基于Dy Res Net-CBAM网络的滚动轴承剩余寿命预测  

Remaining life prediction of rolling bearing based on Dy Res Net-CBAM network

作　　者：向玲[1] 王凯伦 胡爱军[1] 朱浩伟 周福成 XIANG Ling;WANG Kailun;HU Aijun;ZHU Haowei;ZHOU Fucheng(Department of Mechanical Engineering,North China Electric Power University,Baoding 071003,Hebei,China)

机构地区：[1]华北电力大学机械工程系,河北保定071003

出　　处：《中国工程机械学报》2023年第1期6-11,共6页Chinese Journal of Construction Machinery

基　　金：国家自然科学基金资助项目(52075170)。

摘　　要：滚动轴承的工作状况关系到使用滚动轴承的机械能否正常运行,预测轴承的剩余使用寿命(RUL)是避免机械系统失效的关键。针对传统的轴承使用寿命预测方法无法自适应调节特征权重、提取有用特征,造成预测值误差过大的问题,提出了一种带有卷积块注意力模块(CBAM)的动态残差网络(Dy Res Net)用于预测轴承RUL。对振动信号进行快速傅里叶变换求得频域累积幅值特征,在动态残差网络中加入CBAM模块,并利用压缩激励模块进行特征细化得出预测结果,使用公开轴承数据集对所提模型进行评估。实验表明:与其他模型相比,Dy Res Net-CBAM模型能够充分提取特征信息,对轴承RUL预测的准确度高于其他模型。The health of rolling bearings is related to the normal operation of rotating machinery,predicting the remaining useful life of bearings is a key method to avoid the failure of bearings and their systems. Aiming at the problem that traditional bearing life prediction methods cannot adjust the feature weight adaptively and extract useful features,resulting in large error of predicted values,dynamic residual network with convolution block attention module(CBAM) was proposed to predict bearing remaining useful life(RUL). Firstly,the cumulative amplitude in frequency domain is obtained by fast Fourier transform. Secondly,CBAM is introduced into the dynamic residual network,and SE module is used for feature refinement to get the prediction results. Finally,a public bearing data set is used to evaluate the proposed model. Experimental results show that compared with other models,DyResNet-CBAM model has better prediction ability of bearing residual life,higher accuracy than other models.

关 键 词：滚动轴承 剩余使用寿命 频域累积幅值 卷积块注意力模块(CBAM) 动态残差网络 

分 类 号：TH133.33[机械工程—机械制造及自动化]