• Aucun résultat trouvé

Filtered Wrinkled Flamelets model for Large-Eddy Simulation of turbulent premixed combustion

N/A
N/A
Protected

Academic year: 2021

Partager "Filtered Wrinkled Flamelets model for Large-Eddy Simulation of turbulent premixed combustion"

Copied!
17
0
0

Texte intégral

(1)

HAL Id: hal-02129890

https://hal.archives-ouvertes.fr/hal-02129890

Submitted on 7 Jul 2020

HAL is a multi-disciplinary open access archive for the deposit and dissemination of sci- entific research documents, whether they are pub- lished or not. The documents may come from teaching and research institutions in France or abroad, or from public or private research centers.

L’archive ouverte pluridisciplinaire HAL, est destinée au dépôt et à la diffusion de documents scientifiques de niveau recherche, publiés ou non, émanant des établissements d’enseignement et de recherche français ou étrangers, des laboratoires publics ou privés.

Filtered Wrinkled Flamelets model for Large-Eddy Simulation of turbulent premixed combustion

Renaud Mercier, Cédric Mehl, Benoit Fiorina, Vincent Moureau

To cite this version:

Renaud Mercier, Cédric Mehl, Benoit Fiorina, Vincent Moureau. Filtered Wrinkled Flamelets model

for Large-Eddy Simulation of turbulent premixed combustion. Combustion and Flame, Elsevier, 2019,

205, pp.93-108. �10.1016/j.combustflame.2019.03.025�. �hal-02129890�

(2)

Combustion and Flame 205 (2019) 93–108

ContentslistsavailableatScienceDirect

Combustion and Flame

journalhomepage:www.elsevier.com/locate/combustflame

Filtere d Wrinkle d Flamelets model for Large-Eddy Simulation of turbulent premixed combustion

Renaud Mercier

a,

, Cédric Mehl

b

, Benoît Fiorina

b

, Vincent Moureau

c

aSafran Tech, Modelling and Simulation, Rue des Jeunes Bois, Châteaufort, Magny-Les-Hameaux 78114, France

bLaboratoire EM2C, CNRS, CentraleSupélec, Université Paris-Saclay, 3 rue Joliot Curie, Gif-sur-Yvette 91190, France

cCORIA- CNRS UMR 6614-Normandie Université - Université et INSA de Rouen, Campus Universitaire du Madrillet, Saint Etienne du Rouvray, Rouen 76800, France

a rt i c l e i n f o

Article history:

Received 3 October 2017 Revised 12 December 2017 Accepted 18 March 2019 Available online 5 April 2019 Keywords:

Turbulent combustion Premixed flames Flame wrinkling Filtered Wrinkled Flame Large Eddy Simulation

a b s t r a c t

Modelsforcombustion LESbasedonageometricaldescriptionofthereactivelayer arewell suitedto capture the turbulentflame front displacementspeed,butdo not predict the filteredchemical flame structure.Thisarticleaimstodiscussandmodeltheimpactoftheflamesub-filterwrinklinglevelonthe speciesproduction,withafocusoncarbonmonoxideemission.Forthatpurpose,2-D wrinkledflames with asinusoidal pattern, whichincludedetailed chemistryeffects, aremanufactured. Three control- lingparameters areidentified: theflamefilter size,thesub-filterflame wrinklingand the numberof flamepatternscontainedwithinthesub-filtervolume.Thisnewflamearchetype,namedFilteredWrin- kled Flamelets(FWF),may be embeddedinvariouscombustion modeling frameworks.Inthe present paper,itisusedtobuild-upafilteredchemicallook-uptableinordertomodeltheunclosedtermsof thefilteredprogress variable equation.Aprioritests areconductedby analyzinganexistingturbulent premixedflamedatabase.Aposterioritestsconsistinmodelingtheswirlingbluff-bodystabilizedCam- bridgeflame.Resultsanalysisshowsthataccountingforsub-filterflamewrinklingonthechemicalflame structureismandatorytopredictintermediatespeciessuchasCO.

1. Introduction

Introductionofdetailedchemistryeffectsinnumericalsimula- tionofturbulentflamesisanultimateobjectiveforthecombustion modelingcommunity[1].Despitethegreatchallengetoovercome, this effort is mandatory to predict complex phenomena such as flame ignitionor extinctionandalso pollutantformation. Several methodshavebeenconductedtoaccountfordetailedchemistryin aLarge-EddySimulationcontext[2]foraCPUcost,whichremains reasonable, so that the simulation of practical industrial reactive systems is possible. Among them, geometrical formulations have beendesignedtocapturetheflamefrontpropagationandthusto representaccuratelythe flamedynamicsinpremixedorstratified combustionregimesinLES[3].

Examplesofexistingapproachesbasedonageometricalformu- lation are the G-equation [4–7], filtered one-dimensional flames [8–11] andthickened flame model[12–14].These approachesare designed to capture accurately the turbulent flame consumption speed inthe flamelet regimewhether ornot theflame thickness

Corresponding author.

E-mail address: renaud-c.mercier@safrangroup.com (R. Mercier).

and wrinkling are resolved at the LES sub-filter scale. The sub- filterscale flame wrinkling , generally modeled eitherby al- gebraic[15–17]ordynamical[18–20]procedures,iscoupledtothe reactiveflowequationssothatboththesourceandturbulentdif- fusion terms of chemical species balance equations (or progress variable) are enhanced by a factor . Consequently, the turbu- lent flame front propagates at a speed S=S0L, where S0L is the unstrained laminar flameconsumption speed. Arecent com- parative study, conducted on a turbulent stratified flame config- urationexperimentedatT.U.Darmstadt[21],indeedconfirmsthat geometrical-basednumericalapproachesperformwellwhenfocus- ingontheturbulentflamefrontposition[22].

The impact of the sub-filter scale flame wrinkling on the filteredflamechemical structurehashoweverneverbeenconsid- ered. Indeed, geometrical approaches forflamelet regime usually assume that the filtered flame thickness and chemical profiles arenot affected bysub-filter scaleflame wrinkling.This assump- tion has been questioned by Moureau et al. [11] who observed, throughthe analysisof theDNSof a turbulentpremixed swirled flame, that filtered quantities are also influenced by the surface of the flame contained at the sub-filter scale. Several families of modeling approaches implicitly account for the impact of sub-filterstate on the resolved quantitiessuch as(i) Linear-Eddy

(3)

94 R. Mercier, C. Mehl and B. Fiorina et al. / Combustion and Flame 205 (2019) 93–108

Fig. 1. Differences between thickened and filtered flame structures. Both filtered and thickened flames have identical thermal layer thicknesses. Legend : — Laminar detailed chemistry solution, — Detailed chemistry solution, filtered at a size = 2 mm, ••• F-TACLES solution..

model[23]whereempiricalmeanofperturbedstatesareused;(ii) PDF-based models were the sub-filter state is describedthrough a presumed [24] or transported PDF [25]. These approaches are howeverdifficulttoexpressinageometricalformalism.

The objective ofthe presentarticleis to modelthe impact of thesub-filter scale flamewrinkling on thechemical flame struc- tureandthereforeon the minorspeciesproductionand destruc- tion,whichis importantforpredictingpollutantemissions. A fo- cus here is made on the carbon monoxide (CO) species. The unclosedtermsoftheprogressvariableequationaremodeledfol- lowinganapproachsimilartotheFilteredTabulatedChemistryfor LES(F-TACLES)model[10],whereachemicallook-uptableisgen- eratedfromFilteredPlanarFlamelet(FPF)elementsthat arecom- puteda priori usinga 1-D flame solver includingdetailedchem- istryand complex transport. The major issueof this approach is to incorporate the flame wrinkling in the look-up table genera- tion. For that purpose, the standard planar flamelet assumption of F-TACLES is revisited by manufacturing a series of 2-D wrin- kledflames withanalytical wrinklingpatterns.The paperisorga- nizedasfollows.Section 2 presentsthea priori studyof theim- pact of sub-filter scale flame wrinkling on the filtered chemical flamestructure.Section3describestheproceduretogeneratease- riesof2-Dwrinkledflameswithanalyticalwrinklingpatterns.The explicit filtering of these manufactured solutions serves to com- puteadatabasethat relatesfilteredthermochemicalquantitiesto the flame wrinklingpatterns. A turbulent combustion model us- ingtheFilteredWrinkledFlamelets(FWF) isderivedinSection 5. Itsa priorivalidationagainst afilteredDNSdatabaseisthen con- ductedinSection6.Finally,LESoftheCambridgepremixedflame burner [26] is computed and compared to experimental data in Section7.

2. Aprioristudyoftheimpactofsub-filterscaleflame wrinklingonthefilteredchemicalflamestructure

Two scales are often unresolved in Large Eddy Simulation of turbulent premixed orstratified flames: the flame thickness and thesmall-scaleflamewrinkling.

Thefirstone,thelaminarflamethickness,isingeneralsmaller thanthegridsizeinmostpracticalmeshes[3].Toavoidtheunder- resolution of the reactive layer and therefore a mis-prediction of the flame propagation speed, Thickened Flame model forLES (TFLES)artificiallythickenstheflamefrontbyafactorF[12].TFLES doesnotreproducehoweversomespecificsub-gridscalephenom- ena for intermediate species. An alternative to thickened flame modelareFilteredPlanar Flamelet(FPF)based modelssuch asF- TACLES[10],wherereferenceflamelets are filtered,tabulated and storedinalook-uptable.

FilteredPlanarFlameletapproachisillustratedinFig.1bypost- processing the detailed chemistry solution of a laminar freely- propagating 1-D methane-air flame computed with the REGATH solver[27]usingthemechanismproposed byLindstedt[28]atan equivalenceratioφ=0.83andatmosphericpressure.Thereactive flame thickness, definedhereasthe full widthathalf maximum (FWHM) of theheat release,is δr=0.17mm. The thermalflame thickness,definedfromunburntgastemperatureTu andburntgas temperatureTb asδl0=(TbTu)/max(T), is δ0l =0.53mm. To mimic a referenceideal laminar LES flame solution, the detailed chemistry temperature andCO mass fractionprofiles are filtered byaGaussianfunctionofwidth.ProfilesoftemperatureT,nor- malizedprogressvariablereactionrateω˙YC/max˙YC)andCOmass fractionYCOareplottedinFig.1for=2mm.

Theeffectiveratior=11.2iswellrepresentativeofpractical reactive LES configurations, where the flame thicknessis smaller than the filterassociated to the flame [29].An advantage of the filtering formalism is the conservation of the mass of chemical species contained within the flame layer as discussed in [10]. Figure 1 presents an a priori reconstruction of the filtered tem- perature andCOmass fractionsextracted froma FPFlook-up ta- ble.Fora1-Dunstrainedlaminarflame(i.e.withoutanysub-filter wrinkling) F-TACLES recovers the filteredlaminar flame structure asexpected.

The second unresolved scales are flame wrinkling patterns smaller than the filter size. The impact of sub-filter scale flame wrinkling on the filtered flame propagation speed is well accounted for in F-TACLES models [10]. However this formula- tion assumes that the filtered chemical flame structure is not altered by sub-filter scale eddies (i.e. the flame is planar at the sub-filter scale). This assumption is challenged by the post- processing of turbulent premixed flame DNS data obtained by Moureau etal.[11]ofthePRECCINSTA combustionchamber [30]. The configuration, which features a plenum, a swirl-injector and a combustion chamber, is representative ofaeronautical combus- tiondevices.DNSflamedataarefilteredatasize =11.2δr.The Favre-filtered temperature and CO mass fraction fields are then conditionally-averaged on iso-contours ata givendistance to the filtered flame front position. This filtered flame front position is defined as the iso-contour c˜=0.8, where c˜is the Favre-filtered normalizedprogressvariable.Theconditionalaveragesareplotted inFig.2. Diamondandplus symbol linesshow referencesprofiles obtainedattwoaxial distancestotheburner head,z=1 cmand z=4 cm,respectively. While the firstaxial position is locatedat the anchoring ofthe flame,the second position is atits tip. The wrinklingoftheflameincreasesfurther downstreaminthecom- bustor, and the two axial positions correspond to two different sub-filterscaleflamewrinklingconditions:=1.2atz=1cm,

(4)

R. Mercier, C. Mehl and B. Fiorina et al. / Combustion and Flame 205 (2019) 93–108 95

Fig. 2. Filtered temperature and CO mass fraction profiles obtained by post-processing a DNS solution of the PRECCINSTA configuration. Legend : Filtered DNS solution ( = 1 . 2 ), +—+ Filtered DNS solution ( = 2 . 0 ), F-TACLES solution.

whereas =2 atz=4cm.Itcan benotedthat forthesecond position,wheretheflameishighlycorrugated,multiplefront-front interactions occur, which leadstoan increase ofthe temperature andtheCOmassfractiononthefreshgassideforX<−2mm.The filteredchemical flamestructureapriori predictedby F-TACLESis reconstructed by post-processing the DNS progress variable field with a filtered look-up table built fromPlanar Flamelets filtered at =11.2δr (FPF) as in the post-processed DNS. As discussed in [11],the sub-filter scale flamewrinklingtends to increase the filteredflamethickness.Itleadstoaslightunderestimationofthe thermallayerby F-TACLESasobservedinFig.2,whichisempha- sizedwhenincreases.Thediscrepanciesareconsiderablymore significantfortheCOmassfraction.Indeed,neglectingtheimpact ofsub-filterscaleflamewrinklingonthecarbonmonoxideproduc- tioncausesan50%underestimationofCOpeakinthiscase.

While the impact of sub-filter scale flame wrinkling on the flame thermal layer remains marginal, its effect on intermedi- ate species prediction is pronounced. In the following section, a methodology named Filtered Wrinkled Flamelets (FWF) is pro- posedtoincludesub-filterscalewrinklinginthefilteredflamelets formalism.

3. Modelingtheimpactofthesub-filterscalewrinklingonthe filteredflameproperties

The influence of sub-filter scale wrinkling on the chemical flame structure is in practice difficult to anticipate and describe analytically.Inthissection,manufacturedflamesolutionsareused tostudytherelationbetweenthefilteredflameproperties(suchas consumption speed,thickness, andspeciesprofiles)withthegeo- metricalpropertiesofthewrinkledflameatthesub-filterscale.

3.1. Problemmodel

The impact of the geometrical properties of a 2-D wrinkled flame on the filtered flame structure is analyzed. Assuming that the sub-filter flame is in the flamelet regime, the flame front is trackedbyaprogressvariablecequalto0and1infreshandburnt gases,respectively.A2-Dwrinkledflamepattern,showninFig.3, is manufactured by assuming that the coordinates (xc=c0,yc=c0) ofthe iso-linec=c0 are relatedbya sinusoidalfunction: xc=c0= Asin(2πyc=c0/Ly),withAthe amplitudeofthesinusoidaliso-line and Ly its wavelength. This flame patternis embedded in a 2-D square ofsize ,the flamefiltersize,to mimica sub-filterscale domain.Forsimplicity,theflameis notwrinkledinthespanwise direction. Themodelcouldbeeasily extendedwith3-D manufac- tured solutions. A 2-D approach is however preferred hereas it simplifiesthemodelformalismaswellasitsanalysisinafirstap- proach.

Fig. 3. Schematic view of sub-filter sinusoidal flame pattern with n = /L y= 2 . The mapping process of a thermochemical quantity onto a 1-D premixed laminar flame computed with detailed chemistry is also presented.

Thenumberofwavelengths containedwithinthe boxisgiven by n=/Ly.As discussed inthe BML analysis[31], 2n corre- spondstotheaveragenumberofflamecrossingsofaspecifiedline perunitdistance.Wesupposeherethatn isaninteger.Fora givenflamefiltersize,thesub-filterflamepatternisthenperi- odicandparametrizedbytwo parameters:n controlsthewave- length while A controlsthe amplitudeof the c=c0 iso-line. The sub-filter scale wrinkling is determined by the pair (A, n). For n=0orA=0theflameremainsplanarwhileforA>0,theflame iswrinkled assoonasn≥1.The possible rangeofvariation of bothAandn isdiscussedinthefollowing.

3.2.PhysicalrangeofvariationofmodelparametersAandn

Assumingaflameletregime,thesizeofthesmallesteddyhav- ingaturnovervelocitysufficienttowrinkletheflamefrontisthe GibsonlengthLG[32]:

LG=(S0l)3

=

S0

l

u

3

(1)

whereS0l is thelaminar flamespeed, u quantifiesthe sub-filter velocityfluctuationsand isthe kinetic energytransferrate, as- sumedherewithin theinertialrange.Theflame wrinklinglength scale Ly/2, illustrated in Fig. 3, should therefore be larger than LG. As discussed by [3,32] in the flamelet regime (i.e.thin flame regime),LGvariesbetweentheintegrallengthscaleLt andthesize ofthesmallesteddyηwhichislargerthantheflamethicknessδl0. Theminimalvalueoftheflamefrontwavelengthreads:

Lminy =2LG>2

δ

0l (2)

(5)

96 R. Mercier, C. Mehl and B. Fiorina et al. / Combustion and Flame 205 (2019) 93–108

Fig. 4. Representation of 2-D manufactured thermochemical variables for /δl0= 5 , = 3 and n = 2 . Top left: progress variable c . Top right: CO mass fraction Y CO. Bottom left: progress variable reaction rate ω˙ c. Bottom right: CH radical mass fraction Y CH.

Asn isassumedtobean integer,themaximalnumberofflame wrinklingpatternsnmax isthereforegivenby:

nmax =E

Lminy

(3)

whereE[r]istheintegerpartofr.

In the flamelet regime, the local flame front curvature κ has

also to remain smaller that a threshold value so that the flame structure is unaltered. Thus, the radius of curvature R=1 has

tobe largerthan thecut-off lengthscale δ whichisofthe order

ofthelaminarflamethickness[3].Astheflamefrontgeometryis modeledbyasinusoidalpatternofnormaln,theanalyticalexpres- sionforκ=∇·nreads:

κ

=A

2

π

Ly

2

|

sin(2

π

yc=c0/Ly)

|

[1+A2(2

π

/Ly)2cos2(2

π

yc=c0/Ly)]3/2 (4)

Themaximal valueofκ observedalongthesinusoidalflamefront

isobservedforyc=c0=Ly/4:

max

κ

(yc=c0)=A

2

π

Ly

2

(5)

TheflameletconditionR>δcombinedwithEq.(5)givesthemax-

imumpossibleamplitudeAmaxofthesinusoidalpattern:

Amax= L2y

4

π

2

δ

(6)

Equations (3)and(6)give the maximalphysicalvalue ofboth Aandn parameters sothat thewrinkledflamepatternremains within the flamelet regime. Based on this assumption, the wrin- kledflame structure is reconstructed from a 1-D flamelet inthe followingsection.

3.3. Estimatingfilteredquantitiesfromapresumedsub-filterflame pattern

Assuming that the flame is in the flamelet regime, the 2-D flame structure is manufactured from a 1-D laminar premixed flame computed with detailed chemistry. The shortest distance D(x,y) betweeneach point (x, y) contained inthe 2-D sub-filter domainandtheiso-linec=c0 isintroducedasillustratedinFig.3. The thin reaction zone position c=c0 is located at the maximal heat releasepoint ofthe flame.The D field is thenused to map anythermochemicalquantityusingthe1-Dflameletdata:

(x,y)=1D(D(x,y)), (7)

where 1D denotes the 1-D flamelet thermochemical quantities.

An exampleof 2-D manufactured fieldsis givenin Fig.4 forthe progressvariablec,itssourcetermω˙c, COmass fractionYCO and CHmassfractionYCH.

Filteredquantitiesaredefinedby:

(x,y)= +

−∞

+

−∞ (x,y)F2D(xx,yy)dxdy, (8)

where F2D is a 2-D box filterof size .By introducing F1D, the 1-D boxfilterofsize ,F2D(x,y)isdecomposedasF1D(x)F1D(y) andalsoreads:

(x,y)= +

−∞ F1D(xx)

+

−∞ (x,y)F1D(yy)dy

dx. (9)

AstheflamepatternisnLy-periodicatthesub-filterscaleinthe y-direction,thefilteringalongyreducestoanaveragingoperation overasinglewavelength:

(6)

R. Mercier, C. Mehl and B. Fiorina et al. / Combustion and Flame 205 (2019) 93–108 97

Fig. 5. Sub-filter scale wrinkling as a function of the normalized amplitude A/δl0for different n and normalized filter sizes /δl0. Legend: —Manufactured wrinkled flamelet integration ( Eq. (12 )); − −Infinitely thin flame assumption ( Eq. (13) ).

(x)= +

−∞ F1D(xx)L1

y

+L

y/2

−Ly/2 (x,y)dy

y(x)

dx. (10)

Therefore,anyfilteredquantityassociatedtotheperiodicwrinkled flamepatternvariesonlyalongxcoordinateandreads:

(x)= +

−∞ F1D(xx)y(x)dx. (11) 3.4. Linkbetweensub-filterscalepatternandwrinkling

Thesub-filterflamewrinklingisdefinedas:

=S S0l = 1

ρ

0S0l +

−∞

ω

˙c(x)dx, (12) wherethe filteredprogressvariablereaction rateω˙c iscomputed from Eq.(11).For a given LES filtersize , the presumed flame pattern and its wrinkling are parametrized by both n= /Ly and the amplitude A. An analytical relation between andξ=nA/isobtainedunderinfinitelythinflameassumption (0l →∞):

=

(2

πξ

)2+1

2

π

EI

2

π

,1(2

πξ

1)2+1

, (13)

where EI is the incomplete elliptic integral of the second kind.

However,inthegeneralcasewheretheflamethicknessisconsid- ered,anumericalintegrationisneededtocomputeEq.(12). 4. AnalysisofFilteredWrinkledFlameletconsistency

To illustrate the methodology, 2-D wrinkled flames are man- ufactured for a given pair (A, n) using 1D variables from a 1-D CH4-air flame computed at φ=0.83 using detailed chem- istry [28]. The progress variable is defined as c=Yc/Yceq, where Yc=YCO+YCO2+YH2O andeqsuperscriptdenotesequilibriumcon- ditions. Manufactured 2-D variables are then numerically fil- teredaccordingtoEq.(11)foragivenfilterwidth.

4.1. Evolutionof withA

Figure5shows asafunctionofA/δ0l fordifferentvaluesof

n and0l.Solidlinesareobtainedfromthenumericalintegra- tionofEq.(12),whereasdashedlinesrepresenttheanalyticalsolu- tionsgivenbyEq.(13)underinfinitelythinflameassumption.For a planarflame(A=0),theflamewrinkling equals1.WhenA increases, increasesasexpected.Thedependencyoftothe pair(A,n)isclearlyevidencedinFig.5as,foragivenamplitude Aandfiltersize,increaseswiththenumberofwavelengths

n.TheanalyticalmodelgivenbyEq.(13)isretrievedaslongasA andnremainsmall,whichisinlinewiththeboundsproposedin Eq.(6). Flamethermalandreactive layers areindeedoverlapping atthesinewavepeakswhenAreachesthecriticalvalueAmaxand thenumericallycomputedwrinklingratiothusdecreasescompare toinfinitelythinflamecase.Itisimportanttonotethatthemodel degeneratestowards DNSwhen→0: asthefiltersizevanishes theboundimposedonnduetotheflamecut-off impliesthatno analyticalwrinklingispossibleandhencetheflamestaysplanar.

Figure 5 also shows that, for a given and n, there is a uniquecorrespondencebetweenA and. Filteredvariablescan therefore be parametrized by (,n,) instead of (A,n,). Thisformulationispreferredforpractical reasons:manyaccurate LESmodelforthesubgridscaleflamewrinklingexistsintheliter- ature(unliketothesubgridflamewrinklingamplitude)

4.2.Effectofwrinklingonflamethickness

Filtered data are also post-processed to compute, for the set of parameters (,n,), the filtered flame thickness δc˜= 1/max(|∇c˜|). δc˜ is plottedin Fig. 6as a function of fordif- ferentvalues of n.A strong dependency ofthe flame thickness

δc˜ to is observedforany ofthe threefiltersizes considered.

When0l >2,thenumberofwavelengthsninfluences signif- icantlythefilteredflamethickness.Indeed,foragivenc˜de- creaseswhenn increases.Inthislast case,δc˜ tends tothefilter size because,fora given,the amplitudeofthe flamepat- ternAgetssmallercomparedto.Forsmallvaluesofn,Aneeds tobelarger toreachagivenlevelofwrinkling(seeFig.5) which leadsto anincreasing filteredflamethicknessδc˜.Thisbehavior is alsoillustrated by Fig.7 whichshowsfiltered progressvariablec andfilteredchemicalreactionrateω˙cfordifferentvaluesofl0, n and.Fora givenn,theflamethicknessincreasesdueto wrinkling(left plot). Larger wrinklingvalues also modify the fil- tered reaction rates profiles (right plot) to larger maximum val- ues.Foragivenwrinklingvalue,theflamethicknessdecreases whennincreasessincethesineamplitudeAreducesforconstant .

Thisanalysis showsthat fora given filtersize , the filtered flame structure dependson both andn. It means that isa priorinot sufficientto determinethefilteredflame structure, whichisalsoinfluenced bythegeometricalshapeoftheflameat thesub-filterscale.

4.3.Effectofwrinklingonsub-filterspeciesmass

Figure8showstheevolutionofthemassofCOmCOcontained in a sub-filter box of size as a function of the wrinkling oftheflame pattern.Toease theinterpretation,mCO()is nor- malizedbythemassofCOofthe planarflamemCO(=1).For

(7)

98 R. Mercier, C. Mehl and B. Fiorina et al. / Combustion and Flame 205 (2019) 93–108

Fig. 6. Comparison of the ratio δc˜/δl0as a function of the sub-filter scale wrinkling for different values of n and different normalized filter sizes /δ0l.

Fig. 7. Favre-filtered progress variable c as a function of dimensionless position x/δ0l (left) and filtered reaction rate ω˙ cas a function of c (right) for different values of wrinkling ratio and filter width /δl0. Legend: Solid lines: FWF n = 1 ; Dashed lines: FWF n = 2 .

Fig. 8. Normalized mass of CO in the sub-filter box as a function of the wrinkling ratio for different values of /δl0and n . Dashed line shows unitary line f()= .

(8)

R. Mercier, C. Mehl and B. Fiorina et al. / Combustion and Flame 205 (2019) 93–108 99

Fig. 9. Normalized mass of CH in the sub-filter box as a function of the wrinkling ratio for different values of /δ0l and n . Dashed line shows unitary line f()= .

a givenflame patternofwrinkling,mCO()iscomputedby integratingρYCOoverthesurface:

mCO()= /2

x=/2

y=0

ρ

(x,y)YCO(x,y)dydx (14)

Severalcurvesarepresentedfordifferentvaluesofl0 andn. As expected,themainobservationisthatmCOincreaseswith duetotheincreasedflamesurfacewithinthebox.mCOisrelatively independent from the value of n and is however much lower than the unitary line mCO()=mCO(=1) which corre- sponds tothe theoreticalvalue ofthe integral that wouldbe ob- tained if the COmass wasconcentrated on the sine pattern(i.e.

for an infinitelythinfront). Thisdeparture of mCO is mainly due to the non-zeroYCO value in the burntgases whichimplies that some fractionof COmass is located outsideof the filtervolume whenthemanufacturedwrinkledflameisconstructed.Thisisver- ifiedbycarryingoutthesameanalysisforCHradicalwhichhasa muchlower thickness(asseeninFig.4)andazerovalueinboth freshandburntgases.TheresultingCHmassmCH showninFig.9 isaccuratelyapproximatedbytheunitaryline.Theremaininggap increases with andisdueto profilethickness effectsin high curvature regions. In this case only, the following approximation thereforeapplies:

mCH()mCH(=1) (15)

In the more generalcaseof thickerspecies profiles compared to ortoa specieswithnon-zeromass fractionsinfreshor burnt gases, thefiltered speciesmassfractions cannotbe approximated bysimplerelationssuchasEq.(15).

4.4. ComparisonofFilteredWrinkledFlamestructureagainstDNS data

The approximations made to manufacture the Filtered Wrin- kledFlameletsneglectseveralphysicalphenomenasuch ascurva- ture or stretch effects. In order to estimate the impact of these assumptions, the FWF database is compared against DNS solu- tion [11,30] of a turbulent premixed flamepreviously introduced inSection2.ToenableadirectcomparisonbetweenDNSandthe FWFdatabase,thefilteredflamethicknessisnowdefinedas:

δ

0c˜ =1/

|∇

c˜

|

c˜=c˜0 (16) where c˜0 is a progressvariable iso-line value chosen tobe close to the region of maximal |∇c˜|. The sub-grid scale flame wrin-

klingiscomputedintheDNSas=|∇c|/|∇c|.Foragivenfilter

size,boththefilteredflamethicknessandthesub-gridscaleflame wrinklingarethenestimatedintheDNSsolutionsalongtheflame surfacedefined asc˜0 iso-line.The filtered flamethickness condi- tioned by the sub-grid scale flamewrinkling isshown in Fig.10 for three values offilter sizes: l0 = 2, 5 and10. The filtered flamethicknessdefinedatc˜0=0.5iscomputedfromthereference

DNSdata(solidplainline)andtheapproximateFilteredWrinkled Flameletdatabase(solidlinewithsymbols)fordifferentvaluesof n. The gray area represents the variability ofδc0˜0l for a given wrinklingvalue:theupperandlowerthinlinescorrespondto plusandminustheRMS,respectively.As thefilteredflamethick- nessisstronglydependentontheflamewrinklingintheDNSand onthenparameterintheFWFmodel,thisfiguregivesanindica- tiononthenparameterthatshouldbeusedintheFWFmodelto recovertheDNSfilteredflamethickness.Then valuethatshould be chosen dependson the non-dimensional filter size 0l: n should increase withthe filter size. Moreover, a singlevalue per filtersizewouldalreadygiveanacceptablemodelsince1≤n≤2 for0l ≤10fortheDNSconsideredhere.

The prediction of COfiltered reaction rate given by the FWF databaseis also assessed by comparisonwith the DNSdatabase.

Forthatpurpose,Fig.11illustratestheimpactofbothn and onthefilteredCOmassfractionprofileYCOforagivenfilterwidth 0l (left plot).YCO profile thickness andpeak value are clearly impactedbyboth parameters.The solidblue curve(=1) also corresponds to the original F-TACLES. Significant differenceswith the=1.5and=2curvesillustrates thegain ofthemodel improvement. CO filtered reaction rate ω˙CO extracted from FWF database is then compared to the post-processed DNS for both 0l =5 and10 (right plot). DNSmean andRMS valuesof ω˙CO

conditionedtoc are shownby blacklineandgrayfilledarea, re- spectively. The trends of the DNS mean profiles are well repro- ducedbytheFWFdatabase.TheFWFprofilesvariationsdueton andarealsoinlinewiththeRMSvaluesofthepost-processed DNS.Theslightdifferencesobservedarecertainlyduetomodeling hypothesis, which includethe sinusoidal shape ofthe flame sur- faceandthe canonicalunstrained 1-D laminarflameprofile used tomanufacturethewrinkledflamelets.ItconfirmsthattheFiltered WrinkledFlamesarchetypesare abletotacklethe impacts ofac- tualsub-filterscalewrinklingpatternsontheresolvedintermedi- atespeciesprofilesthroughnandparameters.

5. Closureofthefilteredprogressvariablebalanceequation usingFilteredWrinkledFlamelets(FWF)

5.1. FilteredWrinkledFlameletdatabase

Aseriesoffilteredwrinkledflameletsismanufacturedbyvary- ing A and for a given wavelengthparameter n. For each set ofgeometricalparameters,filteredthermochemicalvariablesare mappedinthefilteredprogressvariablec˜spaceasin[10].Asdis- cussedinSection 4.1,a setofparameters (A,) corresponds,for agivenn,toauniquevalueof.Thereforecanbemapped andtabulatedas:

=[c,,]n. (17)

Anillustrationof theFWFtable is showninFig. 7fora CH4- Air premixed flame at φ=0.83 filtered at =5δ0l (top right)

(9)

100 R. Mercier, C. Mehl and B. Fiorina et al. / Combustion and Flame 205 (2019) 93–108

Fig. 10. Comparison of δ0˜c/δl0computed for ˜ c 0= 0 . 5 as a function of the sub-filter scale wrinkling for different values of n and different normalized filter sizes /δl0. Legend: Solid lines with symbols: FWF database. Solid line: DNS Mean value. Filled area: plus and minus the standard deviation.

Fig. 11. Favre-filtered CO mass fraction Y COas a function of dimensionless position x/δl0(left) and filtered chemical reaction rate ω˙ COas a function of the filtered progress variable ˜ c (right) for different values of wrinkling ratio and filter width /δl0. Legend: Solid colored lines: FWF n = 1 ; Dashed colored lines: FWF n = 2 ; Solid black line: DNS Mean value. Filled area: plus and minus the standard deviation.

and =10δl0 (bottom right). The evolution along the c˜ coordi-

nate of the filtered progress variable reaction rate ω˙c is shown for different values of and n. In both figures, the profiles of ω˙c for =1 match as expected regardless of the value of n.Theyalsocorrespondtothefilteredlaminarreactionratethat wouldbeextractedfromthe1-Dfilteredpremixedflame.Itcorre- spondstothecasewheretheflameisplanaratthesub-filterscale.

However,when =1.4 and=2.5˙c differs dependingon n as discussed in Section 3.4. Thiseffect is exacerbated as increases.

5.2. Closureofthefilteredprogressvariableequation

InLES, the filteredprogressvariable c˜is governedby the fol- lowingequation:

ρ

c˜

t +

·

ρ

u˜c˜=

·(

ρ

Vc)+

·

ρ

(c˜u˜cu)+

ω

˙c, (18) where ρ is the filtered density, u˜ is the filtered velocity, Vc is the molecular diffusion velocity of the progress variable and ω˙c is the filtered progress variable reaction rate. The RHS terms of

(10)

R. Mercier, C. Mehl and B. Fiorina et al. / Combustion and Flame 205 (2019) 93–108 101

Eq.(18)arecomputedfromFilteredWrinkledFlamelets(FWF)in- troducedinSection3.Thefilteredchemicalsourcetermisdirectly estimatedfromEq.(11)as:

ω

˙c=+

−∞ F1D(xx)

ω

˙cy(x)dx, (19)

where thesuperscript representsthevalues extractedfromthe manufactured wrinkled flamelet. The filtered molecular diffusion termis closedasρVc=−Dmc˜whereDm isestimatedfromthe

manufacturedwrinkledflameletas:

Dm=−

ρ

Vc/

c˜. (20)

Thesub-filterscaleconvectionterm∇·ρ(c˜u˜cu)ismodeledas:

T =

·

ρ

(c˜u˜cu)=

ρ

0S

|∇

c˜

|

|∇

c

|

, (21)

whereρ0 andS arerespectively thefreshgas densityandcon- sumptionspeed of themanufactured flamelet filtered at com- puted from Eq. (12). This formulation ensures that the filtered progressvariableiso-surfacespropagateatSasshownin[10].For afixedvalueofn,thetermsDm,T andω˙carepre-computedand tabulatedasfunctionsofthefilteredprogressvariablec,theflame filtersizeandthewrinklingfactoraccordingtoEq.(17).The flamefiltersizeischosentoensurethatthefilteredflamefrontis well resolvedontheLESgrid[29,33].Thewrinklingfactorcanbe estimatedfromtheLESusingeitheratransport equation[34],al- gebraicmodels[12,15–17] ordynamical formulations[18–20].The filteredprogressvariabletransportequation closedwithtabulated FilteredWrinkledFlameletsthereforereads:

ρ

c˜

t +

·

ρ

c˜u˜ =

·

Dm[c,,]n

c˜

+T[c,,]n+

ω

˙c[c,,]n. (22) ThismodelingstrategyissimilartotheF-TACLESmodel[10]ex- cept that Filtered Wrinkled Flamelets (FWF) are used instead of FilteredPlanarFlamelets(FPF).Therefore,thewrinklingfactor becomesacoordinateoftheFWFdatabase-whichintegratesthe impactofwrinklingonthefilteredflamefeatures-insteadofmul- tiplyingtheRHStermsofEq.(22).

5.3. Estimationofthewavelengthparametern

The FWF framework assumes that representing the sub-filter scale flame wrinkling with a unique wavelength Ly is sufficient tocapturethecorrectfilteredchemical speciesdistribution atthe sub-filterscale.Thishypothesiswillbeassessedusingaprioriand aposteriorianalysisinSections6and7,respectively.Inthecontext ofrealisticmulti-scaleflamewrinkling,thewavelengthparameter n=/Ly representsthe averagenumber offlame crossingsper filterlength [35].Assumingan infinitelythinflamefront, Bray etal.[35]showthatn reads:

n=

| σ

y

|

2 (23)

where isthesub-filterflamesurfacedensitywhichexpresses =|∇c|=|∇c|.σyisthemeancosineangleoftheinstanta-

neousflamefrontwiththec=c0iso-surfaceandisassumedtobe anuniversalconstant[35,36]sothat|σy|≈0.5.Equation(23)may alsoberecastas:

n=C

|∇

(

ρ

c)

|

(24)

where C=0.25b with ρb is the burnt gases density. In this modelformulationgivenbyEq.(24),nisnotafreeparameteras itisrelatedto∇c)computedfromtheLESaswellasand, whicharealsotwoFWFtablecoordinates.Inthefollowing,ρcwill

becomputeddirectlyfromtheDNSdatabaseintheapriorianaly- sis(Section6)whereasitwillbetransportedfollowingEq.(18)in theaposteriorianalysis(Section7).

5.4.SynopticofFWFmodelworkflow

GivenanLESgridandburneroperatingconditions,thestep-by- stepprocedure togeneratea FWFtableanduseit ina turbulent premixedflameLESissummarizedhere:

1.Compute the1-D premixedflameatselectedoperatingcondi- tionsusingdetailedchemistry.

2.Based on the computed 1-D flame, generate a series of 2-D wrinkled flamelets varying sine amplitude A and wavelength n as detailed in Section 3.1. Knowing the grid cell size x, thesuggestedflamefiltersizeis=5x.

3.FilterthedifferentmappedvariablestocomputerT,T,Yk˙Yk, Dm,T,ω˙c.

4.Tabulatethesevariablesas=[c,,]n

5.PerformtheLESusingrT toclosetheequationofstate.Thefil- teredprogressvariableequationisclosedusingDmandT+ω˙c as diffusion term and reaction rate, respectively. Other op- tional quantities such asYk andω˙Yk can be accessed aspost- processingvariables.

Theimplementationofthe FWFmodelanditstabulationpro- cedurearevalidatedby computingaseriesof1-DfilteredCH4-air flames atφ=0.83for different filtersizes.In these 1-D sim- ulations, Eq. (22) is solved by assuming constant values of the sub-filter scale wrinkling . They are performedto verify that thefilteredflamefronteffectivelypropagatesatS=S0l what- ever the choice of the flame patternparameter n. The simula- tionresults,notdetailedhere,showedthattheresolutioncriterion /x≥5 was neededto ensure an errorof less than 3% on S. Thisconclusionwasalsodrawnforotherfilteredflamemodelsim- plementedintheYALES2code[10,29].

6. AprioricomparisonofFWFdistributionsagainstafiltered DNSdatabase

Anapriorivalidationofthemethodologyisperformedbychal- lenging the Filtered Density Function (FDF) reconstructed using FWF against the DNS data [11,30] introduced in Section 2. Ref- erenceprogress variable FDFare extractedfrom the PRECCINSTA database by using the procedure detailed in [11] and plotted in Fig.12fordifferentpairs(c,Sc),whereSc,theunmixednessfactor, readsSc=c2/(c(1c))andtwofiltersize0l =2and0l = 4.

TheFWFsub-filterFDFiscomputedbyfirstrecastingEq.(10)as follows:

ρ

= 1

0

+

−∞

F1D(xx) Ly

+L

y/2

−Ly/2

ρ

δ

(cC)dy

dxdC (25)

where ρ=ρ(x,y), c=c(x,y) and δ is the dirac distribution.

Then,astatisticalexpressionofisobtainedbyintroducingPFWF, theFDFassociatedtoFWF:

= 1

0 PFWF(C)dC (26)

where PFWF(C)= 1

ρ

+

−∞

F1D(xx) Ly

+L

y/2

Ly/2

ρδ

(cC)dy

dx (27)

Firstandsecond moments ofthe progressvariableare computed fromtheFWF-FDFasfollow:

c= 1

0

cPFWF(C)dC (28)

c2= 1

0 (cc)2PFWF(C)dC (29)

Références

Documents relatifs

The model constraints are both to ensure a correct flame propagation and to recover the chemical structure of the filtered flame under two possible situations : (1) the flame

As a result of the reduced reactivity under MILD combustion conditions, the chemical time scale increases and the strong interaction between chemistry reaction and mixing makes

Pour éclairer le lien qui existe entre la présence par grâce et l’union transformante, Edith Stein recourt dans la Science de la Croix à Thérèse qui insiste dans le

Unit´e de recherche INRIA Lorraine, Technopˆole de Nancy-Brabois, Campus scientifique, ` NANCY 615 rue du Jardin Botanique, BP 101, 54600 VILLERS LES Unit´e de recherche INRIA

Ce qui en train de se passer au moment où Pearltrees a commencé, et notamment via Wikipédia, via YouTube, quelques phénomènes comme ça, à cette époque-là,

For the original network, in the Simmelian method, when we use different m, n to get the new different results of the deleted redundant information network, we compare the new

Il n’est pas contradictoire de constater, d’une part, que la première demande- resse avait signalé à son chirurgien ne pas souhaiter une anesthésie péridurale et qu’elle

Download date: 28.. 104- 105) considère que les frais d’assistance technique, dans le cadre d’une expertise médicale en matière de sécurité sociale, doivent être pris en charge