About

The Turbulence Collaboration started as an exploratory idea between graduate school friends that turned into a potentially important research project. Turbulence is a key driver in climate physics, internal combustion engines, geophysical flows etc.,and getting turbulence right is really key to developing truly predictive modeling frameworks. It is for this reason, Dr. Richard Feynman called turbulence as the “most important unsolved problem in classical physics”.

The primary goal behind this collaborative effort is to develop and apply advanced machine learning algorithms to improve our understanding of the physics of turbulence. These learned algorithms can be then used to improve existing model fidelity.

This is a multi-institution effort between academia, national laboratories and the industry and some of the problems we are currently looking at include improving popular two-equation turbulence models (under simplified assumptions), geophysical turbulence as well as building a pathway to create better climate sub-models.

Participating Institutions

The turbulence collaboration project has participating members from the academia, industry as well as US Department of Energy laboratories

  • Los Alamos National Laboratory
  • University of Massachusetts
  • NVIDIA Corp.
  • University of Toronto/Vector Institute
  • German Research Center for Artifical Intelligence
  • Rice University

Members

Mentors

Publications

  • Turbulence Forecasting via Neural ODE, to appear, NeurIPS ML4PS 2019, Vancouver, BC, Canada
  • A data-driven approach to modeling turbulent decay at non-asymptotic Reynolds numbers, to appear, American Physical Society DFD 2019, Seattle, WA
  • Climate sub-closures using Neural ODE, talk submitted to GTC 2020
  • Continuous time networks for two-equation turbulence models, poster submitted to GTC 2020

On-going Research

  • Modeling buoyancy flux in geophysical flows using continous time networks (for homogenous stratified flows)
  • Explainability of black-box ML models such as Neural ODEs -- deep dive into latent space learnings
  • Aerodynamic drag optimization using Neural ODEs (project in collaboration with Carnegie Mellon University)