基于改进联合分布适配和支持向量机的谐波减速器故障诊断  

Fault diagnosis of harmonic reducer based on improved joint distribution adaptation-support vector machine

作　　者：石超 刘彪 郭世杰[1,2] 唐术锋 吕贺[1] SHI Chao;LIU Biao;GUO Shijie;TANG Shufeng;LV He(School of Mechanical Engineering,Inner Mongolia University of Technology,Hohhot 010051,China;Inner Mongolia Key Laboratory of Robotics and Intelligent Equipment Technology,Hohhot 010051,China)

机构地区：[1]内蒙古工业大学机械工程学院,内蒙古呼和浩特010051 [2]内蒙古自治区机器人与智能装备技术重点实验室,内蒙古呼和浩特010051

出　　处：《机电工程》2025年第3期441-450,共10页Journal of Mechanical & Electrical Engineering

基　　金：国家重点研发计划项目(2018YFB1307501);国家自然科学基金资助项目(52065053,52365064);中央引导地方科技发展专项(2020ZY0002);内蒙古关键技术攻关项目(2020GG0255);内蒙古自然科学基金资助项目(2022FX01,2023LHMS05018);内蒙古自治区高等学校科学研究项目(NJZY21308);内蒙古自治区直属高校基本科研业务费资助项目(JY20220046);内蒙古自治区高等学校青年科技英才支持计划项目(NJYT23043);内蒙古自治区高等学校创新团队发展支持计划支持项目(NMGIRT2213);内蒙古自治区科技计划项目(2021GG0259,2021GG0255)。

摘　　要：在对谐波减速器进行变工况故障诊断时,一般难以获得大量的带标签数据,从而导致所训练的模型识别准确率较低。针对这一问题,提出了一种基于改进联合分布适配和支持向量机的迁移模型(方法),从而对谐波减速器进行了故障诊断。首先,对周期样本进行了时域、频域以及熵特征的多特征提取,构造了样本集;然后,针对联合适配(JDA)对齐两域状态下,未考虑到数据潜在的几何结构问题,在JDA的基础上增加了联合分布的权重因子以及加权流形正则化项,并使用支持向量机(SVM)进行了伪标签的迭代更新,构造了改进联合分布适配-支持向量机(IJDA-SVM)迁移模型;最后,使用实验所得的谐波减速器振动信号数据以及滚动轴承公开数据集对该方法的有效性进行了验证。研究结果表明:IJDA-SVM在谐波减速器单域诊断效果上,最高识别率可达97.25%,平均识别率为94.08%,在谐波减速器多域诊断效果上,最高识别率可达95.25%,平均识别率为92.5%。采用该方法能够实现变工况谐波减速器的故障诊断目的,其具有诊断精度高、泛化效果好的优点。Aiming at the problem that it was difficult to obtain a large amount of labeled data when diagnosing the fault of harmonic reducer under variable working conditions,it leads to the low recognition accuracy of the trained model.A fault diagnosis method of harmonic reducer based on multi-feature construction and transfer learning was proposed.Firstly,the multi-feature extraction of time domain,frequency domain and entropy features was performed on the periodic samples to construct the sample set.Then,aiming at the problem that the potential geometric structure of the data was not considered in the joint distribution adaptation(JDA)aligned two-domain state,the weight factor of the joint distribution and the weighted Manifold Regularization(MR)term were added on the basis of the JDA,and the support vector machine(SVM)was used to iteratively update the pseudo label,and the improved joint distribution adaptation-support vector machine(IJDA-SVM)migration model was constructed.Finally,the vibration signal data of the harmonic reducer obtained from the experiment and the public data set of the rolling bearing were used for verification.The experimental results show that the highest recognition rate of IJDA-SVM can reach 97.25%and the average recognition rate is 94.08%in the single domain diagnosis effect of harmonic reducer.In the multi domain diagnosis effect of harmonic reducer,the highest recognition rate can reach 95.25%and the average recognition rate is 92.5%.This method can realize the fault diagnosis of harmonic reducer under variable working conditions,and its diagnosis accuracy is high and the generalization effect is good.

关 键 词：变速器 多域故障诊断 变工况 迁移学习 改进联合分布适配-支持向量机 流形正则化 

分 类 号：TH132.46[机械工程—机械制造及自动化]