基于图卷积网络的滚动轴承智能故障诊断

Intelligent Fault Diagnosis of Rolling Bearing Based on Graph Convolution Network

基于非欧几里德空间的数据包含着数据点以及数据点之间的关系信息,而基于深度学习模型的故障诊断方法通常忽略了数据点之间的关系信息。对此,通过结合可视图算法和图卷积网络,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。首先,将原始信号利用可视图算法转换为图数据,以图数据显示时域特征,极大丰富了输入信息;其次,利用构建的图卷积网络对故障特征进行学习,以达到故障诊断的目的。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97%以上的准确率,这表明利用可视图算法提取的关系信息对轴承故障的识别具有重要作用。

Data based on non-Euclidean space contains data points and relationship information between data points.However,fault diagnosis methods based on deep learning models usually ignore the relationship information between data points.In this paper,irregular data based on non-Euclidean space is applied to bearing fault diagnosis by combining the visibility algorithm and the graph convolution network.Firstly,the original signal is converted into graph data by viewable algorithm,and the graph data is used to display the time domain characteristics,which greatly enrich the input information.Secondly,the constructed graph convolutional network is used to learn fault features to achieve the purpose of fault diagnosis.In order to verify the effectiveness of the proposed method,experiments are carried out on node-level classification task and graph-level classification task respectively.Experimental results show that graph convolutional network can achieve more than 97%accuracy in single bearing fault classification task,which indicates that relational information extracted by viewable algorithm plays an important role in bearing fault identification.