基于C-GRU的电机轴承故障诊断  

Fault Diagnosis of Motor Bearing Based on C-GRU

作　　者：许爱华[1] 杜洋 段玉波[1] 许瀚锋 XU Ai-hua;DU Yang;DUAN Yu-bo;XU Han-feng(School of Electrical Engineering & Information,Northeast Petroleum University,Daqing Heilongjiang 163318,China;State Grid Shandong Electric Power Company Linyi County Power Supply Company, Dezhou Shandong 251500,China)

机构地区：[1]东北石油大学电气信息工程学院,黑龙江大庆163318 [2]国网山东省电力公司临邑县供电公司,山东德州251500

出　　处：《组合机床与自动化加工技术》2021年第2期47-51,共5页Modular Machine Tool & Automatic Manufacturing Technique

基　　金：国家自然科学基金资助项目(51774088);黑龙江省自然基金资助项目(JJ2019LH0187);东北石油大学引导性创新基金(2019YDL-10)。

摘　　要：针对单一卷积神经网络模型在轴承故障诊断工作对于训练样本需求过多的不足,根据采集到的电机轴承振动数据为时序数据的特点,结合门控循环单元在处理时序数据所具有的优势,采用了基于卷积神经网络和门控循环单元(C-GRU)的电机轴承故障诊断算法。将CNN在特征提取的优点与GRU处理时序数据的优点有机结合起来,在选择合适的网络结构和参数后,深入挖掘振动信号中所蕴含的时序特征。通过选用不同数据集实验对比,在训练样本减少的同时,所使用模型的最终分类准确率分别达到了99.7%与99.6%,验证了文中所采用的模型的可靠性与实用性。Aiming at the problem that the single convolutional neural network model requires too many training samples for bearing fault diagnosis.Based on the characteristics that the collected vibration data of motor bearings are time series data,combined with the advantages of the gated loop unit in processing time series data,the motor bearing fault diagnosis algorithm based on convolutional neural network and gated loop unit is adopted.Combine the advantages of CNN in feature extraction with the advantages of GRU in processing time series data.After selecting the appropriate network structure and parameters,the time series features contained in the vibration signal can be deeply explored.By choosing different data sets for experimental comparison,while the training samples are reduced,the final classification accuracy rates of the models used in the article have reached 99.7%and 99.6%respectively,verifying the reliability and practicability of the models used in the article.

关 键 词：轴承 故障诊断 卷积神经网络 门控循环单元 

分 类 号：TH16[机械工程—机械制造及自动化] TG65[金属学及工艺—金属切削加工及机床]