Introduction k− k−ω SST RSTM References

Introduction to Turbulence Modeling

Syed Imran SHAH

6 janvier 2014

Syed Imran Shah 1/ 81

 Introduction k− k−ω SST RSTM References

Characterization of Turbulent Flows I

High Reynolds Number, Rotational (Vorticity), Shear, Unsteadiness,

Three-Dimensionality, Chaos etc. Can be either wall-bounded or free-shear
ows.

Syed Imran Shah 2/ 81

 Introduction k− k−ω SST RSTM References

Categories of Turbulent Flows I

Examples of Wall-Bounded Flows

Boundary Layers, Pipe Flows, Channel Flows, Wall Jets etc.

Examples of Free-Shear Flows

Jets (Round or Plane), Wakes, Shear (or Mixing) Layers etc.

Some Applications
Turbine Blades, Aircraft Wings/Fuselage, Helicopter Rotors, Ship Hulls, Jet
Engines, Automobiles, High-Speed Trains, Wind Turbines, Weather
Prediction etc.

Syed Imran Shah 3/ 81

 Introduction k− k−ω SST RSTM References

Nomenclature I

ν = Kinematic viscosity
µ = Dynamic viscosity
ρ = Fluid density
p = Instantaneous static pressure
ui = Instantaneous velocity in the i-direction
p = Mean static pressure
ui = Mean velocity in the i-direction
p0 = Fluctuating static pressure
u0i = Fluctuating component of the ui velocity
u0i = Mean uctuating velocity in the i-direction
02 02 02
k = Turbulent kinetic energy,
p (u + v + w )/2
uτ = Wall friction velocity, (τw /ρ)

Syed Imran Shah 4/ 81

 Introduction k− k−ω SST RSTM References

Reynolds-Averaged Navier-Stokes (RANS) I

In the indicial notation, the RANS for a constant viscosity and

incompressible (∂ρ/∂p = 0) uid ow with rst coecient of viscosity and
in cartesian coordinates are given as,

∂ui
=0 (1)
∂xi

∂ui ∂ui 1 ∂p ∂u0i u0j

+ uj = gi − + ν∇2 ui − (2)
∂t ∂xj ρ ∂xi ∂xj
−ρu0i u0j has the units of stress
The correlation
2 2 2 2
(N/m = kg.m/s .1/m = kg.m/s ) and is called as Reynolds stress. It is
a 2nd order tensor with nine components. In most ows, this tensor is
symmetric, i.e., −u0i u0j = −u0j u0i and therefore the total unknown
correlations reduce from nine to six.

Syed Imran Shah 5/ 81

 Introduction k− k−ω SST RSTM References

Task of Turbulence Modeling in RANS I

Closure Problem
For RANS of 3D ows,

I a total of 10 unknown variables,

I 4 are mean velocities u,v ,w and mean static pressure p,

I 6 unknown variables are components of Reynolds Stress Tensor (ui uj )
0 0

I With 1 continuity and 3 momentum equations, we have a total of 4

equations

I Unknowns > equations, therefore the system is not closed.

I Task : Devise approximations for the unknown variables in terms of

known ow parameters.

1 2
1
"Modeling" in US English and "Modelling" in British English
2
No Closure Problem for the Direct Numerical Simulation (DNS)
Syed Imran Shah 6/ 81
 Introduction k− k−ω SST RSTM References

Boussinesq Hypothesis (1877)[3] I

I First to introduce the concept of eddy viscosity (or turbulent viscosity)

I Analogy with shear stress of Newtonian Fluid

I Turbulent Shear Stress proportional to mean velocity gradient

I Linear relation between turbulent stresses and mean velocity gradient

∂ui ∂uj
−ρu0i u0j = µt ( + ) = 2µt Sij (3)
∂xj ∂xi
∂ui ∂uj
where Sij = 12 ( ∂xj
+ ∂xi ) is the mean rate-of-strain tensor.

Syed Imran Shah 7/ 81

 Introduction k− k−ω SST RSTM References

Boussinesq Hypothesis (1877) : Contd I

Characteristics of µt
I µt is apparent, turbulent or eddy viscosity

I µt is scalar but not a constant

I µ dependent upon uid, µt dependent upon ow

I Units of µt are same as µ(P a.s)

I µt is not homogeneous (e.g.,∂µt /∂y 6= 0)

I µt is assumed isotropic (µx (for u02 ) = µy (for v 02 ) = µz )

I µ is physical but µt is hypothetical

I µt > µ at high Reynolds numbers

I kinematic eddy viscosity νt = µt /ρ (m2 /s)

I µt or νt > 0 for RANS turbulence models

Syed Imran Shah 8/ 81

 Introduction k− k−ω SST RSTM References

Boussinesq Hypothesis(1877) : Contd I

Limitations of the Boussinesq Hypothesis

For an incompressible ow, the Reynolds normal stresses (ρu0i u0i ) like u02 ,
v 02 , and w02 are modelled as ;

−ρu0i u0i = µt Sii

∂ui
= µt (2 )
∂xi
∂u ∂v ∂w
= 2µt ( + + )
∂x ∂y ∂z
= 0

danger : Problem of tensorial consistence.

Syed Imran Shah 9/ 81

 Introduction k− k−ω SST RSTM References

Boussinesq Hypothesis(1877) : Contd II

Modied Relationship
2
−ρu0i u0j = − ρkδij + 2µt Sij (4)
3
1 0 0
wherek is the turbulent kinetic energy per unit mass given by
2 ui ui =
1 02 02 02
2 (u + v + w ) and δij is the kronecker delta with :
I δij = 0 for i 6= j (Reynolds Shear Stresses)

I δij = 1 for i = j (Reynolds Normal Stresses)

I δii = δ11 +δ22 +δ33 = 1+1+1 = 3.

So the extra term ensures the correct sum of the Reynolds Normal stresses
which is ρu0i u0i = 2k .

Syed Imran Shah 10/ 81

 Introduction k− k−ω SST RSTM References

Boussinesq Hypothesis(1877) : Contd I

Limitations of the Boussinesq Hypothesis

This hypothesis is mostly useful for the computation of Reynolds Shear
stress (u0 v 0 ) in Turbulent Boundary Layer equations but danger : Reynolds
Normal Stresses become equal (isotropic),

u02 = v 02 = w02 = 2/3k

which is not realistic (except in the isotropic turbulence) as most turbulent
ows are anisotropic with the usual relative magnitudes of,

u02 > w02 > v 02

with typical ratios of 1 : 0.6 : 0.4 in the logarithmic region of a turbulent
boundary layer over a at surface without external pressure gradient(s).

Syed Imran Shah 11/ 81

 Introduction k− k−ω SST RSTM References

KTH Sweden Flat Plate ZPG TBL DNS Data(2010)[15] I

DNS of ZPG FP TBL, Reθ = 2540

8

6
Reynolds Stresses+

<u′2>/u2τ
5
<v′2>/u2τ
4 2 2
<w′ >/uτ
3
−<u′v′>/u2τ
2

0
0 500 1000 1500
+
y

Syed Imran Shah 12/ 81

 Introduction k− k−ω SST RSTM References

Eddy Viscosity Models I

Algebraic Turbulence Models

Several recent and previously proposed models compute the Reynolds stress
tensor using the eddy viscosity through Boussinesq Hypothesis. Such
models are called Eddy Viscosity Models. Among these, several models
construct an algebraic relationship for the eddy viscosity. They are called
Algebraic Models or Zero-equation models and mainly include,

I Prandtl's Mixing-Length Model (1925)[13]

I Cebeci-Smith Model (1967)[17]

I Baldwin-Lomax Model (1978)[2]

Syed Imran Shah 13/ 81

 Introduction k− k−ω SST RSTM References

Prandtl's Mixing-Length Model (1925) I

I Models the Turbulent Viscosity

I Analogy with the kinetic theory of gases

I Mixing length (lm ) : analogy with Mean Free Path of molecules (lmf p )

2  ∂u  ∂u
 
−ρu0i u0j = ρlm  ( ) (5)
∂y ∂y
Reynolds shear stress −ρu0i u0j should change sign with ∂u/∂y thats why
∂u 2
modulus sign has been used. Use of ( ) in the above equation would
∂y
∂u
mask the change of sign. Comparing with −ρu0i u0j = µt , we have,
∂y
2  ∂u 
 
µt = ρlm (6)
∂y
 

Syed Imran Shah 14/ 81

 Introduction k− k−ω SST RSTM References

Prandtl's Mixing-Length Model (1925) : Contd I

Modulus sign has been used in the equation (6) because the eddy viscosity
is always positive (for this and two-equation models) even if the velocity
gradient ∂u/∂y becomes negative.

Limitations of The Mixing-Length Model

∂u
danger : If
∂y = 0, equation (6) implies µt = 0. This happens at the center
of pipes and channels, with wall-jets or a mean velocity prole of turbulent
boundary layer near separation (especially in 2D). This means that this
model can't be used for near separation or separated ows. However, this
model works better for free shear ows like jets, mixing layers and wakes. It
is almost completely ignored by commercial CFD softwares.

Syed Imran Shah 15/ 81

 Introduction k− k−ω SST RSTM References

Experimental Observations for Mixing Length I

Algebraic expressions for mixing-lengths of various ow categories.

Flow Type Mixing Length

Log Layer lm = κy
Mixing Layer lm = 0.071δ0.5
Plane Wake lm = 0.18δ0.5
Plane Jet lm = 0.09δ0.5
Round Jet lm = 0.075δ0.5
Pipe Flow lm /R = 0.14 − 0.08(1 − y/R)2 − 0.06(1 − y/R)4

Here κ=0.4, δ0.5 is half-width (of a jet, wake or mixing-layer) with y = δ0.5
when u(x, δ) = 0.5u(x, 0), R is the pipe radius and y is the wall-normal
distance.

Syed Imran Shah 16/ 81

 Introduction k− k−ω SST RSTM References

Mixing-Length for the Turbulent Boundary Layers I

Using the Prandtl's hypothesis lm = κy in the whole turbulent boundary

layer from y=0 (wall) to y=δ (outer edge) gives poor results because
lm = κy is strictly valid in the log layer. For the inner layer, Van Driest
(1956) proposed that the mixing-length be multiplied by a damping
function as wall is approached. Thus for the internal layer,

+ /A+
lm = κy(1 − e−y ) (7)

where y + = yuτ /ν and for zero pressure gradient turbulent boundary layer
A+ = 26. Damping eects are reduced for y + > A+ . As for the external
layer, lm 6= κy and experiments show that lm ∼ constant, so that,

lm = Co δ (8)

Syed Imran Shah 17/ 81

 Introduction k− k−ω SST RSTM References

Mixing-Length for the Turbulent Boundary Layers II

for the outer layer. Here Co ' 0.085 (varies between 0.04 and 0.09 for
y/δ > 0.2) and δ is the boundary layer thickness. The mixing-length model
for the whole turbulent boundary layer can be written as,

+ /A+
lm = min(κy, Co δ)(1 − e−y ) (9)

i.e., the minimum of κy or Co δ is multiplied with the damping function.

Syed Imran Shah 18/ 81

 Introduction k− k−ω SST RSTM References

Turbulence Models with Transport equations I

I These include eddy-viscosity and Reynolds Stress Transport Models

(RSTM or RSM).

I Transport equation is a rate of change equation which incorporates

the ow history eects in the eddy viscosity.

I The number of equations used in a model depends on the number of

transported variables.

I When one transport equation is employed, it is referred to as

one-equation model. When two transport equations are used, it is
known as two-equation turbulence model and so on.

I Examples include Spalart-Allmaras (1 equation), k − , k − ω , SST ,

k − τ , k − φ, ASM (2 equations), V2−f (4 equations),
RSM/RSTM (7 equations : 6 Reynolds Stresses and 1 turbulent
dissipation) while not using any scalar (temperature, enthalpy, mass
fraction etc) transport equation.

Syed Imran Shah 19/ 81

 Introduction k− k−ω SST RSTM References

Turbulence Models with Transport equations II

I In k− and k−ω models, the transported variables k,  and ω are

used to determine the eddy viscosity µt .
I In most two-equation models, one transported variable represents the
velocity scale and other the length scale assuming that one length and
one velocity scale are enough to characterize turbulence.

I The transported variables must be directly or indirectly specied at the

inlet of computational domain like k and  for the k− model or the
0 0
six components of Reynolds stress tensor -ρui uj and turbulent
dissipation for the RSTM/RSM model.

Syed Imran Shah 20/ 81

 Introduction k− k−ω SST RSTM References

k −  Models I

The purpose of these models is to predict an eddy viscosity. Units of eddy

viscosity νt are m2 /s. Dimensionally,
νt ∼ length × velocity
or,

νt = CLn V n (10)

Therefore, for the eddy viscosity formulation we need to know at least one
length scale and one velocity scale. Here C is a dimensionless constant of
n
proportionality, L is the length scale (to be specied) and Vn is the
velocity scale (to be specied) with n as exponent.

Syed Imran Shah 21/ 81

 Introduction k− k−ω SST RSTM References

k −  Models : Assumptions Used I

I Turbulence is in local equilibrium,i.e., the production rate of k =

dissipation rate  of k (conversion into heat/internal energy)
ut lt νt k2
I Turbulent Reynolds number ReT = ν = ν (= ν for k− model) is
high, which implies isotropy and homogeneity of small dissipative
structures

I Eddy viscosity is positive and isotropic

By denition the rate of dissipation of turbulent kinetic energy per unit

mass is given as,

∂u0 ∂u0i ∂u0

 
=ν i + k 0 S0
= 2νSij ij (11)
∂xk ∂xk ∂xi
0 = 1( i +∂u0 ∂u0j
where Sij 2 ∂xj ∂xi ).

Syed Imran Shah 22/ 81

 Introduction k− k−ω SST RSTM References

k −  Models : Assumptions Used I

If we assume that small dissipatives scales (or structures) of ow are
homogeneous and isotropic, then,

∂u0i ∂u0i
p = ν( ) (12)
∂xj ∂xj
This last form is used in the two-equation models of turbulence. This form
itself is only valid for an incompressible homogeneous turbulent ow
(sometimes called pseudo-dissipation) but when used in the k -transport
equation the combined terms of dissipation rate and viscous diusion also
represent anisotropic eects. Expanding the above equation :

p ∂u0 2 ∂u0 2 ∂u0 2 ∂v 0 2 ∂v 0 2 ∂v 0 2

=( ) +( ) +( ) +( ) +( ) +( )
ν ∂x ∂y ∂z ∂x ∂y ∂z
(13)
∂w0 2 ∂w0 2 ∂w0 2
+( ) +( ) +( )
∂x ∂y ∂z
Syed Imran Shah 23/ 81
 Introduction k− k−ω SST RSTM References

k −  Models : Eddy Viscosity I

At high Reynolds number, the rate of dissipation  and production of

turbulence Pk are of a similar order of magnitude in boundary layers with
local equilibrium, therefore,

νt  = νt P k
∂u
= νt (−u0 v 0 )
∂y
∂u 0 0
= −νt ( )u v
∂y
= −u0 v 0 (−u0 v 0 )
2
= (u0 v 0 ) ≈ 0.09k 2 (14)

Syed Imran Shah 24/ 81

 Introduction k− k−ω SST RSTM References

k −  Models : Eddy Viscosity II

∂u
Here Pk = −u0 v 0 is the production rate of turbulent kinetic energy (in
∂y
2D). The last equality in the above equation stems from the experimental
observation that stress-intensity ratio −u0 v 0 /k ≈ 0.3 in the log layer of the
turbulent boundary layer.
Rearranging,

νt ≈ 0.09k 2 / (15)

This formula is usually written,

νt = Cµ k 2 / (16)

where the usual value of the coecient Cµ is 0.09. As k,  and Cµ are

always positive, νt is also always positive. Reynolds stress tensor is then
computed using the Boussinesq Hypothesis,

Syed Imran Shah 25/ 81

 Introduction k− k−ω SST RSTM References

k −  Models : Eddy Viscosity III

2
−ρu0i u0j = 2µt Sij − ρkδij (17)
3
or,

2
−u0i u0j = 2νt Sij − kδij (18)
3
1 ∂ui ∂uj
where Sij = ( + ) is the mean rate-of-strain tensor and δij is the
2 ∂xj ∂xi
Kronecker delta.

Syed Imran Shah 26/ 81

 Introduction k− k−ω SST RSTM References

KTH Sweden Flat Plate ZPG TBL DNS Data(2010)[15] I

DNS of ZPG FP TBL, Reθ = 2540

80
<−u′v′>/(∂U/∂y)
70 Cµ k2/ε

60

50
νT
+

40

30

20

10

0
0 200 400 600 800 1000
y+

Syed Imran Shah 27/ 81

 Introduction k− k−ω SST RSTM References

k −  Models : k transport equation (Exact) I

∂k ∂k ∂ui ∂u0 ∂u0i 1 ∂u0k p0 1 ∂u0i u0i u0k ∂2k

+ uk = −u0i u0k −ν i − − +ν
∂t ∂x ∂x ∂xk ∂xk ρ ∂xk 2 ∂x ∂xk ∂xk
| {z k} | {z k} | {z } | {z } | {z k } | {z }
(I) (II) (III) (IV) (V) (VI)

Transport equation (Exact) for k

I (I) Convection term

I (II) Production term (source term)

I (III) Viscous Dissipation term (sink term)

I (IV) Pressure Diusion term

I (V) Turbulent Diusion term

I (VI) Viscous Diusion term

Syed Imran Shah 28/ 81

 Introduction k− k−ω SST RSTM References

k −  Models : k transport equation (Exact) :Contd I

The exact transport equation for k can also be written as,

+ div(k U) = −u0i u0j .Sij − 2νSij

0 .S 0 + div(− 1 p0 u0 + 2ν u0 S 0 − 1 u0 u0 u0 )
∂k
ij ij
∂t ρ 2 i i j
(20)
In words,

Rate of transport rate of rate of transport transport

change + of k = production - dissipation + of k + of k by
of k by convection of k of k by pressure viscous stresses
transport
+ of k by
Reynolds stresses
3
3
Introduction to Computational Fluid Dynamics,Versteeg and Malalasekera,2007
Syed Imran Shah 29/ 81
 Introduction k− k−ω SST RSTM References

k −  Models : k transport equation (Exact) :Contd I

0 0
The viscous dissipation term per unit volume (2µSij Sij ) is usually
multiplied by ρ to give the pseudo-dissipation rate of the turbulent kinetic
0 0
energy per unit mass (2νSij Sij ) with units m2 /s3 . It is usually not written
02
as 2νSij to allow summation over the repeated index (Einstein notation).

0 S 0 = 2ν(S 02 + S 02 + S 02 + 2S 02 + 2S 02 + 2S 02 )
 = 2νSij ij 11 22 33 12 13 23
(21)

Syed Imran Shah 30/ 81

 Introduction k− k−ω SST RSTM References

k −  Models : k transport equation (Exact) :Contd I

The isotropic viscous dissipation term of the k -transport equation can be

replaced with the anisotropic viscous dissipation term when combined with
the viscous diusion term
4;

∂u0i ∂u0 ∂u0 ∂u0i ∂u0 ∂2k ∂u0 ∂u0i

   
∂ 0
ν u + k −ν i + k =ν −ν i
∂xk i ∂xk ∂xi ∂xk ∂xk ∂xi ∂xk ∂xk ∂xk ∂xk
(22)
Therefore the k -transport equation is valid for ows with anisotropic
dissipation rate.

4
Statistical Approach to Wall Turbulence by Sedat Tardu
Syed Imran Shah 31/ 81
 Introduction k− k−ω SST RSTM References

KTH Sweden Flat Plate ZPG TBL DNS Data(2010)[15] I

DNS of ZPG FP TBL, Re = 2540

θ

Production
0.2 Dissipation
0.15 V diffusion
T diffusion
0.1 Redistribution
loss gain

0.05
0
−0.05
−0.1
−0.15
−0.2
−0.25
0 100 200 300 400 500
y+

Syed Imran Shah 32/ 81

 Introduction k− k−ω SST RSTM References

Modeled k -transport equation I

LHS terms of the k -equation are used exact. On the RHS, the production
term of the k -equation is modeled using the Boussinesq hypothesis.

∂ui h ∂u
i ∂uj 2 i ∂u
i
−ρu0i u0j = µt ( + ) − δij ρk (23)
∂xj ∂xj ∂xi 3 ∂xj
The viscous dissipation term or  is computed from its own transport
equation. Approximately ≈ u03 /l = k 3/2 /l.
Turbulent diusion term is modeled using the gradient diusion hypothesis :
k is diused from the regions of high k to those of low k. Limitation :If the
gradient of k is zero in a direction, it would imply zero turbulent diusion.

µt ∂k
− 21 u0i u0i u0j = −u0j k ∼
=
σk ∂xj

Syed Imran Shah 33/ 81

 Introduction k− k−ω SST RSTM References

Modeled k -transport equation II

σk is equivalent of a turbulent Prandtl number for k and is usually set to

unity. Pressure diusion term might be modeled using the gradient law but
is not used because of the lack of denitive experimental data (Wilcox,
Turbulence Modeling for CFD, 1st Edition, 1993).

µt ∂p
−u0j p0 ∼
=
σp ∂xj

It is also proposed to regroup the 2 terms of diusion in case of boundary

layers,

  µ ∂k
t
− ku0j + p0 u0j =
σk ∂xj
5

5
"Modeled" in US English and "Modelled" in UK English
Syed Imran Shah 34/ 81
 Introduction k− k−ω SST RSTM References

Modeled k -transport equation : Contd I

The modeled k-equation can now be assembled as,

∂2k
     
∂k ∂k ∂ui ∂uj ∂ui
µt ∂k ∂
ρ + uj = µt + − ρ +
+µ 2
∂t ∂xj ∂xj ∂xi ∂xj
σk ∂xj ∂xj ∂xj
    
∂ui ∂uj ∂ui ∂ µt ∂k
= µt + − ρ + µ+
∂xj ∂xi ∂xj ∂xj σk ∂xj

Where for k− model, µt = ρCµ k 2 /. In case of fully-turbulent ow,

turbulent diusion >> viscous diusion and hence viscous diusion term
can be neglected with resulting equation (for steady ow in the mean) :

    
∂k ∂ui ∂uj ∂ui ∂ νt ∂k
uj = νt + −+ (25)
∂xj ∂xj ∂xi ∂xj ∂xj σk ∂xj

Syed Imran Shah 35/ 81

 Introduction k− k−ω SST RSTM References

The  transport equation (Exact) I

This can be derived from the Navier-Stokes equation and is far more
complex than the corresponding equation for k. For the incompressible ow
case,

∂u0 ν ∂p0 ∂u0j

 
∂ ∂ ∂
uj =− νu0j ( i )2 + 2 −ν
∂xj ∂xj ∂xl ρ ∂xi ∂xi ∂xj
 0 0
∂ui ∂ui ∂uj ∂u0l ∂u0l

− 2ν + (26)
∂xj ∂xl ∂xl ∂xi ∂xj
2
∂u0i ∂ 2 ui ∂u0i ∂u0i ∂u0j ∂ 2 u0i

− 2νuj0 − 2ν −2 ν
∂xl ∂xj ∂xl ∂xj ∂xl ∂xl ∂xj ∂xl

Syed Imran Shah 36/ 81

 Introduction k− k−ω SST RSTM References

The  transport equation (Simplied) I

There are many unknown correlations in the exact equation. To simplify

the equation, terms that are smaller at high Reynolds number Ret (order of
magnitude analysis) are neglected.

∂u0i 2 ∂ui ∂u0i ∂u0j ∂u0l ∂u0l

   
∂ ∂ 0
uj =− νuj ( ) − 2ν +
∂xj ∂xj ∂xl ∂xj ∂xl ∂xl ∂xi ∂xj
| {z } | {z } | {z }
(Advection) (Turbulent Diusion) (Production)
(27)
∂u0i ∂u0i ∂u0j ∂ 2 ui 2 ∂2
 
− 2ν − 2 ν + ν
∂xj ∂xl ∂xl ∂xj ∂xl ∂xj ∂xj
| {z } | {z } | {z }
(Turbulent Production) (Viscous Dissipation) Viscous Diusion

Syed Imran Shah 37/ 81

 Introduction k− k−ω SST RSTM References

Modeled  transport equation I

Most of the terms of this equation have been modeled by their dimensional
equivalents. The turbulent diusion term is modeled using the hypothesis
of transport by the mean gradient,

∂u0
     
∂ ∂ ∂ νt ∂
− νu0j ( i )2 = − u0j  = (28)
∂xj ∂xl ∂xj ∂xj σ ∂xj
Production terms (mean gradient and mixed) are modeled using the
anisotropy tensor (proposed by John Lumley).

∂ui ∂u0i ∂u0j ∂u0l ∂u0l

 
∂ui
−2ν + = −C1 aij  (29)
∂xj ∂xl ∂xl ∂xi ∂xj ∂xj

Syed Imran Shah 38/ 81

 Introduction k− k−ω SST RSTM References

Modeled  transport equation I

Anisotropy tensor aij if modeled through the Boussinesq hypothesis can be

written as,

u0i u0j
2
aij = − δij (30)
k 3
= −2νt Sij
 
1 ∂ui ∂uj
where Sij = + . Thus, the modeled production terms
2 ∂xj ∂xi
become,

∂ui ∂u0i ∂u0j ∂u0l ∂u0l

   
 ∂ui ∂ui ∂uj
−2ν + = C1 νt + (31)
∂xj ∂xl ∂xl ∂xi ∂xj k ∂xj ∂xj ∂xi

Syed Imran Shah 39/ 81

 Introduction k− k−ω SST RSTM References

Modeled  transport equation I

Turbulent production and viscous dissipation terms are modeled together

(both have dimensions of m2 /s4 ) as,

2
∂u0i ∂u0i ∂u0j ∂ 2 u0i 2

−2ν −2 ν = −C2 (32)
∂xj ∂xl ∂xl ∂xj ∂xl k
Assembling terms, the modeled transport equation for the dissipation rate 
can be written as,

∂2 2
   
∂ ∂ νt ∂  ∂ui ∂ui ∂uj
uj = +ν + C1 νt + − C2
∂xj ∂xj σ ∂xj ∂xj ∂xj k ∂xj ∂xj ∂xi k
(33)

Syed Imran Shah 40/ 81

 Introduction k− k−ω SST RSTM References

Choice of the constants for k −  models I

The adjustable constants or coecients in the modeled transport equations

are Cµ , σk , σ , C1 and C2 . These coecients can either have xed values
or can be functions of some parameters. They are meant to recapture the
"lost physics". The derivation of these coecients for the standard model
(Launder and Sharma, 1974)[9] is demonstrated here.

Syed Imran Shah 41/ 81

 Introduction k− k−ω SST RSTM References

Choice of the constants for k −  models II

Cµ - Logarithmic Layer of Turbulent Boundary Layer

Hypothesis of local equilibrium (dissipation rate = production rate),

νt  = νt P k
∂U
= νt (−u0 v 0 ) (34)
∂y
∂U
= νt (−u0 v 0 )
∂y

Using Boussinesq Hypothesis,

∂U
−u0 v 0 = νt
∂y

Syed Imran Shah 42/ 81

 Introduction k− k−ω SST RSTM References

we have,

νt  = (u0 v 0 )2 (35)

As

k2
νt = Cµ or ν t  = Cµ k 2 (36)

therefore comparison with the equation (35) gives,

(u0 v 0 )2
Cµ k 2 = (u0 v 0 )2 or Cµ = (37)
k2
Experimental results show that in the logarithmic layer of turbulent
boundary layer,

−u0 v 0
= 0.3 theref ore Cµ = 0.09 (38)
k
Syed Imran Shah 43/ 81
 Introduction k− k−ω SST RSTM References

C2 - Decaying Grid Turbulence I

Turbulent ow behind a grid is a simple ow which allows us to determine

the C2 coecient. Behind the grid, the velocity gradients are small which
∂U
means that turbulence production P k = −u0 v 0 ≈ 0. Furthermore V =0
∂y
and the diusion terms are negligible. Therefore the modeled k and 
equations become,

dk
U = − (39)
dx
d 2
U = −C2 (40)
dx k
Assuming that the decay of k is exponential, i.e., k ∝ x−n , we have,
dk
∝ −nx−n−1 . Putting it in the equation (39) gives,
dx
Syed Imran Shah 44/ 81
 Introduction k− k−ω SST RSTM References

−unx−n−1 = − or  = unx−n−1 (41)

Derivating the above equation with respect to x,

d
= un(−n − 1)x−n−2 (42)
dx
d
Substitute in the equation (40),
dx
2
u(un(−n − 1)x−n−2 ) = −C2
k
Substitute for  from equation (41) and replace k by x−n ,

(unx−n−1 )2
u(un(−n − 1)x−n−2 ) = −C2
x−n

Syed Imran Shah 45/ 81

 Introduction k− k−ω SST RSTM References

n+1
Solving for C2 gives, C2 = . Experimental results (e.g.,
n
Comte-Bellot and Corrsin, JFM, 1966)[4] show that n = 1.25 ± 0.06 and

therefore C2 = 1.92 .

Syed Imran Shah 46/ 81

 Introduction k− k−ω SST RSTM References

C1 - Logarithmic Layer of Turbulent Boundary Layer I

The modeled transport equation for the dissipation rate applied to the 2D
turbulent boundary layer is,

2
 
∂ ∂ ∂ νt ∂ 
u +v = + C1 P k − C2
∂x ∂y ∂y σ ∂y k k
In the logarithmic layer of the turbulent boundary layer, the convection
term is negligible and making use of the local equilibrium hypothesis
(replacing Pk by ), we have,

Syed Imran Shah 47/ 81

 Introduction k− k−ω SST RSTM References

C1 - Logarithmic Layer of Turbulent Boundary Layer II

2
 
∂ νt ∂
0 = + (C1 − C2 ) (43)
∂y σ ∂y k
 
∂ νt ∂ k
0 = + C1 − C2
∂y σ ∂y 2
 
k ∂ νt ∂
C1 = C2 − 2
 ∂y σ ∂y

Syed Imran Shah 48/ 81

 Introduction k− k−ω SST RSTM References

we need to substitute for , νt and k. Again using the local equilibrium

hypothesis of Pk =  and the relation u2τ ≈ -u0 v 0 , we have,

∂u ∂u u3
 = P k = −u0 v 0 ≈ u2τ = τ (44)
∂y ∂y κy
∂u uτ
where the relation = has been used. For νt ,
∂y κy

∂u ∂u
νt  = νt P k = νt (−u0 v 0 ) = (−u0 v 0 )νt = (u0 v 0 )2 ≈ u4τ (45)
∂y ∂y
νt ≈ u4τ / = uτ κy

For k,

k2 
νt = Cµ or k 2 = νt (46)
 Cµ

Syed Imran Shah 49/ 81

 Introduction k− k−ω SST RSTM References

As νt  ≈ u4τ , therefore

u4τ uτ
k2 = or k=p (47)
Cµ Cµ

Insert the expressions of , νt and k into equation (43) and simplify,

κ2
C1 = C2 − p (48)
σ  Cµ
The last two coecients σk and σ are optimised by applying the model to
various fundamental ows. For σk = 1 and σ = 1.3, C1 = 1.44 and the
set of coecients becomes,

Cµ = 0.09 σk = 1 C1 = 1.44 C2 = 1.92 σ = 1.3

Syed Imran Shah 50/ 81

 Introduction k− k−ω SST RSTM References

Some k −  models I

Model Cµ σk C1 C2 σ κ

Jones and Launder (1972) 0.09 1.0 1.57 2.00 1.3 0.41
Launder and Sharma (1974) 0.09 1.0 1.44 1.92 1.3 0.43
Chien (1982) 0.09 1.0 1.35 1.80 1.3 0.42
Nagano and Tagawa (1990) 0.09 1.4 1.45 1.90 1.3 0.42
Myong and Kasagi (1990) 0.09 1.4 1.40 1.80 1.3 0.395
Michelassi and Shih (1991) 0.09 1.3 1.45 2.00 1.3 0.46
So et al (1992) 0.09 1.0 1.50 1.83 1.45 0.385

From Aupoix(2004)[1]. Commerical softwares like FLUENT allow the user

to modify the closure coecients. Launder and Sharma (1974)[9] is
considered the standard model.

Syed Imran Shah 51/ 81

 Introduction k− k−ω SST RSTM References

Limitations of the k −  models

I Standard model is not realizable (Realizable k− model formulated)

I Does not take into account the curvature eects on turbulence

I Performs poorly in adverse (positive) pressure gradient ows

I Performs poorly in ows with buoyancy eects

I Performs poorly in stagnation ow (rapid strains, RNG k− model

formulated)

I Can't be used very near the wall (viscous sublayer and buer layer)

I Low-Reynolds ReT model/wall function is needed for near-wall ows

The dissipation term in the  equation includes the term 2 /k . When k→

0 in a low-turbulence zone like viscous sublayer, this causes large numerical
problems as generally k and  do not approach zero at the same rate.

Syed Imran Shah 52/ 81

 Introduction k− k−ω SST RSTM References

Coecients of the turbulence models : Limitations I

These coecients have been calibrated under certain conditions.

I Calibrated experimentally for air/water

I Not valid for oil, bear, liquid metals

I Calibrated for one atmospheric pressure

I Calibrated for Newtonian Fluids

CFD software FLUENT allows its users to modify these coecients.

Syed Imran Shah 53/ 81

 Introduction k− k−ω SST RSTM References

k − ω models I

I Concept originally presented by Kolmogorov (1942)[7]

I Modern form presented by Wilcox (1988)

I k transport equation is almost the same as k− models

I Instead of , a transport equation for ω is used

I Revised models are Wilcox (1998 and 2006)

I Can be integrated down to the wall

Wilcox (1988)[18] elected to dene ω as /β ? k with β ? = Cµ , hence it can

be dened as specic dissipation (rate of dissipation per unit turbulence
kinetic energy). The eddy viscosity is dened as νt = k/ω . Boussinesq
hypothesis is used for the computation of Reynolds stress.

Syed Imran Shah 54/ 81

 Introduction k− k−ω SST RSTM References

Transport Equations of the k − ω model (1988)[18]

The modeled k -transport equation is only altered by changing  to β ? kω .

   
∂k ∂ui ∂uj ∂ui ? ∂ ∂k
uj = νt + − β kω + (ν + σk νt ) (49)
∂xj ∂xj ∂xi ∂xj ∂xj ∂xj

   
∂ω ω ∂ui ∂uj ∂ui 2 ∂ ∂ω
uj = γ νt + − βω + (ν + σω νt ) (50)
∂xj k ∂xj ∂xi ∂xj ∂xj ∂xj

The ω transport equation can be used in a low-turbulence zone because its

production term does not go to zero as k→ 0.

ω ∂ui ∂uj ∂ui ∂ui ∂uj ∂ui

γ νt ( + ) = γ( + )
k ∂xj ∂xi ∂xj ∂xj ∂xi ∂xj

Syed Imran Shah 55/ 81

 Introduction k− k−ω SST RSTM References

Closure Coecients of k − ω I
The closure coecients of this model have been derived in the same way as
the k− model (using decaying grid turbulence, logarithmic layer of TBL
and local equilibrium hypothesis) and are given as :

β? = 0.09 σk = 1/2 σω = 1/2 β = 3/40 γ = 5/9

These coecients are slightly modied in the 1998[19] and 2006[20]

models with the addition of some auxilliary functions. The FLUENT
software automatically implements the proper boundary conditions at wall
(like k = 0 at wall) and at freestream.

Limitation of k − ω model
Ask=0 on a non-slip solid boundary while  6= 0 at wall therefore

ω= ? becomes innite at that boundary. Low-Reynolds k−ω model
β k
devised to overcome this default.

Syed Imran Shah 56/ 81

 Introduction k− k−ω SST RSTM References

KTH Sweden Flat Plate ZPG TBL DNS Data(2010)[15] I

DNS of ZPG FP TBL, Reθ = 2540 DNS of ZPG FP TBL, Reθ = 2540
0.5 0.35

0.3
0.4
0.25
0.3

−<u′v′>/k
0.2
Cµ

0.2 0.15

0.1
0.1
0.05

0 0
0 200 400 600 800 1000 0 500 1000 1500
y
+
y+

Syed Imran Shah 57/ 81

 Introduction k− k−ω SST RSTM References

KTH Sweden Flat Plate ZPG TBL DNS Data(2010)[15] I

DNS of ZPG FP TBL, Reθ = 2540 DNS of ZPG FP TBL Reθ 2540
2 14

12

1.5 10

(P + Π )/−ε of k
8
P+/−ε+ of k

+
1 6

+
4

+
0.5 2

0 −2
0 500 1000 1500 0 500 1000 1500
+
y y+

Syed Imran Shah 58/ 81

 Introduction k− k−ω SST RSTM References

KTH Sweden Flat Plate ZPG TBL DNS Data(2010)[15] I

DNS of ZPG FP TBL, Reθ = 2540

DNS of ZPG FP TBL, Reθ = 2540
5
0.9
k+
−ε+ 0.8
4
0.7

3 0.6
k+,−ε+

0.5

ω+
2 0.4

0.3
1 0.2

0.1
0
0
0 50 100 150 200 50 100 150 200
y+ y+

Note : y+ starts from 5 for the ω gure.

Syed Imran Shah 59/ 81

 Introduction k− k−ω SST RSTM References

Realizability Constraints I
Mathematical constraints that a turbulence model should satisfy.
(1) The Reynolds normal stresses, turbulent kinetic energy k and
dissipation rate  should stay positive (or zero) all the time.

u02
i ≥ 0, k ≥ 0, ≥0 (51)

(2) Schwarz inequality or the correlation coecient for Reynolds shear

stresses should not exceed one, i.e.,

u0i u0j
≤ 1, i 6= j, no summation over i and j (52)
(u02 02 1/2
i uj )

(3) When u02

i approaches zero near walls, it should do so smoothly. It can

be accomplished by requiring that the derivative of u02

i should go to zero as

u02
i goes to zero, i.e., Du02
i /Dt → 0.

Syed Imran Shah 60/ 81

 Introduction k− k−ω SST RSTM References

The rst condition is important for the eddy-viscosity models in the

stagnation ows. For the streamwise normal stress u02 , using Boussinesq
hypothesis,

2 ∂u
u02 = k − 2νt (53)
3 ∂x

If ∂u/∂x becomes too large (like in a stagnation ow), then u02 can
become negative or non-realizable. In general, normal Reynolds stresses can
become negative in an eddy-viscosity model if the mean strain rate is large
enough to satisfay,

k ∂u 1
> ≈ 3.7 (54)
 ∂x 3Cµ

The Realizable k− model implements these constraints and is able to

correctly predict the spreading rate of the round or axisymmetric jet (round
jet anomaly of the standard k− model).
Syed Imran Shah 61/ 81
 Introduction k− k−ω SST RSTM References

Shear Stress Transport (SST) Model-1994 I

I Mentor (1994)[10] noted two shortcomings of the basic k−ω model.

Firstly it overpredicts the level of shear stress in adverse pressure
gradient (APG) ows. Secondly it has spurious sensitivity to the
freestream values of ω (freestream boundary condition).

I To overcome the shortcomings of the basic k−ω model, Mentor

(1994)[10] proposed the shear stress transport (SST) model. The SST
model is developped in two stages. The rst to improve predictions in
APG TBLs and the second to solve the problem of sensitivity to the
freestream values of ω.
I Mentor (1994)[10] blended the k−ω model near the wall with the
k− model in the outer layer to get the best parts of each model.
The SST model was able to predict the reductions of eddy viscosity
due to adverse pressure gradient better than any of the two models.

Syed Imran Shah 61/ 81

 Introduction k− k−ω SST RSTM References

Experimental data shows that in TBLs undergoing APG, the production

k is much larger than k
rate of
p its dissipation rate, i.e., P >>  and
−u0 v 0 ' a1 k (where a1 = Cµ ) with consequences for two-equation k − 
and k − ω model. In the boundary layer ows, the Boussinesq assumption
can be written as,

∂u k ∂u Cµ k 2 ∂u
−u0 v 0 = νt = =
∂y ω ∂y  ∂y
 2 2 1/2  2 1/2
k Cµ ∂u 2 1/2 k Cµ ∂u 2
= k ( ) = kCµ ( )
2 ∂y 2 ∂y
Pk
= kCµ1/2 ( )1/2 (55)

When P k >> , equation (55) overpredicts the shear stress. In ZPG at
plate
k
TBL, P ≈  and equation (55) is valid.

Syed Imran Shah 62/ 81

 Introduction k− k−ω SST RSTM References

In the SST model, the denition of νt is modied to account for transport

eects of the shear stress. To reduce the turbulent shear stress |u0 v 0 | in
APG ows, eddy viscosity was redened to have two denitions.

1/2
k Cµ k
νt = = 1/2 (56)
ω Cµ ω
−u0 v 0 = νt ∂u
p
Now from
∂y and −u0 v 0 = Cµ k , we have,

p ∂u
Cµ k = νt = νt Ω
∂y
or
1/2
Cµ k
νt = (57)
Ω
where Ω = ∂u/∂y is the absolute vorticity for boundary layers. We want
equation (57) to be used in boundary layers hence we multiply it with a

Syed Imran Shah 63/ 81

 Introduction k− k−ω SST RSTM References

function 1/F2 with F2 equal to 1 near walls and 0 elsewhere. Then we take
a minimum νt of equations (56) and (57).

 1/2 
k Cµ k
νt = min , (58)
ω F2 Ω
1/2
Cµ k
(inside b.l) F2 = 1 ⇒ νt = if P k >  (Ω is high)
Ω
k
(outside b.l) F2 = 0 ⇒ νt =
ω

For implementing the k − ω model near wall and the k− away from it,
the blending function F1 is used,

(inside b.l) F1 = 1 ⇒ k − ω model

(outside b.l) F1 = 0 ⇒ k −  model (transformed)
Function F2 remains equal to 1 longer than F1 to aect the eddy viscosity
in the wake region of a boundary layer.

Syed Imran Shah 64/ 81

 Introduction k− k−ω SST RSTM References

Transport Equations of the SST model I

 
∂k ∂ ∂ ∂k fk − β ? kω
+ (uj k) = (ν + νt σk ) +P (59)
∂t ∂xj ∂xj ∂xj

 
∂ω ∂ ∂ ∂ω ω
+ (uj ω) = (ν + νt σω ) + α P k − βω 2 + (1 − F1 )CD
∂t ∂xj ∂xj ∂xj k
(60)
where
1 ∂k ∂ω
CD = 2σω2 (61)
ω ∂xi ∂xi
The boxed part of equation (60) corresponds to the k−ω model.
fk = min(P k , 20) = min(P k , 20β ? kω)
P is the production limiter to
prevent the build-up of turbulence in the stagnation regions.

Syed Imran Shah 65/ 81

 Introduction k− k−ω SST RSTM References

Blending Functions I
The viscosity switch F2 and model (inner/outer) switch F1 gradually
change from 1 near walls to 0 outside boundary layer and are given as,

F1 = tanh(φ41 ) (62)

  √  
k 500ν 4σω2 k
φ1 = min max ? , 2 , (63)
β ωd d ω max(CD , 10−20 )d2

F2 = tanh(φ22 ) (64)

 √ 
2 k 500ν
φ2 = max ? , 2 (65)
β ωd d ω
where d is the wall-normal distance to the closest wall (in some
applications like wing-fuselage junctions, closest wall is hard to determine).

Syed Imran Shah 66/ 81

 Introduction k− k−ω SST RSTM References

Blending Function F1 I
DNS of ZPG FP TBL, Re = 2540
θ
1

0.8
F1 of the SST Model

0.6

0.4

0.2

0
500 1000 1500 2000
y+

Syed Imran Shah 67/ 81

 Introduction k− k−ω SST RSTM References

Coecients of the SST model I

The coecients are not constants and are blended from their corresponding
k−ω and k− values using the relation,

C SST = F1 C k−ω + (1 − F1 )C k− (66)

Coecients of the SST model are derived from the following coecients
using equation (66) :

1) k−ω : σk1 = 1/2, σω1 = 1/2, β1 = 3/40, α1 = 5/9

2) k− : σk2 = 1, σω2 = 0.856, β2 = 0.0828, α2 = 0.44

These values of α1 and α2 (or γ1 and γ2 ) are from the SST (2003) model.
Also σk1 = 0.85 in the SST (2003)[11] model.

Syed Imran Shah 68/ 81

 Introduction k− k−ω SST RSTM References

Modications to the SST model (2003) I

The production limiter in the k -equation was modied to :

fk = min(P k , 10) = min(P k , 10β ? kω)

P (67)

The relation φ1 was also modied to :

  √  
k 500ν 4σω2 k
φ1 = min max ? , 2 , (68)
β ωd d ω max(CD , 10−10 )d2

The magnitude of vorticity

q Ω (used in the denition of νt ) was replaced by

|S| = 2Sij Sij .

A production limiter (similar to k -equation) is employed for the ω equation
too.

Syed Imran Shah 69/ 81

 Introduction k− k−ω SST RSTM References

Reynolds Stress Transport Models I

There are 10 unkowns in RANS ; the four mean quantities (ui , p) and six
turbulent stresses, −ρu0i u0j . To close this equation system, we need a
turbulence model. The most comprehensive turbulence model is to derive
exact transport equations for the turbulent stresses. This was rst proposed
by Rotta (1951). An exact transport equation for the Reynolds stresses can
be derived from the NS equations. Solving seven transport equations is
computationally expensive but gives precise results. RSTMs take into
account the curvature eects, buoyancy eects, stagnation regions etc.
Realizability can be a problem. NS equations for an incompressible and
steady ow with constant density and viscosity are,

∂ui
=0 (69)
∂xi

∂ui uj 1 ∂P ∂ 2 ui
=− +ν 2 (70)
∂xj ρ ∂xi ∂xj
Syed Imran Shah 70/ 81
 Introduction k− k−ω SST RSTM References

Reynolds Stress Transport Equation (Exact) I

The exact transport equation for the Reynolds stresses can be written as,

∂ 0 0 ∂ui ∂uj
u u uk = −u0j u0k − u0i u0k
∂xk i j ∂xk ∂xk
| {z } | {z }
convection production
 
∂ 0 0 0 ∂ 0 0 1 0 0
1 0 0
− u u u −ν u u + δjk ui p + δik uj p
∂xk i j k ∂xk i j ρ ρ (71)
| {z }
diusion
p0 ∂u0i ∂u0j ∂u0 ∂u0j
 
+ + − 2ν i
ρ ∂xj ∂xi ∂xk ∂xk
| {z } | {z }
pressure-strain viscous dissipation

or Cij = Pij + Dij + Πij − ij

Syed Imran Shah 71/ 81

 Introduction k− k−ω SST RSTM References

Reynolds Stress Transport Equation (Modeled) I

The term by term modeling of the basic RSTM model (Launder, Reece and
Rodi (1975)[8] with wall correction terms of Shir (1973)[16] and Gibson
and Launder (1978)[6] is demonstrated here.
(1) Production Rate Term Pij : This term is computed exactly without
using the Boussinesq hypothesis.
(2) Diusion Rate Term Dij : This term contains triple velocity correlations

u0i u0j u0k and pressure-velocity correlations u0 p0 . In the eddy-viscosity models,

standard gradient diusion hypothesis was used to model this term in the k
equation. This assumes that k is diused from regions of high k to those of
low k. Hence for the turbulent diusion term (subscript t for turbulent),

k 1 νt ∂k
Dj,t = − u0i u0j u0i = −u0j k =
2 σk ∂xj

Syed Imran Shah 72/ 81

 Introduction k− k−ω SST RSTM References

This scalar diusivity model implies that if there is zero gradient in a

direction (x,y or z ), then there is no turbulent diusion ux in that
direction. Tensor diusivity model, proposed by Daly and Harlow (1970)[5],
also called General Gradient Diusion Hypothsis (GGDH), is used for the
RSTM.
k ∂k
Dj,t ∝ u0j u0k (72)
∂xk
For j=1 and k=1,2,3, we have,

k ∂k ∂k ∂k
D1,t ∝ u01 u01 + u01 u02 + u01 u03 (73)
∂x1 ∂x2 ∂x3
∂k k
Therefore even if
∂x1 = 0, the turbulent diusion term D1,t may still be
∂k
non-zero if
∂x2 6= 0. The repeated index k causes summation over itself.

Syed Imran Shah 73/ 81

 Introduction k− k−ω SST RSTM References

For RSTM, the turbulent diusion term is modeled as,

k ∂u0i u0j
 
t ∂ 0 0
Dij = ck uk um (74)
∂xk  ∂xm

where ck = 0.25. This relation can result in numerical instabilities and

commercial CFD softwares (e.g.,FLUENT) use a simple scalar turbulent
diusivity model (Shir, 1973) [16],

νt ∂u0i u0j
 
t ∂
Dij = (75)
∂xk σk ∂xk

where νt is computed as in the k− model and σk = 0.82. The viscous

diusion is usually neglected. The contribution of pressure-velocity
correlation is either negelected or summed up with the turbulent diusion
term by adjusting the coecient ck . This works in most cases but in ows
with thermal buoyancy, this model is not valid.

Syed Imran Shah 74/ 81

 Introduction k− k−ω SST RSTM References

(3) Viscous Dissipation Rate Term ij : The small or dissipating scales of
turbulence are usually assumed to be isotropic at high Reynolds number
(Kolmogorov theory), i.e., the velocity uctuations of small-scale
turbulence do not have any preferred direction. In the isotropic turbulence
u02 = v 02 = w02 (turbulent normal stresses) and u0i u0j = 0 (turbulent shear
stresses). These properties are true for their gradients too,

∂u0 ∂u0 ∂v 0 ∂v 0 ∂w0 ∂w0 ∂u0i ∂u0j

= = and = 0, i 6= j
∂xk ∂xk ∂xk ∂xk ∂xk ∂xk ∂xk ∂xk

Using the above two relations, expression for the isotropic viscous
dissipation rate can be written as,

2 ∂u0i ∂u0i
ij = δij with =ν (76)
3 ∂xk ∂xk

Syed Imran Shah 75/ 81

 Introduction k− k−ω SST RSTM References

Limitation : The dissipation rate term is not always isotropic, e.g.,

dissipation rate in the near-wall turbulence is highly anisotropic.

Syed Imran Shah 76/ 81

 Introduction k− k−ω SST RSTM References

(4) Pressure-Strain Rate Term Πij : The modeled pressure-strain rate term
is divided into a slow part, a rapid part and a wall part. The wall part is
again divided into a slow and rapid part. It is also called the redistribution
term. It acts such as to make turbulence more isotropic by decreasing large
normal and shear stresses and increasing the small normal stresses. In this
sense it is also called as 'Robinhood term', taking from the rich and giving
it to the poor. If we denote the modeled pressure-strain terms as φij , then,

φij = φij,1 + φij,2 + φij,1w + φij,2w (77)

The subscript 1 denotes the slow term, 2 denotes the rapid term and
w denotes the wall part. A third part can be added for a body force.
(4a) Slow Pressure-Strain Term : It is also called as return-to-isotropy
term. The slow term represents the eects of turbulent uctuations
∂ui
whereas rapid term represents inuence of the mean velocity gradient .
∂xj

Syed Imran Shah 77/ 81

 Introduction k− k−ω SST RSTM References

For the slow term, Rotta (1951)[14] proposed a linear model,

 0 0
ui uj
  
 0 0 2 2
φij,1 = −c1 ρ u u − δij k = −c1 ρ − δij (78)
k i j 3 k 3

To promote isotropy, the coecient c1 > 1 and common value is c1 = 1.8

(experiments)
(4b)Rapid Pressure-Strain Term : Naot et al (1970) [12] proposed a relation
for the rapid term with analogy to the slow term of Rotta (1951)[14],
known as Isotropization of Production (IP) model (used in FLUENT),

 
2 k
φij,2 = −c2 ρ Pij − δij P (79)
3
where c2 can vary with Ret (near-wall ows) but is generally considered
constant and equal to 0.6 (validated in experiments).

Syed Imran Shah 78/ 81

 Introduction k− k−ω SST RSTM References

(4c) Wall-Reection Term : The wall-reection part of the pressure-strain

term dampens the normal stresses perpendicular to the wall and enhances
the stresses parallel to the wall. Both slow and rapid terms have their
wall-reection or wall-echo parts. For the slow part, Shir (1973) [16]
proposed,
 
 0 0 3 0 0 3 0 0
φw
ij,1 = cw
1 u u nk nm δij − uk ui nk nj − uk uj nk ni fw (80)
k k m 2 2
Gibson and Launder (JFM,1978)[6] proposed the wall-part of the rapid
term,
 
 3 3
φw w
ij,2 = c2 φkm,2 nk nm δij − φik,2 nk nj − φkj,2 nk ni fw (81)
k 2 2
Here ni is the ith unit normal vector (orientation of wall), cw
1 = 0.5, cw
2 =
0.4k 3/2
0.3 and fw = where d is the wall-normal distance. These relations
d
are valid for attached ows on impermeable walls. Advanced forms also
exist.
Syed Imran Shah 79/ 81
 Introduction k− k−ω SST RSTM References

Dissipation Rate Equation I

This is a modied  equation from the standard k− model. The exact
dissipation rate equation is the same as equation (26). In the modeled
equation, the diusion term is modeled using the General Gradient
Diusion Hypothesis (GGDH) as compared to the Standard Gradient
Diusion Hypothesis (SGDH) used in the k −  model.

∂u0i 2
     
∂ 0 ∂ 0 ∂ k 0 0 ∂
− νuj ( ) =− u  = C uj ul (82)
∂xj ∂xl ∂xj j ∂xj  ∂xl
Mean gradient and mixed production terms are computed exactly. FLUENT
employs an equation similar to the k− model with SGDH and Boussinesq
hypothesis for the production term. The model equation for  as per
Hanjalić and Launder (1972) is given as,

2
   
∂ ∂ k 0 0 ∂ 
uj = C uj ul + C1 Pii + C3 Gii − C2 (83)
∂xj ∂xj  ∂xl k k

Syed Imran Shah 80/ 81

 Introduction k− k−ω SST RSTM References

References I

[1] B. Aupoix. Introduction to turbulence modelling. Technical report,

VKI Lecture Series, 2004. 51

[2] B.S. Baldwin and H. Lomax. Thin layer approximation and algebraic
model for separated turbulent ows. In 16th Aerospace Sciences
Meeting, Huntsville, Alabama, January 1978. 13

[3] J. Boussinesq. Essai sur la théorie des eaux courantes. In Mémoires

present é par divers savants à l'academie des science de l'institut de

France, volume 23-1. Imprimerie Nationale, Paris, 1877. 7

[4] G. Comte-Bellot and S. Corrsin. The use of a contraction to improve

the isotropy of grid-generated turbulence. Journal of Fluid Mechanics,
25(4) :657682, 1966. 46

[5] B.J. Daly and F.H. Harlow. Transport equations in turbulence. Physics
of Fluids, 13 :26342649, 1970. 74

Syed Imran Shah 81/ 81

 Introduction k− k−ω SST RSTM References

References II

[6] M.M. Gibson and B.E. Launder. Ground eects on pressure

uctuations in the atmospheric boundary layer. Journal of Fluid
Mechanics, 86 :491511, 1978. 73, 80

[7] A.N. Kolmogorov. Equations of turbulent motion in an incompressible

uid. Izvestia Academy of Sciences USSR ;Physics, 6(1-2) :5658,
1942. 54

[8] B.E. Launder, G.J. Reece, and W. Rodi. Progress in the development
of a Reynolds-stress turbulence closure. Journal of Fluid Mechanics,
68 :537566, 1975. 73

[9] B.E. Launder and B.I. Sharma. Application of the energy dissipation
model of turbulence for the calculation of ow near a spinning disk.
Letters in Heat and Mass Transfer, 1(2) :131137, 1974. 41, 51

Syed Imran Shah 82/ 81

 Introduction k− k−ω SST RSTM References

References III

[10] F.R. Mentor. Two-equation eddy-viscosity turbulence models for

engineering applications. AIAA Journal, 32(8) :15981605, 1994. 62

[11] F.R. Mentor, M. Kuntz, and R. Langtry. Ten years of industrial

experience with the sst turbulence model. In K. Hanjalić, Y. Nagano,
and M. Tummers, editors, Fourth International Symposium on
Turbulence, Heat and Mass Transfer, Antalya, Turkey, 2003. Begell
House, Inc. 69

[12] D. Naot, A. Shavit, and M. Wolfshtein. Interactions between

components of the turbulent velocity correlation tensor. Israel Journal
of Technology, 8 :259, 1970. 79

[13] L. Prandtl. Bericht Ü ber untersuchungen zur ausgebildeten turbulenz.

ZAMM, 5(2) :136139, 1925. 13

Syed Imran Shah 83/ 81

 Introduction k− k−ω SST RSTM References

References IV

[14] J. Rotta. Statistische theorie nichthomogener turbulenz (statistical

theory of non-homogeneous turbulence). Zeitschrift für Physik,
129(6) :547572, 1951. 79

[15] P. Schlatter and R. Örlü. Assessment of direct numerical simulation

data of turbulent boundary layers. Journal of Fluid Mechanics,
659 :116126, 2010. 12, 27, 32, 57, 58, 59

[16] C.C. Shir. A preliminary numerical study of atmospheric turbulent

ows in the idealized planetary boundary layer. Journal of Atmospheric
Sciences, 30 :13271339, 1973. 73, 75, 80

[17] A.M.O Smith and T. Cebeci. Numerical solution of the turbulent

boundary layer equations-report dac 33735. Technical report, Douglas
Aircraft Division, 1967. 13

Syed Imran Shah 84/ 81

 Introduction k− k−ω SST RSTM References

References V

[18] D.C. Wilcox. Reassessment of the scale-determining equation for

advanced turbulence models. AIAA Journal, 26(11) :12991310, 1988.
54, 55

[19] D.C. Wilcox. Turbulence Modeling for CFD. DCW Industries,Inc, La

Cañada California, 2nd edition edition, 1998. 56

[20] D.C. Wilcox. Formulation of the k-omega turbulence model revisited.

AIAA Journal, 46(11) :28232838, 2006. 56

Syed Imran Shah 85/ 81