基于多源域混淆蒸馏的变工况轴承故障诊断方法  

Fault Diagnosis Method for Variable Operating Condition Bearings Based on Confusing Distillation of Multi-Source Domains

作　　者：丁建建 朱晓娟[1] DING Jianjian;ZHU Xiaojuan(School of Computer Science and Engineering,Anhui University of Science and Technology,Huainan Anhui 232001,China)

机构地区：[1]安徽理工大学计算机科学与工程学院,安徽淮南232001

出　　处：《重庆科技学院学报（自然科学版）》2024年第3期49-55,共7页Journal of Chongqing University of Science and Technology：Natural Sciences Edition

基　　金：安徽省高校自然科学研究项目“智能煤矿中设备状态感知与诊断研究”(KJ2020A0300)。

摘　　要：针对新工况下标记的故障样本不足和信号分布有差异等问题,提出基于多源域混淆蒸馏的轴承变工况故障诊断方法,将域混淆方法与多教师知识蒸馏方法相融合。首先,令每个源域模型在互相蒸馏的过程中混淆所提取的特征,在提取源域不变信息的同时抑制源域特异信息;然后,在目标域模型中蒸馏整合所有源域不变信息;最后,利用目标域数据进行模型迁移,以有效改善多个源域迁移时效果不稳定的现象。经实验验证,该方法可以通过多种已知工况数据实现对目标工况数据的故障诊断,准确率较高。To address the problems such as the lack of sufficient marker fault samples under the new condition and the difference in signal distribution,A fault diagnosis method for variable operating condition bearings based on confusing distillation of multi-source domains was proposed.The method involving the idea of domain obfuscation was integrated into the multi-teacher knowledge distillation.Firstly,the extracted features in the process of distilling each other was obfuscated in each source domain model and source domain invariant information was extracted while source domain specific information was suppressed.Then,all source domain invariant information was integrated by distilling in the target domain model.Finally,the target domain data was used for model migration in order to effectively solve the problem of unstable effect when migrating multiple source domains.It is experimentally verified that the proposed method can diagnosis fault in the data of the target working condition through multiple known working condition data with high accuracy.

关 键 词：变工况 知识蒸馏 多源域迁移 域混淆 故障诊断 

分 类 号：TH133[机械工程—机械制造及自动化]