Experimental Study on Application of Distributed Deep Reinforcement Learning to Closed-loop Flow separation Control over an Airfoil

摘要

This paper experimentally investigates a closed-loop flow separation control system on a NACA 0015 airfoil using a DBD plasma actuator at the chord-Reynolds number of 63,000. The closed-loop control system is constructed utilizing the Deep Reinforcement Learning(DRL). The plasma actuator is installed to the surface of the airfoil at 5% of the chord length from the leading edge and driven with AC voltage. The time series data of surface pressure are used as the input data to the neural network, and the neural network is trained to select the optimum burst frequency of the actuator at angles of attacks of 15 degrees. Ape-X DQN, which is the latest algorithm of DRL is used to improve the training of the neural network. As a result, the neural network is trained stably in Ape-X DQN compared to Deep Q Network (DQN), which is the old algorithm. The result of the time-averaged pressure measurements indicates that the flow controlled by the network trained by Ape-X DQN is suppressed more than the network trained by DQN at angle of attack of 15degrees.