Effect of Synthetic Jets Actuator Parameters on Deep Reinforcement Learning-Based Flow Control Performance in a Square Cylinder

We utilize deep reinforcement learning (DRL) algorithms to precisely control the mass flow rates of synthetic jets located on the upper and lower surfaces of a square cylinder, achieving effective active flow control. Through DRL-based active flow control (AFC) technology, we significantly reduce the lift and drag coefficients of the square cylinder at Re=100 and Re=500, while completely suppressing vortex shedding in the wake flow field.Additionally, we conduct a sensitivity analysis of the position and width parameters of the synthetic jets regarding flow control performance. Our observations indicate that positioning the synthetic jets near the trailing edge corners of the square cylinder, as opposed to the leading edge corners, completely suppresses vortex shedding, resulting in more stable controlled flow coefficients, C_D and C_L. When the synthetic jets are positioned at the trailing edge corners, flow control reduces the mean drag coefficient by 14.4% and the standard deviation of the lift coefficient by 86.1% for the baseline flow at Re=100. For the baseline flow at Re=500, flow control reduces the mean drag coefficient by 51.4% and the standard deviation of the lift coefficient by 90.5%. At both Reynolds numbers, vortex shedding in the wake flow field is completely suppressed. Furthermore, using narrower synthetic jets results in a lower reduction rate of the standard deviations of C_D and C_L, while increasing the mean and standard deviation of the mass flow rate of the jets used for flow control. This study provides guidance on optimizing the width and position of synthetic jets for DRL-based active flow control.