Advancing Deep Reinforcement Learning Strategies for Active Flow Control in Wings
Please login to view abstract download link
In this work we propose a framework that couples a GPU-enabled CFD solver with a deep reinforcement learning (DRL) agent, accelerating the collection of experiences during training. Additionally, the use of proper orthogonal decomposition (POD) modes is explored to reduce the noise introduced by flow turbulence, which can mask relevant state information at higher Reynolds numbers. In this approach, POD acts as a filter that extracts the most relevant flow features, enabling the agent to make more data-informed decisions. This methodology is first applied to a low Reynolds number case (NACA 0012 wing at Re = 1,000) and then evaluated in the turbulent regime (SD7003 wing at Re = 60,000).
