精密进给系统热误差的协同训练支持向量机回归建模与补偿方法  

Co-Training Support Vector Machine Regression Modeling and Compensation for Thermal Error of Precision Feed System

作　　者：朱星星 赵亮[1,2] 雷默涵 王帅[1,2] 凌正[1,2] 杨军 梅雪松[1,2] ZHU Xingxing;ZHAO Liang;LEI Mohan;WANG Shuai;LING Zheng;YANG Jun;MEI Xuesong(Shaanxi Key Laboratory of Intelligent Robots,Xi’an Jiaotong University,Xi’an 710049,China;State Key Laboratory for Manufacturing Systems Engineering,Xi’an Jiaotong University,Xi’an 710049,China)

机构地区：[1]西安交通大学陕西省智能机器人重点实验室,西安710049 [2]西安交通大学机械制造系统工程国家重点实验室,西安710049

出　　处：《西安交通大学学报》2019年第10期40-47,共8页Journal of Xi'an Jiaotong University

基　　金：国家自然科学基金资助项目（51605375）;数字制造装备与技术国家重点实验室资助项目（DMETKF2019017）

摘　　要：针对精密进给系统热误差的数据稀缺且获取成本高的问题,提出了一种基于协同训练支持向量机回归算法(COSVR)的精密进给系统热误差建模与补偿方法。通过整合标记数据(温度和热误差)及未标记温度数据建立热误差模型,利用基于西门子840D数控系统开发的补偿方法进行补偿。以精密镗床双驱动滚珠丝杠进给系统X轴为研究对象,进行热特性实验,获取24 m/min进给速度下的标记数据和12 m/min进给速度下的未标记温度数据,利用COSVR整合所有数据建立热误差模型,并通过遗传算法优化的支持向量机回归算法(GA-SVR)仅选用标记数据建立对照模型,获取18 m/min进给速度下的标记数据用于模型性能测试。结果表明:与GA-SVR模型相比,COSVR模型的均方根误差减少了34.14%,且在100 min和520 min时的误差范围分别减小了62.62%和55.85%。COSVR模型具有更好的预测性能且能更有效地降低热误差,进一步提高了精密进给系统热误差的建模精度。Considering the problem of high cost and scarcity of thermal error data acquisition for precision feed system,a modeling and compensation strategy for the thermal error of precision feed system based on co-training support vector machine regression is proposed.This strategy establishes thermal error model by integrating labeled data such as temperature error and thermal error and unlabeled temperature data,then compensates the compensation method based on Siemens 840D NC system.Taking X-axes of double-drive ball screw feed system of precision boring machine as the research object,thermal characteristic experiments are carried out.The labeled data at 24 m/min feeding speed and the unlabeled temperature data at 12 m/min feeding speed are obtained.The thermal error model is constructed by integrating all data with COSVR,and only the labeled data are used to construct the control model with support vector machine regression algorithm optimized by genetic algorithm(GA-SVR).The labeled data at 18 m/min feeding speed are obtained for model performance test.Compared with GA-SVR model,the root mean square error of COSVR model is reduced by 34.14%,and the error range at 100 min and 520 min is reduced by 62.62%and 55.85%respectively.The results show that COSVR model has better prediction performance and reduces thermal error more effectively to further improve the accuracy of thermal error modeling of precision feeding system.

关 键 词：精密镗床 进给系统 协同训练 支持向量机回归 无标记数据 

分 类 号：TG502.15[金属学及工艺—金属切削加工及机床]