引用本文:胡艳艳,衣骁捷,彭开香.基于加权图卷积网络的多传感器旋转机械故障诊断[J].控制理论与应用,2026,43(3):521~529.[点击复制]
HU Yan-yan,YI Xiao-jie,PENG Kai-xiang.Multi-sensor rotating machinery fault diagnosis using weighted graph convolutional network[J].Control Theory & Applications,2026,43(3):521~529.[点击复制]
基于加权图卷积网络的多传感器旋转机械故障诊断
Multi-sensor rotating machinery fault diagnosis using weighted graph convolutional network
摘要点击 686  全文点击 95  投稿时间：2023-12-25  修订日期：2025-07-05
查看全文  查看/发表评论  下载PDF阅读器   HTML
DOI编号  10.7641/ CTA.2024.30829
  2026,43(3):521-529
中文关键词  多传感器  故障诊断  希尔伯特–黄变换  图卷积网络
英文关键词  multiple sensors  fault diagnosis  Hilbert-Huang transform  graph convolutional network
基金项目  国家自然科学基金项目(62273038,U21A20483),智控实验室开放基金项目(ZKSYS–KF03–05)资助.
作者单位E-mail
胡艳艳* 北京科技大学自动化学院 huyanyan@ustb.edu.cn 
衣骁捷 北京科技大学智能科学与技术学院  
彭开香 北京科技大学自动化学院  
中文摘要
      多传感器数据能为故障诊断提供更为全面和精确的信息,但现有建模在欧氏空间里的深度学习算法难以 有效处理传感器间复杂的相互影响和空间关系.同时,旋转机械振动信号的非平稳特性也极大影响了故障诊断的效 果. 为解决上述问题,本文提出了一种新的基于加权图卷积网络的多传感器旋转机械故障诊断方法.利用希尔伯 特–黄变换(HHT)自适应地提取故障特征,克服信号非平稳性的影响.考虑到图结构在空间关系上强大的表达能力 以及图卷积网络强大的特征学习能力,依据传感器节点特征向量之间的距离度量构建加权HHT图,并搭建具有两 层结构的图卷积网络进行故障诊断.同时,在网络损失函数中引入两个正则项以提高诊断的精度.公开数据集上的 实验结果验证了所提出方法的有效性及相比其他方法的优越性.
英文摘要
      Multi-sensor data can provide more comprehensive and accurate information for fault diagnosis, but current deep learning algorithms modeled in Euclidean space have difficulty effectively handling the complex interactions and spa tial relationships between sensors. Additionally, the non-stationary characteristics of vibration signals in rotating machinery greatly affect the effectiveness of fault diagnosis. To address these issues, this paper proposes a novel multi-sensor fault diagnosis method for rotating machinery based on a weighted graph convolutional network. The Hilbert-Huang transform (HHT) is used to adaptively extract fault features, overcoming the impact of signal non-stationarity. Considering the strong expressive power of graph structures in spatial relationships and the powerful feature learning capabilities of graph convo lutional networks, a weighted HHT graph is constructed based on the distance metric between sensor node feature vectors, and a two-layer graph convolutional network is built for fault diagnosis. Additionally, two regularization terms are intro duced into the network’s loss function to improve diagnostic accuracy. Experimental results on public datasets verify the effectiveness and superiority of the proposed method compared to other approaches.