用聚类与插值改进深度学习算法实现变工况轴承故障诊断  

Improved Deep Learning Algorithm Based on CAI for Bearing Fault Diagnosis Under Variable Operating Conditions

作　　者：李俊卿[1] 耿继亚 胡晓东 张承志 何玉灵[2] LI Junqing;GENG Jiya;HU Xiaodong;ZHANG Chengzhi;HE Yuling(Department of Electrical Engineering,North China Electric Power University,Baoding 071003,China;School of Energy Power and Mechanical Engineering,North China Electric Power University,Baoding 071003,China)

机构地区：[1]华北电力大学电力工程系,河北保定071003 [2]华北电力大学能源动力与机械工程学院,河北保定071003

出　　处：《电力科学与工程》2024年第6期60-68,共9页Electric Power Science and Engineering

基　　金：国家自然科学基金资助项目(52177042);河北省自然科学基金资助项目(E2020502032);中央高校基本科研版费专项基金资助项目(2020MS114)。

摘　　要：针对基于深度学习轴承故障诊断模型由于工况因素导致诊断效果不佳的问题,提出了一种用聚类与插值(Clustering and interpolation,CAI)改进深度学习算法实现变工况轴承故障诊断的方法。首先,采用有限元法仿真多工况、多故障类型的轴承振动信号数据,获取足够样本;然后,完成宽卷积核深度卷积神经网络(Deepconvolutionalneuralnetworks with widekernel,WDCNN)模型构建,并利用任一工况下的数据完成模型训练;最后,利用CAI算法统一其余工况数据的转速信息,调用WDCNN模型完成对其余工况样本的故障诊断。结果显示,WDCNN模型对训练数据所属工况故障诊断准确率达99.9%,对经过CAI算法处理其他工况数据故障诊断识别率分别为98.7%、99.2%,是一种简单、准确有效、泛化能力强的故障诊断方法。A method of using clustering and interpolation(CAI)to improve deep learning algorithm for bearing fault diagnosis under variable operating conditions is proposed to address the problem of poor diagnostic performance of bearing fault diagnosis model based on deep learning due to operating conditions.Firstly,finite element method is used to simulate bearing vibration signal data with multiple operating conditions and fault types to obtain sufficient samples;Secondly,complete the construction of deep convolutional neural networks with wide kernel(WDCNN)model,and use data from any operating condition to complete model training;Finally,the CAI algorithm is used to unify the data information of the remaining operating conditions,and the WDCNN model is called to complete the fault diagnosis of the remaining operating conditions samples.The results show that the WDCNN model has a fault diagnosis accuracy of 99.9%for the working condition to which the training data belongs,and the fault diagnosis recognition rates for other working condition data processed by the CAI algorithm are 98.7%and 99.2%respectively.It is a simple,accurate and effective fault diagnosis method with strong generalization ability.

关 键 词：深度学习 聚类与插值算法 故障诊断 轴承 有限元分析 

分 类 号：TH13[机械工程—机械制造及自动化] TH17