结合振动信号图像化和RepVGG的滚动轴承故障诊断方法  

Rolling Bearing Fault Diagnosis Method Using Vibration Signal Imaging and RepVGG

作　　者：周建民 王云庆 李家辉 ZHOU Jianmin;WANG Yunqing;LI Jiahui(The Ministry of Education Key Laboratory of Conveyance and Equipment(East China Jiaotong University),Nanchang 330013,China;State Key Laboratory of Rail Transit Infrastructure Performance Monitoring and Guarantee,Nanchang 330013,China)

机构地区：[1]华东交通大学,载运工具与装备教育部重点实验室,南昌330013 [2]轨道交通基础设施性能监测与保障国家重点实验室,南昌330013

出　　处：《机械科学与技术》2024年第12期2024-2031,共8页Mechanical Science and Technology for Aerospace Engineering

基　　金：国家自然科学基金项目(51865010)。

摘　　要：针对滚动轴承故障诊断中一维振动数据的细微特征提取不明显,以及实时诊断速度慢的问题,采用振动数据转图像数据的预处理方法,使用连续小波变换将振动信号转换为二维时频图,同时提出一种基于结构重参数化技术(RepVGG)的轴承故障诊断方法。将训练模型的多分支网络结构等价转换为单路径网络结构,从而提高推理模型的精度和速度。以轴承故障诊断数据集进行实验验证,并与EfficientNet模型和ResNet50模型进行比较分析,结果表明,RepVGG模型能够准确识别轴承故障类别,平均准确率优于其他方法,并且在相同的实验硬件条件下,推理时长分别减少了81%和66.19%,有效提升了诊断的速度和精度,具有较好的适应性和优越性。For fault diagnosis problems of rolling bearing that the nuances of one-dimensional vibration data feature extraction is not obvious and real-time diagnosis speed is low,using vibration data and image data pretreatment method,the vibration signal can be converted into 2D time-frequency diagrams using the continuous wavelet transform,a kind of based on the re-parametric technology(RepVGG)structure of bearing fault diagnosis method is put forward.The multi-branch network structure of the training model is equivalent to the single-path network structure,so as to improve the accuracy and speed of the inference model.The experimental verification is done on the bearing fault data set.The results show that the RepVGG model can accurately identify bearing fault categories,and the average accuracy rate is better than other methods.Moreover,under the same experimental hardware conditions,the RepVGG model is effective and efficient.The reasoning time is reduced by 81%and 66.19%respectively,which effectively improves the speed and accuracy of fault diagnosis,and has good adaptability and superiority.

关 键 词：轴承 故障诊断 时频图 RepVGG模型 

分 类 号：TH165.3[机械工程—机械制造及自动化] TH133.33