Abstract:The Flame Surface Density (FSD) model is an affordable combustion model that has been widely used in simulating turbulent premixed flames. In Large Eddy Simulations (LES) with FSD, the combined effect of reaction and diffusion is governed by the Filtered Flame Front Displacement (FFFD) term. While the existing modelling approaches for this term are computationally cost-effective, their predictions still show inconsistencies in certain cases. This study aims to address these inconsistencies by generating Machine Learning (ML) models for the FFFD and FSD terms using the DNS data of a turbulent premixed jet flame. With this approach, the relevance of certain input parameters as well as certain modelling assumptions used for the FFFD term are assessed. Overall, it is found that the resolved curvature term is the most important input parameter to consider and that the resolved progress variable should also be considered in the models. It is shown that the ML models perform significantly better than legacy, algebraic formulations using a priori testing. To further assess the performance of ML, one of the ML models is employed in a a posteriori LES and compared against simulations with the algebraic model. The ML simulation is stable and yields encouraging improvements on key physical parameters regarding the flame length and the FFFD distribution. Novelty and Significance Statement: This research is of importance because it answers fundamental and practical questions related to the use of combustion modelling approaches, specifically the Flame Surface Density (FSD) and the Filtered Flame Front Displacement (FFFD) models, by means of Machine Learning (ML) algorithms. From a fundamental aspect, we show that two features which are typically not considered as inputs in combustion models, i.e., the progress variable and the resolved curvature, are key to consider for improved predictions of the model, more so than features which are typically used in FSD modelling, i.e., uΔ′,Δ, and |∇c ̄| . From a practical standpoint, we demonstrate a framework to use the developed ML combustion model a posteriori in a LES, without any stability issues. Overall, these findings are key to guide further traditional and ML improvement efforts on combustion models.

Machine Learning of Combustion LES Models from Reacting Direct Numerical Simulation

A Robust Autoregressive Long-Term Spatiotemporal Forecasting Framework for Surrogate-Based Turbulent Combustion Modeling Via Deep Learning

Modelling turbulent premixed flames using convolutional neural networks: application to sub-grid scale variance and filtered reaction rate

Assessment of Machine Learning Techniques for Simulating Reacting Flow: From Plasma-Assisted Ignition to Turbulent Flame Propagation

Turbulent Combustion Simulation by Large Eddy Simulation and Direct Numerical Simulation

Advancing Reacting Flow Simulations with Data-Driven Models

Hydrogen reaction rate modeling based on convolutional neural network for large eddy simulation

Application of machine learning for filtered density function closure in MILD combustion

Large-eddy Simulation of a Multi-Injection Flame in a Diesel Engine Environment Using an Unsteady Flamelet/progress Variable Approach

Prediction of Swirling Flow Field in Combustor Based on Deep Learning

High-resolution Reconstruction and A-Priori Modeling of Turbulent Flames in the Context of Large Eddy Simulation Using the Convolutional Neural Network

Augmenting filtered flame front displacement models for LES using machine learning with a posteriori simulations

Direct Numerical Simulations and Analysis of Three-Dimensional N-Heptane Spray Flames in a Model Swirl Combustor

Machine learning for predictive coal combustion CFD simulations—From detailed kinetics to HDMR Reduced-Order models

Machine Learning-Assisted Soot Temperature and Volume Fraction Fields Predictions in the Ethylene Laminar Diffusion Flames.

Towards a Generalised Artificial Neural Network for Sub-Grid Filtered Density Function Closure in Turbulent Combustion

A Deep Learning-Based ODE Solver for Chemical Kinetics.

DeepFlame: A deep learning empowered open-source platform for reacting flow simulations

Deep learning for presumed probability density function models

Generalization Capability of Convolutional Neural Networks for Progress Variable Variance and Reaction Rate Subgrid-Scale Modeling

A multi-scale sampling method for accurate and robust deep neural network to predict combustion chemical kinetics