基于图卷积网络的轴承故障检测算法  

Bearing fault detection algorithm based on graph convolutional networks

作　　者：邱瑞 于炯[1,2] 冷洪勇[1] 杜旭升 李姝 刘怡然 QIU Rui;YU Jiong;LENG Hongyong;DU Xusheng;LI Shu;LIU Yiran(School of Software,Xinjiang University,Urumqi 830091,China;School of Information Science and Engineering,Xinjiang University,Urumqi 830046,China)

机构地区：[1]新疆大学软件学院,新疆乌鲁木齐830091 [2]新疆大学信息科学与工程学院,新疆乌鲁木齐830046

出　　处：《现代电子技术》2023年第21期89-93,共5页Modern Electronics Technique

基　　金：国家自然科学基金资助项目(61862060);国家自然科学基金资助项目(61462079);国家自然科学基金资助项目(61562086);国家自然科学基金资助项目(61562078);新疆维吾尔自治区自然科学基金项目(2021D01C083)。

摘　　要：轴承故障占到了感应发动机故障总数的13,会造成严重的经济损失甚至生命威胁。准确且及时地检测出轴承故障对于提高生产效率和降低安全风险至关重要。传统检测算法对轴承信号特征的选择较为敏感,而基于机器学习的轴承故障检测算法往往仅考虑了样本的特征信息却忽略了样本之间的联系,从而丢失了许多有用的信息。为解决这个问题,将特征提取迁移到图结构,提出了GE⁃HITS的轴承故障检测方法。该方法首先将样本以德劳内三角网形式连接成图;然后将图结构与样本的特征信息一同作为图卷积神经网络的输入;最后将训练后的数据通过权威值排序来判断故障点。通过与在真实数据集上运行的8种对比算法的3种指标进行比较,GE⁃HITS算法都取得了最佳的结果,充分证明了所提算法的有效性。Bearing failures account for one⁃third of all induction engine failures,which poses significant economic losses and even life⁃threatening risks.Accurate detection of bearing failures is crucial for improving production efficiency and reducing risks.Traditional detection algorithms are sensitive to feature selection,while bearing fault detection algorithms based on machine learning often only consider the feature information of samples and overlook the relationships between samples,which leads to the loss of valuable information.To address this issue,a bearing fault detection method called GE⁃HITS(graph convolutional network and embedding hypertext⁃induced topic search)is proposed by transferring feature extraction to a graph structure.In this method,samples are connected into a graph by the Delaunay triangulation,and the graph structure,along with the feature information of the samples,is used as the input to the graph convolutional neural network.Finally,the trained data are used to confirm the fault points by authoritative value sorting.By comparing the three indexes of the eight benchmark algorithms run on real datasets,it can be seen that the GE⁃HITS algorithm consistently achieves the best results,which fully demonstrates the effectiveness of the proposed approach.

关 键 词：轴承故障检测 深度学习 数据挖掘 图卷积神经网络 GE⁃HITS 权威值排序 

分 类 号：TN911.1-34[电子电信—通信与信息系统] TP311.13[电子电信—信息与通信工程]