Professional Documents
Culture Documents
Turbulence Modeling For Beginners
Turbulence Modeling For Beginners
b y
TONYSAAD
UNIVERSITYOFTENNESSEESPACEINSTITUTE
tsaa d@utsi.ed u htt p:/ / j ed i .k n o w s . it
The purpose of this tiny guide is to summarize the basic concepts of turbulence
modeling and to a compile the fundamental turbulence models into one simple
framework.Intendedforthebeginner,noderivationsareincluded,unlessinsome
simple cases, as the focus is to present a balance between the physical
understandingandtheclosureequations.Ihopethismaterialwillbehelpful.
TABLEOFCONTENTS
Introduction........................................................................................................................ 3
First Order Models: ........................................................................................................... 6
Zero-Equation Models .............................................................................................................. 7
One-Equation Models: .............................................................................................................. 8
Two-Equation Models ( k ): .............................................................................................. 10
Acknowledgments............................................................................................................. 19
INTRODUCTION
Aturbulentflowfieldischaracterizedbyvelocityfluctuationsinalldirectionsand
hasaninfinitenumberofscales(degreesoffreedom).SolvingtheNSequationsfora
turbulentflowisimpossiblebecausetheequationsareelliptic,nonlinear,coupled
(pressurevelocity, temperaturevelocity). The flow is three dimensional, chaotic,
diffusive, dissipative, and intermittent. The most important characteristic of a
turbulentflowistheinfinitenumberofscalessothatafullnumericalresolutionof
the flow requires the construction of a grid with a number of nodes that is
proportionaltoRe9/4.
ThegoverningequationsforaNewtonianfluidare
ConservationofMass
ui
+
= 0
t
xi
(1)
Conservationofmomentum
ui u j ui
+
=
x j
x j
u
i
x j
p
+ gi + sui
xi
(2)
Conservationofpassivescalars(givenascalar T )
c pT
c p u jT
x j
x j
T
k
x j
+ st
(3)
Sohowcanwesolvetheproblem?Oneofthesolutionsistoreducethenumberof
scales(frominfinityto1or2)byusingtheReynoldsdecomposition.Anyproperty
(whether a vector or a scalar) can be written as the sum of an average and a
fluctuation, i.e. = + where the capital letter denotes the average and the
lower case letter denotes the fluctuation of the property. Of course, this
decomposition will yield a set of equations governing the average flow field. The
newequationswillbeexactforanaverageflowfieldnotfortheexactturbulentflow
field. By an average flow field we mean that any property becomes constant over
time.TheresultofusingtheReynoldsdecompositionintheNSequationsiscalled
theRANSorReynoldsAveragedNavierStokesEquations.Uponsubstitutionofthe
Reynolds decomposition (for each variable, we substitute the corresponding
decomposition)weobtainthefollowingRANSequations:
ConservationofMass
U i
+
= 0
xi
(4)
Conservationofmomentum
U i U jU i
+
=
x j
x j
_____
U i
P
ui u j
+ Sui
x
x
j
i
(5)
Conservationofpassivescalars(givenascalar T )
c pT
c pU jT
x j
x j
_____
T
c p u j t + St
k
x j
(6)
Note:aspecialpropertyoftheReynoldsdecompositionisthattheaverageofthefluctuating
componentisidenticallyzero,afactthatisusedinthederivationoftheaboveequations.
However,byusingtheReynoldsdecomposition,therearenewunknownsthatwere
_____
introduced such as the turbulent stresses ui u j and turbulent fluxes (where the
overbar denotes an average) and therefore, the RANS equations describe an open
setofequations.Theneedforadditionalequationstomodelthenewunknownsis
calledTurbulenceModeling.
Wenowhave9additionalunknowns(6Reynoldsstressesand3turbulentfluxes).
Intotal,forthesimplestturbulentflow(includingthetransportofascalarpassive
scalar,e.g.temperaturewhenheattransferisinvolved)there14unknowns!
A straight forward method to model the additional unknowns is to develop new
PDEs for each term by using the original set of the NS equations (multiplying the
momentum equations to produce the turbulent stresses). However, the problem
with this procedure is that it will introduce new correlations for the unknowns
(triplecorrelations)andsoon.Wethenmightthinkofdevelopingnewequationsfor
the triple correlations, nevertheless, we will end up with quadruple correlations
andsoonAnalternativeapproachistousethePDEsfortheturbulentstressesand
fluxes as a guide to modeling. The turbulent models are as follows, in order of
increasingcomplexity:
Algebraic(zeroequation)models:mixinglength(firstordermodel)
Oneequationmodels:kmodel,tmodel(firstordermodel)
Twoequationmodels:k,kkl,k2,lowRek(firstordermodel)
Algebraicstressmodels:ASM(secondordermodel)
Reynoldsstressmodels:RSM(secondordermodel)
ZeroEquationModels
OneEquationModels
FirstOrderModels
TwoEquationModels
AlgebraicStressModels
SecondOrderModels
ReynoldsStressModels
FIRSTORDERMODELS
First order models are based on the analogy between laminar and turbulent flow.
They are also called Eddy Viscosity Models (EVM). The idea is that the average
turbulent flow field is similar to the corresponding laminar flow. This analogy is
illustratedasfollows
u u j 2
u j
ij = i +
ij
x j
x j xi 3
Laminar FLow:
q = k T
i c x
p
i
t
U U j
ij = ui u j = t i +
xi
x j
Turbulent Flow:
q t = u t = kt T
i
i
c p xi
(7)
2
ij k
3
whichisreferredtoasthegeneralizedBoussinesqhypothesis.
Notethat:
t = Turbulent Viscosity
k = Turbulent Kinetic Energy
ZEROEQUATIONMODELS
Inzeroequationmodels,asthenamedesignates,wehavenoPDEthatdescribesthe
transportoftheturbulentstressesandfluxes.Asimplealgebraicrelationisusedto
close the problem. Based on the mixing length theory, which is the length over
which there is high interaction of vortices in a turbulent flow field, dimensional
analysisisusedtoshowthat:
t =
dU
t
lu = lm lm
dy
(8)
lm isdeterminedexperimentally.Forboundarylayers,wehave
lm = y for y <
lm = for y
(9)
and
kt = t
cp
Prt
(10)
Equations(8)through(10)arethenusedinthelaminarturbulentanalogyandthen
backintotheoriginalRANSequations.
ONEEQUATIONMODELS
In oneequation models, a PDE is derived for the turbulent kinetic energy and the
unknowns (turbulent viscosity and conduction coefficient) are expressed as a
functionoftheturbulentkineticenergyas:
K=
1 2 2
u + v + w2
2
(11)
We also make use of the fact that t lu but in this case, the velocity scales is
proportionaltothesquarerootofthekineticenergy(unliketheabovecasewhereu
wasproportionaltothegradientofvelocity).Therefore,wehave:
t = C K lm
kt = t
cp
(12)
Prt
Nowthattheturbulentviscosityandturbulentconductioncoefficientareexpressed
in terms of the turbulent kinetic energy (therefore the turbulent stresses and
turbulent fluxes are functions of the kinetic energy), a PDE is developed for the
turbulentkineticenergy.
U i
K U j K
+
= ui u j
gi ui t ....................(Pk + Gk )
x j
x j
x j
1 2
K
ui u j + pu j
x j
2
U j U j
x j x j
.........(D k )
(13)
.....................................( k )
On the other hand, this equation introduces two new unknown correlations; the
turbulentandpressurediffusions(Dk)andthedissipationrates(k)whichneedto
bemodeled.Finally,weendupwiththefollowing:
K U j K
+
=
x j
x j
t
+
k
K
+ Pk + Gk
x j
(14)
Where
Pk = ui u j
U U j
U i
= t i +
x j
xi
x j
Gk = gi
U i
x
j
kt T
c p xi
(15)
(16)
Andfromdimensionalanalysis,weobtain
U j U j
x j x j
K 3/ 2
K 3/ 2
= Cd
l
lm
(17)
TWOEQUATIONMODELS( k )
In the twoequation models, we develop two PDEs: one for the turbulent kinetic
energyandonefortheturbulentdissipationrate.ThePDEfortheturbulentkinetic
energy is already given by Eq.(13), however, the expression for the turbulent or
eddy viscosity is different. So, the idea is to express the turbulent viscosity as a
functionofKandandthenderivePDEsforKand.
t ul = k
t = C
k2
kt = t
cp
1/ 2
k 3/ 2
(18)
Prt
Theequationfortheturbulentkineticenergyisrepeatedhereforconvenience
t
K U j K
+
=
+
x j
x j
k
Pk = ui u j
U U j
U i
= t i +
x j
xi
x j
Gk = gi
K
+ Pk + Gk
x j
U i
x
j
kt T
c p xi
(19)
(20)
(21)
Nowinsteadofmodeling,weshalldevelopanindependentPDEforitstransport.
Weobtain
U j
+
=
x j
x j
t
+
Pk = ui u j
+ C 1 ( Pk + C 3Gk ) C 2
K
K
x j
U U j
U i
= t i +
x j
xi
x j
U i
x
j
(22)
(23)
10
Gk = gi
kt T
c p xi
(24)
Theconstantsaredeterminedfromsimplebenchmarkexperiments.
C
C 1
C 2
C 3
Prt
0.09
1.0
1.3
1.44
1.92
01.0
0.70.9
11
SECONDORDERMODELS
Thecentralconceptofsecondordermodelsistomakedirectuseofthegoverning
equations for the second order moments (Reynolds stresses and turbulent fluxes)
instead of the questionable Boussinesq hypothesis. The motivation is to overcome
thelimitationsoffirstordermodelsindealingwiththeisotropyofturbulenceand
the extra strains. The overshoot of this approach is the large number of PDEs
inducedwhichinvolvemanyunknownorimpossibletofindcorrelations.Themost
famousmodelsaretheAlgebraicStressModel(ASM)andtheReynoldsStressModel
(RSM). The second order modeling approach shall be illustrated with the RSM
modelonly.
12
THESTANDARDREYNOLDSSTRESSMODEL(RSM)
The RSM involves the modeling of turbulent diffusion, pressure strain correlation
whichisthemostinvolvedpartoftheRSM,andtheturbulentdissipationrate.The
RANSmomentumequationiswrittenas
ui u j
+ Dijt
xk
+ Pij + Gij + ij ij
ui u j U k ui u j
+
=
xk
xk
(25)
Where
Pij = ui uk
U j
Ui
u j uk
xk
xk
Gij = g j ui t gi u j t
(26)
(27)
Andthescalartransportequationis
ui t U j ui t
ui t
+
=
+ Ditt
xj
xj xj
(28)
+ Pit + Git + it it
Where
Ui
T
uk t
xk
xk
Git = gi t 2
(29)
(30)
13
ui u j
k
uk ul
xl
(31)
u jt
ut
k
+ u j uk i
ui uk
xk
xk
(32)
Dijt = Cs
and
Ditt = Cst
wheretheconstantsaredeterminedfromsimpleexperiments.
PRESSURESTRAINCORRELATIONMODELING
The role of the pressurestrain interaction is to redistribute the turbulent kinetic
energyoverthethreenormalstresses.Thegoverningequationfortheevolutionof
thisphenomenontakestheformofaLagrangianintegrodifferentialequation
1
p ui
=
x j 4
2u u u
U k ul ui g k t ui dV
k l
i
2
+
V xk xl x j xl xk x j + T xk x j x x + I S (33)
(34)
(35)
And
where
ij1 , it1 representtheturbulenceturbulenceinteraction,i.e.betweenfluctuations.
14
ij1 = C1 aij = C1 ui u j ij k
k
3
it1 = C1t ui t
= C1t ui t
(36)
(37)
turbulence
ij 2 = C2 Pij ij Pkk
3
(38)
it 2 = C2t Pit 2
(39)
ij 3 , it 3 represent the body force effects. Their modeling is similar to that of the
sheareffects
ij 3 = C3 Gij ij Gkk
3
(40)
it 3 = C3t Git
(41)
w
w
w
w
, ijw2 , it2
, ij3
,it3
Finally, ijw1 ,it1
represent the wall effects. The basic idea is that a
pressurewaveisreflectedatawallandthusaffectsthewholeflowfieldinanelliptic
manner
3
3
ijw1 = C1w uk ul nk nl ij ui uk n j nk u j uk ni nk Fn
2
2
k
3
3
ijw2 = C2 w kl 2 nk nl ij ik 2 n j nk jk 2 ni nk Fn
2
2
3
3
ijw3 = C3 w kl 3 nk nl ij ik 3 n j nk jk 3 ni nk Fn
2
2
(42)
and
15
itw1 = C1tw uk t ni nk Fn
k
w
it 2 = C2tw kt 2 ni nk Fn
w
it 3
(43)
= C3tw kt 3 ni nk Fn
with
C3 4 K 3 2
Fn =
yn
(44)
Where yn represents the distance from the wall to the concerned point in the
domain.
MODELINGOFTHETURBULENTDISSIPATIONRATE
Asimpleisotropicmodelisusedtomodel .
ij = ij
2
3
(45)
it = 0
(46)
Withthefollowingtransportequationforthedissipationrate
U j
=
+
x j
x j
Ui
k
1
(47)
C u j uk
+ ( C 1 Pk C 2 ) Pk = Pii = ui u j
xk k
xj
2
Where
Pk =
Ui
1
Pii = ui u j
2
xj
(48)
TheconstantsappearinginRSMequationsaregiveninthefollowingtable
16
Cs
0.5
0.3
0.3
Ct
C1
C2
C1tw
C2tw
C3tw
0.5
0.0
0.0
17
THEALGEBRAICSTRESSMODEL
Inthealgebraicstressmodel,twomainapproachescanbeundertaken.Inthefirst,
the transport of the turbulent stresses is assumed proportional to the turbulent
kineticenergy;whileinthesecond,convectiveanddiffusiveeffectsareassumedto
be negligible. Algebraic Stress models can only be used where convective and
diffusivefluxesarenegligible,i.e.sourcedominatedflows.
18
ACKNOWLEDGMENTS
I am indebted to my undergraduate advisor and tutor, Prof. Michel ElHayek at
Notre Dame University for introducing me to turbulence modeling. Most of this
tutorialisbasedonhisexplanationandpresentations.
19