Evaluation of deconvolution modelling applied to numerical combustion

Abstract

International audienc

Scalar flux modeling in turbulent flames using iterative deconvolution

“…In the case of signals composed of many different wavenumber components, convergence will be complete for all wavenumbers at a sufficiently large number of iterations provided 0 < b < e c rmin . This simple result is consistent with previous studies where b=1 was used [38,39]. Clearly, this is a very convenient choice which does not require knowledge of the original wavenumbers contained within the original signal since b = 1 is the minimum value of the upper bound e cr which is not realised since k is non-zero.…”

“…Lagrange polynomials. Details of the DNS and LES meshes are given in Table II for cases A and B and in Table III Deconvolution using iterative algorithms was used to deconvolute scalar fields such as the species mass fractions and temperature in the case of laminar flames in [38], and for turbulent flames in [39] (on the DNS mesh) with overall good results.…”

From Discrete and Iterative Deconvolution Operators to Machine Learning for Premixed Turbulent Combustion Modeling

“…This was done here for ∆/δ L = 8 with a second order interpolation of the scalar signals inside computing cells (h = 0.5∆) [13]. Flame deconvolution with an appropriate sub-grid interpolation or a regularisation procedure thus appears as a robust tool [14,15]. Sometimes deconvolution is also associated to a scale similarity hypothesis [16].…”

Direct mapping from LES resolved scales to filtered-flame generated manifolds using convolutional neural networks

“…The DNS database is then used to train convolutional networks in order to directly reconstruct the unresolved scalar sources and transport terms in the framework of tabulated detailed chemistry (premixed flamelet) LES. The major advantage of such a direct reconstruction of unresolved sources and fluxes from mesh-resolved quantities in the LES, is that by doing so there is no need for explicit filtering or solving additional transport equations, both of which save computational time and mitigate possible resolution issues [18].…”