Engineering Analysis with Boundary Elements 44 (2014) 2835

Contents lists available at ScienceDirect

Engineering Analysis with Boundary Elements

journal homepage: www.elsevier.com/locate/enganabound

Panel method for mixed congurations with nite thickness

and zero thickness
Jos M. Ezquerro, Victoria Lapuerta n, Ana Lavern-Simavilla, Jos M. Garca, Taisir Avils
Universidad Politcnica de Madrid, Plaza de Cardenal Cisneros, 3, 28040 Madrid, Spain

art ic l e i nf o

a b s t r a c t

Article history:
Received 23 July 2013
Received in revised form
10 January 2014
Accepted 10 April 2014
Available online 8 May 2014

Panel methods are well-known methods for solving potential uid ow problems. However, mixed
congurations of obstacles with nite thickness and zero thickness have not been solved with these
methods. Such congurations arise naturally in delta wings, sailing boats, and even in complete aircraft
aerodynamics. In this work, a new numerical approach is proposed for solving 2D mixed congurations
of obstacles with nite thickness and zero thickness. The method is based on the Dirichlet and Neumann
formulations and is checked by comparison with analytical results.
& 2014 Elsevier Ltd. All rights reserved.

Keywords:
Aerodynamics
Panel method
Mixed conguration
Potential ow

1. Introduction
Methods for solving potential uid ows are frequently used in
engineering practices, mainly for preliminary design. Although nonpotential models like CFDs (Computational Fluid Dynamics) are widespread, they require long calculation times and the results are not
always reliable so inputs from potential methods are often needed.
One of the most developed potential methods is the panel
method [13]. The main advantage of this method is that it
reduces the dimension of the problem by one order, so the
numerical cost is very low compared with non-potential methods.
The panel methods allow one to calculate numerically the
solution of any given problem as long as the velocity potential
satises the Laplace equation. There has been much work and
many numerical codes based on panel methods [411] since the
pioneering work of Hess and Smith [4].
The panel method based on Green's formula was rst introduced in the work of Morino and Kuo [5], in which the primary
unknown was the velocity potential. There are two main formulations of panel methods based on Green's formula: Neumann and
Dirichlet [3]. The Dirichlet formulation solves the Laplace equation
numerically and the velocity potential is obtained. However,
with the Neumann formulation only differences of potential are
obtained. Dirichlet formulation is more stable, more suitable to
numerical computation than Neumann formulation and leads to
numerical errors of smaller order of magnitude.
n

Corresponding author.
E-mail addresses: jezquerro@eusoc.upm.es (J.M. Ezquerro),
mariavictoria.lapuerta@upm.es (V. Lapuerta),
ana.laveron@upm.es (A. Lavern-Simavilla), taviles@eusoc.upm.es (T. Avils).
http://dx.doi.org/10.1016/j.enganabound.2014.04.011
0955-7997/& 2014 Elsevier Ltd. All rights reserved.

However, these formulations cannot be directly applied to

mixed congurations of both nite and zero thickness, such as
the mast and the sail of a sailing boat, trafc signs [12], delta wings
or congurations like Gurney aps [13]. In [14] a distribution of
vortices is used to model the surface of a mast and sail and sources
and sinks are used to represent the ow separation, which is just
an empirical t to data from wind tunnel and is not related to the
subject of this paper. Nevertheless, in this work we show that
methods based on discrete vortex do not recover correctly the ow
around mixed two-dimensional congurations.
In this paper a numerical scheme for mixed two-dimensional
(2D) congurations of nite and zero-thickness bodies is presented. This method does not introduce spurious singularities in
the numerical resolution and gives very good precision results,
even for very thin airfoils or airfoils with cusped trailing edge.
The paper is organized as follows: in Section 2 Dirichlet and
Neumann formulations are reviewed. In Section 3 a new numerical
scheme for mixed two-dimensional (2D) congurations of nite
thickness and zero-thickness bodies is presented. This formulation
is also applicable to bodies without thickness, providing more
precise results than the discrete vortex solution. In Section 4 an
analytical solution is calculated in order to check the results
and the precision of the numerical method. Section 5 is dedicated
to the analysis of the results and, nally, in Section 6 the main
conclusions are extracted.

2. Review of Dirichlet and Neumann formulations

In this section Dirichlet and Neumann formulations are reviewed
and reformulated.

J.M. Ezquerro et al. / Engineering Analysis with Boundary Elements 44 (2014) 2835

29

2.1. Dirichlet formulation for nite-thickness bodies

The formulation is based on Green's integral equation [3],

xp 

Z
B

m  n ds 

 m  n ds 1

where is the velocity potential at a generic point P, of

coordinates xp , outside the body (see Fig. 1), n is the normal
vector dened such that it always points into the uid region, 1
is the potential of the stationary uid far enough from the body,
is the velocity potential on the upper side of the discontinuity
surface, w, and  on the lower one. m is the potential at xp
due to sources located at the body or at the discontinuity surface
with unit strength and m  n is the potential at xp due to
doublets located at the body or at the discontinuity surface with
their axes perpendicular to the B or w surfaces. The internal
potential inside the body has been considered zero. The velocity
potential dened by Eq. (1) already fullls the zero normal velocity
at the body boundary (=n 0).
Eq. (1) represents the velocity potential of a distribution of
doublets on both the surface of the body and the discontinuity


surface, with axis n and intensities and  , respectively.
The solution of this integral equation is obtained by letting xp
tend to B.
The basic idea of this method consists in solving Green's
integral equation by discretization of the body: the body is
replaced by N straight panels, see Fig. 2, and it is assumed that
the velocity potential is constant on each panel, j, which
correspond to a constant distribution of doublets along the panel.
Each panel is dened by two limiting nodes placed on the body
surface, whose coordinates are xj ; zj . In the middle point of each
panel a collocation point is placed, xcpj ; zcpj . The numbering of the
panels is in a clockwise sense, panel N 1 being the rst panel of
the lower surface starting from the trailing edge of the obstacle.
The discontinuity surface is modeled as a single panel of innite
length (it is numbered panel N 1).
In this formulation the unknown variables of the problem are
the velocity potential values on each panel, j. The equations
which have to be solved are obtained by particularizing Eq. (1) on
the collocation points. The problem is reduced to an algebraic

Fig. 3. Denition of jkF, jkI and 1

k .

system of equations:
Z
N
x  xcpk
k 1k  j
2
j 1 2 panel j jx  xcpk j
Z
x  xcpk
N  1
n ds 
 n ds
2
2
panel N 1 jx  x cpk j

where 1k is the potential of the stationary uid far enough from

the body calculated in the collocation points. Using an angular
reference parallel to the discontinuity surface panel and a local
frame attached to each panel (see Fig. 3), Eq. (2) can be reformulated as
N

k 1k

j j
N  1 1
j
k

kF kI
2

j12

where jkF, jkI and k are dened in Fig. 3. Note that the velocity
j
j
potential of each panel, given by the term j =2 kF  kI in
Eq. (3), does not introduce any discontinuity in the uid ow.
It introduces a discontinuity segment, the straight line between
the two nodes j and j 1, which models the surface of the body.
The velocity potential value changes from j =2 in the upper
j
j
surface to  j =2 in the lower surface of the panel since kF  kI
jumps from to  . As mentioned above, in order to solve Eq. (1),
the point xp , in the ow outside the body, tends to the collocation
points over the body, therefore the potential on the upper surface
of the panel, j =2, which is the potential of the outer ow around
the discretized body, is selected hereafter. The only discontinuity
line in the uid domain is the discontinuity surface which starts in
the trailing edge of the prole and ends at innity.
2.2. Neumann formulation for zero-thickness bodies
Nowadays, the potential methods most widely used to solve
zero-thickness bodies are the vortex-lattice method for 3D congurations and the discrete vortex method for 2D congurations
[3]. These methods can be derived from the Neumann formulation,
but some information of Green's integral is lost in the derivation
and the errors are larger than those obtained using the Dirichlet
formulation.
Here Neumann equations are reformulated by modeling the
zero-thickness bodies by panels with two wet faces (see Fig. 4),
as a degenerate boundary [15].
Eq. (3) leads to
jN

Fig. 1. Fluid domain in Greens integral equation.

2k 21k j kF  kI N  1 k
j

j1

and expanding the summation and the discontinuity term,

jN

j kF  kI N  1 k 1 kF  kI  k
j

j1

N kF  kI k
N

Fig. 2. Discretization of geometry for a nite-thickness obstacle.

j N=2

j kF  kI

j2

j N1

j N=2 1

j jkF  jkI
5

30

J.M. Ezquerro et al. / Engineering Analysis with Boundary Elements 44 (2014) 2835

situated on the lower surface, k r N=2, and upper surface, k 4 N=2:

N=2

k 1k

j 1
jak

N
j j
j j
j
j
kF  kI
 kI
2
2 kF
j N=2 1
jaN1k

k N 1  k N  1 1

k ;
2
2
2
j

N=2

k 1k

Fig. 4. Discretization of geometry for a zero thickness obstacle.

j 1
jaN1k

By using,

1
k

NkF 1
k 0

1
kI

j N=2 1
jak

k N 1  k N  1 1
k ;

2
2
2

k4

16

kF  kI
j

N
2

17

setting,
l N 1 j

and being j the potential jump between upper and lower surfaces
of the same panel,

j j  N 1  j

then, Eq. (5) can be rewritten as

18

N=2 1

k 1k

l N
lak

j N1

j1

j N=2 1

j jkF  jkI
10

kF  kI
l

j N=2 1
jaN1k

k N 1  k N  1 1

k ;
2
2
2

kr

kF  kI
j

N
2

19

N
j j
l l
l
j
kF  kI
 kI
2
2 kF

N=2 1

j kF  kI N  1 k  N kI
j

kF  kI

N
2

and introducing it into the rst sum of Eqs. (16) and (17) we have

jkI NkF 1  j

jN

kr

k 1k

j N=2 1
jak

l N
lak

N  1 1
k ;
k N 1k
2
2
2

N
k4 :
2

20

By introducing Eq. (10) in Eq. (4) and imposing the boundary

condition over the body at the collocation point k,

Reversing the order in the rst sum of Eqs. (19) and (20), taking
into account

k  nk 0

jkF  jkI  NkF 1  j  NkI 1  j ;

11

this leads to
0 2 1k  nk  N

0N
kI

j N1

j N=2 1

0j
kF

j 

0j
kI

12

0j

x k k  n k k
j

N 1  N 0

l1


k

l 0lkF  0lkI 2 1k  nk

15

k 1k

j N=2 1
jak

22

j  j j
j
kF  kI
2

k N  1 1
k

23
2
2
2


Finally, using j j  j , as dened in the previous section,

j  j j
j
kF  kI
2

k N  1 1

k
2
2

14

has been imposed.

Renaming the panels in the sum in order to begin from the
leading edge of the airfoil, reordering terms in Eq. (12) and
changing the dummy index j to l j N=2 we nally have the
algebraic system of equations that allows one to obtain M values of
the jump of the potential. It is remarkable that with Neumann
conditions the N unknowns of the potential cannot be obtained,
but only M N/2 unknowns: the jump of potential between the
faces of the panels, l:
l M1

j N=2 1
jak

13

and the Kutta condition,

M 0M
kI 

noting that k is the same point on the lower and upper surfaces
and identifying the corresponding potentials on the upper surface


(k ) and lower surface (k ), then become

k 1k

where

21

we have
M

k 1k

j 1
jak

k 1k

j 1
jak

kF  kI 

j M 1

2 2 k

24

kF  kI

j M 1

2 2 k

25

3. Method for mixed nite-thickness and zero-thickness

bodies

This system needs the Neumann formulation for zero-thickness bodies, formulated in the previous section, so the combined
equation system is composed by Eqs. (15), (24) and (25).
The velocity on each side of the body can be calculated using

In this section a new formulation is developed to solve zerothickness bodies and mixed congurations with both nitethickness and zero-thickness bodies.

V i7

3.1. Mixed DirichletNeumann formulation for zero-thickness bodies

Here we obtain a new formulation for zero-thickness bodies.
First, Eq. (3) is duplicated and applied at the collocation points k

i7 1  i7
ci

i 2; N;

26

where ci is the length of each panel. The pressure coefcient can

be calculated on the lower and upper surfaces with
7
cpi
1

 7 2
Vi
:
V1

27

J.M. Ezquerro et al. / Engineering Analysis with Boundary Elements 44 (2014) 2835

31

The whole conguration in the t-plane is rotated an angle,

, about the origin,

0

Fig. 5. Mixed conguration (thick and non-thick) panel discretization.

3.2. Formulation for mixed nite-thickness and zero-thickness bodies

The geometry of the body is shown in Fig. 5. The zero-thickness
part of the body is modeled by M panels numbered from the
trailing edge of the nite-thickness airfoil towards the trailing
edge of the airfoil itself and with normal vectors pointing
upwards. The nite-thickness part is divided into N panels
numbered clockwise and starting from the lower surface trailing
edge. The normal vector for the panels in the zero-thickness part
points towards the upper surface. The discontinuity surface is
taken as panel N M 1 and its normal vector is oriented upwards.
As in the previous case, a collocation point is placed at the center
of each panel.
The solution of the coupled problem is achieved by calculating
the potential jump at each of the panels with Eq. (15) and equating
the potential of the inner surface of the nite-thickness panels to
the internal potential, that is, zero. Finally, a set of algebraic
equations is obtained,
M
N M 0N

kI

j NM 1

j1

j N M

j 1
jak

kF  kI 
j

k N M 1

 1k ;
2
2 k

k 1; N:

te  i t 0  a

31

i 0

32

k N 1; ; N M;
j NM

j 1
jak

j j

j
1
 kI k N M k
2 kF
2
2

k k  k

!k

te  i t 0 a

37

and 1 r k r 2.
The second transformation, a Joukowski transformation from
the t-plane to the -plane, maps the circle and zero-thickness plate
onto a single plate (see Fig. 6),

R2
l

t
2R l

38

4RR l l
Rl

The third transformation is a Joukowski transformation from

the -plane to the -plane, the inverse of which is

30


k

36

where

k k

k 1k

1c
1c

29

k 1; ; N

and for

s ka

The potential of each panel is easily obtained with Eq. (25), and
taking into account the denition of the jump of the potential,
Eq. (9), we have
for

and translated to point t 0 a  i, where is the camber

parameter of KarmanTrefftz transformation, is the thickness
parameter of KarmanTrefftz transformation, R is the radius of the
q
2
circular body and a R= 1 2 . Then, the KarmanTrefftz
transformation leads to

28

In matrix form,
Ajk f j g fbg

35

The length of the plate in the -plane is

j 0jkF  0jkI 2 1k  nk ;

k N 1; N M

a
C
arctanB
@q A
2
2
R  a

39

40

with L=4.
The circle theorem is used and an appropriate vortex intensity
is included to obtain the complex velocity potential in plane,


2
i
log
41
g U 1 e  i ei

4U 1 sin

42

Finally, the conjugate velocity is

dg
dg d d dt
dg 1 d 1

ds d d dt ds d d dt ds
d
dt

43

where
33

d
2
1 2
d

44

34

d
R2
1 2
dt
t

45

ds
te  i t 0  ak  1  i
2ka
e
dt
te  i t 0 a3

46

4. Analytical solution
In order to validate the numerical scheme proposed for mixed
zero-thickness and nite-thickness congurations an analytical
expression is obtained for the conguration presented at the top of
Fig. 6, which consists of a curved airfoil and a camber line. Next,
the complete transformation outlined in Fig. 6 is described.
First, to obtain curved airfoils, a generalized KarmanTrefftz
transformation is used to map a symmetric conguration formed
by a circular body and a at plate (t-plane) into a curved airfoil and
a camber line (s-plane).

5. Results and discussion

Fig. 7 compares the pressure coefcient from the analytical
solution described in Section 4 with the pressure coefcient
obtained with the numerical scheme proposed in Section 3. As
this gure shows, the agreement is very good even with only 59
panels for the body and 31 for the plate. The amount of panels in

32

J.M. Ezquerro et al. / Engineering Analysis with Boundary Elements 44 (2014) 2835

Fig. 6. Complete conformal mapping transformation.

the surfaces of the thick prole and the camber line is such that
their lengths are approximately the same.
To check the convergence of the method, the error in the
circulation, , has been calculated. Note that the lift is proportional to the circulation. Fig. 8 shows the relative error of the
numerical method as a function of N, being num N  1 . The
results are very good. For example, for N 100, the relative error is
0.4% and for N 500 the error is 0.07%.
Fig. 9 compares the pressure coefcient from the analytical
solution described in Section 4 with the pressure coefcient
obtained with the numerical scheme proposed in Section 3 and
with the result obtained using the discrete vortex method. All
the calculations of the discrete vortex method presented in the
paper have been performed using the standard method [3] (vortex
placed at a quarter of the panel chord and collocation point
at three quarters of the panel chord). The error in the pressure

coefcient is calculated as error jcp;numerical  cp;analytical j and is

plotted in Fig. 10 computed with each numerical method for an
increasing number of panels. These gures show that:
(i) The error in our method is signicantly smaller than the one
in discrete vortex method, and the difference is more important for a small number of panels. Increasing the number of
panels in the discrete vortex method does not reduce the
error to the level found in our method. For example, in the
case of Fig. 10(c), the mean relative error with respect to
the analytical solution is: 1% in the upper surface and 0.2% in
the lower surface calculated with our method, and 16% in the
upper surface and 10% in the lower surface calculated with
the discrete vortex method.
(ii) When the number of panels increases, the error of the discrete
vortex method decreases, but this method never recovers

J.M. Ezquerro et al. / Engineering Analysis with Boundary Elements 44 (2014) 2835

33

Fig. 7. Comparison of the pressure coefcient obtained with the analytical solution
of Section 4 and the mixed DirichletNeumann numerical approach proposed in
Section 3 (DN) for N 59, M 32. The parameters of the conguration are 121
and l/R 7.

Fig. 8. Relative error in the circulation for the numerical approach proposed in
Section 3. The parameters of the conguration are 121 and l/R 7.

Fig. 10. Error in the pressure coefcient calculated with the mixed Dirichlet
Neumann numerical approach proposed in Section 3 (DN) and the discrete vortex
method for (a) N 49, M 17, (b) N 94, M 32 and (c) N 146, M 49. The data of
the conguration are  21, l/R 3, k 1.8, 0.05, 0.3.
Fig. 9. Pressure coefcient obtained with the analytical solution of Section 4, the
mixed DirichletNeumann numerical approach proposed in Section 3 (DN) and
the discrete vortex method for N 49, M 17. The data of the conguration are
 21, l/R 3, k 1.8, 0.05, 0.3.

correctly the behavior in the junction between the nite

thickness and zero-thickness elements. The maximum error in
the junction obtained with our method in case (a) is 0.4559 for
C p and 0.8388 for C p , in case (b) is 0.3641 for C p and 0.8354

for C p , and in case (c) is 0.3131 for C p and 0.6121 for C p . The
maximum error in the junction obtained with the discrete
vortex method in case (a) is 3.143 for C p and 4.367 for C p , in
case (b) is 3.423 for C p and 5.069 for C p , and in case (c) is 2.406
for C p and 3.819 for C p . Our method does signicantly better in
the junction even with a small number of panels.

34

J.M. Ezquerro et al. / Engineering Analysis with Boundary Elements 44 (2014) 2835

(iii) Our method ts much better than the discrete vortex method
with the analytical solution in the nite thickness part. This is
because the order of magnitude of the error in the discrete
vortex method is the same as the order of magnitude in
the Neumann method error, which is higher than the order of
magnitude in the Dirichlet method error.

solution even for this very thin airfoil. It is well known that the
discrete vortex method is not applicable to very thin congurations, as this one.
The method here described is also capable of providing good
results even for congurations with cusped trailing edges as can
be seen in Fig. 13. In this gure the pressure coefcient computed

Fig. 11 shows the pressure coefcient obtained using the

numerical approach of Section 3 for a KarmanTrefftz body with
a cosine-shaped camber line. Notice that in the lower surface the
angle in the junction is greater than 1801, and therefore the
velocity is innite, whereas in the upper surface the angle is
smaller than 1801 and the velocity is zero. The method recovers
correctly this behavior.
Fig. 12 compares the pressure coefcient from the analytical
solution described in Section 4 with the pressure coefcient
obtained with the numerical scheme proposed in Section 3 for a
very thin airfoil. The prole used is a KarmanTrefftz, with
k 1.915, which gives an airfoil with 10.8% of maximum thickness
and a trailing edge angle of 15.31. The pressure coefcient is
calculated for an angle of the incident ow of 51. We have used 60
panels in the airfoil and 35 panels in the tail. As can be seen the
numerical solution ts extremely well with the analytical exact

Fig. 11. Pressure coefcient calculated with the mixed DirichletNeumann numerical approach proposed in Section 3 (DN) for N 59, M 46. The data of the
conguration are  21, k 1.8, 0.19, 0.2.

Fig. 12. Pressure coefcient calculated with the analytical solution of Section 4 and
the mixed DirichletNeumann numerical approach proposed in Section 3 (DN) for
N 60, M 35. The data of the conguration are  51, l/R 5, k 1.915, 0.04,
0.

Fig. 13. Pressure coefcient calculated with the analytical solution of Section 4, the
mixed DirichletNeumann numerical approach proposed in Section 3 (DN) and
the discrete vortex method for (a) N 31, M 11, (b) N 80, M 26 and (c) N 153,
M 48. The data of the conguration are  21, l/R 3, k 2, 0.2, 0.3.

J.M. Ezquerro et al. / Engineering Analysis with Boundary Elements 44 (2014) 2835

35

discrete vortex method is shown in Fig. 14, for an increasing

number of panels. These gures show that:
(i) Our method ts much better than the discrete vortex method
with the analytical solution. In fact, in the discrete vortex
method the error increases when the number of panels
increases and this method does not converge to the analytical
solution. The maximum error in the junction obtained with
our method in case (a) is 0.1787 for C p and 0.0970 for C p , in
case (b) is 0.2313 for C p and 0.1256 for C p , and in case (c) is
0.2368 for C p and 0.1457 for C p . The maximum error in the
junction obtained with the discrete vortex method in case
(a) is 2.732 for C p and 6.603 for C p , in case (b) is 4.183 for C p
and 2.456 for C p , and in case (c) is 13.367 for C p and 5.583
for C p .
(ii) Our method ts very well with the analytical solution even for
a low number of panels. For increasing number of panels our
method converges to the analytical solution, and the method
performs well even near the junction.

6. Conclusions
In this work a new formulation for combined nite-thickness
and zero-thickness bodies has been developed. This formulation
has been tested by comparison with analytical solutions and gives
very good agreement even for very thin airfoils and airfoils with
cusped trailing edge. The convergence of this new formulation has
also been tested.
The method can be very useful for preliminary design in all
kinds of problems that combine both nite-thickness and zerothickness bodies; these include sailing boats, Gurney ap congurations, and the study of realistic aircraft aerodynamics.
References

Fig. 14. Error in the pressure coefcient obtained with the mixed Dirichlet
Neumann numerical approach proposed in Section 3 (DN) and the discrete vortex
method (a) N 31, M 11, (b) N 80, M 26 and (c) N 153, M 48. The data of the
conguration are  21, l/R 3, k 2, 0.2, 0.3.

with the discrete vortex method and our method is compared to

the pressure coefcient obtained with the analytical solution, for
an increasing number of panels. The error in the calculation of the
pressure coefcient obtained with the numerical scheme and the

[1] Hess JL. Panel methods in computational uid dynamics. Annu Rev Fluid Mech
1990;22(1):25574.
[2] Erickson LL. Panel methodsan introduction. NASA technical paper 2995;
1990.
[3] Katz J, Plotkin A. Low-speed aerodynamics. New York: Cambridge University
Press; 2001.
[4] Hess JL, Smith AMO. Calculation of non-lifting potential ow about arbitrary
three-dimensional bodies. J Ship Res 1964;8(2):2244.
[5] Morino L, Kuo CC. Subsonic potential aerodynamics for complex congurations: a general theory. AIAA J 1974;12(2):1917.
[6] Rubbert PE, Saaris GR. A general three-dimensional potential ow method
applied to V/STOL Aerodynamics. SAE paper no. 680304; 1968.
[7] Hess JL. Calculation of potential ow about arbitrary 3-D lifting bodies.
Douglas Aircraft Company. Report MDC-J5679-01; 1972.
[8] Morino L, Chen LT, Suciu EO. Steady and oscillatory subsonic and supersonic
aerodynamics around complex congurations. AIAA J 1975;13(3):36874.
[9] Ehlers FE, Rubbert PE. A mach line panel method for computing the linearized
supersonic ow. NASA CR-152126; 1979.
[10] Hwang WS. A boundary node method for airfoils based on the Dirichlet
condition. Comput Methods Appl Mech Eng 2000;190(1314):167988.
[11] Ye W, Fei Y. On the convergence of the panel method for potential problems
with non-smooth domains. Eng Anal Bound Elem 2009;33(6):83744.
[12] Sanz-Andrs A, Lavern-Simavilla A, Baker C, Quinn A. Vehicle-induced force
on pedestrian barriers. J Wind Eng Inderodyn 2004;92:41326.
[13] Morishita E. SchwartzChristoffel panel method. Trans Japan Soc Aeronaut
Space Sci 2004;47(156):1537.
[14] Wilkinson S. Simple multilayer panel method for partially separated ows
around two-dimensional masts and sails. AIAA J 1987;26(4):3945.
[15] Chen JT, Hong HK. Review of dual boundary element methods with emphasis
on hypersingular integrals and divergent series. Appl Mech Rev 1999;52
(1):1733.