A deconvolution-based approach for a posteriori validation of subgrid-scale models: Application to agglomerating nanoparticles
Year: 2026
Type: article
Abstract: Large eddy simulation (LES) of turbulent reacting flows relies on several models accounting for subgrid scale effects on flow dynamics, scalar transport, source terms, nanoparticle dynamics or spray evolution for instance. In the literature, many subgrid approaches have been developed in the last decades based on a priori analysis of direct numerical simulations (DNS). Generally, to be validated a posteriori , these models are directly used in LES of reference DNS or of experimental configurations. However, an actual LES combines several models enabling only the validation of the global LES strategy, but not of each individual model. To overcome this issue, the present work proposes a rigorous a posteriori validation strategy based on the transport of a filtered scalar equation by unfiltered flow variables, so that the sole filtered quantity is the scalar itself. By solving these equations on a DNS-like grid, and using deconvolution algorithms, the effect of a single subgrid model can be analyzed, while exactly evaluating the other terms. The proposed approach is illustrated investigating a simple case represented by agglomerating nanoparticles in a turbulent flow. The considered test case will allow to rigorously evaluate the validity of literature subgrid models for the coagulation term, classically used to perform LES of soot production and metal-oxide synthesis in turbulent flames. Novelty and significance statement: This work presents an a posteriori validation strategy for SGS models in LES of agglomerating nanoparticles. Building on DNS-to-LES validation practices, we develop the DNS2LES framework that solves filtered balance equations alongside the unfiltered flow to evaluate unresolved transport from deconvoluted fields. By applying an explicit filter on the DNS grid, the method isolates the model’s contribution and minimizes numerical/model error coupling. The approach is applied to agglomerating nanoparticles to quantify the errors of the intermittency subgrid model used for coagulation of flame-made nanoparticles such as soot and metal-oxides, proving that subgrid models are required especially for high Schmidt number scalars. Results suggest that LES based on this subgrid model should be considered with caution. The DNS2LES framework has the potential to be generalized to other multi-phase reactive flows.
Language: English
Subfield: Computational Mechanics
Field: Engineering
Domain: Physical Sciences
Open Access status: hybrid
Estimated APC paid: $3,570
OpenAlex ID: W7167095256