A-PRIORI ANALYSIS OF THE FLAME STRUCTURE IN THERMODIFFUSIVELY UNSTABLE NH3/H2/AIR MIXTURES

In this investigation, we adopt a data-driven approach utilizing Direct Numerical Simulation (DNS) data from thermodiffusively unstable NH3/H2-Air mixtures to define progress variables, thereby addressing flame dimensionality concerns. Irreducible error analysis highlights the necessity of incorporating at least three progress variables for a comprehensive depiction of hydrogen-enriched ammonia flames. While two-dimensional models suffice for accurately representing temperature, three-dimensional models are essential for effectively reproducing NO dynamics. Additionally, our investigation of Artificial Neural Network (ANN) models reveals their ability to faithfully replicate DNS data when trained on a subset of that data. However, their performance markedly declines when exclusively trained on a dataset comprising unstretched 1D freely propagating flames. This suggests the critical importance of expanding datasets, particularly by incorporating stretched flames, to bolster the ANN model's capacity.