摘要
金属铣削通常伴随着高温高热高压,作为多元非线性过程具有复杂性与无序性。使用M-V5CN组合机床铣削U71Mn高锰钢获取了1000组样本集,基于Xgboost算法增益分析发现表面粗糙度影响因素由大到小依次为:主轴转速、铣削深度、每齿进给量、铣削宽度。提出一种在三折交叉验证下基于Stacking方法的改进集成回归模型,通过集成训练GA-SVM、Xgboost、KNN模型有效避免单一基模型在特定铣削参数区间内误差敏感问题。最终在测试集样本中模型最大误差errmax(0.041μm),平均绝对误差MAE(0.017μm),决定系数r(0.938),较基模型性能提升显著。
Metal milling is usually accompanied by high temperature,high heat and high pressure. As a multivariate nonlinear process,it has complexity and disorder. 1000 sets of samples are obtained by milling U71Mn high manganese steel with M-V5CN combined machine tool. Based on Xgboost algorithm gain analysis,it is found that the influencing factors of surface roughness from the largest to the smallest are spindle speed,milling depth,feed per tooth,milling width. An improved integrated regression model based on Stacking method is proposed under three-fold cross-validation. Through integrated training of GA-SVM,Xgboost and KNN models,the error sensitivity of a single base model in a specific milling parameter range can be effectively avoided. Finally,the maximum error of the model in the test set sample is errmax(0.041μm),mean absolute error MAE(0.017μm),determination coefficient r(0.938),significantly improving the performance of the base model.
作者
奚建峰
史柏迪
庄曙东
陈天翔
陈威
XI Jianfeng;SHI Baidi;ZHUANG Shudong;CHEN Tianxiang;CHEN Wei(Changzhou Branch of Vallourec Tianda(Anhui)Co.,Ltd.,Changzhou 213033;School of Mechanical Engineering,Hohai University,Changzhou 213022;Jiangsu Key Laboratory of Precision Instruments,Nanjing University of Aeronautics and Astronautics,Nanjing 213009)
出处
《计算机与数字工程》
2022年第12期2826-2830,2842,共6页
Computer & Digital Engineering
基金
江苏省精密与微细制造技术重点实验室基金项目(编号:CZ520007812)
中央高校基本科研业务费(编号:2018B44614)
江苏省高校实验室研究会立项资助研究课题(编号:GS2019YB18)资助。
