Vibration control of three coupled flexible beams using reinforcement learning algorithm based on proximal policy optimization

Abstract

In order to suppress the vibration of a class of multi-coupling flexible beams, a three-coupling flexible beam experimental device is constructed. The vibration characteristics and active vibration control algorithms of three coupled flexible beams are studied. Due to the complex residual vibration of multi-coupled flexible beam structure, which shows the interaction between beams and close frequency characteristics, the dynamic model of coupled flexible beam structure is established by finite element method (FEM) and its theoretical modal analysis is carried out. Furthermore, the parameters of the experimental system are identified, and the close modal frequencies and damping ratios of the actual system are obtained. The characteristics of close modal vibration are verified. A reinforcement learning (RL) vibration controller based on proximal policy optimization (PPO) algorithm is designed. The algorithm mainly includes actor neural network and critic neural network, which are used to approximate the control strategy and evaluate state respectively. Based on theoretical modeling and experimental identification correction model, the agent learns offline, and then transfers the learned policy parameters to the experimental system for vibration control experiments. The experimental results demonstrate that the designed RL controller can effectively suppress the residual vibration of three coupled flexible beams.