• CN:11-2187/TH
  • ISSN:0577-6686

机械工程学报 ›› 2021, Vol. 57 ›› Issue (17): 53-63.doi: 10.3901/JME.2021.17.053

• 特邀专栏:智能制造前沿及应用 • 上一篇    下一篇

扫码分享

基于薄壁零件刚度仿真到真实迁移的力控制研究

陈鹏飞, 李祥飞, 何显铭, 蔡元昊, 赵欢, 丁汉   

  1. 华中科技大学数字制造装备与技术国家重点实验室 武汉 430074
  • 收稿日期:2021-02-01 修回日期:2021-04-27 发布日期:2021-11-16
  • 通讯作者: 李祥飞(通信作者),男,1990年出生,博士后。主要研究方向为运动控制、视觉伺服、力控制技术。E-mail:lixiangfei@hust.edu.cn
  • 作者简介:陈鹏飞,1996年出生,博士研究生。主要研究方向为机器人智能力控制磨抛与装配技术。E-mail:d202080198@hust.edu.cn;何显铭,1996年出生,硕士研究生。主要研究方向为机器人智能示教学;习与力控制技术。E-mail:m201970513@hust.edu.cn;蔡元昊,1997年出生,硕士研究生。主要研究方向为机器人加工力控制;技术。E-mail:m202077001@hust.edu.cn;赵欢,男,1983年出生,教授。主要研究方向为机器人智能化加工装备与技术。E-mail:huanzhao@hust.edu.cn;丁汉,男,1963年出生,教授,博士研究生导师。主要研究方向为数字化制造与智能化制造。E-mail:dinghan@hust.edu.cn
  • 基金资助:
    国家自然科学基金(52090054)和湖北省自然科学基金杰青项目(2020CFA077)资助项目。

Research on Force Control Based on Sim2Real Transfer for Stiffness of Thin-walled Parts

CHEN Pengfei, LI Xiangfei, HE Xianming, CAI Yuanhao, ZHAO Huan, DING Han   

  1. State Key Laboratory of Digital Manufacturing Equipment and Technology, Huazhong University of Science and Technology, Wuhan 430074
  • Received:2021-02-01 Revised:2021-04-27 Published:2021-11-16

摘要: 复杂薄壁零件广泛应用于航空航天领域,但由于结构复杂、壁厚较小、受载易变形,因此其加工过程中的恒力控制较为困难,导致零件加工精度难以满足要求。阻抗控制提供了一种控制恒定接触力的有效方式。然而,当零件(即环境)初始位置未知时,现有在线刚度估计方法无法保证环境刚度估计值收敛到真实值。此外,尽管可以通过离线策略精确辨识环境刚度,但力/形变数据获取过程复杂耗时。为此,提出了一种仿真到真实迁移(sim2real)的环境刚度精确估计方法。首先,利用有限元仿真获取零件不同位置的刚度值,并通过域随机化算法得到不同材料、形状零件的仿真环境刚度库。然后,采用神经网络学习仿真环境下的刚度数据,根据该结果预测真实环境中不同种类零件的刚度值。所提出的环境刚度估计方法克服了环境刚度和初始位置无法同时在线精确估计的难点,且比离线辨识策略更为简单高效。采用直线模组与悬臂梁开展试验验证,结果表明所提出的方法可以精确估计环境刚度,估计误差小于10%,利用估计刚度值进行的力跟踪误差小于±0.5 N,均优于现有基于Lyapunov的估计方法。

关键词: 薄壁零件, 恒力加工, 刚度估计, 仿真到真实迁移

Abstract: Complex thin-walled parts are widely used in aerospace field. However, due to its complex structure, small wall thickness and deformability, it is difficult to realize constant force control in the process of machining. As a result, the machining accuracy of parts is difficult to meet the requirements. Impedance control provides an effective way to keep constant contact force. However, the existing on-line stiffness estimation methods cannot guarantee that the environment stiffness estimation converges to the real value when the original position of the part (i.e. environment) is unknown. In addition, although the off-line strategy can accurately identify the environment stiffness, the process of force/deformation data acquisition is complex and time-consuming. Therefore, a sim2real transferring method is proposed to accurately estimate the environmental stiffness. Firstly, the stiffness values of different positions of parts are obtained by finite element simulation analysis, and the simulation environment stiffness library of parts with different materials and shapes is obtained by domain randomization algorithm. Then, the neural network is used to learn the stiffness data in the simulation environment, and the stiffness values of different kinds of parts in the real environment are predicted according to the results. The proposed method overcomes the difficulty that the environmental stiffness and original position cannot be accurately estimated online at the same time. Moreover, it is simpler and more efficient than the off-line identification strategy. The linear motion platform and cantilever beam are used to carry out the experimental verification. The results show that the proposed method can accurately estimate the environmental stiffness, and the estimation error is less than 10%, and the force tracking error based on the estimated stiffness is less than ±0.5 N, which are better than the existing Lyapunov based estimation method.

Key words: thin-walled parts, constant force machining, stiffness estimation, sim2real

中图分类号: