Deep reinforcement learning for flow-induced noise control: Multi-objective optimization of acoustic suppression and drag reduction

Environmental noise from transportation systems poses critical challenges to human environments, necessitating effective flow control strategies. This study examines the coupling between near-field pressure fluctuations and far-field overall sound pressure levels (OASPLs) in multi-cylinder rotating flows. Numerical simulations employ the immersed boundary method with the Ffowcs Williams–Hawkings acoustic analogy, while active flow control is achieved using deep reinforcement learning (DRL). The trained agent optimizes cylinder rotation to suppress wake vortex shedding, substantially reducing near-field pressure fluctuations, acoustic source intensity, and far-field OASPL. To balance acoustic mitigation with aerodynamic performance, a multi-objective optimization framework is developed, achieving simultaneous drag reduction and noise suppression. Analysis of control strategies shows that the agent suppresses boundary-layer vortex formation and delays wake separation through rotational control. The findings provide new insights into adaptive noise control with direct relevance to aerospace applications.