Documenti di Didattica
Documenti di Professioni
Documenti di Cultura
DOI: 10.2514/1.36541
This paper presents a reformulated version of the authors k-! model of turbulence. Revisions include the addition
of just one new closure coefcient and an adjustment to the dependence of eddy viscosity on turbulence properties.
The result is a signicantly improved model that applies to both boundary layers and free shear ows and that has
very little sensitivity to nite freestream boundary conditions on turbulence properties. The improvements to the k-!
model facilitate a signicant expansion of its range of applicability. The new model, like preceding versions, provides
accurate solutions for mildly separated ows and simple geometries such as that of a backward-facing step. The
models improvement over earlier versions lies in its accuracy for even more complicated separated ows. This paper
demonstrates the enhanced capability for supersonic ow into compression corners and a hypersonic shock-wave/
boundary-layer interaction. The excellent agreement is achieved without introducing any compressibility
modications to the turbulence model.
Ue , U1
Nomenclature
Clim
Cp
c
cf
Dk
E
e
f
H
h
k
ks , k
s
=
=
=
=
=
=
=
=
=
=
=
=
M1
p
=
=
pw , p1
Pk
PrL , PrT
ReH , Re
=
=
=
=
Re1
r
SR , SB
=
=
=
Sij
Sij
S^ij
Tw , Taw
t
tij
=
=
stress-limiter coefcient
2
pressure coefcient, p p1 =12 U1
chord length
skin-friction coefcient
dissipation of turbulence kinetic energy
Favre-averaged total energy, e 12 ui ui k
Favre-averaged specic internal energy
round-jet function
backward-facing-step height
Favre-averaged specic enthalpy
Favre-averaged specic turbulence kinetic energy
dimensional and dimensionless surface-roughness
height, u ks =w
freestream Mach number
mean static pressure, nite-difference-scheme
order of accuracy
surface and freestream mean static pressure
production of turbulence kinetic energy
laminar and turbulent Prandtl number
Reynolds number based on step height and
momentum thickness
Reynolds number per unit length
ne-grid point to coarse-grid point number ratio
dimensionless surface value of ! for a surface
with roughness and mass injection
Favre-averaged strain-rate tensor,
1
@ui =@xj @uj =@xi
2
zero-trace Favre-averaged strain-rate tensor,
Sij 13 @um =@xm ij
Galilean-invariant Favre-averaged strain-rate
tensor, Sij 12 @um =@xm ij
Favre-averaged wall temperature, adiabatic wall
temperature
time
mean viscous stress tensor
u
ui
u
vw , v
w
x
xi
xr , xs
y
, , , d
,
0
0o
ij
"
, T
, w
do
ij
xy
!
ij
!
!~
I. Introduction
2823
2824
WILCOX
B.
Constitutive Equations
2
ij 2 T Sij kij
3
1 @uk
S ij Sij
3 @xk ij
k
T ;
!~
s9
2Sij Sij =
!~ max !; Clim
:
;
8
<
Clim 78
Mean-Flow Equations
(1)
@
@p
@
@
ui
uj ui
tji ji
@x
@x
@x
@t
j
i
j
(2)
1
@
1
@
e ui ui k
uj h ui ui k
2
@xj
2
@t
@
T @h
@k
k
u t ij
@xj i ij
PrL PrT @xj
! @xj
(3)
Note that the energy-conservation equation (3) ensures conservation
of total energy E, which includes the kinetic energy of the turbulence.
(4)
(5)
(6)
(7)
For the sake of clarity, this paper will refer to the reformulated k-!
equations as the new k-! model. This paper focuses on whats new
about the model relative to previous versions. Readers interested in
all aspects of the model and its development can nd a complete
presentation by Wilcox [14].
A.
ij
@ui
@
k!
@xj
@xj
k @k
! @xj
@
@
!
@u
!
uj ! ij i !2
@t
@xj
k
@xj
@k @!
@
k @!
d
! @xj @xj @xj
! @xj
(8)
(9)
2825
WILCOX
D.
Closure Coefcients
9
100
;
(
d
o f ;
PrT 89 (10)
SB
0;
do ;
@k
@xj
@k
@xj
@!
@xj
@!
@xj
0
>0
o 0:0708;
ij jk S^ki
;
!
!3
Downloaded by University of Newcastle on March 12, 2014 | http://arc.aiaa.org | DOI: 10.2514/1.36541
35;
12;
1
do
8
f
(11)
1 85
!
1 100
!
1 @um
S^ki Ski
2 @xm ki
(12)
(13)
For surfaces that include surface roughness, the model uses the noslip condition for velocity and k. The surface value of ! depends
upon the dimensionless surface-roughness height k
s . The boundary
condition for ! is
!
u2
S
R
at y 0
(14)
where SR was chosen to provide a close match to measured roughsurface boundary-layer data of Nikuradse, as noted in [16]. The
following correlation between SR and k
s reproduces measured
effects of sand-grain roughness for values of k
s up to about 400:
( 200 2
k ;
k
s 5
SR 100s
(15)
200 2
100 5ks
k k
e
; k
s >5
k
s
40; 000
k2s
at y 0
(16)
u2
S
B
at y 0
Table 1
Flow
Far wake
Mixing layer
Plane jet
Round jet
Radial jet
(17)
25
v
1
5v
w
w
(18)
Speziale [6]
Peng [18]
Kok [9]
Wilcox [4]
New k-!
Measured
0.221
0.082
0.089
0.102
0.073
0.206
0.071
0.096
0.040
0.191
0.056
0.083
0.107
0.068
0.496
0.141
0.135
0.369
0.317
0.326
0.096
0.108
0.094
0.099
0.3200.400
0.1030.120
0.1000.110
0.0860.096
0.0960.110
2826
WILCOX
rst that, based on the analysis of a turbulent front by Lele [20], there
are two necessary conditions for the front to propagate. Specically,
we must have
do >
and
> do
(20)
Figure 1 shows how predicted spreading rates vary with do for the
far wake, the mixing layer, and the plane jet. The curves shown were
computed with all other closure coefcients as specied in Eqs. (10)
and (12). To isolate effects of cross diffusion, results shown
correspond to having no stress limiter: that is, !~ ! in Eq. (6). The
limiter has virtually no effect on the far wake and the plane jet. It
reduces the mixing-layer spreading rate by less than 6%. Of greatest
relevance to the present discussion, the value of is 3=5. As shown,
spreading rates for all three cases are greatest when do is equal to its
minimum permissible value according to Eq. (20) (viz.,
do ). Predicted values decrease monotonically as do
increases and fall below the lower bound of measured spreading rates
for all three cases when do 15, which is much less than the
maximum allowable value of 35.
Figure 2 shows how predicted spreading rates vary with when
we set do equal to its minimum permissible value. As noted,
computations were done with all closure coefcients other than ,
as specied in Eqs. (10) and (12) in the absence of the stress limiter.
Computed spreading rates for all three cases decrease monotonically
as increases. Computed 0 values lie above the range of measured
0 for all three cases when < 0:55 and below when < 0:70.
Thus, we conclude that
0:55 < < 0:70
for 12
(21)
Fig. 1
Fig. 2
B.
Round-Jet/Plane-Jet Anomaly
"2
"2
! C"2 C"3
p
k
k
(22)
where C"2 and C"3 are closure coefcients. In terms of k-! model
parameters, " / k!. The parameter
p is a nondimensional measure
of vortex stretching dened as
p
(23)
Effect of cross diffusion on free-shear-ow spreading rates for 35 and 12; shaded areas depict measured-value ranges.
Effect of cross diffusion on free-shear-ow spreading rates for do and 12; shaded areas depict measured-value ranges.
2827
WILCOX
model predictions for the radial jet, which also has nonzero
p .
Without the Pope correction, the k-" model predicts a radial-jet
spreading rate of 0.094, which is close to the measured range of 0.096
to 0.110 (see Tanaka and Tanaka [23] and Witze and Dwyer [24]).
Using the Pope [21] correction for the radial jet reduces the k-"
model-predicted spreading rate to 0.040. Hence, as noted by Rubel
[22], the round jet/plane jet anomaly has been exchanged for a round
jet/radial jet anomaly.
In contrast to the k-" model, as indicated in Table 1, the Wilcox [4]
k-! model predicts comparable spreading rates for both the round
and radial jets, both larger than the predicted plane-jet spreading rate.
Similarly, when a constant value of 0:0708 is used for the new
k-! model, the predicted round- and radial-jet spreading rates are
0.177 and 0.168, respectively. Numerical experimentation shows
that if is reduced to 0.06, the models spreading rates for both the
round and radial jets are close to the measured values. Because
Popes [21] argument implies nothing regarding the functional
dependence of the modication upon
p , it is completely consistent
to propose that depends upon this parameter in a manner that
reduces the value of as needed for both ows. Thus, as a
generalization of the Pope modication, the reformulated k-! model
uses the following prescription for :
o f
(24)
where
o 0:0708;
f
1 85
!
1 100
!
(25)
and
ij jk Ski
!
!3
(26)
Fig. 3 Far-wake and radially spreading-jet solutions: new k-! model (solid line), k-" model [19] (dashed line), RNG k-" model [25] (dotted line), and
measured [24,26,27] (open and lled circles).
2828
WILCOX
Fig. 4 Sensitivity of free-shear-ow spreading rates to freestream conditions: new k-! model (solid line), Wilcox [4] k-! model (dashed line), and k-"
model [19] (dotted line).
E.
F.
Turbulent/Nonturbulent Interfaces
as
y!
(27)
2829
WILCOX
Fig. 6 Attached boundary layers with strong adverse pressure gradient: computed values (solid line) and Samuel and Joubert experiment [29] (circles)
and Stratford [30] (circles).
Fig. 7 Supersonic and hypersonic boundary layers: Coles experiment [31] (circles) and Watson experiment [31] (circles).
9
>
nu do >
=
nk do
do >
>
;
n!
do
(28)
and
> do
(29)
do
nu
nk
n!
1.000
0.500
0.856
0.500
1.100
0.667
1.000
0.600
0.400
0.500
1.712
0.125
3.333
1.000
0.546
20
3.333
1.500
0.546
20
2.333
0.500
0:454
19
2830
WILCOX
Fig. 8 Computed and measured velocity defect near the boundarylayer edge for a at-plate boundary layer using three k-! models: new
(solid line), Kok (dashed line), Hellsten (dotted line), Klebanoff (circles),
Wieghardt and Tillman [33] (squares), and Winter and Gaudet [34]
(triangles).
IV.
Stress Limiter
The second key modication in the new k-! model occurs in the
expression for the eddy viscosity. In the new model, T is the ratio of
k to ! multiplied by a factor that is, effectively, a function of the
turbulence-kinetic-energy production-to-dissipation ratio. This
modication greatly improves the models predictions for supersonic
and hypersonic separated ows.
Note that this modication pertains to the proposed constitutive
relation between the Reynolds stresses and mean-ow properties,
rather than to the k-! model per se. The virtues of the stress limiter
(often referred to as a weakly nonlinear stress/strain-rate relationship) can be realized by using a nonlinear stress/strain-rate
relationship or even by computing the Reynolds stresses with a
stress-transport model based on the k and ! equations. As noted
p
k for
Pk
Cl2
lim
Dk
(34)
Stress-limiter coefcients
Reference
Coakley [11]
Durbin [35]
Menter [7]
Moore and Moore [36]
New k-!
C0
Clim
0
0
0
2.85
0
1.00
1.03
1.00
0.75
0.88
1
1
0
1
1
0
0
1
1
0
2831
WILCOX
Fig. 11 Computed and measured skin friction and surface pressure for
ow past a backward-facing step; ReH 37; 500: new k-! model (solid
line) Wilcox [4] k-! model (dashed line), and Driver-Seegmiller [42]
(circles).
2832
WILCOX
Fig. 12 Computed and measured skin friction and surface pressure for
ow past a backward-facing step; ReH 5000; new k-! model (solid
line), Wilcox [4] k-! model (dashed line), and Jovic and Driver [44]
(circles).
Fig. 13 Effect of the stress-limiter coefcient on computed reattachment length for a backward-facing step with ReH 37; 500.
Fig. 15 Effect of the stress limiter on the k-! model for Mach 2 ow past
a backward-facing step: with limiter (solid line), without limiter (dashed
line), and Samimy et al. [47] (circles).
WILCOX
2833
Fig. 16 Schematics of supersonic ow into a compression corner and shock-wave/boundary-layer interaction (reecting shock).
C.
computation was done with the new k-! model with and without the
stress limiter.
As shown in Fig. 15, with Clim 7=8, the stress limiter has a
barely noticeable effect on the computed surface-pressure
coefcient. Computed and measured values of Cp differ by less
than 7% for the entire oweld. Predicted reattachment length with
Fig. 17 Comparison of computed and measured surface pressure and skin friction for Mach 3 shock-separated ows using the new k-! model: with
limiter (solid line), without limiter (dashed line), Settles et al. [50] (circles), Dolling and Murphy [51] (squares), Reda and Murphy [52] (diamonds), and
Murthy and Rose [53] (circles).
2834
WILCOX
D.
rise is nearly the same for the two ows, this similarity conrms that
the k-! models predictions are consistent with the free-interaction
concept.
The numerical separation points for these ows are further
upstream than indicated by oil-ow measurements. Marshall and
Dolling [54] indicated that such ows include a low-frequency
separation-shock oscillation. Adams [55] found this oscillation in a
direct numerical simulation of a Mach 3 compression-corner ow.
The time-mean pressure distribution upstream of the corner is
affected by these oscillations, for which the frequency content
includes substantial energy at time scales of the mean motion. This
unsteadiness is responsible for the apparent mismatch between the
beginning of the pressure rise and the separation point. Because
computations with the k-! model are so close to measured properties,
yet display no low-frequency oscillation of the shock, we can
reasonably conclude that the computations effectively incorporate
the slow oscillation into the Favre-averaged ow variables.
Figure 18 indicates how the separation-point location for the
24 deg compression-corner ow varies with Clim . As shown, similar
to the effect for an incompressible backward-facing step (see
Fig. 13), xs increases monotonically as Clim increases. Selecting
Clim 7=8 yields a value of xs 1:82, which provides a very
close match to most details of this oweld.
Figure 19 shows that using Clim 1 produces a separation bubble
roughly twice the measured size. This explains why Menters [7]
model fares so poorly for this ow [56].
E.
We turn now to a hypersonic ow: namely, the Mach 11 shockwave/boundary-layer interaction investigated by Holden [57]. The
incident shock makes a 17.6 deg angle with the surface and increases
the static pressure by a factor of 70. The shock angle was adjusted
from the precise value used in the experiment to match the overall
Fig. 18 Effect of the stress-limiter coefcient on computed separationpoint location for Mach 3 ow into a 24 deg compression corner.
Fig. 19 Comparison of computed and measured surface pressure and skin friction for Mach 3 ow into a 24 deg compression corner: Menter [7]
k-!=k-" model (solid line), new k-! model with Clim 1 (dashed line), Settles et al. [50] (circles), and Dolling and Murphy [51] (squares).
2835
WILCOX
Fig. 20 Effect of the stress limiter on the new k-! model for a Mach 11
shock-wave/boundary-layer interaction: with limiter (solid line), without
limiter (dashed line), and Holden [57] (circles).
inviscid pressure rise for an assumed specic heat ratio 1:4. The
surface is highly cooled with a wall-to-adiabatic-wall temperature
ratio of Tw =Taw 0:2.
Figure 20 compares computed and measured surface pressure for
the new k-! model with and without the stress limiter. As shown, the
limiter increases separation-bubble length from 0.34 to 1:53o . The
computed surface-pressure rise is much closer to the measured rise
when the limiter is used. As with the Mach 3 applications of the
preceding subsection, this indicates that the predicted shock pattern
closely matches the experimental pattern. Holden [57] estimated the
size of the separation bubble to be about 1:00o . The surface-pressure
data suggest a separation bubble about twice that size.
As with all of the computations discussed in this paper, the
turbulent Prandtl number was chosen to be PrT 0:89. In general,
for this and other hypersonic shock-separated ows done with the
new k-! model (see Wilcox [14]), computed heat transfer at the
reattachment point is about 50% higher than measured. This
difculty is characteristic of turbulence models that base the
turbulent heat-ux vector on Reynolds analogy. As shown by Xiao
et al. [58], realistic predictions for hypersonic reattachment point
heat transfer can be achieved by constructing additional model
equations to compute the heat-ux vector.
V.
Conclusions
I.
j"h j
;
rp 1
"h
fine
coarse
fine
(A1)
Free-Shear-Flow Programs
Table A1
Flow
Far wake
Mixing layer
Plane jet
Round jet
Radial jet
0coarse
0fine
GCI
201
201
201
201
201
2
2
2
2
2
0.32640
0.09599
0.10780
0.09411
0.09914
0.32600
0.09643
0.10740
0.09388
0.09890
0.05%
0.19%
0.16%
0.10%
0.10%
2836
WILCOX
Table A2
Flow
Samuel and Joubert experiment [29]
Stratford [30]
Coles Mach 4.5 experiment [31]
Watson Mach 10.3 experiment [31]
Table A3
Fine-grid size
GCI
6
6
7
7
201
201
201
201
2
2
2
2
1.30875
0.67697
1.21393
0.23706
1.31154
0.67203
1.21390
0.23741
0.09%
0.31%
0.00%
0.06%
Flow
High Reynolds number incompressible backstep
Low Reynolds number incompressible backstep
Transonic bump
Mach 2 backstep
Mach 3 20 deg compression corner
Mach 3 24 deg compression corner
Mach 3 shock/boundary-layer interaction
Mach 11 shock/boundary-layer interaction
Figure
Fine-grid Size
xcoarse
xfine
GCI
11
12
14
15
17
17
17
19
301 163
201 163
201 101
401 201
401 201
401 201
401 201
500 300
1.25
1.25
1.25
1.25
1.41
1.41
1.41
1.19
7.158
7.451
0.414
2.728
1.151
2.785
4.603
1.519
7.070
7.280
0.410
2.672
1.160
2.825
4.650
1.530
3.1%
5.9%
2.6%
5.3%
1.1%
2.0%
1.5%
2.5%
Acknowledgment
The research presented in this paper was totally funded by DCW
Industries, Inc.
References
II.
A program named EDDY2C was used for ows with boundarylayer separation. The program uses the MacCormack [41] fully
implicit ux-splitting method with GaussSeidel line relaxation.
Computations on numerous ows with three different grids show
that the effective order of accuracy of the numerical algorithm is
typically p 1:8.
The ow property that generally takes longest to converge is the
size of the separated region. All computations reported in this paper
were run long enough to insure iteration convergence, with the
WILCOX
2837
2838
WILCOX
[61] Crank, J., and Nicolson, P., A Practical Method for Numerical
Evaluation of Solutions of Partial Differential Equations of the HeatConduction Type, Proceedings of the Cambridge Philosophical
Society (Mathematical and Physical Sciences), Vol. 43, No. 50, 1947,
pp. 5067.
[62] Rubel, A., and Melnik, R. E., Jet, Wake and Wall Jet Solutions Using a
k-" Turbulence Model, AIAA Paper 84-1523, 1984.
[63] Blottner, F. G., Variable Grid Scheme Applied to Turbulent Boundary
Layers, Computer Methods in Applied Mechanics and Engineering,
P. Givi
Associate Editor