Bio-inspired hybrid flow control and gust mitigation using reinforcement learning

Passive flow control using a covert-feather inspired passively deployable flap attached on an airfoil provides significant lift improvements at post-stall angles of attack involving massive flow separation and vortex shedding. In this talk, we will describe a hybrid flow control strategy to achieve even greater aerodynamic benefit, where the stiffness of the hinge is actively varied in time to yield more favorable fluid-structure interaction (FSI) than for a constant stiffness. The control of such a strongly-coupled FSI system is challenging due to the nonlinear coupling between the passively oscillating flap and the unsteady flow. Therefore, we design a closed-loop feedback controller for temporally varying the stiffness using reinforcement learning (RL). In RL, the optimal control strategy is learned by trial-and-error where the actuator performs a variety of control actions, analyzes the resulting performance of the system and updates the controller strategy to maximize performance. The controller is constructed to (a.) maximize aerodynamic lift under steady free-stream conditions and (b.) minimize fluctuations in lift due to vortex gusts. The appropriate cost functions and learning algorithms for achieving the lift maximization and gust mitigation goals will be presented.