Reducing the Drag of a Bluff Body by Deep Reinforcement Learning

• DOI: 10.48550/arxiv.2305.03647
• Publication Date: 2023-01-01
• Authors: Enrico Ballini; Alberto Silvio Ch...; Stefano Micheletti
• Keywords: Bluff

We present a deep reinforcement learning approach to classical problem in fluid dynamics, i.e., the reduction of drag bluff body. cast as discrete-time control with continuous action space: at each time step, an autonomous agent can set flow rate two jets fluid, positioned back The agent, trained Proximal Policy Optimization, learns effective strategy make interact vortexes wake, thus reducing drag. To tackle computational complexity dynamics simulations, which would training procedure prohibitively expensive, we train on coarse discretization domain. provide numerical evidence that policy this approximate environment still retains good performance when carried over denser mesh. Our simulations show considerable consequent saving total power, defined sum power spent by system and force, amounting 40% compared reference body without any jet. Finally, qualitatively investigate learnt neural network. observe it achieves frequency formation activating accordingly, blowing them away off rear surface.