Frequency band attention network-based small sample fault diagnosis method of gearbox

Objectives To address the issue of deep learning-based fault diagnosis methods relying on a large number of fault samples, a novel fault diagnosis method based on frequency band attention network is proposed to achieve precise gearbox fault diagnosis under small sample conditions. Methods Initially, through the reconstruction-encoding layer, the vibration signals are transformed into sub-band encoded signals that are easy to classify. Subsequently, the intrinsic band attention layer is constructed to fully mine the representative time-frequency features from sub-band encoded signals. Finally, the multiple feature fusion module is used to integrate the time-frequency feature information for fault recognition under small-sample conditions. Results The proposed method is experimentally validated based on a self-built gearbox fault simulation test bench. Experimental results show that the proposed method achieves a fault diagnosis accuracy of 99.85% under small-sample conditions, outperforming the comparison models. Conclusions The research results can provide a reference for the fault diagnosis of gearbox under small sample conditions.