Professional Documents
Culture Documents
Abstract
A traDSport equation for the turbulent viscosity was
usembled, using empiricism and arguments of dimenIIional analysis, Galilean invarianee, and selective dependence on the molecular viscosity. It has similarities
":'lith the models of Nee &: Kovasznay, Secundov et .1.,
and ~aldwin &: Barth. The equation includes a destruction term that depends on the distance to the wall, related to the one in Secundov's model and to one due
to Hunt. Unlike early one-equation models the resulting turbulence model is local (i.e . the equation at one
point does not depend on the solution at other points),
and therefore compatible with &rids of any structure
and Navier-Stokes solvers in two or three dimensions.
It is numerically forgiving, in terms of near-wall resolution and stil"ness, and yields fairly rapid convergence to
steady state. The wall and freestream bonndary conditions are trivial. The model yields relatively smooth
Iaminar-tur!rulent t_ition, at points epecified by the
user. It is powerful enough to be calibrated on 2-D mixing layers, wake&, and flat-pla~ bonndary layers, wh~h
we Itonsider to be the building blocks for aerodynanuc
lows. It yields satisfactory predictions of bonndary
layers in preaure gadients. Its Dumerical implementation in a 2-D steadY.fltate Navier-Stokes solver has
beeR completed and is discussed. The cases presented
include .bock-induced eep&ration and a blunt trailing
edge. The model locates Iihocks slightly farther forward
than the JoImsoD-King model. It perfonm well in the
near wake and appears to be a good candidate for more
complex flows such as high-lift systems or wing-body
junctions. However, _ it is not clear whether steady
soIutioc'! will or should be obtained.
_ics.
,+
two-equation models are the simplest complete models. The Nee-Kovasznay model was not followed upon
partly because it was not affordable at the time (1969).
Tbe Secundo\' model is currently entered in the Collaborative Testing of Turbulence Models (CTTM, [13])
and Prof. Bradsbaw was kind enough to pro\'ide the
one-page description that was submitted. Dr. Secundov pro\'ided a few details in a personal communication
as well as a list of publications ranging from 19i! to
1986, but none in English. This model is presented as
an evolution of the Nee-Ko\'&Sznay model but is rich in
ilear-wall and compressibility corrections. In particular
we "reinvented" their near-wall destruction term. It is
expected for simple empirical models, developed under
roughly the same constraints (invariance, and so on).
to exhibit strong similarities. However, the leeway is
large enough to produce models with widely dift'erent
perCormance.
"oe
c.
are basing this criticism on experimental and
direct-simulation results [15].
.vu::;u:.;
(,v,r"J
We break our convention for 0'. which belongs to the 1I\'eak solutions such as (3), and raises the possibility of
C6 series, because of the traditional notation of Prandtl non-unique solutions. Indeed if tbe initial condition is
numbers.
Izl, we have a weak solution in which II, behaves lih
1
0, and a smooth solution with
The diffusion term of (2) conserves the integral of the IzI /(2+< ,,) near z
c
/I,
>
0
at
% = O.
The
difference is confined to a
.,. Recall the lack of a destruction term.
quantity
This lack wu responsible for a mild inconsistency in boundary layer near % = O. In a numerical setting with
iIotropic turbulence. It could also inva1idate the model straightforward second-order centered differencing. the
in the c.... or Ihear flowl in which v, decreues (nep- weak solution will be obtained if the diffusion term is
tive Dv,/Dt) luch at an axilymmetric wake. However, written /I, V2 /1,+( 1+cu)(V 11,)2, but the smooth solution
the difruaion term can eaaily brine down the centerline will be obtained if it is written as in (2). Other forms
value or v" and the true constraint is that under (2) the tbat give the smooth solution are V2(vl/2)+Ct2(V/l1)2
intesra1 or v:+ cu cannot decrease. With tbe cl.-ical and (l+eu)V(/I, VV,)-Ct2/1r V2/11' The later addition of
exponents of the aelf-.imilar axisymmetric wake (length a term proportional to the molecular viscosity formally
scale oc 11/3, velocity oc t- 2/ 3 ), we find that the intecral resolves this non-uniqueness and leads to the smooth
increases provided tbat cn ~ 1. Even iCthe calibration solution with /I, > O. Howe\'er, particularly at high
does not include the axisymmetric wake, it is preferable Reynolds numbers, it is desirable to use a favorable form
of the diffusion term.
to satisfy this constraint.
v:+
(0.
[Z
tD,
O,"t---t--~t---hf4-f:,.o'~-!-----;
11:+'".
1.70
1.115
1.10 .~
1.55
I ~
UO
'----.,
10Cll1
1.30
US
0.1
~I
---0.7
11'
0.8
500
I
50
0.'
I~=~ I
0.1 +---;.+--+---t--...,~-l..-i!---ll
0.04
0.06
/UhCw2
"'"
:-------
II
\.Cb2
"'"I
U5
0.3+----+--- -f--il'i:*-+---+---+----!
0.02
i'..
1.50
--- --
o.5T-~-""----r---..,..--.....,.-..,-"'_-....,
1.0
Fisure 1: Calibrated model constants. - locus of solutiOlll; +, x, point selected for the calculations.
~ar-wall Region. high Reynolds Number
In a boundary layer the blocking effect of a wall is
felt at a distance through the pressure term, which acts
as the main dest ruction term for tbe Reynolds shear
stress. This susgests a destruction term in the transport
equation for the eddy viscosity. Dimensional analysis
leads to a combination -CVJ I(IIr/d)2 , witb d the distance
to the wall. The subscript It' stands for 'walr . This
1.0
~I! V
0.1
1'--j'LOQ
1\ I layer
[II
I",
I
i
0.6
/'
F_
....
8hlar
0.2
,
j
HIe,..."
FPG
I/V
0.0
0.0
/ 'V
I",
,.rt
0uIIr
olBL
I,
0.
Figure 3:
see (4-6).
0.8
1.2
[!.i)
Figure 5 shows the velocity profile in wall coordinates, illustrating the log layer and the smooth depar[
ture in the wake. Again, the shape of the outer region
appears good, showing that the destruction term and
This function is shown in Fig. 3. The results are the !'" function are fair approximations, at least in thi5
most sensitive to the slope of Iv. at r
1, which is flow. The arrival at the freestream velocity is a little too
controlled by C",2. The step from 9 to /", is merely a abrupt, as it 1\'as in the mixing layer, in Fig. 2. This belimiter that prevents large values of I"" which could up- havior cannot be corrected except by making the model
set the code and give an undeserved importance to the very diffusive (low u) .
I.,(r)
=, 9 ++c",3
1
/I ] 1/6
3
C'6
,
= r + C.,2 (r
r). (6)
'rr~ff.
1\
I
,
!
I
'\""- .---
r---- ~. r"
,I
'\
I>"<-
-1
7
fo.t
-4
I
!
I
!
I
i
I
,
I
I
I
o
yW
y/S"
Jj
V"'I ,
"
"
"
,I
,
i
'I
I;
Ii
I:
U+
Ii
I,
i
i
i
i
I Ii I
i
I
I
Ii
'I
I:
I!
I:
II
Ii
,I I
I
I
10000
then roughly follows a ramp up to the edge of the turbulent region. Its outer part is representative of the
outer part of either one of the free shear flows, including the vanishing contribution of the destruction
term. The production is equal to the shear stress.
In the outer part the dift'usion is primarily responsi.
We follow Baldwin .\: Barth [8] in choosing a transported quantity which behaves linearly near the wall.
This is beneficial for numerical solutions: ;; is actually
easier to resolve than U itself, in contrast with (, for
instance. Therefore. the model will not require a finer
grid than an algebraic model would. To arrive at this
cODIider the c1aaical log layer and de\'ise near-wall made of the momentum equation and (9), as it has been
"damping functions~ that are compatible with known obtained starting with a wide range of initial conditions.
results. These functions are distinct from the I. near- This includes the results of Figs. 4 and 5; in particular
Fig. 5 displays the viscous and buffer layers.
wall inv"lCid destruction term.
we
The final set of terms provides control over the laminar resions of the sbear layers, a control which has
two aspects: keeping the 8011' laminar where desired .
and obtaining transition where desired. Navier-Stokes
codeiJ with &Igebraic models usually have crude "off-on"
devices or short ramps hued on the grid index along the
wall. These do not help the convergence of the codes. In
addition we require a device that is useable on unstructured grids. The subscript t will stand for "trip". We
i.Jse this word to mean that the transition point is im
p06ed by an actual trip, or natural but obtained from a
The production term also needs attention. III it S is separate method. 0"
scco."t do,d' tAt t.rhlt"cf
replaced with S, given by
mollel 6( tM/std to prellict tAt transition location. Tbis
is true of all the models we kno,,. Some models. in
X
(8) c1uding Ir-(, predict relaminarization. In situations tbat
/~2 1 - 1 + xl,,)
should induce relaminarization, this one tends to drop
The function /,,2 is constructed, just like /01> 110 that the eddy viscOCIity to low levels, but without "snapping"
S would maintain its log-layer behavior (S
/(KY to O. We may be able to improve on tbis using the f'2
all the way to the wall. S is singular at the wall, but function below.
ii is 0 there, 10 that the production is well-behaved '
We delCribed how transition was expected only in
Note that there is a range of X in which S is _ than thin shear layers. The linearized version of (9) for small
S and may become negative. This should DOt upset v only contains the production term: DvlDt cuSii;
the Dumerial salven. Other quantities involved in the therefore,
0 is an unstable IOlution of (9) (going
"inviKid" model are redefined in terms of if iDst.ead of in the direction of DIDt). In a boundary-layer code
the uro solution is easily maintained, but in a Na\'ier/I" for instance r == ii/(SK 2tP).
We fiwly add a viscous diffusion term, cOMistent Stokes code exactly-aero values are rarely preserved, so
with a Dirichlet boundary condition at the wall, if O. that the model is "primed" by numerical errors upThis term too is based on an analogy, rather tbaa a rig- stream of the trip. It then transitions at a rate that
oroua equation. In addition ii behaves linearly, 110 that depends on numerical details and has little to do wit h
its Laplacian will be small in an established dution. the boundary layer's true propensity to transition, as
Accordin&ly, we insert the molecular viscOCIity ia a con- controlled by pressure gradient, suction, and so on. We
venient place and pay little attention to a factor of tT . verified this behavior, and it is not acceptable.
"0
=...
v=
Dv
-_ 1 [
Dt =c61 S/I+; V./I+V)VV)+C.2(VV)
-c",t!tIJ
2]
[~r
(9)
C'4
2. In any case c,3 must be larger than 1. As
for Ct. it could be decreased several-fold , if a code still m=CbJ
D"V
[1 - f 12 J S- 11+;;
- 1 [....
- \ . ((1I+1I)\"V)+Cb
2 (\"vf"]
yielded premature transition. The cross-over point of
1- / 12, i.e., the bound of the basin of attraction), is at
X=
og CI3 C,.. However, 2 was small enough in our
- [cu,du - 112]
+ III AU 2. (11)
code. Values much smaller than 1 would start affecting
the results in the turbulent region.
with
~;
111
=CII gl
[~r
[d 2 +9~d;1) ,
(12)
and 9, == min(O.I, taU /w,AZ) ",here Il.z is the grid spacing along the wall at the trip. This equation specifies
the two I, terms, and the trip term is the last in (11).
The Gaussian in 111 confines the domain of influence of
the trip terms as needed ; it is roughly a semi-ellipse.
The magnitude is adjusted so that the integrated contribution for a particle crossing the domain of influencl"
is on the order or U.6, 6 the boundary-layer thickness,
as is ensured by typical algebraic models [2]. The odd
factor 91 is passive in a situation with an extremely fine
grid, but is quite active il.nd necessary in practice. This
is because the domain ofinfluence Dfthe trip scales with
the boundary layer thickness, which is very small in the
laminar part. As a result, that domain easily falls between two streamwise grid points, so that the trip is not
felt at all. The 9, factor guarantees that the trip term
will be nonzero over a few streamwise sta(ions.
Note that the grov.th of ;; to nonlinear levels under the effect of the production term occurs in a few
boundary-layer thicknesses (C6I being roughly 0.13) .
This is consistent with the idea that the trip term mimics the secondary instabilities invoked by recent transi
tion theories, which have growth rates on the order of
1/6. However since the streamwise grid is often much
coarser than 6, transition will still appear very steep to
that grid . The contribution of the trip source term is
Dimensional analysis points to AU 2 as a proper rapidly overwhelmed by the exponential amplification
scale for the source term. and we arrive at due to production . Thus. we have a formal ad\"antag'
fmstream C9nditjons
',,1
(19)
The exact solution of the turbulence transport equation cannot become negative. It can be shown that if
ii 0 at some location and the surrounding values are
non-negative, then 8v/8t ~ O. An underlying goal of
tbe solver is to reproduce this analytic behavior-a nonnegative turbulence field-throughout the solution process (i.e. ii ~ 0 at all grid points and at all time steps).
':: =
M(V)ii=
c., (VV)2] ,
(14)
1121 S V,
(15)
D(V)v
= [coud.. -
eu
v_]2
[
,,2/t2] d '
(16)
=IIIAU 2
(17)
10
(20)
(218)
(21 b)
(2Ic)
[M -
KIJ v ~ 0
[p - JS] v ~ 0
[D -I>] v:S 0
for all v
(22a)
(22b)
1 + Cb2 1 [(
;";\
ii.+ 1 - V.
[M(V)V]\2) = - - - - v+ vIHl/2-"-'""--"':'
(22c)
forallv~O
CT
Ax.
A.r.+I/2
_ (v
Vi_I] ,
(27)
i-I/2
cn
1
[M(V)VW') :: - -(v
+ ii)i--"
Ax.
CT
iii - Vi-I]
[ AXi+I/2 - AXi-l/2 .
iii+! - Vi
(28)
Description of the discretization and construction of The diffusion coefficients (V+V)i+I/2 and (v+ ii).-1/2 at
implicit matrices will be given in detail for I-D, where cell faces are taken as averages of adjacent cell-centered
the notation is simpler. Extension to 2-D and 3-D is values,
straightforward because there are no cross derivatives
in the one-equation model.
(v + ii)i:H/2 = ~ [(v + iI). + (v + ii).:i:Il,
(29)
The model is discretized using a cell-centered finiteand AXi+I/2 and Ari_l/2 are distances between cell
difference scheme.
centers.
.
J acob'lans """,2)
Th e approXimate
M
an d """'M 3) are 0 I)Advection Operators
tained by freezing the diffusion coefficients.
The advection terms are discretized using f:rst-order
...,{2)
1 + C62
_
1
accurate upwinding.
Mi,i_1
+ ---(V+V)i_l/2 "A
= 1_ (Ui -
lUi/).
(24)
M,
','
CT
1 + Cb"
_
1
- - - - -(v + 11).+1/2--":-'-CT
""{2)
M i.i+ 1
1 + Cb2
+ ---(v
CT
""",3)
t.t-
M!~Ll
= -- (-Un/AXi,
= -. (+ut - Un/AXi,
fV(~i~1 = - (+Un/AXi.
Note that _}i;fl) is M-type and (M - IVI)(1lv
satisfying the positivity constraints.
cb2
(f
~Zi~J't-1/2
1
= + ""::'(v+ ii)i--':""-AXi~.ri_l/
...,{.3,l
C6"
Mi,.
(25)
(30)
~.r'~X'+1/2
1
+ ii)'+I/2"',,-.,-... .r ...... .rHln
-(V+ii)i
M; I:: -
K(~/
~x.~r'_l/2
CT
:-:{2)
(23)
where the advection velocities U+ and U- are defined.
ut = ~(Ui + IUil).
=
1
= - -1 +-Cb2
( v + ii)i-I/2 ~ ~
Xi x._I/~
CT
Cb2
1-
+ -(v+ ii)i
(31)
,
AXiA.r.+I/2
1
-(v + ii)i
.
CT
Ax.Axi+l/2
CT
""{3)
= 0,
Mi,i+l
=-
C62
[M -
M](2) v
=0,
i71(3)
[M-M J
:-:{2)
v::O,
(32)
:-:{3)
',i+1
+ Kr 3 )
',HI - CT~XiA.r'+1/2
[(1 +
Cb2)(V
+ ii),+1/2 -
Cb2( II
+ ii).]
(3:3)
M'2)
i"+l
K(3 )
i.i+l
[0 +
:s
= 20'aZ,aZi + 1/ 2
Ct2)(V
+ ii)i+l + (1-
Cb2)(V
:s
=
=
=
=
In a typical fiow situation, both production and destruction increase in magnitude as v is increased (i.e ..
p, D and their derivatives are all positive) . In this case
our third strategy chooses,
Source Operators
12
With the _ree term6 taken ""ether. only two posit h'ity coutramts are required.
1) -
[1>-D]-lP'-P]
P' ~ 0,
(41a)
= [1>-P}-[D-P] ~ O.(4Ib)
=~tRY". (45)
where R is the residual or risht-band-side operator in
(18). We have aI.o tried the approximate factorization
developed by Shih &.: Ch}'u [23J for finite-rate chemistry.
[N - ~tUd N-I [N -
.1tU,,] = ..1tR," ,
(46a)
(46b)
ti,,"',
(4i)
where
y .. +l
pves.
= [1+~t(M(-Ud
(4E-)
(44)
t"
13
[1- ..1ttr(]
=
[1- ..1tU.,J ..1y(tl = ~v ( i
The second .tep equation bas been used to rewrite the o.ooa
eplittiq error 011 the ri&ht-band side in the first ,tep;
~
~v- is a coovenieat definition.
0.0021
0.11024
D~
00022
I
I
.
0.0031
0.001'
o~
0.0014
----l
0.0001
f\
0.0010
!
!
I
i
0 I'
0.000& 1
0;
0.022
0.020
I
Mode/_
,~7
O.ClOl
0.005
0.004
0.002
/
V
E"'~
Boundary-Lug Calculatjons
'\
0.0012
Results
C}
= 0.0015
The model shows promisin, conver,ence to steady8ta&e wbeo coupled to the solver for the velocity field.
Cooversence is typically as ,000 or hetter than tbat
I
I
....r
~
...- V
/'
,
I
!
j..-I
Only iDcompreaible boundary layers have been conIidered. With zero prel5ure padient, the model obeys
the accepted ReyuoId..number Kalin" 10 tbe results
IhowD at R.
104 enlUre apeement with the curmat tIaeories. The model &ives satisfactory results in
attadaed boundary layers with preeaure padients, typical of tille Stanford lQ68 cues. We only present reault.
for the .mk low aDd the Samuel-Joubert low [24] as
tile other CUllS with moderate &r&dients Ihow the same
tread. Darmofal's code was used asain.
14
---,--,---,.---,----
0.12'
1-~!:---,:
1.65
1.10
0.1'
0
--!-----;--+--+---.1--+---,..--
.!
O.'Or-+--+--1r---t--+--+----i-~~
i--+--+--+-+----if---+--.;...!~~
i JIt::
0.08 +---f---+--If--+---;---+--i--J-
1.515
J...
0.08
II
1.80
Exp' -4::
OJ
0
1.45
c
0
1.40
0
.f1
O.07+--+_+_-+--t-~--I--+JSoL-1
0.06+-_+_+_-+--t-~-_1-"-"4---i
~~I
O.05+-_+_+_-+_-t__
"/
"i~odel !
_+--+_+--+.-.<-..j,4'7-+_~' _:
1
0.03+-
"/
0.02+-_+_+_-+--;;"..q-r~--1--+----.;
0.01
,./
i--1.35
Exp.....
~_--:!!~~~L_
!
O.04+--+--+---1:..---t--+'<----o,.q:.=~--
V :
II
,,"/
t---t--r--::lI!F;.,-&--+--r--+--f----'
~v
0.00 ~_....-_~:....JL-.J--..L-__1_--.J----"
0.2
0..
0.6
0.11
' .0
Fig. 8. The model produces slightly higher skin friction, but lower thicknesses. The shape factors (Fig. 9) 0.'2
are in disa&reement even before the pressure gradient is 0."
I
applied, and are the root of the disagreement in thickI
nesses. With an adverse gradient, the skin-Criction term 0.'0
loses its authority in the momentum equation. The ex- 0.08
perimental values for z between 1 and 2m, H ~ 1.39,
I
are surprisingly high considering the weak pressure gra- 0.011
dient and the Reynolds number, R, ~ 6500. Interest- 0.07 !
ingly, the calculated shape factor is catching up with Y
0.06
the experimental one Cor z > 3m.
"
.
i
~ "
.............
!
..............
""k
-.........
"-......
jlAodeI !
,
Figure 10 shows the velocity at z = 3.4m. The posi- 0.05
tion oC the boundary-layer edge is good, but the com- 0.04
~ ,
Exp. ~~i
puted profile is Cuner than the experimental profile. The
shear-stress profiles in Fig. 11 show good agreement for 0.03
7)
the outer values, but the near-wan agreement may be 0.02
,..
poor enough to partly explain the differences in the ve~ I
locity profiles. Stress disagreements can of course be 0.01
P
t--,
l-i
compounded by the convection and pressure terms. Cu- 0.00
0.0016
0.0000
0.0004
0.0008
0.0020
0.00'2
riously Dr. F. Menter, who was kind enough to test the
r
model in his Navier-Stokes code [25], obtained similar
a&reement Cor the stress profiles, but better agreement Figure 11: Shear-stress profile at :r = 3.39m in SamuelJoubert flow . T normalized with edge velocity. y in
for the velocity.
meters.
The Samuel-Joubert results suggest a mild but genuine tendency to underpredict the shape factor and
thickneMeI in adverse pressure sradients. This may this type of 80w. Neverthl!less, the behavior of the
make the model a little more resilient to separation than new mod~l is not disappointing and vindicates the noit would ideally be. The tendency is not as strong as tion that a model calibrated Cor mixing layers, wakes.
with the Cebeci-Smith, Baldwin-Lomax, and IN mod- and zero-gradient boundary layers has a gooa chance in
els, but our comparison with experiment is not quite adverse-gradent boundary layers, which are an interas Sood as that obtained by Menter \II-ith the Johnson- mediate sit'Jation. In addition, an improvement of the
King and Ir-w models [25J. Both models have been dt'<;truction !erm 1"Iay yet be found sufficient to obtain
finely tuned over years. with an emphasis on precisely resuits 011 <i ;"! .... ith the best models.
~
15
~avier-Stokes
Calculations
'1.4
'1.2
..
I
~I-
".0
--
~
L..-" ~
xperl~t
ptesent~ :
- - - . a.tdwi".lomu
- . - . .IcIhNon-KJng
-0.1
\..
\;\
V'
~
l.r' ~
i
I
I !
!
,I
'
'"
~ .:J,
I
AU calculations for Case 6 were performed at tbe same 0.4
0.4
0.0
0.2
0.6
0.8
1.0
conditions: M
0.725. He
6.5 X 106 a prescribed
z/c
lift coefficient of Cl
0.743. and transition trips at
Figure
12:
Pressure
distribution
for Case 6, RAE 2822
3% chord. Results on the 768 x 160 grid are shown in
Figs. 12 and 13. AU models converged solidly on both airfoil.
grids.
0.007,..,..-....---.--,..-....--....----,r-------_
Figure 13 shows the upper-surface skin-friction coefficient for the same case. Between the three models there
is a difference or up to 10% upstream of the shock. and
similar but reversed differences downstream of it. The
new model d0e8 not predict wall-shear reversal. The
other two predict reversal at the foot of the shock, but
only on the finest grid (768 x 160). Reversal was never
predieted at the VTAW, where the finest grid used was
369 x 65. It appears that with current codes the simple
qUe8tion of whether reversal occurs is not firmly answered even at grid resolutions that are considered very
fine. At the shock. the new model produces a larger
step up for the boundary-layer thicknesses than with
Baldwin-Lomax, resulting in a mor-e for~'ard position.
Intere8tingly, the size or the displacement effect is not
correlated with the occurrence of reversal at the wall.
-O.OO,+--~-4--L-+---L---<~~-+-~--
16
0.0
0.2
0.4
0.6
0.8
' .0
z/c
Figure 13: Skin-friction coefficient on upper surface for
Case 6, RAE 2822 airfoil. C, based on (;at:.
-.
-'.4
-1.2
f'
+- j
--
-,.0 ~.
-, .... :;'
-aa
'-
...
~~
.--1
~'
~\
.
~;.-
--~_'
----8eldwi~Lomu
-'-"~King
-aa
-C.,
~\.\1'
I\~,~
"
V ~
V"
"\
,. .
"
"
0.007
O.CIIII
~I
- - - '
8eIdwi~.
'.
,~
1I
---- .-
\\ .
\\
--prwent~
,.
.-
Ii
:I
ylC
UminIU.dg.
0.6
I
0.5
....
..,..V
0.4
)'
0.3
0.2
0.0
.. '""
..- - - ...~.----.,..,
.o:-,-----:,""
02::---x-IC--:-'' -.,:-----:--..,..
-D.1
. ,
~
:
II
0.1
...OIL-_----------------
1/
0Y
1.00
!
I
1.02
1.0<1
1.06
1.08
1.10
zlc
Figure 17: Minimum velocity in wake for Case 1, blunt
RAE 2622 airfoil. 0 experiment, - - - 768 x 160 grid,
- - - 384 x 80 grid.
18
Xlc
103
...
O'Olf===:-~~==~~~g::~~~~~~~~~~~~
::
:
....
eddy
blends
boundary-layer
behavior
(as
in fig.visc06ity
4) into its
wakeitsbehavior
(a bell-shaped
distribution). There is little basis to judge its level in the
recirculation region.
Finall)' Fig. 20 shows the pressure distribution. The
agreement is good, especially considering the 0.415' difference in angle ofattack, except that the calculated Cp
is shifted down. This was also observed by Drela for his
numerical method (31). We also ran Case 4 to measure
the Reynold&-number effect. In the experiment, Case 4
had the same angle of attack as Case 1, but a Reynolds
number of only 1.9 x 106 , resulting in a 1066 of 0.036 in
C/. We ran Case 4 with imposed C/ 0.415, and the
angle of attack settled at 1.940 The lift-curve slope is
about 0.155 per degree. This implies that the C/ I06S
due to the lower Reynolds number is about 0.022. This
is only about 60% of the lift I06S observed in the experiment .
L--------:=====
01
...
..... .~-~--:~-~-==---=---=-~--:-:
.
1.00
1.0'
1.at
XIC
1.04
l.OJ
-O.s
,
.....
,I
ElOP
'-,.- o.t.!.:
-0.11
!!!: ~
-Cp
;Y
-0.2
lY
'j
0.4
0.0
0.2
0.11
0.4
ric
'''\
1I
0.0
0.2
"f\
-o.~
I! !
I
,"
~
~xl,.J
0.'
1.0
19
Outlook
The development of the model will nevertheless continue. It will praerve the basic favorable features of
the model: ainpe trauport equation, local formulation,
moderate resolution requirements, good numericalltability, iD8eDlitivity to the freestream value, ready control of trauition. The invariance principles wiII allO be
upheld. Additional tests will be made, notably in three
dimelllioal. No additional difficulties are anticipated.
Tests in more Itronpy stimulated Oows such a massive
IeparatioD, wakes in pressure gradients, or free vortices,
are likely to reveal weak_ in the model. They will
aIIo aue. the current Navier-Stokea codes, in two respects. The first .. the detai1accuracy, for instance the
cooaervatioo of momentum in boundary layers, which
is far from perfect in our experience. This problem haa
been ot.tructed by the difficulty in computing the thicknellel6 and IJ in Navier-Stokes codes, but it must now
be addreased, if only on a flat plate at low Mach number.
Even when unsteady solutions of the modeled equations are obtained, their meaning will need scrutiny.
Many of the properties of Reynolds-averaging hold approximately if there is a separation of scales (i.e., a spectral gap) hetween the resolved motions and those that
are left to the turbulence model. We have little evidence
in that domain. An example of legitimate decomposition would be a mixing layer that wanders on a time
scale much longer that its internal time scale which, after modeling, is 6/AU (with 6 the thickness, and A(,
the velocity difference). Note that what we envision
here is different from Large-Eddy Simulation. As the
The other aspect is the question of steady solutions grid is refined the model does not change, the way it
in Oows with medium- or large-scale separation. An does in LES (through a narrowing of the filter, so that
classic example is the flow past a circular cylinder; a the limit is direct simulation). The difficulty of LES is
pressing industrial example is the Oow past a staDed the filtering without spectral gap. Instead, v..e converge
airfoil, near its maximum lift coefficient C"mazo. Except to a smooth solution of the modeled equations .
at very low Reynolds number the Oow is unsteady and
Some of the near-future directions have been hinted
three-dimensional. On the other hand, provided the ge- at above. There is the choice of S, between the vorticometry is time-independent, it is legitimate to define its ity, the strain rate, or another scalar norm of the detime-average and to hope for a code that would com- formation tensor. There is the use of an approximation
pute it as a ateady solution. It is .Iso legitimate to re- of the turbulent kinetic energy Ii: to give the Reynoldsquest the low-frequency component (i.e., with Strouhal stress tensor a plausible trace (or set of eigen\lllues).
number of order 1) of the solution, particularly the Oue- There is a modification of the /", function in the region
tuating loads for Itructural purposes. The ideal turbu- r > 1, that would alter the results only on adverse preslence would include a Iwitch between these two options. lure gradients. The third modification could be assessed
We do not know, in general, whether the exact solu- in the Samuel-Joubert or airfoil flows, but obvious test
tion of the Navier-Stokes/turbulence-model system in cases for the first two are not at hand.
its pretent form is steady. In the design of some models,
The current model has no compressibilit~ terms. Emdecisions have been made solely on the basis of a preferpirical terms baaed on the turbulent Mach number, such
ence for the candidate that yielded steady solutions. We
as the one in the Secundov model, or on the quantity
are far from having the capability of routinely performVp.V;; are available. The former may be calibrated
ing time-accurate calculations to explore the issue. All
in supersonic mixing layers. The latter may assist the
the routine calculations aim at Iteady solutions. Failure
model in shock/boundary-layer interactions if a consisto converge, and the generation of a limit cycle,
be
tent trend is found that shows an erroneous shock loa lign that the exact solution is unsteady. It may allo
cation, and we can extricate that trend from numer ibe just a numerical problem.
cal concerns (e.g., artificial dissipation) and from the
Enviaioo a ateady solution, say past a cylinder. An endIeaa corrections of tranlOnic airfoil testing. In that
aJebraic modelluch as, for instance, Baldwin-Lomax range of density variations it may be enough to write
would be UIed out of its range, but this does not stop ea- the transport equation in terms of p instead of ii. takger UIetI. In the recirculating re&ion it would yield eddy ing advantage of the flexibility in placing p inside or
viscosities on the order of U00 R, with U00 the freestream outside various derivatives. This is consistent ",ith the
velocity and R the cylinder radius. We presume here spirit in which the model was devised. To tJ .:s date.
that the Baldwin-Lomax fmn condition would occur efforts to devise a curvature term with the required inin the wake, and not near the wall. A balance be- variance pr(\perties and no d.dependence have failed.
tween production, on one side, and diffusion and de- Note that curvature effects have been obsef\ed in free
struction, on the other, suggests that the new model shear flows. Three-dimensional effects in boundary layalso could produce a IOlution with an eddy vilCOIiity ers (Le., pressure-gradient vector at an angle to the ve
m.,
20
locity vector) are also delicate to introduce even empirThe trip function III is as follows: d, is the distance
ically without violating the in~-ariance principles. No from the field point to the trip. which is on a wall. ,,:, is
plana have been made to depart (rom a acalar eddy vis- the wall vorticity at the trip, and ~U is the difference
c05ity.
between the velocity at the field point and that at the
trip. Then" == min(O.I.~U/IoII~Z) where ~.r is th.,
grid spaein, along the wall at the trip. and
Acknowled,ements
:J%
g~d;]
=
=
=
=
References
=ii 1,,1.
1~1
=X
X3
3
3'
+C.,1
ii
X:: -.
If
(AI)
- [C.,tltI/ - :: 112]
[~r + In ~U2
I.,
(A3)
is
1+ I ] 1/'
I.,=.?
, ,=,,+cw,(,,6_,,). ,,:.....!--.
[ +)3
tl/3
S,,2d2
(A4)
[3] Jobnson, D. A., &: King. L. S., "A mathematically simple turbulence closure model for attached and separated turbulent boundary layer5~,
AIAA JoUrnal. Vol. 23, No. 11. 1985, pp. 16841692.
[4] Mavriplis, D. J . "Algebraic Turbulence Modeling
for Unstructured and Adaptive Meshes", AIAA-901653.
(A2)
ii
S == S + 242 1"2,
[1] Baldwin, B. S,' &: Lomax, H., "Thin layer approximation and algebraic model for separated turbulent
flows". AIAA-7S-257.
(A5)
21
[10] Nee, V. W., &: Kovuznay, L. S. G., "Simple phe- [24] Samuel, A. E., &: Joubert, P. N. "A boundar~'layer
developing in an increasingly ad\'erse pressure granomenological theory of turbulent ahear flows",
Physics of Fluids, Vol, 12, No.3, 1969, pp. 473dient", J. Fluid Mech., Vol. 66, 3, 19i4, pp. 4$1484.
505.
[11] Secundov, Smirnova, Koclov, &: Gulyaev, "One- [25] Menter, F. R., "Performance of popular turbulence
equation eddy vileosity model (modified L, S. J.
Kovumay model)". Short aummary of the equatioaa. Personal communication, 1990.
C. P., "Aerofoil RAE 2822 - Pressure distributions, and boundary layer and wake measurements", AGARD-AR-138, 1979.
Swanson, R. C., &: Turkel. E., "Artificial Dissipation and Central Difference Schemes for the Euler and Navier-Stokes Equations", AIAA-Si-llOi,
1987.
[28] Wigton, L. B., "High Quality Grid Generation Using Laplacian Sweeps", Fourth International Symposium on Computational Fluid Dynamics, U .C.
Da\'is.Sep. 1991, pp. 1222-1228.
[29] Cook, P. H., &: McDonald, M. A., "Wind tunnel measurements in the boundary layer and wake
of 6.n airfoil with a blunt base at high subsonic
speeds", RAE Tech. Rept. 84002. 1984.
[16] Baldwin, B. S., &: Barth, T. J., "A one-equation [30J Henne, P. A., &: Gregg, R. D., "A new airfoil design
turbulence transport model for high Reynolds numconcept", AIAA-89-2201-CP
ber waIl-bounded flows", NASA TM 102847, 1990.
[31] Orela, M., "Integral boundary layer formulation for
[171 Townsend, A. A. The structure of turbulent shear
blunt trailing edges", AIAA-89-2200.
flow. Cambridge University Press, New York, 1976.
[18] Schlichting, H. Boundary-layer theory. McGrawHill, New York, 1979.
[19] Hunt, J. C. R. "Turbulence structure in thermal convection and shear-free boundary layers".
J. Fluid Mech., Vol. 138, 1984, pp. 161-184.
[23] Shih, T. I.-P., &: Chyu, W. J., "Approximate Factorisation with Source Terms" , AIAA Journal, Vol.
29, No. 10, October 1991, pp. 1759-1760.
22
..j