基于改进深度残差网络的轴承故障诊断方法  

Bearing Fault Diagnosis Method Based on Improved Depth Residual Network

作　　者：高淑芝[1] 韩晓亮 张义民[1] GAO Shuzhi;HAN Xiaoliang;ZHANG Yimin(Institute of Equipment Reliability,Shenyang University of Chemical Technology,Liaoning Shenyang 110142,China;College of Information Engineering,Shenyang University of Chemical Technology,Liaoning Shenyang 110142,China)

机构地区：[1]沈阳化工大学装备可靠性研究所,辽宁沈阳110142 [2]沈阳化工大学信息工程学院,辽宁沈阳110142

出　　处：《机械设计与制造》2025年第3期241-244,249,共5页Machinery Design & Manufacture

基　　金：NSFC-国家自然科学重点基金-辽宁联合基金(U1708254);辽宁省特聘教授(No.[2018]3533)项目。

摘　　要：针对卷积神经网络结构因深度的增加导致的网络退化以及准确率饱和问题,本文改进深度残差网络应用于故障诊断。首先,改进的残差网络包含三个残差单元模块,改进后的残差结构去掉了不必要的非线性层,在模块首尾都加入批量归一化层提升了网络性能;其次,采集的轴承故障样本分为训练集与测试集,将训练集数据样本输入到网络模型中进行训练优化,输入测试集数据到诊断模型中得出诊断结果;最后,利用t-SNE可视化方法对模型中每一个残差模块学习特征的过程进行分析。经轴承寿命试验台数据结果表明,本方法对滚动轴承发生故障的诊断识别率均达到100%。可见该模型具有非常好的诊断识别效果。Aiming at the network degradation and accuracy saturation caused by the increase of depth of convolutional neural network structure,it improves the depth residual network and applies it to fault diagnosis.Firstly,the improved residual network includes three residual unit modules.The improved residual structure removes the unnecessary nonlinear layer,and adds batch normalization layer at the beginning and end of the module to improve the network performance;Secondly,the collected bearing fault samples are divided into training set and test set.The training set data samples are input into the network model for training optimization,and the test set data are input into the diagnosis model to obtain the diagnosis results;Finally,the t-SNE visualization method is used to analyze the learning process of each residual module in the model.The results of bearing life test-bed show that the diagnosis and recognition rate of rolling bearing faults by this method is 100%.It can be seen that the model has a very good diagnosis and recognition effect.

关 键 词：滚动轴承 故障诊断 深度残差网络 t-SNE可视化 

分 类 号：TH16[机械工程—机械制造及自动化] TP18[自动化与计算机技术—控制理论与控制工程]