Copyright© 1997, American Institute of Aeronautics and Astronautics, Inc.

AIAA-97-1829
MODERN PANEL METHOD TECHNIQUES FOR
MODELLING WAKE-BODY INTERFERENCE
A Terzi and T W Chiu

School of Engineering, University of Exeter

North Park Road, Exeter EX4 4QF, UK

long distance even within a very small time-step. On

the other hand, if one tries to evaluate the surface
pressure close to a panel edge, the pressure will be
unduly low for the same reason. In most higher order
panel method formulations, the tendency of the
Abstract velocity to approach infinity at the edges of the panels
may also arise.
This paper* describes some techniques recently
developed for modelling the aerodynamics of bodies The Close Approach Method (CAM) uses
using the panel method. The computational scheme is interpolation between the velocities at the control
based on 'External Neumann' boundary conditions points to determine the velocities near the surface.
and it uses a piecewise constant distribution of source The process ensures a smooth velocity field around
on non-lifting bodies and source plus doublet on solid bodies modelled with panels even when the
lifting bodies. Vortex filaments are used to model the observation point is very close to the surface. The use
trailing vortex wake behind lifting bodies. The new of this method is particularly beneficial in computing
techniques are developed for the interaction between wake relaxation because it assumes that the flow is
vortex wakes and solid bodies and for the evaluation parallel to the surface, thus strictly enforcing the
of velocity components close to a body surface. boundary condition. Therefore the method can
prevent the relaxed wake vortices from passing
through the surface despite the modification of the
Introduction velocity field due to the moved wake.

In modelling the aerodynamics of complex bodies, The Slice Method (SM) aims to avoid the assumption
the study of wake/body interference is of great that the panel strengths are fixed during the wake
importance (see, for instance, Dutt et al9, Katz et al13 relaxation. With this method the influence of the
and Lamarre et al15). In a panel method computation, wake deformation is taken into account and the body
such interactions have always been difficult to deal panel source or doublet strengths are updated in each
with. This is because the shape of the vortex wake is wake iteration. However, to improve the efficiency of
usually not known a priori. When the common wake the computation, only the strength of the panels on
relaxation technique is applied, the vortex filaments the part (or slice) of the body closest to the vortex
in the wake may not always be able to circumvent filament being considered is updated. In this way the
obstacles downwind of the wing, and can easily go body panel strengths are constantly updated by the
through the surface panels of the obstacles. Another relaxing wake, which in turn update the strengths of
difficulty is usually called the Close Approach the vortex filaments in the wake. This results in
Problem. The induced velocity at any point close to unproved modelling of the interaction between the
the edge of a surface panel with a low order body and the wake. In particular, the velocity field
singularity distribution is usually unduly high. If a within the slice due to the combined influence of the
vortex filament moves towards an edge, it will surface panels and the vortex filaments now satisfies
experience very high velocities and will be moved a the boundary condition on the body, and therefore the
vortices would not pass through the body.
1
Copyright © 1997 by A Terzi and T W Chiu. Finally, the Choice of the Control Point Method
Published by the American Institute of Aeronautics which uses the 'cosine law' to evaluate both the
and Astronautics, Inc. with permission.
 Copyright© 1997, American Institute of Aeronautics and Astronautics, Inc.

discretisation of the surface into panels and the Making use of Green's theorem and of potential flow
position of the control point on the states panels on theory, the potential at an observation point P can be
lifting bodies surface allow to get more accurate expressed by its value on the boundary S as
velocity values.

General Background
(2)
When compared with the other schemes, the inviscid where <7 is the source density per unit area on S, ju
incompressible panel method has the principal is the doublet density per unit area on S and r is the
advantage that the elements (and hence the length of the vector f from the point Q on S and the
unknowns) are situated only on the surface point P.
(boundary) of the configuration and not throughout
the external space. Thus the order of the problem is The gradient of equation (2) gives the induced
reduced by one, which enables very complex
geometries to be handled, while the number of velocity UP at the observation point P
unknowns is kept to a tolerable level. Furthermore,
since only the surface of the configuration is (3)
discretised, the boundary mesh generation is much
simpler than most other numerical schemes. With the Imposing the External Neumann Boundary conditions
Panel Method, the aerodynamics of a large number of on a set of control points, a system of linear equations
designs can be investigated very quickly with is obtained, the matrix form of which is
computers affordable by small companies.

By allowing some approximations, the Panel Method (4)

can produce acceptably accurate results for a large
number of designs within a realistic time scale. The in which the unknowns %j give the strength of the
fundamental mathematics underlying the Panel
Method techniques applies rigorously only to singularities on the panels. After the solution of this
incompressible, inviscid flow for which the governing system of linear equations, velocities, pressures,
equation is Laplace's equation forces and moments can be evaluated both on the
surface of the bodies and in the space around them.
For more details on the various Panel Method
(i) formulations see, for instance, Chiu2"3'5"7, Chiu et al4,
Hess10, Hunt11'12, Katz et al14, Roggero et al19 and
where <j) is the velocity potential, while satisfying Smith et al20.
some boundary conditions on the boundary S. By
analysing the expression of potential in terms of Panel Methods can be used to study hypothetical
surface integrals (Green's theorem) the total field aerodynamic cases of zero Mach number and infinite
(related to the idealised physical field and subject to Reynolds number for which there is no effect of
its boundary conditions) is formed by the onset field compressibility and for which viscous effects are
(the combined 'unperturbed' field associated with the confined to infinitesimally thick layers of fluid
'outer boundary' and with any such fixed adjacent to the configuration surface and in its wake.
singularities) and the perturbation field (the The methods can also give useful results for non-zero
perturbation potential and perturbation velocity due (sub-critical) Mach numbers and for flows in which
to the introduction of the inner surfaces). By looking viscous effects are not negligible. This explains the
at the velocity potential form, it can be seen that the big effort on research aimed to develop new panel
singularity distributions (source and/or doublets) of method techniques which allow more realistic
unknown density may be considered to produce the numerical modelling of the problem to be studied in
perturbation potential that together with the onset order to complement the more traditional and very
potential gives the total potential. This concept forms expensive facilities offered by the wind tunnel.
the basis of the 'Surface Singularity Techniques'
(Panel Method) of which numerous varieties
exist5-12-14-19.
 Copyright© 1997, American Institute of Aeronautics and Astronautics, Inc.

Close Approach Method

The three-dimensional panel method used for the (5)

studies reported in this paper is based on the External
Neumann boundary condition and it adopts a
piecewise constant source and doublet distribution on where A, B and C are the coefficients obtain from the
the panels. This formulation leads to the problem that coefficients of the plane of interpolation, k is the
the velocities tend to infinity at the panel edges and selected control point number, xk and zk are the co-
reliable surface velocities can only be found at the ordinates of the k* owned control point expressed in
control points12. Indeed this can be easily explained the global reference system and M is the number of
by recalling, for instance, that a panel with a control points selected. By imposing the 'least'
piecewise constant doublet distribution is equivalent condition we obtain a system of linear equations that
to having a vortex ring round its edges11. Therefore can be easily solved by the Lower-Upper
singularities exist in the induced velocity field close Decomposition method. Once the coefficients A, B
to the edges. and C are known they can be used to evaluate the
flow velocity component at P in the / direction, Vpi.
The Close Approach Method uses the method of A similar approach can be used to evaluate the
interpolation between the velocities at the control A

points (which can be found reliably) to determine the velocity component at P in the j direction, Vpj.
velocities near the surface while strictly imposing
fluid continuity. More specifically, considering a
local reference system (that will be described later)
the velocity components in two of the directions are
obtained by interpolation while the one in the third
direction is determined by the continuity equation.

If the velocity is required at a particular point P, the

nearest node N1 on the body is determined first. If the
perpendicular distance from this node to the
observation point is less than a certain distance (circa
1 panel size) then the interpolation scheme is used.
The velocities at the control points of the panels
which have this node as one of the corners are used.
Let us call them "owned control points". These are
transformed from the global co-ordinate system to a Fig.l: Selection of the panels that belong to
local one. One of the three principal directions of this the 'nearest node N^.
orthogonal local co-ordinate system is the normal unit
vector calculated as the mean of the normal unit
vectors of the panels concerned (nm ). (see Fig 1)
The normal velocity component can be calculated by
applying the continuity equation. By making use of
The other two principal directions are thus tangential
^ *. the truncated Taylor series expansions for Vpi and Vpj,
to the body surface (i,j). the equations used to evaluate them by means of the
least squares method and the continuity equation
Now the velocities in the direction of the two
tangential principal axes need to be evaluated. This
Pi Pnm
can be done by linear interpolation using the (6)
velocities calculated at the control points of the dn
panels concerned. If there are more than three control
points involved, the least squares method is (Fig.2) we can finally evaluate Vpn .
performed. The sum of the squared difference K Finally, the velocity components at P are transformed
between the value of the velocity component Vki in to the global reference system.
s*.

direction / at each point and the corresponding point

on the plane of interpolation has got the form
 Copyright© 1997, American Institute of Aeronautics and Astronautics, Inc.

Velocity Potential and the direction of the tangential

velocity are discontinuous across such a sheet. The
normal derivative of the potential is continuous (and
equal to zero) by virtue of the zero-force condition if
it is assumed that the wake entrains no fluid from the
adjacent domains.Hunt12

Wake Modelling

The lifting bodies vortex sheets are discretised by

doublet panels of constant doublet distribution. This
is then equivalent to having a vortex lattice extending
downstream and, because of the constancy of the
doublet distribution, we end up dealing with vortex
filaments. The vortex circulation is calculated by
applying the Kutta-Joukowski condition. The wake is
defined in a geometric way and the initial position is
Fig.2: Imposition of the continuity equation for the prescribed. The doublet strength over the wake is
velocity component in nm direction. found by iterating until the pressures on the upper and
lower sides of the trailing edge are equal. During this
iteration the wake geometry is modified.
The use of this method is particularly beneficial
because it assumes that the flow is parallel to the
A vortex wake is required behind lifting surfaces to
surface, thus automatically enforcing the boundary
satisfy both the Helmholtz circulation constraints and
condition. Therefore it will prevent the relaxed wake
the Kutta condition for which the flow must leave the
vortices from passing through the surface despite the
trailing edge smoothly1. Since the circulation of each
modification of the velocity field due to the moved
vortex segment relates directly and linearly to the
wake.
doublicity of the trailing edge panels, the influence
coefficients can be included in the corresponding
entries of the influence coefficient matrix.
Slice Method
The wake is modelled by vortex filaments of constant
In some panel method application codes, the panel
circulation along its length. Each vortex filament is
strengths were assumed to be fixed during the wake
made up of a number of straight vortex segments. In
relaxation. The assumption was made simply because
order to get more accurate results the vortex segments
resolving for the panel strengths on the entire body at
closer to the bodies should be of shorter length with
every stage would be too time-consuming. Clearly,
respect to the ones further away. Initially the wake is
this is an incorrect assumption since the movement of
assumed to be fixed and with this configuration the
the vortices does induce a velocity field which will
influence coefficient matrix is evaluated. Then the
affect the global velocity field and hence the panel
strengths. Furthermore, the investigation of the strength of the singularities is evaluated by means of
an iterative solution method. Finally, the wake
induced velocity field- showed that if the panel
strengths are not constantly updated, the wake will relaxation procedure is performed as an iterative
procedure. At each iteration the wake is analysed
tend to pass through the body surface. This is
extremely undesirable and must be avoided. station by station (Fig.3) and at each station the
nodal displacements are calculated and the nodes
(vortex segments joints) are properly displaced.
WAKE RELAXATION PROCEDURE
As a compromised measure, the present method
A 3D-\vake is a region of rotational fluid which carry updates the strength of only a small number of panels,
away the vorticity generated within the Boundary which are the closest to the vortex filaments
Layer on the physical configuration; in the limit of concerned. In other words, only a small but greatly
infinite Re the wake is of infinitesimal thickness and influenced part of the system of linear equation is
force-free stream surface of the mean flow. The solved at each relaxation step. In the present
formulation, the panels are chosen form a "slice" of
 Copyright© 1997, American Institute of Aeronautics and Astronautics, Inc.

the body (Fig.4a, 4b), hence the method is called the and will be solved by a modified Gauss-Seidel
Slice Method. iteration technique.

Note that solving the system which contains the

modifications induced by the wake means to impose
the boundary conditions of non-permeability to the
new flow configuration. This prevents the vortices
from passing through the body surface at each
relaxation step. The velocities in the wake vortex
filaments are men determined and the wake relaxation
can be carried on. The Slice Method procedure will
have to be performed at each relaxation step.

nodo N sulta scia

Fig.3: y-stations for the wake relaxation procedure.

Before calculating the velocities on the filaments at a

particular y-station (being the free-stream in the y
direction) the velocities on the panels upstream and
downstream from this y-position within a certain
distance are found, and then the panels strengths are
resolved. Then, by using the Biot-Savart law, the
velocity induced by all the wake vortices on the
panels within a certain streamwise distance from the
node are found. The normal components of these
N
induced velocities will be subtracted from the 'initial'
normal velocity components (in this context the word
'initial' means 'previous' since the wake relaxation is
an iterative procedure) and the resulting velocity
variations AF^ (where / goes through the panels of
slice) can be considered as due to the freestream.
Recalling the linear system of equations to be solved
for the singularities strength (equation (4)), where Fig.4a,4b: Slice method within the wake relaxation
iterative procedure.
\/3k\ is the vector that contains the boundary
conditions, the AFj values will have to be added to
the relative entries of the right hand side vector. The
The updating of panel strengths of the panels
obtained system of linear equations will have the
contained in the slice will significantly improve the
form accuracy while keeping the computational time within
a reasonable tolerance. It is worth noting that the
(7) wake deformation greatly affects the flow field
(Lamarre et al15). Therefore the Slice Method, which
 Copyright© 1997, American Institute of Aeronautics and Astronautics, Inc.

ensures more precise evaluation of wake deformation, experimentally showed that the cosine spacing
produces more accurate results. significantly improves the accuracy of the drag
calculated by the PPA. Nevertheless the Trefftz-plane
integration of the rolled-up wake was found to be
Choice Of The Control Points very sensitive to errors in wake shape, Smith and
Kroo20. Since the wake geometry is generated from a
The local induced velocities are very important for vortex-lattice method, the use of the PPA on lifting
the design optimisation of devices such as winglets, bodies panelled in according to the cosine law results
wingtip stores, closed-coupled tandem wings. remarkably accurate.
Therefore, when studying wake-body and bodies
interference the correct evaluation of the local
induced velocities is desirable. Since the induced Results
velocities cause downwash the effective angle of
attack changes which causes induced drag, a better The CAM and the SM were implemented in a panel
evaluation of this effect gives a better estimation of method code. The code was then tested using a case
the velocity values and hence of the force on the of an obstacle (an ellipsoid) placed behind a
wing. rectangular wing of NACA 0010 section. 523 source
panels were used on the ellipsoid while a combined
Dumitrescu8 demonstrated that by both panelling a 334 source plus doublet panels were used on the
wing body and choosing the control points of the wing. Each wake vortex filament consists of 20
panels accordingly to the 'cosine law', more accurate streamwise segments. The symmetry of the
induced velocity values can be obtained. configuration was made use of in the computation.

From the mathematical consistency of the relationship Figures 5 - 7 show the bodies and the rolled up
between the analytically correct formulation of the vortex wake from three different points of view at an
induced velocity calculation and the approximation angle of attack of 0°. Figure 5 shows the
by the Panel Method, it follows that, by using the configuration viewed from downstream. It is clear
lifting line theorem for wings such that the circulation that the wake is able to circumvent the ellipsoid
distribution can be developed as a trigonometric smoothly. One can see the interaction between the
polynomial form, a discretised approximation of the two bodies. The shape of the wake clearly differs
Prandtl's equation and of the integral for the induced from what one would normally expect without an
velocity can be obtained by replacing the wing with a obstacle.
system of horse-shoe vortices IF the division in
panels is operated according to the cosine law and IF
the control points are chosen at half distance of the
edges of the panel considering as a spatial variable
the local angle 9 , which is related to the co-ordinate
that goes along the chord, by the expression

y=- (8)

where b is the wing span.

As far as the authors are aware of, the theory above

has been tested only for isolated wings. In the
research reported in this paper the same theory has
been used within the calculation of wing-wake and
wing-body interference. The results showed that also
for these configurations the accuracy had improved.

Letcher16 outlined the superior accuracy and

convergence properties of the far-field analysis by
Trefftz-plane integration (FFA) with respect to the
near-field force calculation (PPA). Beside that, he Figure 5
 Copyright© 1997, American Institute of Aeronautics and Astronautics, Inc.

Figure 6 Figure 8

Figure 7 Figure 9

Figures 8 - 1 0 show the same configuration at an

angle of attack of 16°. The ability of the vortex wake
to go round the ellipsoid close to its surface is shown
clearly in figure 9. Once again the CAM and the SM
ensure that the vortices do not go through the body.
 Copyright© 1997, American Institute of Aeronautics and Astronautics, Inc.

Conclusions

The use of the CAM and the SM has greatly

improved the ability of the panel method in modelling
the flow around bodies in which the interaction
between the vortex wakes and body parts is common.
While the CAM allows the correct evaluation of the
velocity field close to a boundary surface, the SM
enables the precise modelling of wake-body
interference. The simple example illustrated in this
paper was chosen because, despite its simplicity,
remarkable interference was likely to be observed.

—— AnalyticarResults
O Panel Method with Close Approach Method

Figure 10

To illustrate numerically the advantage of the close

approach method, we computed the pressure
coefficients close to the surface of the same ellipsoid
in a potential flow in the x-direction. The ellipsoid
surface is described by the ellipsoidal co-ordinates, £,
u, and q> which are related to the Cartesian co-
ordinates:

x = c/<
Figure 11

z=
References
where c is half the distance between the foci of the [1] Anderson, J.D., "Fundamentals of
ellipsoid, "£ = constant" represents an ellipsoidal Aerodynamics" McGraw Hill, 1985.
surface for -l<u<l and -7c<<p<7i. For the ellipsoid we
used, c = 0.284 m, £ = 1.157. The points at which [2] Chiu, T.W., "Aerodynamic Loads on Railway
analytical and computed pressure coefficients are Train in a Cross-Wind at Large Yaw Angles",
compared are along the line of intersection of the PhD Thesis, Cambridge University, Engineering
plane 9 = 0 and the ellipsoidal surface C, = 1.169. The Departments, 1990.
points are thus effectively about 5 mm from the [3] Chiu, T.W., "The use of boundary element
surface. In Figure 7, the pressure coefficients are method (panel method) in train aerodynamics",
plotted against the values of u. It can be seen that the Presented in the Wind Engineering Society
results from the close approach method agree well meeting on Wind Effects on Road and Rail
with theory even though the points are very close to Vehicles, 17 February 1992, London, Abstract
the surface. published in the Wind Engineering Society
Newsletter no.8, April 1993.
 Copyright© 1997, American Institute of Aeronautics and Astronautics, Inc.

[4] Chiu, T. W., Wood, B.D. and Berney, A.H., [15] Lamarre, F. and Paraschivoiu, I., "Efficient
"Application of B.E.M. C.F.D. in Formula-One panel method for vortex sheet roll-up", AIAA
car aerodynamics" Proceedings of the Journal of Aircraft, 29 (1), 28-33, 1992.
International Conference on Numerical Methods
[16] Letcher, J.S., "Convergence of Lifting and drag
in Continuum Mechanics, High Tatras, Slovakia,
predictions by a Morino Panel Method
19-22 September, 1994.
(VSAERO), AIAA Journal, Vol.26, No.9,
[5] Chiu, T.W., "Prediction of the aerodynamic pp.1019-1020, 1989.
loads on a railway train in a cross-wind at a large
[17] Press, W.H., Flannery, B.P., Teukolsky, S.A.
yaw angles using an integrated two- and three- and Vetterling, W.T., "Numerical Recipes, the
dimensional source/vortex panel method" art of Scientific Computing (Fortran Version)"
Journal of Wind Engineering and Industrial
Cambridge University Press.
Aerodynamics 57, 1995, 19-39.
[18] Rachason, T.F. and Katz, J., "Unsteady Panel
[6] Chiu, T.W., "The application of modern panel
Method for flows with multiple bodies moving
method techniques to sport aero/hydrodynamics"
along various paths", Journal of Aircraft, Vol.32,
AIAA paper 95-2210, AIAA 26th Fluid
No. 1, July-Aug. 1994.
Dynamics Conference, 19-22 June 1995, San
Diego, U.S.A. [19] Roggero, F. and Larguier, R., "Aerodynamic
calculation of complex three-dimensional
[7] Chiu, T.W., "Panel Methods in sports
configurations", Journal of Aircraft Vol. 30,
aerodynamics-racing cars", Summaries of Paper,
No.5, Sept-Oct.1993.
Conference on the Mathematics of Finite
Elements and Applications, Brunei University, [20] Smith, S.C. and Kroo, I.M., "Computation of
25-28 June, 1996,26-27. induced drag for elliptical and crescent-shaped
wings", Journal of Aircraft, Vol. 32, No.l, July-
[8] Dumitrescu, L.Z., "Sur la tegitimite" et la
Aug. 1994.
precision des calculs des vitesses induites par la
me"thode des panneaux sur les ailes d'envergure
finie", La recherche ae~rospatiale, 1993 (1), SI-
35.
[9] Dutt, H.N.V. and Rajeswari, S.R., "Wing-Body
interference using a hybrid Panel Method", Acta
Mechanica 106, 111-126, 1994.
[10] Hess, J.L. and Smith, A.M.O., "Calculation of
potential flow about arbitrary bodies" Progress
in Aeronautical Science Series, 1996.
[11] Hunt, B., "Relationship between volume, surface
and line distribution of vorticity, source and
doublicity", British Aerospace Notes, 1977.
[12] Hunt, B., "The panel method for subsonic
aerodynamic flows: A survey of mathematical
formulations and numerical models and outline
of the new British Aerospace scheme",
Computational Fluid Dynamics, Lecture Series
4, 1978, Von Karman Institute for Fluid
Dynamics.
[13] Katz, J. and Dykstra, L., "Effect of wing body
interaction on the aerodynamics of two generic
racing cars", SAE Paper No. 920349, 1992.
[14] Katz, J. and Plotkin, A., "Low-speed
aerodynamics: from wing theory to Panel
Method", McGraw Hill, New York, 1991.
Copyright© 1997, American Institute of Aeronautics and Astronautics, Inc.