面向顺应性加工的机器人自主学习、自主编程和自律跟踪

During the grinding and polishing of large and complex components with multiple varieties and small batch in the harmful environment, the workers have not been completely replaced by the robots because of the following problems: (a) The off-line programming of complex surfaces and structures is complicated and the preparation takes a long time; (b) The motion control programming cannot guarantee the force and position tracking performance; (c) The robots usually do not possess the ability of artificial experience drawing and self-skill promotion. Therefore, this project takes the large steam turbine blade as the object for study of robotic grinding and polishing. To solve the key problems of self-learning, self-programming and self-tracking, the following research topics are focused: (i) Interaction mechanism among “human-teaching robot-operating robot-environment”; (ii) Artificial force-control behavior coding and unstructured environment modeling with the visual and force telepresence; (iii) Motion planning based on self-learning and self-programming; (iv) Real-time feedrate generation and force-position control with force sensing; (v) Construction of machining system for robot learning and machining verification. This project is aimed at: (a) Integrating both the advantages of the artificial intelligence and the robot machining; (b) Constructing skillful robots which own the abilities of high flexibility, experience accumulation and generalization; (c) Validating the robotic imitation machining of the large turbine blades. The project is expected to provide new technological solution for the compliant machining of large and complex components, and improve the intelligent level of robotic machining.

在多品种、小批量、作业环境恶劣的大型复杂零件磨抛加工中，机器人完全取代人工尚且存在如下瓶颈：复杂曲面和复杂结构离线编程困难，准备时间长；底层运动控制编程无法保证力位跟踪性能；机器人通常不具备工匠经验汲取和自我技能提升能力。为此，本项目以大型汽轮机叶片机器人磨抛为研究对象，围绕机器人自主学习、自主编程和自律跟踪这一关键科学问题，开展“人-示教机器人-操作机器人-环境”交互机理、力觉和视觉临场感下工匠力控示教行为编码和非结构化环境建模、自主学习和自主编程综合的路径规划、力觉引导下实时速度轨迹生成和力位自律跟踪、机器人学习加工系统构建与加工验证等五方面研究工作，旨在融合工匠智慧与机器人精细加工双重优势，构建具有高度柔性、经验沉淀和泛化能力的“能工巧匠”型机器人，实现大型汽轮机叶片的机器人模仿加工。项目预期将为大型复杂零件的顺应性加工提供新的工艺思路，提升机器人加工的智能化水平。

复杂曲面零件离线编程困难，工匠经验难以指令化。如何架构顺应性磨抛中机器人学习加工的范式，实现零件加工过程中动觉示教的人机技能迁移，保障零件表面质量和精度、提升磨抛加工效率，仍然是机器人智能化制造尚未解决的突出难题。本项目以复杂曲面机器人模仿学习磨抛加工为对象，研究了“人-机-环境”交互机理，提出了力觉-视觉下流形结构建模方法，突破了阻抗行为的模仿学习技术，研制了学习型机器人顺应性加工系统并得到应用验证，取得如下4个主要进展：.(1) 建立了变阻抗的拖动意图预测方法，提出了异构式遥操作机器人运动映射算法，开发了提升示教编程效率的虚拟夹具技术，最大示教操作力降低约50%，示教精度平均提升7.5倍。.(2) 提出了环境参数学习与局部特征描述的三维重构算法，建立了机器人示教数据积空间位姿学习模型，力估计精度提升4.62%，特征提取速度提高近15倍，位姿流形编码精度提高15.62%。.(3) 发展了环境刚度学习的自适应阻抗控制策略，提出了保障跟踪精度的降维行为再现方法，建立了经验加速的机器人操作路径规划与柔顺示教分段路径泛化方法，力跟踪误差小于0.1N，关键点跟踪精度提升近1倍，路径规划效率提高约20%。.(4) 研制了面向复杂曲面零件的学习型机器人顺应磨抛系统，开发了工艺集成的离线编程软件，提出了光笔式机器人原位测量方案，在中国航发应用于高压压气机叶片打磨，轮廓精度±0.05mm，粗糙度Ra0.4μm。.项目团队在IEEE RAL和Robot CIM等国际权威期刊发表SCI论文12篇，发表国际会议论文和国内EI论文9篇，参与撰写出版著作2部（获湖北出版政府奖），申请国家发明专利13项（其中授权4项），参与专利成果转让/许可2次，共计700万元，登记软件著作权2项。获江苏省科学技术一等奖，共融机器人挑战赛冠军2次，IROS云机器人挑战赛全球冠军。组织大型学术会议1次，作特邀报告3次。项目负责人获评教育部青年长江、湖北省杰青。项目共培养博士后2人、博士生6人、硕士生10人，研究生获得智湖奖学金、国家奖学金3人次。