Recent advances in combustion modelling for Large Eddy Simulation (LES) have increasingly utilised lower-dimensional manifolds, such as Flamelet Generated Manifolds and Flamelet/Progress Variable methods, due to their computational efficiency. These methods typically rely on one-dimensional representations of flame structures, often assuming premixed or non-premixed configurations. However, practical combustion devices frequently operate under partially-premixed conditions and present challenges due to mixture inhomogeneities and complex flow features. The Linear Eddy Model (LEM) offers an alternative by directly simulating turbulence-chemistry interactions without presuming specific flame structures. However, traditional LES-LEM approaches are computationally quite expensive due to the need for resolved LEM domains to be embedded in every LES cell.The authors developed the Super-Grid LEM (SG-LEM) method (Comb. Theor. Model. 28, 2024) to address these computational challenges by coarse-graining the LES mesh and embedding individual LEM domains within clusters of LES cells. This study evaluates SG-LEM in the context of the Multi-Regime Burner (MRB) introduced by Butz et al. (Combust. Flame, 210, 2019), which features both premixed and non-premixed flame characteristics. SG-LEM simulations of the MRB case demonstrate the method’s sensitivity to clustering parameters, with flow-aligned clusters significantly improving flame stability. LEM domains on the super-grid were able to represent the MRB flame topology while LES radial profiles including velocity, mixture fraction, temperature, and $${\textrm{CO}}$$
CO
mass fraction, were validated against experimental data and also reference simulations using standard combustion closures. The work also investigates discrepancies in CO profiles using conditional statistics and stand-alone LEM simulations. Finally, the work identifies areas of improvement for the SG-LEM framework, in particular relating to cluster generation, and (advective and diffusive) mass exchange between neighbouring LEM domains, as well as possible solutions for future SG-LEM implementations which could improve the model’s predictive capability.
