Application of Data-Driven Manifold Modeling for Turbulent Combustion (NREL Internal Use Only)

Research output: NRELPoster

Abstract

While the observed thermochemical states in turbulent combustion systems typically are confined to a low-dimensional manifold in state space, complexities in real combustion systems often mean that the observed manifolds differ from those implied by common physical assumptions. The increasing availability of high-fidelity simulation and experimental data presents an opportunity for data-driven development of manifold-based models for turbulent combustion. In this work, we apply the recently developed Co-optimized Machine-Learned Manifolds (CMLM) approach, which combines the identification of manifold parameters, nonlinear mapping to reaction rates and physical properties, and closure for LES into a single neural network structure. This model is integrated into the Pele suite of reacting flow solvers to allow for a posteriori evaluation. Previous work has demonstrated improved predictive performance for reaction rates in an a-priori setting, both for idealized data from 1D flames and for data from direct numerical simulations (DNS) of a turbulent ignition kernel. Here, we demonstrate that these improvements carry over when the model is coupled with the flow solver for a-posteriori evaluation. We also evaluate how the model robustness is affected by the data selected for training, including training with multi-fidelity data. The CMLM approach provides a method of augmenting the one-dimensional data sets typically used to generate physics-based manifolds with more complex data sets to generate data-driven manifold models, which is now demonstrated in a-posteriori combustion simulations.
Original languageAmerican English
StatePublished - 2022

Publication series

NamePresented at the 39th International Symposium on Combustion, 24-29 July 2022, Vancouver, British Columbia, Canada

NREL Publication Number

  • NREL/PO-2C00-83441

Keywords

  • machine learning
  • reduced-order manifolds
  • turbulent combustion

Fingerprint

Dive into the research topics of 'Application of Data-Driven Manifold Modeling for Turbulent Combustion (NREL Internal Use Only)'. Together they form a unique fingerprint.

Cite this