Modeling differential diffusion and local extinction in partially cracked ammonia flames using principal component transport

Partial cracking of ammonia (NH3) into hydrogen (H2) and nitrogen (N2) enhances flame stability by increasing the mixture reactivity. Modeling turbulent jet flames with such mixtures remains challenging due to complex turbulence-chemistry interactions, local extinction, and differential diffusion. This study presents a reduced-order large eddy simulation (LES) framework coupling principal component (PC) transport with deep neural networks (DNNs), while incorporating differential diffusion and subgrid-scale (SGS) closures. A mixture-averaged (MA) transport model using a varimax-rotated PC basis enables species-specific transport. The model is validated against the KAUST ammonia flames D and F at high Reynolds numbers. The proposed approch satisfactorily captures mixture fraction decay, temperature and species (e.g., OH, NH2) distributions, as well as extinction-reignition dynamics. Compared to a PC-DNN model with a unity Lewis number transport model, the proposed approched improves significantly the prediction of extinction with minimal added computational cost.