Report

Avancées en Modélisation de la Turbulence SFEN ST 6 CFD pour conception & sureté des réacteurs, Chatou, le 29 Avril 2009 D. Laurence & F. Archambeau, MFEE What is new in Turbulent Modelling? …. Computers work more so modellers can sit back • Osborne Reynolds (1895): RANS “Re Averaged Navier Stokes” => exact expression for the production of stresses (&fluxes), • Joseph Valentin Boussinesq (1897) Eddy Viscosity • Ludwig Prandtl (1925): mixing length • Andrei Komogorov (1942): energy cascade production – dissipation scales separation • LES 1 4 3 Joseph Smagorinsky (1964): Lt and 0 1980-1990 Stanford CTR, Kim, Moin, Moser, Ferziger, Piomelli … k 3 2 • k-eps : Jones & Launder (1972), Launder & Sharma (1974), • Re Stress Transport: Launder, Reece, Rodi (1975), Speziale Sarkar Gatsky(1991) • 2000 – 2009 , lower cost HPC => LES for everyone and everything ? 2 LES preamble: Only Large Scales Matter (most of the time) Large Scales & Human activity Drag, mixing, heat transfer, chem. reactions … Large Scales dictate physics Generated by/scale with obstacle Impose dissipation rate Exceptions: noise, combustion, 2-phase flows …and NEAR WALLs ! 3 LES => great, colourful results … IF .. the mesh is such that large scales are well captured (which means ... solution is already known ?) 4 Under-Resolved Channel LES Coarse-mesh LES is more dangerous than coarse RANS ! Channel Flow LES on structured grid at Re*=395 (Re*=y+ at centre ) From N. Jarrin 2006 www.cfdtm.org % error on friction 5 Detached Eddy Simulation (DES) Scale Adaptive Simulation (SAS) - LES limitation: too strongly dependent on mesh, unreliable for true “predictions” - Hybrid RANS-LES methods developed in EU AEROSPACE project DESIDER 2004-2007. - Validation limited to aerodynamics, seems premature for Reactor thermal hydraulics ! “SAS-SST model formulation dos not involve the grid spacing, it avoids the undefined model regimes of DES. In case of overly coarse grids, the model reverts back to the underlying RANS formulation. Fig. 5.2: The SST-SAS model can represent the true nature of the flow, by allowing a break-down of the turbulent structures into a spectrum (down to the grid limit)” From “Best Practice Guidelines for the use of CFD in Nuclear Reactor Safety Applications” OECD NEA workgroup on CFD in reactor studies www.nea.fr/html/nsd/csni/cfd.htm 6 Flow ventilation of room with release of passive scalar DES error SAS error RANS error From Dr S. Gant. 'Challenges in validation of unsteady modeling approaches (DES/SAS/URANS)‘ presented at: http://cfd.mace.manchester.ac.uk/Main/KnooWorkshop To appear http://www.hse.gov.uk/ and J Flow, Turbulence and Combustion 7 Better RANS models to predict Length scales Kolmogorov lengthscale 3 1 f Taylor microscale Turbulent energy scale: Lturb 4 k 3 2u 2 u1 x1 2 2 First 3 are only “models” but can be obtained from “good” RANS models(*) “True” Taylor and Integral scale are only given by 2 point correlations, i.e. curvature and integral, e.g.: 1 L11 x, t R11 e1r , x, t dr R11 0, x, t 0 More work needed to extract “true” scales from DNS or expt. and test predictions from recent advanced RANS models (*) e.g. SST model in a simple channel flow will dramatically underestimate k near wall by a factor 2-3, and over-estimate lenghscale at centre by a factor 2! 8 Channel flow LES with “Taylor scale” mesh, N total = 0.44 Million , Re = 395 U U y+ y/h y+ 9 Unstructured meshing strategy for LES a) Possible FV near-wall refinements: a) dichotomy, b) non-conforming, c) & d) polyhedral & zoom Lenghtscale predicted by PRECURSOR RANS, then L= (k**3/2)/eps 10 Conclusions Can we design next gen. nuclear power plants using LES/DES ? Errors can be much larger than in RANS Beware vast majority of LES are “post-dictions”, or “explanations” how many failed LES were never published ? Mesh influence is tremendous LES quality criteria progressing (see Q-LES workshops B. Guerts et al.) but trial and error most expensive with LES Use “good” RANS model as precursor for LES feasibility study and mesh generation (but more user practice and control and over meshing software is needed) Progress in RANS-LES coupling (google ANR STURM4) RANS models are progressing quietly but surely (elliptic blending, algebraic structure based models… ), but no time to present… Beyond “cold / aerodynamic” flows “good” RANS models will still be major engineering tool for complex physics (2 phases flows, near wall effects, heat transfer, combustion, natural convection) which need more than just eddy viscosity predictions (time-scales, lenghtscales, anisotropy…) 11 Zonal modelling / RANS-LES coupling - LES community very active on this topic - First idea (DES) reduce viscosity when turbulent scale > cells - Bad transition region (kink) where RANS viscosity is reduced but resolved structures are still too weak Current solution is to OVERLAP RANS and LES 12 Zonal RANS-LES coupling, Uribe model Re Cells 395 40x30x30 590 40x40x30 1100 50x50x40 2000 50x50x40 4000 64x80x64 Uribe, J., Jarrin, N., Prosser, R., Laurence, D. 2009 "Two velocities hybrid RANS-LES of a trailing edge flow". J. Flow, Turbulence and Combustion. y+ @ y+ @ Dx+ Dz+ f=0.5 f=0.99 59 39 26 158 88 59 32 184 140 88 46 266 256 160 83 485 400 200 99 574 Integral length scales in channel flow Turbulent energy scale is easy RANS “model” but does not represent true (2 point correlation) integral scale for channel flow Lturb k 3 2 1- x : streamwise 2- y : wall normal 3 – z : spanwise streaks Addad, Y.,Gaitonde, U., Laurence, D., Rolfo, S. Optimal unstructured meshing for large eddy simulations, ERCOFTAC Series, Springer, Quality and Reliability of Large-Eddy Simulations. Meyers, Geurts & Sagaut, (Eds.), Vol. 12, pp. 93-103. Solid: stream-wise separation Dashed: span-wise separation Near wall RANS model developments – heated pipe 1.5 1.3 Nu/Nu 0 1.1 Launder & Sharma Model (CONVERT) Cotton & Ismael Model (CONVERT) Suga Non-Linear Eddy Viscosity Model (CONVERT) Lien-Chen-Leschziner k-eps Model (STAR-CD) k-omega-SST Model (STAR-CD) Lien & Durbin v2f Model (STAR-CD) k-omega-SST Model (Code_Saturne) Manchester v2f Model (Code_Saturne) Large Eddy Simulation (STAR-CD) DNS - You et al (2003) 0.9 0.7 0.5 buoyancy aiding 0.3 0.01 15 0.1 1 Bo 10