Modeling local extinction and differential diffusion in KAUST ammonia flames D and F using principal component transport

Partial cracking of ammonia (NH 3 ) into hydrogen (H 2 ) and nitrogen (N 2 ) enhances flame stability in carbon-free combustion by increasing fuel reactivity. However, modeling turbulent jet flames fueled by such multicomponent mixtures remains challenging due to complex turbulence-chemistry interactions (TCI), localized extinction, and the role of differential diffusion. This study develops a reduced-order large eddy simulation (LES) framework that couples principal component (PC) transport with deep neural networks (DNNs), incorporating differential diffusion and subgrid-scale (SGS) closures for efficient and accurate simulation of NH 3 /H 2 /N 2 -air jet flames. A mixture-averaged (MA) transport model is formulated using a Varimax-rotated PC basis, resulting in a diagonally dominant diffusion matrix that enables species-specific transport. SGS closures for PC variance and scalar fluxes are integrated to improve representation of unresolved turbulence. The framework is validated against experimental data for two piloted jet flames from the KAUST NH 3 flame series, referred to as flames D and F, with Reynolds numbers of 24,000 and 36,000, corresponding to 59% and 89% of the blowoff threshold, respectively. The proposed MA-Varimax-SG model captures key features including mixture fraction decay, temperature profiles, and the distributions of major and intermediate species (e.g., OH, NH 2 ), while accurately predicting localized flame holes and extinction-reignition dynamics. Compared to the PC-DNN model with unity Lewis number, the proposed approach improves extinction prediction in highly strained regions with less than 10% additional computational cost. These findings highlight the critical role of differential diffusion and SGS modeling in partially cracked ammonia flame simulations and demonstrate the scalability and robustness of the proposed framework for complex combustion systems. • NH 3 /H 2 /N 2 -air piloted flames at high Reynolds numbers are investigated numerically. • PC-transport MA-Varimax-SG model accurately predicts key flame characteristics. • Subgrid-scale closures improve prediction of local extinction and radicals. • Differential diffusion remains relevant at high Reynolds numbers.