Centre National de la Recherche Scientifique
Site d'essais en mer de Centrale Nantes
Statistical modeling of turbulence for strong (106) or very high (109) Reynolds numbers for the viscous hydrodynamics to the real and for the unsteady aerodynamics within the limit of the incompressibility hypothesis. Non-isotropic model EARSM with and without wall functions and hybrid RANS-LES. Modeling of cavitation by a liquid / vapor interface capture model.
In the context of hydrodynamics, it is a question of increasing the existing capacities of the digital basin to get rid of the effects of scales but always with a fine evaluation of the modeling compared to experimental data carefully Chosen and whose measurement uncertainty is available.
The scope of application aims, among other things, to continue the analysis and the objectification of performance for water sports such as rowing and canoeing, in collaboration with the CREPS de Nantes and the federations With the use of the coupling already implemented with the multi-body dynamics code MBDyn to model the complete systems. In the continuity of the previous works, it is the coupling of the resolution of the flow with deformable structures that feeds this topic. The coupling with structural models reduces dimensionally beam (1D) or membranes / shell (2D) is now acquired. The increasingly common use of complex structures with non-homogeneous materials such as composites in hydrodynamic applications makes coupling with dedicated 3D structure codes more and more prominent. These digital tools require revisiting the coupling algorithms so as not to lose effectiveness.
Although RANS modeling remains widely used, these approaches seem to be slowly supplanted by the LES approaches, but they can not yet approach the real situations of the hydrodynamics of viscous fluids. The LES approach remains prohibitive for high-reynolds simulations. Consequently the hybrid modeling technique RANS-LES, whose DES (Detached Eddy Simulation) approach and its variants, is an unavoidable alternative. We have already implemented it on bodies of low elongation but it remains to be evaluated and validated in the field of hydrodynamics. As far as the contribution of calculated turbulence (instationalities) is concerned, it is then essential to be able to develop high order methods of precision which are stable and which remain usable on geometries of great complexity. With the non-structured finite volume approach and implicit method, the main difficulty then lies in the discretization of the second derivatives and in the formation of a pressure operator on collocative meshing. The automatic mesh refinement implemented is a fundamental asset which now possesses a high degree of reliability and maturity and is essential to access the fine details of the flow. In the framework of hybrid modeling, it will be necessary to develop a criterion of automatic choice of the scales of local length of the turbulent model. Indeed, in LES hybrid modeling, the choice between RANS and LES as well as the filtration length in LES depends directly on the size of the cells.
The pointed limitations of the two-order modeling developed in the finite volume method and which will have to be lifted are mainly (1) numerical dissipation and dispersion which, too important, degrade the Monitoring and transport of flow structures, and (2) access to small scales requires fine meshes and time steps with high computing times. In the field of aeronautics high order methods as alternatives to methods of order two already exist. With regard to the calculation means and the expected CPU costs, in the use of computers in parallel mode, it will be necessary to exceed the limits of the conventional MPI approach used in the exchange of information between the processors and MPI / OPENMP hybrid parallelization by exploitation of co-processors.
This objective is related to the notion of targeted calculation, which consists in automatically making a simulation aimed at, for example, calculating a specific quantity related to the solution (such as the drag of a body). Every effort of calculation must aim at the control of the precision of this quantity. The simulation tool must then automatically adjust the parameters of the calculation, globally or locally (mesh, etc.) to produce the best approximation of this quantity and the error must be estimated in order to guarantee the accuracy of the result. The local contributions to the error can be used as a criterion of refinement in the method of automatic refinement that we have in order to construct an optimal mesh. A strategy for estimating the discretization error consists in forming and solving a linearized transport equation of the error whose source term corresponds to the differential residue of the problem considered but evaluated to a higher order.
It is a question of adding, and not of substituting, to the existing finite volume method the possibility of being able to represent the modeling of the interactions between an incompressible viscous fluid and deformable or dislocated structures. The typical example is the boat / ice interaction. The implementation should take advantage of the automatic mesh adaptation and the developments made on the generation of iso-surfaces as part of the modeling of the surface tension.
