Deep reinforcement learning for active flow control in a turbulent separation bubble

Nature Communications, Год журнала: 2025, Номер 16(1)

Опубликована: Фев. 6, 2025

Abstract The control efficacy of deep reinforcement learning (DRL) compared with classical periodic forcing is numerically assessed for a turbulent separation bubble (TSB). We show that strategy learned on coarse grid works fine as long the captures main flow features. This allows to significantly reduce computational cost DRL training in turbulent-flow environment. On grid, able TSB area by 6.8%, while DRL-based achieves 9.0% reduction. Furthermore, agent provides smoother conserving momentum instantaneously. physical analysis reveals production large-scale counter-rotating vortices adjacent actuator pairs. It shown acts wide range frequencies sustain these time. Last, we also introduce our fluid dynamics and open-source framework suited next generation exascale computing machines.

Язык: Английский

Deep reinforcement learning for active flow control in bluff bodies: A state-of-the-art review

Ocean Engineering, Год журнала: 2025, Номер 327, С. 120989 - 120989

Опубликована: Март 18, 2025

Язык: Английский

On low-frequency pressure fluctuations and their impact on vehicle aerodynamics in a full-scale automotive wind tunnel

Physics of Fluids, Год журнала: 2025, Номер 37(4)

Опубликована: Апрель 1, 2025

Low-frequency pressure fluctuations pose significant challenges in the design and operation of full-scale automotive wind tunnels. However, mutual interactions between tunnel test vehicle aerodynamics remain unclear. This study employs high-fidelity numerical simulations to investigate characteristics low-frequency their influence on aerodynamics. A virtual was constructed replicate geometry a real-world facility, Detached Eddy Simulations were utilized capture dynamics within section. performed for three configurations: (a) an empty section, (b) mounted (c) open-road setup. Reduced order analyses applied compare spatiotemporal velocity fields across all configurations. The simulation results show that have minor time-averaged drag coefficient, despite noticeable enhancement temporal variation drag. global standing-wave pattern governs field pulsating is revealed by reduced-order analyses, where characteristic frequency found decrease due installation vehicle. In addition, coupling effect observed coherent structures shear layer wake, elucidating complex aerodynamic are inherent These provide useful insights guide optimization experimental execution, which will ultimately enhance precision reliability evaluations

Язык: Английский