Criteria to switch from tabulation to neural networks in computational combustion

Nikolaou, Zacharias M.; Domingo, Pascale

doi:10.1016/j.combustflame.2022.112425

Cited by 4 publications

(2 citation statements)

References 29 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…The DNN requires only 2% of the QFM's memory, while the computational cost remains similar. However, the performance heavily depends on DNN architecture [48] and hardware, which will be the subject of future studies. This work can serve as the basis to investigate more complex flame configurations in future works, involving aspects such as differential diffusion and stretch effects, turbulent combustion, or hydrogen/hydrocarbon fuel blends.…”

Section: Discussionmentioning

confidence: 99%

Application of dense neural networks for manifold-based modeling of flame-wall interactions

Bissantz¹,

Karpowski²,

Steinhausen³

et al. 2023

Applications in Energy and Combustion Science

View full text Add to dashboard Cite

Section: Discussionmentioning

confidence: 99%

Application of dense neural networks for manifold-based modeling of flame-wall interactions

Bissantz¹,

Karpowski²,

Steinhausen³

et al. 2023

Applications in Energy and Combustion Science

View full text Add to dashboard Cite

“…Data-based methods on the other hand (Maulik & San 2017; Nikolaou & Vervisch 2018) require high-quality and relatively large amounts of data. In addition, their computational efficiency depends strongly on the structure (number of layers, number of neurons) of the network, and there exist in fact bounds above which a neural network will always underperform (more floating-point operations) in comparison to using a simple tabulation approach (Nikolaou, Vervisch & Domingo 2022). One solution to this problem is to use approximate reconstruction operators derived from truncated Taylor-series expansions of the inverse filter operator.…”

Section: Introductionmentioning

confidence: 99%

Revisiting the modelling framework for the unresolved scalar variance

Nikolaou,

Domingo,

Vervisch

2024

J. Fluid Mech.

Self Cite

View full text Add to dashboard Cite

The unresolved scalar variance in large-eddy simulations of turbulent flows is a fundamental physical and modelling parameter. Despite its importance, relatively few algebraic models have been developed for this important variable with the most prominent models to date being the classic scale-similarity and gradient models. In this work a new generalized modelling framework based on reconstruction has been developed, which in contrast to classic modelling approaches allows the construction of base static variance models of arbitrary accuracy. It is demonstrated that higher-order reconstructions naturally lead to base static variance models of increased accuracy, and that the classic scale-similarity and gradient models are subsets of more general and higher-order models. The classic scale-similarity assumption for developing dynamic models is also revisited, and it is demonstrated that this can essentially be reinterpreted as a two-level reconstruction approach. Based on this result, a new general methodology is proposed that allows the construction of dynamic models for any given base static model, and a corresponding general reconstruction operator, algebraic or iterative. Consequently, improved static and dynamic models for the scalar variance are developed. The newly developed models are then thoroughly tested a priori using two high-fidelity direct numerical simulation databases corresponding to two substantially different flame and flow configurations, and are shown to outperform classic algebraic models for the variance.

show abstract

Acceleration of the complex reacting flow simulation with a generalizable neural network based on meta-learning

Bai,

Huai,

Liu

et al. 2024

Fuel

View full text Add to dashboard Cite

Criteria to switch from tabulation to neural networks in computational combustion

Cited by 4 publications

References 29 publications

Application of dense neural networks for manifold-based modeling of flame-wall interactions

Application of dense neural networks for manifold-based modeling of flame-wall interactions

Revisiting the modelling framework for the unresolved scalar variance

Acceleration of the complex reacting flow simulation with a generalizable neural network based on meta-learning

Contact Info

Product

Resources

About