ebook img

Numerical Study Comparing RANS and LES Approaches on a Circulation Control Airfoil PDF

29 Pages·2010·5.81 MB·English
by  
Save to my drive
Quick download
Download
Most books are stored in the elastic cloud where traffic is expensive. For this reason, we have a limit on daily download.

Preview Numerical Study Comparing RANS and LES Approaches on a Circulation Control Airfoil

Numerical Study Comparing RANS and LES Approaches on a Circulation Control Airfoil ChristopherL.Rumsey∗ NASALangleyResearchCenter,Hampton,VA23681-2199 TakafumiNishino† NASAAmesResearchCenter,MoffettField,CA94035-1000 Anumericalstudyoveranominallytwo-dimensionalcirculationcontrolairfoilisperformedusingalarge- eddysimulationcodeandtwoReynolds-averagedNavier-Stokescodes. DifferentCoandajetblowingcondi- tionsareinvestigated.Inadditiontoinvestigatingtheinfluenceofgriddensity,acomparisonismadebetween incompressibleandcompressibleflowsolvers. Theincompressibleequationsarefoundtoyieldnegligibledif- ferencesfromthecompressibleequationsuptoatleastajetexitMachnumberof0.64.Theeffectsofdifferent turbulencemodelsarealsostudied.Modelsthatdonotaccountforstreamlinecurvatureeffectstendtopredict jetseparationfromtheCoandasurfacetoolate,andcanproducenon-physicalsolutionsathighblowingrates. Threedifferentturbulencemodelsthataccountforstreamlinecurvaturearecomparedwitheachotherand withlargeeddysimulationsolutions.AllthreemodelsarefoundtopredicttheCoandajetseparationlocation reasonablywell,butoneofthemodelspredictsspecificflowfielddetailsneartheCoandasurfacepriortosep- arationmuchbetterthantheothertwo. AllReynolds-averagedNavier-Stokescomputationsproducehigher circulationthanlargeeddysimulationcomputations,withdifferentstagnationpointlocationandgreaterflow accelerationaroundthenoseontotheuppersurface. Theprecisereasonsforthehighercirculationarenot clear,althoughitisnotsolelyafunctionofpredictingthejetseparationlocationcorrectly. Nomenclature A Planformarea P Productionterms,Eqs.(2)and(8) a Speedofsound p Pressure C Sectionliftcoefficient,L/(q A) q Dynamicpressure,ρU2/2 L ∞ C Pressurecoefficient,(p−p )/q Re Reynoldsnumber,ρU c/µ p ∞ ∞ ∞ C ConstantforSSTRCmodel R GradientRichardsonnumber,Eq.(10) rc i C Jetmomentumcoefficient,Eq.(11) r Radiusofcurvature µ c Airfoilchordlength r˜ TerminSARCmodel c ConstantinSAmodel r∗ S/Ω b1 c ,c ,c ConstantsinSARCmodel S Magnitudeofstrain,Eq.(5) r1 r2 r3 c0 coefficientinEASM-komodel S˜ TerminSAmodel µ D DiffusionterminSAmodel S Strainratetensor diff ij D DissipationterminSAmodel t time diss dist Normaldistancefromthewall/c U Velocity F FunctioninSSTandSSTRCmodels u,v,w Cartesianvelocitycomponents 1 F Sensitizationfactor,Eq.(9) u Cartesianvelocitycomponent 4 j f ,f FunctionsinSAmodel u Streamwisevelocity r1 t2 θ h Slotheight u0u0 Specificturbulentstresstensorcomponent i j k Turbulentkineticenergy u0w0 Specificturbulentshearstress,Eq.(12) θ θ L Liftforce W Vorticitytensor ij M Machnumber,U/a x,y,z Cartesiancoordinates m˙ Massflowrate x Cartesiancoordinate j ∗SeniorResearchScientist,ComputationalAeroSciencesBranch,MailStop128,AssociateFellowAIAA. †PostdoctoralFellow,NASAAdvancedSupercomputingDivision,MailStopT27B-1,MemberAIAA. 1of29 AmericanInstituteofAeronauticsandAstronautics Greek: Subscript: α Angleofattack(deg) inlet Plenuminletcondition β,β∗ ConstantsinSSTandSSTRCmodels j Jetcondition ∆s+ Streamwisespacinginwallunits max Maximumcondition ∆n+ Normalspacinginwallunits mean Meancondition ∆b+ Spanwisespacinginwallunits ref Reference(∞)condition γ ConstantinSSTandSSTRCmodels sep Jetseparationlocation µ Coefficientofviscosity ∞ Free-streamquantity µ Turbulenteddyviscosity t ν˜ TurbulencevariableinSAmodel θ Angle(deg)aroundCoandasurface ρ Density σ ,σ ,σ ConstantsinSSTandSSTRCmodels k ω ω2 τ Turbulencetimescale τ Shearstress ij Ω Magnitudeofvorticity,Eq.(4) ω Specificdissipationrate I. Introduction Circulationcontrolcandramaticallyenhancetheliftofairfoilsandwings. Byblowingajetofairalonganairfoil surfacenearitscurvedtrailingedge,separationcanbedelayedbecauseoftheCoandaeffect(whichcausesthejetto “stick”tothecurvedsurface). SeeFig.1. Manycomputationalstudieshavebeendoneforcirculationcontrolairfoils; forexample,seeRefs.1–9. However,ingeneralmosteffortshaveonlyattainedlimitedsuccess. Inparticular,many Reynolds-averagedNavier-Stokes(RANS)modelscandoreasonablywellforlowblowingconditions, butathigher blowingratestheytendtopredictjetseparationtoolate. Infact,sometimesthecomputedjetwrapsun-physicallytoo farontotheairfoil’slowersurface,orevencompletelyaroundtheairfoil. Therearemanydifficultiesassociatedwithvalidatingcodes,methods,andmodelsoncirculationcontrolairfoils. Duetothetypicallyverynarrowblowing-slotopenings,thenear-wallregionofcirculationcontrolairfoilexperiments can be very difficult to measure. In particular, it is important to know the jet exit conditions very accurately, so that the CFD can apply the same boundary conditions. The behavior of the jet can be very sensitive to the exit conditions. The integral quantity jet momentum coefficient, typically defined from experiment based on total mass flowrateandderivedjetexitconditions, maybeinsufficientinformationforthepurposeofCFDvalidationbecause of possible nonuniform blowing across the entire airfoil span and errors from the isentropic flow assumption.10 It is also particularly difficult to maintain/guarantee two-dimensionality in the wind tunnel flowfield at strong blowing conditions. Furthermore,three-dimensionalvorticalstructuresinthewindtunnelduetointeractionwiththesidewalls cause a net downwash (or effective negative angle of attack) that increases in magnitude with increasing blowing. Some 3-D RANS studies that included side walls2,10,11 have demonstrated this effect. The effect of this downwash canbedifficulttoaccuratelyaccountforwhenperforming2-DCFD,especiallywhenalsoattemptingtoaccountfor the upper and lower tunnel walls. To date, 3-D LES computations have typically lacked the resources for attaining sufficientgriddensitytoadequatelyresolvetheside-wallvorticalstructures. OfallRANSmodelsappliedtocirculationcontrolairfoils,arguablythemostsuccessfultodatehaveincorporated somemodelingofcurvatureeffects,whichcanbesignificantincirculationcontrolcasesnearthetrailingedgeCoanda surface. Forexample,theSpalart-AllmarasforRotationandCurvature(SARC)model12 (seeSwansonandRumsey1) andafullReynoldsstressmodel(seeSlomskietal.13)tendedtoyieldbetterresultscomparedtoexperimentthanother RANS models like Spalart-Allmaras (SA)14 or Menter’s shear-stress transport (SST) k-ω,15 which do not directly account for curvature. However, even curvature-corrected models have not done particularly well at very strong blowingconditions;1theystilltendtoyieldtoohighliftcomparedwithexperiment. Recently,LEScomputationshavealsobeenappliedtocirculationcontrolflows. Therehasbeensomesuccessat lowblowingconditions,16–19 butathigherblowingratesLEShasnotbeenassuccessful.20 Thelackofbetteroverall success for LES to date may be due to insufficient computational resources. Since computational capacity changes veryrapidly,thisavenueofpursuitisbeingcontinued. An on-going collaborative experiment at NASA Langley and Georgia Tech has been studying the flow around a circulationcontrolairfoildesignedattheGeorgiaTechResearchInstitute(GTRI).21Becauseonlylimitedexperimental 2of29 AmericanInstituteofAeronauticsandAstronautics dataspecificallyforthepurposeofCFDvalidationhavebeenobtainedtodateforthismodel(see,e.g.,Allanetal.10), thispaperwillfocusonmakingcomparisonsbetweenRANSandLESforflowovertheGTRImodel,withtheintentof discerningkeydifferencesbetweenthemethods. Thefocusistoemploythemethodsonidenticalgridsandmatching boundary conditions at zero angle-of-attack, including the top and bottom walls. Three-dimensional effects due to side-wallsareneglected,althoughtheLESincludes3-Dflowfieldstructures. Wewouldliketoanswerthequestion: In what ways do the 2-D RANS and spanwise-averaged LES methods differ? Attempting to answer this question involves determining the influence of grid density, code, and turbulence modeling. A curvature correction for SST (fromHellsten22 andManietal.23),thathasnotbeenappliedtocirculationcontrolairfoilsbefore,isalsoexamined. Eventually,itishopedthatLESmaybeusedtoimproveRANSturbulencemodelsforthesekindsofflowfields. Thispaperisorganizedasfollows. AfterabriefdescriptionoftheRANSandLEScomputercodesemployedfor this effort, the grid and test cases are described. Then, in the Results section, four areas of focus are discussed: (a) agridandcodesensitivitystudy,(b)comparisonsbetweencompressibleandincompressibleRANSsolutions,(c)the effects of turbulence models and curvature corrections for a case with high blowing, and (d) comparisons between RANSandLES.Finally,conclusionsaredrawn. II. NumericalMethodsandTurbulenceModels The RANS computer codes employed for this study were FUN3D24 and CFL3D.25 FUN3D is an unstructured finitevolumeupwind-biasednode-centeredcode,andCFL3Disamultiblockstructuredfinitevolumeupwind-biased cell-centeredcode. BothsolvethecompressibleRANSequationsandarenominallysecondorderspatiallyaccurate. FUN3Dsolvestheequationsonmixedelementgrids,includingtetrahedra,pyramids,prisms,andhexahedraand also has a two-dimensional path for triangular and quadrilateral grids. It employs an implicit upwind algorithm in whichtheinviscidfluxesareobtainedwiththefluxdifferencesplittingschemeofRoe.26 Forsecond-orderaccuracy, interfacevaluesareobtainedbyextrapolationofthecontrolvolumecentroidalvalues, basedongradientscomputed at the mesh vertices using an unweighted least-squares technique. The solution at each time-step is updated with a backwardEulertime-differencingscheme. Ateachtimestep,thelinearsystemofequationsisapproximatelysolved withamulti-colorpoint-implicitprocedure. Localtime-stepscalingisemployedtoaccelerateconvergencetosteady- state.Foralltheapplicationsinthispaper,FUN3Dsolvestheturbulenceequationswithafirst-orderadvectionscheme. InCFL3D,theconvectivetermisapproximatedwiththird-orderupwind-biasedspatialdifferencing,andthevis- cous terms are discretized with second-order central differencing. The flux difference-splitting method of Roe26 is employedtoobtainfluxesatthecellfaces. AdvancementintimeisaccomplishedviabackwardEuler,withanimplicit approximatefactorizationmethod. Theturbulenceequationsaresolvedwithafirst-orderadvectionscheme. The LES code employed is CDP, an incompressible Navier-Stokes code developed at the Center for Turbulence Research(CTR)atStanfordUniversity.27 CDPisafinitevolumeunstructuredflowsolverwithsecondorderspatial accuracy. Itisbasedontheleast-squarespressuregradientreconstructionofMaheshetal.28 forhybridunstructured grids. Themethodologyisnearlyenergyconservingintheinviscidlimit. CDPusescentraldifferencinganddoesnot introduceanynumericaldissipationorsmoothing. Thenecessarydissipationcomessolelythroughthesubgrid-scale model. Thesubgrid-scalestressesintheCDParemodeledwiththedynamicSmagorinskymodel.29 TheRANSturbulencemodelsSA14 andSST15 andtheLESdynamicSmagorinskymodel29 arestandardandwill notbedescribedhere. Detailscanbefoundintheirrespectivereferences. However,abriefdiscussionwillbegiven forthelessstandardmodelsSARC,EASM-ko,andSSTRC. IntheSARCmodel,12theproductiontermoftheSAmodelismodified. Intheoriginalmodel, ∂ν˜ ∂ν˜ +u =P +D +D (1) ∂t j∂x diff diss j whereν˜isrelatedtothekinematiceddyviscosity,P = c (1−f )S˜ν˜,andS˜isafunctionofthemagnitudeofthe b1 t2 vorticity. AllthetermsaredefinedintheoriginalSAreference.14 IntheSARCmodification, P =c (f −f )S˜ν˜ (2) b1 r1 t2 where f =(1+c ) 2r∗ (cid:2)1−c tan−1(c r˜)(cid:3)−c (3) r1 r1 1+r∗ r3 r2 r1 withr∗ =S/Ω,and p |Ω|= 2W W , (4) ij ij 3of29 AmericanInstituteofAeronauticsandAstronautics p |S|= 2S S , (5) ij ij andW =(∂u /∂x −∂u /∂x )/2isthevorticitytensorandS =(∂u /∂x +∂u /∂x )/2isthestrain-ratetensor. ij i j j i ij i j j i Thefunctionr˜dependsontheLagrangianderivativeofthestrain-ratetensor(seeoriginalreference),andtheconstants aretakentobec =1,c =12,andc =1. r1 r2 r3 The EASM-ko model30 is an explicit algebraic stress model in k-ω formulation. In this model, the turbulent eddy viscosity µ is given by µ = c0 ρkτ, where τ = 1/ω for the k-ω version, and c0 is a variable coefficient t t µ µ obtainedthroughthesolution ofacubicequation. This modelmakesuseofanon-linearconstitutive relationrather thantheBoussinesqmodel. AsdiscussedinRumseyetal.,31 becauseEASMformulationsarederiveddirectlyfrom theReynoldsstressmodel,theyretainsomeoftheinvariancepropertiesofthefulldifferentialformeventhoughthe assumption of anisotropy equilibrium in the Cartesian frame of reference is not strictly correct for curved flows. A curvature corrected form of EASM was derived, but curved flow results were very similar to those predicted by the standard form of the model. Because the standard model is more robust than the curvature corrected version, only EASM-ko(thestandardmodel)wasusedinthecurrentwork. SSTRC is Hellsten’s curvature correction22 to the standard SST model. In SSTRC, the destruction term in the ω-equation is multiplied by a curvature sensitization factor F , which is a function of the local flow curvature. The 4 equationsare: (cid:20)(cid:18) (cid:19) (cid:21) ∂k ∂k ∂ µ ∂k ρ +ρu =P −β∗ρkω+ µ+ t (6) ∂t j∂x ∂x σ ∂x j j k j (cid:20)(cid:18) (cid:19) (cid:21) ∂ω ∂ω γρ ∂ µ ∂ω 1−F ∂k ∂ω ρ +ρu = P −F βρω2+ µ+ t +2ρ 1 (7) ∂t j∂x µ 4 ∂x σ ∂x σ ω ∂x ∂x j t j ω j ω2 j j wheretheturbulenceproductiontermis ∂u P =τ i (8) ij∂x j Thesensitizationfactorisgivenby 1 F = . (9) 4 1+C R rc i ThecurvaturesensitizationfactorF isalocalformulationthatisindependentofthecoordinatesystemandbasedon 4 Hellsten’sapproximationtothegradientRichardsonnumber: (cid:18) (cid:19) |Ω| |Ω| R = −1 , (10) i |S| |S| This gradient Richardson number is easily computed, and is independent of coordinate system and grid topology. Near-wall convex curvature yields F levels below 1, which decreases ω-destruction and in turn increases ω. The 4 near-walleddyviscositydecreasesasaresult. Therefore,usageofthecorrectioncoefficientF weakenstheturbulent 4 boundarylayeraroundconvexcurvedsurfaces,whichcreatesagreatertendencytoseparate. Hellsten originally recommended C = 3.6, but this value was found to diffuse the jet excessively in some rc applications. Manietal.23 recommendedavalueofC =1.4forcasesinvolvingground-jetinteractions. Thisvalue rc isalsousedforallresultshere. Seetheoriginalreferencesfordetailsconcerningothertermsintheequations. III. DescriptionofGridandTestCases Anearviewofthe2-Dgrid(witheveryothergridpointremoved)isshowninFig.2. Thisfigurealsoshowsthe geometryoftheairfoil,includingthedual-chamberedplenum. TheearliersketchofFig.1showedthegeometrynear theCoandasurface(notethatthelowersurfaceincludedaslotforblowing, butitwasclosedforthecurrentstudy). Although not shown, the top and bottom tunnel walls were located at z = ±0.508 m (z/c = ±2.327). A wavy transition strip was located near the lower surface leading edge (for details, see Nishino et al.17). The airfoil chord wasc = 0.2183m. Thejetslotheightwash = 0.000503m(h/c = 0.0023). TheradiusoftheCoandasurfacewas r =0.02066m(r/c=0.09463). ThefinegridemployedfortheLEScomputationshadapproximately116milliongridpoints. Ithadaspanwise (y-direction)extentof0.014m(ory/c=0.0641),with256pointsinthatdirectionandperiodicboundaryconditions 4of29 AmericanInstituteofAeronauticsandAstronautics at the side boundaries. For the lowest jet blowing condition, the coarsest grid resolution (in wall units) over the Coandasurfacewasapproximately∆s+ ≈42,∆n+ ≈0.6,∆b+ ≈25,inthestreamwise,wall-normal,andspanwise directions,respectively. NotethattheresolutioninwallunitsreportedinTable1ofNishinoetal.17 wasnotcorrect. The subgrid-scale eddy viscosity in the attached jet region (i.e., before the separation of the jet from the Coanda surface)wasuptoabout5timeslargerthanthemolecularviscosity. Forthehigherjetblowingcondition,thecoarsest resolutionwasapproximately∆s+ ≈60,∆n+ ≈0.8,∆b+ ≈36,andthesubgrid-scaleeddyviscosityintheattached jetregionwasuptoabout10timeslargerthanthemolecularviscosity. Thegridincludedthetunneltopandbottom walls,andalsomodeledthetwoinnerplenumchambersaswellasthewavytransitionstriponthelowersurfacenear the leading edge. The fine grid used for the RANS computations was the same, but was solved two-dimensionally using451,400gridcellsinonespanwiserow. Itisimportanttonotethatforthe2-DRANScomputations,thewavy strip(describedindetailinNishinoetal.17),issimplyasmallrectangularbumplocatednearx/c=0.021withheight 0.00256candlength0.01368c. Thegridwasoriginallyconstructedasamulti-zonegridwith73zones. Forusewith FUN3D,asinglezoneconsistingofhexahedralelementswascreatedfromtheoriginalgridbyremovingallduplicate pointsatthezoneinterfaces. The free-stream flowconditions were: M = 0.099, Re = 2.24×106 per m, or Re = 488,992 per chord. The RANS computations were run in steady-state mode, whereas the LES computations were run time-accurately and averaged to obtain statistically steady solutions, as described in Nishino et al.17 Because the simulations included wind tunnel walls, no angle of attack corrections were applied to any cases; all were run at α = 0◦. Viscous wall boundaryconditionswereappliedonallsurfacesoftheairfoil,includingtheinteriorplenum. Theonlyexceptionwas theleft-mostfaceoftheinterior-mostplenum,atwhichavelocityanddensitywerespecified(thisfacehadaheight of0.01261m). Thetopandbottomtunnelwallsweretreatedasinviscidsurfaces. Theupstreaminflowplane(located atx/c=−3.621)hadafarfieldRiemanninvariantboundarycondition,andthedownstreamoutflowplane(locatedat x/c=10.005)hadaspecifiedbackpressurep/p =1.0. ref The various plenum boundary conditions employed are listed in Table 1. Conditions were chosen to match the massflowrate,withtotaltemperatureatinflowof295K.Thejetmomentumcoefficientisdefinedas: m˙ U C = j j,mean (11) µ q A ref wherem˙ isthemassflowrateatthejetexitandU isthemeanjetvelocityatthejetexit. Notethatthespecific j j,mean conditionsforcompressibleandincompressibleboundaryconditionsattheinletfortheRANSandtheLESneededto besomewhatdifferentinordertoachieveasimilarvelocityatthejetexit. TheincompressibleRANScomputations inthispaperusedthe(A)conditions;however,thedifferencebetweenresultsusing(A)and(B)conditionswasfairly small. The reference velocity was U = 34 m/s. The reference speed of sound was a = 344 m/s and the ref ref referencedensitywasρ = 1.2kg/m3. Mostcomparisonswereperformedforthetwolowestblowingconditions ref (M ≈ 0.39 and 0.64). Only a few compressible RANS computations (and no LES) were performed at the highest j blowing rate of M ≈ 0.90, in order to demonstrate the effect that the various turbulence models have at that high j blowingrate. ItisrecognizedthatthelocalMachnumberoftheflownearthejetexitcanexceedlevelsforwhichincompressible flow assumptions are considered valid. However, at least for the two lower blowing cases, it was felt that the high- Mach-numberregionswerelimitedtoafairlysmallareasothatuseoftheincompressibleLEScodewasjustified.17 As willbeshownbelow,FUN3Dwasrunbothincompressibleandincompressiblemodesanditsresultswerecompared inordertoquantifytheinfluenceofsolvingtheincompressibleequationsandjustifytheirusefortheseflows. IV. Results A list of the CFD computations performed is provided in Table 2. The table includes computed lift coefficients andjetseparationangle(θ ),asdefinedinFig.1. NotethattheM = 0.39jetseparationanglefortheLESlisted sep j hereisslightlydifferentfromthatgiveninNishinoetal.18 becausehereweareaccountingfortheverythinregionof meanreverseflownearthebodythatoccursnearthejetseparationpoint. Inotherwords,theseparationangleforall resultsinthispaperisdefinedasthelocationwherewallskinfrictiongoestozerointhetime-averagedflowfield. In the“extra-fine”gridusedforoneoftheLEScomputations,thenumberoftangentialcellsaroundtheCoandasurface wasdoubled(from800to1600). Allotheraspectsofthegridremainedthesameasthefinegrid. Ontheotherhand, the“medium”gridlevelusedfortheCFL3Dgrid-independencestudiesconsistedofeveryotherpointremovedfrom thefinegridineachcoordinatedirection. Fig.3showstime-averagedsurfacepressurecoefficientsfromtheLEScomputations,demonstratingthat–other 5of29 AmericanInstituteofAeronauticsandAstronautics Table1. SummaryofCoandablowingconditions conditionsperunitspan U ,m/s ρ ,kg/m3 inlet inlet m˙ =0.0636kg/(ms)(C ≈0.044,M ≈0.39,U ≈130m/s) µ j j,max compressibleRANS 3.925 1.285 incompressibleRANS(A) 4.268 1.182 incompressibleRANS(B) 4.203 1.2 incompressibleLES 4.203 1.2 m˙ =0.1048kg/(ms)(C ≈0.120,M ≈0.64,U ≈210m/s) µ j j,max compressibleRANS 5.703 1.457 incompressibleRANS(A) 7.126 1.166 incompressibleRANS(B) 6.924 1.2 incompressibleLES 6.924 1.2 m˙ =0.1403kg/(ms)(C ≈0.215,M ≈0.90,U ≈300m/s) µ j j,max compressibleRANS 6.248 1.781 thanintheimmediateregionofthewavytransitionstrip–theresultswereessentiallystatisticallytwo-dimensional(see alsoNishinoetal.17). Therefore,directcomparisonsbetweenspanwise-averagedLESand2-DRANSwerepossible. A. GridandCodeSensitivityStudy LEScomputationswereperformedfortheM = 0.64caseusingboththefinegrid(approximately116milliontotal j grid points) and the extra fine grid with double the number of tangential cells around the Coanda surface (approxi- mately175milliontotalgridpoints). Resultswereverynearlythesame. Visualizationsoftheflowstructuresaround theCoandasurfaceaswellasaveragedsurfacepressurecoefficientsareshowninFig.4. AsdescribedinNishinoet al.,17 anumberofhairpin-likeflowstructureswereobservedintheLESaroundthemixinglayerofthehigh-speedjet flowandthelow-speedexternalflow. Here,theseflowstructuresappearedtobesimilarbetweenthetwogrids. Many2-DRANScomputationswereperformedonboththefinegrid(451,400gridcells)aswellasonthemedium gridwitheveryotherpointremoved(112,850gridcells).Furthermore,thetwocodesFUN3DandCFL3Dwererunon thesamefinegridusingthecommonSA,SST,andSSTRCturbulencemodelsasacode-independencecheck. FUN3D did not have the ability to run the SARC or EASM-ko turbulence models, but these were run with CFL3D on the two grid sizes. Results are tabulated in Table 2, and surface pressure coefficients are shown in Fig. 5. For the most part,resultswerenearlythesameonthetwogridsizesandwiththetwocodes. (However,notethatEASM-kodidnot produceasteady-statesolutionatM =0.39onthemediumgrid,soonlythefinegridresultisgiven.)Twoexceptions j to this were the SA and SST models on the M = 0.90 case. These particular solutions were non-physical and not j well-converged: theSAmodelallowedthejettocompletelywraparoundtheairfoil,andtheSSTmodelallowedthe jet to stick to the Coanda surface all the way to the lower surface, so that the jet faced forward into the oncoming stream. Inthec plots,thisnon-physicalbehaviorshowsupasalarge“kink”or“step”inthelowersurfacepressure p coefficientbetweenx/c = 0.90and1.0. Thisbehaviorwasconsistentbetweencodesandoccurredonbothfineand mediumgrids. Overall,thedifferencesbetweentheturbulencemodelsweremuchgreaterthanthedifferencesduetogridorcode. Thisfactisimportant, becauseitmakesitpossibletodrawconclusionsaboutthebehavioroftheturbulencemodels withsomedegreeofconfidencethatdiscretizationerrorsandcodebiasareprobablynotplayingasignificantpart. As seeninTable2,theeffectofgrid/codeonliftcoefficientandjetseparationlocationwerealsominimal. B. ComparisonsBetweenCompressibleandIncompressibleRANSSolutions Forevenmoderateblowingconditions,thejetbehavescompressiblynearthejetexit. Nonetheless,researchershave employed both incompressible and compressible CFD codes to circulation control airfoil flows, but until now the impactofthetypeofsolverhasnotbeenexplored. Itcanbearguedthatthecompressibleregionisverylocalizedfor manyblowingrates,andhencetheuseofincompressibleequationsmaynothavemuchnegativeconsequenceforthe 6of29 AmericanInstituteofAeronauticsandAstronautics Table2. Summaryofcomputationsperformed Method M Code Grids Turbulencemodel C Jetsepar. (θ ) j L sep IncompressibleLES 0.39 CDP fine dynamicSmag 1.36 67◦ 0.64 CDP fine/extra-fine dynamicSmag 3.50/3.51 95◦/96◦ IncompressibleRANS 0.39 FUN3D fine SA 1.85 73◦ 0.64 FUN3D fine SA 4.67 107◦ CompressibleRANS 0.39 FUN3D fine SA 1.84 72◦ 0.39 FUN3D fine SST 1.67 68◦ 0.39 FUN3D fine SSTRC 1.56 64◦ 0.64 FUN3D fine SA 4.65 107◦ 0.64 FUN3D fine SST 4.49 106◦ 0.64 FUN3D fine SSTRC 3.93 95◦ 0.90 FUN3D fine SA 4.91 wrap 0.90 FUN3D fine SST 6.46 174◦ 0.90 FUN3D fine SSTRC 6.69 117◦ 0.39 CFL3D fine/med SA 1.86/1.82 73◦/73◦ 0.39 CFL3D fine/med SARC 1.72/1.71 68◦/67◦ 0.39 CFL3D fine/med SST 1.70/1.69 69◦/68◦ 0.39 CFL3D fine/med SSTRC 1.59/1.57 65◦/64◦ 0.39 CFL3D fine/med EASM-ko 1.70/unst 68◦/unst 0.64 CFL3D fine/med SA 4.72/4.66 108◦/108◦ 0.64 CFL3D fine/med SARC 4.04/4.02 95◦/95◦ 0.64 CFL3D fine/med SST 4.42/4.40 105◦/103◦ 0.64 CFL3D fine/med SSTRC 3.92/3.90 95◦/93◦ 0.64 CFL3D fine/med EASM-ko 4.10/4.07 97◦/96◦ 0.90 CFL3D fine/med SA 4.61/4.50 wrap/wrap 0.90 CFL3D fine/med SARC 6.96/6.84 112◦/112◦ 0.90 CFL3D fine/med SST 6.88/7.02 169◦/155◦ 0.90 CFL3D fine/med SSTRC 6.79/6.69 119◦/118◦ 0.90 CFL3D fine/med EASM-ko 6.74/6.71 116◦/115◦ 7of29 AmericanInstituteofAeronauticsandAstronautics globalflowfield. Byusingasinglecode(FUN3D)thathasbothincompressibleandcompressiblecapabilities,theimpactisexplored forthefirsttimehere.Figs.6(a)and(b)showstreamlinesfortheM =0.64case,usingFUN3Dinincompressibleand j compressible mode, respectively. Fig. 6(c) shows the streamwise velocity (u nondimensionalized by the reference θ velocity) profile comparisons. Incompressible and compressible results were essentially the same, including the jet separationlocation. Althoughnotshown,resultsfortheM =0.39casewereevencloser,aswouldbeexpected. The j M =0.90caseproducednon-physicalresultsusingtheSAmodel,sodidnotprovideanyusefulinformation. These j resultsappeartojustifytheuseofincompressibleflowsolvers(liketheCDPcodeemployedforLEScomputations), atleastforblowingratesuptoM =0.64. j C. EffectofTurbulenceModelandCurvatureCorrectionforM =0.90 j Results from Swanson et al.2 indicated that a rotation/curvature correction to the SA turbulence model can signifi- cantlyimproveresultsforcirculationcontrolcases. Furthermore,Swansonetal.showedthattheSSTmodeltended toproducepoorsolutionsathigherblowingrates. Here,twootherturbulencemodelsthatprovidesomelevelofsen- sitivitytoflowcurvature–EASM-koandSSTRC–arealsoexaminedfortheM = 0.90case. EASMmodelswere j alsoinvestigatedpreviouslyforcirculationcontrolairfoilsbyGrossandFasel.9 Earlier,Fig.5showedsurfacepressurecoefficientsforalltheRANSmodelsonallthreecases. Thetwocasesthat stoodout(inanegativesense)wereSAandSSTatM = 0.90. Figs.7(a)–(e)showstreamlinesfortheM = 0.90 j j case, using all five RANS models. The jet in the SA result wrapped completely around the airfoil (Fig. 7(a)), and the jet in the SST result separated close to the end of the Coanda surface near θ = 170◦ (Fig. 7(b)). Neither of these models accounts for curvature. The three models that account for curvature yielded separation locations near θ ≈110◦−120◦. Basedonothercirculationcontrolairfoilexperiments(e.g.,Novaketal.32),thisseparationlocation appearsmuchmorereasonablethanthatpredictedbySAorSST. D. ComparisonsBetweenRANSandLESSolutions ThemeanLESliftcoefficientsasafunctionofC arecomparedwiththesteady2-DRANSresultsinFig.8.Generally, µ theSARC,SSTRC,andEASM-koresultswerewithinafewpercentofeachotheratallblowingvalues. TheSAand SSTlevelswerehigherattheM =0.64condition. Atthehighestblowingcondition,theSAwrap-aroundproduced j averylowC value,whiletheSST(coincidentally)producedC inagreementwithSARC,SSTRC,andEASM-ko L L in spite of the fact that its jet separation angle was so much larger. The LES results were lower than the best of the othermodelsbybetween12and17%. 1. M =0.39Case j Fig.9showssurfacepressurecoefficientsfortheM = 0.39case. AsdiscussedinNishinoetal.,17 theLESyielded j excellentagreementwith(preliminary)experimentaldataatthisblowingcondition. TheLESalsoagreedwellwith independent implicit LES computations of Madavan and Rogers19 (note that the latter computations were not direct numericalsimulations,asimpliedbythepaper’stitle). Here,allRANSresultspredictedmoreuppersurfacesuction thantheLES.ThejetintheSAmodelresultseparatedslightlylater(by5-6◦)thanLES,butalltheotherRANSmodels predictedtheseparationwithin1-3◦ofLES. BothhereandinNishinoetal.,18C resultsusingSAwerefurthestfromLES.Here,theSST,SARC,andEASM- p kowereallaboutthesame,andSSTRCwastheclosesttoLES.AdditionaldetailsareshowninFigs.10and11. (In these figures, only the best RANS results SSTRC, SARC, and EASM-ko are shown.) The SSTRC and EASM-ko models both predicted a higher peak jet velocity and narrower jet width compared to LES, while the SARC model was much closer. In fact, SARC and LES velocity profiles were very similar prior to jet separation. In Fig. 11, the nondimensional turbulent quantity u0w0 is given in a coordinate system aligned with the local near-wall flow and θ θ flow-normaldirections,where 1 u0w0 =− (w0w0−u0u0)sin(2θ)+u0w0cos(2θ) (12) θ θ 2 with u0u0, w0w0, and u0w0 the Cartesian specific turbulent shear stress components. Note that the LES results in- cludedonlytheresolvedcomponent(i.e.,thesubgrid-scalecontributions,whichwerefairlysmallintheattachedjet region,werenotincluded). Onthewhole,allRANSmodelsoverpredictedtheu0w0 peakmagnitudeclosetothewall θ θ 8of29 AmericanInstituteofAeronauticsandAstronautics compared to LES. In the outer part of the jet, the SSTRC and EASM-ko models severely underpredicted the peak magnitudeandextent,whereastheSARCmodelwascloserintheregionspriortojetseparation. ComparingthestreamlinesaroundthebackoftheairfoilinFig.12, itcanbeseenthatSSTRCyieldedabubble that was much too large compared to LES, and EASM-ko yielded a separation bubble that was very different in character from the LES. The SARC solution was very close to the LES solution, but the separation bubbles were slightlydifferentinshape,andtheSARCstreamlinesdroveverticallydownwardfurtherthanforLES.Fig.13shows jet/wakeprofilesbehindtheairfoil.Atthefirstpositionneartheclosureoftheseparationbubble,alltheRANSmodels werefairlyconsistentwithLES,butfurtherdownstreamtherewerebiggerdifferences. TheRANSmodelsallalmost completelydissipatedtheveryweakjet,andexhibitedmorewake-likebehaviorthanLES.Withthesedifferencesin bubblesize/positionandjet/wakebehaviorandposition,theRANSsolutionshadaresultinglargeroverallcirculation thanLES.Thus,asshowninFig.14,theRANSstagnationpointneartheairfoilleadingedgewasfurtherdownstream (note that the region near the trip is not shown in the LES solution because the 3-D spanwise waviness of the trip itselfprecludedspanwiseaveraginginthisarea). Althoughnotshown,theotherRANSmodelsweresimilartoSARC. ThereforetheRANSflowacceleratedmorearoundthenose. Boundarylayerprofilesontheupperandlowersurfaces nearx/c = 0.8areshowninFig.15. Bothupperandlowersurfaceswereturbulentattheselocations. Becausethe stagnationpointlocationsdidnotagree,theedgevelocitiesforRANSweredifferentthanforLES,asexpected. 2. M =0.64Case j The M = 0.64 surface pressure coefficients and Coanda profile results are shown in Figs. 16, 17, and 18. Similar j tothe lowerblowing case, here theSAand SSTyielded theworst agreementwithLES. TheSA andSST produced similarjetseparationlocationsbetween105-108◦,whereasLES,SSTRC,SARC,andEASM-koallseparatedearlier, near 95-97◦. Among the RANS models that accounted for curvature, all produced very similar c distributions, but p noneagreedwithLES.Again,SARCcametheclosesttoagreeingwiththevelocityandu0w0 profilesfromLESprior θ θ toseparation. TheM = 0.64blowingcasewasdifferentincharacterthantheM = 0.39case,becausetherewasnolongera j j largeseparationbubbledirectlynexttotheundersideoftheseparatingjet,butratheramuchsmalleronefurtheraround, nearer to the bottom of the Coanda surface. (As was seen in Fig. 7, at the even stronger blowing with M = 0.90 j this separation bubble went away completely.) Fig. 19 shows details of the flowfields near the trailing edge. The separationbubblewasslightlysmallerforSARCthanforLES,butSSTRCandEASM-koagreedwellwithLESin this regard. In spite of the fact that the RANS solutions separated at about the same location as the LES solution, theRANSjetsmoveddownwardfurtherverticallythanLESdownstream. AsseeninFig.20,inthishigherblowing casethejetremainedfairlystrongpost-separation,andintermsofpeakvelocityandspreadingrate,theSSTRCand EASM-kobothdidareasonablygoodjobpredictingit. However,theSARCmodeldissipatedthejettooquickly. This isaknownproblemwiththeSAmodelitself,33 andtheproblemnodoubtcarriesovertotheSARCvariant. Again, allRANSmodelsproducedmorecirculationthanLES.AsseeninFig.21,theleadingedgestagnationpointforthe RANSmodelwasagainlocatedfurtheraftonthelowersurface,causinghighervelocitiesastheflowacceleratedonto theuppersurface(althoughnotshown, theleadingedgebehaviorofSSTRCandEASM-koweresimilartoSARC). Boundary layer profiles on the upper and lower surfaces near x/c = 0.8 in Fig. 22 indicate that both upper and lowersurfaceswereturbulentattheselocations. Again,becausethestagnationpointlocationsdidnotagree,theedge velocitiesforRANSweredifferentthanforLES. V. DiscussionandConclusions Inpaststudies,ithasbeengenerallyassumedthatifonecouldcapturethecorrectjetseparationlocationfromthe Coandasurfaceofcirculationcontrolairfoils,thenthesimulationwouldyieldaccurateairfoilforcesandpressuredis- tributions. Many“standard”turbulencemodelsinearlierstudiestendedtopredictjetseparationtoolate,consequently yieldingliftpredictionsthatweretoohigh(comparedtoexperiment)byasignificantamount. Here,althoughdetailed experimentaldata(includingmeasurementsofjetboundaryconditions)fortheparticularconfigurationbeingstudied haveyettobecompleted,wehavecompared2-DRANScomputationsusingvariousturbulencemodelswith3-DLES. Wehavedemonstratedthatturbulencemodelsthataccountforcurvaturecanproducejetseparationlocationsthatagree verywellwithLES.Threedifferentmodels–SSTRC,SARC,andEASM-ko–allperformedaboutthesameinterms of the jet separation location and surface pressure coefficient predictions. However, the SARC model yielded much betterdetailedagreementintermsofvelocityandturbulenceprofilesovertheCoandasurfacecomparedtoLES.Past studies(e.g., SwansonandRumsey,1 Pfingstenetal.7) havealsoshowntheSARCmodeltobereasonablygoodfor 9of29 AmericanInstituteofAeronauticsandAstronautics thesetypesofflows. Yet, in spite of agreeing fairly well for many details of the flowfield over the Coanda surface, the best RANS models still produced larger airfoil circulation (higher lift) than LES, by between about 12 and 17%. The specific reasonsforthelargercirculationhavenotbeendeterminedyet.Somepossibilitiesinclude:(a)differencesintheshape of the separation bubble underneath the jet separation and the direction and location of the separated jet sheet, (b) inadequatespreadingordissipationofthedownstreamjetintheRANSmodels,and(c)insufficientgridresolutionfor LES, RANS (or both) in regions farther from the airfoil. Also, although SARC was the best at predicting the flow very near the Coanda surface, it was worse than the SSTRC and EASM-ko models in predicting the free-shear jet characteristics downstream in the M = 0.64 case. The turbulence model must be able to accurately capture both j near-wallcurvatureeffectsandfree-shearplanarjetdecayfortheseproblems. Itappearsthatwithoutasharptrailing edge (Kutta condition), the jet behavior downstream can be extremely sensitive in these circulation control airfoil cases. Andtheairfoilcirculationinturnappearstobeverysensitivetosmalldifferencesinjetbehaviorandlocation. In terms of grid resolution, the RANS made use of one plane of the 3-D LES grid; this was very fine compared to grid sizes typically used for 2-D RANS in past studies. It was demonstrated that this grid size was fine enough toyieldRANSsolutionswithverylittledifferencesfromsolutionsonhalfthegridsizeineachcoordinatedirection. TwoindependentRANScodesalsoproducedalmostthesameresults. ItisnotknownatthistimewhethertheLES computationswereadequatelygrid-converged.Best-practicesforairfoil-typeLEScomputationswerefollowedwithin the available resource constraints. One LES computation required on the order of 2 months to complete using 256 processors. (Each fine grid RANS computation took on the order of several hours on 6 processors). A refinement of the grid in the region of the Coanda surface made little difference to the LES solution, but only the streamwise directionwasrefinedduetoresourceconstraints. Inallofthecomputationsperformed,carewastakentosettheboundaryconditionsinsidethejetplenuminorder to match both the jet mass flow rate and the velocity profile at the jet exit. With this same starting condition at the jetexitforallcodes,itwaspossibletoconfidentlyconcludethatdifferencesdownstreamwereduetotheturbulence modeling,andnotduetootherfactorssuchasboundaryconditioninfluence. Asmentionedabove,detailedexperimentaldataforthisconfigurationforthepurposeofCFDvalidationisbeing acquired, but is not yet available. Preliminary comparisons between LES and experiment17 indicated near perfect agreement between LES and experiment at the lowest blowing condition, but poorer agreement at a higher blowing condition. Measured experimental jet boundary conditions at the slot exit will be required before drawing any firm conclusions, however. Also, it is well-known that interactions of the jet sheet with tunnel side walls cause roll-up vorticesdownstream,whichinduceanetdownwashontheairfoil. Thiseffectincreaseswithincreasingblowingrate. TodatethisdownwashinfluencehasnotbeenaccountedforintheLEScomputations. In conclusion, this paper has highlighted some key differences between RANS and LES for circulation control airfoilflow. ItwasshownthatuseofanincompressibleLEScodewaslikelytobevalidforthisconfigurationforjet blowinguptoatleastM =0.64. Thiswasdemonstratedbyusinganincompressibleandcompressibleversionofthe j sameRANScode,andshowingessentiallyidenticalresults.Inotherwords,forthesecasesthesmallregionnearthejet exitwherethelocalMachnumberexceededlevelswheretheincompressibleassumptionisstrictlyvaliddidnotaffect thesolutionnoticeably. SeveralRANSturbulencemodelswerecomparedwitheachotherandwithLES.Atveryhigh blowingrates, thestandardSAandSSTmodels, whichdonotaccountforflowcurvatureeffects, hadatendencyto producenon-physicalsolutions,withSAsomewhatworsethanSST.Threedifferentmodelsthataccountforcurvature –SSTRC,SARC,andEASM-ko–allbehavedbetter,andweregenerallysimilartoeachotheroverall. Thebehavior ofSSTRConacirculationcontrolairfoilcasehasnotbeendocumentedpreviously. Allthreemodelsdidagoodjob predicting the jet separation location. SARC agreed much better with LES in terms of Coanda profile details, but worse in terms of downstream jet dissipation and spreading. All RANS models produced higher circulation (higher lift) than LES, with different stagnation point location and greater flow acceleration around the nose onto the upper surface. Theprecisereasonsforthehighercirculationarenotclear,althoughitisnotsolelyafunctionofpredicting thejetseparationlocationcorrectly, aspreviouslybelieved. Acurrenthypothesisisthatdifferencesinthepredicted characterandlocationofthejetsheetafteritseparatesfromtheairfoilareasignificantcontributingfactor. Acknowledgments ThefirstauthorwouldliketothankMs. ShellyJiang,whoservedasaninternatNASALangleyduringsummer 2009 and helped to implement and test the SSTRC model. The second author would like to thank Dr. Seonghyeon HahnofStanfordUniversityforprovidingtheLEScode. ThisworkwasfundedbytheSubsonicFixedWingProject oftheFundamentalAerodynamicsProgram. 10of29 AmericanInstituteofAeronauticsandAstronautics

Description:
A numerical study over a nominally two-dimensional circulation control airfoil is performed . blowing rates they tend to predict jet separation too late The EASM-ko model30 is an explicit algebraic stress model in k-ω formulation. This model makes use of a non-linear constitutive relation rather.
See more

The list of books you might like

Most books are stored in the elastic cloud where traffic is expensive. For this reason, we have a limit on daily download.