Reinforcement learning vibration control of a multi-flexible beam coupling system

Abstract

An active vibration control algorithm based on reinforcement learning (RL) is applied to suppress the coupling vibration of a multi-flexible beam coupling system. The experimental setup of four-flexible beam coupling system is constructed. Piezoelectric sensors/actuators are used to detect vibration signals and suppress vibration. The finite element method (FEM) is used to establish the system dynamics model, and the model is modified by identifying parameters using the experimental data to obtain an accurate system model. The identified model is used as the simulation environment of RL algorithm. The multi-agent twin delayed deep deterministic policy gradient (MATD3) algorithm is designed to train the RL vibration controller through interaction with the simulation environment. The trained RL vibration controller is used to suppress the vibration of the four-flexible beam coupling system in simulation and experimental environment. Simulation and experimental results show that compared with proportional and derivative (PD) controller, the RL controller trained by the MATD3 algorithm has better control effect, especially for small amplitude vibration.