A well-known Kármán vortex street is typically formed in the wake of the flow over a bluff body exerting an oscillating value of the force. This unsteadiness may cause structural damages due to the coupling of the body vibrations and pressure fluctuations of the fluid. Over time, many flow control strategies have been proposed to influence and suppress these unwanted dynamical features. The overall classification includes passive and active methods due to the possible energetic input to the flow. Active methods represent open- and closed-loop control, depending on the presence of feedback from sensors to actuators with a further update of a control signal. An appealing way to design new closed-loop control strategies is to rely on the so-called data-driven and learning-based methods, which lately receive well-deserved attention.
Employees of the Laboratory of Applied Digital Technologies of the International Mathematical Center of Mathematics and Mechanics Department of NSU examined the problem of flow around a cylinder that can have rotational oscillation around its axis. Thus, the function of angular velocity was the only parameter. The deep reinforcement learning optimization algorithm that was used changes the angular velocity of the cylinder to reduce the oscillating forces acting on it. With the help of minor rotational vibrations of the cylinder around its axis, it was possible to reduce the drag force by 16% and stabilize the unstable system. The results of this work are published in the Energies journal.