基于神经网络补偿的零力控制研究  

Force-Free Control Based on Neural Network

作　　者：江春林 梁润黎 华国环[1] JIANG Chun-lin;LIANG Run-li;HUA Guo-huan(College of Electronic and Information Engineering,Nanjing University of Information Science and Technology,Nanjing 210044,China;Zhongke Yitong Technology(Nanjing)Co.,Ltd.,Nanjing 211135,China)

机构地区：[1]南京信息工程大学电子与信息工程学院,江苏南京210044 [2]中科医通科技(南京)有限公司,江苏南京211135

出　　处：《中国电子科学研究院学报》2025年第1期68-74,共7页Journal of China Academy of Electronics and Information Technology

基　　金：国家重点研发计划(2022YFB4401301)。

摘　　要：针对机器人直接示教中摩擦力难以准确补偿的问题,文中提出了一种基于力矩指令下的零力控制方法,通过神经网络和摩擦力模型结合来提高摩擦力辨识精度,实现机器人的直接示教。首先,建立机器人动力学模型,确定数据测量方案,并进行数据采集;然后,针对采集到的电流数据含有高频噪声的问题,采用二阶巴特沃斯低通滤波器进行滤波;接着,针对机器人关节摩擦力在低速阶段的非线性的问题,采用傅里叶级数对机器人关节低速运行时的摩擦力进行建模,采用Stribeck模型对机器人关节高速运行时的摩擦力进行建模;最后,利用神经网络对重力和摩擦力补偿后的力矩残差进行再补偿,并在机器人关节模组上进行实验验证。结果表明,加入神经网络补偿后,机器人关节力矩补偿更均匀,可以消除关节力矩过补偿现象,且力矩补偿均方根误差可以降低20%以上,所需拖拽力更小。Aiming at the difficulty of accurate compensation of friction in direct teaching of robots,a zeroforce control method based on torque instruction was proposed,which combined neural network and friction model to improve the identification accuracy of friction and realize the direct teaching of robots.Firstly,the robot dynamics model is established,the data measurement scheme is determined,and the data collection is carried out.Secondly,the collected current data contains high-frequency noise,and the second-order Butterworth low-pass filter is used to filter it.Thirdly,aiming at the non-linearity of robot joint friction at low speed,Fourier series is used to model the friction of robot joint at low speed,and Stribeck model is used to model the friction of robot joint at high speed.Finally,neural network is used to compensate the residual torque after gravity and friction compensation,and the experiment is verified on the robot joint module.The results show that with the addition of neural network compensation,the robot joint torque compensation is more uniform,the phenomenon of joint torque overcompensation can be eliminated,and the root mean square error of torque compensation can be reduced by more than 20%,and the dragging force is smaller.

关 键 词：零力控制 摩擦力 傅里叶级数 神经网络 

分 类 号：TP242[自动化与计算机技术—检测技术与自动化装置]