multi-Domain Acoustic Fusion (MDAF), which combines the advantages of easy detection in the audible domain and high sensitivity and anti-interference capability in the ultrasonic domain. First, Wavelet Threshold Denoising (WTD) is applied to preprocess the signal matrix, then the microphone array signals are beamformed using GSC based on Linearly constrained minimum variance (LCMV). Subsequently, the beamformed signals are integrated with the ultrasonic domain signals of the acoustic emission through a multi-domain acoustic fusion process. Finally, multiple high-dimensional time-frequency feature representations of the fused signals are used as model inputs using a lightweight RepViT network, enabling their use in mobile devices in the experiment. The results show that IGSC-MDAF-RepViT achieved an accuracy rate of 99.98% on the used features, which was 1.12% and 4.3% higher than that when only using a single sound domain signal, respectively. It can more accurately characterize the operating status of the bearing.