一种改进CNN在轴承故障诊断中的应用

针对轴承智能故障诊断过程中的特征自适应提取和在变工况下诊断能力差的问题,提出了一种基于特征通道权重调整的“端对端”一维卷积神经网络(Squeeze-Excitation Convolutional Neural Network,SECNN)滚动轴承故障诊断模型。首先采用一维卷积神经网络自适应地从原始振动信号中提取特征进行分类;然后通过增加特征通道权重模块来获取通道全局信息,学习特征通道之间的依赖关系;再据此对特征通道权重进行调整,增强滚动轴承故障诊断模型在变工况下的特征自适应提取能力。通过轴承实验台数据的验证结果表明:SECNN在多个变载荷工况下的故障诊断准确率均值达到97%,相比于传统方法提高了20%左右。同时利用t-SNE技术可视化特征提取过程,进一步验证了所提取的诊断模型的有效性。

Application of an improved CNN in fault diagnosis of bearings

Aiming at the problem that adaptive feature extraction and poor diagnostic capability under variable working conditions in the intelligent fault diagnosis of bearing,an end-to-end rolling bearing fault diagnosis model(SECNN)based on one-dimensional convolutional neural network was proposed.Firstly,the convolutional neural network was used to adaptively extract features from the raw vibration signals for classification.Then the feature channel weight block was added to obtain the global information of the feature channels and learn the dependence relationship between channels.According that,the weight of channel was adjusted to enhance the adaptive feature extraction ability of the model under variable working conditions.The verification results of the bearing test bench data show that the average diagnostic accuracy of SECNN can achieve 97%when the load varies,20%higher than the traditional methods.Meanwhile,the feature extraction process was visualized by t-SNE technology to further verify the effectiveness of the network.