基于迁移学习的滚磨光整加工工艺要素决策  

Mass Finishing Process Elements Decision Based on Transfer Learning

作　　者：史玉皓 田建艳 杨英波 李文辉[3] 杨胜强 SHI Yuhao;TIAN Jianyan;YANG Yingbo;LI Wenhui;YANG Shengqiang(College of Electrical and Power Engineering,Taiyuan University of Technology,Taiyuan 030024,China;College of Mechanical and Vehicle Engineering,Taiyuan University of Technology,Taiyuan 030024,China;College of Aeronautics and Astronautics,Taiyuan University of Technology,Taiyuan 030024,China)

机构地区：[1]太原理工大学电气与动力工程学院,山西太原030024 [2]太原理工大学机械与运载工程学院,山西太原030024 [3]太原理工大学航空航天学院,山西太原030024

出　　处：《控制工程》2025年第3期481-491,552,共12页Control Engineering of China

基　　金：山西省重点研发计划项目(201903D121057);山西省研究生教育创新计划项目(2021Y229)。

摘　　要：在滚磨光整加工工艺要素决策中,当特征信息分布差异较大时,采用案例推理和专家推理模型会导致出现决策不准的问题。为此,提出了加权多源自适应迁移学习算法,构建了基于迁移学习的滚磨光整加工工艺要素决策模型。首先,对待加工零件信息、加工要求、工艺要素进行了特征表征;然后,为提高算法的适用范围,在基于流形结构的多源自适应迁移学习算法(multi domain adaptation-manifold regularization,MDA-MR)中引入适配因子与Wasserstein距离;最后,针对预测目标分类过多引发的负迁移问题,设计了工艺要素相似度匹配算法,综合构建了决策模型并设计了决策界面。仿真结果表明,该模型的决策准确率更高,可以为滚磨光整加工提供更有价值的决策支持。For the problem that the CBR and ER models are inaccurate when the distribution of feature information of new parts and the original machining experience is quite different in the process elements decision of mass finishing,the weighted multi domain adaptation transfer learning algorithm is proposed and built the decision model of mass finishing process elements based on transfer learning.Firstly,the information of the processed parts,processing requirements,and process elements are characterized.Secondly,in order to improve the application scope of the transfer learning,the adaptation factor and the Wasserstein Distance are introduced on the MDA-MR algorithm.Then,in order to solve the problem of negative transfer caused by excessive classification of predicted targets,the process element similarity matching algorithm is designed.Finally,the decision model and interface of mass finishing process elements are built.The simulation results show that this model has higher decision accuracy,which can provide more valuable decision support for mass finishing.

关 键 词：迁移学习 滚磨光整 工艺要素决策 相似度匹配算法 决策界面 

分 类 号：TP29[自动化与计算机技术—检测技术与自动化装置] TG356.28[自动化与计算机技术—控制科学与工程]