A meta-reinforcement learning method by incorporating simulation and real data for machining deformation control of finishing process
提出一种元强化学习模型，利用仿真环境训练并结合少量真实数据更新，解决精加工过程变形控制中数据标签不足的问题，在仿真和实际加工中验证了有效性。
Finishing determines the final dimension and geometric accuracy of parts, and the finishing process directly affects the stiffness and residual stress redistribution of the workpiece, so the optimisation of the finishing process plays a very important role in deformation control. At present, existing data-driven methods for deformation control need a large amount of labelled training data, which is always a challenge in the manufacturing area, especially for machining deformation. To address the above issues, this paper presents a meta-reinforcement learning model incorporated by simulation and real data, which is trained in a simulation environment with a piecewise sampling strategy for data collection, and can be updated in a real machining environment through a very small number of real monitoring data. The finishing process optimisation for deformation control can be realised using the proposed approach. Finally, the effectiveness of the proposed method is verified both in simulation environment and actual machining, and better results are obtained compared with other existing methods.