Grinding constant force control method based on deep reinforcement learning PPO algorithm

Abstract

The invention provides a grinding constant force control method based on a deep reinforcement learning PPO algorithm, and the method comprises the steps: firstly carrying out the stress analysis of an end effector during the grinding process of a spherical grinding wheel, and obtaining the real grinding force through the force processing; a relation model between the difference between the current grinding force and the preset grinding force and the normal compensation displacement of the end effector is trained through a deep reinforcement learning PPO algorithm, and training data normalization preprocessing, Euclidean distance-based reward function design, targeted deep neural network structure design and algorithm convergence promotion design are included; training a controller based on a PPO algorithm and using a grinding robot constant force controller. According to the method, a prior model between the grinding force error and the normal displacement does not need to be established in advance, the task that the grinding force is controlled within the target grinding force during plane grinding and curved surface grinding can be completed, and meanwhile the good self-adaptive capacity is achieved.

Description

technical field [0001] The invention relates to grinding machine control, specifically a grinding constant force control method based on a deep reinforcement learning PPO algorithm. Background technique [0002] The traditional manual grinding not only has uneven grinding quality and low efficiency, but also the flying dust produced by grinding has caused harm to the health of workers. The use of grinding robots can solve the problem very well. Installing a grinding end effector at the end of a multi-degree-of-freedom robot can machine workpieces with complex surface geometries, and can also avoid the problems caused by manual grinding. [0003] Robot grinding force control can be divided into passive force control and active force control. Passive power control mainly relies on some auxiliary compliance mechanisms, so that the robot can naturally comply with the grinding force when it is in contact with the abrasive belt wheel. Although this control method can effectively...

AI Technical Summary

Problems solved by technology

Although the traditional force control strategy can achieve a certain control effect, it is difficult to achieve satisfactory results due to nonlinearity and a large number of uncertainties in robot grinding.

The force control based on the intelligent strategy can realize the constant force control through the intelligent algorithm, but the traditional intelligent algorithm generally needs to establish a priori model in advance, such as the relationship model between the grinding force difference and the compensation displacement. Modeling is very difficult

Images