Research on Fault Diagnosis of Rolling Bearing Based on Lightweight Model With Multiscale Features

Abstract

In the rolling bearing fault diagnosis, it is difficult to deploy the model with high computational cost and slow computational speed in engineering applications. Therefore, this article proposes a lightweight model combined with a multiscale features (MSFs) module. An improved sampling method—random window sampling (RWS)—is adopted, which makes the number of samples unlimited by the length of the original data. The lightweight model based on the depthwise separable (DS) convolution and channel shuffle is built, which effectively reduces the calculation parameters. However, the lightweight model may lead to insufficient feature extraction capability, so this article introduces the MSFs module and the squeeze-and-excitation (SE) module. The SE module enhances the extraction of the key feature in the model and the MSFs module improves the feature extraction capability of the model with less calculation cost. The experimental verification is carried out on the rolling bearing fault datasets of Case Western Reserve University (CWRU) and Yanshan University (YSU). The accuracy of the proposed model can reach more than 99% under different datasets. Compared with other models, the computation parameters and floating point of operations (FLOPs) are reduced to varying degrees. The model achieves lightweight while ensuring great fault diagnosis accuracy.