Viscoelastic Turbulent Channel Drag Reduction via Opposition Control Using Multi-Agent Reinforcement Learning
Please login to view abstract download link
Drag reduction in turbulent flows is critical for reducing energy consumption in pipelines and transportation systems. While polymer additives are known to attenuate turbulence in wall-bounded flows, the interplay between viscoelastic turbulence and active flow control remains poorly understood. We perform direct numerical simulations of viscoelastic turbulent channel flow at a bulk Reynolds number of 2800 and Weissenberg number of 4, using the FENE-P constitutive model under constant mass flux conditions. Opposition control is implemented by sensing wall-normal velocity fluctuations at an off-wall detection plane and applying spatially distributed blowing/suction at the walls. Control decisions are made by a multi-agent reinforcement learning framework in which agents share a Deep Deterministic Policy Gradient (DDPG) policy, enabling coordinated, decentralized actuation with centralized training. This approach leverages localized sensing while maintaining a globally consistent strategy. The ongoing work investigates the effectiveness of such control in reducing drag, modifying near-wall coherent structures, and uncovering possible synergies between polymer-induced and control-induced turbulence suppression. Anticipated findings aim to advance both the physical understanding of viscoelastic drag reduction and the development of robust, adaptive control methods for complex fluids. (General: 800)
