# Laboratoire de Mécanique des Fluides et d’Acoustique - UMR 5509

LMFA - UMR 5509
Laboratoire de Mécanique des Fluides et d’Acoustique
Lyon
France

## Nos partenaires

Soutenance de thèse ECL

Jeudi 28 octobre 2021 à 16h00, amphi 203, bât. W1, ECL

### Numerical simulation and modeling of compressible turbulence in dense gas flows

Soutenance en mode mixte, présentiel possible et souhaité en amphi 203.

Lien pour la visioconférence
Meeting ID : 990 5716 9156
Passcode : 483185

Jury :
Guillaume Balarac, Prof. Grenoble-INP, LEGI – Rapporteur
Gianluca Iaccarino, Prof. Stanford, ICME – Rapporteur
Luminita Danaila, Prof. Univ. Caen, M2C – Examinateur
Alberto Guardone, Prof. Politecnico di Milano, Crealab – Examinateur
Christophe Corre, Prof. ECL, LMFA – Directeur de thèse
Alexis Giauque, MCF ECL, LMFA – Co-directeur de thèse

Summary :
The present work is devoted to the analysis and modeling of turbulence in flows of dense gases (DG). The interest for these gases mainly comes from the Organic Rankine Cycles (ORC) turbine industry. Indeed, their use enables a great adaptability for ORCs. The main advantage of DG is their capacity to exchange large amount of energy at low to moderate temperatures for the heat source.
DG are single-phase vapors characterized by long chains of atoms and medium to large molecular weights. In the vicinity of the critical point, DG exhibit an unusual behavior when compared with classical gases. Their use in ORCs raises modeling issues when numerically designing ORC turbines since the turbulent flows at stake include both significant compressibility effects and differences with respect to perfect gases (PG). However, up to now, turbulence closure models developed for PG have been applied for RANS simulations and Large Eddy Simulation (LES) of DG flows, for lack of a better option. The peculiar thermodynamic behavior of DG questions the relevance of this choice, which implicitly assumes that turbulent structures are not affected by DG effects.

The present thesis tackles the DG LES modeling issue by considering 3 main steps :
1) the detailed analysis of DG mixing layers direct numerical simulation (DNS) ;
2) an a priori assessment of LES subgrid-scale (SGS) terms using filtered DNS (DNS of homogeneous isotropic turbulence is also used) ;
3) the construction and a posteriori validation of a new LES SGS model using supervised machine learning algorithms.

DNS of mixing layers are computed for DG and PG flows for three values of the convective Mach numbers ($M_c =0.1-1.1-2.2$). Results show major differences for the momentum thickness growth rates at $M_c =2.2$, which is twice as large for DG when compared to PG.
The a priori evaluation of SGS terms is performed from filtered DNS. The SGS pressure term appearing in the filtered momentum equation needs to be modeled in addition to the SGS terms usually modeled in PG flows.
To answer the need for a specific SGS modeling, a modeling methodology using artificial neural networks (ANN) is presented. The method is then applied to the SGS pressure term showing a proper a priori prediction of the term. The a posteriori assessment is carried out for mixing layers at $M_c =1.1$ and $M_c =2.2$ with several filtering sizes.

Keywords : Dense Gas, Compressible Turbulence, Mixing Layer, Direct Numerical Simulation (DNS), a priori and a posteriori Large Eddy Simulation (LES), Machine Learning