Turbulence control for drag reduction through deep reinforcement learning

Deep reinforcement learning was applied to turbulence control for drag reduction in direct numerical simulation of turbulent channel flow. The learning determines the optimal distribution of wall blowing and suction based on the wall shear stress information. From an investigation of the optimal actuation fields, two distinct drag reduction mechanisms were identified. One of them, which had not previously been recognized, attempts to cancel the near-wall sweep and ejection events.