Learning in Multi Scale Models with Stochastic Source Coupling

Abstract

A learning framework is proposed for the simulation of multi physics and multiphase environments whose solution is rendered stochastic because of empiricism or inaccuracy of data in interphysics and interphase source coupling. Based upon the identification of regions in the time dependent local physics where uncertainties are amplified at the fastest rate, and its connection with global measures of the solution, a feedback learning is designed. By means of local learning the framework limits uncertainty in the source coupling and the solution without a need for sizeable global estimates. This makes the learning framework computationally feasible for large problems with intrinsic multi physics. An improved predictive capability of multi phase, multi physics is of importance to a wide range of technologies that are relevant to the Air Force, ranging from propulsion devices such as ramjets, scramjets, rotating detonation engines, rocket engines to energetic materials in blast waves and explosives. In general, the physics of chemically reacting turbulent flow with gasliquid solid interfacial dynamics involve significant modeling that on the process scale of these technologies can hardly be predicted with engineering accuracy. A learning approach that quantifies and reduces uncertainty and enhances accuracy will lead to improved designs and prototyping.

Document Details

Document Type
DoD Grant Award
Publication Date
Jan 14, 2022
Source ID
FA95501910387

Entities

People

  • Gustaaf B Jacobs

Organizations

  • Air Force Office of Scientific Research
  • Salk Institute for Biological Studies
  • United States Air Force

Tags

Fields of Study

  • Physics

Readers

  • Agent-Based Social Robotics and Mobile-Assisted Learning in Virtual Environments.
  • Combustion and Flow Dynamics.
  • Computational Fluid Dynamics (CFD)