Active control of milling vibration in thin-walled parts using a deep Q-network algorithm

Abstract

Vibration is easily generated in thin-walled parts during milling. This leads to problems such as reduced machining quality and increased tool wear. The vibration is characterized by strong nonlinearity and complexity. To solve these problems, an active vibration control method based on deep reinforcement learning (DRL) is proposed. The deep Q-network (DQN) algorithm is employed to intelligently suppress milling vibration. A dynamic model of the thin-walled part is established. A control framework using piezoelectric patches as actuators is designed. The milling process is modeled as an interaction problem between an agent and the environment. The state, action, and reward function are defined. In simulation, the vibration suppression effects of the DQN controller and the PID controller were compared. Excitations included impulse, sinusoidal, white noise, and actual milling force. The results show that under milling force excitation, the vibration reduction rate of DQN control was improved by 49.27%, compared with PID control. The effectiveness of the algorithm was verified by a milling experiment on a thin-walled part. The experimental results show that the vibration response is significantly reduced under DQN control. The results are in good agreement with the simulation. It is demonstrated that the DQN algorithm has a good suppression effect on milling vibration of thin-walled parts.