基于多源域深度迁移学习的舵机在线故障诊断  

Online fault diagnosing of Rudders based on multi-source domain deep transfer learning

作　　者：吕丞辉 程进军[1] 胡阳光 文斌成 李剑峰 LYU Chenghui;CHENG Jinjun;HU Yangguang;WEN Bincheng;LI Jianfeng(Air Force Engineering University Aeronautics Engineering College,Xi’an 710038,China;The No.63768 th Troop of PLA,Xi’an 710000,China)

机构地区：[1]空军工程大学航空工程学院,西安710038 [2]63768部队,西安710000

出　　处：《兵器装备工程学报》2022年第9期60-67,共8页Journal of Ordnance Equipment Engineering

摘　　要：针对航空武器不同舵机轴承在不同负载力矩下呈现特征数据与工作状态映射关系难以定量表达,开展未知领域的状态识别是一条可行的技术路线;引入多源域深度迁移学习的思想,提出具有多核MMD的MSFAN故障诊断方法。采用傅里叶变换提取不同域原始数据的时频域特征,通过多核MMD距离度量方式减小源域和目标域之间的特征分布差异;利用特定域分类器降低不同域对目标样本在类边界附近的分类损失,提高模型在目标域中的分类精度。试验分别采用公开轴承数据集作为源域数据,使用该方法对目标域数据进行状态识别,与Alxnet、Rexnet18等诊断算法相比,所提方法获得较好的转移性能,基本达到100%的故障识别率。Different steering gear bearings of aviation weapons show different characteristic data under different load moments,and it is difficult to quantitatively express the mapping relationship between the characteristic data and the working state.Under the condition of existing feature data and state mapping,it is a feasible technical route to carry out state recognition in unknown fields.The idea of deep transfer learning in multi-source domains was introduced,and a fault diagnosis method for MSFAN with Multi Kernel MMD was proposed.The Fourier transform was used to extract the time-frequency domain features of the original data in different domains,and the feature distribution difference between the source domain and the target domain was reduced by the Multi Kernel MMD distance measure;The classification loss of samples near the class boundary,which improves the classification accuracy of the model in the target domain.The test used the public bearing data set as the source domain data,and used this method to identify the target domain data.Compared with the diagnosis algorithms such as Alxnet and Rexnet18,the proposed method achieves better transfer performance and basically achieves 100%fault identification Rate.

关 键 词：舵机 轴承 多源域深度迁移学习 MSFAN 故障诊断 

分 类 号：TP18[自动化与计算机技术—控制理论与控制工程]