无标定视觉伺服多轴孔装配定位方法研究  

Research on positioning method of uncalibrated visual servo multiple peg-in-hole assembly

作　　者：焦建军[1,2] 李宗刚[1,2] 李龙雄 陈引娟[1,2] 夏广庆[3] JIAO Jianjun;LI Zonggang;LI Longxiong;CHEN Yinjuan;XIA Guangqing(School of Mechatronic Engineering,Lanzhou Jiaotong University,Lanzhou 730070,China;Robot Research Institute,Lanzhou Jiaotong University,Lanzhou 730070,China;State Key Laboratory of Structural Analysis,Optimization and CAE Software for Industrial Equipment,Dalian University of Technology,Dalian 116024,China)

机构地区：[1]兰州交通大学机电工程学院,甘肃兰州730070 [2]兰州交通大学机器人研究所,甘肃兰州730070 [3]大连理工大学工业装备结构分析优化与CAE软件全国重点实验室,辽宁大连116024

出　　处：《中南大学学报（自然科学版）》2024年第10期3731-3741,共11页Journal of Central South University:Science and Technology

基　　金：国家自然科学基金资助项目(61663020);甘肃省高等学校产业支撑计划项目(2022CYZC-33);大连理工大学工业装备结构分析国家重点实验室开放课题(ZG22119);兰州交通大学军民融合创新团队培育基金资助项目(JMTD20221)。

摘　　要：针对机器人多轴孔铆接装配件装配过程易出现装配件定位不准、孔与铆钉对准困难和装配效率低等问题,提出一种基于鲁棒卡尔曼滤波和前馈神经网络的铆接件视觉伺服精确定位方法,实现多轴孔铆接工件的自动定位。首先,搭建以工件顶点为特征点的机器人视觉伺服定位系统;其次,利用鲁棒卡尔曼滤波在线预测图像雅可比矩阵,并借助前馈神经网络在线动态补偿系统的估计误差,解决未知系统噪声的互相关性及“图像-操作空间”延时性造成的图像雅可比矩阵辨识精度低的问题;最后,采用最小二乘法设计无标定视觉伺服深度估计器,通过机器人的运动状态和图像特征的变化数据估计深度,解决铆接件定位时的深度问题。研究结果表明:相较于采用传统卡尔曼滤波的视觉伺服方法,机器人末端速度收敛时间减少30%,收敛曲线平滑,机器人末端运行轨迹平稳且接近直线,无振荡回馈现象,图像特征点定位误差绝对值最大为0.228像素,能够实现公板和母板快速定位,保证机器人快速完成铆接件的定位对准及后续装配任务。To address the challenges in the robot-assisted multiple peg-in-hole riveting assembly process,including inaccurate positioning of assembly parts,difficulties in aligning holes with rivets,and low assembly efficiency,the precise visual servo positioning approach for riveting parts was proposed based on robust Kalman filtering and feedforward neural networks,which enables the automatic positioning of multi-hole riveting workpieces.Initially,a robot visual servo positioning system was established,characterized by workpiece vertices as feature points.Secondly,Robust Kalman filtering was employed for online prediction of the image Jacobian matrix,and feedforward neural networks were used for online dynamic compensation of the system's estimation error,so that the low recognition accuracy of the image Jacobian matrix caused by the cross-correlation of unknown system noise and the latency in the image-operation space was effectively addressed.Additionally,to resolve the depth issue during the positioning of riveting parts,an uncalibrated visual servo depth estimator was designed using the least squares method,which estimated the depth through the robot's motion state and changes in image features.The results show that compared to the visual servo method using traditional Kalman filtering,the proposed approach reduces the convergence time of the robot's end speed by 30%and offers a smoother convergence curve.The robot's end trajectory is stable and closely linear,with no oscillatory feedback.The absolute value of the image feature point positioning error is at most 0.228 pixels,ensuring rapid positioning of the male and female plates and enabling the robot to align and proceed with subsequent assembly tasks quickly.

关 键 词：多轴孔 铆接 机器人定位 无标定视觉伺服 卡尔曼滤波 图像雅可比矩阵 

分 类 号：TP242[自动化与计算机技术—检测技术与自动化装置]