A non-contact fault diagnosis method based on multi information fusion networks for rolling bearings

Traditional fault diagnosis of rolling bearings primarily depends on vibration signal analysis, however, the physical contact requirement of vibration sensors significantly limits their practical application. To overcome this limitation, a novel non-contact diagnostic approach utilizing sound array technology, the Multi-Information Fusion Network (MIFNet) is proposed. Firstly, a multi-scale feature fusion module with information enhancement (IE-MSFFM) is developed, which adaptively enhances the sound signals of each channel to reduce signal noise and extract multi-scale characteristics for information fusion. Secondly, a multi-channel information selection fusion module (MCISFM) is developed to remove redundant information between multi-channel sound array signals and perform further information fusion to extract deep fault features of rolling bearings. Finally, the fault diagnosis module (FDM) is used to obtain the fault diagnosis results. The effectiveness of MIFNet is evaluated based on experimental data acquired by circular array sound sensors. The results show that MIFNet has excellent robustness and fault feature extraction performance in processing sound array signals. In addition, compared to existing advanced bearing fault diagnosis methods, MIFNet can faster and more accurate diagnose faults based on sound array signals. This study provides a new diagnostic method for non-contact fault diagnosis of rolling bearings.