东北大学学报(自然科学版) ›› 2025, Vol. 46 ›› Issue (5): 62-70.DOI: 10.12068/j.issn.1005-3026.2025.20230300

• 机械工程 • 上一篇    下一篇

基于GRM-IConvNeXt模型的滚动轴承故障诊断方法

罗亨发, 于天壮, 周世华()   

  1. 东北大学 机械工程与自动化学院,辽宁 沈阳 110819
  • 收稿日期:2023-11-01 出版日期:2025-05-15 发布日期:2025-08-07
  • 通讯作者: 周世华
  • 作者简介:罗亨发(1998—),男,江西赣州人,东北大学硕士研究生
  • 基金资助:
    中央高校基本科研业务费专项资金资助项目(N2303011);辽宁省自然科学基金资助项目(2022-MS-125)

A Fault Diagnosis Method of Rolling Bearings Based on GRM-IConvNeXt Model

Heng-fa LUO, Tian-zhuang YU, Shi-hua ZHOU()   

  1. School of Mechanical Engineering & Automation,Northeastern University,Shenyang 110819,China.
  • Received:2023-11-01 Online:2025-05-15 Published:2025-08-07
  • Contact: Shi-hua ZHOU

摘要:

针对复杂的轴承运行工况以及传统深度学习故障诊断方法泛化能力弱、模型识别准确率不高等问题,提出了一种基于GRM-IConvNeXt模型的滚动轴承故障诊断方法.首先,提出了一种全局关系矩阵(global relationship matrix, GRM)的编码方法,利用其保留原始信号特征的优点将一维振动信号转换为二维图像.然后,构造了一个针对轴承故障诊断小样本分类的改进ConvNeXt(improved ConvNeXt, IConvNeXt)模型,并选用大小为5×5的卷积核和多个BN层与Hardswish激活函数以强化特征提取性能,同时通过CBAM机制根据GRM图像特征自适应地生成权重.实验结果表明,GRM-IConvNeXt模型在变工况和小样本的情况下都具有良好的特征提取能力和泛化性.

关键词: 滚动轴承, 全局关系矩阵, IConvNeXt模型, CBAM, 故障诊断

Abstract:

Aiming at the problems of complex bearing operating conditions, weak generalization ability and low accuracy of model recognition of traditional deep learning fault diagnosis methods, a rolling bearing fault diagnosis method based on the GRM-IConvNeXt model is established. Firstly, a coding method of global relationship matrix (GRM) is proposed, which can transform one-dimensional vibration signals into two-dimensional images by taking the advantage of preserving the original signal features. Then, an improved ConvNeXt (IConvNeXt) model for small sample classification of bearing fault diagnosis is constructed, and a convolution kernel with a size of 5×5, multiple BN layers and Hardswish activation function are selected to enhance the feature extraction performance. At the same time, weights are adaptively generated according to the GRM image features through the CBAM(convolutional block attention module) mechanism. The experimental results show that the GRM-IConvNeXt model has good feature extraction ability and generalization under off-design conditions and small samples.

Key words: rolling bearing, global relationship matrix(GRM), IConvNeXt model, CBAM(convolutional block attention module), fault diagnosis

中图分类号: