摘要
针对变工况下的滚动轴承无法获得大量带标签样本数据以及传统深度学习诊断方法识别率低的问题,提出一种基于迁移学习的卷积神经网络模型滚动轴承故障诊断方法。首先,采用短时傅里叶变换处理滚动轴承振动信号获得源域、目标域样本集;其次,利用源域样本预训练卷积神经网络模型;最后,通过目标域样本微调卷积神经网络模型实现滚动轴承故障诊断。通过2种不同滚动轴承振动数据进行迁移故障诊断实验,实验结果表明:相对于卷积神经网络的故障诊断方法,基于迁移学习的卷积神经网络故障诊断识别率提高了7%。
Aiming at the problem that rolling bearings under variable operating conditions cannot obtain a large number of labeled sample data and the low recognition rate of traditional deep learning diagnostic methods,a convolutional neural network rolling bearing fault diagnosis method based on transfer learning is proposed.First,the short-time Fourier transform is used to process the vibration signal of the rolling bearing to obtain the source domain and target domain sample sets;second,the source domain samples are used to pre-train the convolutional neural network model;finally,the target domain samples are used to fine-tune the convolutional neural network model to implement the rolling bearing troubleshooting.Two different rolling bearing vibration data are used to carry out migration fault diagnosis experiments.The experimental results show that:compared with the fault diagnosis method of convolutional neural network,the fault diagnosis recognition rate of convolutional neural network based on transfer learning is increased by 7%.
作者
唐波
陈慎慎
郭必奔
郝家琦
TANG Bo;CHEN Shen-shen;GUO Bi-ben;HAO Jia-qi(College of Metrology and Measurement Engineering,China Jiliang University,Hangzhou,Zhejiang 310018,China)
出处
《计量学报》
CSCD
北大核心
2022年第3期386-391,共6页
Acta Metrologica Sinica
基金
国家自然科学基金(11872061)
浙江省自然科学基金(LY21E050017)
国家市场监督管理总局科技计划(2020MK189)。
关键词
计量学
滚动轴承
卷积神经网络
迁移学习
故障诊断
metrology
rolling bearing
convolutional neural network
transfer learning
fault diagnosis