Documente Academic
Documente Profesional
Documente Cultură
www.elsevier.com/locate/ijh
Abstract
The paper presents a numerical investigation of buet ¯ows using various turbulence models, including linear and non-linear
low-Re eddy-viscosity models (EVM). The accuracy of the models is assessed against experimental data for transonic ¯ows around
the NACA-0012 aerofoil. The study shows that non-linear two-equation models in conjunction with functional cl coecient for the
calculation of the eddy-viscosity (henceforth labelled NL-cl ), provide satisfactory results for transonic buet ¯ows. The compu-
tations also reveal that the Spalart±Allmaras one-equation model provides comparable results to the NL-cl models, while larger
inaccuracies are introduced by linear and non-linear models based on constant cl coecient. Moreover, the buet onset boundaries
are similarly predicted by the one-equation and NL-cl models. The study has been performed using a second-order time accurate
implicit-unfactored method which solves in a coupled fashion the Navier±Stokes and turbulence transport equations. The spatial
discretisation of the equations is obtained by a Riemann solver in combination with a third-order upwind scheme. Ó 2000 Begell
House Inc. Published by Elsevier Science Inc. All rights reserved.
0142-727X/00/$ - see front matter Ó 2000 Begell House Inc. Published by Elsevier Science Inc. All rights reserved.
PII: S 0 1 4 2 - 7 2 7 X ( 0 0 ) 0 0 0 5 3 - 9
G. Barakos, D. Drikakis / Int. J. Heat and Fluid Flow 21 (2000) 620±626 621
models have been validated for steady ¯ows, mainly two-di- part of the turbulent dissipation rate (in the case of the
mensional and incompressible, (e.g., Craft et al., 1996, Launder±Sharma model).
amongst others), while more recently experience has been ac- The matrix H J H~ has non-zero entries for the source
quired from applications to compressible ¯ows with shock/ terms of the turbulence model equations. J is the Jacobian of
boundary-layer interaction (e.g., Barakos and Drikakis, the transformation from Cartesian to curvilinear co-ordinate
2000b). system. E, G and R, S are the inviscid and viscous ¯uxes, re-
In the present work, various turbulence closures including spectively. The total energy per unit volume e is given by
algebraic, one-equation as well as linear and non-linear low-Re e qi
1=2q
u2 w2 qk, where i is the speci®c internal
two-equation models, are validated in transonic buet ¯ows. energy. The pressure is calculated by the ideal gas equation of
The assessment of the models is performed against experi- state.
mental results for buet around the NACA-0012 aerofoil at A third-order upwind scheme in conjunction with a char-
Reynolds number Re 107 , a range of Mach numbers between acteristic-based ¯ux averaging is used to calculate the inviscid
0:7 and 0:85, and for incidence angles between 0° and 5° ¯uxes at the cell faces (Eberle et al., 1992; Drikakis and Durst,
(McDevitt and Okuno, 1985). 1994). Limiters based on the squares of pressure derivatives
have been used in detecting shocks and contact discontinuities.
An implicit-unfactored solver (Barakos and Drikakis, 1998b,
2. Numerical method 1999) has been employed for the solution of the equations. A
sequence of approximations qm such that: limm>1 qm ! U n1 is
The numerical simulations have been carried out using an de®ned between two time steps n and n 1. Using implicit time
implicit CFD solver (Barakos and Drikakis, 1998b, 1999) de- discretization and after linearizing the ¯uxes around the sub-
veloped for unsteady and turbulent aerodynamic ¯ows. The iteration state m the following form is derived:
main feature of the method is the strong coupling of turbu- Dq
lence models with the Navier±Stokes equations, via an implicit
Aminv Dqn
Cinv
m
Dqf ÿ
Amvis Dqn ÿ
Cvis
m
Dqf RHS;
Dt
unfactored scheme and a Riemann solver, the latter being used
in conjunction with a third-order upwind interpolation scheme
3
(Drikakis and Durst, 1994). where
The compressible Navier±Stokes equations for a two-
dimensional curvilinear co-ordinate system
n; g, in conjunc- qm ÿ U n
RHS ÿ Enm Gmf ÿ Rmn ÿ Sfm ÿ H m ;
4
tion with the transport equations of the turbulence model, are Dt
written in matrix form as
Dq qm1 ÿ qm
5
oU oE oG oR oS
H:
1 and
ot on of on of
U is the six-component vector of the conservative variables oE oG oR oS
Ainv ; Cinv ; Avis ; Cvis :
6
oU oU oU oU
T ;
U J
q; qu; qw; e; qk; q~
2
At each time step the ®nal system of algebraic equations is
where q is the density, u, w are the velocity components in the solved by a point Gauss±Seidel relaxation scheme. According
x- and z-directions, respectively, e the total energy per unit to the present method, the transport equations for the turbu-
volume, k the turbulent kinetic energy and ~ is the isotropic lence model are solved coupled with the ¯uid ¯ow equations.
622 G. Barakos, D. Drikakis / Int. J. Heat and Fluid Flow 21 (2000) 620±626
This strategy provides fast convergence and compact numeri- This cubic expansion has been utilized here to calculate the
cal implementation. components of the Reynolds-stress tensor ÿq ui uj . In the
For unsteady ¯ow simulations the discretisation of the time above, Sij and Xij are the strain and vorticity tensors, while S~
derivative is obtained by a second-order scheme (Barakos and and X ~ are their normalized invariants
Drikakis, 1999)
k q q
~ k Xij Xij =2;
1:5U n1 ÿ 2U n 0:5U nÿ1 S~ Sij Sij =2; X
14
~ ~
Ds and the coecients ci take the values: c1 ÿ0:1; c2 0:1;
ÿ Enn1 Fgn1 ÿ Rn1
n ÿ Sgn1 ÿ H n1 :
7 c3 0:26; c4 ÿ10c2l . The eddy viscosity is calculated by
lT cl qfl
k 2 =~
, where
In time accurate computations, the time marching must be
performed using the same time step in all cells of the compu- 0:31 ÿ exp fÿ 0:36 exp
0:75gg
tational domain. This global time step is de®ned for a given cl ;
15
1 0:35g1:5
CFL number by
Ds 6 Dsmax 8 !1=2 !2 9
0 1 < R~t R~t =
fl 1 ÿ exp ÿ ÿ ;
16
B CFL C : 90 400 ;
min @ J q A ;
8
2 2 2 2
kmax 2
lcp =Pr
nx nz fx fz
i;k
where kmax is the maximum eigenvalue calculated using the ~X
g max S; ~ :
17
solution from the previous time step.
Such functional form of cl was found to be bene®cial in ¯ows
far from equilibrium and similar conclusions have also been
3. Turbulence modelling reported by Liou and Shih (1996), Huang (1999) and Bardina
et al. (1997) for a variety of compressible ¯ows. The non-linear
In the present study, the following models have been em- eddy-viscosity model of So®alidis and Prinos (1997) is actually
ployed: the algebraic Baldwin and Lomax (1978) model, the the k±x version of the non-linear k± model of Craft et al.
one-equation model of Spalart and Allmaras (1992), the (1996).
Launder and Sharma (1974) and Nagano and Kim (1988)
linear k± models, as well as the k±x version (So®alidis and
Prinos, 1997) of the non-linear eddy-viscosity model of Craft
et al. (1996). 4. Simulation of transonic buet
In the case of linear EVM the stress tensor sij is modelled
using the Boussinesq approximation 4.1. Test cases
sij slij sR
ij ;
9 Computations were carried out for the experimental cases
of McDevitt and Okuno (1985). Their experiments have been
where performed for the NACA-0012 aerofoil at Mach numbers
oui ouj 2 ouk between 0:7 and 0:8, angles of incidence less than 5° and Rec
slij l ÿ l dij ;
10 number between 1 and 14 millions. McDevitt and Okuno
oxj oxi 3 oxk
identi®ed the incidence-angle and Mach number as the most
important parameters for the buet onset. Their wind-tunnel
oui ouj 2 ouk 2
sR
ij lT ÿ lT dij ÿ qk dij
11 results are particularly suitable for validating CFD codes
oxj oxi 3 oxk 3 because they are free of wall eects in contrast to previous
and lT is the eddy-viscosity. experimental studies (McDevitt et al., 1976).
Non-linear EVM use an expansion of the Reynolds stress McDevitt and Okuno (1985) organized their experiments in
components in terms of the mean strain-rate and rotation six sets and the corresponding parameters are shown in Table 1.
tensors For the sets numbered as 4, 5 and 6, buet was reported and,
ÿ ÿ consequently, these sets were considered in the present work.
Sij Ui;j Uj;i =2; Xij Ui;j ÿ Uj;i =2:
12 As has also been reported by Mateer et al. (1992), the eects of
In the case of the non-linear k± EVM of Craft et al. (1996) a boundary-layer tripping on the obtained results for Re about
cubic expansion for the anisotropy of the Reynolds stress 106 , is negligible. Therefore, in the present study computations
tensor, aij ui uj =k ÿ
2=3dij , is employed were carried out for Reynolds number Re 106 which pro-
vides fully turbulent ¯ow (McDevitt and Okuno, 1985).
l l 1
aij ÿ T Sij c1 T Sik Skj ÿ Skl Skl dij
qk q~ 3
Table 1
l ÿ l 1 Nominal conditions for the experiments of McDevitt and Okuno
c2 T Xik Skj Xjk Ski c3 T Xik Xjk ÿ Xlk Xlk dij
q~
q~ 3 (1985)
l kÿ Set Incidence a (deg) Mach number Re (10ÿ6 )
c4 T 2 Ski Xlj Skj Xli Skl
q~
1 2 0.75 1.2±13.9
2 0 0.75 4.0±12.2
l k 2
c5 T 2 Xil Xlm Smj Sil Xlm Xmj ÿ Slm Xmn Xnl dij 3 0 0.8 1.2.0±12.1
q~
3 4 1 0.8 1.0±10.3
lT k l k 5 2 0.775 1.0±9.9
c6 Sij Skl Skl c7 T 2 Sij Xkl Xkl :
13 6 4 0.725 1.0±9.3
q~
2 q~
G. Barakos, D. Drikakis / Int. J. Heat and Fluid Flow 21 (2000) 620±626 623
4.2. Results the case of transonic buet, the self-excited shock oscillations
are also associated with shock/boundary-layer interaction and
Flow unsteadiness around a lifting surface may originate ¯ow separation. It is thus important for a turbulence model to
from the motion of the boundary or from unsteady free-stream
conditions. However, in the case of buet the induced un-
steadiness is due to ¯ow non-linearities associated with certain
combinations of Re, Mach number, and angle of incidence. In
Table 2
Details of the computational grids employed in the calculations; grid
G4 was selected for buet calculations
Grid i-direction k-direction Far-®eld location
G1 180 60 5c
G2 241 80 5c
G3 291 85 7c
G4 361 90 7c
predict accurately the separation induced by the interaction of the past for steady compressible ¯ow computations (Bardina
the shock with the boundary layer and, subsequently, the et al., 1997; Huang, 1999). All linear k± models employed in
buet onset. this study predicted the shock position shifted downstream and
In the present study, several computational grids have been underestimated the length of separation region. The same was
employed to ensure grid-independent solutions and their de- also the case for the algebraic Baldwin±Lomax model
tails are given in Table 2. In addition, calculations have been (Fig. 1(b)).
performed for various dimensions of the computational do- In Fig. 2, comparison of numerical and experimental results
main to ensure independence of the solution from the far-®eld for the buet onset is presented. There is a well-de®ned region
boundary conditions. For buet predictions the grid G4 of Mach and incidence angle where buet occurs. Initially,
(Table 2) was used. four computations (Fig. 2) were performed at conditions below
In Fig. 1, the Cp distributions, for M 0:775 and a 4°, the experimentally reported buet onset and steady-state so-
using various closures and dierent grids are compared with lutions were achieved (symbols in Fig. 2 labelled ``no SIO
the experimental results. For this Mach number and incidence (shock-induced oscillation)''). Afterwards, the incidence-angle
angle, the ¯ow has been found (McDevitt and Okuno, 1985) to was slowly increased to obtain unsteadiness and it was found
be steady and all turbulence models predicted steady ¯ow, as that after the initial peak of the Cl curve (Fig. 3) the compu-
well. As can be seen, none of the models was able to capture tations resulted either in periodic loads, thus indicating buet
exactly the experimental shock position. The non-linear (symbols in Fig. 2 labelled ``SIO''), or in steady-state ¯ow. In
models were used in conjunction with both functional cl (Eq. the latter case, the computations were repeated for a higher
(15)) and constant cl (cl 0:09) coecient. When the models incidence-angle until buet is captured. Once buet was pre-
were employed with a constant cl , were found to give results dicted, the incidence-angle was again decreased and the com-
(Fig. 1(c)) similar to the ones obtained by the linear k± and putation was repeated to check whether the experimental
algebraic models (Fig. 1(b)). The Launder±Sharma and Nag- boundary (solid line in Fig. 2) for buet onset could be closer
ano±Kim models provided similar predictions (plots are not approached. Computations were performed for a long time
shown here). The results obtained by using functional cl were interval to verify that almost periodic loads are obtained for
in better agreement with the experimental data. Computations the buet conditions (see Fig. 3(c)).
without the non-linear expansion revealed that the models For all combinations of Mach number and incidence angle
predictions were mainly dominated by their damping functions considered here, the linear k± models led to a steady solution
and functional cl , and it seems that the anisotropic stress ex- (Fig. 3(a), (b)), thus failing to predict buet. As can be seen in
pansion does not play any important role in this case. The Fig. 2, the computations predict the buet onset boundary
results using the Spalart±Allmaras model were comparable to slightly shifted to higher incidence angles and Mach number.
those obtained by the non-linear models using functional cl . This is similar to what Girodroux-Lavigne and LeBalleur
Similar conclusions about the eects of varying cl coecient (1988) have obtained. Edwards (1996), however, reported
on turbulence models performance have also been reported in results closer to the experimental data using an inverse
Fig. 4. Mach number ®eld around a NACA-0012 aerofoil at dierent time instants during the buet development.
G. Barakos, D. Drikakis / Int. J. Heat and Fluid Flow 21 (2000) 620±626 625
Acknowledgements
References
Marvin J.G., Huang, G.P., 1996. Turbulence modeling ± progress and McDevitt, J.B., Levy Jr., L.L., Deiwert, G.S., 1976. Transonic ¯ow
future outlook. In: Keynote Lecture Presented at the 15th about a thick circular-arc airfoil. AIAA J. 14, 606±613.
International Conference on Numerical Methods in Fluid Dynam- Nagano, Y., Kim, C., 1988. A two equation model for heat transport
ics, June 1996, Monterey, CA, USA. in wall turbulent shear ¯ows. J. Heat Transfer 110, 583±589.
Mateer, G.G., Seegmiller, H.L., Hand, L.A., Szodruch, J., 1992. An So®alidis, D., Prinos, P., 1997. Development of a non-linear strain-
experimental investigation of a supercritical airfoil at transonic sensitive k±x turbulence model. In: Proceedings of the 11th
speeds, NASA TM-103933, NASA Ames, CA, USA. Symposium on Turbulent Shear Flows, TSF-11, Grenoble, France,
McDevitt, J.B., Okuno, A.F., 1985. Static and dynamic pressure p2-89±p2-94.
measurements on a NACA 0012 airfoil in the Ames high Reynolds Spalart, P.R, Allmaras, S.R., 1992. A one-equation turbulence model
number facility, NASA-TP-2485, NASA Ames, CA, USA. for aerodynamic ¯ows, AIAA Paper 92-0439.