Active flow control of square cylinder adaptive to wind direction using deep reinforcement learning

Abstract

This study proves the effectiveness of deep reinforcement learning (DRL) as a valuable tool for addressing complicated active flow control challenges, especially when employing flow fields characterized by strong nonlinearity and various wind attack angles. It demonstrates that employing multiple jets and surface pressure probes can achieve an ideal control performance, effectively diminishing aerodynamic forces and optimizing flow stability around the square cylinder under different wind attack angles. These findings enhance the potential for the practical application of DRL-based flow control strategies in engineering, and further progress toward real-world applications.

Topics & Concepts