• CN:11-2187/TH
  • ISSN:0577-6686

机械工程学报 ›› 2019, Vol. 55 ›› Issue (3): 130-137.doi: 10.3901/JME.2019.03.130

• 数字化设计与制造 • 上一篇    下一篇

基于双深度神经网络的轮廓误差补偿策略研究

喻曦, 赵欢, 李祥飞, 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:2018-04-19 修回日期:2018-09-22 出版日期:2019-02-05 发布日期:2019-02-05
  • 通讯作者: 赵欢(通信作者),男,1983年出生,副教授。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
  • 作者简介:喻曦,女,1993年出生,主要研究方向为运动控制技术。E-mail:yuxi@hust.edu.cn;李祥飞,男,1990年出生,博士研究生。主要研究方向为多轴运动控制与机器人视觉伺服。E-mail:lixiangfei@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机器人技术。E-mail:dinghan@hust.edu.cn
  • 基金资助:
    国家重点研发计划(2017YFB1301501)和国家自然科学基金(91748114,51535004)资助项目。

Research on Contouring Error Compensation Method using Dual Deep Neural Networks

YU Xi, ZHAO Huan, LI Xiangfei, DING Han   

  1. State Key Laboratory of Digital Manufacturing Equipment and Technology, Huazhong University of Science and Technology, Wuhan 430074
  • Received:2018-04-19 Revised:2018-09-22 Online:2019-02-05 Published:2019-02-05

摘要: 五轴数控机床的加工精度通常由轮廓误差指标来衡量。传统的轮廓误差降低策略主要包括精确的轮廓误差估计和有效的轮廓控制器设计。然而,传统策略存在刀具路径轮廓误差在线估计或控制器设计复杂等问题。为此,从机床输入驱动指令和输出末端位姿的映射出发,针对五轴数控机床加工大批量工件提出基于数据驱动的轮廓误差补偿策略。调整PID控制器参数保证系统单轴伺服的稳定跟踪,同时采集各伺服轴的输入指令和机床的实际输出位姿。针对五轴数控机床的刀具位姿和刀轴方向分别搭建位姿和方向两个深度神经网络,并基于数据训练所得的神经网络模型预测系统新的输入参考指令。采用五轴刀具路径开展轮廓跟踪试验。试验结果表明:所提出的基于深度神经网络的轮廓误差补偿策略不需要刀具路径轮廓误差的在线估计和控制器的有效设计,即可有效降低刀具路径的位置和方向轮廓误差。

关键词: 参考输入, 轮廓控制, 轮廓误差, 深度神经网络

Abstract: The machining accuracy of five-axis CNC machine tools is usually measured by contour error. The traditional contour error reduction strategies mainly include accurate contour error estimation and effective contouring controller design. However, there are problems in traditional strategies, such as online contour error estimation or complex controller design. To this end, based on the mapping between the input drive commands of machine tool and the output pose, a data-driven contour error compensation strategy is firstly proposed for five-axis CNC machine tools. First, a PID controller is adjusted to ensure the stable tracking of single axis, and the input commands and actual output pose of machine tool are collected at the same time. Then, according to the tool pose and orientation of five-axis CNC machine tool, dual deep neural network models for the tool pose and orientation are built respectively, and the new reference inputs can be predicted based on the neural network models obtained from training data. Finally, a five-axis tool path is used to carry out the experiments. The experimental results show that the proposed contour error compensation strategy does not require the online contour error estimation and effective controller design, which can reduce the position and orientation contour errors effectively.

Key words: contour error, contouring control, deep neural network, reference input commands

中图分类号: