In order to improve the quality of the industrial robot automatic polishing on curved surfaces and ensure the constant polishing pressure during polishing process, a method for polishing complex concave cavity surfaces with industrial robot is proposed in this article. The method can achieve stable force control and precise position control and is easy to be realized online. In order to ensure the removal rate uniformity of surface material at different normal vectors, a method for adjusting the speed of motorized spindle in real time according to the surface normal vector is proposed. After planning the trajectory and normal vectors, combined with the feedback force signal from the sensor and the proportional–integral controller in the direction of the normal vector, the robot terminal tool corrects the trajectory in the direction of the surface normal vector, indirectly realizing force control between the tool and the surface. The robot polishing system with different polishing tools has different system stiffness. In order to ensure the polishing system with different stiffness to have a better tracking performance of the contact force, an adaptive proportional–integral control algorithm proposed in this article can be used to evaluate the stiffness of polishing system and to adjust proportional–integral parameters. The simulation and experimental results indicate that the method can realize the polishing of concave cavity surface commendably.

Polishing robot is an automatic system in which the robot controls the end effector to fix the polishing tool and finish the workpiece polishing efficiently. In order to solve the problem of how to maintain the stability of actuator contact force in the robot automatic polishing system, a learning algorithm of robot impedance control parameters based on reinforcement learning is proposed and the impedance control model is established in this paper. The influence parameters (inertia M, damping B, stiffness K) of impedance performance are analyzed by numerical simulation method and the optimized impedance parameters are obtained at last. Due to the small number of iterations and high data utilization rate, reinforcement learning algorithm is more suitable for robot constant force tracking. In the process of applying reinforcement learning algorithm, a combination of dynamic matching method and linearization method is proposed to predict the output distribution of the state, which greatly improves the cost function of the evaluation strategy, and impedance parameters corresponding to the optimal strategy are obtained. Finally, steam turbine blade is taken as polishing test part. The average roughness of the selected points of test part after polishing is only 0.302μm, and much less than 1.151μm before polishing, which verifies the feasibility of the proposed impedance control method.