Reinforcement-Learning-Driven Active Flow Control for Airfoil Pitching Moment Trim

Deep Reinforcement Learning (DRL) is performed to discover closed-loop fluidic Active Flow Control (AFC) strategies for achieving zero pitching moment, i.e. the aerodynamic trim condition, about a moment reference point of a NACA0012 airfoil. The DRL agent controls the velocity of the AFC fluidic jets, while observing a sufficiently-Markov state in the form of velocity measurements probed from several locations around the airfoil in two-dimensional compressible Navier-Stokes simulations. The reward function is a function of the pitching moment and drives the learning agent towards discovering optimal control strategies to obtain a net-zero pitching moment. This study is a step towards control of aerodynamic moments that can yield novel approaches for replacing or enhancing the mechanical aerodynamic control surfaces on aircraft wings and wind turbine blades.