基于AMCNN-BiGRU的滚动轴承故障诊断方法研究  

Fault diagnosis method for rolling bearings based on AMCNN-BiGRU

作　　者：徐鹏 皋军[2] 邵星[2] XU Peng;GAO Jun;SHAO Xing(School of Mechanical Engineering,Yancheng Institute of Technology,Yancheng 224051,China;School of Information Engineering,Yancheng Institute of Technology,Yancheng 224051,China)

机构地区：[1]盐城工学院机械工程学院,江苏盐城224051 [2]盐城工学院信息工程学院,江苏盐城224051

出　　处：《振动与冲击》2023年第18期71-80,共10页Journal of Vibration and Shock

基　　金：国家自然科学基金(62076215);教育部新一代信息技术创新项目(2020ITA02057)。

摘　　要：为克服传统滚动轴承故障诊断方法需要人工提取特征的缺点,提出一种基于注意力模块的卷积神经网络-双向门控循环单元的滚动轴承故障诊断方法。该方法利用下采样后的原始振动信号作为输入,通过具有两种不同核大小的并行卷积块从采样后的数据中提取特征,并使用注意力模块对提取的特征进行加权融合处理,最后将具有不同权重的特征输入到双向门控循环单元进行故障分类,从而实现端到端的诊断。为了理解所提出模型的诊断过程,对所学习的特征进行可视化,分析发现模型可以有效映射不同类型的故障。经试验表明,该模型使用下采样后的原始数据有效缩短了网络的训练时间,同时还可以保持100%的诊断准确率。In order to overcome the disadvantage of manual feature extraction in traditional rolling bearing fault diagnosis methods,a rolling bearing fault diagnosis method using the convolution neural network-bi-directional gated cycle unit(AMCNN-BiGRU)based on an attention module was proposed.In the method,the original vibration signal after downsampling was used as input,features were extracted from the original data by parallel convolution blocks with two different core sizes,and the extracted features were weighted and fused by the attention module.Finally,the features with different weights were input to the bi-directional gated cycle unit for fault classification,so as to realize end-to-end diagnosis.In order to understand the diagnosis process of the proposed model,the learned features were visualized,and it is found that the model can effectively map different faults.The experimental results show that the model can effectively shorten the network training time and maintain 100%diagnostic accuracy by using the downsampled original data.

关 键 词：卷积神经网络 门控循环单元 注意力机制 轴承故障诊断 可视化 

分 类 号：TH212[机械工程—机械制造及自动化] TH213.3