NASA Technical Memorandum 88242

NASA-TM-8824219860018578

G~eometry Definition and Grid

Generation for a Complete Fighter
Aircraft
Thomas A. Edwards

April 1986
Ani)
r ~\
n
(, {'I'. 1i\Pt';
1:::1,),

L',NGLEY f\E~)EAf\CH CENTER

LIBRARY, NASA
{if'.MPTON, VIRGINIA

111111111111111111111111111111111111111111111
NF00951

1:'oH R.E.F'EREl\JCr'

NI\S/\
National Aeronautics and
"'01 'to HI': 1'AV,Rff f'lh/M THIll nOOlli

Space Administration
NASA Technical Memorandum 88242

Geometry Definition and Grid

Generation for a Complete Fighter
Aiircraft
Thomas A. Edwards, Ames Research Center, Moffett Field, California

April 1986

NI\S/\
National Aeronautics and
Space Administration
Ames Research Center
Moffett Field, California 94035
 GEOMETRY DEFINITION AND GRID GENERATION
FOR A COMPLETE FIGHTER AIRCRAFT

Thomas A. Edwards
NASA Arnes Research Center
Moffett Field, California 94035
USA

SUMMARY

RIlcent advances in computing power and numerical solution procedures have enabled computational
fluid dynamicists to attempt increasingly difficult problems. In particular. efforts are focusing
on computations of complex three-dimensional flow fields about realistic aerodynamic bodies. To
perform such computations, a very accurate and detailed description of the surface geometry must
be provided. and a three-dimensional grid must be generated in the space around the body. The
geornetry must be supplied in a format compatible with the grid generation requirements, and must be
verified to be free of inconsistencies. This paper presents a procedure for performing the geometry
definition of a fighter aircraft that makes use of a commercial computer-aided design/computer-aided
manufacturing system. Furthermore, visual representations of the geometry are generated using a
computer graphics system for verification of the body definition. Finally, the three-dimensional
grids for fighter-like aircraft are generated by means of an efficient new parabolic grid generation
method. This method exhibits good control of grid quality, and generates grids about 50 times faster
than comparable grids generated via elliptic algorithms.

1. INTRODUCTION

Many applications of computational fluid dynamics (CFD) involve solving the equations governing
fluid flow using a finite-difference or finite-volume approximation. The solution procedure commonly
involves inverting large. banded matrices in an iterative mode until a steady-state solution is
obtained. Depending on the form of the governing equations being solved, the matrices can be
scalar tridiagonal (e.g., potential formulations) or block tridiagonal (Euler and Navier-Stokes
formulations), where the blocks are commonly 4x4 for two-dimensional problems and 5x5 for
three-dimensional problems. To obtain the numerical solution to the flow field, the domain of
interest must first be discretized with a grid. Typically, a body-fitted grid is used where the grid
boundaries conform to the body surfaces. The grid can be generated by an algebraic interpolation
procedure (Ref. 1) or by the solution of partial differential equations (elliptic (Ref. 2,3);
parabolic (Ref. 4,5); or hyperbolic (Ref. 6,7)). In any case, some or all of the grid boundaries
must be prescribed prior to generating the interior grid points. Thus, the body surface must be
defined and a grid of surface points supplied as input to the three-dimensional grid generation
procedure.

The numerical solution to the flow field is obtained at each grid node (or at the center of the
grid cells for finite-volume formulations), and the accuracy of the solution depends partly on the
number and location of the nodes. The Euler and lIavier-Stokes equations account for more physical
phenomena, but they require a finer computational mesh. Similarly, as the body geometry increases in
complexity, more surface grid points are required to maintain an accurate definition of the surface
Shape. The number of points needed in the computational mesh to solve a given problem, then, is
dependent on the level of detail in both the geometry and the flow physics. However, the computer's
memory capacity limits the maximum number of points that can be used for a flow calculation. Until
recently, computer resources have limited researchers to problems representing a compromise between
physical complexity and geometrical complexity: problems involved either complex geometry at the
expense of simple flow physics, or complex flow physics at the expense of simple geometry (Fig. 1).

However, the newest and most powerful supercomputers available have for the first time presented
the CFD researcher with the tools to attempt problems incorporating both complex flow physics and
complex geometry. This, in turn, has created a demand for a new level of accuracy in defining the
body geometry. To obtain accurate solutions to viscous flows over complex bodies, the grid must be
refi.ned, clustered. and stretched to make the most efficient use of the grid points. When this is
done, the distribution of points on the surface of the body must be adjusted accordingly, requiring
interpolation of the geometry data base. To minimize errors inherent in the interpolation procedure.
the data base should be interpolated as accurately and as few times as possible. The problem of
geometry definition includes the modeling of surfaces, along with the process of redistributing
points on these surfaces. The resulting surface definition must satisfy the requirements of the
three-dimensional grid generator, while maintaining an accurate representation of the actual body
surface.

Another requirement in performing geometry definition is error-checking to verify that the

geometry is being represented accurately and in a manner consistent with the data base. For most
applications. it is sufficient and expedient to verify visually the surface definition by displaying
the surface graphically. Toward this end. both shaded-surface and wire-frame displays have proven
useful (Ref. 8).
 Once the surface definition has prescribed the points on the aircraft surface. a three-dimensional
grid must be generated to discretize the space around the aircraft. Grid generation methods
frequently involve solving partial differential equations (PDEs) in the coordinate variables.
since CFD researchers already have a great deal of experience in the numerical solution of PDEs.
Each class of PDE offers certain advantages and disadvantages with regard to the problem of grid
generation. Thompson (Ref. 2) has shown that elliptic equations have desirable smoothing properties
and permit boundary conditions to be imposed on all boundaries. However. they require iterative
solution techniques. which are computationally expensive. Steger and Chausee (Ref. 6) subsequently
demonstrated that hyperbolic equations can be used to generate body-fitted grids as well. The
solution procedure is noniterative. which is computationally efficient. but boundary discontinuities
propagate into the interior field. and outer boundary conditions cannot be prescribed. The present
method solves a parabolic equation set. which combines the attributes of the two former methods.
The algorithm has a marching (noniterative) solution procedure and smoothing properties. and permits
specification of the outer boundary grid. The method was first introduced by Nakamura (Ref. 4). and
later was extended by Edwards (Ref. 5).

This paper presents a method for performing geometry definition of fighter-like geometries of
interest to current CFD efforts. Throughout the description. the F-16 aircraft will be used to
demonstrate the effectiveness of this procedure. since this aircraft is the test case for an ongoing
effort to compute transonic. viscous flows over realistic aircraft (Ref. 9). Also. the parabolic
grid generation technique will be described in detail. Sample grids have been generated for the
F-16 geometry. and the computational efficiency of the parabolic algorithm is compared to that of an
elliptic algorithm.

2. GEOMETRY DEFINITION

2.1 CAD/CAM System

In the present study. a computer-aided design/computer-aided manufacturing (CAD/CAM) system

is used to define the geometry of a modern fighter jet with minor idealizations. This system
uses a VAX 11/780 as a host computer. and consists of an alphanumeric display screen and a
graphics display screen (see Fig. 2). The system is suitable for multiple applications ---
for example. printed-circuit design. mechanical drawing/drafting. finite-element modeling. and
numerically-controlled tool-path computation. Here. the CAD/CAM system performs the surface
constructions and geometrical computations of interest to CFD applications. Among the benefits in
this approach are that it saves time and effort by using existing. proven technology (Ref. 10) as
opposed to developing new software for this specialized application. In addition to being able to
perform many geometrical constructions. the system can compute surface intersections. parameterize
surfaces. and sample surfaces in several different modes for new point distributions. The system
also features a versatile programming language to execute repetitive or computationally laborious
tasks.

2.2 Surface Fitting

The input-geometry data base for the example used here consists of (x,y,z) coordinates of points on
the cross-sections of various parts of the F-16 aircraft. Figure 3 presents a shaded-surface image
of the input geometry. Since the fighter has a very complicated surface. including an inlet. ventral
fins. and speed brakes. it was subdivided into 50 simpler components before the cross-sections were
taken for the data base. Hence. the input geometry is essentially a file of Cartesian coordinates
of cross-sectional points on many small pieces that. when patched together. constitute the exterior
surface of the fighter.

The first step in the geometry definition procedure is to fit each subcomponent of the body with
B-surfaces. In the syntax of Calma. the term "B-surface" refers to an analytic product surface
of one-dimensional B-splines along the parametric directions. This type of surface is somewhat
less general than a parametric bicubic (Coons) patch. because it neglects the cross-derivative or
"twist" vector at the nodes (Ref. 11). The B-surface is a sufficiently general description for the
small. regularly shaped components used in this study. Where appropriate. other surfaces can be
generated. such as planar. ruled. Coons. or Bezier surfaces. or surfaces of revolution. Various
surface trimming and editing operations are also available. enabling modification of the geometry to
suit the constraints of the particular application.

Frequently. it is necessary to make modifications to the actual geometry to reduce the scope of the
problem or to focus only on its essential elements. For the F-16 aircraft. initial implementation of
the flow solution scheme required that the inlet be faired over and modeled as a streamlined solid
surface. This idealization substantially simplifies the grid generation and boundary condition
procedures. To do this. the forward section of the inlet was trimmed away using utility functions
available on the Calma. A new surface was then defined by generating splines at the boundaries
of the region to be faired over. The splines were defined so as to be tangent at both ends to the
actual body geometry. Figure 4 shows the new surface definition for the faired-over inlet. Hote
also that the ventral fins have been removed and that a fillet has been constructed in the nozzle
region where the actual geometry has a gap.

2
 2.3 Surface Sampling

Once the input points are fit with B-surfaces. the three-dimensional grid topology must be
determined in order to define the type of surface grid needed on the body. The flow solution scheme
fOI' the presEmt application uses a cylindrical (O-H) topology. shown in Fig. 5. The fuselage maps
into one sidE! of the computational cube. and the wing and horizontal tail correspond to an adjacent
side. This means that the surface grid on the fuselage is independent of the grid on the wing and
tail surface!l. except that the fuselage cross section points must match up with the wing and tail
coordinates where these surfaces intersect. For many transport-type aircraft. a clear demarcation
exists between the wing and fuselage. but the fighter in this example has a blended wing-body. which
leaves analysts free to define the location of the wing root themselves. This arbitrariness is
used to advantage in the present problem since. as seen in Fig. 6. there is a sharp corner in the
fuselage croEls-section aft of the wing. If the wing root is defined as a straight line parallel
to the centerline. the sharp corner in the body cross-section is included as part of the fuselage.
introducing t;he three singular points in the grid shown in Fig. 7a. If. instead. the wing root
is defined so as to follow the corner of this shelf. two of the singularities are eliminated. as
shown in Fig. 7b. Since topological peculiarities such as this are not easily generalized. they are
currently dealt with on a case-by-case basis.

After the analytic surface patches are generated and the topology for the surface grids are
determined. the surfaces must be sampled for a distribution of points to supply as a boundary
condition for the three-dimensional grid generator. For the current application. axis-normal body
cross sections were desired at prescribed body stations. A code was written in DAL. the Calma
programming language. which computes the intersections of a plane of constant x with all of the
surfaces on the fuselage. A B-spline is constructed at the intersection of the plane and each
surface that it intersects. For example. if the fuselage is represented by 30 surfaces. any given
plane of constant x may cut. say. 10 of them. so the result of the intersection calculation is 10
B-splines. However. to construct one continuous cross section from these 10 splines. they must
be linked together in the proper order and fit as a single spline. This is done by searching from
the endpoint of one spline for the nearest spline endpoint (disregarding any splines already linked
together) until all 10 splines are ordered end to end. Once the splines are arranged end to end.
a single spline is fit through points on the 10 individual splines. These points are obtained by
recovering the original control points used to construct the 10 individual splines. An alternative
method would be to generate new points by a parametric sampling of the 10 splines. but this would
introduce further errors as a result of the additional interpolation. A parametric curve sampling is
then calculated on the cross-sectional splinE! to obtain the fuselage surface grid at that particular
body station. The increment of the curve parameter can be adjusted to achieve equally spaced points
around the body. or clustering of points near areas of detailed geometry or anticipated large flow
gradients. The point coordinates are written into a disk file for subsequent input to the grid
generation routine.

Use of the Calma CAD/CAM system in this segment of the geometry definition process has proven
useful in representing the complex three-dimensional surfaces associated with flight vehicles.
The surface-editing functions are capable of modifying the surface geometry to suit the particular
application. and the programming language (DAL) permits nearly the entire process to be automated and
made transparent to the user. As stated earlier. the most significant benefit of using the CAD/CAM
sys·tem for this application is that the technology is immediately available; development time and
duplication of effort are circumvented.

3. GEOMETRY VERIFICATION

3.1 Graphics Workstation

Before passing the surface grid to the grid generator and flow solution code. it should be checked
for smoothness and continuity. For the present study. two graphical techniques. described below.
were employed to verify the geometry. Both exhibit unique advantages and disadvantages with regard
to geometry verification. The Silicon Graphics Iris 1500 Workstation was used to display the
graphical images. This system features a special hardware transformation system. enabling real-time
image manipulation on the display screen. and color display capability. As will be explained below.
these tools greatly accelerate the geometry verification step. .

3.2 Shaded-surface images

A shaded-surface image is essentially a photograph of the geometrical information used to define

the surface. The image is faceted because the location of the surface is defined only at discrete
points in space. The shaded-surface display is effective for revealing holes in the surface. and for
indicating discontinuities in the surface slope. Degenerate grid cells (cells with fewer than four
sides) are difficult to detect using this approach. however.

To produce the shaded-surface image of the surface grid. nonplanar. four-sided polygons are
defined on the aircraft surface by connecting the surface grid points. These polygons are then
sorted according to the distance from their centroids to a prescribed light source that illuminates
thern. The orientation of the polygon with respect to the light source determines the shading of the
polygon. so that polygons facing the light source are shaded white. and those aligned with the light
source are black (Fig. 8). By coloring the polygons in the order of the decreasing distance to the
light source. a hidden-surface representati.on is effected. The shaded polygons are loaded into the

3
display memory of the Iris workstation, and by virtue of the real-time rotation and translation, a
quick visual check of the surface representation is enabled. The shading of the polygons is fixed
before they are loaded into display memory since the machine is currently not capable of dynamic
reshading. Examples of the display are shown in Figs. 3, 4, and 6.

3.3 Wire-frame images

Another useful display technique is the wire-frame image. In this approach, the points defining
the surface shape are connected with straight lines, yielding an image such as that shown in Fig. 9.
Since the individual grid cells are outlined on the surface, degenerate cells are easily detected.
This type of display is also useful to inspect topological boundaries in the surface grid. A
disadvantage of the wire-frame image is that, as the mesh is refined, the image becomes progressively
more difficult to visualize. This problem is somewhat alleviated through the use of color, and a
significant advancement could be made with hidden-line removal. But the most important element
in both display techniques, shaded-surface and wire-frame, is the ability to perform real-time
manipulations of the image, and as yet, hidden-line removal cannot be done in real time.

3.4 General remarks

The process of defining the geometry is complete when the surface has been sampled for points in
a format compatible with the requirements of the grid generator, and the resulting grid surface
has been verified to be free of inconsistencies such as holes, degenerate cells, and inaccurate
representation of the geometry. It was mentioned earlier that it is desirable to minimize the
number of times the data are interpolated to obtain the final surface grid. By following the
foregoing procedure for defining the fighter, the data have been interpolated only twice: once
in generating the surfaces from the input data, and a second time in constructing the splines at
the surface intersections. Any loss in accuracy using this procedure is compensated for by its
generality. Use of the same procedure permits any type of cutting surface to be used to compute
surface intersections. This flexibility can be exercised to compute nonaxis-normal sections, and
even nonplanar cuts through the body. Moreover, the procedure applies to nearly any continuous
surface that can be represented by a set of surface patches. For surfaces with significant twist,
the same definition procedure can be applied, making use of the more general Bezier surfaces rather
than B-surfaces. A flow chart describing the entire geometry definition and verification process is
shown in Fig. 10.

4. GRID GENERATION

The surface grid obtained by the foregoing procedure is acceptable as input to a variety of
three-dimensional grid generation methods. Since many existing methods are difficult to use or
require a great deal of computing power, analysts are always looking for ways to simplify and
expedite the generation of the grid. The method described below implements a new differencing scheme
which permits a marching (noniterative) solution procedure. As will be shown, this method results
in a substantial savings in both the computer time required to generate the grid and the core memory
necessary during the calculation. The algorithm has been used to generate grids about isolated wings
for viscous transonic flow computations (Ref. 9) and also for the F-16 geometry described above.

4.1 Equations

The proposed grid generation algorithm is derived from a set of three-dimensional elliptic
PDEs. Let (x,y,z) be the coordinates in the physical space and (e,ry,l) be the coordinates in the
computational space (Fig. 11). Then the mapping between the two spaces is given by Laplace's
equation:

exx + eyy + e•• = 0

'Ixx + 'Iyy + 'I •• = 0 (1)
Ixx + Iyy + I.z = 0
These equations are inverted to make (e,ry'l) the independent variables, giving

(2)
where r = x,y, or z, and

Al =(x~ + y~ + z;)(x; + Y; + z;) - (x~x, + Y~Y, + Z,/Z,)2

ill =(x~x, + Y~Y, + z~z()(x,xe + Y(Ye + z(ze)- (3)
(xex~ + YeY~ + zez~)(x; + Y; + z;J
A 2 , A3 , Ih, and 13 3 are found by rotating C '7, and I in Al and B I •

4
 4.2 Differencing Scheme

To form the finite-difference equations from the PDEs, elliptic grid generation algorithms employ
central differences to approximate all derivatives:

(4)

The resulting equations are then solved iteratively using relaxation schemes such as successive
overrelaxation (SOR) or alternating-direction implicit (ADI) until some convergence criterion is
satisfied. In the present study, a "parabolic" differencing scheme is used to approximate the
derivatives in q and~. In this scheme, the derivative of a quantity at a point corresponding to
index n is approximated by a difference between its value at index (n .- 1) and its value at index
nmax. These quanti ties are both known at the first marching step (n = 2) from boundary data,
permitting a marching solution procedure. For example. the finite-difference approximation to
equation (2) above for x at an interior point in the grid with indices i,j, k becomes

Al (Xi+I,j,k - Xi,j,kl{li - (Xi,j,k - Xi-I,1,k)/ 1.-1 +

2(fi + Ii-I!
A (Xi,jmax,k - Xi,j,k)/G) - (xi,1,k - Xi,j-I,k)/g}-I +
2 !(G) + gj-d
A (Xi,1,kmax - xi,1,k)/lh - (xi,1,k - Xi,J,k-l!/hk-1 +
3 ~(1h + hk-d
Xi+l,}max,k - Xi-l,}max,k - Xitl,}-I,k + Xi-I,j-I,k
213I ' +
(fi + li-d(G) + gj-d
2B2 Xi,jmax,kmax -.Xi,j-l,kmax - xi,imax,k-l + Xi,j-l,k-l +
(G j + gj-d(Ih + hk-d
213 Xi'H,1,kmax _. Xi-I,},kmax - Xi+I,j,k-1 + Xi-I,j,k-I = 0
(5)
3
(Ii + li-d(Hk + hk-d
where

Ii = ei+1 - ei li-I = ei - ei-l

gj-I = qj - qj-I Gj = g) + gi+1 + '" + g}max-I (6)
hk ... l = Ik - ~k-l Ilk = hk + h k + 1 + '" + hkmax-l
The unknown quantities in this equation are the Xi-I,1,k, Xi,j,k, and Xi+I,},k. All other quantities are
known from the boundary grids or previously solved planes. The coefficients AI, A2 , dc., are lagged
one ~-step to linearize the equations. By approximating derivatives in '1 and ~ with the parabolic
differencing scheme, while still using central differencing for derivatives in e, the algorithm can
march in both the ~- and ~- directions. A scalar tridiagonal system is obtained for the equations in
the e-direction. Thus, the formulation is equivalent to the numerical solution of a parabolic POE
with source terms added such that the outer boundary points are obtained exactly when the marching
solution reaches the jmax or kmax index.

4.3 Boundary Gondi tions

Boundary data are required on all six sides of the computational cube. For the purposes of this
study, a two-dimensional, parabolic grid generation algorithm generated the outer boundary grids.
Body surface grids are supplied from the geometry definition routine described above.

4.4 Clusterinl~

Often, a more accurate flow solution can be obtained by clustering the grid points near regions
of antiCipated large gradients in the flow variables. To control the spacing between grid lines
of the same family, an intermediate computational space is used as described in Ref. 4. The result
of clustering grid lines in this space is to make a corresponding displacement of grid lines in the
physical space. This is implemented by def ining functions I, g, and h that describe the nonuniform
values of (t.O, (t. I I) , and (t.O , respectively. For a uniformly spaced computational grid, these
functions would be unity at every point. To cluster or stretch the grid, the value of the function
at a particular point would be the distance from that point to the next point along the particular
coordinate direction, as shown in Fig. 12. To generate values for the functions j, g, and h.
any convenient stretching formula can be used. such as exponential. cosine. or geometric. These
functions are used in forming the finite difference equations so that the differencing schemes become

a() ()i+1 - ( )i-1

aT ""'-Ii + li--I-
a() ()1max _. ( )j-1
(7)
a;J"'" Gj+g j _ 1

a() (hmax - ( )k-I

a~- "'" Ilk + hk-I

5
 The computational grid used for the flow calculation is uniformly spaced, and the transformation of
variables is contained entirely in the grid metrics.

4.5 Angle Control

Accuracy and stability of the flow calculation are also enhanced by requiring that the grid cells
be nearly orthogonal, although for most applications, the only region in which angle control is
desired is near the inner boundaries, where the solid body surfaces lie. The angular orientation
of grid lines is controlled using a fictitious outer boundary grid to determine the source terms.
Since the parabolic algorithm causes the grid lines to march toward the outer boundary grid, the
orientation of these grid lines can be controlled by temporarily shifting the outer boundary grid.
If the grid is shifted in such a way that the new outer boundary points lie on lines perpendicular
to the inner boundary surfaces, the grid lines will march away from the inner surfaces orthogonally,
as Fig. 13 shows. In the present study, the shifted outer grid is computed by first calculating the
slope of the body surface and the distance from inner boundary points to corresponding outer boundary
points. If this distance is designated by Ri,i,k, then the shifted outer boundary points are given by

(8)

where

azl
ln x ::::: - - (9)
ax aur/ace

The shifted outer boundary is scaled so that it lies at the same distance from the inner boundary
as the actual outer boundary. This scaling is needed to preserve the desired clustering or
stretching of grid lines. As the solution marches toward the outer boundary, the temporary grid
is gradually shifted back to the desired outer boundary point distribution. In this manner, angle
control is maintained at the body surfaces, and a smooth transition is made to the outer boundary.

5. RESULTS

The parabolic algorithm has been coded to generate O-H grids about wing-fuselage geometries. Using
the surface grid generated for the F-16 on the CAD/CAM system, coarse grids have been generated for
use in zonal flow solution applications. A grid-embedding approach is used, wherein the coarse
outer grid is generated using the parabolic or elliptic scheme and the inner grids are interpolated
from the outer grid (Ref. 9). This allows different equation sets to be solved in different grid
zones, as appropriate. The Reynolds-averaged, thin-layer Navier-Stokes equations are solved in the
zones that have solid surface boundaries, and the Euler equations are solved in the outer zones.
Where warranted, it is possible to further refine the grid with additional zones. Since the grid
refinement necessary to resolve viscous flows is achieved through grid-embedding, the principal grid
is rather coarse and is only slightly clustered near the solid surfaces. Figure 5 shows the O-H
topology used. The grid dimensions are 65 (axial) by 20 (circumferential) by 20 (radial). Other
detail views are shown in Figs. 14-16. Note also the control of angle and spacing that has been
enforced at the body.

6. COMPUTATIONAL EFFICIENCY

An important attribute of the parabolic algorithm is its computational efficiency. Since the
solution procedure is noniterative, only one sweep through the domain is required to generate the
grid. Also, to generate points on any coordinate plane of constant k, information is needed on
the edges of this plane, on the (k - 1) plane, and on the krnax plane. Hence, previously computed
coordinate planes can be saved permanently, resulting in a substantial economy in core memory
requirements. The CPU time used per grid point to generate the grid is roughly independent of the
size or refinement of the desired grid; about 62 Msec per grid point on the NASA Ames Cray X-MP22,
or 2.44 sec for the 39,OOO-point grid shown in Figs. 14-16. This is about 50 times faster than a
comparable elliptic grid generator that uses SOR on the coarse grids used for these examples. The
difference is expected to be even greater as the grid is refined. since the iteration count for
elliptic algorithms increases with mesh refinement. while the parabolic algorithm experiences little
change in CPU use. Because of the small amount of CPU time required to generate a grid using this
scheme. the method is appropriate for use in the interactive development of a grid.

6
 7. CONCLUSIONS

A proCedUrE! has been described to define and verify complex aircraft surfaces for use in CFD
calculations. By using current technology in CAD/CAM and graphics display systems. the procedure
has been made very general and reliable. It has been found that by partitioning the surface into
small. regularly-shaped surfaces and using B--surfaces to fit the points. surface sampling can be
applied to continuous bodies of arbitrary shape. Also. the geometry verification has been greatly
improved by using graphics display systems capable of real-time manipulations of the image. Both
the shaded-surface and wire-frame images have proven useful in verifying the accuracy of the surface
definition. A parabolic differencing scheme has been implemented to generate three-dimensional
grids about fighter-like geometries. and has been shown to control grid spacing and angles in a
straightforward manner. The method is also very efficient in terms of computer memory and CPU time.

8. References

1. Smith. Robert E .• "Algebraic Grid Generation." Numerical Grid Generation. Joe Thompson (ed. ).
ElsE!vier Science Publishing Co. • Inc. 1982. pp. 137-170.

2. Thompson. J. • "Elliptic Grid Generation." Numerical Grid Generation. Joe Thompson (ed. ). Elsevier
Science Publishing Co .• Inc. 1982. pp. 79-106.

3. Sorenson. R. L.• "Three-Dimensional Elliptic Grid Generation About Fighter Aircraft for Zonal
Finite-Difference Computations." AIAA paper 86-0429. Reno. Jan. 1986.

4. Nakamura. S .• "Marching Grid Generation Using Parabolic Partial Differential Equations."

Numerical Grid Generation. Joe Thompson (ed. ). Elsevier Science Publishing Co. • Inc. 1982. pp.
776 .. 786.

6. Edwards. T. A.• "Noniterative Three-Dimensional Grid Generation Using Parabolic Partial

Differential Equations." AIAA paper 86-0485. Reno. Jan. 1986.

6. Steger. J. L. and Chaussee. D. S . . "Generation of Body Fitted Coordinates Using Hyperbolic

Part;ial Diffe:rential Equations." SIAM J. Sci. Stat. Compu! .• Vol. 1. No.4. Dec. 1980.

7. Steger. J. L. and Rizk. Y. M.• "Generation of Three-Dimensional Body-Fitted Coordinates Using

Hyperbolic Pa:rtial Differential Equations." NASA TM-86763. June 1986.

8. Cozzolongo. J. V.• "Aircraft Geometry Verification with Enhanced Computer-Generated Displays."

NASA TM-84264. June 1982.

9. Holst. T. L. . Gundy. K. L.• Flores. J . . Chaderjian. N. M. . Kaynak. U. . and Thomas. S. D.•

"Numerical Solution of Transonic Wing Flows Using an Euler/Havier-Stokes Zonal Approach." AIAA paper
86-1.640. Cincinnati. July 1986.

10. Enomoto. F. . and Keller. P. • "Using a Commercial CAD System for Simultaneous Input to
Theoretical A,~rodynamic Programs and Wind-Tunnel Model Construction." NASA CP-2272. April 1983.
11. Craidon, C. G. . "A Computer Program for Fitting Smooth Surfaces to an Aircraft Configuration
and other Three-Dimensional Geometries." NASA TM X-3206. 1976.

Figure 1. Current computing power enables solution of complex flow physics over detailed.
three-dimensional surfaces.
 300 i
I

200 I

100 i

z O!

-100
I

-200

Figure 2. Calma CAD/CAM workstation used

for geometrical calculations. o
o 2~~ 100
Y

Figure 5. O-H grid topology.

Figure 3. Shaded-surface image of input

geometry data base.

Figure 6. Detail of wing-fuselage junction.

Figure 4. Shaded-surface image of modified

geometry with faired-over inlet.

8
 • STRAIGHT WING ROOT
3 GRID SINGULARITIES

• CURVED WING ROOT

1 GRID SINGULARITY

Figure 7. Wing-root definitions and

resulting grid singularities.

,,-<f 100
/
/ 75
/
/
/

/
/
/
/
/
/
/
/

/
/
z
/~ fI
\/
___L------- Q -15
-100 -
400
350 300 _ ~

250 ;oo-~~o~_
X 100~~'~
50 50
o
o Y

Figure 8. Polygon shading technique. Figure 9. Wire-frame representation of

surface grid.

9
 .
.
•
(a) READ INPUT COORDINATES (b) FIT POINTS WITH B-SURFACES (e) CUT SURFACES WITH
SECTIONING PLANE

.
\
• )
(g) VERIFY NEW GEOMETRY (f) ASSEMBLE NEW (e) GENERATE ONE (d) GENERATE SPLINES
WITH SHADED SURFACE CROSS SECTIONS CONTINUOUS AT INTERSECTIONS
AND WIRE FRAME IMAGES; SPLINE CROSS-
SUPPLY DATA TO GRID SECTION;SAMPLE
GENERATOR FOR NEW POINTS

Figure 10. Geometry definition/verification procedure.

PHYSICAL DOMAIN COMPUTATIONAL DOMAIN

F E
E
.. ..

'I t D t = f(x. y. z)
'(
c:==-C A 'I = f(x. y. z)
B l' = f(x. y. z)

II .. 11
F
A B C D

Figure 11. Transformation of variables from physical domain to computational domain.

10
 i, j + 1

'''_f;-,~i~ - .
-
i - 1, j I i,j
_ f i -_ _

-- I
i + 1,j

T i, j - l

Figure 12. Definition of stretching Figure 13. Temporary outer boundary for
functions. grid orthogonality.

-300.-==~~------------------------.

-250

-200

-150

-100

-50

z 0

50

100

150

200

250

300+-====;~---.----.----'r----.--~
o 50 100 150. 200 250 300
Y
Figure 14. Cross section of grid through canopy and strake.

11
 -300 800

700
-200

600

-100
500

Z 0 Y 400

300
100
200

200
100

o 100 200 300 400 500 600 700 800

X

Figure 15. Cross section of grid through Figure 16. Planform plane of grid with body
wing. and wing (top view).

12
 2. Government Accession No.

I
1. Report No. 3. Recipient's Catalog No.
NASA TM·-88242
4. Title and Subtitle 5. Report Date

GEOMETRY DEFINITION AND GRID GENERATION FOR A April 1986

6. Performing Organization Code
COMPLETE FIGHTER AIRCRAFT
7. Authods) 8. Performing Organization Report No.

rhomas A. Edwards A-86208

10. Work Unit No.
9. Performing Organization Name and Address

11. Contract or Grant No.

Ames Research Center
Moffett Field, CA 94035
13. Type of Report and Period Covered
12. Sponsoring Agency Name and Address
Technical Memorandum
N ationalAeronautics and Space Administration 14. Sponsoring Agency Code
Washington, DC 20546
505-60
15. Su pplementary Notes
loint of Contact: Thomas A. Edwards, Ames Research Center, MS 202A-14,
Moffett Field, CA 94035 (415) 694-5369 or FTS 464-5369
16. Abstract

Recent advances in computing power and numerical solution procedures

h ave enabled computational fluid dynamicists to attempt increasingly
d ifficult problems. In particular, efforts are focusing on computations of
c omplex three-dimensional flow fields about realistic aerodynamic bodies.
To perform such computations, a very accurate and detailed description of
t he surface geometry must be provided, and a three-dimensional grid must be
g enerated in the space around the body. The geometry must be supplied in a
f ormat compatible with the grid generation requirements, and must be
verified to be free of inconsistencies. This paper presents a procedure for
Performing the geometry definition of a fighter aircraft that makes use of a
c ommercial computer-aided design/computer-aided manufacturing system.
F urthermore, visual representations of the geometry are generated using a
computer graphics system for verification of the body definition. Finally,
t he threE~-dimensional grids for fighter-like aircraft are generated by means
o f an efficient new parabOlic grid generation method. This method exhibits
good control of grid quality, and generates grids about 50 times faster than
c omparable grids generated via elliptic algorithms.

17. Key Words (Suggested by Author(s)) 18. Distribution Statement

G'eometry Unlimited
D efinition
Grid generation
Subject Category - 01
19. Security Oassif. (of this report) 20. Security Classif. (of this page) 21. No. of Pages 22. Price·

unclassified Unclassified 14 A02

·For sale by the National Technical Information Service, Springfield, Virginia 22161
End of Document