Manual assembly of aero-engine casings with dissimilar flanges suffers from low efficiency,poor mating accuracy and high safety risk.To overcome these problems,this paper proposes a compliant control scheme for a six-degree-of-freedom Stewart platform that learns to insert a flange test specimen via reinforcement learning.Firstly,the kinematic and dynamic models of the Stewart platform are established,and its ability to adjust posture under complex constraints is analyzed.Secondly,a data-driven compliant strategy that fuses admittance control with a reinforcement-learning algorithm is designed to realize high-precision mating and gentle insertion.Finally,a co-simulation framework combining Webots and PyCharm is built to verify the effectiveness and robustness of the proposed method under various initial misalignments and disturbances.Results show that,compared with traditional fixed parameter admittance control methods,the proposed method reduces the maximum contact force of the platform by 30.4% and improves the assembly efficiency by about 90% during the assembly process,and achieves a steady-state position error of only 7.5mm.This indicates that the proposed method significantly improves work efficiency and accuracy while ensuring assembly safety,and has good engineering promotion value.
1)时间复杂度:单步推理的主要计算量来自于矩阵乘法,其时间复杂度可表示为$O({\sum }_{l=1}^{L-1}{n}_{l}·{n}_{l+1})$。在本文的网络结构设置下(输入层12维,隐藏层256维,输出层6维),单次前向传播涉及的浮点运算次数(Floating Point Operations Per Second,FLOPS)在105数量级。对于现代工业控制计算机,如IPC或嵌入式GPU模块,其计算能力通常在109数量级,单步推理耗时可控制在1ms以内,远低于法兰装配场景要求的10ms控制周期,满足硬件实时控制需求。
QIAOZ Y. Current status of industrial robot technology and industrialization development strategy in my country[J]. ChinaNon-Governmental, 2023(10):203-205.
[2]
DANESHMANDM, NOROOZIF, CORNEANUC, et al. Industry 4.0 and prospects of circular economy:a survey of robotic assembly and disassembly[J]. The International Journal of Advanced Manufacturing Technology, 2023, 124(9):2973-3000.
LIUJ H, SUNQ C, CHENGH, et al. The state-of-the-art,connotation and developing trends of the products assembly technology[J]. Journal of Mechanical Engineering, 2018, 54(11):2-28.
[4]
RUSSOM, ZHANGD, LIUX J, et al. A review of parallel kinematic machine tools:design,modeling,and applications[J]. International Journal of Machine Tools and Manufacture, 2024, 196:104118.
DAVLIAKOSI, PAPADOPOULOSE. Impedance model-based control for an electrohydraulic Stewart platform[J]. European Journal of Control, 2009, 15(5):560-577.
SUND Y, XUR H, WANGJ Y, et al. Parallel robotic automated docking method for realizing space segment assembly[J]. Scientific Reports, 2025, 15(1):5754.
DEHGHANS A M, DANESHM, SHEIKHOLESLAMF. Adaptive hybrid force/position control of robot manipulators using an adaptive force estimator in the presence of parametric uncertainty[J]. Advanced Robotics, 2015, 29(4):209-233.
DUZ J, WANGW, YANZ Y, et al. Variable admittance control based on fuzzy reinforcement learning for minimally invasive surgery manipulator[J]. Sensors, 2017, 17(4):844.
ZHANGH Y, LIL, ZHAOJ B, et al. The hybrid force/position anti-disturbance control strategy for robot abrasive belt grinding of aviation blade base on fuzzy PID control[J]. The International Journal of Advanced Manufacturing Technology, 2021, 114(11):3645-3656.
CHENK. Research on robot assembly contact state recognition method based on machine learning[D]. Hangzhou: Zhejiang University of Technology, 2020.
[15]
LEVINES, PASTORP, KRIZHEVSKYA, et al. Learning hand-eye coordination for robotic grasping with large-scale datacollection[EB/OL].(2016-08-28)[2026-01-24]. https://arxiv.org/abs/1603.02199.
[16]
ZENGA, FLORENCEP R, TOMPSONJ, et al. Transporter networks:rearranging the visual world for robotic manipulation[C]// Proceedings of the 4th Conference on Robot Learning (CoRL 2020).[S.l.]:PMLR, 2021:726-747.]
[17]
ZHUY K, MOTTAGHIR, KOLVEE, et al. Target-drivenvisual navigation in indoor scenes using deep reinforcement learning[EB/OL].(2016-09-16)[2026-01-15]. https://arxiv.org/abs/1609.05143.
[18]
OPENAI, ANDRYCHOWICZM, BAKERB, et al. Learning dexterous in-handmanipulation[EB/OL].(2019-01-18)[2026-01-17]. https://arxiv.org/abs/1808.00177.
[19]
INOUET, DE MAGISTRISG, MUNAWARA, et al. Deep reinforcement learning for high precision assemblytasks[EB/OL].(2017-09-21)[2026-01-24]. https://arxiv.org/pdf/1708.04033v2.
[20]
MARTÍN-MARTÍNR, LEEM A, GARDNERR, et al. Variable impedance control in end-effector space: an action space for reinforcement learning in contact-richtasks[EB/OL].(2019-08-02)[2026-01-24]. https://arxiv.org/pdf/1906.08880.
[21]
SCHULMANJ, WOLSKIF, DHARIWALP, et al. Proximal policy optimizationalgorithms[EB/OL].(2017-08-28)[2026-01-24]. https://arxiv.org/abs/1707.06347.