Application of deep reinforcement learning for aerodynamic control around an angled airfoil via synthetic jet.

Abstract

This study introduces a Deep Reinforcement Learning (DRL) framework for aerodynamic control of weakly compressible flow around an angled airfoil, aiming to optimize lift and drag through synthetic jet actuation. By combining pressure and velocity information from strategically placed sensors, the framework enhances the agent's learning efficiency and control precision. Results demonstrate that integrating both variables improves total reward, reduces vortex shedding, stabilizes the wake, and minimizes lift fluctuations while lowering drag. The control policy, trained through 300 episodes using a Deep Q-Network (DQN) with five hidden layers of 128 neurons, achieves stable convergence and effective wake stabilization. Among the tested architectures, i.e. Traditional DQN, Double DQN, and Dueling DQN, the latter yields the most consistent learning behavior and highest performance by distinguishing state-value and advantage functions. Overall, the proposed DRL-based approach provides an efficient and robust strategy for active flow control in compressible aerodynamic applications, highlighting its potential for future engineering and aerospace systems.