A VOLUME FORCING METHOD BASED ON A RECONSTRUCTION-LIKE PROCEDURE FOR HYBRID RANS/LES

International audienceExplosion risks, especially those related to hydrogen releases, are encountered in a large number of situations. In nuclear facilities, these include for instance hydrogen leak from pipes or storages in ventilated rooms (Taveau, 2011) or, in nuclear reactor containment vessels, hydrogen production due to the oxidation of zirconium during a severe accident (Bentaib et al., 2015). In evaluating explosion hazards, the first stage consists in predicting the release and the turbulent mixing of flammable species. The cost and reliability of numerical simulations depend on the approach adopted for the turbulence modeling. One usually distinguishes two modeling approaches: the Reynolds-Averaged Navier-Stokes (RANS) statistical approach, that consists in modeling all turbulent scales, and the Large Eddy Simulation (LES) approach that consists in a direct calculation of the large scales and a modeling of the small ones. The RANS approach has the advantage of presenting reasonable computational costs but the reliability of the simulations may depend strongly on the level of sophistication of the turbulence model. On the other hand, LES approach is much more predictive but requires larger computing times that can be out of reach for large facilities and long transients. In this frame, hybrid RANS/LES methods are a very attractive alternative but pose the difficulty of generating turbulent fluctuations in transition zones. In this work, this difficulty is adressed by means of a volume forcing method. Numerical calculations using the open-source software CALIF3S-P2REMICS are carried out on a planar turbulent jet at Re = 10e4.Forcing methods to promote the development of fluctuations and to bridge continuously from a RANS to LES domain consist usually in adding a body force to the Navier-Stokes equations. Most of the time, target statistical quantities such as mean velocity and turbulent kinetic energy need to be defined as inputs for the forcing term. These statistical quantities could come from either a precursor simulation as in Pamier et al. (2009) or the RANS upstream domain as in de Laage de Meux et al. (2015). For the latter, the forcing is applied to a channel flow which has an homogeneous energy distribution in the streamwise direction. Therefore, the use of values fromthe RANS upstream domain as targets appears relevant. For non-homogeneous flows the turbulent kinetic energy can significantly vary in the streamwise direction. For this reason, we propose here another approach which consists of deriving the target quantities from a prescribed kinetic energy ratio between resolved and subgrid scales. In this study we suggest to use a reconstruction-like procedure as in Janin et al. (2021) to trigger turbulence in transition zones. The reconstruction procedure consists in adding a fluctuating part, e.g. a synthetic velocity, to the resolved velocity. This operation introduces additional terms into the filtered Navier-Stokes equations. Usually, only the unsteady term is retained (de Laage de Meux et al., 2015) requiring, at least formally, that turbulence is statistically homogeneous. However, the flow of interest involves shear flow turbulence and this calls to retain a convective term related to the reconstruction procedure. We follow here the motivations of Lundgren (2003) regarding the physical meaning of his linear forcing and we also retain here the term proportional to the resolved velocity gradient