Numerical characterization of ammonia–hydrogen combustion in internal combustion engines using ECFM under RANS and LES frameworks

The transition towards low-carbon transport has accelerated the exploration of carbon-free alternative fuels for internal combustion engine applications. Among these, ammonia enriched with hydrogen represents a promising candidate for application in internal combustion engines. In this context, computational fluid dynamics (CFD) plays a pivotal role in supporting engine design; consequently, the predictive capability of the combustion model is of primary importance, particularly within a fuel-flexible framework. In the present work, a numerical campaign is conducted and validated against an extensive experimental dataset obtained from a commercial light-duty engine operated in a single-cylinder configuration and fuelled with 𝑁𝐻3∕𝐻2 mixtures. The dataset spans a wide range of equivalence ratios and intake pressures. The adoption of a fully premixed charge, together with precise control of engine speed, enables the decoupling of turbulence effects from those associated with mixture properties, thereby allowing a more rigorous assessment of the combustion process. The well-established ECFM combustion model is systematically investigated using both Reynolds-Averaged Navier–Stokes (RANS) and Large Eddy Simulation (LES) turbulence modelling approaches, while maintaining a consistent formulation. Results indicate that model calibration is required to accurately reproduce experimental observations when varying hydrogen content and equivalence ratio. Notably, similar calibration trends are observed for both RANS and LES simulations, suggesting that the required adjustments are primarily linked to the turbulence–flame interaction, as modelled through the flame surface density (Σ) transport equation, rather than to the specific turbulence closure. These findings highlight the need for an improved formulation of the flame surface density model, capable of adapting to variations in mixture composition and accurately capturing both strain- and curvature-induced flame stretch. Furthermore, the incorporation of a dedicated sub-model to account for flame instabilities, particularly under lean conditions, is shown to be essential for enhancing predictive accuracy.