Enhancing Flamelet Generated Manifold Model Through Artificial Neural Networks

Abstract

The Flamelet Generated Manifold (FGM) model is a widely used tabulated-chemistry approach for turbulent combustion modeling, providing an effective compromise between accuracy and computational cost. However, a major limitation arises when the manifold is extended to incorporate additional physical effects, such as the dilution level, since high-dimensional look-up tables significantly increase memory requirements. To address this challenge, and motivated by recent advances in machine learning methods, the present work introduces an Artificial Neural Network (ANN) as a surrogate for the conventional tabulated manifold within the FGM framework. To validate the model, the three-stream Sandia-D flame is investigated. In the first part of the analysis, the ANN-based formulation is compared with the standard FGM model, considering air as the only oxidizer. The results obtained provide the basis to extend the ANN model through the introduction of an auxiliary control variable, allowing the pilot to be considered as an additional oxidizer stream, thereby extending the applicability of the FGM model with a significant reduction in memory requirements.