基于SK-ResNet和迁移学习的滚动轴承故障诊断  

Rolling Bearing Fault Diagnosis Based on SK-ResNet and Transfer Learning

作　　者：潘雪娇 董绍江[2] 邹松[1] 吕智明 宋锴 PAN Xuejiao;DONG Shaojiang;ZOU Song;LYU Zhiming;SONG Kai(College of Traffic&Transportation,Chongqing Jiaotong University,Chongqing 400074,China;School of Mechatronics and Vehicle Engineering,Chongqing Jiaotong University,Chongqing 400074,China;Chongqing Changan Automobile Co.,Ltd.,Chongqing 401120,China)

机构地区：[1]重庆交通大学交通运输学院,重庆400074 [2]重庆交通大学机电与车辆工程学院,重庆400074 [3]重庆长安汽车股份有限公司,重庆401120

出　　处：《组合机床与自动化加工技术》2024年第10期166-170,共5页Modular Machine Tool & Automatic Manufacturing Technique

基　　金：国家自然科学基金资助项目(51775072);重庆市科技创新领军人才支持计划项目(CSTCCCXLJRC201920)。

摘　　要：针对传统深度学习模型在变工况环境下泛化能力差、诊断精度低的问题,提出了一种基于SK-ResNet和迁移学习的滚动轴承故障诊断方法。首先,对采集到的时域信号进行快速傅里叶变换(FFT)获得频域信号,并进行加权融合得到新的时频域数据集;其次,将选择性内核网络(SKNet)融入到残差网络(ResNet)中提高特征提取能力;然后,采用基于多核最大均值差异(MK-MMD)和相关对齐(CORAL)改进的差异对齐损失(DDM)缩小变工况下滚动轴承故障数据特征分布差异,并将其应用到模型的多个模块中进一步缩小特征之间的分布距离。实验结果表明,与传统滚动轴承故障诊断方法相比,本文方法具有更好的诊断精度和泛化能力。Aiming at the problems of poor generalization ability and low diagnosis accuracy of traditional deep learning model under variable working conditions,a fault diagnosis method of rolling bearing based on SK-ResNet and transfer learning was proposed.First,the frequency domain signal is obtained by fast Fourier transform(FFT),and a new time-frequency domain data set is obtained by weighted fusion.Secondly,the selective kernel network(SKNet)is integrated into the residual network(ResNet)to improve the feature extraction capability.Then,the differential alignment loss(DDM)based on multi-core maximum mean difference(MK-MMD)and correlated alignment(CORAL)is used to reduce the feature distribution difference of rolling bearing fault data under varying working conditions,and it is applied to multiple modules of the model to further reduce the distribution distance between features.The experimental results show that the proposed method has better diagnostic accuracy and generalization ability than the traditional rolling bearing fault diagnosis method.

关 键 词：选择性内核网络 残差网络 迁移学习 差异对齐损失 

分 类 号：TH133.3[机械工程—机械制造及自动化] TG66[金属学及工艺—金属切削加工及机床]