Reinforcement learning vibration control for a flexible hinged plate

Abstract A reinforcement learning (RL) vibration control method for a flexible hinged plate is developed, including combined bending and torsional vibrations. The experimental setup is constructed. Two laser displacement sensors are used as detection sensors, and two-channel piezoelectric actuators are used to suppress both bending and torsional vibrations simultaneously. A method combining finite element modeling (FEM) and experimental identification is applied to obtain an accurate model of the system. The obtained model is constructed as an RL environment, and a deep deterministic policy gradient (DDPG) RL algorithm based on the priority experience replay (PER) is designed to train modal controllers. The RL modal controllers were transferred to the real experimental system by using the parameter transfer method, and the vibration control experiments were conducted by combining the bending and torsional vibration control method. Simulation and experimental results demonstrate that the controller trained by the proposed deep RL algorithm has better control effects compared with PD control, especially for small amplitude vibration.