基于ECAProNet模型的小样本变工况轴承故障诊断
Bearing Fault Diagnosis under Variable Working Conditions with Few Samples Based on ECAProNet Model
王勉 1吴东升 1王笑臣1
作者信息
- 1. 沈阳理工大学 自动化与电气工程学院,沈阳 110159
- 折叠
摘要
针对小样本变工况轴承故障诊断中易出现过拟合的问题,提出一种在原型网络(Pro-Net)中加入高效通道注意力机制(ECA)的轴承故障诊断方法(ECAProNet).基于度量学习的思想,在改进原型网络框架下将通过特征提取模块的样本信号映射至特征度量空间,在该空间内确定查询样本与各类原型间的欧式距离,得到损失函数,进而优化特征提取网络框架;为达到元学习目的,采用基于episodes的训练策略,将算法泛化到不同工况的测试诊断中.在CW-RU数据集上设置5-way 5-shot和5-way 1-shot验证实验,结果表明,ECAProNet在小样本变工况轴承故障诊断中表现出较好的性能.
Abstract
For the problem of over-fitting in bearing fault diagnosis under variable working condi-tions in the case of few samples,a bearing fault diagnosis method(ECAProNet)with efficient chan-nel attention mechanism(ECA)added to the prototype network(ProNet)is proposed.Based on the idea of metric learning,the sample signal of the feature extraction module is mapped to the feature metric space under the framework of the improved prototype network.In this space,the Euclidean distance between the query sample and various prototypes is determined,and the loss function is ob-tained to optimize the feature extraction network framework.In order to achieve the purpose of me-ta-learning,the training strategy based on episodes is used to generalize the algorithm to the test di-agnosis of different working conditions.The 5-way 5-shot and 5-way 1-shot verification experiments are set on the CWRU dataset.The results show that ECAProNet has good performance in bearing fault diagnosis with few samples and variable working conditions.
关键词
故障诊断/小样本学习/变工况/原型网络/高效通道注意力机制Key words
fault diagnosis/few-shot learning/variable working conditions/prototypical network/efficient channel attention引用本文复制引用
基金项目
辽宁省教育厅高等学校基本科研项目(LJKMZ20220618)
出版年
2024
NETL
NSTL
万方数据