• Aucun résultat trouvé

Combining a Mass Matrix formulation and a high order dissipation for the discretisation of turbulent flows

N/A
N/A
Protected

Academic year: 2023

Partager "Combining a Mass Matrix formulation and a high order dissipation for the discretisation of turbulent flows"

Copied!
32
0
0

Texte intégral

(1)

a p p o r t

d e r e c h e r c h e

9 -6 3 9 9 IS R N IN R IA /R R -- 7 0 7 9 -- F R + E N G

Thème NUM

Combining a Mass Matrix formulation and a high order dissipation for the discretisation of turbulent

flows

Anca Belme, Hilde Ouvrard

N° 7079

Octobre 2009

(2)
(3)

ThèmeNUMSystèmesnumériques

ProjetTropis

Rapportdereherhe n°7079 Otobre200923pages

Abstrat: This study fouses on analysis of a massmatrix shemewith high-ordernumerial

dissipation for thedisretisation ofthree-dimensional turbulent ows.We used amixed nite ele-

ment/nitevolumeformulationforthedisretisationofompressibleNavier-Stokesequations.VMS-

LESturbulenemodelisperformedtosimulatetheowaroundairularylinderat asubritial

Reynoldsnumber

Re

of3900.

Key-words: numerialshemes,turbulenesimulation,ompressible

INRIA,2004RoutedesLuioles,BP.93,06902Sophia-Antipolis,Frane

UniversityofMontpellierII,Frane

(4)

Résumé: Leprésentrapportestfoalisésurl'analysed'unshémanumériqueàmatriedemasse

et ave une dissipationnumérique d'ordreélévépour ladisrétisationdes éoulementsturbulents

tridimensionelles. Onautiliséuneformulationmixteéléments/volumesniespourladisrétisation

deséquations deNavier-Stokesompressible. UnmodèledeturbuleneVMS-LESest utilisépour

lasimulationd'unéoulementsubritiqueautour d'unylindreànombredeReynolds

Re

3900.

Mots-lés : shémasnumériques,simulationdelaturbulene,matriedemasse

(5)

by Hughes et al. in Ref. [4℄. The VMS-LES diers fundamentallyfrom the traditional LES in

a number of ways. In this approah, one does not lter the Navier-Stokes equations but uses

instead a variational projetion. This is an important dierene beause as performed in the

traditional LES, ltering works well with periodi boundary onditions but raises mathematial

issues in wall-bounded ows. The variational projetion avoids these issues. Furthermore, the

VMS-LES method a priori separates the sales that is, before the simulation is started. And

mostimportantly, itmodels theeet of theunresolved-salesonlyin the equationsrepresenting

the smallest resolved-sales, and not in the equations for the large sales. Consequently, in the

VMS-LES,energyisextratedfromtheneresolved-salesbyasubgridsale(SGS)eddy-visosity

model, but no energy is diretly extrated from the large strutures in the ow. The proposed

modelhasbeenimplementedinanumerialsolver(AERO)fortheNavier-Stokesequationsinthe

ase of ompressible ows and perfet Newtonian gases, based on a mixed nite-element/nite-

volume sheme formulated for unstrutured grids made of tetrahedralelements. Finite elements

(P1 type) and nite volumes are used to treat the diusive and onvetive uxes, respetively.

Conerning the VMS approah, the version proposed in Ref. [5℄ for ompressible ows and for

the partiular numerial method employed in AERO has been used here. The disretisation of

the onvetive uxes for the nite volume sheme used in the software AERO exhibits a lak of

auray in theaseof irregularmesh. Inorder tooveromethis problem, weinvestigateon this

work a new mass matrix sheme with high-order numerial dissipation for the disretization of

turbulent ows. Thepresentreport isorganisedasfollows: wereviewin Setion2amassmatrix

sheme forthe1Dadvetionequationand weanalysethestabilityofthis shemefor expliitand

impliittimeadvaningusingVonNeumannanalysis. Setion3ontainsthenumerialmethodin

3Dase appliedto theompressibleNavier-Stokesequations,whih hasbeenimplementedin our

CFDsoftwareandashortdesriptionoftheVMS-LESturbulenesimulationmodel. Finallyinthe

lastsetion,onegivessomeresultsofthesimulationofturbulentowaroundairularylinder.

2 Mass Matrix Central Dierening sheme for the advetion

equation

2.1 Spatial 1D MUSCL formulation

Letusrstonsidertheone-dimensionalsalaradvetionmodel

u t + c(u) x = 0

(1)

2.1.1 Spatialdisretisation

The nite-volume method is used forthe disretization in spae. Let

x j

,

1 ≤ j ≤ N

denote the

disretization points of the mesh. For eah disretization point, we state :

u j ≈ u(x j )

and we

denetheontrolell

C j

astheinterval

[x j1

2 , x j+ 1

2 ]

where

x j+ 1

2 = x j +x 2 j+1

.

We dene the unknown vetor

U = {u j }

aspoint approximation valuesof the funtion

u(x)

at

eahnode

j

ofthemesh.

(6)

Thetimeadvaningshemeiswritten :

U j,t + Ψ j (U ) = 0

wherethevetor

Ψ j (U )

isbuiltasfollows:

Ψ j (U ) = 1

∆x (Φ j+ 1

2 − Φ j− 1

2 ); Φ j+ 1

2 = Φ(u j+ 1 2

, u + j+ 1 2

);

(2)

where

u 1 2

,

u + 1 2

are integration values of

u

at boundaries of ontrol volume

C j

. The

Φ

's are

dened as:

Φ(u, v) = cu + cv 2 − δ

2 c(v − u)

where

δ

isaparameterontrolingthespatial dissipation,suhas

δ ∈ [0, 1]

.

Thereonstruted valuesare then givenby:

u j+ 1 2

= u j + 1 2 ∆u j+ 1 2

and

u + j+ 1 2

= u j − 1 2 ∆u + j+ 1 2

(samereonstrutionfor

u j− 1 2

and

u + j− 1 2

)wheretheslopes

∆u j+ 1 2

and

∆u + j+ 1 2

aredenedby:

∆u j+ 1 2

= (1 − β)(u j−1 − u j ) + β(u j − u j−1 ) +θ c (−u j−1 + 3u j − 3u j+1 + u j+2 ) +θ d (−u j − 2 + 3u j − 1 − 3u j + u j+1 )

and

∆u + j+ 1 2

= (1 − β)(u j+1 − u j ) + β(u j+2 − u j+1 ) +θ c (−u j−1 + 3u j − 3u j+1 + u j+2 ) +θ d (−u j + 3u j+1 − 3u j+2 + u j+3 )

It hasbeenshownthathoosing

β = 1 3

,

θ c = − 10 1

and

θ d = − 15 1

,thisshemebeomesfth-order

aurate. Thenumerialdissipationintroduedbythisshemeisthenmadeofsixth-orderderiva-

tivesandanbewrittenas:

D j (u) = D j+ 1

2 (u) − D j− 1

2 (u)

∆x

where

D j+ 1 2 (u) = δc

60 (−u j − 2 + 5u j − 1 − 10u j + 10u j+1 − 5u j+2 + u j+3 )

(3)

2.2 Mass Matrix Sheme with entral dierening

Theusualentral-dierenesthree-pointsshemeispenalizedbyadispersionleadingerror,see[1℄.

This errorisompensatedintroduingthenite-element

P 1

onsistentmassmatrix. That is, the

temporaltermofequation(1)isevaluatedbyFiniteElements. Thetimeadvaningisthuswritten

(7)

The nite-element

P 1

mass matrix M is dened as

M

ij = Z

Φ (i) Φ (j)

where

Φ (i)

is the basis

funtion forthenite-element

P1

formulation. Thatis:

Φ (i)

isapieewiselinearfuntionsuhas

Φ (i) (S i ) = 1

and

Φ (i) (S j ) = 0

forall

i 6= j

,where

S j

denotesthe

j

-thvertex. Misathree-diagonal matrixequalto:

M =

Three-diag

( 1 6 ∆x, 2

3 ∆x, 1 6 ∆x)

Oneobtainsthusforthe

j

-thomponentofvetor

M U (t)

:

(M U (t)) j = m j,j−1 U j−1,t + m j,j U j,t + m j,j+1 U j+1,t

where

m j,j−1 = 1/6∆x m j,j = 2/3∆x m j,j+1 = 1/6∆x .

Then oneanombine this disretisationof the temporal term with anumerialux forthe ap-

proximationoftheonvetifterm:

Φ j+ 1

2 = Φ(u j , u j+1 ) = c u j + u j+1

2

− δ 2 T j+ 1

2

Terms

c u j + u j+1

2

will ontributeto theentraldierenedux.

Aording to the Pasal triangle, a fth-order dierene evaluated between

j

and

j + 1

an be

written asfollows(

k > 0

):

T j+1/2 = kc(−u j − 2 + 5u j − 1 − 10u j + 10u j+1 − 5u j+2 + u j+3 )

Onehoosestheonstant

k

as:

k = 1

30

(5)

in ordertohavethesamelevelofdissipationasintheupwind ase,see(3).

Finally,startingfromexpression(2),wegetfortheoperator

Ψ j

dened:

Ψ j (U ) = c

2∆x ( k u j − 3

+ (−6k) u j−2

+ (−1 + 15k) u j − 1

+ (−20k) u j

(6)

+ (1 + 15)k u j+1

+ (−6k) u j+2

k u j+3 )

(8)

2.3 Time advaning stability

2.3.1 Expliittimestepping

Let us onsider a time integration of the system

M U t = AU

, with A the spatial approximation matrix and M the P1 nite element mass matrix. We an ombine the abovesheme with the

linearisedsix-stageRunge-Kuttasheme:

U (0) = U n

U (k) = U (0) + N −k+1 ∆t M 1 Ψ U (k 1)

, k = 1 . . . N U n+1 = U (6)

(7)

The stability study of the sheme is made with the lassial Fourieranalysis. Letus inlude in

equation(6)theFouriermode:

u ˆ n j = u k e ijθ k

where

θ k

isthefrequeneparameter.

Weobtain:

m θ

dˆ u n j

dt = − Ψ ˆ δ

where

m θ = 1 3 (2 + cos(θ))

Ψ ˆ δ = 2∆x c (R 3 cos(3θ) + R 2 cos(2θ) + R 1 cos(θ) + R 0 + iI 1 sin(θ))ˆ u n j

,with:

 

 

 

 

 

R 3 = 2δk R 2 = −12δk R 1 = 30δk R 0 = −20δk I 1 = 2

denoting

λ θ

thelinearoperatorsuhas

Ψ ˆ δ = −λ θ u ˆ n j

.

WeintroduetheCourantnumber

ν = c∆t ∆x

and theampliationfator

G θ = g(z θ )

.

z θ = λ m θ ∆t

θ

,

g

istheRK6 harateristipolynomial:

g(z) = 1 + z + z 2 2 + z 3

6 + z 4 24 + z 5

120 + z 6 720

Finaly,wehave:

 

z θ = − 2(2+cos(θ)) (z R θ + iz I θ )

z θ R = R 3 cos(3θ) + R 2 cos(2θ) + R 1 cos(θ) + R 0

z θ I = I 1 sin(θ)

Plotting thegainfuntion

Ga(ν) = max

θ ∈ [0,π] g(z θ )

, weandeterminize

ν max

, themaximum valueof

CFLnumberto obtainastableshemewith,that isthemaximumvalueof

ν

suhas

|g(z θ )| ≤ 1

.

ForaCFLsmallenough,thewholeof

z θ

omplexnumbersremainsinsidetheA-Stabilityregionof theRK6timeadvaningshemeasillustratedonFIG.(1).

(9)

−4 −3 −2 −1 0 1

−4

−3

−2

−1 0 1 2 3

Real part of z

θ

Imaginary part of z θ

Figure1: Stabilityanalysis: wedepitwiththeline theboundaryoftheregionforwhih

g(z) ≤ 1

and(inside)dashed

z θ

omplexnumbersfor

δ = 1

and

ν max = 1.11

2.3.2 Impliittime stepping

The purpose ofthis setion isthe stability analysis ofthe impliit sheme.This anbealso done

withFourieranalysis. Theomputation ismadeonthesameone-dimensionalsalaronservation

law.Letususetheimpliitshemea

δ

-shemeusingamass-matrixformulation:

T n δU n+1 = ∆t n Ψ(U n )

with

δU n+1 = U n+1 − U n

and

T n

istheimpliitmatrix.

Inaseofarstordersheme,

T n

representsthefollowingthree-diagonalmatrix:

T n =

diag

( 1 6 − ν, 2 3 + ν, 1 6 )

With theFourieranalysisweobtain:

t θ = 1 + ν(1 − cos(θ)) + 1 3 (2 + cos(θ)) + iν sin(θ)

Theampliationfatoristhengivenby:

G(∆t) = t θ +z t θ

θ

We are must interestedon the behaviorof theampliation fator when thetime step

∆t

tends

to

+∞

. The aim is to know if the builds ones shemes are preonditionated at rst ordre with satisfatory fators of onvergene. Let us denote

lim

∆t →∞ G(∆t) = f θ (δ)

. We are searhing the

Fourier'smodesfordierentshemeswithmaximisesthefuntions:

f θ (δ) = 1 − 1 4(1 − cos(θ))m θ

[(1 − cos(θ))z θ R + sin(θ)z θ I ] + i[ sin(θ)z R θ − (1 − cos(θ))z I θ

4m θ (1 − cos(θ)) ]

(10)

Thefuntion

f δ

aredepitedwiththreedierentvaluesof

δ

inFIG.2,3and4. Weanobservethat

ourimpliit shemeis inonditionallystable. Letus denotethat theoptimal valuethat minimise

thegainfuntion when

∆t

goesto

is

δ = 1

.

Figure 2: Ampliationfator,

∆t

goesto

,ase

δ = 1

Figure3: Ampliationfator,

∆t

goesto

,ase

δ = 0.5

(11)

Figure4: Ampliationfator,

∆t

goesto

,ase

δ = 0.3

3 Numerial Method 3D

3.1 Introdution

InthepresenthaptertheodeAERO,usedin thepresentstudy, isdesribed. Theodepermits

to solve the Euler equations, the Navier-Stokesequations for laminar ows and to use dierent

turbulenemodelsforRANS,LESandhybridRANS/LESapproahes. Theunknownquantitiesare

thedensity,theomponentsofthemomentumandthetotalenergyperunitvolume. AEROemploys

a mixed nite-volume/nite-element formulation for the spatial disretization of the equations.

Finite-volumes are used for the onvetive uxes and nite-elements (P1) for the diusive ones.

The resultingshemeis seond order aurate in spae. Theequations an be advaned in time

withexpliitlow-storageRunge-Kuttashemes. Alsoimpliittimeadvaningispossible,basedon

alinearisedmethod thatisseondorderaurateintime.

3.2 Set of equations

In the AERO ode the Navier-Stokes equations are numerially normalised with the following

referenequantities:

ˆ

L ref = ⇒

harateristilengthoftheow

ˆ

U ref = ⇒

veloityofthefree-streamow

ˆ

ρ ref = ⇒

densityofthefree-streamow

ˆ

µ ref = ⇒

moleularvisosityofthefree-streamow

(12)

Theowvariablesanbenormalisedwiththereferenequantitiesasfollows:

ρ = ρ

ρ ref u j = u j

U ref p = p p ref

E = E

ρ ref U ref 2 µ = µ µ ref

t = t L ref

U ref

.

Thenon-dimensionalformoftheNavier-Stokesequationsforthelaminarasearereportedin the

following:

∂ρ

∂t + ∂(ρ u j )

∂x j = 0

∂(ρ u i )

∂t + ∂ρ u i u j

∂x j = − ∂p

∂x i + 1 Re

∂σ ij

∂x j

∂(ρ E )

∂t + ∂(ρ E u j )

∂x j = − ∂(p u j )

∂x j + 1 Re

∂(u j σ ij )

∂x i − γ ReP r

∂x j h

µ E − 1

2 u j u j i

(8)

where the Reynolds number,

Re = U ref L ref /ν

, is based on the referenes quantities,

U ref

and

L ref

,thePrandltnumber,Pr,anbeassumedonstantforagasandequalto:

P r = C p µ k

and

γ = C p /C v

is the ratio between the spei heats at onstant pressure and volume. Also

theonstitutiveequationsfor thevisousstressesandthe stateequations may bewritten in non-

dimensionalform asfollows:

σ ij = − 2

3 µ ∂u k

∂x k δ ij

+ µ ∂u i

∂x j + ∂u j

∂x i

p = (γ − 1)ρ E − 1

2 u j u j

.

(9)

In orderto rewrite thegoverningequations in aompat form moresuitablefor thedisrete for-

mulation,thefollowingunknownvariablesaregroupedtogetherin the

W

vetor:

W = (ρ, ρu, ρv, ρw, ρE) T

.

Iftwoothervetors,

F

and

V

aredenedasfuntionof

W

,asfollows:

F =

 

ρu ρv ρw

ρu 2 + p ρuv ρuw ρuv ρv 2 + p ρvw ρuw ρvw ρw 2 + p

 

and

V =

 

 

0 0 0

σ xx σ yx σ zx

σ xy σ yy σ zy

σ xz σ yz σ zz

uσ xx + vσ xy + wσ xz − q x uσ xy + vσ yy + wσ yz − q y uσ xz + vσ yz + wσ zz − q z

 

 

(13)

∂ W

∂t + ∂

∂x j

F j ( W ) − 1 Re

∂x j

V j ( W , ∇ W ) = 0 .

(10)

It is importantto stressthat thevetors

F

and

V

are respetivelytheonvetiveuxesand the diusiveuxes.

3.3 Spatial disretization

Spatialdisretizationisbasedonamixednite-volume/nite-elementformulation. Anitevolume

upwind formulation is used for the treatment of the onvetive uxes while a lassial Galerkin

nite-elemententredapproximationisemployedforthediusiveterms.

Theomputationaldomain

isapproximatedbyapolygonaldomain

Ω h

. Thispolygonaldomain

isthendividedin

N t

tetrahedrialelements

T i

byastandardnite-elementtriangulationproess:

Ω h =

N t

[

i=1

T i .

(11)

Thesetofelements

T i

formsthegridusedinthenite-elementformulation. Thedualnite-volume gridanbebuiltstartingfromthetriangulationfollowingthemediansmethod.

Inthemediansmethodanite-volumeellisonstrutedaroundeahnode

a i

ofthetriangulation, dividingin4sub-tetrahedraeverytetrahedronhaving

a i

asavertexbymeansofthemedianplanes.

C i

isthe unionofthe resultingsub-tetrahedra having

a i

asavertexand theyhavethe following

property:

Ω h =

N c

[

i=1

C i .

(12)

where

N c

is thenumberofells,whihisequalto thenumberofthenodesofthetriangulation.

3.3.1 Convetive uxes

Indiatingthebasisfuntionsforthenite-volumeformulationasfollows:

ψ (i) (P ) =

1

if

P ∈ C i

0

otherwise

theGalerkinformulationfortheonvetiveuxesisobtainedbymultiplyingtheonvetiveterms

of(10)bythebasisfuntion

ψ (i)

,integratingonthedomain

Ω h

andusingthedivergenetheorem.

Inthiswaytheresultsare:

ZZ

h

∂F j

∂x j

ψ (i) x y = ZZ

C i

∂F j

∂x j Ω = Z

∂C i

F j n j σ

where

dΩ

,

and

n j

aretheelementarymeasureoftheell,ofitsboundaryandthe

jth

omponent

ofthenormalexternalto theell

C i

respetively.

Thetotalontributiontotheonvetiveuxesis:

(14)

X

j

Z

∂C ij

F( W , ~n) σ

where

j

are alltheneighbouringnodesof

i

,

F ( W , ~n) = F j ( W )n j

,

∂C ij

istheboundarybetween

ells

C i

and

C j

,and

~n

istheouternormaltotheell

C i

.

Thebasiomponentfortheapproximationoftheonvetiveuxesis theRoesheme,Ref. [15℄:

Z

∂C ij

F( W , ~n) σ ≃ Φ R (W i , W j , ~ν ij )

where

~ν ij = Z

∂C ij

~n σ

and

W k

isthesolutionvetoratthe

k

-thnodeofthedisretization.

IntheV6shemedeveloppedin[7℄and[6℄. Thenumerialuxes,

Φ R

,areevaluatedasfollows:

Φ R (W i , W j , ~ν ij ) = F(W i , ~ν ij ) + F(W j , ~ν ij )

| {z 2 }

centred

− γ s d R (W i , W j , ~ν ij )

| {z }

upwinding

where

γ s ∈ [0, 1]

isaparameterwhihdiretlyontrolstheupwindingoftheshemeand

d R (W i , W j , ~ν ij ) =    R(W i , W j , ~ν ij )    W j − W i

2 .

(13)

R

istheRoematrixandisdenedas:

R(W i , W j , ~ν ij ) = ∂F

∂ W ( W c , ν ij )

(14)

where

W c

istheRoeaveragebetween

W i

and

W j

.

The lassial Roesheme is obtainedas a partiular aseby imposing

γ s = 1

. The auray of

thisshemeisonly

1st

order. InordertoinreasetheorderofaurayoftheshemetheMUSCL

(Monotone Upwind Shemes for Conservation Laws) reonstrution method, introdued by Van

Leer,Ref. [19℄,isemployed. ThismethodexpressestheRoeuxasafuntionof

W ij

and

W ji

,the

extrapolatedvaluesof

W

attheinterfaebetweentwoells

C i

and

C j

:

Z

∂C ij

F( W , ~n) σ ≃ Φ R (W ij , W ji , ~ν ij )

where

W ij

and

W ji

aredenedasfollows:

W ij = W i + 1

2 ( ∇ ~ W ) ij · ij , ~ W ji = W j + 1

2 ( ∇ ~ W ) ji · ij . ~

Toestimatethegradients

( ∇ ~ W ) ij · ij ~

and

( ∇ ~ W ) ji · ij ~

,oneusedthese expressions,Ref. [6℄:

(15)

ξ c [( ∇ ~ W ) U ij · ij) ~ − 2( ∇ ~ W ) C ij · ij) + ( ~ ∇ ~ W ) D ij · ij)] + ~ ξ d [( ∇W ~ ) M · ij) ~ − 2( ∇W ~ ) i · ij) + ( ~ ∇W ~ ) D j · ij)] ~ , ( ∇ ~ W ) ji · ji ~ = (1 − β)( ∇ ~ W ) C ji · ij) + ~ β( ∇ ~ W ) U ji · ij) + ~

ξ c [( ∇ ~ W ) U ji · ij) ~ − 2( ∇ ~ W ) C ji · ij) + ( ~ ∇ ~ W ) D ji · ij)] + ~ ξ d [( ∇ ~ W ) M · ij) ~ − 2( ∇ ~ W ) i · ij) + ( ~ ∇ ~ W ) D j · ij ~ )] ,

where

( ∇ ~ W ) i

and

( ∇ ~ W ) j

are the nodal gradients at the nodes

i

and

j

respetively and are alulatedastheaverageofthegradientonthetetrahedra

T ∈ C i

,havingthe node

i

asavertex.

Forexamplefor

( ∇ ~ W ) i

weanwrite:

( ∇ ~ W ) i = 1 V ol(C i )

X

T ∈C i

V ol(T ) 3

X

k∈T

W k ∇Φ ~ (i,T) .

(15)

where

Φ (i,T )

is the P1 nite-element basis funtion. In the 3D ase, these funtions are so dened:

Φ (i) (P) =

1

if

P = S i

0

if

P = S j

,

i 6= j

( ∇ ~ W ) M · ij ~

, forthe 3D ase,is the gradientat thepoint

M

in Fig. 5and itis omputed by

interpolationofthenodalgradientvaluesatthenodesontainedinthefaeoppositetotheupwind

tetrahedron

T ij

.

( ∇ ~ W ) M · ij ~

isthe gradientat thepoint

M

in Fig.5and itis evaluated in the

samewayas

( ∇ ~ W ) M · ij ~

. Theoeients

β

,

ξ c

,

ξ d

are parametersthat ontrolthe ombination

offullyupwindandentredslopes. TheV6shemeisobtainedbyhoosingthemtohavethebest

aurayonartesianmeshes,Ref.[7℄:

β = 1/3, ξ c = −1/30, ξ d = −2/15 .

Thevariantoftheaboveshemealled mass-matrixwithentraldierening, sets

β

,

ξ c

and

ξ d

to

zeroandinvolvesatimederivativeevaluatedwiththenite-elementonsistentmassmatrix. This

massmatrixasinthe1Daseisexpressed intermsoftheusualP1testfuntions asfollows:

M ij = R

φ (i) φ (j)

As in Setion 2, we an ombine this time derivative with a ux. Weget thus the following

semi-disretizationequation:

∂ W i

∂t = ∆tM 1 X

j

Φ( W i , W j , ~ ν ij )

withthenumerialux denedby:

Φ(W i , W j , ~ ν ij ) = F(W i , ~ ν ij ) + F(W j , ~ ν ij )

2 − γ s

2 |R(W i , W j , ~ ν ij )| ∆W ij

(16)

where:

∆W ij = C(2( ∇W ~ ) M .~ ij − 5( ∇W ~ ) u ij .~ ij + 6( ∇W ~ ) c ij .~ ij − 5( ∇W ~ ) d ij .~ ij + 2( ∇W ~ ) M .~ ij) .

Followingthestudydoneforthe1Dadvetionequation(seeequation5),theonstant

C

issetto:

C = δ

60

(16)

A variant of this sheme is also investigated and onsists in using a projeteddissipation in

ij ~

:

|R(W i , W j , ~ ν ij )|

beomes

R(W i , W j , ν ~ ˜ ij )

in whih

ν ~ ˜ ij = ( ν ~ ij . k ij ij ~ ~ k ) k ij ij ~ ~ k

This optionhas been

Figure5: Skethofpointsandelementsinvolvedintheomputationofgradient

investigatedfortheappliation ofaGaussianfuntion translationin setion5.1

3.3.2 Diusiveuxes

TheP1nite-elementbasisfuntion,

φ (i,T)

,restritedtothetetrahedron

T

isassumedtobeofunit

valueonthenode

i

andtovanishlinearlyattheremainingvertexesof

T

. TheGalerkinformulation forthediusivetermsisobtainedbymultiplyingthediusivetermsby

φ (i,T )

andintegratingover thedomain

Ω h

:

Z Z

Ω h

∂V j

∂x j

φ (i,T ) Ω = Z Z

T

∂V j

∂x j

φ (i,T ) Ω .

(17)

Integratingbypartstheright-handsideofequation(17)weobtain:

ZZ

T

∂V j

∂x j φ (i,T ) Ω = ZZ

T

∂(V j φ (i,T) )

∂x j Ω − ZZ

T

V j ∂φ (i,T)

∂x j Ω = Z

∂T

V j φ (i,T) n j σ − ZZ

T

V j

∂φ (i,T)

∂x j Ω .

(18)

(17)

X

T,i ∈ T

Z

∂T

V j φ (i,T ) n j σ − Z Z

T

V j ∂φ (i,T )

∂x j

=

− X

T,i ∈ T

Z Z

T

V j ∂φ (i,T)

∂x j Ω + Z

Γ h =∂Ω h

φ (i,T ) V j n j σ .

(19)

IntheP1formulationforthenite-elementmethod,thetestfuntions,

φ (i,T )

,arelinearfuntions

on theelement

T

and sotheirgradient isonstant. Moreover, in thevariational formulationthe unknown variables ontained in

W

are also approximated by their projetion on the P1 basis funtion. Forthese reasonstheintegralanbeevaluateddiretly.

3.4 Boundary onditions

Firstly, therealboundary

Γ

isapproximatedbyapolygonalboundary

Γ h

thatanbesplitin two

parts:

Γ h = Γ ∞ + Γ b

(20)

where the term

Γ ∞

represents the far-elds boundary and

Γ b

represents the body surfae. The

boundaryonditionsareset usingtheSteger-Warmingformulation([18℄)on

Γ ∞

andusing slipor

no-sliponditionson

Γ b

.

3.5 Time advaning

One the equations have been disretized in spae, the unknown of the problem is the solution

vetorat eah node of the disretization asa funtion of time,

W h (t)

. Consequently the spatial disretizationleadstoasetofordinarydierentialequationsintime:

d W h

dt + Ψ( W h ) = 0

(21)

where

Ψ i

isthetotalux,ontainingbothonvetiveanddiusiveterms,of

W h

throughthe

i

-th

ellboundarydividedbythevolumeoftheell.

3.5.1 Expliittimeadvaning

Intheexpliitasea

N

-steplow-stokageRunge-Kuttaalgorithmis usedforthedisretization of Eq.(21):

 

W (0) = W (n) ,

W (k) = W (0) + ∆t α k Ψ( W (k−1) ), k = 1, ... , N W (n+1) = W (N ) .

in whih the sux

h

has been omitted for sakeof simpliity. Dierent shemes anbeobtained

varyingthenumberofsteps,

N

,andtheoeients

α k

.

(18)

3.5.2 Impliittime advaning

Forthe impliit time advaning sheme in AERO the following seond order aurate bakward

diereneshemeisused:

α n+1 W (n+1) + α n W (n) + α (n − 1) W (n 1) + ∆t (n) Ψ( W (n+1) ) = 0

(22)

wheretheoeients

α n

anbeexpressedasfollows:

α n+1 = 1 + 2τ

1 + τ , α n = −1 − τ, α n − 1 = τ 2

1 + τ

(23)

where

∆t (n)

isthetimestepusedatthe

n

-thtimeiterationand

τ = ∆t (n)

∆t (n 1) .

(24)

Thenonlinearsystemobtainedanbelinearisedasfollows:

α n+1 W (n) + α n W (n) + α (n−1) W (n−1) + ∆t (n) Ψ( W (n) ) =

− h

α n+1 + δt (n) ∂Ψ

∂ W ( W (n) ) i

( W (n+1) − W (n) ).

(25)

Following thedefet-orretionapproah, thejaobians are evaluated using the1st order ux

sheme (fortheonvetivepart), whilethe expliituxesare omposed with 2ndorder auray.

TheresultinglinearsystemissolvedbyaShwarzmethod.

4 Variational Multisale approah for Large Eddy Simulation

AnewapproahtoLESbasedonavariationalmultisale(VMS)frameworkwasreentlyintrodued

inHughesetal. TheVMS-LESdiersfundamentallyfromthetraditionalLESinanumberofways.

Inthisnewapproah,onedoesnotltertheNavier-Stokesequationsbutusesinsteadavariational

projetion. Thisis animportant dierenebeauseasperformed in thetraditional LES,ltering

works aswellwith periodi boundaryonditions but raises mathematial issues in wall-bounded

ows. Thevariationalprojetionavoids thisissues. Furthermore,the VMS-LESmethod a priori

separatesthesalesthat is,before thesimulationisstarted. Andmostimportantly,itmodelsthe

eet of theunresolved-salesonly in theequationsrepresentingthe smallestresolved-sales,and

notintheequationsforthelargesales. Consequently,in theVMS-LES,energyisextratedfrom

the neresolved-salesby atraditionalmodelsuh asSmagorinskyeddyvisosity model, but no

energyisdiretlyextratedfromthelargestruturesinteow. Forthisreason,oneanreasonably

hopetoobtainabetterbehaviornear walls,and lessdissipationin thepreseneof largeoherent

strutures.

A less fundamental, yet noteworthy, dierene between the VMS-LES and traditional LES

methods isthattheVMS-LES approahleadstogoverningequationsthat arewrittenin termsof

theoriginal(orundeomposed)owvariablesandthemodeledeetoftheunresolved-salesonthe

(19)

intotheirlteredandutuatingpartsinthesubgrid-tensor,thenfaestheissuesofmodelingthe

subgrid-sales. In the VMS-LES approah, one does nothave to deompose into spae-averaged

and utuatingparts eahourenein theNavier-Stokesequationsofeah owvariable beause

thenal equationsareexpetedto beexpressedin termsoftheimportane asitanbeexploited

to bypassthemodeling ofsomeutuatingquantities,aswillbeillustratedhereforompressible

turbulentows.

Theinitial developement ofthe VMS-LES method foused oninompressible turbulent ows,

regular grids, and spetral disretisations where the separaton apriori of the salesis simpleto

ahieve.Forniteelementapproximations,ahierarhialbasisapproahandanalternativemethod

basedonellagglomerationwerereentlyproposedforseparatingaprioritheoarse-andne-sales.

Inmostases,theVMS-LESmethodwasappliedmainly tohomogeneousisotropiinompressible

turbulene, and reently to inompressible turbulent hannel ows, fo with it demonstrated an

improvementoverthetraditionalLESmethod.

Inthis Variational Multisale approah for Large Eddy Simulation (VMS-LES) approah the

owvariablesaredeomposed asfollows:

w i = w i

|{z}

LRS

+ w i

|{z}

SRS

+w i SGS

(26)

where

w i

are the largeresolved sales (LRS),

w i

are the small resolved sales (SRS) and

w i SGS

aretheunresolvedsales. ThisdeompositionisobtainedbyvariationalprojetionintheLRSand

SRSspaesrespetively. Inthepresentstudy,wefollowtheVMSapproahproposedinRef.[5℄for

thesimulationofompressibleturbulentowsthroughanitevolume/niteelementdisretization

onunstruturedtetrahedralgrids. If

ψ l

arethe

N

nite-volumebasisfuntionsand

φ l

the

N

nite-

elementbasisfontionsassoiatedtotheusedgrid,inordertoobtaintheVMSowdeomposition

inEq. (26),thenitedimensionalspaes

V F V

and

V F E

,respetivelyspannedby

ψ l

and

φ l

,anbe

in turndeomposedasfollows[5℄:

V F V = V F V

M V F V ; V F E = V F E

M V F E

(27)

in whih

L

denotesthediret sumand

V F V

and

V F V

arethe nitevolumespaes assoiatedto

thelargestand smallestresolvedsales,spanned bythebasis funtions

ψ l

and

ψ l

;

V F E

and

V F E

arethenite elementanalogous. InRef.[5℄aprojetoroperator

P

in theLRSspaeisdened by

spatialaverageonmaroellsinthefollowingway:

W = P ( W ) = X

k

 

V ol(C k ) X

jǫI k

V ol(C j ) X

jǫI k

ψ j

 

| {z }

ψ k

W k

(28)

(20)

fortheonvetiveterms,disretizedbynitevolumes,and:

W = P ( W ) = X

k

 

V ol(C k ) X

jǫI k

V ol(C j ) X

jǫI k

φ j

 

| {z }

φ k

W k

(29)

forthediusiveterms,disretizedbyniteelements. InbothEqs. (28)and(29),

I k = {j/C j ∈ C m(k) }

,

C m(k)

being the maro-ell ontaining the ell

C k

. The maro-ells are

obtained by a proess known as agglomeration [9℄. The basis funtions for the SRS spae are

learlyobtainedasfollows:

ψ l = ψ l − ψ l

and

φ l = φ l − φ l

.

A keyfeature of the VMS-lesapproahis that the modeled inuene of the unresolvedsales on

largeresolvedonesissettozero,andsotheSGSmodelisaddedonlytothesmallestresolvedsales

(whih modelsthedissipativeeetoftheunresolvedsalesonsmall resolvedones). Thisleadsto

thefollowingequationsaftersemi-disretizations[5℄.

Z

C i

∂ρ

∂t Ω + Z

∂C i

ρ~ V .~nΓ = 0 Z

C i

∂ρ~ V

∂t Ω + Z

∂C i

ρ~ V ⊗ V ~nΓ + ~ Z

∂C i

p~nΓ + 1

Re Z

σ∇Φ i Ω + 1 Re

Z

τ ∇Φ i Ω = 0 Z

C i

∂E

∂t Ω + Z

∂C i

(E + p) V .~nΓ + ~ Z

σ ~ V .∇Φ i Ω + γ

ReP r Z

∇e.∇Φ i Ω + γ ReP r t

Z

µ t ∇e .∇Φ i Ω = 0

(30)

where

e

denotes the internal energy (

E = e + 1 2 V ~ 2

) with

V ~ = (u 1 , u 2 , u 3 )

theveloity.

τ

is the

smallresolvedsalesSGS stressgibenby:

τ = µ t (2S ij − 2 3 S kk δ ij )

with

S ij = 1 2 ( ∂u

′ i

∂x j + ∂u

′ j

∂x i )

and

µ t

, thesmall resolved saleseddy visosity (whih depends on the

hosenSGSmodel).

Oneannotiethat thelaminar Navier-Stokesequations arereoveredbysubstituting

τ = 0

and

µ t = 0

in Eq. (30)aboveanthattheSGSmodelisreoveredbysubstituting

τ = τ

,

µ t = µ t

,

e = e

and

Φ i = Φ i

in theequations,where

τ

and

µ t

denotetheusualSGS stresstensorandSGS

eddyvisosity,respetively.

Moredetails aboutthis VMS-LESmethodologyanbefoundinRef. [5℄and[2℄.

(21)

Thegraphofagaussianisaharateristisymmetri"bellshapeurve" thatquiklyfallstowards

plus/minusinnity.

Forour2Dstudy,weonsider theadvetionequation:

∂ρ

∂t (x, y, t) + c ∂ρ

∂x i (x, y, t) = 0

usinganadvetionvetor

c = (0, 1)

andagaussianfuntion asinitialondition:

ρ(x, y, z, 0) = 1 + exp 150(x+0,3)

Theadvetionequation is solvedusing theMass Lumping V6sheme and theMass MatrixCen-

tral Dierening shemewith twooptions: projeteddissipation andnon projeteddissipationas

desribed in setion3.3. Realling thata"Mass lumping"shememeans that thetemporal term

oftheequationsistreatedbyFiniteVolumewhereasMass-Matrixshememeansatemporalterm

treatedbyFiniteElement. Weareinpartiularinterestedonthedissipationofthegaussianwhen

it istranslated usingdierentmeshes. The rstmesh is regularasshown FIG.6 andthe seond

oneisirregularwithstrongvariationsoftheloalmeshsize(seeFIG.9)

Figure6: Regularmesh

Intheregular ase, weplot respetivelyin FIG. 8and FIG. 7theGaussian translation using

theMassMatrixCentralDiereningshemeandtheMassLumpingV6sheme. Oneanobserve

that the translation is well predited. In the irregular ase, as shown FIG. 11, the translation

is well preditedwith the MassMatrix Central sheme and using theV6 sheme, onehavesome

perturbations,seeFIG.10.

(22)

Figure7: Gaussian translatedwiththeMassLumpingV6shemeonregularmesh

Figure 8: Gaussian translatedwiththeMass MatrixCentral Dierening sheme(with projeted

dissipation)onregularmesh

(23)

Figure9: Irregularmesh

Figure10: GaussiantranslatedwiththeMassLumpingV6shemeonirregularmesh

(24)

Figure11: GaussiantranslatedwiththeMassMatrixCentralDiereningsheme(withprojeted

dissipation)onirregularmesh

(25)

and atasubritial ReynoldsnumberRe

D

of3900 (Re

D = u ∞ D

ν

)basedon ylinderdiameter

D

andfree-steam veloity

u

. Theomputationaldomainasshownin FIG.12 is

−10 ≤ x/D ≤ 25

,

−20 ≤ y/D ≤ 20

and

−π/2 ≤ z/D ≤ π/2

where

x

,

y

and

z

denote thestreamwise,transverseand spanwisediretion respetively. Theharateristisofthedomainarethefollowing:

L i /D = 10, L 0 /D = 25, H y /D = 20

and

H z /D = π

Theylinderofunitdiameter isenteredon

(x, y ) = (0, 0)

.

Theow domain isdisretized byan unstruturedtetrahedralgrid whih onsists of approxi-

matively

2.9 × 10 5

nodes. Theaverageddistane ofthenearestpointtothe ylinderboundaryis

0.017D

whihorrespondsto

y + ≈ 3.31

.

Forthepurposeofthesesimulations,theSteger-Warmingonditionsareimposedattheinow

and outow as well as on the upper and lower surfae (

y = ± H y

). In the spanwise diretion

periodi boundaryonditionsis applied. Onthe ylindersurfaeno-slip boundary onditionsare

set.

Figure12: Computationaldomain

ToinvestigatetheinueneofnumerialshemesontheVMS-LESapproah,threesimulations

havebeenarriedoutusingtheWALEsubgrid-salemodel. Theharateristisofthesimulations

performedforthisstudyaresummarized onTAB.1. OneusestheRoe-Turkelsolveraspreondi-

tioning. Forstabilityreasons,themass-matrixshemehasbeenrunsettingthenumerialvisosity

(26)

parameter

γ

to its maximalvalue1. In thesamemanner, theprojeteddissipation optionis not presentedhere.

Simulation NumerialSheme Valueofonstant

C

Dissipation

γ

CFL

Simu1 MassMatrix+CentralDi.

1

30

Nonprojeted 1 20

Simu2 MassLumping+V6

1

30

Nonprojeted 1 20

Simu3 MassLumping+V6

1

30

Nonprojeted 0.3 20

Simu4 MassMatrix+Centraldierening

1

15

Nonprojeted 1 20

Table1: Simulationsusing VMS-LES approah withthe WALEsubgrid-sale model, onstant

C

dened insetion3.3byequation16

The CFL number has been hosen sothat avortex shedding yle is sampled in alittle less

than1000timesteps. Time-averagedvaluesandturbuleneparametersaresummarizedinTAB.2

andomparedto datafrom experimentsofNorberg,OngandWallaeandParnaudeauetal..

C d

denotesthemeandragoeient,

C d

and

C l

respetivelytherootmeansquarevaluesofthedrag andlift,

St

theStrouhalnumber,

U min

themean enterlinestreamwiseveloity,

Cp back

themean

bak-pressureoeientand

l r

themeanreirulationlength.

St C d C d C l U min Cp back l r

Simulations

Simu1 0.252 0.77 0.0157 0.0367 -0.20 -0.66 1.76

Simu2 0.213 1.08 0.0490 0.4912 -0.29 -1.09 0.83

Simu3 0.215 1.01 0.0639 0.5920 -0.29 -1.02 1.08

Simu4 0.215 1.02 0.0103 0.0303 -0.29 -0.94 1.12

Experiments

[11℄ 0.215

±

0.05 0.99

±

0.05 -0.24

±

0.1 -0.88

±

0.05

[12℄ 0.21

±

0.005 1.4

±

0.1

[13℄ -0.34 1.51

Table2: Flowparametersforthepresentsimulations

As shown in TAB. 2, for the samelevel of numerial dissipation (Simu1, Simu2 and Simu3),

the owparametersobtained withthe mass-matrixshemeare generallyless well preditedthan

thoseobtainedwiththelassialV6sheme. Inpartiularthemeandragisunder-estimatedwith

themass-matrixsheme,whihinvolvesanover-estimationofthereirulationlength. Comparing

Simu1andSimu4,oneannotethattointrodueanumerialdissipationtwiebiggersigniantly

improvetheresults.

FIG. 13showsa proleof time-averagedand

z

-averagedstreamwiseveloity. Onean deter-

minize from this plot the most dissipative sheme, that is the sheme produing themost short

reirulationlength. This assumptionis veriedonFIG. 14,showingthe pressuredistribution on

the ylinder surfae averaged in time on homogeneousz diretion. The most dissipative sheme

is indeed the one being the most far from the experimental data. From this twoplots, onean

onludethatthemass-matrixshemeisnotenoughdissipative. Thisanexplaintheenountered

robustness problem. Introduing a bigger dissipation level redue onsiderably the reirulation

length. Oneanalso note FIG. 14 thedisrepanies with theexperimental data, for

θ

beingbe-

(27)

2 4 6 8 10

−0.5 0 0.5

x

<u> Simu1

Simu2 Simu3 Simu 4

Experiment Parnaudeau et al.

Figure13: Time-averagedstreamwiseveloityontheenterlinediretion

tween

60

and

100

. Thisisprobablydueto theoarsenessoftheusedgrid.

0 20 40 60 80 100 120 140 160 180

−1.5

−1

−0.5 0 0.5 1

Angle θ (0 at stagnation point)

<Cp>

Simu1 Simu2 Simu3 Simu 4 Experiment

Figure 14: Time-averaged and z-averaged pressure distribution on the surfae of the ylinder,

experiment: Norberg

FIG.15displaysthetotalresolvedReynoldsstress

u′u′

in thewakeat

x = 1.54

andFIG.(16)

displaysthetotalresolvedReynoldsstress

v′v′

at

x = 1.54

. OneanobserveFIG.15thattheplot

is not symmetri using the mass-matrixsheme. With alowdissipation level, the better results

(28)

are obtainedwith theV6sheme. The behaviourof theMass-Matrixsheme isimprovedusinga

biggernumerialvisosity.

−2 0 −1 0 1 2

0.05 0.1 0.15 0.2 0.25

y

<u’u’>

Figure 15: Total resolved streamwise Reynolds stress

< u′u′ >

at

x = 1.54

, experiments: Par- naudeau et al.;

−−

: Simu1;

: Simu2;

...

: Simu3;

−.

: Simu4;

+

: PIV1 experiment;

×

: PIV2

experiment

−2 0 −1 0 1 2

0.1 0.2 0.3 0.4 0.5

y

<v’v’>

Figure 16: Total resolved streamwiseReynolds stress

< v′v′ >

at

x = 1.54

, experiments: Par- naudeau et al.;

−−

: Simu1;

: Simu2;

...

: Simu3;

−.

: Simu4;

+

: PIV1 experiment;

×

: PIV2

experiment

(29)

fortheMass-Matrixsheme(Simu4)givesabetteragreementwiththeexperiments.

6 Conlusion

WehavepresentedinthisworkarstinvestigationofanewMassMatrixsheme,inordertoobtain

abetterauraythanforMassLumpingFVshemeonirregularmeshes.

Thisnumeriswasinstalledinaparallel odeAEROofresearhandprodution.

Therstresultsarepromising. FortheGaussiantranslation,theMass-Matrixlearlyimprovedthe

preditionusinganirregularmesh. Wehavealsoshownthattheresultsobtainedforthesimulation

oftheowaroundairular ylinderareimprovedwith theMass-Matrixshemewhenasuitable

levelofnumerialdissipationishosen.

This rstwork ontheMass-Matrixshemeneedsto be ontinued byinvestigating thedissipation

term and the behavior of this model for the simulation of turbulent ows on strongly irregular

three-dimensionalmeshes.

Aknowlegments: WethankEriLamballaisforkindlysendingusdatarelatedto[13℄. CINES

andINRIA aregratefully aknowledgedforhavingprovided theomputationalresoures.

(30)

Referenes

[1℄ C.Debiez,Approximationet linéarisationdeoulementsaérodynamiques instationnaires,Ph.D

Thesis,UniversitédeNieSophia-Antipolis,Éoledotorale-Sienespourl

'ingénieur,(1996).

[2℄ C.Farhat,A.KajasekharanandB.Koobus,A dynami variationalmultisalemethodfor large

eddy simulations on unstrutured meshes, Computational Methods in Applied Mehanis and

Engineering,(2005),1668-1691.

[3℄ J. Hinze,Turbulene,MaGraw-Hill,NewYork,(1959).

[4℄ T.J.R.Hughes,L.MazzeiandK.E.Jansen.Largeeddysimulationandthevariationalmultisale

method,Comput. Vis.Si.,3,(2000),47-59.

[5℄ B. Koobus and C. Farhat, A variational multisale method for the large eddy simulation of

ompressible turbulent ows on unstrutured meshes-appliation to vortex shedding, Comput.

MethodsAppl. Meh. Eng.,193,(2004),1367-1383.

[6℄ B. Koobus, S. Wornom, S. Camarri, M.-V. Salvetti, A. Dervieux Nonlinear V6 shemes for

ompressible ow, Institut National de Reherhe en Informatique et Automatique,30 janvier

2008(2005),1668-1691.

[7℄ S. Camarri,M.V. Salvetti, B.Koobus,and A. Dervieux A low diusion MUSCL sheme for

LESon unstruturedgrids,Comput. Fluids,33,(2004),11011129.

[8℄ G.W.Jones,J.J.CinottaandR.W.Walker,Aerodynamiforesonastationaryandosillating

irular ylinder at high Reynolds numbers, National Aeronautis and Spae Administration,

NASATRR-300,(1969).

[9℄ M.H. Lallemand,H. SteveandA. Dervieux.Unstruturedmultigridding by volume agglomera-

tion: urrentstatus,Comput. Fluids,21,(1992),397-433.

[10℄ M.H. Lallemand, Shemas deentres multigrilles pour la resolution des equations d'Euler en

elementsnis,Ph.DThesis,UniversitédeProvene,CentreSaintCharles,(1988).

[11℄ C.Norberg,EetsofReynoldsNumberandLow-IntensityFree-SreamTurbuleneontheFlow

aroundaCirularCylinder, Publ.No.87/2,DepartmentofAppliedTermos. andFluidMeh.,

ChalmerUniversityof Tehnology,Gothenburg,Sweden,(1987).

[12℄ L.OngandJ.Wallae,Theveloityeldoftheturbulentverynearwakeofairularylinder,

Exp.in Fluids,vol.20,SpringerVerlagBerlin,(1996),441-453.

[13℄ P.Parnaudeau,J.Carlier,D. Heitz andE.LamballaisExperimentelandnumerial studies of

the ow overairular ylinder atReynoldsnumber3900, Phys.of Fluids,20,(2008).

[14℄ R.Peyret,Handbook of omputational uidmehanis,AademiPress,2004

[15℄ P. L. Roe, Approximate Riemann solvers, parameters, vetors and dierene shemes,

J.Comp.Phys.,43,(1981),357-372.

[16℄ S. Shmidt and F. Thiele, Comparison of numerial methods applied to the ow over wall-

mountedubes,Int.J.ofHeat andFluidFlow,23,(2002),330-339.

(31)

[18℄ J.L. Steger and R.F. Warming, Flux vetor splitting for the invisid gas dynami equations

withappliations tothe nite dierenemethods,J.Comp. Phys.,40,(1981),263-293.

[19℄ B. van Leer, Towards the ultimate onservative sheme. IV: A new approah to numerial

onvetion,J.Comp. Phys.,23,(1977),276-299.

[20℄ D.C. Wilox,Turbulenemodeling,DCW IndustriesIn.,(1993).

[21℄ M.M.Zdravkovih,Flow aroundirular ylinders,OxfordUniversityPress,(2003).

(32)

2004, route des Lucioles - BP 93 - 06902 Sophia Antipolis Cedex (France)

Unité de recherche INRIA Futurs : Parc Club Orsay Université - ZAC des Vignes 4, rue Jacques Monod - 91893 ORSAY Cedex (France)

Unité de recherche INRIA Lorraine : LORIA, Technopôle de Nancy-Brabois - Campus scientifique 615, rue du Jardin Botanique - BP 101 - 54602 Villers-lès-Nancy Cedex (France)

Unité de recherche INRIA Rennes : IRISA, Campus universitaire de Beaulieu - 35042 Rennes Cedex (France) Unité de recherche INRIA Rhône-Alpes : 655, avenue de l’Europe - 38334 Montbonnot Saint-Ismier (France) Unité de recherche INRIA Rocquencourt : Domaine de Voluceau - Rocquencourt - BP 105 - 78153 Le Chesnay Cedex (France)

Éditeur

INRIA - Domaine de Voluceau - Rocquencourt, BP 105 - 78153 Le Chesnay Cedex (France)

Références

Documents relatifs

Crois en toi, fais toi confiance, tu as tout ce qu’il faut pour réussir à apprendre de nouvelles

We overcome this central difficulty by providing a scenario for singularity-mediated turbulence based on the self-focusing non-linear Schr¨odinger equation. It avoids the

We now explore the connection between the intermittency study presented above and the statistics of the local energy transfer and viscous dissipation. Note that in our

In a phase time approach Hartman has shown that the tunneling time of wave packets is independent of the barrier length for opaque barriers (where kL &lt; 1 holds, with k the

Figure 5 shows the mass distribution of populations of DNDs and clusters of different generations evaluated within the model of discrete-size diamond nanospheres for

In this study we have shown that experimental factors such as TEM sample thickness, ADF detector inner angle, and strain at the atomic scale can influence the ADF-STEM image

• The conversions of the subcognitive questions to search engine queries may be cognitive, but the choice of which conversions to make may have been influenced by

To determine if modulating cell elasticity by changing the substrate rigidity produced similar effects as actin inhibitors on LPS response, murine macrophages or differentiated