• CN:11-2187/TH
  • ISSN:0577-6686

机械工程学报 ›› 2026, Vol. 62 ›› Issue (18): 268-286.doi: 10.3901/JME.260771

• 特邀专栏:大模型驱动的高端装备智能运维 • 上一篇    

扫码分享

图知识赋能残差空时信息融合的轴承可信故障诊断方法

杨同光1,2, 周坪1,2, 李学军3, 韩清凯3, 马辉3, 朱真才1,2   

  1. 1. 中国矿业大学机电工程学院 徐州 221116;
    2. 智能采矿装备技术全国重点实验室 徐州 221116;
    3. 东北大学佛山研究生创新学院 佛山 528312
  • 收稿日期:2025-11-07 修回日期:2026-06-30 发布日期:2026-11-20
  • 作者简介:杨同光,男,1995年出生,博士,助理研究员。主要研究方向为旋转机械故障诊断、状态评估与预测技术。E-mail:15242208931@163. com;周坪(通信作者),男,1992年出生,博士,副教授,硕士研究生导师。主要研究方向为机械装备健康监测、无损检测。E-mail:zhoup@cumt. edu. Cn
  • 基金资助:
    中央高校青年教师科研创新能力(ZYGXQNJSKYCXNLZCXM-E2P)、国家重点研发计划(2020YFB2007800)、江苏省自然科学基金(BK20231068)和江苏省优势学科建设(PAPD)资助项目。

Graph Knowledge-empowered Residual Spatiotemporal Information Fusion for Credible Bearing Fault Diagnosis

YANG Tongguan1,2, ZHOU Ping1,2, LI Xuejun3, HAN Qingkai3, MA Hui3, ZHU Zhencai1,2   

  1. 1. College of Mechanical and Electrical Engineering, China University of Mining and Technology Xuzhou 221116;
    2. National Key Laboratory of Intelligent Mining and Equipment Technology, Xuzhou 221116;
    3. Key Laboratory of Rotor Vibration Monitoring and Diagnosis Technology for Machinery Industry, School of Foshan Graduate Innovation, Northeastern University, Foshan 528312
  • Received:2025-11-07 Revised:2026-06-30 Published:2026-11-20

摘要: 深度学习凭借复杂结构实现了机械故障数据间非线性关系的深层挖掘。然而,现有深度学习模型对风电机组的发电机轴承不同健康状态的电蚀故障存在难以识别的难题。此外,传统图神经网络模型不同特征维度之间存在过相关现象,导致邻居增益存在信息冗余度高的问题。为了填补上述缺陷,开发一种图知识赋能的残差空时信息融合的发电机轴承诊断模型,其着眼于解决实际工程中轴电流损伤导致轴承不同健康状态电蚀故障无法准确识别的问题。具体来讲,创新性地设计一种新的图结构,并且推导出最大互信息系数公式,构建新的邻接矩阵并证明该矩阵的有效性。基于该图结构设计空间信息融合模块,使得所提模型具有很强的推理和表达能力。设计时间信息融合模块,从而深度挖掘不同健康状态轴承电蚀故障的时序信息。同时设计特征关注模块,对空时域提取的轴承故障特征实现信息互补。最后,基于数据集对所提模型的性能进行验证,结果表明,所提模型具有良好的优越性和可信性,为轴承故障提供一种识别精度更高的新思路。

关键词: 智能诊断, 发电机轴承, 空时信息融合, 可解释性

Abstract: Deep learning realizes the deep mining of the nonlinear relationship between mechanical fault data by virtue of complex structure. However, the existing deep learning models are difficult to identify the electrical erosion faults of generator bearings in different health states of wind turbines. In addition, there is an over-correlation between different feature dimensions of the traditional graph neural network model, which leads to the problem of high information redundancy in neighbor gain. In order to fill the above defects, a generator bearing diagnosis model based on graph knowledge-enabled residual space-time information fusion is developed, which focuses on solving the problem that the shaft current damage leads to the inability to accurately identify the electrical corrosion faults of bearings in different health states in practical engineering. Specifically, a new graph structure is innovatively designed, and the maximum mutual information coefficient formula is derived. A new adjacency matrix is constructed and the validity of the matrix is proved. Based on the graph structure, a spatial information fusion module is designed, which makes the proposed model have strong reasoning and expression ability. A time information fusion module is designed to deeply mine the time series information of bearing electrical erosion faults in different health states. At the same time, a feature attention module is designed to complement the information of bearing fault features extracted in space-time domain. Finally, the performance of the proposed model is verified based on the data set. The results show that the proposed model has good superiority and credibility, and provides a new idea with higher recognition accuracy for bearing faults.

Key words: intelligent diagnosis, generator bearings, spatiotemporal information fusion, interpretability

中图分类号: