Liutex and Deep Reinforcement Learning Based Active Flow Control

The mainstream flow control methods can be broadly categorized into two types: passive or active flow control. Passive flow control has numerous examples, so attention has shifted more towards active flow control. Over recent years, machine learning has become a popular technology to fluid dynamics problems. Meanwhile, the Liutex vector has emerged as a new vortex identification standard, which outperforms previous methods. In this study, we use the classic two-dimensional flow around a cylinder ( \(Re = 100\) ). Adapting an open-source reinforcement learning based strategy for active flow control, we modify the input state parameters and reward functions to obtain a new control strategy. Specifically, we increase the number of state parameters from 151 to 453 which include Liutex vector and Shear vector, and also propose a new reward function utilizing Liutex vectors to control the vortex shedding. We find that the new state and reward function are effective for control. It can achieve drag reduction effects, and weaken the Karman vortex street.