Computational Methods for Data-Driven and Reduced-Order Modeling of Multiscale Physics

Abstract

Multi-resolution DMD is a critically enabling algorithm which allows for the data-driven construction of ROMs at each level of temporal activity. Data-driven models can be inferred through a number of techniques, including DMD, Koopman theory and sparse identification of nonlinear dynamics. And importantly, the innovation of DMD with control can be used to connect ROMs at different temporal scales. Specifically, fast scales act as input for slow scales and vice-versa. Thus the mathematical architecture allows for the construction of heterogeneous, multi-scale ROMs, each of which acts as an input to the other levels. The quality and appropriateness of the multi-scale models can be evaluated through information theory criteria, providing a robust cross-validation of the inferred ROMs.

Document Details

Document Type
DoD Grant Award
Publication Date
Sep 11, 2017
Source ID
FA95501710329

Entities

People

  • J. Nathan Kutz

Organizations

  • Air Force Office of Scientific Research
  • United States Air Force
  • University of Washington

Tags

Readers

  • Computational Fluid Dynamics (CFD)
  • Computational Modeling and Simulation

Technology Areas

  • AI & ML