Professional Documents
Culture Documents
H
CFL3D has several turbulence model capabilities. Appendix H provides the derivation
for the advanced turbulence model equations. Be aware that while some variables in this
appendix are consistent with the rest of the manual (and are listed in the Nomenclature
section), many are introduced, defined, and used only in these sections and may or may
not appear in the Nomenclature listing. Also, for simplicity’s sake in this Appendix,
Re = Re L̃ .
R
Following Wilcox46, Favre19 averaging can be used with the Navier-Stokes equations to
account for turbulent fluctuations. The resulting equations of motion can be written using
the summation convention as follows. The full Navier-Stokes equations are shown here,
but in CFL3D, they are solved as the thin-layer approximation in pre-selected coordinate
direction(s). The ~ indicates a dimensional quantity.
∂ρ̃ ∂
------ + ( ρ̃ũ j ) = 0 (H-1)
∂t̃ ∂ x̃ j
∂ ∂ ∂ p̃ ∂τ̃
( ρ̃ũ i ) + ( ρ̃ũ j ũ i ) = – ------- + -------ji- (H-2)
∂ t̃ ∂ x̃ j ∂x̃ i ∂x̃ j
∂ ∂ ∂
( ρ̃Ẽ ) + ( ρ̃ũ j H̃ ) = [ ũ i τ̃ ij – q̇˜ j + ψ̃ j ] (H-3)
∂ t̃ ∂ x̃ j ∂ x̃ j
where
1 2 2 2
p̃ = ( γ – 1 ) ρ̃Ẽ – --ρ̃ ( ũ + ṽ + w̃ ) (H-4)
2
1 2 2 2
Ẽ = ẽ + -- ( ũ + ṽ + w̃ ) (H-5)
2
p̃
H̃ = Ẽ + --- (H-6)
ρ̃
1 µ̃
˜q̇ = – ---------- µ̃ T ∂ã 2
- ----- + -------- --------
γ – 1 Pr Pr T ∂x̃ j
j (H-7)
2 γ p̃
ã = ------ (H-8)
ρ̃
Note that the kinetic energy of the fluctuating turbulent field k̃ is ignored in the defini-
1 2 2 2
tion of Ẽ in CFL3D (it is assumed that k̃ << ẽ + -- ( ũ + ṽ + w̃ ) ). Define
2
1 ∂ũ ∂ũ
S̃ ij = -- -------i + -------j
2 ∂x̃ j ∂x̃ i
(H-9)
1 ∂ũ ∂ũ
W̃ ij = -- -------i – -------j
2 ∂x̃ j ∂x̃ i
Ω̃ = 2W̃ ij W̃ ij (H-10)
L T
τ̃ ij = τ̃ ij + τ̃ ij (H-11)
where
1 ∂ũ k
τ̃ ij = 2µ̃ S̃ ij – -- -------- δ ij
L
3 ∂x̃ k
(H-12)
For all eddy-viscosity models in CFL3D, the following approximations are made:
1 ∂ũ k
τ̃ ij = 2µ̃ T S̃ ij – -- -------- δ ij
T
3 ∂x̃ k
(H-13)
ψ̃ j = 0 (H-14)
˜ µ̃ T 1 ∂ũ k
Note that u i u j = – τ̃ ij ⁄ ρ̃ , or – 2 ------ S̃ ij – -- -------- δ ij in this case.
T
ρ̃ 3 ∂x̃ k
1 ∂ũ k
τ̃ ij = 2µ̃ T S̃ ij – -- -------- δ ij – --ρ̃k̃δ ij +
T 2
3 ∂x̃ k 3
(H-15)
k̃
2µ̃ T K 1 -- ( S̃ ik W̃ kj – W̃ ik S̃ kj ) –
ε̃
µ̃ ∂k̃
ψ̃ j = µ̃ + -----T- -------
σ k ∂x̃ j
(H-16)
µ̃ is replaced by µ̃ + µ̃ T
µ̃ µ̃ µ˜T
and ----- is replaced by ----- + --------
Pr Pr Pr T
where µ̃ T is the eddy viscosity value obtained by whatever turbulence model is used, and
Pr = 0.72 , Pr T = 0.9 . For the nonlinear models, both the above substitutions must be
made and the following additions as well. The term
∂ µ̃ ∂k̃
µ̃ + -----T- -------
∂ x̃ j σ k ∂x̃ j
is added to the energy equation. These additions are made in subroutines gfluxv, hfluxv,
and ffluxv.
H.2 Nondimensionalizations
k̃ µ̃ ∞ ω̃
k = -----
2
ω = -----------
-
2 P̃ k L̃ R
2
ã ∞ ρ̃ ∞ ã ∞ P k = -----------
-
2
µ̃ ∞ ã ∞
µ̃ ∞ ε̃ ρ̃
ε = -----------
- ρ = ------ 2
ρ̃ ∞ ã ∞
4 ρ̃ ∞ P̃ ω L̃ R
Pω = -----------
-
2
ũ p̃
p = -----------2- ρ̃ ∞ ã ∞
u = ----- (H-17)
ã ∞ ρ̃ ∞ ã ∞ 2
P̃ ε L̃ R
x̃ µ̃ Pε = -----------4-
x = ----- µ = ------ ρ̃ ∞ ã ∞
L̃ R µ̃ ∞
Ω̃L̃
τ̃ ij L̃ R t̃ ã Ω = ---------R-
τ ij = ------------ t = -------∞- ã ∞
µ̃ ∞ ã ∞ L̃ R
d = rk ⋅ n – r0 ⋅ n (H-18)
where r k is the vector from the origin to k given and r 0 is the vector from the origin to the
k = 1 point. Note that if grid lines curve significantly d can become negative as in
Figure H-1(b). If the grid lines do this, CFL3D currently does not allow the distance to go
negative. Instead, it computes d out to its maximum, then sets all distances thereafter to
that maximum value. However, if the grid behaves like the grid in Figure H-1(b), the Bald-
win-Lomax model, which is inherently dependent on the grid structure, is probably not a
good choice anyway. The use of any of the other field-equation models is recommended
instead.
fixed i, j line
k given n
n
d fixed i, j line
k = 1 k = 1
k given d < 0
(a) (b)
In the current implementation in CFL3D, if the grid contains multiple blocks, Baldwin-
Lomax should only be implemented on blocks that contain bodies. This restriction applies
to the Baldwin-Lomax model only. It does not apply to the more general field-equation
models.
µ T = µ T, inner y ≤ y crossover
(H-19)
µ T = µ T, outer y > y crossover
where y crossover is the location along the constant index line where µ T, inner exceeds
µ T, outer , marching away from the wall. The inner eddy viscosity is
µ T, inner = ρl Ω --------
2 Re
M ∞
(H-20)
+
l = y [ 1 – exp ( – y ⁄ 26 ) ] (H-21)
+
Since it now appears to be fairly widely accepted that the definition of y for the damp-
ing term needs to be modified in Baldwin-Lomax to give better answers when there are
large temperature gradients near the wall,
+ ρ̃τ̃ w
y = ------------ ỹ (H-22)
µ̃
+ ρ̃ w τ̃ w
y = ---------------- ỹ (H-23)
µ˜w
where
2
F wake = min [ y max F max, 1.0 y max u dif ⁄ F max ]
(H-26)
+
F ( y ) = yΩ [ 1 – exp ( – y ⁄ 26 ) ]
+
In wakes, exp ( – y ⁄ 26 ) is set to zero. F max is the maximum value of F ( y ) that occurs in
a profile and y max is the value at which F max occurs. Also,
0.3 y 6 –1
F kleb = 1 + 5.5 ----------
y max
(H-27)
2 2 2 2 2 2
u dif = ( u + v + w ) max – ( u + v + w ) min
The second term in u dif is taken to be zero, except in wakes. The region for the search for
F max is currently bound by
where η , in this case, is the index direction away from the body.
The test in CFL3D is very simple-minded and can often fail. However, it is quite diffi-
cult to find a test that works for all circumstances; so, for lack of anything better, the follow-
ing is used. Marching outward away from the body, F max is updated index by index. Then,
if
All of the one- and two-equation model equations can be written in the general form
∂ ∂
( X ) + u j ( X ) = SP + SD + D (H-30)
∂t ∂x j
All of the field-equation models are solved uncoupled from the Navier-Stokes equa-
tions. All of the models are solved in essentially the same fashion. Details are given in
“Solution Method” on page 299 in the form of an example.
Also, all of the one- and two-equation models are based on incompressible turbulence
equations. No compressibility corrections have been added. Hence, for problems where the
2
turbulent Mach number, M T = 2k̃ ⁄ ã , is high, these turbulence models may not be
applicable. Note that for most subsonic through low supersonic aerodynamic applications,
the incompressible forms of the turbulence models are generally expected to be valid.
All of the field equation models except for Wilcox k – ω make use of the minimum dis-
tance function, or the distance to the nearest wall, smin. This distance differs from the
directed distance used by the Baldwin-Lomax model in that it does not follow grid lines and
can be computed across zone boundaries. Hence, it is more easily applicable to multiple
zone applications. The minimum distance function represents the distance to the nearest
viscous wall and takes into account grid skewness when computing the nearest wall-point
location (in subroutine findmin_new). The exception is the Baldwin-Barth model which
uses a minimum distance algorithm (subroutine findmin) that does not take grid skew-
ness into account. (The reason for this exception is that the Baldwin-Barth model requires
other variables not currently implemented in subroutine findmin_new.)
The Baldwin-Barth9 model solves a single field equation for a turbulent Reynolds num-
ber term R :
∂R ∂R M∞ ν T ∂ 2 R M ∞ 1 ∂ ∂R
------ + u j ------
- = ( C ε f 2 – C ε ) RP + -------
- ν + ----- -------2- – -------- ----- ν -------
Re σ ε ∂x Re σ ε ∂ x j T ∂x j
(H-31)
∂t ∂x j 2 1
j
where
1 Cµ
----- = ( C ε – C ε ) ---------
- (H-32)
σε 2 1
κ
2
ν T = C µ RD 1 D 2 (H-33)
y+
D 1 = 1 – exp – -----+- (H-34)
A
y+
D 2 = 1 – exp – -----+- (H-35)
A2
Cε Cε 1
f 2 = -------1 + 1 – -------1 --------+ + D 1 D 2 D 1 D 2 +
Cε κ y
(H-36)
Cε
2 2
+
y 1 y+ 1 y+
---------------- -----+- exp – ------ D 2 + -----+- exp – ------ D 1
D1 D2 A A +
A2 A2 +
µ T = ρν T (H-37)
∂u k 2
P = 2ν T S ij S ij – --ν T --------
2
3 ∂x k
(H-39)
but is approximated as
2
P ≅ νT Ω (H-40)
S P ≅ ( C ε f 2 – C ε ) C µ D 1 D 2 ΩR (H-41)
2 1
M∞ ν ∂2 R M ∞ 1 ∂ ∂R
D = -------- ν + ----T- -------2- – -------- ----- ν T -------
∂x j
(H-42)
Re σ ε ∂x Re σ ε ∂ x j
j
+
Since the Baldwin-Barth model requires y (rather than just the minimum distance to
the wall), additional information about the nearest wall point needs to be stored as well as
the minimum distance function d . For simplicity, it is currently assumed that in the
+
regions where y has an effect, these additional wall values are in the same grid zone as the
point in question.
The Spalart-Allmaras34 model solves a single field equation for a variable ν̂ related to
the eddy viscosity through
µ T = ρν̂ f v (H-43)
1
where
3
χ
f v = ------------------
3 3
(H-44)
χ + Cv
1
1
ν̂
χ ≡ -- (H-45)
ν
The equation is
∂ν̂ ∂ν̂
------ + u j ------- = C b [ 1 – f t ]Ων̂ (H-46)
∂t ∂x j 1 2
M∞ ν̂ 2
+ -------- C b [ ( 1 – f t ) f v + f t ] ----2- – C w f w --
1
Re 1
2 2 2
κ 1
d
M ∞ C b2 ∂ 2 ν̂
– -------- -------ν̂ -------2-
Re σ ∂x
j
M∞ 1 ∂ ∂ν̂
+ -------- --- ( ν + ( 1 + C b )ν̂ ) -------
Re σ ∂ x j 2 ∂x j
1 1
6 -- –6 – 6 – -6-
1+ Cw 6 g + Cw
3 3
f w = g ------------------
6 6
- = ---------------------
–6
- (H-49)
g + Cw 1 + Cw
3 3
6
g = r + Cw (r – r ) (H-50)
2
ν̂
r = --------------------------- (H-51)
Ŝ -------- κ d
Re 2 2
M ∞
where
ν̂ f v
Ŝ = Ω + ------------------------
2
(H-52)
-------- κ 2 d 2
Re
M ∞
χ
f v = 1 – ------------------- (H-54)
2 1+χfv
1
CFL3D currently uses Spalart’s Version Ia34. The fv3 term was employed as a smooth fix to
prevent Ŝ from going negative prior to 12/97, but was removed after an error was discov-
ered ( Ŝ and fv2 were also different). The constants are
2
C b = 0.1355 σ = -- C b = 0.622 κ = 0.41
1 3 2
C w = 0.3
C w = 2.0 C t = 1.2 C t = 0.5 2
3 C v = 7.1 3 4
1 (H-55)
Cb (1 + Cb )
C w = ------2-1 + --------------------
2
(note typo in ref. 34)
1
κ σ
X = ν̂ (H-56)
S P = C b [ 1 – f t ]Ων̂ (H-57)
1 2
M∞ ν̂ 2
S D = -------- C b [ ( 1 – f t ) f v + f t ] ----2- – C w f w --
1
(H-58)
Re 1 2 2 2
κ 1
d
M ∞ C b2 ∂ 2 ν̂ M ∞ 1 ∂ ∂ν̂
D = – - -------ν̂ -------2- + -------- ---
------- ( ν + ( 1 + C b )ν̂ ) ------- (H-59)
Re σ ∂x Re σ ∂ x j 2 ∂x j
j
Note that in CFL3D, the S P and S D terms are grouped differently than Spalart’s. Part of
Spalart’s production term is grouped into S D because it has the common factor M ∞ ⁄ Re ,
like the other destruction terms.
∂k ∂k M∞ 1 ∂ µ ∂k M ∞
----- + u j ------- = --- P k -------- – β′kω -------- + --- µ + -----T- ------- --------
1 Re
ρ Re M ∞ ρ ∂ x j σ k ∂x j Re
(H-60)
∂t ∂x j
∂ω ∂ω M∞ 1 ∂ µ ∂ω M ∞
------ + u j ------- = --- P ω -------- – βω -------- + --- µ + -----T- ------- --------
1 2 Re
ρ Re M ∞ ρ ∂ x j σ ω ∂x j Re
(H-61)
∂t ∂x j
ρk
µ T = ------ (H-62)
ω
2
P ω = γρΩ (H-64)
The minimum distance function d is not required by this model. However, at the present
time it is still computed and stored by CFL3D. For the general form in Equation (H-30),
Xk = k (H-66)
2 M∞
S P, k = --- µ T Ω --------
1
Re
(H-67)
ρ
S D, k = – β′kω --------
Re
M ∞
(H-68)
1 ∂ µ ∂k M ∞
D k = --- µ + -----T- ------- --------
ρ∂x j σ k ∂x j Re
(H-69)
and
Xω = ω (H-70)
2 M∞
S P, ω = γ Ω --------
Re
(H-71)
S D, ω = – βω --------
2 Re
M ∞
(H-72)
1 ∂ µ ∂ω M ∞
D ω = --- µ + -----T- ------- --------
ρ∂x j σ ω ∂x j Re
(H-73)
Menter’s k-ω model is assessed in reference 27. Note that there are two corrections to
this reference. Equation (1) should be
σµ t = µ + σµ t
(H-74)
σ∗ µ t = µ + σ∗ µ t
Γ = max(2Γ 3, Γ 1 ) (H-75)
∂k ∂k M∞ 1 ∂ µ ∂k M ∞
----- + u j ------- = --- P k -------- – β′kω -------- + --- µ + -----T- ------- --------
1 Re
ρ Re M ∞ ρ ∂ x j σ k ∂x j Re
(H-76)
∂t ∂x j
∂ω ∂ω M∞ 1 ∂ µ ∂ω M ∞
------ + u j ------- = --- P ω -------- – βω -------- + --- µ + -----T- ------- --------
1 2 Re
ρ Re M ∞ ρ ∂ x j σ ω ∂x j Re
(H-77)
∂t ∂x j
1 1 ∂k ∂ω M ∞
+ 2 ( 1 – F 1 ) -------- --- ------- ------- --------
σ ω ω ∂x j ∂x j Re
2
In this model, C µ is incorporated into the definition of ω . The eddy viscosity is given by
ρk a 1 ρk Re
µ T = min ------, ----------- -----
ω ΩF 2 M
(H-78)
2
Pk = µT Ω (H-79)
2
P ω = γρΩ (H-80)
The constants are calculated from φ = F 1 φ 1 + ( 1 – F 1 )φ 2 , where the φ ’s are the con-
stants:
β κ
2 β2 κ
2
γ 1 = -----1- – ------------------ γ 2 = ------ – ------------------
Cµ σ C Cµ σ C
ω1 µ ω2 µ
σ k = 1 ⁄ 0.85 σ k = 1.0
1 2
β 1 = 0.075 β 2 = 0.0828
κ = 0.41 a 1 = 0.31
β′ = C µ = 0.09
4
F 1 = tanh ( Γ )
Γ = min [ max ( Γ 1, Γ 3 ), Γ 2 ]
500ν M ∞ 2 k M∞
Γ 3 = ------------- --------
4ρk
Γ 1 = ----------
2
- -------- Γ 2 = -----------------------------------
-
C µ ωd Re
d ω Re
2
d σ ω ( CD k – ω )
2 (H-82)
2 ∂k ∂ω
CD k – ω = max ρ ------------ ------- -------, 1 ×10
– 20
σ ω ω ∂x j ∂x j
2
2
F 2 = tanh ( Π )
Π = max ( 2Γ 3, Γ 1 )
Xk = k (H-83)
2 M∞
S P, k = --- µ T Ω --------
1
Re
(H-84)
ρ
S D, k = – β′kω --------
Re
M ∞
(H-85)
1 ∂ µ ∂k M ∞
D k = --- µ + -----T- ------- --------
ρ∂x j σ k ∂x j Re
(H-86)
and
Xω = ω (H-87)
2 M∞
S P, ω = γ Ω --------
Re
(H-88)
1 ∂k ∂ω M ∞
S D, ω = – βω -------- + 2 ( 1 – F 1 )σ ω --- ------- ------- --------
2 Re
M ∞ 2 ω ∂x ∂x Re
(H-89)
j j
1 ∂ µ ∂ω M ∞
D ω = --- µ + -----T- ------- --------
σ ω ∂x j Re
(H-90)
ρ∂x j
∂k ∂k M∞ 1 ∂ µ ∂k M ∞
----- + u j ------- = --- P k -------- – ε -------- + --- µ + -----T- ------- --------
1 Re
ρ Re M ∞ ρ ∂ x j σ k ∂x j Re
(H-91)
∂t ∂x j
∂ε ∂ε 1 M ∞ ε
2
1 ∂ µ ∂ε M ∞
– C ε - f 2 -------- + --- µ + -----T- ------- --------
Re
----- + u j ------
- --
-
= Pε -------
- ---
ρ Re M ∞ ρ ∂ x j σ ε ∂x j Re
(H-92)
∂t ∂x j 2 k
2
ρk
µ T = C µ f µ -------- (H-93)
ε
C ε = 1.45
1
σ k = 1.0
C ε = 1.83 σ ε = 1.4 (H-95)
2
C µ = 0.09
– 0.75
f µ = [ 1 + 4 ( Re τ ) ] tanh ( 0.008Re k )
(H-96)
Re
f2 = 1 – exp – -------k-
12
where
2
ρk
Re τ = --------
µε
(H-97)
ρ kd Re
Re k = ------------- -----
µ M
where d is the distance to the nearest wall. For the general form in Equation (H-30),
Xk = k (H-98)
2 M∞
S P, k = --- µ T Ω --------
1
Re
(H-99)
ρ
S D, k = – ε --------
Re
M ∞
(H-100)
1 ∂ µ ∂k M ∞
D k = --- µ + -----T- ------- --------
ρ∂x j σ k ∂x j Re
(H-101)
and
Xε = ε (H-102)
ε 2 M∞
S P, ε = --- C ε --µ T Ω --------
1
Re
(H-103)
ρ 1k
2
ε
= – C ε ---- f 2 --------
Re
S D, ε
M ∞
(H-104)
2 k
1 ∂ µ ∂ε M ∞
D ε = --- µ + -----T- ------- --------
ρ∂x j σ ε ∂x j Re
(H-105)
EASM stands for Explicit Algebraic Stress Model.3 Ivisc = 8 is the “linear” k – ω ver-
sion, treated as an eddy viscosity model, whereas ivisc = 12 is the fully nonlinear k – ω
version. The two versions obtain µ T with identical methods. However, the nonlinear
model includes nonlinear terms added to the Navier-Stokes equations, whereas the linear
model does not. (Its effect is felt only through the µ T term.)
The k – ω equations are of the same form as in the Wilcox k – ω model (Equation (H-
60) and Equation (H-61)). However, C µ is not incorporated into the definition of ω (it is
now a variable coefficient), so,
k
µ T = C µ ρ --- (H-106)
ω
σ k = 1.4
2 σ ω = 2.2
κ
γ = β – ------------------- β′ = 1.0 (H-107)
σ ω C µ∗
β = 0.83
κ = 0.41
T ∂u i
P k = τ ij -------
∂x j
(H-108)
ω T ∂u i
Pω = γ ---τ ij -------
k ∂x j
T
where τ ij is given (dimensionally) in Equation (H-15). The variable C µ and the coeffi-
cients K 1 and K 2 in Equation (H-15) are determined as follows:
2 6 6
3 ( 1 + η ) + 0.2 ( η + ζ )
2 2 2 2 6
-α
C µ = ----------------------------------------------------------------------
6 1
(H-109)
3 + η + 6ζ η + 6ζ + η + ζ
1
α2 -- M
( S S ) --------
2 ∞
η = ----
-
ω ij ij Re
(H-110)
1
α3 -- M
ζ = ----- ( W ij W ij ) --------
2 ∞
Re
(H-111)
ω
K 1 = α3
(H-112)
K 2 = 2α 2
where
α 1 = -- – C 2 --
4 g
3 2
g
α 2 = ( 2 – C 3 ) --
2
(H-113)
g
α 3 = ( 2 – C 4 ) --
2
C –1
g = -----1 + C 5 – 1
2
(Note that η, ζ , and α are just used here to denote terms in the equations and do not
reflect the definitions in the Nomenclature list.) Currently, the pressure-strain correlation
is modeled with the Speziale-Sarkar-Gatski (SSG) correlation, which uses:
T
To improve convergence, the µ T terms in τ ij multiplying the nonlinear terms
1
S ik W kj – W ik S kj and S ik S kj – --S S lk δ ij
3 kl
are replaced by
k
µ T ′ = C µ ′ρ --- (H-115)
ω
where
2 –8 6 6
3 ( 1 + η ) + 0.2 × 10 ( η + ζ )
2 2 2 2 6
-α
C µ ′ = -----------------------------------------------------------------------
6 1
(H-116)
3 + η + 6ζ η + 6ζ + η + ζ
k
µ T∗ = C µ∗ ρ --- (H-117)
ω
The minimum distance function d is not required by this model. However, at the present
time it is still computed and stored by CFL3D. For the general form in Equation (H-30),
Xk = k (H-118)
1 T ∂u i M ∞
S P, k = --- τ ij ------- --------
ρ ∂x j Re
(H-119)
S D, k = – β′kω --------
Re
M ∞
(H-120)
1 ∂ µ T∗ ∂k M ∞
D k = --- µ + -------- ------- -------- (H-121)
ρ∂x j σ k ∂x j Re
and
Xω = ω (H-122)
1 ω T ∂u i M ∞
S P, ω = --- γ ---τ ij ------- --------
ρ k ∂x j Re
(H-123)
S D, ω = – βω --------
2 Re
M ∞
(H-124)
1 ∂ µ T∗ ∂ω M ∞
D ω = --- µ + -------- ------- -------- (H-125)
ρ∂x j σ ω ∂x j Re
C ε = 1.83
2
κ
2
σ k = 1.0
C ε = C ε – ------------------ (H-127)
σ ε C µ∗ σ ε = 1.3
1 2
κ = 0.41
T ∂u i
P k = τ ij -------
∂x j
(H-128)
ε T ∂u i
P ε = C ε --τ ij -------
1k ∂x j
The variable C µ , and all constants are identical to those given for the EASM k – ω model
in “EASM Gatski-Speziale k-Omega Model” on page 286, except that ω is replaced by
ε ⁄ k . The damping functions are given by
Re
f 2 = 1 – exp – -------k-
12
(H-129)
where
ρ kd Re
Re k = ------------- --------
µ M ∞
(H-130)
T
The µ T terms in τ ij multiplying the nonlinear terms are replaced in exactly the same way
as for the EASM k – ω model. Also, the µ T terms in the D k and D ε diffusion terms are
replaced by
2
ρk
µ T∗ = C µ∗ -------- (H-131)
ε
Xk = k (H-132)
1 T ∂u i M ∞
S P, k = --- τ ij ------- --------
ρ ∂x j Re
(H-133)
S D, k = – ε --------
Re
M ∞
(H-134)
1 ∂ µ T∗ ∂k M ∞
D k = --- µ + -------- ------- -------- (H-135)
ρ∂x j σ k ∂x j Re
and
Xε = ε (H-136)
ε T ∂u i M ∞
S P, ε = --- C ε --τ ij ------- --------
1
ρ 1 k ∂x j Re
(H-137)
2
ε
= – C ε ---- f 2 --------
Re
S D, ε
M ∞
(H-138)
2 k
1 ∂ µ T∗ ∂ε M ∞
D ε = --- µ + -------- ------- -------- (H-139)
ρ∂x j σ ε ∂x j Re
∂k ∂k M∞ 1 ∂ µ ∂k M ∞
----- + u j ------- = --- P k -------- – ε -------- + --- µ + -----T- ------- --------
1 Re
ρ Re M ∞ ρ ∂ x j σ k ∂x j Re
(H-140)
∂t ∂x j
∂ε ∂ε 1 M ∞
2
ε Re 1 ∂ µ ∂ε M ∞
----- + u j ------
- --
-
= Pε -------
- – C ε - -------- + ---
--- µ + -----T- ------- --------
ρ Re 2 k M ρ∂x j σ ε ∂x j Re
(H-141)
∂t ∂x j ∞
ε 2µ ∂ k 2 M ∞
+ C ε -- ------ --------- --------
2 k ρ ∂x Re
j
2
ρk
µ T = – G 1 f µ -------- (H-142)
ε
2
The extra ( ∂ k ⁄ ∂x j ) term in the ε equation was added to replace the f 2 damping func-
2 2
tion term on the ε ⁄ k term, which is needed because ε ⁄ k → ∞ at the wall. The con-
stants are
C ε = 1.44
1
σ k = 1.0
(H-143)
C ε = 1.83 σ ε = 1.3
2
T ∂u i
P k = τ ij -------
∂x j
(H-144)
ε T ∂u i
P ε = C ε --τ ij -------
1k ∂x j
T
τ ij is given (dimensionally) in Equation (H-15). However, K 1 and K 2 in that equation are
now given by
G
K 1 = -----2-
G1
(H-146)
G
K 2 = – -----3-
G1
and
0
L1 L2
for η1 = 0 G 1 = --------------------------------------
-
0 2 2
( L 1 ) + 2η 2 ( L 4 )
0
1 L1 L2
for L1 = 0 G 1 = -----------------------------------------------------------------
-
0 2 2 2 2
( L 1 ) + --η 1 ( L 3 ) + 2η 2 ( L 4 )
3
1 1 (H-147)
-- --
3 3
G 1 = – -- + – --- + D + – -- – D
p b b
for D>0
3 2 2
θ
G 1 = – -- + 2 – -- cos --
p a
for D < 0, b < 0
3 3 3
θ 2π
G 1 = – -- + 2 – -- cos -- + ------
p a
for D < 0, b > 0
3 3 3 3
where
k 2 M∞ 2
η 1 = -- S ij S ij --------
ε Re
(H-148)
k 2 M∞ 2
η 2 = -- W ij W ij --------
ε Re
0 0
2L 1 L1 L2
p = – ----------
-
1
r = – ----------------- (H-149)
1 2
η1 L1 ( η1 L1 )
1 0 2 1 2 2 2
q = -----------------2 ( L 1 ) + η 1 L 1 L 2 – --η 1 ( L 3 ) + 2η 2 ( L 4 ) (H-150)
1 3
( η1 L1 )
2
a = q – -----
p 1 3
b = ----- ( 2p – 9pq + 27r )
3 27
2 3 (H-151)
b a –b ⁄ 2
D = ----- + ----- cos θ = --------------------
4 27 3
– a ⁄ 27
–L4 G1 +2L 3 G 1
G 2 = ---------------------------
0 1
- G 3 = ---------------------------
0 1
- (H-152)
L1 – η1 L1 G1 L1 – η1 L1 G1
C 2
L 2 = -----2 – --
0 2 3
0 C
L1 = -----1 – 1 C
2 L 3 = -----3 – 1 (H-153)
1 1
2
L1 = C 1 + 2 C
L 4 = -----4 – 1
2
0 C 2 = 0.36
C 1 = 3.4
C 3 = 1.25 (H-154)
1
C1 = 1.8 C 4 = 0.4
Xk = k (H-155)
1 T ∂u i M ∞
S P, k = --- τ ij ------- --------
ρ ∂x j Re
(H-156)
S D, k = – ε --------
Re
M ∞
(H-157)
1 ∂ µ ∂k M ∞
D k = --- µ + -----T- ------- --------
ρ∂x j σ k ∂x j Re
(H-158)
and
Xε = ε (H-159)
ε T ∂u i M ∞ ε 2µ ∂ k 2 M ∞
S P, ε = --- C ε --τ ij ------- -------- + C ε -- ------ --------- --------
1
ρ 1 k ∂x j Re 2 k ρ ∂x Re
(H-160)
j
2
ε Re
S D, ε = – C ε ---- --------
2 k M
(H-161)
∞
1 ∂ µ ∂ε M ∞
D ε = --- µ + -----T- ------- --------
ρ∂x j σ ε ∂x j Re
(H-162)
∂k ∂k ∂k ∂k
u j ------- = U ----- + V ------ + W ----- (H-163)
∂x j ∂ξ ∂η ∂ζ
where
U = ξx u + ξ y v + ξz w + ξt
V = ηx u + η y v + ηz w + ηt (H-164)
W = ζx u + ζ y v + ζz w + ζt
∂ ∂k ∂ ∂k ∂ ∂k ∂ ∂k
µ ------- ≅ ξ x ξ x µ ----- + ξ y ξ y µ ----- + ξ z ξ z µ -----
∂ x j ∂x j ∂ξ ∂ξ ∂ξ ∂ξ ∂ξ ∂ξ
(H-165)
∂ ∂k ∂ ∂k ∂ ∂k
+ η x η x µ ------ + η y η y µ ------ + η z η z µ ------
∂η ∂η ∂η ∂η ∂η ∂η
∂ ∂k ∂ ∂k ∂ ∂k
+ ζx ζ x µ ----- + ζ y ζ y µ ----- + ζ z ζ z µ -----
∂ζ ∂ζ ∂ζ ∂ζ ∂ζ ∂ζ
An example of the discretization is given for the first term on the right-hand side of Equa-
tion (H-165):
∂ ∂k ∂k ∂k
ξx ξ x µ ----- → ξ x ξ x µ 1 ----- 1 – ξ x ξ x µ 1 ----- 1
(H-166)
∂ξ ∂ξ i
i + -- i + -- ∂ξ i + --
1 i
i – -- i – -- ∂ξ i – --
1
2 2 2 2 2 2
where
∂k
----- = ki + 1 – ki
∂k
----- = ki – ki – 1
∂ξ i + 1-- ∂ξ i – 1--
(H-167)
2 2
i–1 i+1
i
1 1
i – -- i + --
2 2
Figure H-2. Definition of left and right states for generalized coordinate transformation.
For the Baldwin-Barth and Spalart-Allmaras models, the levels of the turbulence quan-
tities R and ν̂ are initialized at their free-stream levels everywhere in the flow field. For the
two-equation models, it helps the solution start-up phase to initialize the k and ω (or k
and ε ) values to crude profiles that simulate typical profiles in a boundary layer. Currently,
for k – ω :
2
k IC = max ( k ∞, – C 1 d + C 2 d )
(H-168)
C µ k IC
ω IC = max – 12444d + 0.54, --------------
v3d
where d is the distance to the nearest wall, scaled by Re/5.e6; and, for EASM models,
C µ = 0.09 ; otherwise, C µ = 1.0 . Also,
100k IC
v3d = max ---------------, 0.009
t max
2
t max = – C 1 S max + C 2 S max
(H-169)
C2
S max = --------
2C 1
C 1 = 45.8 C 2 = 1.68
For k – ε :
zk1 = k ∞
– 471d + 0.47
zk2 = 10
(H-171)
– 37.5d – 3.7
zk3 = 10
zk4 = 6.7d
ep1 = ε ∞
– 555d – 6
ep2 = 10
(H-173)
– 280d – 9.2
ep3 = 10
20 13333d – 9.8
ep4 = min ( 1 ×10 , 10 )
R ∞ = 0.1 (H-174)
ν̂ ∞ = 1.341946 (H-175)
For the standard k – ω or Menter’s SST models ( C µ is incorporated into the definition
of ω ),
–9
k ∞ = 9 × 10
(H-176)
–6
ω ∞ = 1 × 10
–9
k ∞ = 9 × 10
(H-177)
–8
ω ∞ = 9 × 10
–9
k ∞ = 1 × 10
(H-178)
– 17
ε ∞ = 1 × 10
µ T, ∞ ≅ 0.009 (H-179)
–8 – 17
in the free stream. Note that ω ∞ = 9 × 10 and ε ∞ = 1 × 10 for the EASM models
are based on C µ = 0.09 , which is only approximate since C µ is variable. Also note that
for two-equation models in the code, ε (or ω ) is represented by the 1st turbulence variable
(e.g., tur10), and k by the 2nd (e.g., tur20).
With the parameters known at the cell-centers near the solid wall boundary (see the
wall region schematic in Figure H-3), the following boundary conditions are applied.
For Baldwin-Barth,
Rw = 0 (H-180)
1
d1
For Spalart-Allmaras,
ν̂ = 0 (H-181)
kw = 0 (H-182)
60µ 1 M ∞ 2
ω w = --------------------2- --------
(H-183)
ρ 1 β ( d 1 ) Re
where
β = 0.075 otherwise.
kw = 0 (H-184)
2µ 1 ∂ k 2 M ∞ 2
εw = -------- --------- --------
ρ 1 ∂n w Re
(H-185)
Note that the actual wall boundary conditions for the turbulence quantities (unlike the
primitive variables) are applied at ghost cells. Hence,
χ BC = 2χ w – χ 1 (H-186)
Version 4.1 of CFL3D applied the above boundary condition for ω (Equation (H-183))
at the ghost cell center rather than at the wall. This is not correct, but supposedly any wall
boundary condition for ω that is bigger by some factor than the “analytical” value
2
6µ 1 ⁄ ( ρβ y ) gives basically the same result. (See reference 26.) The current version of
CFL3D (Version 5.0) obtains the ghost cell value using linear interpolation from ω w and
ω 1 (or ε w and ε 1 ).
The one or two equations are solved, decoupled, using implicit approximate factoriza-
tion (AF). The S P terms are treated explicitly, lagged in time while the S D and D terms are
treated implicitly (they are linearized and a term is brought to the left-hand-side of the
equations). This procedure is described by Spalart and Allmaras35 (strategy #3). The advec-
tive terms are discretized using first-order upwinding. Treating the destruction terms
implicitly helps increase the diagonal dominance of the left-hand-side matrix. Most of the
S D terms are linearized in a simplified fashion by assuming no coupling between the vari-
ables. For example,
(n + 1) (n) ∂
β′kω ≅ β′kω + ( β′kω )∆k (H-187)
∂k
(n)
≅ β′kω + β′ω∆k
For the S D terms in the k – ε equations, however, a more sophisticated approach was
taken. When linearizing, the k and ε variables are assumed to be coupled, with the eddy
viscosity µ T assumed to be fixed. For example, in the k equation, S D = – ε ( Re ⁄ M ∞ ) is
linearized as follows:
(n + 1) (n) ∂
ε ≅ε + ( ε )∆k (H-188)
∂k
2
(n) ∂ C µ f µ ρk
≅ε + --------------------- ∆k
∂ k µT
(n) C µ f µ ρk
≅ε + 2 ------------------- ∆k
µT
(n) ε
≅ε + 2 --∆k
k
2
ε
The S D = – C ε ---- f 2 ( Re ⁄ M ∞ ) term in the ε equation is treated similarly, for consis-
2 k
tency:
2(n + 1) 2(n) 2
ε ε ∂ ε
---- ≅ ---- + ---- ∆ε
∂ε k
(H-189)
k k
1 3
2(n) --
µ f µ ρ 2
∂ 2 --
ε C
≅ ---- + --------------- ε ∆ε
k ∂ε µT
1
2(n) --
ε 3 C µ f µ ρε 2
≅ ---- + -- ------------------ ∆ε
k 2 µT
2(n)
ε 3ε
≅ ---- + -- --∆ε
k 2k
Menter’s SST k – ω model has an additional cross-diffusion term in the ω equation. This
term is treated implicitly (See reference 26).
(n + 1) (n) ∂
Cω ≅ Cω + C ∆ω (H-190)
∂ω ω
(n) Cω
≅ C ω – --------- ∆ω
ω
(The negative sign insures that, when taken to the left-hand side, this term increases the
diagonal dominance of the implicit matrix.)
H.8.1 Example
This example shows the solution method for the k equation in Menter’s SST model. All
other equations are solved in a similar fashion.
∂k ∂k M∞ 1 ∂ µ ∂k M ∞
----- + u j ------- = --- P k -------- – β′kω -------- + --- µ + -----T- ------- --------
1 Re
ρ Re M ∞ ρ ∂ x j σ k ∂x j Re
(H-191)
∂t ∂x j
Written in generalized coordinates, with the production term treated explicitly (lagged in
time), this equation becomes:
∂k ( n + 1 ) ∂k ( n + 1 ) ∂k ( n + 1 ) ∂k ( n + 1 )
----- = – U ----- – V ------ – W ----- (H-192)
∂t ∂ξ ∂η ∂ζ
1 M ∞ ( n ) Re ( n + 1 )
+ - P k --------
-- – β′kω --------
ρ Re M ∞
1 ∂ 2 µ ∂k
( ξ x + ξ y + ξ z ) µ + -----T- -----
2 2
+ ---
ρ ∂ξ σ k ∂ξ
∂ 2 µ ∂k
( η x + η y + η z ) µ + -----T- ------
2 2
+
∂η σ k ∂η
µ ∂k (n + 1)
∂ 2 M ∞
( ζ x + ζ y + ζ z ) µ + -----T- ----- -------
2 2
+
∂ζ σ k ∂ζ Re
-
= RHS k
2 2 2
where the superscripts denote the time level. Note that the terms ( ξ x + ξ y + ξ z ) etc. are
not strictly correct as written. This is really short-hand notation. The correct way to
expand the diffusion terms is given in “Generalized Coordinate Form” on page 294. This
section also indicates how these metric terms are discretized.
U = ξx u + ξ y v + ξz w + ξt
V = ηx u + η y v + ηz w + ηt (H-193)
W = ζx u + ζ y v + ζz w + ζt
Define
(m + 1) (m)
∆k = k –k (H-194)
Then, t-TS subiterations are used (see Appendix B). (If no subiterations, then m = n .)
(m + 1) (n) (n) (n – 1)
(1 + φ)(k – k ) – φ(k – k )
---------------------------------------------------------------------------------------------- = RHS k (H-195)
∆t
φ = 0 for 1st order in time (also used for non-time-accurate runs), φ = 1 ⁄ 2 for 2nd
order in time. Linearize the right-hand-side terms that are taken at time level ( m + 1 ):
∂k ( m + 1 ) ∂k ( m ) upwind
– U ----- ≅ – U ----- – U δ ξ ( ∆k ) (H-196)
∂ξ ∂ξ
(m + 1) (m)
– -------- β′kω ≅ – -------- β′kω – -------- β′ω ( ∆k )
Re Re Re
M ∞ M ∞ M ∞
(H-197)
(m + 1) (m)
M ∞ 1 ∂ ∂k
------- M ∞ 1 ∂ ∂k M∞ 1 ∂
χ ----- ≅ -------- --- χ ξ ----- + -------- --- { χ ξ δ ξ ( ∆k ) }
Re ρ ∂ ξ ξ ∂ξ
- ---
Re ρ ∂ ξ Re ρ ∂ ξ
(H-198)
∂ξ
µ
where χ ξ represents ( ξ x + ξ y + ξ z ) µ + -----T- .
2 2 2
σ k
The equation is approximately factored and the contribution of the S D term to the left-
hand side is added in the first factor only.
M∞ 1 ∂
+ ∆t -------- β′ω – ∆t -------- --- ( χ ξ δ ξ )
1 upwind Re
-----------------2- ( 1 + φ )I + ∆tU δ ξ
M ∞ Re ρ ∂ ξ
(H-199)
(1 + φ)
M∞ 1 ∂
– ∆t -------- --- ( χ η δ η )
upwind
( 1 + φ )I + ∆tV δ η
Re ρ ∂ η
M∞ 1 ∂
– ∆t -------- --- ( χ ζ δ ζ ) ∆k = R
upwind
( 1 + φ )I + ∆tW δ ζ
Re ρ ∂ ζ
(n – 1) (m) (n)
R = ∆ tRHS k + φ∆k – (1 + φ)(k –k ) (H-200)
This equation is solved in a series of three sweeps in each of the three coordinate direc-
tions:
M∞ 1 ∂
+ ∆t -------- β′ω – ∆t -------- --- ( χ ξ δ ξ ) ∆k∗ = R
upwind Re
( 1 + φ )I + ∆tU δ ξ
M ∞ Re ρ ∂ ξ
M∞ 1 ∂
– ∆t -------- --- ( χ η δ η ) ∆k∗∗ = ( 1 + φ )∆k∗
upwind
( 1 + φ )I + ∆tV δ η
Re ρ ∂ η (H-201)
M∞ 1 ∂
– ∆t -------- --- ( χ ζ δ ζ ) ∆k = ( 1 + φ )∆k∗∗
upwind
( 1 + φ )I + ∆tW δ ζ
Re ρ ∂ ζ
(m + 1) (m)
k = k + ∆k
Each sweep requires the solution of a scalar tridiagonal matrix. In versions of CFL3D prior
to March 2002, φ = 0 always for the turbulence equations, regardless of the temporal
order of accuracy of the mean flow equations.
H.9 Limiting
As a precaution, many of the turbulence quantities are limited in practice. Whether all
of these conditions are necessary or not for the models to work and/or converge well is
unknown. Some are probably included only because they were tried at some point and
never removed.
•In all models, the linearized S D term is added to the left-hand-side matrix only if its
contribution is positive.
•In all models, the terms D, ν̂, k, ω , and ε are not allowed to become negative. If the
equation “wants” to yield a negative value, it is instead set to some very small positive
number. (A counter also sums the number of times this happen and it is output in the
*.turres output file.)
•In all the k – ω models as well as the k – ε models for Abid and Gatski-Speziale, the
production term in the k equation (i.e.
1 T ∂u M
--- τ ij -------i -------∞-
ρ ∂x j Re
•In all EASM models, the τ 11, τ 22 , and τ 33 terms are limited to be positive (realizabil-
ity constraint).
T ∂u i
•In all EASM models, the production term τ ij ------- is limited to be positive.
∂x j
1 T ∂u M
--- τ ij -------i -------∞-
ρ ∂x j Re
is limited (in both the k and ε equations) to be less than or equal to 40ε -------- .
Re
M ∞
µ ∂ k 2 M∞
ε -------- – 2 --- --------- --------
Re
M ∞ ρ ∂x j Re
is limited to be positive.
2 2
•In the Gatski-Speziale EASM models, η and ζ are each limited to be less than or
equal to 10.
•In the Gatski-Speziale EASM models, C µ is limited to be between 0.005 and 0.187.
•In the Girimaji EASM models, η 1 and η 2 are each limited to be less than or equal to
1200.
•In the Girimaji EASM model, G 1 is limited to be between -0.005 and -0.2.
T
•In the EASM k – ε models, “ ε ”in the denominator of the nonlinear terms in τ ij is
limited to be greater than or equal to the free-stream value of ε only in terms that
appear in the Navier-Stokes equations. This limiting is not done for the τ ij terms that
appear in the S P terms in the k and ε equations.
•In the one- and two-equation models, the diffusion term in one coordinate direction,
for example, can be written as
∂ ∂χ
B ------ = B 1 χ i + 1 – B 1 + B 1 χ i + B 1 χ i – 1
∂ ξ ∂ξ i + -- i + -- i – -- i – --
2 2 2 2
B = max B 1, 0
1
i + -- i + --
2 2
B = max B 1, 0
1
i – -- i – --
2 2
A simple approach for wall functions is taken from Abdol-Hamid, Lakshmanan, and
Carlson1. This approach modifies the eddy-viscosity value µ T, w at the wall only. The wall
shear stress is estimated from law-of-the-wall values at the first cell center off the wall and
is used to determine an “equivalent” eddy viscosity at the wall.
Wall functions can be employed when the grid spacing is too coarse in the direction
+ + +
normal to the wall; typically y lies in the range 30 < y < 200 . (The y value of the first
grid point off the wall should be O(1) to ensure decent turbulent computations when no
wall functions are used. See Section 2.2.) Wall functions are invoked in CFL3D by setting
the value of ivisc(m) to be negative.
A caution: While wall functions can work very well, they can also sometimes cause
problems. They are not strictly valid for separated flows, although many people use them
anyway with reasonable success. It is recommended that wall functions not be used with
the Baldwin-Lomax model in CFL3D. With other turbulence models, it is recommended
that they be used sparingly, if at all. In our opinion, it is better to make a turbulent grid
+
( y = O ( 1 ) ) whenever possible and avoid the use of wall functions.
The wall function replaces the eddy viscosity at the ghost cell center at walls as follows:
µ T, ghost = 2µ T, w – µ T, 1
+2
n µ1 M∞
µ T, w = µ 1 ------------------------ -------- – 1
∂u 2 Re
ρ 1 ------ d (H-202)
∂y 1
2 2 2
∂u ( u1 – uw ) + ( v1 – vw ) + ( w1 – ww )
------ = ---------------------------------------------------------------------------------------
-
∂y 1 d1
where d 1 is the distance to the nearest wall from the first cell center off the wall and
+
for R c ≤ 20.24 n = Rc
7
∑ a0 ( n )Rc
+ n–1
for 20.24 < R c ≤ 435 n = a0 ( 1 ) +
n=2
5 (H-203)
∑
+ n–1
for 435 < R c ≤ 4000 n = a1 ( 1 ) + a 1 ( n )R c
n=2
3
∑ a2 ( n )Rc
+ n–1
for R c > 4000 n = a2 ( 1 ) +
n=2
2 2 2
ρ 1 ( u 1 – u w ) + ( v 1 – v w ) + ( w 1 – w w ) d 1 Re
R c = --------------------------------------------------------------------------------------------------- --------
M ∞
(H-204)
µ1
a 1 ( 1 ) = 5.777191
a 0 ( 1 ) = 2.354039 a 1 ( 2 ) = 6.8756983 ×10
–2
a 0 ( 2 ) = 0.117984 –6
a 1 ( 3 ) = – 7.1582745 ×10
–4
a 0 ( 3 ) = – 4.2899192 ×10 –9
a 1 ( 4 ) = 1.5594904 ×10
–6
a 0 ( 4 ) = 2.0404148 ×10 – 13
a 1 ( 5 ) = – 1.4865778 ×10 (H-205)
–9
a 0 ( 5 ) = – 5.1775775 ×10
a 0 ( 6 ) = 6.2687308 ×10
– 12 a 2 ( 1 ) = 31.08654
–2
a 0 ( 7 ) = – 2.916958 ×10
– 15 a 2 ( 2 ) = 5.0429072 ×10
–8
a 2 ( 3 ) = – 2.0072314 ×10