摘要
在选择滚抛磨块时,针对案例推理技术在案例相似性较低情况下无法选定有效滚抛磨块的问题,提出基于迁移学习的滚抛磨块优选方法。通过缩小源案例和新问题的分布差异,将源案例的信息迁移到新问题的求解中,优选得到滚抛磨块。首先,通过分析大量实际加工工艺案例构建案例特征E-R图;然后,对不同类型的案例特征值进行归一化预处理,经过集成流形约束及条件分布适配缩小不同类型的零件案例分布差异,并通过特征变换矩阵将不同类型的零件案例集的特征信息投射到公共空间;最后,通过在公共空间内建立滚抛磨块优选模型,得到滚抛磨块参数值。通过大量仿真研究可知,对于相似性较低的案例,通过迁移学习方法可以得到比案例推理技术更准确的结果,验证了模型良好的泛化性和实用性。基于迁移学习的滚抛磨块优选模型可以弥补传统案例推理技术进行磨块选择时的不足,为滚抛磨块的选择提供决策指导。
When choosing the abrasive media,in view of the problem that case-based reasoning technology cannot select the effective abrasive media due to the low case similarity,an optimal model of the abrasive media based on transfer learning was proposed.By reducing the distribution difference of the source case and the new problem,the knowledge of the source case was transferred to the solution of the new problem so as to optimize the abrasive media.Firstly,the E-R diagram of case characteristics was constructed according to the large number of actual processing cases.Then,the feature values of different types of cases were normalized to make a pre-processing.The case distribution differences of different types were reduced based on the integrated manifold constraints and conditional distribution adaptation,and the feature information of different types of part case sets was projected to the public space by finding the feature transformation matrix.Finally,the optimal model of the abrasive media was established in public space to obtain the parameters value.A large number of simulations show that more accurate results can be obtained for cases with low similarity,which proves the good generalization and practicality of the model.The proposed model can make up for the shortcomings of traditional case-based reasoning and provide decision guidance for the selection of the abrasive media.
作者
闫瑞斌
田建艳
杨胜强
高云松
刘军军
高炜
YAN Ruibin;TIAN Jianyan;YANG Shengqiang;GAO Yunsong;LIU Junjun;GAO Wei(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 Mathematics,Taiyuan University of Technology,Taiyuan 030024,China;North Tianyu Electromechanical Technology Co.,Ltd.,Langfang 065000,China)
出处
《现代制造工程》
CSCD
北大核心
2021年第1期75-80,共6页
Modern Manufacturing Engineering
关键词
滚磨光整加工
滚抛磨块
迁移学习
条件分布适配
流形正则化
磨块优选
barrel finishing
the abrasive media
transfer learning
Conditional Distribution Adaptation(CDA)
Manifold Regularization(MR)
optimization of the abrasive media
