RepVGG与CapsNet融合的轴承故障诊断

Bearing fault diagnosis based on the combined use of RepVGG and CapsNet

如何提高特征提取能力和提取特征空间信息是实现高精度故障诊断的关键。RepVGG等深层卷积神经网络忽视了特征空间信息,而CapsNet由于网络层次较浅导致特征提取能力受限。为解决上述问题,提出了RepVGG与CapsNet融合的轴承故障诊断方法。首先获取不同测点的振动信号,通过格拉姆角差场转换成二维特征图并沿通道方向拼接;然后选取RepVGG网络作为前置卷积层,实现多维振动信号的特征提取与融合;最后由CapsNet提取特征空间信息,实现轴承故障诊断。试验结果表明,RepVGG与CapsNet融合的故障诊断方法具有优良的故障识别效果和抗噪性能。

How to improve feature extraction ability and extract the spatial information of features is the key to achieve high-precision fault diagnosis.RepVGG and other deep convolutional neural networks ignore the spatial information of features,while CapsNet has limited feature extraction ability due to its shallow network level.In order to solve the above problems,a bearing fault diagnosis method based on the combined use of RepVGG and CapsNet was proposed.First,the vibration signals at different measuring points were obtained,converted into two-dimensional feature maps by gramian angular difference field,and concatenated along the channel direction.Then,the RepVGG network was selected as the pre-convolutional layer to realize the feature extraction and the fusion of multi-dimensional vibration signals.Finally,by CapsNet,the features’spatial information was extracted to realize bearing fault diagnosis.The experimental results show that the fault diagnosis method based on the combined use of RepVGG and CapsNet has excellent fault recognition performance and noise resistance.