Sie sind auf Seite 1von 10

See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/298895489

THE k-epsilon-R-t TURBULENCE CLOSURE

Article  in  Engineering Applications of Computational Fluid Mechanics · June 2009

CITATIONS READS

0 364

4 authors, including:

U. Goldberg Oshin Peroomian


Metacomp Technologies, Inc. 53 PUBLICATIONS   521 CITATIONS   
96 PUBLICATIONS   1,588 CITATIONS   
SEE PROFILE
SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Rough surface predictions by single equation turbulence models. View project

Hybrid RANS/LES View project

All content following this page was uploaded by U. Goldberg on 24 March 2016.

The user has requested enhancement of the downloaded file.


Engineering Applications of Computational Fluid Mechanics Vol. 3, No. 2, pp. 175–183 (2009)

THE k- ε -R t TURBULENCE CLOSURE


U. Goldberg*, O. Peroomian, P. Batten and S. Chakravarthy

Metacomp Technologies, Inc., 28632 Roadside Drive, Suite 255, Agoura Hills, California 91301, U.S.A.
* E-Mail: ucg@metacomptech.com (Corresponding Author)

ABSTRACT: A turbulence closure, based on transport equations for the turbulence kinetic energy, k, its
dissipation rate, ε and the undamped eddy viscosity, R, is presented. The model, which is free of topography-
dependent parameters, combines a k-ε closure with the R t model so that no inflow turbulence decay occurs in
external flows, an attribute often sought by aeronautical engineers using CFD for flow computations. The model is
shown to revert to the k-ε closure in near-wall flow regions. Two aerodynamic flow cases are presented, comparing
the original k-ε closure to the current 3-equation model.
Keywords: turbulence decay, CFD, 3-equation closure

1. INTRODUCTION maintains inflow turbulence levels undecayed all


the way to the body. By setting the final eddy
A common attribute of the various k-ε models is viscosity as the maximum of the two models’
the decay of external flow turbulence from its contributions, a non-decaying eddy viscosity field
inflow level to that prevailing in the vicinity of results. Unlike the freestream production
the body’s front portion. While this is correct approach, the present method achieves the goal
behavior in the absence of turbulence generation without imposing any of the restrictions and
mechanisms (such as atmospheric disturbances) it limitations mentioned above, thus rendering the
is often inconvenient from a computational view engineer’s task considerably easier and less error
point: in subsonic/transonic flows the outer prone.
boundary must be placed at a large distance from In the present paper the k-ε model of Goldberg,
the body (30–50 characteristic body lengths is Peroomian and Chakravarthy (1998) is coupled
typical) to prevent body-induced pressure signals with the R t closure of Goldberg (2003) to produce
from reaching the outer boundary. Without the 3-equation k-ε-R t turbulence closure (also
extreme diligence in the specification of inlet called the “KERT” model). Two external flow
conditions, turbulence levels predicted with k-ε cases are presented, comparing the performance
or k-ω closures, can, under such conditions, decay of this closure with that of the corresponding k-ε
to the point of causing laminar flow at the body model.
surface. One way to keep inflow turbulence levels
all the way to the body is to add a freestream 2. HIGHLIGHTS OF THE NUMERICAL
production term to the transport equations so as to APPROACH
cancel the forcing terms, leaving only advection
active. While this method certainly works, special CFD++ (Chakravarthy, 1999), a general Navier-
treatment is necessary in formulating the Stokes flow solver, was used here. This code
freestream production term in order to prevent it features a second order Total Variation
from influencing near-wall regions. Also, the Diminishing (TVD) discretization based on a
freestream turbulence length-scale must be large multi-dimensional interpolation framework. For
enough to maintain ε ∞ relatively small, otherwise the results presented here, an HLLC (Harten, Lax,
it may overwhelm the near-wall levels of ε . In van Leer, with Contact wave) Riemann solver
addition, the modified sink term in the ε equation was used to define the (limited) upwind fluxes.
must be kept non-positive to avoid the occurence This Riemann solver is particularly suitable for
of unbounded growth in ε which would cause high-speed flow applications since, unlike
local laminarization. These restrictions impose an classical linear solvers such as Roe’s scheme, it
extra burden on engineers trying to compute such automatically enforces entropy and positivity
flows with confidence. An alternative approach, conditions (Batten, Leschziner and Goldberg,
presented here, is to couple the k-ε model with 1997). Further details on the numerical
another type of turbulence closure which naturally methodology in CFD++ can be found in

Received: 22 Aug. 2008; Accepted: 20 Oct. 2008

175
Engineering Applications of Computational Fluid Mechanics Vol. 3, No. 2 (2009)

Chakravarthy (1999), Chakravarthy, Peroomian and Sekar (1996), Peroomian, Chakravarthy and Goldberg
(1997) and Peroomian et al. (1998).

3. THE k- ε -R t TURBULENCE MODEL

3.1 Model formulation


In this linear model Reynolds stresses are related to the mean strain through the Boussinesq approximation:
⎛ 2 ⎞ 2
− ρ u i u j = μ t ⎜ U i , j + U j , i − U k , k δ ij ⎟ − ρ k δ ij (1)
⎝ 3 ⎠ 3
Transport equations for k, ε and R:
∂ ( ρk ) ∂ ∂ ⎡ ⎛⎜ μt ⎞ ∂k ⎤

+ ( U j ρk ) = ⎢⎜ μ+ ⎟ ∂x ⎥ + Pk − ρε (2)
∂t ∂x j ∂x j ⎣⎢ ⎝ σk ⎠ j ⎥⎦

∂ ( ρε ) ∂ ∂ ⎡ ⎛⎜ μt ⎞ ∂ε


+ ( Uj ρε ) = ⎢⎜ μ+ ⎥ + ( C ε 1 Pk − C ε 2 ρε + C ε 3 E ) / T t (3)
∂t ∂x j ∂x j ⎢⎣ ⎝ σε ⎟ ∂x ⎥⎦
⎠ j

∂ ( ρR ) ∂ ∂ ⎡⎛ μt ⎞ ∂R ⎤
+ ( Uj ρR ) = ⎢ ⎜⎜ μ + ⎟
⎟ ∂x ⎥ + ( C1 − C 2 f 2 ) ρRPk − ρC 3 D (4)
∂t ∂x j ∂x j ⎢⎣ ⎝ σR ⎠ j ⎥⎦

P k is the turbulence production − ρ u i u j U i , j which, when modeled in terms of the Boussinesq concept,
Eq. (1), reduces to:
⎡ ⎛ ∂U i ∂U j 2 ∂U k ⎞ 2 ⎤ ∂U i
Pk = ⎢ μ t ⎜ + − δ ij ⎟ − ρ kδij ⎥ (5)
⎢⎣ ⎜ ∂x j ∂x i 3 ∂x k ⎟ 3 ⎥⎦ ∂x j
⎝ ⎠
The extra source term in the ε equation is

E=ρ εT t max { k , (ν ε ) 1 / 4 } ⋅ max ⎧⎪⎨ ∂k ∂τ ⎫⎪


,0⎬ (6)
⎪⎩ ∂x j ∂x j ⎪⎭

The large eddy time-scale is


⎧ ∂R ∂R
⎪ λ>0
τ = k /ε (7) D = ⎨ ∂x j ∂x j (10)
and the realizable time-scale, which reduces to the ⎪
⎩0 λ≤0
corresponding Kolmogorov scale for small
(dissipative) eddies, is where

{
T t = τ ⋅ max 1 , 2/ Rt } (8) λ=
∂Q ∂R
(11)
∂x j ∂x j
where R t = k τ / ν , the turbulence Reynolds number.
The R equation’s destruction term involves the Q= ( U − U 0 ) 2 + ( V − V0 ) 2 + ( W − W0 ) 2 (12)
damping function
This term is active only in the immediate vicinity
tanh χ of walls; further away it vanishes. In principle,
f2 = , χ = R /( C μ ν ) (9)
tanh 2C 1μ/ 4 χ
near-wall damping functions should vary as 1/y at
the immediate vicinity of walls to enforce
f 2 = 1 except in the near vicinity of a wall, where μ t ~ y 3 . However, in the close proximity of walls
it grows to very large values (see Fig. 2 in μ t << μ and it is advantageous, in practice, to
Goldberg (2003)). The extra sink term is avoid the 1/y behavior. In the present work the
damping functions suggested in Goldberg,
Peroomian and Chakravarthy (1998) and
Goldberg (2003) are adopted:

176
Engineering Applications of Computational Fluid Mechanics Vol. 3, No. 2 (2009)

− Aμ R t kw =0 (20)
f μ , k −ε =
1− e

{
max 1 , 2 / Rt } (13)
1− e
Rt
And, since R ~ k 2 / ε with ε finite at walls (see
Eq. (22),
tanh α χ 2
f μ, R = (14) Rw=0 (21)
tanh β χ 2
whereas
The eddy-viscosity field is given by:
{
μ t , k −ε = min C μ f μ , k −ε ρk τ , φ ρ k / S } (15) εw = 2 ν ∂( k / ∂y ) 2
1 (22)

μ t , R = f μ , R ρR (16) where y is the wall-normal coordinate.


The following changes to the current R t model,
μ t = max { μ t , k −ε , μ t , R } (17) compared to the original stand-alone closure
(Goldberg, 2003) have been made: (1) the
where S is the mean strain magnitude. In coupled model uses the turbulence production
Eq. (15) the eddy-viscosity from the k-ε model is term P k , also used by the k and ε transport
limited by imposing the isotropic version of the equations. (2) The value of α changed, based on
2 model re-calibration.
Schwartz Inequality, e.g., uυ ≤ uu ⋅ υυ . This
realizability constraint is responsible, among 4. MODEL EVALUATION
other things, for avoiding the unphysical build-up
of turbulence kinetic energy in flow stagnation In the following external aerodynamic flow cases
regions (e.g., a blunt nose). Two choices exist for the k-ε-R t model’s performance is compared with
the parameter φ : 2/3 (Schwartz or “weak” experimental data as well as with predictions
realizability) and 0.31 (Bradshaw or “strong” by the corresponding k-ε closure (Goldberg,
realizability). The latter is usually recommended Peroomian and Chakravarthy, 1998).
for impinging or high speed flows.
4.1 NACA 63210 airfoil
3.2 Model constants
This case is presented to demonstrate the essence
σ k = 1.0, σ ε = 1.3, σ R = 1.0,C ε 1 = 1.44, of the difference between the k-ε-R t and k-ε
C ε 2 = 1.92, C ε 3 = 0.3, A μ = 0.007, α = 0.00015, closures under typical aerodynamic flow
β = 0.2, C 1 = C 2 + κ 2 ( C 3 − 1 / σ R ) = 39.918 , conditions, with a far-field boundary placed 100
C 2 = ( 12 / 11 ) β / α = 39.834 , C 3 = 3 /( 2 σ R ) = 1.5 . cord lengths away from the body. The 90,000 size
grid consists of a prism layer around the airfoil,
From Eqs. 13–16 and the above constants, the +
with y max < 0.7 , and triangular elements in the
following relation is derived when R t << 1 :
rest of the domain. Figure 1 shows the mesh
μ t , R / μ t , k −ε = α / ( )
2 A μ β ≅ 0.076 (18) in the vicinity of the airfoil while Figs. 2 and
3 show grid details in the leading
Thus, according to Eq. (17), and trailing edge regions, respectively.
μ t = μ t , k −ε , R t << 1 (19) Flow conditions are as follows:
α = 0 , M ∞ = 0.17 , Re ∞ = 9.2 × 10 6 , Tu ∞ = 0.5 %
Analyzing Eqs. (13)–(16) in the logarithmic and l ∞ = 1 mm . Figures 4 and 5 are eddy
overlap leads to the conclusion that Eq. (19) is
viscosity contour plots resulting from the k-ε and
valid from the wall to y + ≈ 200 , beyond which k-ε-R t computations, respectively. The former
μ t , R = μ t , k −ε . This indicates that in near-wall exhibits strong turbulence level decay between
regions (viscous sublayer and lower portion of the inflow and airfoil; the latter shows no decay.
log layer) the eddy viscosity field is determined Consequetly, there is no turbulence present at the
by the k-ε model. airfoil's leading edge area when the k-ε closure is
used (see Fig. 6) but it is present when the k-ε-R t
3.3 Wall boundary conditions model is invoked (Fig. 7). C p and C f profiles are
seen in Figs. 8 and 9, respectively. The pressure
The k and R transport equations are subject to
distribution predicted by the two models is
simple Dirichlet boundary conditions at solid
practically identical but differences are observed
walls.
in the skin friction profiles. Specifically, the k-ε

177
Engineering Applications of Computational Fluid Mechanics Vol. 3, No. 2 (2009)

result shows laminar-to-turbulent flow transition


on both pressure and suction sides of the airfoil.
This is a direct consequence of turbulence decay
and its effect, as seen also in Fig. 6. There is,
however, an overshoot of C f on both pressure and
suction sides at the locations where the flow
becomes fully turbulent. These overshoots raise
skin friction and impose a net increase in viscous
drag, relative to the k-ε-R t prediction, translating
into a 0.4% decrease in L/D from the k-ε-R t
model solution to that of the k-ε closure.

Fig. 4 NACA 63210: eddy viscosity in


computational domain using k-ε closure.

Fig. 1 NACA 63210: airfoil mesh.

Fig. 5 NACA 63210: eddy viscosity in


computational domain using k-ε-R t closure.

Fig. 2 NACA 63210: grid in leading edge zone.

Fig. 6 NACA 63210: turbulence kinetic energy at


leading edge using k-ε closure.

Fig. 3 NACA 63210: grid in trailing edge zone.

178
Engineering Applications of Computational Fluid Mechanics Vol. 3, No. 2 (2009)

Fig. 7 NACA 63210: turbulence kinetic energy at


leading edge using k-ε-R t closure.

Fig. 10 Trap Wing: photo of test model in wind-


tunnel.

4.2 Trapezoidal high-lift wing


Experimental data for this wing configuration are
provided in Johnson, Jones and Madson (2000).
The wind-tunnel model consisted of a body pod, a
half-span wing, a slat and a flap. Figure 10 shows
the model inside the tunnel’s test section. Two
grids were generated using Metacomp’s MIME
Fig. 8 NACA 63210: pressure coefficient
(Multipurpose Intelligent Meshing Environment)
profiles predicted by the k-ε and k-ε-R t which, in turn, used a CATIA V5 geometry model.
closures. The “baseline” mesh consisted of 23 million
elements; the “refined” grid had 39 million
elements. Freestream conditions were M ∞ = 0.2 ,
Re c = 4.3 ×10 6 and α = 30°. Wall boundary
conditions: tunnel walls were treated as inviscid,
all wing surfaces were treated as adiabatic viscous
walls. Since the grids had y + < 1 at the first off-
wall layer, all transport equations were solved to
the walls; no wall functions were used. There was
only a 0.13% difference in L/D between the
baseline and refined grids (4.111 and 4.116,
respectively) and all results presented below are
from the refined mesh. Figure 11 is a pressure
contour plot which also provides a view of the
overall topology. Figure 12 shows the mesh
around the wing at the mid-span section.
Figure 13 is the residual plot from the k-ε-R t
Fig. 9 NACA 63210: skin friction profiles predicted
calculation. Residuals dropped seven or more
by the k-ε and k-ε-R t closures.
orders of magnitude and convergence was
achieved after about 500 iterations. Figure 14
shows the evolution of lift and drag coefficients,
indicating that forces converged in less than 300
iterations. Both models’ results are plotted,

179
Engineering Applications of Computational Fluid Mechanics Vol. 3, No. 2 (2009)

showing close agreement. This is a consequence


of the fact that the flow over the body is predicted,
in both cases, by the k-ε closure (Eq. 19). Figures
15 and 16 show C l vs. α and C l vs. C d (drag
polar) respectively. Predictions at several angles-
of-attack compare favorably with corrected wind-
tunnel data. Figure 17 shows eddy viscosity
contours on the mid-span plane, resulting from
the k-ε-R t computation. There is no change in μ t
levels from inflow to close vicinity of the wing.
Consequently, as seen in Fig. 18, wall turbulence
starts evolving from the slat’s leading edge. In
contrast, the k-ε calculation exhibits strong Fig. 13 Trap Wing: k-ε-R t residual plot.
turbulence decay, seen in Fig. 19, resulting in
delayed build-up of wall turbulence over the slat
(Fig. 20). This imposes less viscous drag on the
wing, as seen in Fig. 21.

Fig. 14 Trap Wing: k-ε-R t and k-ε forces evolution.


Fig. 11 Trap Wing: k-ε-R t surface pressure plot and
wing topology.

Fig. 12 Trap Wing: mid-span mesh section.

Fig. 15 Trap Wing: Cℓ vs. α.

180
Engineering Applications of Computational Fluid Mechanics Vol. 3, No. 2 (2009)

Fig. 19 Trap Wing: k-ε mid-span eddy viscosity


contours.

Fig. 16 Trap Wing: drag polar.

Fig. 20 Trap Wing: k-ε turbulence kinetic energy


contours at wing leading edge.

Fig. 17 Trap Wing: k-ε-R t mid-span eddy viscosity


contours.

Fig. 21 Trap Wing: viscous drag comparison between


k-ε and k-ε-R t models.
Fig. 18 Trap Wing: k-ε-R t turbulence kinetic energy
contours at wing leading edge. 5. CONCLUDING REMARKS

This paper tested the performance of the k-ε-R t


turbulence closure and compared it with that of
the corresponding k-ε model. Whereas the former
maintains turbulence levels from inlet to body, the
latter’s inflow levels usually decay and may
disappear completely by the time the body is

181
Engineering Applications of Computational Fluid Mechanics Vol. 3, No. 2 (2009)

reached. The effect of these two behaviors on α constant in damping function


predicted aerodynamic forces (eg., lift and drag) (Eq. 14), angle-of-attack (context
was shown. In many flow cases of engineering dependent)
interest, turbulence levels are better estimated in β constant in damping function
the vicinity of the flight vehicle than far upstream, (Eq. 14)
yet the computational domain must often be δ boundary layer thickness
extended well upstream of the body. This renders δij Kronecker delta
the k-ε-R t model a practical turbulence closure for ε turbulence kinetic energy
a large class of external aerodynamic flow dissipation rate
problems. The model was shown to revert back to κ 0.41 (von Kármán constant)
the corresponding k-ε closure in near-wall flow μ dynamic molecular viscosity
regions. μt dynamic eddy viscosity
ν μ / ρ , kinematic molecular viscosity
NOMENCLATURE ρ density
σk , σε , σR turbulent diffusion coefficients
Aμ coefficient in damping function τ large eddy time-scale, shear stress
(Eq. 13) (context dependent)
Cf skin friction coefficient
φ realizability constant (Eq. 15)
Cp pressure coefficient
χ turbulence Reynolds number in R
Cε 1 , Cε 2 , Cε 3 constants (Eq. 3)
model (Eq. 9)
Cμ 0.09, eddy-viscosity coefficient
D damping function (Eqs. 4, 10) ∇ gradient operator
E extra source term in ε equation
(Eqs. 3, 6) Subscripts
f μ , k −ε damping function for k-ε model
(Eq. 13) c corrected for compressibility
f 2 , fμ,R damping functions for R model effects
(Eqs. 9, 14) i or j Cartesian component in the i- or j-
k turbulence kinetic energy direction
l turbulence length scale i,j j-direction derivative of the
M Mach number i-component
Pk turbulence production (Eq. 5) t turbulent
Q relative velocity magnitude w evaluated at the wall
(Eq. 11) x based on streamwise direction
R undamped eddy viscosity 0 reference state, stagnation
Rt turbulence Reynolds number conditions (context dependent)
(Eq. 8) 1 wall-adjacent cell centroid
Re Reynolds number ∞ evaluated at the freestream
S mean strain τ based on friction velocity
T time scale, temperature (context
dependent)
Tu turbulence intensity REFERENCES
t time
U,V,W Cartesian mean velocity 1. Batten P, Leschziner MA, Goldberg UC
components (1997). Average-state Jacobians and implicit
u,v,w Cartesian fluctuating velocity methods for compressible viscous and
components turbulent flows. Journal of Computational
ui u j Reynolds stress tensor Physics 137:38–78.
2. Chakravarthy S (1999). A unified-grid finite
uτ ( τ / ρ ) 1W/ 2 , friction velocity volume formulation for computational fluid
x,y,z Cartesian streamwise, normal and dynamics. Int. J. Numer. Meth. Fluids
transverse coordinates 31:309–323.
y+ yu τ / v w , inner layer 3. Chakravarthy S, Peroomian O, Sekar B
nondimensional coordinate (1996). Some internal flow applications of a

182
Engineering Applications of Computational Fluid Mechanics Vol. 3, No. 2 (2009)

unified-grid CFD methodology. AIAA Paper


96-2926.
4. Goldberg U (2003). Turbulence closure with
a topography-parameter-free single equation
model. Int. J. of CFD 17(1):27–38.
5. Goldberg U, Peroomian O, Chakravarthy S
(1998). A wall-distance-free k-ε model with
enhanced near-wall treatment. ASME J.
Fluids Engrg. 120(3):457–462.
6. Johnson P, Jones KM, Madson M (2000).
Experimental investigation of a simplified 3D
high lift configuration in support of CFD
validation. AIAA Paper 2000-4217.
7. Peroomian O, Chakravarthy S, Goldberg U
(1997). A “grid-transparent” methodology for
CFD. AIAA Paper 97-0724.
8. Peroomian O, Chakravarthy S, Palaniswamy
S, Goldberg U (1998). Convergence
acceleration for unified-grid formulation
using preconditioned implicit relaxation.
AIAA Paper 98-0116.

183

View publication stats

Das könnte Ihnen auch gefallen