###
工程科学与技术:2024,56(1):256-266
←前一篇   |   后一篇→
本文二维码信息
码上扫一扫!
基于类对比簇分配异构迁移学习的空间滚动轴承寿命阶段识别
(1.四川大学 机械工程学院,四川 成都 610065;2.重庆大学 机械传动国家重点实验室,重庆 400044)
Life Stage Identification of Space Rolling Bearings Based on Class-contrast Cluster-allocation Heterogeneous Transfer Learning
(1.School of Mechanical Eng., Sichuan Univ., Chengdu 610065, China;2.The State Key Lab. of Mechanical Transmissions, Chongqing Univ., Chongqing 400044, China)
摘要
图/表
参考文献
相似文献
附件
本文已被:浏览 152次   下载 128
投稿时间:2023-04-16    
中文摘要: 针对变工况条件下因样本数据分布差异大、可训练用样本较少以及不同寿命阶段样本数量不均等造成的空间滚动轴承寿命阶段识别准确率较低的问题,提出一种无监督迁移学习方法--类对比簇分配异构迁移学习(CAHTL)。在CAHTL中,通过异构迁移学习将历史工况下少量有类标签样本和当前工况的无类标签样本(即待测样本)迁移到公共特征空间内,使得不同工况样本之间的分布差异最小化;利用源域聚类簇点构建目标域样本特征的正负样本实现两域样本的数量再分配,再对两域正负样本进行对比学习以使待测样本分类性更好;通过计算待测样本与聚类簇点的相似度完成待测样本分类,且该分类过程无需参数学习,因此可避免样本不均等情况下对于不同寿命阶段样本识别准确率差距过大和在少有类标签训练样本情况下网络出现过拟合的问题;利用随机梯度下降和动量更新对CAHTL参数进行不同步更新,以保持样本特征的一致性并提高CAHTL的收敛速度。CAHTL可利用空间滚动轴承历史工况下的少量、非均等的已知寿命阶段的训练样本对当前工况的待测样本进行较高精度的寿命阶段识别。空间滚动轴承寿命阶段识别实例验证了该方法的有效性。
Abstract:Life stage identification accuracy of space rolling bearings is low due to the large difference of sample distribution, the small number of available training samples, and the unequal number of samples at different life stages under variable working conditions. Therefore, this paper proposes a novel unsupervised transfer learning method called class-contrast cluster-allocation heterogeneous transfer learning (CAHTL). This method firstly transfers a small number of labeled samples under historical working conditions and the unlabeled samples (i.e., the testing samples) under the current working conditions into a public feature space through heterogeneous transfer learning, so as to minimize the distribution difference between the samples under different working conditions. After that, the positive and negative samples of sample features in target domain are constructed by using cluster points in source domain to achieve the number redistribution of two domain samples, and then contrastive learning on the positive and negative samples in two domains is carried out to make a better classification characteristic on the testing samples. Then, the classification of testing samples is completed by calculating the similarities between the testing samples and the cluster points without parameter learning, which can prevent the large difference in identification accuracy of samples at different life stages in the case of unequal samples and the over fitting of CAHTL in the case of few labeled training samples. Finally, the stochastic gradient descent and momentum renewal are used to asynchronously update the CAHTL parameters to maintain the consistency of sample features and improve the convergence speed of CAHTL. CAHTL can use few and unequal training samples at known life stages under historical working conditions of space rolling bearings to identify the life stages of testing samples under the current working conditions with high accuracy. The effectiveness of the proposed CAHTL is verified by an experiment of identifying the life stage of a space rolling bearing.
文章编号:202300291     中图分类号:TH113;TP391.4    文献标志码:
基金项目:中央高校基本科研业务费(2022CDZG–12);机械传动国家重点实验室开放基金资助项目(SKLMT–KFKT–201718);四川省重点研发项目(2020KJT0117–2020YFQ0039)
作者简介:第一作者:刘峰良(1998-),男,硕士生.研究方向:机械设备状态监测;故障诊断.E-mail:1562021338@qq.com;通信作者:李锋,副教授,E-mail:lifeng19820501@163.com
引用文本:
刘峰良,李锋,汤宝平,汪永超,田大庆.基于类对比簇分配异构迁移学习的空间滚动轴承寿命阶段识别[J].工程科学与技术,2024,56(1):256-266.
LIU Fengliang,LI Feng,TANG Baoping,WANG Yongchao,TIAN Daqing.Life Stage Identification of Space Rolling Bearings Based on Class-contrast Cluster-allocation Heterogeneous Transfer Learning[J].Advanced Engineering Sciences,2024,56(1):256-266.