摘要
针对机械设备的状态监测和故障诊断面临的先验样本数据少、样本空间不完备的"小数据"困境,提出了基于测地流核函数的域自适应故障诊断方法。以有限的先验样本数据作为源域,以实际监测数据作为目标域,分别提取设备状态特征并将特征分布子空间嵌入格拉斯曼流形,基于测地流核函数对源域和目标域在特征分布结构上的相似性进行度量,从而实现域自适应基础上的故障诊断。基于轴承振动数据的试验验证表明,基于测地流核函数的域自适应故障诊断能够有效抑制工况变化、采样母体差异的影响,提高故障诊断正确率。
The shortage of prior faulty data or the incompleteness of sample space,which leads to a"small data"trap,is a common situation encountered when implementing intelligent condition monitoring and fault diagnosis of machines.To overcome this problem,a domain adaptive fault diagnostic scheme was proposed based on the geodesic flow kernel method.With prior data samples as the source domain and monitored data as the target domain,status features of the machine were extracted and selected separately to construct the subspaces of machine conditions.By embedding the subspaces in a Grassmann manifold,the structural similarity of subspaces was evaluated based on the geodesic flow kernel to achieve the domain adaptive fault diagnosis.The verification based on the bearing vibration data demonstrates that the domain adaptive fault diagnosis based on the geodesic flow kernel can effectively reduce the impact of the variety in working conditions and the physical differences underlying sampling populations,so that the accuracy of fault diagnosis can be improved.
作者
刘海宁
宋方臻
窦仁杰
黄亦翔
刘成良
LIU Haining;SONG Fangzhen;DOU Renjie;HUANG Yixiang;LIU Chengliang(School of Mechanical Engineering,Jinan University,Jinan 250022,China;School of Mechanical Engineering,Shanghai Jiao Tong University,Shanghai 200240,China)
出处
《振动与冲击》
EI
CSCD
北大核心
2018年第18期36-42,共7页
Journal of Vibration and Shock
基金
国家自然科学基金(51605191)
山东省中青年科学家科研奖励基金(ZR2016EEB37)
济南大学博士基金(XBS1524)
关键词
域自适应
故障诊断
小数据
测地流核函数
domain adaptation
fault diagnosis
small data
geodesic flow kernel
