You are on page 1of 10

A CALCULATION METHOD FOR THE THREE-DIMENSIONAL

BOUNDARY-LAYER EQUATIONS IN INTEGRAL FORM


Bilal Mughal and Mark Drelayz
Computational Aerospace Sciences Laboratory
Department of Aeronautics and Astronautics
Massachusetts Institute of Technology, Cambridge, MA 02139

Abstract x, y, z Cartesian coordinates


generic angle
A new numerical method is developed to solve the three- streamline angle w.r.t. to external ow
dimensional boundary-layer equations, in integral form, on  characteristic line angle w.r.t. external ow
non-orthogonal grids. The nite-volume scheme employed () displacement thickness
eliminates the need to compute metric-gradient terms found () density thickness
in curvilinear-coordinate nite-di erence methods. The in-  auxiliary density thickness
tegral method is based on two equations for momentum and ()() orientation-dependent momentum thickness
one for kinetic energy with empirical equilibrium compress- () orientation-dependent kinetic-energy thickness
ible turbulent- ow closure relations selectively extracted  laminar viscosity, residual distribution factor
from the literature. Johnston's model is used for the cross-  density
ow. The non-linear discrete equations are solved simulta-  shear stress
neously using the Newton-Raphson method along a row of ()w value at wall
cells and the solution is marched successively downstream. ()e value at boundary-layer edge
Along each row, cell residuals are distributed to nodes in
a manner consistent with the local direction of character-
istic lines. Results are computed for a well-known in nite 1 Introduction
swept wing case to evaluate empirical-closure accuracy, and
for a nite swept wing case to demonstrate the full three- There have been astonishing advances, over the last sev-
dimensional capability. eral years, in the development of Navier-Stokes solvers ca-
pable of generating solutions over realistic aircraft con g-
urations. The generation of each solution, however, re-
Nomenclature quires considerable resources thus leaving these solvers out-
side the realm of routine design activities, especially those
A cross ow parameter involving optimization. Coupled Euler/boundary-layer or
cD()
coecient of dissipation Potential/boundary-layer methods, on the other hand, are
cf()
skin friction far more economical; they have proved very useful in two
D dissipation or shear-work integral dimensions and o er the same potential in three dimen-
E()() components of () sions. Furthermore, diculties in comprehensive modeling
F ux vector of turbulent ows puts Navier-Stokes solvers in the same
G ux vector league as these coupled methods in terms of solution ac-
H() shape factor curacies. The present study is undertaken to model the
j grid counter boundary-layer part of a coupled solver.
M() Mach number An integral form of the equations is chosen over the usual
q speed di erential form primarily because interest is largely con-
R, R~ vector of residuals ned to integral quantities. There is also an order of mag-
Re() Reynolds number nitude decrease in computational e ort. Furthermore, inte-
S Source term vector gral schemes tend to be more robust and are better suited
s, n streamline coordinates for viscous/inviscid calculations. On the other hand, an ob-
U vector of primary variables vious disadvantage is the increased reliance on empiricism.
u, v , w velocity components In reality, however, integral methods have historically pro-
 Graduate Research Assistant, Student Member AIAA duced results of accuracy comparable to those generated by
y Associate Professor, Member AIAA di erential schemes [13].
z Copyright c 1993 by the American Institute of Aeronautics
and Astronautics, Inc. All rights reserved. The prediction of three-dimensional turbulent boundary-

1
layer ows has only recently begun to garner attention,
relative to two-dimensional ows, sucient to match their y
ubiquitous presence in nature. One prominent reason is
the sheer geometrical complexity of general extensions into
a third dimension. The traditional nite-di erence curvi-
linear formulation of the equations on non-orthogonal grids
requires the evaluation of cumbersome metric and metric-
gradient terms. This new method uses a nite-volume dis- Crosswise profile
cretization approach to pose the computational problem
in a simple locally-Cartesian form, without any additional
metric-gradient terms. Streamwise profile
A method of residual distribution is developed to account
for the local ow-of-information directions in the system β
of hyperbolic equations. The resulting non-linear discrete w
equations are solved using the Newton-Raphson method. n
s

2 Governing Equations Figure 1: Velocity distribution in a typical 3D boundary


layer
The steady compressible di erential boundary-layer
equations, in a Cartesian system, are integrated normal to
the wall to obtain a system of three equations involving For the special case where the coordinates (x; z ) are
integral quantities (see, for example, [7]). In conservation aligned with the coordinates (s; n), formed by the exter-
form, they are: nal ow streamline and cross ow directions respectively,
numeric subscripts (1; 2) are used instead of (x; z ).
@F + @G + S = 0 ; (1) The components of the vector U are the three chosen
@x @z primary unknowns,
where the ux vectors are 0 1
0 1 BB 11 CC
e qe2 xx U = B w CA ;
B C @ 
F =B
B 2
@ eqe zx
C
C
A 1
eqe3 x where w is the limiting streamline angle, as shown in Fig-
ure 1.
and 0 1 The three rst-order integral equations are hyperbolic.
B eqe2 xz C The system can be diagonalized into decoupled equations
G=B
B 2
@ eqe3zz
C
C
A; involving the characteristics. Eigenvalues of the system are
real and distinct and correspond to the the tangent of the
eqe z angle that the characteristics make with the external ow
and the source vector is streamline.
0  
1 For an incom-
BB eqex @ue=@x + eqez@ue =@z xw CC pressible, turbulent system of momentum-integral and en-
S = B eqe x @we =@x + eqe z @we=@z zw CA : trainment relations, Cousteix and Houdeville [3] show that
@  2  2
one of the characteristic lines lies close to the limiting wall
e qe x @qe =@x + eqe z @qe =@z 2D streamline. In Figure 2, this characteristic line is labeled
1 . The other two characteristic lines always lie between
The dissipation or shear-work integral D can be written the external- ow streamline and the limiting wall stream-
as the sum of two components, one due to the wall-normal line. The calculation of reasonable characteristic directions,
gradient of velocity in the x direction and the other due to therefore, constitutes a demonstration of the validity of the
the gradient in the z direction, such that closure employed.
D = Dx + Dz :
The rst two equations involve the momentum de cit and
3 Empirical relationships for closure
are simple 3D extensions of the 2D von Karman equation Empirical relations are necessary to make the problem
while the last is a relation for shear work or kinetic-energy determinate. All non-primary variables are correlated to
dissipation. These equations are valid in laminar and tur- the three primary unknowns, 11 , 1 and tan( w ), using
bulent ow regimes; in turbulent ow, the variables are assumed forms of velocity pro les together with empirical
interpreted to be mean ow quantities.

2
The assumption that the streamwise component of wall
shear is related to the streamwise velocity pro le implies
that the law-of-the-wall applies to the velocity pro le in its
usual two dimensional form. Strictly, this is incorrect. This
γ assumption is likely to be of more consequence, however, in
1 methods employing di erential equations while in integral
γ methods, errors thus introduced are unlikely to be more
2 serious than those originating from other sources [8].
A thickness, E11 , that depends only on the streamwise
β γ velocity in the boundary layer is de ned such that
w 3 s Z
e qe3 E11 = u(u2e u2 ) dy :
0
Figure 2: Characteristic directions in a 3D boundary layer This thickness is the streamwise component of 1 and can
be correlated to H and Re11 through the streamwise veloc-
ity pro le. The correlation employed is one due to Drela [5].
skin-friction and dissipation relations derived from experi- In addition to this relation, another proposed by Whit-
ment. eld et. al. [14] was also considered. It produced results
There isn't a substantial observed improvement in accu- almost indistinguishable from those using Drela's equation.
racy when using eld methods instead of equilibrium ones However, for one calculation (the van den Berg/Elsenaar
in the equations [13]. Therefore, in the interests of simplic- in nite wing) it failed to yield characteristic directions co-
ity, closure of the integral equations is achieved by equi- herent with the computed result in that one or more of these
librium empirical models. These models are obtained from characteristic directions was signi cantly outside the range
the literature. of the limiting and external- ow streamlines. The fail-
The traditional methodology employed in three- ure was presumably due to an assumption in their deriva-
dimensional closure has been to assume two-dimensional- tion that e ectively eliminated the dependence on Reynolds
like behavior of the boundary layer in the streamwise di- number as noted by Drela [5]. Consequently, the sensitiv-
rection because there is a close experimental resemblance ity of the characteristic directions to the Jacobian-matrix
between three-dimensional streamwise pro les and two- elements involving HE11 , especially at low Reynolds num-
dimensional boundary layers [8]. One fortunate outcome of bers, was probably erroneous. Drela's expression does not
this assumption is that empirical correlations of the stream- include this simpli cation.
wise variables in three-dimensional ows can draw a con- The correlation for compressible density thickness em-
siderable amount of information from the better-explored ployed is only a function of the shape parameter and Mach
two-dimensional ows. Clearly, this approach works best number and is given by Whit eld et. al. [14].
when the cross ow is small. The coecient of dissipation due to the surface-normal
gradient of velocity in the streamwise direction was taken
3.1 Streamwise Closure from Whit eld et. al. [14]. It was derived by numerically
evaluating the dissipation integral using a constant laminar
Turbulent boundary layers have a two-layer structure plus turbulent shear stress in the region very near the wall
where the thickness of each layer scales di erently with Re . and an eddy-viscosity model in the inner and outer regions.
Therefore, at least two independent parameters are needed
to adequately describe a velocity pro le. The analytical 3.2 Crosswise Closure
formula that is employed in this calculation method was
developed by Swa ord [11]. It is the sum of two indepen- The cross ow model employed in this calculation method
dent transcendental functions, one expressed in terms of is Johnston's hodographically triangular one [6]. In this
the inner variable y+ and the other expressed in terms of model, within a thin layer adjacent to the surface, the cross-
the outer variable y=. wise velocity is related to the streamwise one by
Following the observation that the velocity distribution w = tan ( w ) :
in the streamwise direction may be represented by pro le u
families developed for two-dimensional ows, it has often Over the remaining portion of the boundary layer,
been assumed that the streamwise component of skin fric-   u 
tion may also be obtained by using skin-friction relations w =A 1
applicable to two-dimensional ows [8]. The presence of q q :
the law-of-the-wall as one component of the two-parameter
family of pro les enables one to correlate u+e in terms of The parameter A is a measure of the cross ow magnitude
two integral parameters, for example H and Re11 , thereby and can be related to w . Smith [10] provides a relation,
yielding an expression for the streamwise component of skin " p #
friction. The relation employed in this calculation method c f cos( w )(1 + 0:18 Me2 )
pc cos( )(1 + 0:18 M 2 ) 1 ;
A= tan( w ) 1

was also developed by Swa ord [11]. 0:10 f1 w e

3
which includes the e ects of compressibility.
While A appears explicitly instead of w in the cross-
ow variable relations, the latter is, nevertheless, chosen as
a primary variable because it is uniquely related A. The global cell
reverse is not true. Y y
The cross ow component of wall shear stress is related to x
the streamwise one through the limiting-streamline angle,
z

Z X
zw = tan( w ) xw :
The dissipation coecient due to the cross ow is ex-
pressed in terms of that due to the streamwise ow and
the parameter A,
  jAj+10:0
cD = Dq23 = 14667 cjAj + 3:0
1020cD1 +4:0
;
2
e e D 1
α
s n
where cD1 = D1 =eqe3 . This correlation was derived by
numerically evaluating the integral 4

Z 
x

j tan( )j1 @u
3

0 @y dy ; y z

using Cebeci-Smith's two-layer eddy-viscosity model [1]. 1 2


The remaining thickness variables can be readily related
to the streamwise ones through the shape factors using the
assumed shape of the cross ow velocity pro le. As an ex- Figure 3: Global, cell and nodal streamline coordinate sys-
ample, tems
21 = A 11 :
Finally, the identity 4.1 Coordinate Systems
12  21 2 The basic computational element is a planar cell, indi-
completes the determinancy of the system of equations and vidually constructed from body coordinates in the global
variables. Cartesian system. A local Cartesian coordinate system,
denoted (x;z ), is de ned at the cell as illustrated in Fig-
It is important to point out that these relationships were ure 3. In the current formulation, this system is oriented
derived using only the outer portion of Johnston's model, so that the z-axis is parallel to face 1-2 of the the cell. The
which is equivalent to assuming that the pro le w(u) is x-axis corresponds roughly to the external streamline direc-
linear in the hodograph plane. Obviously, the hypothesis tion and to the solution marching direction. A right-handed
of linearity is wrong near the wall because of the no-slip coordinate system will, therefore, correspond to the y-axis
condition, but the concerned region is very thin. Cousteix pointing out of the cell.
and Houdeville [3] claim that the resulting error in the esti- An additional local coordinate system, denoted by (s; n),
mated thicknesses is negligible even if the maximum value is de ned at each of the four nodes of the cell such that s
of w occurs where u=qe is approximately 0.6. Consequently, is aligned to the external ow streamline direction and n
this approximation is expected to be valid and is unlikely to the cross ow direction. This is the discrete version of
to signi cantly a ect the accuracy of the cross ow results. the analytic streamline system discussed earlier. It is nec-
essary to de ne this system to accommodate the empirical
4 Numerical Scheme relations discussed in the previous section.
Most variables are orientation dependent. Transforma-
Discrete coupled equations are solved simultaneously for tion relationships between the (s; n) and (x; y) coordinates
rows of cells, each considered individually, as part of space are readily obtained by substituting the transformed veloc-
marching the solution along a direction corresponding ap- ity components (~u; w~),
proximately to the primary external- ow direction. The 0 1 0 10 1
procedure involved in constructing the discrete equations
requires the de nition of global and local Cartesian coordi- @ u~ A = @ cos( ) sin( ) A @ u A
;
nate systems. w~ sin( ) cos( ) w
into variable de nitions, where is the angle between the
coordinate directions s and x. For example, the two com-

4
ponents of the displacement thickness in (x;y) are enough to be entering the calculation domain along this initial row
to uniquely determine the components in (s; n), of cells to satisfy the characteristic direction requirement.
Along a row of N cells, there are 3N+3 primary unknowns
1 = cos( ) x sin( ) z but only 3N discrete equations. Therefore, three additional
2 = sin( ) x + cos( ) z equations need to be supplied to determine the solution.
These equations can be speci ed at the boundary nodes or
at an interior node.
Rotation of all vector quantities into the local cell sys-
tem before di erencing properly accounts for the grid non- It is clear from the domain of in uence principle that
orthogonality. These transformations are e ectively equiv- boundary conditions must be speci ed along those edges of
alent to adding the necessary metric-gradient terms that the domain where characteristic directions indicate the pas-
arise when the equations are written for a general curvilin- sage of information into the domain. In principle, as many
ear coordinate system. boundary conditions as there are characteristics crossing
the boundary are required. An immediate corollary is that
in regions where characteristics indicate that information
4.2 Derivative Terms is outbound, no conditions are needed; if any are applied,
they will exert no in uence on the solution in the interior of
The rst-order derivative terms are discretized using the domain. By the same reasoning, the e ect of boundary
Green's theorem by performing a discrete contour integral conditions on the solution is limited to a region bounded
around a cell. The value at an edge of the cell is simply an by the path traced by the most angularly-directed charac-
average of the values at the two neighboring nodes. The teristic.
discrete terms are formally second-order accurate.
For situations where Dirichlet conditions are not known
at the domain boundary, zero-curvature boundary condi-
4.3 Local Construction of Discrete Equations tions, where the value at the boundary node is extrapo-
lated from the two adjacent ones, are found to perform
The discrete equations along each row of cells are solved adequately. These are preferred over Neumann conditions
separately, as indicated earlier. For each of these rows, the where an arbitrary gradient is speci ed because the latter,
constituent cells are visited individually, at each iteration, in general, will not be consistent with the gradients of ex-
and a residual-vector, R, is constructed for the cell. R is a ternal velocity components and edge density.
function of the variables at the four nodes of the cell. The
solution is stored at the nodes.
4.5 Residual distribution to nodes
The vector of primary variables U is known at nodes 1
and 2 from the solution at the previous upstream row. The The directions of the characteristic lines have to be con-
values at nodes 3 and 4 are unknown. They are initialized sidered in the assembly of the system of cell residuals R.
to be equal either to the values at nodes 1 and 2, respec- This is achieved by de ning a new vector of residuals R~
tively, or to the linear extrapolation of the values at nodes such that
1 and 2 and next upstream nodes. It is necessary that these R~j = (1 k )Rj + k Rj+1 ;
initialized values be considered to be in the local (s; n) co-
ordinate system. where k is a function of the angle that the local kth
characteristic makes with the corresponding cell edges. In
The procedure employed at each of the nodes at a cell Figure 4, which can be considered a scalar case, is positive
is as follows. First, the closure formulae are applied to clockwise so that for the characteristic oriented towards the
determine all the non-primary unknowns as a function of right in Figures 5 and 6,  will be greater than 0.5. Thus,
U in the (s; n) system. Next, these unknowns are rotated R~ corresponds to the net residual contributions at a node of
into (x; z ) coordinates. Then, in Equation 1 the derivative the residual vectors of the two neighboring cells in propor-
terms are approximated by the nite-volume method and tion to the local direction of information ow. The slope of
the source term components by averaging the nodal values. the  function at = 0 is a user control.
The derivatives of velocities are also evaluated using the In principle, the residual of the equation in terms of a
nite-volume method from their known components in the characteristic variable should be distributed in proportion
(x; z ) system at the nodes and then simply multiplied to to the angle of the corresponding characteristic. In the
the remaining source-term cell-averaged variables. current formulation, however, the distribution of the en-
The system of three discrete equations, thus developed, tire vector of residuals for the conservative formulation of
can be expressed as a function of the unknowns by writ- Equation 1 is governed by the same value of . This is
ing the residual vector R as a function of U at the two coupled to the 1 characteristic direction computed at the
downstream nodes 3 and 4, R = R(U3 ; U4 ). upstream row of nodes. While not ideal, this choice is found
to perform satisfactorily in practice.
4.4 Boundary Conditions Two residual equations at either end of the row of cells
are provided as boundary conditions. These will blend into
An initial condition needs to be prescribed along the rst the global system of equations when the 1 characteristic is
row of nodes to start the calculation. Physically, uid has inbound. In Figure 5, a \shock" is schematically illustrated;

5
µ
1

Measurement plane

1 2 3 4 5 6 7 8 9 10
Guiding vane

0.90 m
−π/2 π/2 α

0.52 m
Figure 4: Residual distribution factor variation with angle

o
35
Flow

Figure 7: Schematic of van den Berg/Elsenaar experimental


setup
24000:

16000:
Figure 5: \Shock-cell" con guration
Re11
8000:

0:
0:40 0:65 0:90 1:15 1:40
x=L
Figure 8: Computed variation of Reynolds number

Figure 6: \Expansion-cell" con guration


conditions, the blocks in the next o -diagonal will be zero
the characteristic reverses direction in a manner where both at one or both ends of the matrix.
boundary conditions are factored into the global system. In
the limit that the slope of ( ) becomes in nite at = 0,
this method is equivalent to discarding the residual equa-
5 Results
tion at the shock cell. An \expansion" case is shown in Some results from two calculations are presented in this
Figure 6. Here, in the limit of in nite d =d( = 0), the section. The rst was performed to simulate a quasi-two-
global system becomes singular because the same residual dimensional ow. The primary intention was to compare
is distributed to two nodes. Hence, one of the equations is the computed results with a well-known experiment in or-
replaced by an internal boundary condition which simply der to verify the empirical closure employed. The latter
forces the variables at one of the two singular nodes to be calculation was performed over a nite swept tapered wing
the average of the values at the two adjacent nodes. mainly to demonstrate this method's three dimensional ca-
The set of coupled non-linear discrete equations, R(U ) = pabilities.
0, are solved using the well-known Newton-Raphson
method. The Jacobian-matrix [@R=@U ]m elements are con- 5.1 An In nite Swept Wing
structed at the FORTRAN level by successive multiplica-
tion and addition of constituent partial derivatives using the The experimental results presented here were obtained
chain rule. The global Jacobian-matrix structure is block- on an \in nite swept wing" under incompressible ow con-
tridiagonal in the interior. For zero-curvature boundary ditions. The experiment was conducted by van den Berg

6
and Elsenaar at The Netherland's National Aerospace Lab-
oratory in 1972 [12].
The wing is actually a at plate with an adverse pres-
sure distribution induced on it by a suitably-shaped body
positioned near the plate. The experiment was designed so
that, over the forward portion of the experimental model,
the pressure remained nearly constant before being made
to increase gradually through the presence of the body.
The pressure-gradient and sweep combine to induce three-
dimensional separation in the vicinity of the trailing edge.
Considerable care was exercised in the attempt to sim-
ulate in nite-wing conditions. The boundary layer could
be maintained very nearly quasi-two-dimensional with the
help of guide vanes on either side of the region of interest.
The positions of the measurement stations relevant to this
Figure 9: Computed limiting-streamline vector plot calculation are depicted on the experimental model along
with other notable features in Figure 7. The line of equally-
spaced measurement stations was 0.90 m long. The angle
of sweep of the wing was 35o .
Detailed measurements and descriptions of the experi-
mental techniques are provided by the authors. This ex-
40:0 periment has been extensively used to compare details of
w
closure assumptions in many di erent kinds of numerical
methods (see, for example, [3, 10, 13]).
26:7
1
The calculation was performed on a 2003 grid assuming
; w quasi-two-dimensional ow. One set of grid-lines was par-
allel to the freestream direction and the other to the wing
13:3 2 leading edge. The computational domain corresponded to
the region enclosed by stations 1 and 10 in the experiment.
3
A cubic spline-interpolation technique was employed
0:0 to smooth the supplied external velocity measurements.
0:40 0:65 0:90 1:15 1:40
x=L Clearly, the spanwise component of velocity gradually in-
creases downstream. The initial condition used was the
Figure 10: Computed variation of limiting-streamline and set of experimentally-obtained values at the rst station.
characteristic angles A simple boundary condition was applied where zero-
gradients were enforced in the direction of the wing iso-
bars, and thus one set of grid lines, along the side of the
computational domain with the in ow.
Computed results are presented in Figures 8 to 11 as the
solid line. The marching coordinate has been scaled by a
0:00360 length of 1 m which is on the order of the wing chord.
The comparison for Re11 is good up to the station near
x=L  1:0, where the spanwise ow is still a small frac-
0:00280 tion of the ow in the marching direction. This is also the
cf case with the comparison for w . The negative value of w
indicates that the limiting streamlines are turning span-
0:00200 wise faster than the external ow streamlines with march-
ing distance. The tuft plot in Figure 9 vividly illustrates
this turning. The calculated three-dimensional separation
0:00120 line occurs where the limiting streamlines coalesce and this
0:40 0:65 0:90 1:15 1:40 is found to be close to the experimentally observed loca-
x=L tion. Experimentally, separation is observed between sta-
Figure 11: Computed variation of skin friction tions 8 and 9, or where x=L  1:3. Figure 10 is a plot
of the variation of the three calculated characteristic direc-
tions, shown in broken lines, juxtaposed against the solid
line for w . The three characteristic lines are observed to
be oriented largely within the angular limits de ned by the
external ow streamline ( w , = 0) and the limiting wall

7
Lower surface Upper surface
0:65 0:65

0:40 0:40
x 30k x 30k

0:15 10k
0:15 10k

0:10 0:10
0:00 0:25 0:50 0:75 1:00 0:00 0:25 0:50 0:75 1:00
z z
Figure 12: Computed Reynolds number contour plots

Lower surface Upper surface


150000 150000:

100000 100000:
Re11 Re11
50000 root 50000:
mid root

tip mid tip


0 0
0:000 0:125 0:250 0:375 0:500 0:000 0:125 0:250 0:375 0:500
x x
Figure 13: Computed Reynolds number chordwise variation

streamline. Furthermore, the line labeled 1 is evidently The external ow is nominally two-dimensional over most
that associated with the limiting wall streamline. of the wing except in the leading edge and tip regions, where
The computed values of cf and H compare less favorably the spanwise components of velocity are signi cant.
with the experimental data. The trend of cf closely fol- The calculation is performed for a freestream Reynolds
lows that of other published numerical case studies, employ- number, based on the mid-wing chord, Rec , of 50  106 .
ing equilibrium closure, for the same experiment [10, 13]. This is typical for a large transport in cruise where transi-
Comparisons are good to x=L  1:0 but almost univer- tion occurs at the attachment line.
sally poor beyond, providing strong evidence for the widely- The initial condition is estimated from the solution of the
acknowledged inadequacy of equilibrium closure in situa- incompressible equations for laminar ow, using Falkner-
tions with separation and large cross ow. Skan pro les, to generate a value for 1 at the nodes on
either side of the attachment line [2]. The assumption that
5.2 A Finite Swept Tapered Wing the ow is locally identi able with the ow on the lead-
ing edge of an equivalent in nite swept wing is made. In
The development of a compressible boundary layer over addition, a value of H = 1:3 is speci ed to yield the ini-
a nite tapered swept wing, at the moderate angle of at- tial condition for 11 . These thicknesses need to be scaled
tack of 1:5o and a Mach number of 0.3, was calculated to slightly to insure that Re11 is within the range necessary
demonstrate the full three-dimensional capabilities of this to sustain turbulent ow and consequently also within the
method. The wing section is a NACA 0012 airfoil. The range of the empirical correlations. w is approximated to
external inviscid velocity and density distributions are sup- be zero. This process only requires @ue =@n at the attach-
plied by a potential ow solver [9]. ment line and the laminar viscosity  as inputs. While
The wing-surface meshes measure 5722 on both the up- this approach is approximate it is nevertheless employed
per and lower surfaces with the greater number of nodes in because the strong favorable pressure gradient which ex-
the chordwise direction. It is important to note that the ists just downstream of the attachment line tend to make
global z-coordinate direction on the lower surface is oppo- conditions over the major part of the wing comparatively
site to that on the upper surface. This is necessitated by insensitive to conditions on the attachment line [4].
the convention that the y-coordinate increase in the direc- Two sets of computed results, one each for Re11 and w
tion, normal to the surface, on the side with the ow. All are shown in Figures 12 to 15. Each set consists of contour
coordinates are normalized using the wingspan.

8
Lower surface Upper surface
0:65 0:65

0:40 0:40 :o
2 0 1 0 :o
1:0o 0 0:o
x 0:5o x
0:0o
1:0o
0:15 0:15

0:10 0:10
0:00 0:25 0:50 0:75 1:00 0:00 0:25 0:50 0:75 1:00
z z
Figure 14: Computed limiting-streamline-angle contour plots

Lower surface Upper surface


4:0 20:0

0:0 12:0
root mid tip
w (o ) w (o )
4:0 root 4:0
mid tip

8:0 4:0
0:000 0:125 0:250 0:375 0:500 0:000 0:125 0:250 0:375 0:500
x x
Figure 15: Computed limiting-streamline-angle chordwise variation

plots and line plots along three chordwise stations for both non-orthogonal grids. The chief singular achievement of
the upper and lower surfaces: the root, near midwing and the method is the application of a nite-volume formula-
at the tip. Note that counterclockwise angles are considered tion that keeps the equations independent of surface and
positive. grid metrics. This results in substantial syntactic simpli -
Unlike the in nite swept wing case where a sustained cation and great exibility in the choice of grids. In addi-
adverse pressure gradient was applied to induce three- tion, a method of residual distribution from cells to nodes
dimensional separation, it is found here that, for the Rec is developed to placate system hyperbolicity.
considered, the ow is attached over the entire wing surface. The use of an integral form of the equations yields a
The variation of w con rms that the most severe three- very fast method; the calculation on the nite wing takes
dimensional e ects are largely restricted to areas near the only about 15 seconds on a DECstation 5000. In terms of
leading edges and wing tips. Immediately downstream of integral variables, this method is as accurate as many dif-
the leading edge, drastic variations are seen. These are nu- ferential, equilibrium-closure methods reported in a recent
merical e ects, presumably a consequence of the slightly Eurovisc-Workshop [13].
incorrect initial condition for w , the low values of Re11 The contextual basis of this paper is to develop a bound-
and the strong pressure gradient associated with the at- ary layer method with the eventual aim of incorporating
tachment line. However, the solution quickly becomes more it into a viscous/inviscid code. The use of the Newton-
placid with marching distance. As noted earlier these vari- Raphson technique is very helpful in this regard; the ele-
ations are not expected to have a substantial e ect on the ments of the Jacobian-matrix can be appended to the global
physical accuracy of the solution over the rest of the wing. Jacobian matrix in a Newton-Raphson-based Full-Potential
Sawtooth oscillations seen in the contour plots are a re- code to develop a global coupled scheme.
sult of numerical sti ness. A ner grid will alleviate the
problem.
7 Acknowledgements
6 Concluding Remarks This research was supported by The Boeing Company
under the supervision of Dr. Wen-Huei Jou.
A novel numerical method is developed for solving the
three-dimensional integral boundary layer equations on

9
References [14] D.L. Whit eld, T.W. Swa ord, and T.L. Donegan. An
inverse integral computational method for compress-
[1] T. Cebeci and A. M. O. Smith. Analysis of Turbulent ible turbulent boundary layers. In W. Haase, editor,
Boundary Layers. Academic Press, New York, 1974. Recent Contributions to Fluid Mechanics. Springer-
Verlag, 1982.
[2] J. Cousteix. Three-dimensional boundary layers. in-
troduction to calculation methods. In Computation of
Three-Dimensional Boundary Layers Including Sepa-
ration, Apr 1986. AGARD-R-741.
[3] J. Cousteix and R. Houdeville. Singularities in three-
dimensional turbulent boundary layer calcuations and
separation phenomena. AIAA Journal, 19(8), Aug
1981.
[4] N.A. Cumpsty and M.R. Head. The calculation of
three-dimensional turbulent boundary layers. part II:
Attachment-line ow on an in nite swept wing. The
Aeronautical Quarterly, 28:150{164, 1967.
[5] M. Drela. Two-Dimensional Transonic Aerodynamic
Design and Analysis Using the Euler Equations. PhD
thesis, MIT, December 1985. Also, MIT Gas Turbine
& Plasma Dynamics Laboratory Report No. 187, Feb
1986.
[6] J.P. Johnston. Three-dimensional turbulent boundary
layers. PhD thesis, MIT, 1957. Also, MIT Gas Turbine
Laboratory Report No. 39.
[7] B. H. Mughal. A calculation method for the three-
dimensional boundary-layer equations in integral form.
Master's thesis, MIT, September 1992.
[8] J.F. Nash and V.C. Patel. Three-Dimensional Turbu-
lent Boundary Layers. SBC Technical Books, Atlanta,
1972.
[9] P.B. Poll. Full potential analysis and design of tran-
sonic propellers. CFDL TR-91-1, MIT, Feb 1991.
[10] P. D. Smith. An integral prediction method for three-
dimensional compressible turbulent boundary layers.
R & M Report 3739, Aeronautical Research Council,
HMSO, London, 1972.
[11] T. W. Swa ord. Analytical approximation of two-
dimensional separated turbulent boundary-layer veloc-
ity pro les. AIAA Journal, 21(6):923{926, 1983.
[12] B. van den Berg and A. Elsenaar. Measurements in
a three-dimensional incompressible turbulent bound-
ary layer in an adverse pressure gradient under in -
nite swept wing conditions. Technical Report NLR
TR 72092U, Nationaal Lucht-En Ruimtevaartlabora-
torium, Amsterdam, 1972.
[13] B. van den Berg, D.A. Humphreys, E. Krause, and
J.P.F. Lindhout. Three-dimensional turbulent bound-
ary layers { calculations and experiments. Notes on
numerical uid mechanics, Germany, 1988.

10

You might also like