NUS Turbulence Workshop, Aug. ‘04 Large Eddy Simulation in Aid of RANS Modelling M A Leschziner Imperial College London RANS/LES simulation of flow around a highly-swept wing Collaborators Lionel Temmerman Anne Dejoan Sylvain Lardeau Chen Wang Ning Li Fabrizio Tessicini Yong-Jun Jang Ken-ichi Abe Kemo Hanjalic The Case for RANS RANS may be something of a ‘can of worms’, but is here to stay Decisive advantages: Economy, especially in statistical homogeneous 2d flows when turbulence is dominated by small, less energetic scales in the absence of periodic instabilities Good performance in thin shear and mildly-separated flows, especially near walls Predictive capabilities depend greatly on appropriateness of closure type and details relative to flow characteristics quality of boundary conditions user competence Challenges to RANS Dynamics of large-scale unsteadiness and associated non-locality Massive separation – large energetic vortices Unsteady separation from curved surfaces Reattachment (always highly unsteady) Unsteady instabilities and interaction with turbulence Strong non-equilibrium conditions Interaction between disparate flow regions post reattachment recovery wall-shear / free-shear layers Highly 3d straining – skewing, strong streamwise vorticity Separation from Curved Surfaces - Tall Order for RANS? LES instantaneous realisations AL model with k - w equation Reverse flow Y/H (Separation : X/H = 0.26, Reattachment : X/H = 4.7) 3 RANS 2 1 X/H 0 0 1 2 3 4 5 6 7 8 9 Dynamics of Separated Flow Steady Unsteady Separation Dynamics of Separated Flow Steady Reattachment Attached Recovery RANS Developments Desire to extent generality drives RANS research Non-linear eddy-viscosity models Explicit algebraic Reynolds-stress models Full second-moment closure Structure-tensor models multi-scale models… Simulation plays important role in aiding development and validation Traditionally, DNS for homogeneous and channel flow at low Re used Increasingly, LES exploited for complex flow The Argument for Resolving Anisotropy Generalised eddy-viscosity hypothesis: U i U j ui u j t xi xi 2 k ij ; U i {U ,V ,W }; xi {x, y, z} 3 Wrongly implies that eigenvalues of stress and strain tensors aligned Wrong even in thin-shear flow: Channel flow 2 u 2 v 2 w2 k 3 Which is wrong u2 v2 The Argument for Resolving Anisotropy Exact equations imply complex stress-strain linkage U j U i uiu j uiuk + u j uk Body -force production xk xk Pij Analogous linkage between scalar fluxes and production ui uiuk U i + ui Body -force production xk xk Pu i Can be used to demonstrate Origin of anisotropy in shear and normal straining Experimentally observed high sensitivity of turbulence to curvature, rotation, swirl, buoyancy and and body forces Low generation of turbulence in normal straining Inapplicability of Fourier-Fick law for scalar/heat transport Inertial damping of near-wall turbulence by wall blocking Reynolds-Stress-Transport Modelling Closure of exact stress-transport equations Duiu j Dt U j U i = uiuk + u j uk Redistribution xk xk Cij AdvectiveTransport Pij Production Diffusion Dissipation Modern closure aims at realisability, 2-component limit, coping with strong inhomogeneity and compressibility Additional equations for dissipation tensor ij At least 7 equations in 3D Numerically difficult in complex geometries and flow Can be costly Motivated algebraic simplifications Homogeneous Straining Axisymmetric expansion Homogeneous Straining Homogeneous shear and plain strain Near-Wall Shear Channel flow Explicit Algebraic Reynolds-Stress Modelling Arise from the explicit inversion of Duiu j Dt Cij AdvectiveTransport U j 2 Dk U i ij = uiuk + u j uk Redistribution 3 Dt xk xk Pij Production 0 Diffusion Dissipation Transport of anisotropy (and shear stress) ignored Redistribution model linear in stress tensor Lead to algebraic equations of the form uiu j f (Sij , ij .....k , ) Most recent variant: Wallin & Johansson (2000) Recent modification (Wallin & Johansson (2002/3)): approximation of anisotropy transport by reference to streamlineoriented frame of reference Non-linear EVM Constitutive equation Sij a 2 t s ui u j k ij 3 2 ij 1 (s 2 13 {s 2 }I ) 2 (ws sw ) 3 (w 2 13 {w 2 }I ) Quadratic 1{s 2 }s 2 {w 2 }s Quasi-cubic 3 (w 2 s sw 2 {w 2 }s 23 {wsw}I ) 4 (ws 2 s 2 w ) Cubic (=0 in 2d) Transport equation for turbulence energy and length-scale surrogate (ε, ω…) Coefficients determined by calibration Large Eddy Simulation – An alternative? Superior in wall-remote regions Resolution requirements rise only with Re 0.4 Near wall, resolution requirement rise with Re 2 Near-wall resolution can have strong effect on separation process Sensitivity to subgrid-scale modelling At high Re, increasing reliance on approximate near-wall treatments Wall functions Hybrid RANS-LES strategies DES Immersed boundary method Zonal schemes Spectral content of inlet conditions Achilles heal of LES Realism of LES – Channel Constriction Effects of Resolution – no-slip condition x=2h x=6h Re=21900 Distance of nodes closest to wall 5 y/H 4 Sensitivity of Reattachment to Separation Abe, Jang and Leschziner 3 2 1 0 0 10 x/H 20 Δxreat=7 Δxsep 0.4 0.05 30 Realism of LES – Channel Constriction Effects of near-wall treatment (WFs) on 0.6M mesh Realism of LES – Channel Constriction Sensitivity to SGS modelling Realism of LES – Stalled Aerofoil High-lift aerofoil – an illustration of the resolution problem Re=2.2M Experiments Realism of LES – Stalled Aerofoil High-lift aerofoil Realism of LES – Stalled Aerofoil Effect of the spanwise extent Realism of LES – Stalled Aerofoil 5 • Mesh 1: 320 x 64 x 32 = 6.6 • 10 cells Effect of the mesh 6 • Mesh 2: 768 x 128 x 64 = 6.3 • 10 cells 6 • Mesh 3: 640 x 96 x 64 = 3.9 • 10 cells 6 • Mesh 4: 1280 x 96 x 64 = 7.8 • 10 cells Streamwise velocity at x/c = 0.96 Prediction of the friction coefficient High-Lift Aerofoil - RSTM & NLEVM RSTM NLEVM The Case for LES for RANS Studies Experiments traditionally used for validation Very limited data resolution Boundary conditions often difficult to extract Errors – eg 3d contamination in ‘2d’ flow Reliance on wind-tunnel corrections Example: 3d hill flow (Simpson and Longe, 2003) Y Mid Coarse Grid Tauw vector Z X TAUW 0.006 0.0055 0.005 0.0045 0.004 0.0035 0.003 0.0025 0.002 0.0015 0.001 0.0005 0 New Experimental Information Flow visualisation vs. LDV 1.2 0.3 Uref x/H 0.18 separation in oilflow 1 -2 TKE contour, velocity vector and streamline + yL = 145 micron, y = 8 based on 2D Utau -1.8 TKE/Uref 2 -1.6 4.37E-02 4.12E-02 3.87E-02 3.62E-02 3.37E-02 3.12E-02 2.87E-02 2.62E-02 2.37E-02 2.12E-02 1.87E-02 1.62E-02 1.37E-02 1.12E-02 8.65E-03 -1.4 z/H -1.2 -1 -0.8 -0.6 0.8 Uref -0.4 -0.2 y/H 0 x/H 0.7 attachment in oilflow 0.6 0 0.5 1 1.5 2 x/H x/H 2.0 attachment in oilflow Large bump#3 0.4 Separation in CCLDV data 0.2 x/H 1.5 separation in oilflow 0 0 0.2 0.4 0.6 0.8 1 x/H 1.2 1.4 1.6 1.8 2 The Case for LES for RANS Studies Well-resolved LES a superior alternative Close control on periodicity and homogeneity Reliable assessment of accuracy SGS viscosity and stresses relative to resolved Spectra and correlations Ratio of Kolmogorov to grid scales Balance of budgets (eg zero pressure-strain in k-eq.) Reliable extraction of boundary conditions Second and possibly third moments available Budgets available Attention to resolution and detail essential LES for RANS Studies Considered are five LES studies contributing to RANS 2d separation from curved surfaces 3d separation from curved surfaces Wall-jet Separation control with periodic perturbations Bypass transition Study of Non-Linear EVMs for Separation Constitutive equation Sij a 2 t s ui u j k ij 3 2 ij 1 (s 2 13 {s 2 }I ) 2 (ws sw ) 3 (w 2 13 {w 2 }I ) Quadratic 1{s 2 }s 2 {w 2 }s Quasi-cubic 3 (w 2 s sw 2 {w 2 }s 23 {wsw}I ) 4 (ws 2 s 2 w ) Cubic (=0 in 2d) 2C-Limit Non-linear EVM Recent forms aim to adhere to wall-asymptotic behaviour Example: NLEVM/EASM of Abe, Jang & Leschziner (2002) Anisotropy cannot be represented by functions of Sij , ij alone Thus, addition of near-wall-anisotropy term, calibrated by reference to channel-flow DNS Involve “wall-direction indicators”, d i , Kolmogorov as well as macro time scales and viscous-damping function aij 1aij 2 aij w aij aij Cd f w ( Rt ) di d j ij d k d k f d Sij , d ij ,invariants 3 ld Ni l di Ni d ld - wall distance xi Nk Nk w 2C-Limit Non-Linear EVM Performance of AJL model in channel flow ( and variants) by reference to DNS 2C-Limit Non-linear EVM Performance of AJL model in channel flow ( and variants) Turbulence energy budget w A-priori Study of Non-Linear EVMs Quadratic terms represent anisotropy ‘cubic’ terms represent curvature effects Example: Streamwise normal stress across separated zone 2d periodic hill, Re=21500 Accurate simulation data used for model investigation Modelled stresses determined from constitutive equation with mean-flow solution inserted Comparison with simulated stresses Linear, quadratic and cubic contributions can be examined separately Jang et al, FTC, 2002 Highly-Resoved LES Data Two independent simulations of 5M mesh Highly-Resolved LES Data Turbulence-energy budget at x/h = 2.0 Near-wall velocity profiles at 3 streamwise locations (wall units) IJHFF (2003), JFM (2005) Highly-Resolved LES Data - Animations U-velocity Q-criterion W-velocity Pressure Streamlines 5 4 Abe, Jang and Leschziner y/H 3 2 1 0 LES solution (Temmerman & Leschziner, 2001) (Separation : X/H = 0.22, Reattachment : X/H = 4.72) Y/H 0 10 20 x/H 30 RSM (Jakirlic & Hanjalic, 1995) (Separation : X/H = 0.26, Reattachment : X/H = 5.9) Y/H 3 3 LES 2 2 1 1 X/H 0 0 1 2 3 4 5 6 7 8 9 0 k - (Launder & Sharma, 1974) (LS) (Separation : X/H = 0.35, Reattachment : X/H = 3.42) Y/H X/H 0 1 2 3 4 5 6 7 8 9 Cubic k - (Apsley & Leschziner, 1998) (AL) (Separation : X/H = 0.26, Reattachment : X/H = 5.3) Y/H 3 3 k- 2 2 1 1 X/H 0 0 1 2 3 4 5 6 7 8 9 0 Cubic k - (Craft, Launder & Suga, 1996) (CLS) (Separation : X/H = 0.26, Reattachment : X/H = 5.9) Y/H 3 2 2 1 1 X/H 0 1 2 3 4 5 6 7 1 8 9 2 3 4 5 6 7 8 9 Abe k - w (Separation : X/H = 0.31, Reattachment : X/H = 4.90) Y/H 3 0 X/H 0 Abe et al X/H 0 0 1 2 3 4 5 6 7 8 9 Velocity Profiles (x/h=2.0) (x/h=6.0) LS- AL- WJ- w CLS- AJL- w LES LS- AL- WJ- w CLS- AJL- w LES -0.2 0 0.2 0.4 0.6 U/Ub 0.8 1 -0.2 0 0.2 0.4 0.6 U/Ub 0.8 1 Shear-Stress Profiles 3 (x/h=2.0) (x/h=6.0) LS- AL- WJ- w CLS- AJL-w LES LS- AL- WJ- w CLS- AJL-w LES 2.5 2 1.5 1 0.5 0 -0.04 -0.02 uv/U2b 0 0.02 -0.04 -0.02 uv/U2b 0 0.02 A-priori Study – modelled vs. simulated stresses Linear EVM Symbols : a-priori analysis Lines : LS model (X/H=2.0) L* L L* L** L L* ñ 2/3k L ** * L * L * L * ñ L L * L * L * L * ñ uu(=2/3k+L) L ** L L ** L L uu-2/3k (actual L ** L L ** L L ** L L * L L ** L ** L L ** L L * L * L L ** L * L * L L ** L * L * L L ** 3 2.5 Y/H 2 1.5 1 0.5 0 0 0.02 0.04 0.06 0.08 0.1 2/3k, mean strain and normal stress Symbols : a-priori analysis Lines : Abe-w model 3 2.5 Y/H 2 1.5 1 0.5 0 LES) L L L (X/H=2.0) (X/H=2.0) LS model a-priori analysis uu-2/3k (actual LES) L L L uu L L L L L L L L L L L L L L L L 0 0.01 0.02 (L=)Linear term of uu/U2b L L L L L L L L L L L L L L L L L L L L L L L L L L L LS model a-priori analysis Actual LES L L L L L L L L L L L L L -0.04 -0.03 -0.02 -0.01 uv/U2b L L L 0 uv 0.01 0.02 Quadratic 2c limit EVM Abe,Jang &Leschziner, 2003 L * (X/H=2.0) L L* * L* L L ** ñ 2/3k L * * L * L * L * ñ L+Q L L ** L ** L ñ uu(=2/3k+L+Q) L ** L L ** L L uu-2/3k (actual LES) L ** L L ** L L ** L L ** L L ** L L ** L L L ** L * L * L L ** L * L L ** L L ** L * *L 0 L L L L L L L L L L L L L L L L L L L L L L L 0.02 0.04 0.06 0.08 0.1 2/3k, mean strain and normal stress (X/H=2.0) (X/H=2.0) Abe-w model a-priori analysis Abe-w model a-priori analysis uu (quadr) uu (linear) 0 0.01 0.02 (L=)Linear term of uu/U2b 0.03 0 0.01 0.02 (Q=)Quadratic term of uu/U2b 0.03 A-priori Study – modelled vs. simulated stresses Linear EVM Symbols : a-priori analysis Lines : LS model (X/H=2.0) L* L L* L** L L* ñ 2/3k L ** * L * L * L * ñ L L * L * L * L * ñ uu(=2/3k+L) L ** L L ** L L uu-2/3k (actual L ** L L ** L L ** L L * L L ** L ** L L ** L L * L * L L ** L * L * L L ** L * L * L L ** 3 2.5 Y/H 2 1.5 1 0.5 0 LES) 3 2.5 Y/H 2 1.5 1 0.5 0 L L 0 0.02 0.04 0.06 0.08 0.1 2/3k, mean strain and normal stress Symbols : a-priori analysis Lines : CLS model L* L L* L* L ** L * L * * L * L * L ** L L ** L L ** L L ** L L L ** L L ** L L * L ** L L ** L L ** L L L ** L * L L ** L L * L* L** L L * L * L * L ** *L 0 L L L L L L L L L L L L L L L L L L L L L L L L (X/H=2.0) (X/H=2.0) LS model a-priori analysis uu-2/3k (actual LES) L L L L L L uu L L L L L L L L L L L L L L L L L 0 0.01 0.02 (L=)Linear term of uu/U2b L L L L L L L L L L L L L L L L L L L L L L L LS model a-priori analysis Actual LES L L L L L L L L L L L uv L L L L -0.04 -0.03 -0.02 -0.01 uv/U2b L 0 0.01 0.02 Cubic EVM Craft, Launder & Suga, 2003 (X/H=2.0) ñ 2/3k ñ L+Q+C (X/H=2.0) (X/H=2.0) (X/H=2.0) CLS model a-priori analysis CLS model a-priori analysis CLS model a-priori analysis ñ uu(=2/3k+L+Q+C) uu-2/3k (actual LES) 0.02 0.04 0.06 0.08 0.1 2/3k, mean strain and normal stress uu (linear) 0 0.01 0.02 (L=)Linear term of uu/U2b uu (‘cubic’) uu (quadr) 0.03 0 0.01 0.02 (Q=)Quadratic term of uu/U2b 0.03 0 0.01 0.02 (C=)Cubic term of uu/U2b 0.03 A-priori Study – modelled vs. simulated stresses Linear EVM Symbols : a-priori analysis Lines : LS model (X/H=2.0) L* L L* * L L* L* ñ 2/3k L ** * L * L * L * ñ L L * L * L * L * ñ uu(=2/3k+L) L ** L L ** L L uu-2/3k (actual L ** L L * L ** L L ** L L L ** L ** L L ** L L * L * L * L * L L ** L L ** L * L * L L ** 3 2.5 Y/H 2 1.5 1 0.5 0 2.5 Y/H 2 1.5 1 0.5 0 L* L L* L* L* * L* L* L* L* L* L* L * L * L * L * L * L * L * L * L L ** L L ** L * L * L * L * L * L L ** L L ** L * L L * L ** L L * L * L ** L * L * L* L L 0 L LS model a-priori analysis uu-2/3k (actual LES) L L L L L L L L L L L L L L L L L L L L L L L L L L L L L L LS model a-priori analysis Actual LES L uv L L L L L L L L L L L L L L L L 0 0.01 0.02 (L=)Linear term of uu/U2b L L L L L L L L L L L L L -0.04 -0.03 -0.02 -0.01 uv/U2b L L L 0 0.01 0.02 Explicit ASM Wallin &Johansson, 2000 (X/H=2.0) * (X/H=2.0) (X/H=2.0) uu LES) 0 0.02 0.04 0.06 0.08 0.1 2/3k, mean strain and normal stress Symbols : a-priori analysis Lines : WJ-LS model 3 L L L L L L L L L L L L L L L L L L L L L L L ñ 2/3k ñ L+Q (X/H=2.0) (X/H=2.0) WJ-LS model a-priori analysis WJ-LS model a-priori analysis ñ uu(=2/3k+L+Q) uu-2/3k (actual LES) 0.02 0.04 0.06 0.08 0.1 2/3k, mean strain and normal stress uu (linear) 0 0.01 0.02 (L=)Linear term of uu/U2b uu (quadr.) 0.03 0 0.01 0.02 (Q=)Quadratic term of uu/U2b 0.03 3D-Hill - Motivation Efforts to predict flow around 3d hill with anisotropy-resolving closures Y Z X 3 -4 1 y/H 2 -2 0 0 0 x/H 2 4 6 4 8 2 z /H 6 LDA Experiments by Simpson et al (2002) Re=130,000, boundary-layer thickness = 0.5xh Computations with up to 170x135x140 (=3.3 M) nodes Several NLEVMs and RSTMs Topology – Experiment vs. NLEVM Computation Y Mid Coarse Grid Tauw vector Z X TAUW 0.006 0.0055 0.005 0.0045 0.004 0.0035 0.003 0.0025 0.002 0.0015 0.001 0.0005 0 Chen et al, IJHFF, 2004 Pressure and Skin Friction on Centreline 0.6 AJL- w AL- SSG- WJ- w Exp., z>0 Exp., z<0 0.05 0.4 0.2 0.04 0 U /U ref Cp -0.2 AJL- w AL- SSG- WJ- w Exp. -0.4 -0.6 0.03 -0.8 -1 -1.2 -2 0.02 -1 0 1 2 x/H 3 4 5 6 0 0.5 1 1.5 z/H 2 2.5 3 Corrected Experimental Information 1.2 0.3 Uref 1 y/H 0.8 0.6 0.4 0.2 0 0 0.2 0.4 0.6 0.8 1 x/H 1.2 1.4 1.6 1.8 2 3D Hill - LES Can origin of discrepancies be understood? Wall-resolved LES at Re=130,000 deemed too costly LES and RSTM computations undertaken at Re=13,000 Identical inlet conditions as at Re=130,000 Grid: 192 x 96 x 192 = 3.5M cells (y+=O(1)) LES scheme Second-order + ‘wiggle-detection’, fractional-step, AdamsBashforth Solves pressure equation with SLOR + MG Fully parallelised WF and LES/RANS hybrid near-wall approximations SGS models: Smag + damping, WALE Temmerman et al, ECCOMAS, 2004 Computational Aspects - LES Re=13000 128 x 64 x 128 cells CFL=0.2 d=0.003 CPU cost: 32 x 30 CPUh on Itanium2 cluster Statistics connected over 10 flow-through times, after initial 6 initial sweeps Near-wall grid: Overall View - LES Short-span integral of skin-friction lines LES / RANS Comparison – topology maps LES /RANS Comparisons – pressure & skin friction LES – subgrid-scale viscosity LES/RANS – cross-sectional flow field x/h = 1.5625 AL Model U: -0.10 -0.01 0.08 0.17 0.26 0.35 0.44 0.53 0.61 0.70 0.79 0.88 0.97 1.06 1.15 x/h = 1.5625 SSG+Chen Model U: -0.10 -0.01 0.08 0.17 0.26 0.35 0.44 0.53 0.61 0.70 0.79 0.88 0.97 1.06 1.15 LES/RANS – cross-sectional flow field x/h = 2.8125 AL Model U: -0.10 -0.01 0.08 0.17 0.26 0.35 0.44 0.53 0.61 0.70 0.79 0.88 0.97 1.06 1.15 x/h = 2.8125 SSG+Chen Model U: -0.10 -0.01 0.08 0.17 0.26 0.35 0.44 0.53 0.61 0.70 0.79 0.88 0.97 1.06 1.15 LES / RANS Comparison – streamwise velocity LES / RANS Comparison – turbulence energy Subgrid-scale energy LES, Re=130,000 – near-wall grid Grid: 192 x 96 x 192 cells LES / RANS Comparison, Re=130,000 – pressure coefficient LES /RANS Comparison, Re=130,000 – Topology maps RANS LES LES / RANS Comparison, 130,000 - velocity LES / RANS Comparison, 130,000 - velocity Wall Jet Motivation RANS models perform poorly in ‘interactive’ flows Example: post-reattachment recovery Key feature: interaction between upper (free) shear layer and developing boundary layer Wall jet is a similar ‘interactive’ flow LES solutions allow physics of interaction to be studied Budgets allow a-priori studies of closure assumptions Requirements for highly-resolved LES Simulations for real wall as well as shear-free wall allow viscous and blocking effects to be separated Dejoan & Leschziner, PoF, 2005 Wall Jet Geometry and flow conditions Grid: 420x208x96 Wall Jet Time evolution Grid: 420x208x96 Wall Jet Resolution indicators Wall Jet Q-structure criterion Wall Jet Log law & Equilibrium Wall Jet Shear-strain / stress dislocation Wall Jet Stress diffusion and budgets Wall Jet Comparison of stresses: AJL NLEVM / SSG RSTM / LES Wall Jet Comparison of budgets: SSG RSTM / LES xk k u2u2 Cd uk ul x l Wall Jet Comparison of budgets: SSG RSTM / LES xk k u1u2 Cd uk ul x l Wall Jet Comparison of stress diffusion: SSG RSTM / LES Transitional Wake-Blade Interaction - NLEVM Key issue: unsteady transition in high-pressure turbine blades Lardat & Leschziner, J AIAA, C&F, FTC, 2004 Transitional Wake-Blade Interaction - NLEVM Unsteady wake-induced separation on suction side Transitional Wake-Blade Interaction - NLEVM Time-space representation of shape factor (unsteady transition) Bypass Transition Experimental evidence: substantial ‘turbulence’ ahead of ‘transition’ T3A Concept of laminar turbulence Necessity to include a specific transport equation for so-called “laminar-kinetic energy” fluctuations (Mayle and Schulz, 1997) Dkl kl k Pk ,l 2 2 2 Dt y y In laminar region, shear production assumed = 0; rise in k attributed to k- diffusion by pressure fluctuation (pressure diffusion) Model proposed: Pk ,l Cw U 2 k k e y / C LES of Bypass Transition Finite-volume code, second order in time and space Localized Lagrangian-averaged dynamic eddy-viscosity model Re=500 (based on displacement thickness * at the inflow) Correlated perturbation field in the free-stream, with a specified energy spectrum Dimension of the domain : (Lx,Ly,Lz)=(200*,40*,35*), (nx,ny,nz)=(256,84,92) LES of transition over a flat plate Streamwise fluctuations in the near-wall region LES of transition over a flat plate Turbulence budgets in the transitional region New transition-specific model T3A T3B New transition-specific model T3A T3B New transition-specific model VKI blade test case, TUFS = 1% Displacement thickness Momentum thickness Shape factor New transition-specific model VKI blade test case, TUFS = 1% Turbulence energy profiles along the suction side of the blade Separation Control Context: flow control Reducing recirculation by external periodical forcing Exploiting sensitivity of mean flow and turbulence to forcing frequency Specific motivation of study Ability of URANS to emulate response observed experimentally and in simulations Challenge: coupling between turbulence time scale and perturbation time scale Subject of study External periodical forcing by mass-less jet applied to a separated flow over a backward-facing step Assessment of URANS modelling by reference to well-resolved LES data and experiment Dejoan & Leschziner, IJHFF, 2004, ASME FED, 2004 Separation Control - LES Expts. by S.Yoshioka, S. Obi andS. Masuda (2001)) Inlet channel: -3h <x<0, 0<y<2 Downstream the step: 0<x<12h; -h<y<2h Spanwise direction homogeneous, Lz=4p/3 Reynolds number: Re=Uc h / =3700 Strouhal number: St=fe h / Uc = 0.2 Optimum frequency Separation Control - Effect on Recirculation Length St Exp. LES LS AJL- 0.0 x r,o=5.5 h x r,o=7 h x r,o=6.5 h x r,o=8 h 0.2 x r=3.8 h x r=5.5 h x r=4.6 h x r=5.5 h Skin Friction Coefficient Separation Control – Effect on Shear Stress LES LS model Separation Control – Effect on Shear Stress LES AJL- model Separation Control – Phase-Averaged Features Streamlines Reynolds Shear Stress Separation Control – Phase-Averaged Features Streamline Reynolds Shear Stress Separation Control – Phase-Averaged Features Streamline Reynolds Shear Stress Concluding Remarks RANS will remain principal approach for many years to come Recognised by industry – hence increased interest in model generality Research pursued on two-point closure, structure modelling, multilength-scale modelling…. but practical prospects are uncertain Progress is incremental, slow and costly Serious model improvements must encompass a broad range of conditions – homogeneous 1D flows to complex 3D flows There is a need to extend further efforts to complex 3D conditions – the real challenge Concluding Remarks LES (& hybrid LES/RANS) of increasing interest – periodicity, shedding, large-scale motion LES is no panacea and faces significant obstacles in near-wall flows Poses serious problems in high-Re conditions – wall resolution, grid, cost… LES can play a very useful role in support of RANS modelling elucidating physics providing wealth of data for validation and a-priori study of closure proposals, especially budget Necessarily a very costly approach, because of resolution demands can only be done at relatively low Re Hybrid RANS / LES strategies hold some promise, but difficult very active area of research NUS Turbulence Workshop, Aug. ‘04 Near-Wall Modelling in LES M.A. Leschziner Imperial College London Hybrid RANS-LES Wall resolved LES is untenable in high-Re near-wall flow Near-wall treatment is key to utility of LES in practice Several approaches: Wall functions Zonal methods – thin-shear-flow equations near wall Hybrid RANS-LES (+ synthetic turbulence) All pose difficult fundamental and practical questions: Compatibility of averaging with filtering Applicability of RANS closure – time-scale separation Interface conditions LES / Wall-Functions Channel flow, Re=12000, 96x64x64 grid LES / Wall-Functions 2d hill flow, Re=2.2x104, 0.6M nodes LES / Wall-Functions Hydrofoil trailing edge, Re=2x106, 384x64x24 grid Hybrid RANS-LES Methodology Interface conditions Superimposed RANS layer Target Velocity U RANS int Turbulent viscosity Turbulence energy U LES int LES tRANS ,int t ,int k RANS mod,int k LES mod,int Hybrid RANS-LES Implementation RANS mod RANS mod C l k LES mod C ,int 0.5 or LES mod,int 0.5 l k RANS ,int RANS mod C k / < . > : spatial average in the homogeneous directions. Alternative: instantaneous value C ( y ) 0.09 C ,int 2 1 exp y 0.09 1 exp yint int Hybrid RANS-LES Typical variation of mean C model at interface in channel flow , 1-eq. RANS across RANS layer Hybrid RANS-LES Variations of mean and instantaneous C in channel flow, 1-eq. RANS model Hybrid RANS-LES Channel flow, Re=42200 64 64 32 512 128 128 yint 135 - j 17 Hybrid RANS-LES Channel flow, Re=42200 Resolved Modelled DES Hybrid RANS-LES Channel flow, Re=42200, velocity and shear stress distributions for two interface positions Hybrid RANS-LES Variations of mean and instantaneous C in channel flow, 2-eq. RANS model, Re=2000 Hybrid RANS-LES Velocity in channel flow, 2-eq. RANS model, Re=2000, average and instantaneous input of C Hybrid RANS-LES Structure (streamwise vorticity) in channel flow, 2-eq. RANS model, Re=2000 Interface y+=120 Interface y+=610 Hybrid RANS-LES 2d-hill flow, Re=21500, interface conditions Grid: 112x64x56=4x105 against reference of 4.6x106 Hybrid RANS-LES 2d-hill flow, Re=21500, variations of C Hybrid RANS-LES 2d-hill flow, Re=21500, variations of velocity and shear stress Hybrid RANS-LES 2d-hill flow, Re=21500, variations of velocity and shear stress Hybrid RANS-LES 2d-hill flow, Re=21500, variations of velocity against log-law Hybrid RANS-LES 2d-hill flow, Re=21500, variations of turbulent viscosity Two-Layer Model Methodology Low-Re solution In sublayer Two-Layer Model Methodology Near-wall control volume divided into subgrid volumes Transport equations solve across the subgrid for: Mean-flow parameters: U, W Wall-normal V-velocity from continuity within subgrid Two-Layer Model Methodology Wall-parallel pressure gradient (dP/dx) calculated from main-grid and assumed constant across subgrid wall calculated from subgridPk , solution wall applied to main-grid as in standard wall-function treatments U U U dP U V t x y dx y y Two-Layer Model Numerical solution in sublayer Similar to 1-D convection-diffusion problem Finite-volume method Central differences for diffusion and for convection Tri-diagonal matrix algorithm Average solution in time No need to solve Poisson equation Very fast! Desider: 6 month meeting Two-Layer Model Trailing-edge separation from hydrofoil; Re=2.2x106 512x128x24 nodes Comparison with highly-resolved LES by Wang, 1536x96x48 nodes Sub-layer thickness y 40 Streamwise velocity pressure Two-Layer Model Streamwise-velocity contours Wall model Two-Layer Model Velocity magnitude B C D X/h |U|/U_e Full LES Wall model (dynamic SGS) E F G Two-Layer Model Turbulence energy Full LES Wall model (dynamic SGS) Two-Layer Model Streamwise velocity in wake Full LES Wall model (dynamic SGS) Two-Layer Model Skin friction Full LES Wall model (dynamic SGS) Concluding Remarks The jury is out on the prospect of approximate wall modelling as a general approach There is evidence that some offer ‘credible’ solutions and gains in economy There is a price to pay (sometimes high) in terms of physical realism (e.g. near-wall structure) Particular problem: loss of small-scale near-wall components It is not clear what to do in very complex near-wall flow – separation, severe 3d straining Particular problems when near-wall flow has a strong effect on global flow features Hybrid RANS-LES and zonal modelling work, but much more research is required to identify applicability and limitations