Fault Diagnosis of Rolling Bearings Based on Spectral Kurtosis Graph and LFMB Network

Rolling bearings usually operate under a time-varying speed. However, most technologies for diagnosing bearing faults are based on a constant working speed. The energy change in the spectral kurtosis images of bearings represents the characteristic frequency change of the bearings under time-varying conditions. Considering the running characteristics of rolling bearings under a time-varying speed and taking advantage of the MBConv and Fused-MBConv structures to extract image change features, we built a lightweight network focused on extracting the change features of the spectral kurtosis images of bearings. This paper presents a fault diagnosis method for rolling bearings based on a spectral kurtosis graph and lightweight Fused-MBConv neural network. This end-to-end method can diagnose bearings with not only constant speed but also time-varying speeds. The effectiveness of the method is verified using constant-speed and time-varying-speed bearing datasets. The results show that the accuracy of the rolling bearing diagnosis can reach 98%.

Automated summary

This paper presents a fault diagnosis method for rolling bearings based on a spectral kurtosis graph and lightweight Fused-MBConv neural network. The method can diagnose bearing faults under time-varying speeds and achieves high accuracy in experiments.