基于边云协同的数控机床故障诊断联邦学习研究  

Research on Federated Learning of Numerical Control Machine Fault Diagnosis Based on Edge Cloud Cooperation

作　　者：徐玲艳 陆艺 赵静 XU Lingyan;LU Yi;ZHAO Jing(College of Metrology and Measurement Engineering,China Jiliang University,Hangzhou,Zhejiang 310018,China;Hangzhou Wolei Intelligent Technology Co.Ltd,Hangzhou,Zhejiang 310018,China)

机构地区：[1]中国计量大学计量测试工程学院,浙江杭州310018 [2]杭州沃镭智能科技股份有限公司,浙江杭州310018

出　　处：《计量学报》2024年第6期873-880,共8页Acta Metrologica Sinica

基　　金：浙江省科技计划项目省级重点研发计划(2021 C01136)。

摘　　要：为解决数控机床主轴轴承和刀具在进行故障诊断模型训练时需要大量数据且耗时长的问题,提出一种基于边云协同架构的联邦平均学习故障诊断模型。首先,设计一维卷积神经网络模型架构,在各边缘客户端进行本地模型训练,以减小数据上传规模和分担云服务器端计算压力;其次,在云服务器端基于准确率优化模型聚合算法,改进边缘客户端筛选算法,以加快模型收敛速度,提高模型准确率;再次,在云服务器端搭建基于Kubernetes的KubeEdge边云协同平台,以缩短数据传输的通信时间。实验结果表明,模型在各边缘客户端故障诊断的准确率最终稳定在87.5%左右;且收敛速度、训练时长等方面与对照组相比,均获得优化。In order to solve the problem when the spindle bearing and cutter fault diagnosis model of NC machine tool is trained,it needs massive data and takes a long time.A federated mean learning fault diagnosis model based on edge-cloud collaborative architecture is proposed.Firstly,one-dimensional convolutional neural network model architecture is designed,and local model training is carried out on each edge client to reduce data upload scale and share computing pressure on cloud server side.Then,the model aggregation algorithm is optimized based on the accuracy on the cloud server side,and the edge client screening algorithm is improved to accelerate the convergence rate and improve the accuracy of the model.Thirdly,the KubeEdge cloud collaborative platform based on Kubernetes is built on the cloud server side to shorten the communication time of data transmission.Finally,the experimental results show that the accuracy of fault diagnosis in each edge client of the model is stable at about 87.5%.Compared with the control group,the convergence speed and training time are optimized.

关 键 词：故障诊断 数控机床 边云协同 联邦学习 振动信号 

分 类 号：TB936[一般工业技术—计量学] TB973[机械工程—测试计量技术及仪器]