基于图卷积网络的轴承故障诊断

Bearing Fault Diagnosis Based on Graph Convolutional Network

针对基于图卷积网络(GCN)的故障诊断方法大多默认节点间的权重相同、导致诊断精度较低与鲁棒性较差的问题,提出了一种基于欧式距离和余弦距离的GCN故障诊断方法(EC-GCN)。首先,充分考虑节点间的数值特征和方向特征,利用欧式距离和余弦距离构建加权样本关联图。其次,利用GCN对样本关联图的故障特征进行提取。最后,将故障特征输入分类器进行分类识别。对公开轴承数据集进行测试,EC-GCN方法能够有效提取轴承故障特征,故障分类精度达到了98.93%。

Most of the fault diagnosis methods based on graph convolutional network(GCN)have the same weight between the default nodes,which result in low diagnosis accuracy and less robustness.A GCN fault diagnosis method based on euclidean distance and cosine distance(EC-GCN)was proposed.Firstly,the numerical and directional features between nodes were fully considered,and the euclidean distance and cosine distance were used to construct the weighted sample association graph.Secondly,the GCN was used to extract the fault features of the sample correlation graph.Finally,the fault features were inputted into the classifier for classification and recognition.Tested on a public bearing dataset,EC-GCN method can effectively extract bearing fault features,the fault classification accuracy reaches 98.93%.