Skip to main content
U.S. Department of Energy
Office of Scientific and Technical Information

DPM: A deep learning PDE augmentation method with application to large-eddy simulation

Journal Article · · Journal of Computational Physics
 [1];  [2];  [3]
  1. Univ. of Oxford (United Kingdom). Mathematics; Univ. of Illinois at Urbana-Champaign, IL (United States). Dept. of Industrial & Systems Engineering; Univ. of Illinois at Urbana-Champaign, IL (United States)
  2. Univ. of Illinois at Urbana-Champaign, IL (United States). The Center for Exascale Simulation of Plasma-coupled Combustion. Coordinated Science Lab.
  3. Univ. of Illinois at Urbana-Champaign, IL (United States). Aerospace Engineering

A framework is introduced that leverages known physics to reduce overfitting in machine learning for scientific applications. The partial differential equation (PDE) that expresses the physics is augmented with a neural network that uses available data to learn a description of the corresponding unknown or unrepresented physics. Training within this combined system corrects for missing, unknown, or erroneously represented physics, including discretization errors associated with the PDE's numerical solution. For optimization of the network within the PDE, an adjoint PDE is solved to provide high-dimensional gradients, and a stochastic adjoint method (SAM) further accelerates training. Additionally, the approach is demonstrated for large-eddy simulation (LES) of turbulence. High-fidelity direct numerical simulations (DNS) of decaying isotropic turbulence provide the training data used to learn sub-filter-scale closures for the filtered Navier–Stokes equations. Out-of-sample comparisons show that the deep learning PDE method outperforms widely-used models, even for filter sizes so large that they become qualitatively incorrect. It also significantly outperforms the same neural network when a priori trained based on simple data mismatch, not accounting for the full PDE. Measures of discretization errors, which are well-known to be consequential in LES, point to the importance of the unified training formulation's design, which without modification corrects for them. For comparable accuracy, simulation runtime is significantly reduced. A relaxation of the typical discrete enforcement of the divergence-free constraint in the solver is also successful, instead allowing the DPM to approximately enforce incompressibility physics. Since the training loss function is not restricted to correspond directly to the closure to be learned, training can incorporate diverse data, including experimental data.

Research Organization:
Univ. of Illinois at Urbana-Champaign, IL (United States)
Sponsoring Organization:
USDOE National Nuclear Security Administration (NNSA); National Science Foundation (NSF)
Grant/Contract Number:
NA0002374
OSTI ID:
1850305
Alternate ID(s):
OSTI ID: 1763780
Journal Information:
Journal of Computational Physics, Journal Name: Journal of Computational Physics Journal Issue: C Vol. 423; ISSN 0021-9991
Publisher:
ElsevierCopyright Statement
Country of Publication:
United States
Language:
English

References (38)

Effects of the Computational Time Step on Numerical Solutions of Turbulent Flow journal July 1994
An Analysis of Numerical Errors in Large-Eddy Simulations of Turbulence journal April 1996
On the Effect of Numerical Errors in Large Eddy Simulations of Turbulent Flows journal March 1997
Application of a fractional-step method to incompressible Navier-Stokes equations journal June 1985
A further study of numerical errors in large-eddy simulations journal January 2003
The use of explicit filters in large eddy simulation journal August 2003
Linear eddy mixing based tabulation and artificial neural networks for large eddy simulations of turbulent flames journal January 2010
Large eddy simulation of extinction and reignition with artificial neural networks based chemical kinetics journal March 2010
Machine learning-assisted early ignition prediction in a complex flow journal August 2019
High order conservative finite difference scheme for variable density low Mach number turbulent flows journal July 2008
Machine learning strategies for systems with invariance properties journal August 2016
Semi-implicit iterative methods for low Mach number turbulent reacting flows: Operator splitting versus approximate factorization journal December 2016
DGM: A deep learning algorithm for solving partial differential equations journal December 2018
Evaluation of subgrid-scale models using an accurately simulated turbulent flow journal March 1979
Numerical simulation of compressible homogeneous flows in the turbulent regime journal September 1987
Large-eddy simulation of the turbulent mixing layer journal May 1997
Optimal LES formulations for isotropic turbulence journal November 1999
Reynolds averaged turbulence modelling using deep neural networks with embedded invariance journal October 2016
Deep learning of vortex-induced vibrations journal December 2018
Numerical Calculation of Time-Dependent Viscous Incompressible Flow of Fluid with Free Surface journal January 1965
Grid-independent large-eddy simulation using explicit filtering journal October 2010
A dynamic regularized gradient model of the subgrid-scale stress tensor for large-eddy simulation journal February 2016
Investigations of data-driven closure for subgrid-scale stress in large-eddy simulation journal December 2018
A dynamic subgrid‐scale eddy viscosity model journal July 1991
Subgrid‐scale backscatter in turbulent and transitional flows journal July 1991
A proposed modification of the Germano subgrid‐scale closure method journal March 1992
Numerical solution of the Navier-Stokes equations journal January 1968
Energy spectrum in high-resolution direct numerical simulations of turbulence journal December 2016
Perspective on machine learning for advancing fluid mechanics journal October 2019
Adjoint Equations in Stability Analysis journal January 2014
Machine Learning for Fluid Mechanics journal September 2019
Discrete Conservation Properties of Unstructured Mesh Schemes journal January 2011
Numerical Simulation of Turbulent Flows journal January 1984
New Trends in Large-Eddy Simulations of Turbulence journal January 1996
DIRECT NUMERICAL SIMULATION: A Tool in Turbulence Research journal January 1998
Scale-Invariance and Turbulence Models for Large-Eddy Simulation journal January 2000
Shape Optimization in Fluid Mechanics journal January 2004
General Circulation Experiments with the Primitive Equations: i. the Basic Experiment* journal March 1963

Cited By (4)


Similar Records

Neural networks for large eddy simulations of wall-bounded turbulence: numerical experiments and challenges
Journal Article · Mon Jul 17 00:00:00 EDT 2023 · European Physical Journal. E, Soft Matter (print) · OSTI ID:2311707

Multi-resolution partial differential equations preserved learning framework for spatiotemporal dynamics
Journal Article · Fri Jan 12 23:00:00 EST 2024 · Communications Physics · OSTI ID:2527398

Learning generative neural networks with physics knowledge
Journal Article · Tue May 17 00:00:00 EDT 2022 · Research in the Mathematical Sciences · OSTI ID:2481112