An interactive platform of deep reinforcement learning and wind tunnel testing

Flow around bluff bodies is a classic problem in fluid mechanics, and flow control is a critical approach for manipulating the aerodynamic characteristics of bluff bodies. Recently, deep reinforcement learning (DRL) has emerged as a highly potential method of flow control. However, the application of DRL to wind tunnel testing involves significant obstacles, which can be classified into software, hardware, and interaction challenges. These challenges make the application of DRL-based wind tunnel testing particularly complex and challenging for many researchers. To address these challenges, this paper proposes a novel DRL-based wind tunnel testing platform, named DRLinWT. DRLinWT introduces a universal adapter capable of managing interactive communications across multiple mainstream communication protocols and integrates commonly used reinforcement learning libraries, thereby significantly reducing the interaction cost between DRL algorithms and wind tunnel tests. Using this platform, a DRL-based flow control experiment for a square cylinder in three flow fields of varying complexity was conducted.