Flow topology of deep reinforcement learning drag-reduced bluff body wakes

The wake topology of a bluff body, representative of a commercial road vehicle, manipulated by different control laws for pulsed jets located at the trailing edges of the model is presented and discussed. The parameters of the control laws have been identified through previous work, in which a deep reinforcement learning (DRL) algorithm was trained under different conditions to achieve drag reduction first and also taking the energy budget into account. The focus of this work is to understand the mechanisms through which the DRL agent can reach the objective in four distinct cases, with different sizes of the state and reward definition. Planar and stereoscopic particle image velocimetry measurements were carried out at different planes in the body's wake. The findings suggest that, when large drag reduction conditions are achieved, the recirculating flow bubble is shortened in the streamwise direction, the wake becomes symmetrical in the streamwise-vertical plane at the symmetry station along the cross-stream direction, and there is a substantial pressure recovery at the base of the model. In these conditions, the wake topology drastically changes with respect to that of the natural case. Conversely, when the energy budget is introduced, the modification of the recirculating flow bubble is smaller as a consequence of the reduced actuation. This study, thus, while complementing previous work with flow physics analyses, gives valuable insights on the wake topologies to aim for when targeting pressure drag reduction through active flow control strategies.