Machine Learning Assisted Mimetic Framework For Burgulence Modeling

October 11, 2024

TIME: 3:30 PM

LOCATION: GMCS 314

SPEAKER: Anand Srinivasan, San Diego State University

ABSTRACT:

The modeling of the subgrid scale stress tensor closure term for turbulent flows is a challenge and is an area of active research. The underlying partial differential equations for turbulent flows contain nonlinear convective terms which can lead to aliasing errors, and requires special attention for spatial discretization. The mimetic difference methods are based on mirroring the fundamental vector calculus identities of div and grad, and possess a discrete equivalent of a global conservation law, thereby leading to accurate discretizations of the nonlinear convective terms. In this work, a machine learning based mimetic framework is proposed to uncover the subgrid scale stress tensor terms for modeling turbulence of the Burgers equation (i.e., Burgulence modeling). The one-dimensional nonlinear Burgers equation is used to evaluate the subgrid scale turbulence model of large eddy simulations. The numerical solution of the Burgers equation capturing all scales of motion is evaluated using direct numerical simulations (DNS) by implementing the high order Corbino-Castillo mimetic operators, along with the third order Runge Kutta temporal scheme. The filtered-DNS velocity profiles are then used as the ground truth in a priori testing for training a neural network to uncover turbulence modeling using a data-driven framework.

HOST: Jose Castillo

VIDEO: