Deep reinforcement learning-based active control for drag reduction of three equilateral-triangular circular cylinders

Abstract

Deep reinforcement learning (DRL) is gaining attention as a machine learning tool for effective active control strategy development. This study focuses on employing DRL to develop an efficient active control strategy for flow around three circular cylinders arranged in an equilateral-triangular configuration in a two-dimensional channel. The analysis of control outcomes reveals that DRL induces vortices of varying sizes between the cylinders, resulting in large elliptical vortices at the rear. This enhancement in flow stability leads to a significant 40.40% reduction in cylinder drag force and an approximate 8.23% decrease in overall drag oscillations. Our research represents a pioneering application of DRL for stabilizing complex flow around multiple cylinders, yielding remarkable control effectiveness. The noteworthy outcomes in controlling the stability of complex flows highlight the capability of DRL to grasp intricate nonlinear flow dynamics, showcasing its potential for investigating active control strategies within complex nonlinear systems.