Reinforcement learning-based separated flow control over NLF(1)-0115

We use a reinforcement learning (RL) algorithm to design active flow controls for separated flows over a 15%-thick, natural laminar-flow airfoil NFL(1)-0115. Flow over the airfoil at a chord-based Reynolds number of 20.000 and free-stream Mach number of 0.1 are considered with an angle of attack of 10 degrees. Around the leading edge, two localized pulsed jets actuation are introduced in a feedback loop with multiple frequencies. To provide physical information guidance for the practical choice of control frequencies in RL, we conducted a resolvent analysis on the baseline time-mean flow to identify the frequency that provides substantial energy amplification. The present research also considers spatial windowing to localized forcing where an actuator may be placed to assess energy amplification. The combination of RL algorithms, pulsed jet actuators, and theoretical guidance derived from resolvent analysis holds promising potential for achieving enhanced flow control and improved aerodynamic performance in practical applications.