Large eddy simulation with flamelet progress variable approach combined with artificial neural network acceleration

Abstract

In the context of large eddy simulation of turbulent reacting flows, flamelet-based models are key to affordable simulations of large and complex systems. However, as the complexity of the problem increases, higher-dimensional look-up tables are required, rendering the conventional look-up procedure too demanding. This work focuses on accelerating the estimation of flamelet- based data for the flamelet/progress variable model via an artificial neural network. The neural network hyper-parameters are defined by a Bayesian optimization and two different architectures are selected for comparison against the classical look-up procedure on the well known Sandia flame D. The performance in terms of execution time and accuracy are analyzed, showing that the neural network model reduces the computational time by 30%, as compared to the traditional table look-up, while retaining comparable accuracy.

           Type               Conference paper
           Publication   In American Institute of Aeronautics and Astronautics

BibTex

@inproceedings{angelilli2021large,
  title={Large eddy simulation with flamelet progress variable approach combined with artificial neural network acceleration},
  author={Angelilli, Lorenzo and Ciottoli, Pietro Paolo and Malpica Galassi, Riccardo and Hernandez Perez, Francisco E and Soldan, Mattia and Lu, Zhen and Valorani, Mauro and Im, Hong G},
  booktitle={AIAA Scitech 2021 Forum},
  pages={0412},
  year={2021}
}
Mattia Soldan
Mattia Soldan
PhD Student - Electrical and Computer Engineering

My research interests are settled at the intersection between Computer Vision and Natural Language Processing.