Communication Dans Un Congrès Année : 2023

Development, Validation and Application of an ANN-based Large Eddy Simulation Subgrid-Scale Turbulence model for dense gas flows

Alexis Giauque
Aurélien Vadrot
  • Fonction : Auteur
  • PersonId : 1501898
Christophe Corre

Résumé

Using a DNS database (available to the NICFD research community) including homogeneous isotropic turbulence (HIT), mixing layers and channel flows, the authors have developed a novel LES model for the Subgrid-Scale (SGS) turbulent Reynolds tensor based on artificial neural networks (ANN) with optimized hyperparameters. Particular attention has been paid to ensure Galilean invariance properties are satisfied by the ANN-based model. To accommodate the large size of the database (over 500 million samples), the ANN is trained on a parallel CPU architecture. A priori performance reveals determination coefficients (r 2 -score) larger than 0.9 can be achieved, yielding an accurate prediction of the SGS turbulent Reynolds tensor magnitude for flow conditions never met during training. A posteriori validation is performed by implementing the model into AVBP, a LES solver developed at CERFACS [10]. The computation of a dense gas mixing layer at a convective Mach number of 2.2 shows that the newly developed ANN-based model provides levels of accuracy comparable with or even better than those provided by existing models (Implicit LES, Sigma model [13]). To further assess the ANN-based model, a 3D ORC turbine configuration is computed and the numerical prediction is compared with measurements gathered by Baumgartner, Otter and Wheeler [1] at Cambridge University and presented at NICFD 2020. Preliminary results from this recently launched European PRACE project are presented.

Fichier principal
Vignette du fichier
Giauque_NICFD.pdf (4) Télécharger le fichier
Origine Fichiers produits par l'(les) auteur(s)

Dates et versions

hal-04928384 , version 1 (04-02-2025)

Identifiants

Citer

Alexis Giauque, Aurélien Vadrot, Christophe Corre. Development, Validation and Application of an ANN-based Large Eddy Simulation Subgrid-Scale Turbulence model for dense gas flows. 4th International Seminar on Non-Ideal Compressible Fluid Dynamics for Propulsion and Power. NICFD 2022, Nov 2022, London, United Kingdom. pp.43-52, ⟨10.1007/978-3-031-30936-6_5⟩. ⟨hal-04928384⟩
0 Consultations
0 Téléchargements

Altmetric

Partager

More