Attenuating wake features of an underwater vehicle using deep reinforcement learning in an experimental environment

The wake features of underwater vehicles play a crucial role in many underwater detection technologies. Effectively attenuating these features holds significant value in naval engineering applications. In this study, we propose a closed-loop control strategy aimed at attenuating wake features using deep reinforcement learning (DRL)-guided active flow control. An underwater vehicle model is towed in an automatic water tank at a Reynolds number of 105, based on the model length and towing velocity. Jet actuators positioned at the tail regulate wake flow dynamics, while velocities at 25 monitoring locations provide real-time feedback via a particle image velocimetry system and a graphics processing unit-accelerated optical flow algorithm. The DRL framework determines the control strategy through stochastic training in the towing tank. The trained DRL agent learns to suppress the variation rate of wake velocity to nearly zero and reduces turbulent kinetic energy by over 16% at 2.5 diameters downstream of the stern. Transfer learning applied to a higher Reynolds number case (Re=2×105) demonstrates superior performance compared to direct learning and realizes similar wake feature attenuation performance. This study validates the concept of hydrodynamic stealth in experimental and turbulent flow environments and confirms the effectiveness of DRL in developing robust flow control strategies for complex hydrodynamic scenarios.