摘要
分析刀具-刀柄结合状态变化时的数控铣削稳定性,其效率因需重复测量刀尖点频响函数而降低。针对此问题,引入迁移学习提出仅需测量目标刀具少量悬伸量下刀尖点频响函数的铣削稳定性预测方法。首先,测量源刀具多个悬伸量和目标刀具少量悬伸量的刀尖点频响函数,采用铣削稳定性解析法获取各铣削参数组合的加工振动状态信息,构建充足的源域数据和少量的目标域数据,并通过源域和目标域的相似匹配筛选源域样本,然后结合神经网络和TrAdaBoost迁移学习算法,自适应更新源域与目标域混合样本权重,建立目标刀具的铣削加工振动状态分类器。以三组刀柄-刀具组合进行实例分析,少样本下采用迁移学习后两组目标刀具的分类器精度分别提升了10.93%、6.25%,并通过铣削实验验证了所提方法的有效性。
Repeated impact tests will be conducted to obtain the tool tip frequency response functions(FRFs) for analyzing the milling stability with variable tool-holder clamping conditions.Considering the low efficiency,a transfer learning-based method is proposed to predict the milling stability of the target tool using the tool tip FRFs measured under a few overhang lengths.The tool tip FRFs under multiple overhang lengths of the source tool and a few overhang lengths of the target tool are measured,which are taken to perform the milling stability analysis and construct the source and target domain data.A similarity matching is conducted on the source and target domains to select appropriate source domain samples.Then,the neural network and TrAdaBoost transfer learning algorithm are combined to adaptively update the weights of target and source samples,and a classifier for obtaining milling vibration states is trained iteratively.Three tool-holder combinations are taken to perform the case study.The accuracies of two classifiers for two target tool-holder combinations are improved by 10.93% and 6.25% respectively after introducing the transfer learning,and the feasibility of the proposed transfer learning-based milling stability prediction method is validated by the milling experiments.
作者
邓聪颖
邓子豪
林丽君
陈翔
马莹
禄盛
DENG Congying;DENG Zihao;LIN Lijun;CHEN Xiang;MA Ying;LU Sheng(School of Advanced Manufacturing Engineering,Chongqing University of Posts and Telecommunications,Chongqing 400065;School of Mechanical Engineering,Chengdu University,Chengdu 610106)
出处
《机械工程学报》
EI
CAS
CSCD
北大核心
2024年第3期296-304,共9页
Journal of Mechanical Engineering
基金
国家自然科学基金(51705058)
中国博士后科学基金(2018M633314)
重庆市教委科技计划(KJQN202300640、 KJZD-K202300611)资助项目。
