1336 AIAA JOURNAL VOL. 6, NO.

Solution of a Three-Dimensional Boundary-Layer

Flow with Separation
HARRY A. DWYER*
General Electric Space Sciences Laboratory, King of Prussia, Pa.

A method of calculating accurate solutions of the three-dimensional laminar boundary-

layer equations has been developed. The method is applied to a problem that exhibits inter-
esting crossflow phenomena, such as a flow reversal driven by crossflow convective terms and
crossflow influences with zero freestream crossflow velocity. The method of solution used
is an implicit finite difference scheme, and the stability and convergence properties of this
scheme seemed to be good. Also, the important question of initial conditions for three-
dimensional boundary-layer flow is discussed, and a method of obtaining initial conditions
is derived.
Downloaded by RYERSON UNIVERSITY on February 3, 2015 | http://arc.aiaa.org | DOI: 10.2514/3.4745

I. Introduction remains a serious doubt that these solutions are good approxi-
mate solutions to the three-dimensional boundary-layer
T HE object of this investigation is to obtain accurate
solutions of the three-dimensional boundary-layer equa-
tions for the incompressible flow over a flat plate with pres-
equations. The crossflow equations violate the region of
influence principle as stated by Der and Raetz,6 and, since
the region of influence contains momentum and heat-transfer
sure gradients and flow separation. The method used to information about the previous history of the boundary-layer
obtain solutions was a numerical one, and has exhibited development, this could be quite serious. Also, in most
itself to have the advantages of being fast and stable with crossflow problems, the curvature of the fluid streamlines
good convergence. It also behaves well in the vicinity of the near the wall of the boundary layer behaves in a much
flow reversal line, and, although not reported in this article, different manner than the inviscid streamlines. In the
the method has been applied to compressible flow. vicinity of the wall it is not necessarily true that the cross-
The importance of three-dimensional effects in boundary- flow derivative is of small order just because the crossflow
layer analysis has long been recognized, but, due to the lack velocity is small. There are many examples of flows along
of adequate analytic and numerical methods, progress in a line of symmetry where the crossflow velocity is zero, but
solving these types of problems has been slow. The results the derivative of crossflow is a first-order term. Therefore,
of analytic or semianalytic studies have generally been the first-order continuity equation may contain crossflow
limited to situations where one of the derivatives in the plane terms even though the crossflow itself is small. The final
of the body surface disappears, as in flow over a yawed objection to the crossflow flow method is that there is really
cylinder of infinite length.1 Although these solutions are no systematic way of calculating an error bound on the solu-
important in their own right, they offer only a limited amount tions obtained by this method.
of insight into the larger class of three-dimensional problems The only other attempt at obtaining a solution of the full
with derivatives in all coordinate directions. Another semi- three-dimensional boundary-layer equations, to the au thor's
analytic technique is the one used by Sowerby,2 which con- knowledge, was that of Der and Raetz.6 Der and Raetz did
sisted of using a series expansion about the leading edge of a a very good job of analyzing the mathematical and physical
flat plate. This method is useful in the vicinity of the lead- characteristics of the three-dimensional boundary-layer
ing edge, but leads to very complicated systems of nonlinear equations, and properly posed their problem mathematically.
ordinary differential equations away from the leading edge. However, their choice of coordinate transformations was
Therefore, the method seems to be limited as a practical tool awkward and complicated, and their finite-difference method
in boundary-layer analysis. seems to have a practical problem of slow convergence. Der
In recent times, the so-called crossflow method has been and Raetz used a finite-difference method which can be said
very popular in attempting to solve three-dimensional bound- to be the three-dimensional analogue of the Dufort-Frankel
ary-layer problems. In this method, the boundary-layer explicit method commonly used on two-dimensional parabolic
equations are written in terms of the inviscid streamline problems. Chang and Cheng7 have discussed the conver-
coordinate. A perturbation expansion technique is then gence problems of this method.
used to generate sequences of two-dimensional problems with
the zeroth approximation being identical with the small
crossflow approximation of Hayes.3 Fannelop4 has recently II. Analysis
applied the implicit finite-difference techniques developed A. Mathematical and Physical
by Blottner5 for two-dimensional problems to these equations
and has had success in solving them. However, there still The three-dimensional boundary-layer equations in their
incompressible and constant transport property form are
Received September 29, 1967; revision received March 4,
(5E7/5Z) + (57/5 F) + = 0 (1)
1968. The author would like to thank General Electric for its
generous support. The discussions with F. G. Blottner, H. Li, *"* ^ _ _ -1- ^J1J i ^_21 /o\
H. Lew, and C. Strom were also very helpful in formulating and
carrying out the present work. Last but not least, the author 57 bZ ~ ~ 7 *>X dY2
would like to thank V. Johnson for her excellent help in carrying
out the numerical computations. (3)
* Specialist, Gas Dynamics; now Assistant Professor, Mechan- • 5^ ' 57 ' 5Z p 5Z ' 572
ical Engineering, The University of California, Davis, Calif.
Member AIAA. where 7 is the boundary-layer coordinate, and U, V, and TF
 JULY 1968 THREE-DIMENSIONAL BOUNDARY LAYER WITH SEPARATION 1337

are the velocities in the X, Y, and Z directions, respectively.

The terms involving spatial derivatives of p, the pressure,
represent the pressure gradients imposed on the boundary
layer by the inviscid flow, and v represents the constant
kinematic viscosity.
Equations (1-3) are quasi-linear, and this property allows
for their mathematical characteristics to be calculated. An
analysis of the characteristics of the aforementioned system
of equations reveals that any surface parallel to the F-axis is
a characteristic one, and that the characteristics are degener-
ate because the equations do not contain second-order deriva-
tives with respect to the independent variables X and Z.
In general, for three-dimensional problems, the classifications
elliptic, parabolic, and hyperbolic lose their meaning, but
the three-dimensional boundary-layer equations can be
classified as being essentially parabolic. This parabolic na-
ture is easily recognized by replacing the variables X and Z
by a streamline variable S, thereby forming a parabolic
equation with two independent variables. This parabolic
system of equations poses a combined initial and boundary-
Downloaded by RYERSON UNIVERSITY on February 3, 2015 | http://arc.aiaa.org | DOI: 10.2514/3.4745

value problem, which will be discussed further on in the

text.
Having discussed the mathematical nature of the boundary-
layer equations, it is important to investigate the mechanisms
of physical transport in three-dimensional boundary layers
Fig. 1 Geometry of the flow problem.
before introducing a method of solution. In a three-dimen-
sional boundary layer, momentum is transferred in the fol-
lowing manner. First, due to the diffusive nature of vis- will be used in the example given in this paper and has been
cosity, momentum will be transferred along the boundary- used by Sowerby, is
layer coordinate across its entire length. Second, due to
convection and the divergence of streamlines across the bound- x =X z =Z 7] = (Ui/2va X)l'*Y (5)
ary-layer coordinate, the fluid in which the momentum
which is almost identical to a Blasius transformation, except
was transferred will move downstream over a limited range
that Ui is now a function of the two nonboundary-layer
of influence determined by the divergence of the streamlines
coordinates. For further extensions of the two-dimensional
across the boundary layer. Since these regions of influence
transformations to three dimensions, one should consult
contain the previous history of the boundary layer, they
the work of Moore.1
should be taken into account in any method of solution.
If the transformations given in Eq. (5) are applied to the
In this investigation, a numerical method of solution is de-
system of Eqs. (1-3), the following new set of equations is
veloped which obeys the region of influence and also models
obtained:
the diffusive nature of the boundary-layer coordinate by
solving the net points across the boundary layer simul-
taneously.
/ JL » 1 U±*
57] \ 2z 2 Ui
B. Coordinate Transformation ~ = 0 (6)
517 57?
When solving the boundary-layer equations by finite-
difference techniques, the boundary-layer equations repre-
sent a free boundary problem, since the distance of the edge (7)
of the boundary layer away from the wall is an unknown.
The nonuniformity of the thickness of the boundary layer
can cause complications in obtaining solutions; therefore, it (8)
is of great advantage to use a coordinate system that makes
the boundary layer have a nearly uniform thickness. Ex- where
amples of these types of transformations in two-dimensional
problems have been associated with the names of Blasius, V = [ +
Mangier, Howarth, and Dorodnitsyn. For three-dimen-
sional boundary-layer problems, the transformations men-
tioned previously are not as useful; however, they still contain Ulx =
a lot of the major boundary-layer growth phenomena. An The preceding system of equations will be used to obtain ac-
example of a possible extension of the two-dimensional trans- curate solutions to a three-dimensional boundary-layer flow.
formation to three dimensions for a Cartesian geometry is
x = X z =Z 7] = (4) III. Problem
where Ui and Wi represent the inviscid flow velocities and The three-dimensional boundary-layer problem which we
x is the distance from the leading edge of the flow. This shall investigate is shown in Fig. 1. The geometries of the
transformation takes into account the boundary-layer bodies in the figure consist of an infinite cylinder joined with
variation due to changes in the freestream velocity and also its axis perpendicular to a thin flat plate. It will be as-
due to the shear stress induced growth from the leading sumed that the thin plate does not affect the inviscid flow
edge of the flow. It should be pointed out, however, that around the cylinder. We will then solve for the three-
this transformation will not remove all the boundary-layer dimensional boundary-layer flow on the flat plate due to the
variation, and one should test for large nonuniformities in inviscid flow created by the cylinder. Since the inviscid
boundary-layer thickness. Another transformation, which flow solutions are generally only valid on the forward portion
 1338 H. A. DWYER AIAA JOURNAL

of the cylinder, the boundary-layer development will only along the edge of the flat plate, the boundary-layer profiles
be found in front of the cylinder. for U and W may be obtained from the Blasius results.
The following equations, which are a result of potential It should be mentioned as a word of caution that other
flow theory, will be used to calculate the inviscid flow or three-dimensional boundary-layer problems involving com-
boundary conditions for our three-dimensional boundary- pressible flow or axisymmetric bodies at angle of attack will
layer problem: result in completely different initial condition equations. An
example of this point is the sharp cone at angle of attack where
z2 - (x - x) 2 the initial condition equations are a system of partial differ-
ential equations in two independent variables.
The initial conditions along X = 0 are not enough to begin
to numerically solve the three-dimensional boundary-layer
(I/a2) [(x - equations because we must also have initial or starting condi-
tions along some z equal to a constant line. To obtain these
added conditions one must have a line of symmetry or know
the initial conditions from some independent source. For
the cylinder-plate problem, the z equal constant initial condi-
Also, we shall choose the following values of the constants tions can be obtained by solving a modified system of bound-
shown previously: V^ = 3050 cm/sec (freestream velocity, ary-layer equations along the line of symmetry of the flat-
100 fps), x = 45.7 cm (distance of cylinder axis from leading plate boundary-layer flow. Along this line W equals zero
edge), a = 6.1 cm (cylinder radius), and v = 0.279 cm2/sec and the ^-momentum equation is identically zero. How-
Downloaded by RYERSON UNIVERSITY on February 3, 2015 | http://arc.aiaa.org | DOI: 10.2514/3.4745

(kinematic viscosity). ever, since dTF/dz is not zero along this line, we must generate
a new equation to obtain dTF/dZ. This equation can be
IV. Initial Conditions gotten by differentiating the W equation and taking ad-
vantage of symmetry. The result is
As mentioned previously, the three-dimensional boundary-
layer equations pose a combined initial and boundary-value
problem. The boundary conditions are usually obtained
from the independent inviscid flow equations, whereas the
initial conditions must be obtained somehow from the (11)
boundary-layer equations themselves. The method used in
this paper to obtain initial conditions is to solve the equations The independent variable is dW/dz and this equation should
resulting from the first term in a series solution of the bound- be solved along the z equal to zero line in the x-rj plane simul-
ary-layer equations at the leading edge. Actually, as is taneously with the [/-momentum and continuity equations.
well known, the boundary-layer equations break down in Also, it can be shown that the variable dTF/dz has a Blasius
the vicinity of the leading edge of a sharp body, but in prac- functional form at x = 0.
tice the distance over which they do not hold is so small that
a series solution evaluated at the leading edge represents
an excellent approximate solution to the flow. Since it can V. Numerical Method
be shown that taking the limit of the boundary-layer equa-
tions as £ approaches zero is equivalent to obtaining the The three-dimensional boundary-layer equations consist
first term in a series solution, we shall use the limit procedure. of a set of parabolic equations which must be solved as a com-
In the limit of £ approaching zero, we have bined initial and boundary-value problem. If one is to ob-
tain accurate solutions of these equations by a numerical
Continuity : method, this method must be capable of starting from the
= 0 initial conditions and then marching downstream satisfying
all the boundary conditions. Because of the fact that
X-momentum : boundary layers start out from almost zero thickness near
- (rjvU/U^U/brj) =v the leading edge of sharp bodies, the explicit numerical meth-
ods2 commonly used for parabolic problems impose an un-
Z-momentum : realistic step size restriction in the x and z direction. There-
fore, it seems that one should develop an extension of the
Crank-Nicolson implicit method with its nice stability fea-
Letting f^ = U/Ui and gn = W/Wi, integrating the con- tures which has been used in the past by Blottner and Fanne-
tinuity equation, and substituting the results into the mo- lop for two-dimensional boundary-layer problems.
mentum equations, we have The most important thing to recognize in the three-
dimensional boundary-layer equations is that there is only
= 0 (9) one coordinate where second derivatives occur. If one then
evaluates derivatives along the boundary-layer coordinate
(10) in an implicit manner, the resulting set of simultaneous
equations will have a tridiagonal matrix form, and can be
For the problem to be solved in this paper, Eqs. (9) and (10) evaluated by the efficient algorithm developed for those
must be solved along the X = 0 line of the flat plate as a types of simultaneous equations. The problem then is to
function of the inviscid flow. However, since the problem obtain a good way of evaluating the x and z or nonboundary-
is posed with the boundary condition layer coordinate derivatives. To illustrate our method of
setting up the finite difference equations, we will use Fig. 2.
f = f = ff* = fv = 0 at 77 = 0 In Fig. 2 is shown the system of net points where the indices
gn = fr, = 1 at rj = oo m, n, and o define the grid location of points with respect
to the x, ifj, and z axes, respectively. It is assumed that
all solutions will have the same functional form. From in- the boundary-layer flow is known for all x = m and z — 0 — 1
spection, it is obvious that the form of the solution is a Blasius (note this is consistent with the initial value problem
flat-plate solution, and this can be obtained from tabulated posed), and it is required to calculate the velocities at the net
values. Therefore, if one knows the inviscid flow velocities point m + 1, 0 for all 77.
 JULY 1968 THREE-DIMENSIONAL BOUNDARY LAYER WITH SEPARATION 1339

The following difference quotients have been used in evalu- where — (l/p)(dp/dz) should be evaluated at the center of
ating derivatives in the ^-momentum equation the four grid points. Equation (13) can be evaluated by
the same algorithm that is used for two-dimensional parabolic
(5U _ Um+l,n,o — Um,n,o . Um+l,n,o—l _ Um,n,o—l
problems.5-8
"" The ^-momentum equation can be put into a finite-differ-
ence form that is essentially identical to Eq. (13); however,
Um+l,n,o l, n,o—\ . Um,n,o ~
the continuity equation requires a different treatment. Since
2Az there is no equation that contains V derivatives in the x and
z directions, we have chosen to calculate FOT+i,TC,0 after U and
W have been calculated at the unknown grid location. The
finite-difference form that was used for the continuity equa-
rjm-\-\,o —1 - U T]m,o — 1
tion is
~ ~

(12)

X
2Ax 2V x + Ui )
Um+l,n,o ~ J7m+l..-l.» n--
Downloaded by RYERSON UNIVERSITY on February 3, 2015 | http://arc.aiaa.org | DOI: 10.2514/3.4745

4Arj

rj I hz fWm+l no — Wm+l 1o + Wm n o ~ Wm n-l o + Wm+l n o-l Wm+l n-l o-l + Wm „ „-! ~ ^Fm n-l o-l'

2 I 7i \ 4A>7 4AZ

m n m n
£ 1 ' '° ' '°~ m,n,o m_,n,o— \
(14)
2Az
With the known boundary condition on V at the wall, Eq.
~T~
2 ' + (14) can be solved for Vr*+i,n,o from the wall out.
4A77 The analysis of the stability and convergence properties
\ o —1 l~ of the previous equations is at present not available, because
of the nonlinearity of the basic partial differential equations.
However, some insight into the behavior of the equations
Um U i+I,n,o— I W m+l,n,o-l Wm
U' = PF' = may be gained from the two-dimensional results since the
three-dimensional boundary-layer equations may be put
+ into a locally two-dimensional streamline form. It has been
V' = found by numerical experimentation with an implicit finite-
where Urj and UTJTJ are the first and second derivatives evalu- difference method that the two-dimensional boundary-
ated at points where U is known. As can be seen from the layer equations have a stability and convergence behavior
difference forms and Fig. 2, the derivatives have been evalu- similar to the unsteady one-dimensional heat conduction
ated at the center of four grid points, and this averaging is equation. An analysis of the properties of the heat con-
meant to be an extension of the Crank-Nicolson method for duction equation can be found in Richtmyer's book.8 In
two-dimensional problems. Also, the finite-difference equa- our results with the finite-difference forms shown in Eqs.
tions are linear and decoupled since the quantities U and W (13) and (14), we have found that the step sizes in the x and
have been evaluated at known grid points. A necessary z directions do not seem to be restricted by stability con-
condition for this finite-difference scheme to give stable siderations except when U or W become negative.
solutions is that U and W must be positive, since a nega-
tive U or W is equivalent to a time reversal in a linear para- VI. Results
bolic problem. Positive U and W are equivalent to assuming As has been stated, the problem to which we will apply
that no time reversal occurs in the problem. (Note that the foregoing analysis consists of an infinite cylinder mounted
there are two timelike coordinates in this problem, x/U and onto a semi-infinite flat plate, Fig. 1. Since the cylinder
z/W.) causes an adverse pressure gradient in front of itself, it is
Applying the results of Eq. (12) to the ^-momentum equa- expected that the U component of velocity will go through
tion, the following finite-difference equation results: a flow reversal along a line in front of the cylinder. In the
Um+i,n+i,o ^~:~ — , " ie
region of flow reversal, the implicit method develops a nu-
merical instability and is not capable of calculating accurately
the fluid flow. Since the adverse pressure gradient is a
Um+l,n-l,o ~ 0A
maximum along the line of symmetry, the flow reversal in
the boundary layer will first occur for the line of symmetry.
.i1-n-°~ii——
— U i 4- U The ^/-velocity profiles along the line of symmetry are
D
-<K ^^n.o-i-r o
shown in Fig. 3. Near the leading edge of the flat plate, the
17-velocity profiles have a Blasius flat-plate shape, and are
not greatly affected by the cylinder's adverse pressure
gradient. At a distance of z = 18.3 cm, almost half the
W
U m,n,o Um,n,o-l

2A2
Urmm.o ~T ^7;
~ U
i m+i,

-)
,-i +
1 dp
p dx
^ rjT)m,o— 1 i
(13)
distance to the cylinder, one begins to see the effect of the
adverse pressure gradient on the velocity profile. The
condition of incipient flow reversal is seen to occur at x =
24.4 cm, and a flow reversal at x = 25 cm. The step sizes
€ V
" (
4 ) along the x and 77 coordinates were A# = 0.61 cm and AT? =
 1340 H. A. DWYER AIAA JOURNAL

(it should also be pointed out that V will probably become

large and negative after the flow reversal point).
Before presenting the three-dimensional boundary-layer
results away from the line of symmetry, we must discuss the
m+1 , n, o
• • calculation of the flow reversal line. In order to discuss this
m+1, n, 0 + calculation, we will use Fig. 5a where a typical section of the
U flow reversal line is shown among grid points. If the
m,n,0+1
finite-difference formulas from Eq. (12) are used to calculate
Fig. 2 Grid used to
obtain difference equa-
the flow reversal line, then all of the grid points shown in
tions. Fig. 5a would have to be used to calculate the point m + 3,
n, o + 4. Since this is clearly impossible because of the
numerical instability in the reversed flow region, the typical
point m + 1, n, o — 1 was dropped in the finite-difference
scheme and the following finite-difference quotients were
used:
= Um
(15)

0.2, and it was found by numerical experimentation that one

ZW.-H.. - Um+l,n-I,o , U ~~ Um,n—l,o
Downloaded by RYERSON UNIVERSITY on February 3, 2015 | http://arc.aiaa.org | DOI: 10.2514/3.4745

more step in the x direction could be taken before the flow

reversal numerical instability occurred. When the x step etc. With this finite-difference method, only the net points
size was reduced to 0.305 cm it was found that the velocities in the inmediate vicinity of the flow reversal line have to be
before the point of flow reversal and the location of the flow used to continue on the calculation. [If the flow reversal
reversal point changed by less than l|%. At the point line had a negative slope then the formulas in Eq. (12) could
of flow reversal and for 77 < 1.0, it was difficult to obtain good be used without change. This occurs for cones at angle of
convergence since the velocities were very small and sensitive attack.] When the finite difference quotients in Eq. (15)
to the location of the point of flow reversal. However, it were used to calculate the velocities in front of flow reversal,
was encouraging to find that the location of the point of the results agreed within 2% of the results obtained from
flow reversal seemed to be converged. Another interesting Eq. (12).
result was obtained by neglecting Eq. (11) and dw/dz along The flow reversal line in front of the cylinder is shown in
the line of symmetry. This caused the flow reversal point Fig. 5b where the finite-difference quotients in Eq. (15) have
to move to x = 23.8 cm. been used. This line represents the point where the CT-com-
It can be seen in Fig. 3 that the boundary layer seemed to ponent of velocity first goes negative in the boundary layer.
converge asymptotically at large rj even though flow reversal The "bump" in the flow reversal line is due to the fact that
had occurred. This asymptotic convergence at large 77 does d2pi/dz2 has a very sharp minimum on the line of symmetry
not necessarily mean that the boundary-layer equations are for the inviscid flow given [remember that d2pi/d22 is used
valid around the point of low reversal as can be seen in Fig. explicitly in Eq. (11)]. This behavior of the boundary-
4. Figure 4 shows the values of the 7-velocity as a function layer equations about a line of symmetry occurs in a region
of x for n = 0.2 and z = 0. Near the leading edge of the plate, of almost zero lateral inviscid streamline curvature, and seems
the behavior is very similar to that of the Blasius flat-plate to bring up serious doubts about the validity of the small
solution; however, as the point of flow reversal is approached, crossflow equations. The bump in the flow reversal line
V becomes relatively large. The rapid growing of V as the is caused by the crossflow terms in the continuity equation.
flow reversal point is approached might indicate that the In Fig. 6 is shown the derivatives dt//dx and dTF/dz, which
boundary-layer approximation is becoming invalid, but since enter into the continuity equation, at constant x and 77 = 0.2
V is still an order of magnitude smaller in absolute value than as a function of z. Figure 6 shows that the crossflow term in
most values of £7, a definite conclusion cannot be reached the continuity equation at the bottom of the boundary layer
is first order even though the inviscid lateral curvature is very
small. The small crossflow approximation assumes that
the momentum and continuity equations can be expanded
into an asymptotic sequence of equations based on an ex-

762 1525 2290 3050

U-(x-VELOCITY) CM/SEC (100 FT/SEC)

Fig. 3 [/-velocity profiles along the line of symmetry. Fig. 4 V-velocity development along the line of symmetry.
 JULY 1968 THREE-DIMENSIONAL BOUNDARY LAYER WITH SEPARATION 1341

pansion parameter proportional to the lateral inviscid

streamline curvature. The results of Fig. 6 indicate that
the small crossflow approximation can be a poor one for the »l s <30.5 U,>2740
continuity equation. This point is further illustrated by look-
ing at the inviscid flow. The continuity equation for the
inviscid flow is
(dtfi/ds) + (bTFi/ds) = 0
which implies that crossflow derivatives are equal to the
negative of the mainflow derivatives.
Figure 5b contains some very interesting crossflow phe-
nomenon. The line shown crossing the flow reversal line
at z = 9.05 cm represents the point where dpi/dx changes
0 0.61 1.22 1.83 2.44 3.05 3.66 4.26 4.88
from an adverse pressure gradient to a favorable one. The
amazing result is that the ^/-component of velocity is still
going through a flow reversal while bpi/dx is negative. This Fig. 6 Velocity derivatives near the line of symmetry.
result was at first very puzzling, but a careful analysis of the
flow calculations revealed the following cause. The flow
reversal of the [/-component of velocity was driven by the and that the flow reversal is confined to a smaller region near
convective term W&Ufbz in the boundary-layer equations, the wall. This result agrees with our previous remarks con-
Downloaded by RYERSON UNIVERSITY on February 3, 2015 | http://arc.aiaa.org | DOI: 10.2514/3.4745

which was positive and overcame the dpi/dx term. The cerning the influence of crossflow on the flow reversal away
WdU/dz was very large because dpi/dz was much larger from the line of symmetry. The W profiles are seen to de-
than dpi/cXr and U also changed very rapidly around the velop from a Blasius flat-plate shape at the leading edge and
point of flow reversal. Therefore, this problem presents an then to become more like those associated with crossflow.
example of flow reversal caused by crossflow effects near the At the point of flow reversal, the crossflow is increasing
bottom of the boundary layer. rapidly and has values of velocity much larger than the U-
Shown in Fig. 7 are the U and W velocity profiles as func- component of velocity.
tions of x and 77 for z — 6.1 cm, the radius of the cylinder.
The [/-velocity profiles are seen to develop in approximately
the same manner as along the line of symmetry. The one
big difference is that the velocity profiles are more favorable,

• -NET POINTS USED AND CALCULATED

X -NET POINTS NOT USED

U-FLOW REVERSAL LINE

762 1525 2290 3050

U ( x - VELOCITY) OH/SEC

a) {/-velocity
a) With grid points

X-CM

b) Without grid
points

0 152 305 457

~- Z - C M K ( Z - V E L O C I T Y ) CM/SEC Z = 6.1 CM
0 3.05 6.1 9.14 12.2 b) JF-velocity
Fig. 5 Flow reversal line. Fig. 7 Profiles away from the line of symmetry.
 1342 H. A. DWYER AIAA JOURNAL

1) A fast, stable, and convergent numerical method of

solving the three-dimensional boundary-layer equations has
been developed and used.
2) The numerical method is basically "implicit." How-
. 229 -
ever, since there are three independent variables, the method
is different from problems with two independent variables.
Only along the boundary-layer coordinate are the derivatives
evaluated implicitly, whereas a directed marching scheme
is used for the derivatives in the plane of the body surface.
Downloaded by RYERSON UNIVERSITY on February 3, 2015 | http://arc.aiaa.org | DOI: 10.2514/3.4745

3) Initial conditions for the three-dimensional boundary-

layer equations have been defined and derived. These
conditions matched up smoothly with the results of the
finite-difference calculations.
4) The line of symmetry equations were found to be sub-
stantially affected by three-dimensional influences through
15.2 22.9 30.5 the continuity equation. These results may restrict the
validity of some applications of the small crossflow approxi-
mation.
Fig. 8 Velocity development approaching flow reversal.
5) The method of solution has shown itself capable of pre-
dicting the point of flow reversal, but a numerical instability
The development of the U flow reversal along the line of associated with negative velocities (time reversal) occurs
symmetry and at a z station where dpi/dz is favorable is after flow reversal.
shown in Fig. 8. The {/-component of velocity is shown at 6) Examples of flow reversal in the three-dimensional
17 = 0.2 and z = 0 and 9.76 cm as a function of x. At the z = boundary-layer have been found that are due to adverse
9.76 cm station, the velocity remains relatively high until the pressure gradients and crossflow. The flow reversal, caused
region of large dpi/dz is reached; then a very sharp flow by crossflow, was due to the convective term WdU/dz, and
reversal is caused by the term WdU/dz as mentioned previ- was indirectly largely affected by the crossflow pressure
ously. The flow reversal along the line of symmetry is seen gradient dp/dz.
to be more gradual than away from the line of symmetry.
References
VII. Computer Time 1
Moore, F. K., "Three-Dimensional Boundary Layer Theory,"
The total computer tune to calculate the preceding three- Advances in Applied Mechanics, edited by H. L. Dryden and T.
dimensional flow problem was 5 min on a General Electric von Karman, Vol. IV, Academic, New York, 1956, pp. 159-228.
2
635 computer. The 5 min calculation involved 72,000 net Sowerby, L., "The Three-Dimensional Laminar Boundary
points where the following step sizes and number of steps Layer on a Flat Plate," Journal of Fluid Mechanics, Vol. 22, Pt.
were taken: A# = 0.61 cm, Nx = 60; Az = 0.61 cm, Nv = 3, pp. 587-598.
3
Hayes, W. D., "The Three-Dimensional Boundary Layer,"
40; A,, = 0.2, Nr, = 30; total number of net points, 60 X Navord Kept. 1313 (NOTS 384), May 1951.
40.X 30 = 72,000. 4
Fannelop, T. K., "A Method of Solving the Three-Dimen-
Therefore, it may be possible that the method of solution sional Laminar Boundary Layer Equations with Application to
is suitable for compressible flow, even with energy and mass a Lifting Re-Entry Body," Paper 67-159, Jan. 1967, AIAA.
5
transfer, without the computer time becoming excessive. Blottner, F. G., "Chemical NonequHiDrium Boundary Layer,"
AIAA Journal, Vol. 2, No. 2, Feb. 1964, pp. 232-240.
6
Der, J., Jr. and Raetz, G. S., "Solution of General Three-
VIII. Conclusions Dimensional Laminar Boundary Layer Problems by an Exact
A method of calculating accurate solutions of the three- Numerical Method," Paper 62-70, Jan. 1962, Institute of Aero-
dimensional laminar boundary-layer equations has been given. space Science.
7
Chang, A. L. and Cheng, H. K., "A Study of Finite-Differ-
The method was applied to a problem that exhibited many ence Methods as Applied to Hypersonic Viscous Shock-Layer
interesting crossflow phenomena, such as flow reversal Equations," Rept. AF-1285-A-9, March 1963, Cornell Aero-
caused by crossflow terms in the momentum equation and nautical Lab.
large crossflow effects caused by the continuity equation. 8
Richtmyer, R. D., Difference Methods for Initial-Value
The main results and conclusions of this study are: Problems, Interscience, New York, 1957.