ABSTRACT: The predictive capabilities of turbulent flow simulations, critical for aerodynamic design and weather prediction, hinge on the choice of turbulence models. The abundance of data from experiments and simulations and the advent of machine learning have provided a boost to turbulence modeling efforts. However, simulations of turbulent flows remain hindered by the inability of heuristics and supervised learning to model the near-wall dynamics. We address this challenge by introducing scientific multi-agent reinforcement learning (SciMARL) for the discovery of wall models for large-eddy simulations (LES). In SciMARL, discretization points act also as cooperating agents that learn to supply the LES closure model. The agents self-learn using limited data and generalize to higher Reynolds numbers in reproducing key flow quantities. We test the discovered wall model to canonical flat plate boundary layers, which shows good predictable capabilities outside the Reynolds numbers used to train the model. We will discuss extensions to this model for flows with pressure-gradient effects.

BIO: Jane Bae is an Assistant Professor of Aerospace at the Graduate Aerospace Laboratories at Caltech. She received her Ph.D. in Computational and Mathematical Engineering from Stanford University in 2018.  She was a postdoctoral fellow in the Graduate Aerospace Laboratories at Caltech and the Institute for Applied Computational Science at Harvard University before joining the Caltech faculty. Her main research focuses on computational fluid mechanics, in particular on modeling and control of wall-bounded turbulence.