Aerodynamic force reduction of rectangular cylinder using deep reinforcement learning-controlled multiple jets

Abstract

The drag and lift forces of rectangular cylinders with four aspect ratios (AR) are mitigated at a Reynolds number of 1000 using deep reinforcement learning (DRL) controlled multiple jet actuators at four corners of the cylinders in this study. Surface pressure probes are set to monitor the flow state, featuring a practical engineering solution. For each jet actuator, the control law is optimized using the powerful capabilities of DRL with the surface pressure signals as the control input and the jet velocities as the control output. Different jet configurations are tested on the rectangular cylinder with AR = 1, 2, 3, and 5 to choose the best multiple jet control configurations. The results have shown that under eight independent jets at the rectangular cylinder corners, the mean drag coefficient is reduced by 77.1%, 39.9%, 52.4%, and 2.8% for the rectangular cylinder with AR = 1, 2, 3, and 5, respectively. Furthermore, the lift fluctuation is reduced by 94.2%, 66.6%, 77.1%, and 59.8%, indicating a significant stabilization of the wake. This configuration with eight independent jets shows the top performance among these jet configurations. This study highlights the significance of the multiple jet actuation and can guide the practical application of DRL-based active flow control of rectangular cylinders.