数据驱动的6R型串联工业机器人精度性能提升  

Improvement of Accuracy Performance for 6R Serial Industrial Robot Based on Data-Driven

作　　者：乔贵方[1] 高春晖 蒋欣怡 聂新港 刘娣[1] QIAO Guifang;GAO Chunhui;JIANG Xinyi;NIE Xingang;LIU Di(School of Automation,Nanjing Institute of Technology,Nanjing 211167,China)

机构地区：[1]南京工程学院自动化学院,南京211167

出　　处：《组合机床与自动化加工技术》2024年第8期66-69,74,共5页Modular Machine Tool & Automatic Manufacturing Technique

基　　金：国家自然科学基金项目(51905258);中国博士后科学基金项目(2019M650095);江苏省研究生科研与实践创新计划项目(SJCX23_1164)。

摘　　要：随着工业机器人在高端制造领域应用中的不断深入,其绝对定位精度低的问题越加凸显。研究了基于神经网络的机器人实际位姿误差预测问题,针对Staubli TX60型串联工业机器人进行了运动学建模和误差分析,并搭建了基于Leica AT960激光跟踪仪的机器人测量实验平台,测量并计算了大量末端位姿误差数据。设计并优化最佳DNN神经网络结构,利用该神经网络预测机器人实际位姿误差,避免了基于模型的机器人精度提升方法中复杂的误差建模。补偿后的机器人平均绝对位置误差和平均绝对姿态误差分别由补偿前的(0.671 12 mm, 0.002 86 rad)降低至(0.048 58 mm, 0.000 46 rad),位置精度提升92.76%,姿态精度提升83.92%。最后,通过与BP、Elman神经网络以及传统LM最小二乘标定方法进行对比实验,验证了基于DNN深度神经网络进行机器人位姿精度提升具有更好的平衡性。With the deepening of the application of industrial robots in the high-end manufacturing field,the problem of low absolute positioning accuracy has become increasingly prominent.In this paper,the problem of robot pose error prediction based on neural network is studied.The kinematic modeling and error analysis of Staubli TX60 series industrial robot are carried out.A robot measurement experiment platform based on Leica AT960 laser tracker is built,and a large number of end position error data are measured and calculated.The optimal DNN neural network structure is designed and optimized,and the actual pose error of the robot is predicted by the neural network,which avoids the complicated error modeling in the model-based robot precision improvement method.After compensation,the average absolute position error and average absolute attitude error of the robot decreased from(0.67112 mm,0.00286 rad)before compensation to(0.04858 mm,0.00046 rad),and the position accuracy increased by 92.76%and the attitude accuracy increased by 83.92%.Finally,through the comparison experiment with BP and Elman neural network,it is verified that the improvement of robot pose accuracy based on deep neural network has better balance.

关 键 词：数据驱动 工业机器人 非模型标定 精度性能 机器人标定 

分 类 号：TH164[机械工程—机械制造及自动化] TG659[金属学及工艺—金属切削加工及机床]