Spalart SanFran 06

33 %
67 %
Information about Spalart SanFran 06
Entertainment

Published on October 30, 2007

Author: Melinda

Source: authorstream.com

Slide1:  Turbulence: Are We Getting Smarter? P. Spalart* Boeing Commercial Airplanes *Training by NASA *Many CFD runs by and discussions with M. Strelets, M. Shur, K. Squires Slide2:  RANS models appear frozen at their 1992 level S-A and SST “rule” in Aerospace, with minor tweaks We lost the Karman constant! Experiments challenge its accepted values… and each other’s Direct Numerical Simulation fails to find the log law Models are rebellious around laminar regions S-A can refuse to “wash away” eddy viscosity SST can refuse to transition Setting inflow values of turbulence quantities is delicate Turbulence theories come and go Examples: Power laws; RNG; Lie groups LES is coming strong; but is it smart? If we’re not smarter, in what sense? RANS models frozen at their 1992 level:  RANS models frozen at their 1992 level This is meant in “practical” CFD Research Community spent much effort on: Reynolds-Stress Transport models Algebraic Reynolds-Stress models Realizable, two-component limit, elliptic effect, etc. Some propagated in vendor CFD codes, but none “won workshops” Separation prediction is paramount The elaborate models still do not win at that game They win in some vortex flows, e.g. the spheroid, but SARC is not bad Extra CPU power is going into finer grids and complexity Recent drag workshop: (courtesy E. Tinoco) We are far from grid overkill, even for simple wing-body case Grid differences exceed model differences The “worst flaw” in each model is still not clear (unlike for k-e) Did Menter and Spalart stop working? F6 WB w/wo FX2 – Total Drag Convergence:  F6 WB w/wo FX2 – Total Drag Convergence Tinoco Charts at the Drag-Prediction Workshop (FX2 is a fairing at the wing root) SST and SA have very different trends for drag vs. grid count F6 WB w/wo FX2 – Skin Friction Drag Convergence:  F6 WB w/wo FX2 – Skin Friction Drag Convergence F6 WB Separation Bubble on Wing – Turbulence Modeling:  F6 WB Separation Bubble on Wing – Turbulence Modeling Edge of Separation Bubble on Wing Wind Tunnel Oil Flow Photo, Re=3M Overlay of Computed Streamlines, SST Turbulence Model, Re=5M It would be fun to test the nonlinear constitutive relation, for the corner flow (or a Reynolds-Stress model) Did Menter and Spalart stop working?:  Did Menter and Spalart stop working? No. Turbulence research is like nicotine Both worked on: Curvature and rotation corrections RANS-LES hybrids: DES and SAS There is also work on: Wall roughness Compressibility Nonlinear constitutive relation Over-tripping More general wall functions R. Langtry with Menter created a transition-prediction method by transport equations that looks like dynamite Slide8:  RANS models appear frozen at their 1992 level S-A and SST “rule” in Aerospace, with minor tweaks If we are not smarter, in what sense? We lost the Karman constant! Experiments challenge its accepted values Direct Numerical Simulation fails to find the log law We Lost the Karman Constant in Experiments:  We Lost the Karman Constant in Experiments In the Good Old Days… The accepted values were 0.40 to 0.41 Probably dominated by the Stanford Olympics Although k-e gave 0.433 all along! Direct Numerical Simulation (DNS) seemed to agree (more on this) The Superpipe experiments gave 0.436 And later 0.42 (smaller Pitot tube, rethink of corrections, etc) The NDF and KTH boundary-layer experiments gave 0.385 or 0.38 The two teams are talking Still, they don’t use the same instrumentation The variation of Cf with Reynolds number is more conclusive than U+ profiles If k is different in these two flows, I’ll quit! We Lost the Karman Constant in Experiments:  We Lost the Karman Constant in Experiments k controls the trend of Cf versus Re Experiments, even ancient, disagree with [0.40,0.41] when Cf is examined over a wide range Inject 0.436 into a model, and Cf goes the wrong way in boundary layer The rule used by Boeing since 1961 agrees with the latest data and 0.38! The 2% difference in Cf means about 1% airplane drag S-A model: matches the NDF-KTH log law with k = 0.38 and C = 4.1, and also the CTTM point, by adjusting to k = 0.38, cv1 = 6.08, cw2 = 0.58 -- A better fv1 is in the works; practical impact will be slight Wings, bodies 10% (Collaborative Testing of Turbulence Models, Bradshaw-Launder-Lumley) Effect of k = 0.38 on RAE2822 Flow is Minimal:  Effect of k = 0.38 on RAE2822 Flow is Minimal Note, Cf is higher with 0.38 because the Reynolds number is not very high: about 4 106 to the shock Case 10, M=0.75, arecom=2.57 Guidance for a new fv1 in S-A model:  Guidance for a new fv1 in S-A model fv1 is reverse-engineered from velocity profiles Agreement between DNS and experiment is compelling up to y+ ~ 200 Very near the wall, trust DNS For high Reynolds number, trust experiment Including k = 0.38 Skin-friction was strongest argument in favor It’s still a “big jump!” NDF-KTH B-layer Hoyas-Jimenez channel We Lost the Karman Constant in U+:  We Lost the Karman Constant in U+ Define a local k: as d ( log y+ ) / d U+ it is constant in a log layer in theory, this plateau stretches with Reynolds number Experiments show this, but have noise in k NDF and KTH Experiments also have an overshoot near y+ = 80 Which is nothing to worry about It can be erased by inserting ( y+ + 10 ) instead of y+ , which is entirely justified, as the virtual origin does not have to be y = 0 DNS Hoyas & Jimenez 2006 reached Ret = 2000 in channel, but… overshoot to ~ 0.43, and have no plateau at all! Spalart 1988 thought he had just reached plateau (@ 0.406) in a BL Johnstone & Coleman 2006 also have overshoot in the Ekman layer RANS models S-A behaves as expected (while settling a bit too late near y+ = 80) SST does not, at relevant Reynolds numbers “Local” Karman Constant:  “Local” Karman Constant Expected qualitative behavior Channel flow Increasing Re Velocity Profile Overshoots the Log Law:  Velocity Profile Overshoots the Log Law Courtesy Prof. Nagib and Mr. Kapil We Lost the Karman Constant in DNS:  We Lost the Karman Constant in DNS Expected qualitative behavior High-Reynolds-number DNS Increasing Re Disaster struck We Lost the Karman Constant in DNS:  We Lost the Karman Constant in DNS If not the Law of the Wall! We seem secure only to y+ ~ 200 Rq = 1410, 1988 BL-channel and DNS-expt agreement We Lost the Karman Constant in SST (k-w):  We Lost the Karman Constant in SST (k-w) S-A model closer to DNS and experiment than SST (k-w) (in which k =0.406) Not having “an fv1” does not pay off. This presentation magnifies differences Slide19:  RANS models appear frozen at their 1992 level S-A and SST “rule” in Aerospace, with minor tweaks We lost the Karman constant! Experiments challenge its accepted values Direct Numerical Simulation fails to find the log law If we are not smarter, in what sense? Models are rebellious around laminar regions S-A can refuse to “wash away” eddy viscosity; “Transition Creep” SST can refuse to transition Setting inflow values of turbulence quantities is delicate Rebellious Models: S-A “Transition Creep”:  S-A is supposed to go turbulent from either: A turbulent inflow value, n = 3 n or more, FT mode (more soon) A trip Separation; “Trip-Less mode” I am not fond of “production blanking” A forward creep is seen: In attached boundary layers When eddy viscosity is supposed to “wash away” because: run was started in FT mode, and then switched to laminar inflow trip creates eddy viscosity where desired Causes, remedies, actions Eddy viscosity dynamic range is huge No conclusive remedy known yet, other than very fine x grid spacing, and details of discretization Users need to plot S-A “Turbulence Index” (0 laminar, 1 turbulent) Rebellious Models: S-A “Transition Creep” Rebellious Models: S-A “Transition Creep”:  Rebellious Models: S-A “Transition Creep” This is a tripped case Eddy viscosity rises by more than a factor of 10 within one cell, twice Any averaging between values < 1 and values O(10) defeats the ft2 term, and production starts. The cb2 term is also a threat. Rebellious Models: Refusing to Transition :  Rebellious Models: Refusing to Transition First, S-A Rumsey reports trouble at this meeting (AIAA-2006-3906) Both “balking” and non-uniqueness It is with n / n = 1.341946… a “NASA value” derived from nt / n = 0.009… itself derived from RT = 0.1 in Baldwin-Barth… The trouble goes away with inflow n = 3 to 5 n (as recommended) FT behavior is easily obtained Rumsey also advocates setting ct3 to 0, which is OK if FT Two-equation models Again, Rumsey has issues with “balking” He wanted FT behavior Again, baseline has inflow nt / n = 0.009 Inflow nt / n = 1.294 helps a little k, e and nt decay considerably during the approach (more soon) Finer grids make the problem worse Rebellious Models: Setting Inflow Values:  Rebellious Models: Setting Inflow Values First issue: desirable ambient values (near the wing) Laminar regions nt needs to be well below n (but Official SST has no trip term) No laminar regions, FT nt may be well above n! Only the Reynolds number based on flap gap or leading-edge radius and nt needs to be large The eddy viscosity must be smaller outside the boundary layer than inside, so that its value does not control the BL This is not satisfied by the inflow values recommended by some CFD vendors This implies k cannot have a “realistic” ambient value, such as 10-4 U2, giving “1% FST” (time scale has grown too much) Second issue: reverse-engineer inflow values With S-A, there is no distinction between inflow and ambient values Rebellious Models: Setting Ambient Values:  Rebellious Models: Setting Ambient Values Courtesy S. Allmaras McDonnell-Douglas three-element airfoil Boeing GGNS adaptive code, S-A model Chord Reynolds number 9 x106 FT mode: requesting transition in all B-layers Inflow n /n = 5, 500, and 5000! Gap Reynolds number: ~ 40,000, 400, and 40 Only highest value has high impact Setting Ambient Values. Eddy Viscosity:  Setting Ambient Values. Eddy Viscosity nt / n = 5 nt / n = 5000 Setting Ambient Values. Velocity:  Setting Ambient Values. Velocity nt / n = 5 nt / n = 5000 Setting Ambient Values. Eddy Viscosity:  Setting Ambient Values. Eddy Viscosity nt / n = 5 nt / n = 5000 Setting Ambient Values: Lift:  Setting Ambient Values: Lift Lift is seriously affected only once ambient nt / n is 5000 Re based on chord and ambient nt is then ~ 2000 Rebellious Models: Setting Inflow Values:  Reverse-engineer inflow values with two equations Use the desired ambient values, and the decay laws: linear growth rapid decay slower decay where x is the distance since the inflow These place limits on the achievable ambient values, most notably for the time scale k / e This is work in progress Hoping to pursue with Rumsey Strelets and Menter groups do not have severe problems CFD code manuals often not steering users too well Rebellious Models: Setting Inflow Values Slide30:  RANS models appear frozen at their 1992 level S-A and SST “rule” in Aerospace, with minor tweaks We lost the Karman constant! Experiments challenge its accepted values Direct Numerical Simulation fails to find the log law Models are rebellious around laminar regions S-A can refuse to “wash away” eddy viscosity SST can refuse to transition Setting inflow values of turbulence quantities is delicate If we are not smarter, in what sense? Turbulence theories come and go Examples: Power laws; High-Order log laws; RNG; Lie groups Turbulence Theories Come and Go:  Turbulence Theories Come and Go Power laws for the velocity profile Nostalgia value; the U ~ y1/7 law of the 1930’s Physical view of outer layer is not Galilean-invariant Incompatible with classical thinking (inner and outer layer) Incompatible with all prevailing transport-equation turbulence models Rapidly defeated by roughness, sliding walls, etc. Not supported by data Experimental, or DNS as a layer that expands with increasing Reynolds number Higher-Order concepts than the wall-and-defect law No rigorous basis, or prospects outside the simplest of cases We are not producing Matched Asymptotic Expansions of a known equation. We only have a large parameter: the Reynolds number Re-Normalization Group; Lie Groups Ostensibly, pure theory Very hard to follow, for me Not supported by expanding layers either; e.g., the exponential profile Does RNG k-e model amount to more than dimensional analysis? Slide32:  RANS models appear frozen at their 1992 level S-A and SST “rule” in Aerospace, with minor tweaks We lost the Karman constant! Experiments challenge its accepted values Direct Numerical Simulation fails to find the log law Models are rebellious around laminar regions S-A can refuse to “wash away” eddy viscosity SST can refuse to transition Setting inflow values of turbulence quantities is delicate Turbulence theories come and go Examples: Power laws; Lie groups If we are not smarter, in what sense? LES is coming strong; but is it smart? And what about DNS? LES is Coming Strong, but is it Smart?:  LES is Coming Strong, but is it Smart? LES is coming to real-life CFD… Largely as a component of hybrid RANS-LES approaches, such as DES Pure LES of a wing remains scheduled for 2045 Only LES can compute large-scale unsteadiness well It’s excellent for jet noise, bluff bodies, and cavity flows 3D URANS is: not powerless not very predictable not responsive to grid refinement. “VLES” is not a sound concept; LES should have been called “Large-and-Medium-Eddy Simulation” LES conflicts with shock-capturing… Not in the right hands! LES is not intellectually very “elegant,” but it is right At least, away from walls Wall modeling, by DES or other, is crucial at useful Reynolds numbers Atmospheric turbulence is just as important as Aerospace turbulence, and they are also “resorting” to LES LES is Coming; what about Shocks?:  LES is Coming; what about Shocks? LES requires low enough numerical dissipation This has slight upwind bias and van Albada limiters The shocks are rather weak Shock capturing requires some dissipation In the NTS code, run by Shur, it works! DNS is Stronger now, but Smarter?:  DNS is Stronger now, but Smarter? Hoyas & Jimenez, 2006 Phys Fluids “Some of the fluctuation intensities… do not scale well in wall units.” Spalart, 1987 JFM “The scaling of turbulent quantities is compared with accepted laws, and the significant deviations are documented.” Not referenced by H&J u’+ w’+ v’+ Back to Work?:  Back to Work? RANS models frozen in time Having two is much better than one Offer version with “modern” k, primarily for skin friction at very high Re Offer “early separation” and “late separation” versions? Shock-induced and low-Mach separation may have different preferences Avoid proliferation of versions DES removes some of the burden from RANS Look for great new players, and help them Karman constant Firming up would be so, so nice! I’m leaning towards 0.38, based on Boundary-Layer Cf DNS is stronger and less naïve in 2006 than 1986, but the goal has moved RANS models rebellious Confirm two-equation approach decay behavior in practice Establish consensus on actual requirements for ambient values Combat Transition Creep, spread transition control in Navier-Stokes CFD LES Enjoy! Prudently upgrade DES and Wall Modeling Build up and test unsteady capabilities in every NS code

Add a comment

Related presentations

Related pages

Stress: The Constant Challenge | PPT Directory

Stress: The Constant Challenge Chapter 2 . 2 . Concepts in Stress . Stressor--Situations that trigger physical and emotional responses (stress responses)
Read more