Title: Leveraging modeling hierarchies in the Exascale era: applications to combustion technologies

Abstract: As we approach the confluence of widespread use of machine learning techniques and simulations running at exascale, several important challenges will need to be addressed. In this talk, we explore some of these challenges, with a specific focus on combustion applications. We discuss a combustion simulation code, PeleC, and its performance characteristics on the fastest supercomputers available today. We look at leveraging the resulting high-fidelity simulations to construct data-driven models for lower-fidelity simulations. We then examine how to adapt reinforcement learning methods to explore a modeling hierarchy and determine adequate control strategies for combustion technologies.

Bio: Marc Henry de Frahan is a computational scientist at the National Renewable Energy Laboratory, where he works on improving next-generation wind and combustion processes. As part of the Exascale Computing Project, Marc develops high-fidelity turbulence models to enhance simulation accuracy and efficient numerical algorithms for future high-performance computing hardware architectures. In addition to traditional physics-based modeling, he is integrating deep neural networks into modeling and reinforcement learning into advanced control strategies. Marc obtained his PhD in Mechanical Engineering in 2016 from the University of Michigan.

Zoom information to connect:

Link: https://umich.zoom.us/j/98133041706

Passcode: 762808