基于边缘图注意力网络的轴承智能故障诊断

基于欧几里德空间的数据包含着节点和边的关系信息，比传统的欧几里得空间的数据具有更多信息。然而，传统的图卷积以及图注意力网路注重于节点信息的提取，对于边的信息利用不够充分。对此，通过结合可视图算法和边缘图注意力网络（EGAT），将基于非欧几里德空间的不规则数据应用到轴承故障诊断领域。诊断过程分为两步：利用可视图算法将原始信号转化为图数据；利用EGAT对故障特征进行学习，然后即可进行故障诊断。实验结果表明：图卷积网络在单一轴承故障分类任务上能够达到 100%的准确率，表明所提出的方法对于轴承故障诊断具有明显的作用。

Bearing Intelligent Fault Diagnosis Based on Edge Graph Attention Network

The data based on Euclidean space contains the relation information of nodes and edges,which has more information than the data in traditional Euclidean space.However,the traditional graph convolution and graph attention network focus on the extraction of node information,while the edge information is not fully used.Aiming at this,by combining viewable algorithm and edge graph attention network (EGAT),irregular data based on non-Euclidian space were applied to bearing fault diagnosis.The diagnosis process was divided into two steps:the viewable algorithm was used to convert the original signal into graph data;EGAT was used to learn fault features,and then fault diagnosis could be carried out.The experimental results show that the graph convolutional network can achieve 100% accuracy in a single bearing fault classification task,which indicates that the proposed method has a distinct role in bearing fault diagnosis.