Vous êtes sur la page 1sur 73

Bachelor Thesis

May 29 2009
Aerodynamic Performance and
Stability Simulation of Dierent
Flying Wing Model Airplane
Congurations
Author:
Flavio Gohl
Supervisor:
Stefan Leutenegger
Dario Schafroth
Professor:
Prof. Dr. Roland Siegwart
Abstract
In ying wing design, the stability criteria often decrease the aircrafts
performance and vice versa. Therefore a pitch unstable wing can have a
higher performance. The question to be answered in this thesis is, how
much performance can be won by a pitch unstable wing, stabilised with
a PID controller.
An application is implemented for studying ying wing dynamic and
stability behaviour. The simulation is based on a vortex lattice method
integrated in a rigid body simulation. The vortex lattice method is mod-
elled with singularity elements in chord and spanwise direction on a three-
dimensional wing. With this method, three-dimensional wing geometries,
asymmetrical ap deection and asymmetrical inow can be simulated.
The application is qualitatively evaluated with real ight tests of a ying
wing with measurements of the period time of a phygoid oscillation.
Three dierent wings, amongst them a pitch unstable wing with high
performance, are analysed in their dynamic stability behaviour and per-
formance. The pitch unstable wing has a slightly higher performance than
the optimised stable wing.
i
Acknowledgement
I would like to express my sincere appreciation to all those who have contributed,
directly or indirectly, to this Bachelors thesis in form of technical or other
support. I give my special thanks to Stefan Leutenegger and Dario Schafroth,
Michael Mller, Jens Walther and Philippe Chatelain and Ursina Gysi.
ii
Contents
1 Introduction 1
1.1 Objectives . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
1.2 Work structure . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
2 State of the Art 3
2.1 State of the Art of Aerodynamical Force Calculation . . . . . . . 3
2.2 State of the Art of Flight Simulators . . . . . . . . . . . . . . . . 3
3 Methods for Aerodynamic Force Calculation 5
3.1 Vortex Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
3.1.1 Theoretical Background on Vortex Methods . . . . . . . . 7
3.1.2 Lifting Line Method and Nonlinear Lifting Line Method
in General . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
3.1.3 Lifting Line Method . . . . . . . . . . . . . . . . . . . . . 11
3.1.4 Nonlinear Lifting Line Method . . . . . . . . . . . . . . . 12
3.1.5 Vortex Lattice Method . . . . . . . . . . . . . . . . . . . . 14
4 Implementation of a Suitable Method 18
4.1 Nonlinear Lifting Line Method . . . . . . . . . . . . . . . . . . . 19
4.1.1 Inputs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
4.1.2 Mesh Generation . . . . . . . . . . . . . . . . . . . . . . . 19
4.1.3 Sideslip Angle . . . . . . . . . . . . . . . . . . . . . . . . . 20
4.1.4 Inuence Coecients . . . . . . . . . . . . . . . . . . . . . 21
4.1.5 Initial Guess . . . . . . . . . . . . . . . . . . . . . . . . . 21
4.1.6 Iterative Process / Coupling the Prole Information . . . 22
4.1.7 Force and Moment Calculation . . . . . . . . . . . . . . . 22
4.1.8 Evaluation of the Nonlinenar Lifting Line Method . . . . 23
4.2 Vortex Lattice Method . . . . . . . . . . . . . . . . . . . . . . . 25
4.2.1 Inclusion of the Prole . . . . . . . . . . . . . . . . . . . . 26
4.2.2 Mesh Generating . . . . . . . . . . . . . . . . . . . . . . . 26
4.2.3 Sideslip Angle . . . . . . . . . . . . . . . . . . . . . . . . . 27
4.2.4 Inuence Coecients . . . . . . . . . . . . . . . . . . . . . 28
4.2.5 Flap deections . . . . . . . . . . . . . . . . . . . . . . . . 30
4.2.6 Force and Moment Calculation . . . . . . . . . . . . . . . 30
4.2.7 Stall . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
4.2.8 Evaluation of the Vortex Lattice Method . . . . . . . . . 31
4.3 Integrating the Methods in the Simulation . . . . . . . . . . . . . 34
5 Evaluation of the Simulation 36
5.1 Flight Test Results . . . . . . . . . . . . . . . . . . . . . . . . . . 36
5.2 Simulation Results . . . . . . . . . . . . . . . . . . . . . . . . . . 37
6 Results of the Simulation 39
6.1 Flying Wing, FG-WingX-02 . . . . . . . . . . . . . . . . . . . . . 39
6.1.1 Stability Results . . . . . . . . . . . . . . . . . . . . . . . 41
6.2 Optimised Stable Wing . . . . . . . . . . . . . . . . . . . . . . . 43
6.2.1 Stability Results . . . . . . . . . . . . . . . . . . . . . . . 45
6.3 Dutch Roll Mode (germ. Taumelschwingung) and Spiral Mode . 47
6.4 Unstable Wing . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47
iii
6.4.1 Geometry . . . . . . . . . . . . . . . . . . . . . . . . . . . 48
6.4.2 Static Results . . . . . . . . . . . . . . . . . . . . . . . . . 48
6.4.3 Controller . . . . . . . . . . . . . . . . . . . . . . . . . . . 49
6.4.4 Simulation results . . . . . . . . . . . . . . . . . . . . . . 50
7 Discussion 55
8 Conclusion 56
9 Future Work 57
iv
1 Introduction
An everlasting problem in aircraft design, especially in ying wing design is the
requirement of stability in pitch roll and yaw axis which is strongly coupled with
the aircraft performance and the manoeuvrability. The stability is coupled with
a loss of the aircrafts performance and vice versa. A demonstrative example
is the elevator of a classic aircraft conguration, it produces a negative lift
(excluded canard congurations).
An aircraft with an optimal performance and optimal stability is a require-
ment, which probably is never reached, it is always the challenge to nd an
optimum. A main step to improve the aircraft performance is, to integrate the
n, elevator and fuselage in one wing. No fuselage, elevator and n would be
needed, therefore weight can be saved and drag reduced. Even the civil avia-
tion recognises, that such ying wing combinations are forward looking, due to
integrating the big fuselage in the wing and transforming it into a lift creating
element.
Due to this trade-o between the stability and the performance, the question
is: What happens with the performance and the stability if the wing is unstable
and a controller garantees the stability? With a close look at ying wing model
airplanes combined with a controller, the question can be advanced to: How
good is the ight performance of a pitch unstable ying wing without taking
regard on pitch stability? Is it possible to reach better glide ratios and sink
rates if the unstable wing is guided by a pitch controller?
In order to optimise the power consumption of a new prototype of conven-
tional solar UAV airplane in higher altitude, it might be advantageous to switch
to a ying wing if a signicant dierence in performance is found.
To answer this question, a fast dynamic simulation application is built for
studying aircraft stability and performance. The simulation is based on a com-
plex aerodynamic model for complex, three-dimensional wing congurations
with twist, tapering, sweep and dihedral integrated in a six degree of freedom
rigid body motion simulation. Inuences, such as ap deections, sideslip angle
and angular rates are considered in the calculation of the aerodynamic forces
and moments. The aerodynamic force calculation method is a Vortex Lattice
method, modied for short computing time. For each time step the method
calculates a force distribution in spanwise and chordwise direction, for a three
dimensional wing.
To answer the question about the unstable wings gain of performance, a
pitch unstable wing is designed and compared with self stable ying wings.
For testing the ight characteristics and the dynamic stability, the ying wings
are simulated in the dynamic simulation. The unstable wing is stabilised by a
controller.
1
1 INTRODUCTION
1.1 Objectives
The objectives of this work are:
Literature review about existing software solutions
Establishing a Matlab/Simulink nonlinear dynamic model in order to sim-
ulate the dynamic behaviour of dierent ying wing model airplane con-
gurations
Verication of the model with ight experiments using an existing RC
ying wing model
Simulation of dierent ying wing model airplane solutions
Self stable ying wing with high ight performance
Low-sweep pitch unstable wing with high performance pitch sta-
bilised by a controller.
Comparison of the performances
1.2 Work structure
This thesis is structured in three parts. In the rst part, a theoretical back-
ground in vortex methods is given. The Nonlinear Lifting Line Method and
the Vortex Lattice Method are described in more details. Information about
existing software solutions are given. A list of existing simulations and their
capacities is provided.
The second part describes how the methods are implemented and what as-
sumptions with regard to a dynamic simulation are made.
In the third part results of the evaluation are presented and results are shown
and discussed in the section results of the simulation.
In the Appendix some important code fragments, a manual of the code and
the most important information about designing ying wings are shown.
2
2 STATE OF THE ART
2 State of the Art
In the section 2.1 it is presented what methods are commonly used for aerody-
namic force calculation. In the section 2.2 dierent existing simulators, which
try to integrate an aerodynamical model, are presented.
2.1 State of the Art of Aerodynamical Force Calculation
Today high order panel codes are commonly used. For dynamical simulation,
an unsteady Kutta-Joukovsky boundary condition is made, as well as unsteady
wake arrangements by modelling the wake with taking regard to the ight state
of earlier simulated time steps. First panel codes were developed in 1962. The
strongest eorts in these codes were made in pre- and post-processing and also
in the automatical coupling of prole information, where inner viscous eects
are separately analysed. With vortex methods, eects of propulsion and internal
ows trough turbines etc. can be modelled. In addition ow separation can be
modelled with vortex methods and give the wing a nonlinear behaviour. There
are even methods which can couple the inner viscous eects in the boundary
layer with the potential ow, so that also turbulent boundary layers can be
modelled. Detailed information can be found in [18, 8, 12, 22]
For real time simulations many simplications are commonly made ore huge
look up tables are generated from measurement or calculations. See in the
section Simulation of the dynamic.
2.2 State of the Art of Flight Simulators
There are various aircraft simulators which try to simulate the behaviour of
aircrafts. A simulation of aircraft with a suciently complex aerodynamic force
calculator for 3D wings was not found. Most simulators are interested in best
graphical visualisation, but not in physical aircraft behaviour. Some simulators
which try to include a more complex integration of aerodynamic forces are listed
below. A very advanced simulator is X-plane. Not only is the graphical engine
outstanding, but also the aerodynamic forces calculation is at a high level.
JSBSim An Open Source project. It calculates aerodynamic forces from look
up tables. Polar data are included for a range of angles from 0...180.
Propulsion and ground eects are implemented as well, and even a con-
troller for auto piloting is included. All stability derivates are modelled
linearly. [2]
Flight Gear An Open Source project. The ight gear uses ight dynamic
models from JSBsim. [15]
X-Plane A well known simulator. The user can dene his own airplane and let
it y in X-plane. X-plane tries to approximate the wing as a nite wing.
Eects as aspect ratio, taper ratio, and sweep of the wing are inuencing
the aerodynamics. Even compressible eects are implemented. A 2D
airfoil polar maker is included. [14]
Andr Noth The simulator simulates an innite wing and neglects the eects
of a nite wing on the lift distribution. For induced drag simple approx-
imations are done to integrate the eects of taper ratio and aspect ratio.
3
2 STATE OF THE ART
Xfoil is generating 2D prole coecients which are integrated over the
wingspan. A dynamic of six degrees of freedom is implemented in mat-
lab simulink. The simulator is made for controller design.
The simulators listed above are not suited to analyse dierent wing geometries
and to study stability and performance behaviour of ying wings. For this
reason a simulator is established and will be introduced in this thesis.
4
3 METHODS FOR AERODYNAMIC FORCE CALCULATION
3 Methods for Aerodynamic Force Calculation
The aim is, to design an algorithm which can calculate the aerodynamic forces
introduced by the wing. Then these forces are given into the dynamic simulation
and must be calculated at each time step.
There are many methods to calculate the forces over a wing. The program
should in principle calculate the forces for a ying wing conguration.
The specications for an implementation with regard to a dynamic simula-
tion are:
very short calculation time for real time simulation
no elevator and n
complex wing (dihedral, winglets, swept wings , etc.)
asymmetrical incident ow
rotation speed of the wing which must be regarded for damping the move-
ment about pitch, roll and yaw axis
asymmetrical ap deection
sideslip angle which must be regarded
uncomplicated implementation
consideration of the prole information
Methods which achieve these specications:
lifting line method (described in section 3.1.3)
nonlinear lifting line method (described in section 3.1.4 )
lifting surface method or even more complex panel methods (described in
section 3.1.5 )
5
3 METHODS FOR AERODYNAMIC FORCE CALCULATION
3.1 Vortex Methods
Inner eects Inner eects are eects which are acting in the boundary layer
due to viscous motion. The boundary layer is extremely small in comparison
with other dimensions. The results shown in Table 3.1 give an impression of the
boundary layers dimensions.
Table 1: Boundary layer, laminar and turbulent [19]
These inner eects can not be neglected, they comprise the information
about the viscous drag. This problem can be split up and separately be calcu-
lated, because its inuence on the outer boundary is small. So the calculation
of the viscous drag is carried out separately. However, there are many programs
which can calculate viscous drag, for example:
Xfoil
Wineppler
Outer eects Pressure forces are eects which act outside of the boundary
layer. They are calculated with vortex methods. Details are described in [8],
page 18. Examples of programs, which can calculate potential ow, are listed
below.
Tornado (MATLAB code)
AVL
XFLR
Miarex (MATLAB code)
6
3 METHODS FOR AERODYNAMIC FORCE CALCULATION
3.1.1 Theoretical Background on Vortex Methods
With the vortex methods listed above, the ow around the wing gets modelled
as a potential ow. This means that only linear aerodynamics can be studied,
eects of stall and other eects at high angles of attack are negligible. The ow
behaviour at high mach numbers is also negligible.
The process of modelling the potential ow happens with an intelligent vor-
tex arrangement, which models the ow around the wing. The mathematical
background for these methods are given in this section.
Biot Savart The induced velocity from a vortex line is calculated with formula
1. This is the velocity eld, which generates a vortex line with a constant
circulation strength.
u =

4

ds r
| r |
3
(1)
10 Drehungsbehaftete Stromungen 81
Gesetz von Biot-Savart
Bisher haben wir das Wirbelstarkefeld eines gegebenen Geschwindigkeitsfeldes u betrachtet, welches
man einfach durch Rotationsbildung erhalt. Jetzt fragen wir, welches Geschwindigkeitsfeld zu einem
gegebenen Wirbelstarkefeld gehort. Zur Vereinfachung betrachten wir den freien Raum ohne Begren-
zungen durch Wande.
Analog zum Magnetfeld eines stromdurchossenen Leiters in der Elektrodynamik ist die azimutale
Geschwindigkeit u

um einen unendlich langen geraden Wirbelfaden (Potentialwirbel) mit Zirkulation


im Abstand r gegeben durch
u

=

2r
.
Allgemeiner induziert ein der Raumkurve C folgendes Wirbelfadenst uck mit konstanter Zirkulation
das Geschwindigkeitsfeld
u(x) =

4

C
ds r
|r|
3
.
C
x
y
z
x
r
du
ds

F ur den Beitrag eines endlich langen, geraden Wirbelfadenst ucks ergibt sich damit in einem durch den
Abstand r und die Winkel
1
und
2
bestimmten Punkt die induzierte azimutale Geschwindigkeit
u

=

4r
(cos
1
cos
2
) .
1
2
!
Wirbelfadenstck
u senkrecht zur Zeichenebene
!
"
r
Beim Grenz ubergang zu einem unendlich langen geraden Wirbelfaden (
1
0,
2
) ergibt sich
daraus wieder die Geschwindigkeit um einen Potentialwirbel. Der Fall eines halbunendlichen Wirbelfa-
dens ist ebenfalls erfasst.
Ist die Wirbelstarke (x) im Raum verteilt, so erhalt man das von induzierte Geschwindigkeitsfeld
durch ein Volumenintegral, das die Beitrage aller Wirbelelemente zur Geschwindigkeit u(x) im Punkt
x aufsummiert:
Stand 30. September 2008
Figure 1: Vortex lament and the induced velocity
If the vortex lament is a line, the integral is transformed to:
u

=

4r
(cos
1
cos
2
) (2)
10 Drehungsbehaftete Stromungen 81
Gesetz von Biot-Savart
Bisher haben wir das Wirbelstarkefeld eines gegebenen Geschwindigkeitsfeldes u betrachtet, welches
man einfach durch Rotationsbildung erhalt. Jetzt fragen wir, welches Geschwindigkeitsfeld zu einem
gegebenen Wirbelstarkefeld gehort. Zur Vereinfachung betrachten wir den freien Raum ohne Begren-
zungen durch Wande.
Analog zum Magnetfeld eines stromdurchossenen Leiters in der Elektrodynamik ist die azimutale
Geschwindigkeit u

um einen unendlich langen geraden Wirbelfaden (Potentialwirbel) mit Zirkulation


im Abstand r gegeben durch
u

=

2r
.
Allgemeiner induziert ein der Raumkurve C folgendes Wirbelfadenst uck mit konstanter Zirkulation
das Geschwindigkeitsfeld
u(x) =

4

C
ds r
|r|
3
.
C
x
y
z
x
r
du
ds

F ur den Beitrag eines endlich langen, geraden Wirbelfadenst ucks ergibt sich damit in einem durch den
Abstand r und die Winkel
1
und
2
bestimmten Punkt die induzierte azimutale Geschwindigkeit
u

=

4r
(cos
1
cos
2
) .
1
2
!
Wirbelfadenstck
u senkrecht zur Zeichenebene
!
"
r
Beim Grenz ubergang zu einem unendlich langen geraden Wirbelfaden (
1
0,
2
) ergibt sich
daraus wieder die Geschwindigkeit um einen Potentialwirbel. Der Fall eines halbunendlichen Wirbelfa-
dens ist ebenfalls erfasst.
Ist die Wirbelstarke (x) im Raum verteilt, so erhalt man das von induzierte Geschwindigkeitsfeld
durch ein Volumenintegral, das die Beitrage aller Wirbelelemente zur Geschwindigkeit u(x) im Punkt
x aufsummiert:
Stand 30. September 2008
Figure 2: Vortex lament as a line
If the vortex lament is innitely long, the formula is:
u

=

4r
(3)
Force on a vortex line According to the Kutta-Joukovsky theorem, a
vortex

moving with the velocity v experiences a force

F.

F = (

) l (4)
7
3 METHODS FOR AERODYNAMIC FORCE CALCULATION
3.1.2 Lifting Line Method and Nonlinear Lifting Line Method in
General
Vortex arrangement / Singularity Element The ow is modelled as a
potential ow. As singularity element, for the Lifting Line Method as well as
for the Nonlinear Lifting Line Method, a horseshoe vortex is chosen.
The singularity element is placed, in order that the side vortex lines are
leaving the wing in ow direction, and the bound vortex is laying on the wing in
the direction of the leading edge. Assuming that the angle are small the trailing
vortices can be laid in the x-y plane of the body co-ordinate system.
A better physical arrangement is, if the trailing vortices are leaving the wing
in ow direction. These dierent arrangements are shown in gure 5. The
trailing vortices in the wake (behind the wing) have to be aligned in x direction
or better in ow direction, because in this case there will not be any force acting
on the trailing vortices.

Wake
= 0 (5)
The arrangement is shown in Figure 3.1.2.
Figure 3: Horseshoe vortex arrangement
a) with elementary wings
b) Prandtl method
gure copied from [24], page 24
The bound vortex is laid on the 1/4 line of the wing, and the collocation
points are lying on the 3/4 line of the wing in the middle of the trailing vortices.
On the collocation points the no slip condition is made. The detailed vortex
arrangement of the Lifting Line method is shown in Figure 4.
8
3 METHODS FOR AERODYNAMIC FORCE CALCULATION
Figure 4: Vortex arrangement and normal vectors in the collocation points with
the starting vortex in innity, copied from[8]
Figure 5: Trailing vortex arrangement
left: trailing vortex leaving the wing in ow direction but following the prole
right: trailing vortex leaving the wing in ow direction
Copied from [8]
Vortex Theorems A vortex is always closed. This means that vortices are
closed laments, or vortex rings. The horseshoe vortex is also closed, with the
starting vortex in innity. Thus its inuence is negligible. A second important
assumption is, that the circulation strength along the vortex ring is constant.
Forces The lift force of a single element is calculated (Kutta Joukovsky
Theorem):
9
3 METHODS FOR AERODYNAMIC FORCE CALCULATION
Lift(i) =

(i) (i) y
V

speed of the ow

density of air (1.225 kg/m


3
)
y width of a horseshoe vortex
(i) circulation strength of element i
The lift force is generated by the innite ow speed on the bound vortex. The
lift force is aligned vertically to the ow speed (see formula 4 on page 7).
The induced drag is calculated with the induced ow speed of the trailing
vortices. They generate a downwash, which induces a velocity on the bound
vortex. The induced velocity on the bound vortex generates a force. This force
is the induced drag. Only the trailing vortices have an inuence on the induced
drag.
Drag
induced
=

V
ind
(i) (i) y
The forces are placed on the 1/4 line on each bound vortex.
Tretz Analysis The induced drag can also be calculated in the so called
Tretz plane, far behind the wing. The induced velocity is much easier to
calculate in the Tretz plane than over the wing, because the trailing vortices
can be modelled as innitely long in both directions.
10
3 METHODS FOR AERODYNAMIC FORCE CALCULATION
3.1.3 Lifting Line Method
Boundary Condition In the classic Lifting Line Method, the circulation
strength is calculated with a no slip condition. No ow can penetrate the prole.
The sum of induced velocity, innite velocity and induced velocity of the wake,
projected on normal direction, is zero. This equation is valid at the collocation
points.
w
induced
n
solid
+ w
i
n
solid
+

V

n
solid
= 0 (6)
w
induced
velocity of the bound vortex
w
i
induced velocity of the wake (trailing vortices)

incident ow velocity
The normal vectors of the wing n
solid
are vertical to the chord line, so the
boundary condition takes only the chord as prole information, cambering is
negligible. To have better prole information in the boundary condition, the
following methods integrate the prole better.
Equation 6 gives for each collocation point a linear equation with the un-
known circulation strength (i). The enormous advantage of the vortex method
is, that the circulation is linear in equation 1, 2 and 3. So equation 6 can be
written as a matrix and vector equation:
K = RHS (7)
K nxn matrix
vector with length n
RHS incident ow (

V

) projected on normal vector n


In order to solve this equation for the circulation vector , only a matrix inver-
sion has to be done.
Closed analytical solutions of equation 6 are given in [24], page 7-10. The
numerical solution of this equation is described in section Implementation.
It is more or less arbitrary to evaluate this equation in a point, which is
laying on the 3/4 line. However this alignment is commonly used and provides
good results. It might be useful to analyse, where that point has to lie with
dierent proles. A sample calculation, why the collocation point is laying on
the 3/4 line is shown in [12], page 23
11
3 METHODS FOR AERODYNAMIC FORCE CALCULATION
3.1.4 Nonlinear Lifting Line Method
The Nonlinear Lifting Line Method was developed in 1946, see [25].
Circulation Strength In Weissingers Nonlinear Lifting Line Method, the
circulation strength is calculated by iteration. At rst the induced angle must
be calculated:

ind
= arctan(V
ind
/V

) (8)
The downwash is calculated on the collocation points. See Figure 6 depicting
a geometrical interpretation of equation 8.
Figure 6: Induced angle and angle of attack
V
induced angle
effective angle
geometrical angle
V
Vind
induced angle
Prole data With the induced angle, the angle of attack can be calculated:

effective
=
geometric

induced
(9)
With the eective angle, the local lift force can be calculated. This is done
with known airfoil polars (measurements, Xfoil, Wineppler). With this informa-
tion, nonlinear prole behaviour can be coupled with the circulation distribution
and the lift force. Especially at high angles of attack this information is essen-
tial, as the prole then shows a highly nonlinear behaviour and can not be
modelled as a perfect at plate with a linear behaviour.
Iterative Process
1. An initial circulation distribution is estimated.
2. The induced angle of attack
ind
is calculated.
3. The local angle of attack (angle between innite velocity and chord minus
the induced angle) is calculated.
4. The lift distribution with known airfoil polar data is calculated.
5. With the Kutta Joukovsky theorem, a new circulation distribution can be
calculated.
6. The old and the new circulation distributions are compared. A new circu-
lation distribution with both of them is iteratively generated. The iterative
process begins with point 1.
12
3 METHODS FOR AERODYNAMIC FORCE CALCULATION
The new circulation is calculated with the following formula: [18], page 277

n+1
(i) = (1 D)
n
(i) +Dnew(i) (10)
D is the damping factor, D 0.05. The iteration is made, until the maximal
dierence between new and old circulation is small enough.
References: [8, 18, 17]
Advantages, Disadvantages and Limitations of the Nonlinear Lifting
Line Method
It is important to know the limitations of the Nonlinear Lifting Line Method.
The main advantage is the inclusion of the nonlinear prol data. Even experi-
mental prol data can be coupled with the potential ow.
Advantages
Only few calculations are necessary.
Nonlinear eects of the prole can be studied.
Flap deections are not integrated in the K matrix.
Disadvantages
Momentum distribution: The fact is, that the lift force is not always placed
in the quarter of the chord, as it is for non swept wings (the sweep angle
is measured at the chord quarter line). For a ying wing the momentum
equilibrium is essential. It has an inuence on the centre of pressure, thus
on the positioning of the centre of gravity. It is of great importance to
know the centre of gravity, not only for the construction, but also for
the mass of stability. The measure of stability is dened as the distance
between the centre of gravity and the neutral point of the aircraft
1
. This
means that the prole data of C
M
is not exactly the value which can be
extracted from the local angle of attack. This problem can only be solved
with a panel method with several panels in the chord direction, see section
3.1.5.
For the lift calculation, all prole data is needed. So a look up table of
prole data is necessary, which is also needed for viscous drag calculation.
The wake is modelled in a simple way.
Limitations
According to NACA technical note [26], The calculations are subject to
the limitations of lifting line theory and should not be expected to give
accurate results for wings of low aspect ratio and large amounts of sweep
1
There are some dierent denitions of the neutral point, there is a geometrical and an
aerodynamical neutral point, mostly they are nearly the same points. Commonly the measure
of stability is given in percent of the mean aerodynamical chord length. More details of
geometrical and aerodynamical neutral points see [19], page 104
13
3 METHODS FOR AERODYNAMIC FORCE CALCULATION
3.1.5 Vortex Lattice Method
The Vortex Lattice Method, implemented in this thesis, is a Lifting Surface
Method. Therefore the theoretical information which is given in this section is
limited to the Lifting Surface Method. The method is similar to the Lifting Line
Method, it is only enriched with more singularity elements in chord direction.
The boundary condition includes prole information, so the cambering has
an inuence on the lift coecient. However, the inuence of the whole prole
is not as strong as in the Nonlinear Lifting Line Method. Viscous drag can not
be calculated with these methods either, this is done with experimental prole
polars or with software solutions.
More complex panel methods are mathematically similar to the low order
panel method, but there are other boundary conditions and other singularity
elements. They use a Dirichlet boundary condition for a thick body, and for
modelling the potential ow they use doublet panels (vortex rings) or constant-
strength sources.
The method, implemented in this thesis, is similar to the method presented
in [6], this method is also known as A Multi-lifting Line Method and its Ap-
plication on Design and Analysis of Nonplanar Wing Conguration.
Singularity element Basically a vortex ring or a combination of a ring
and a horseshoe vortex may be chosen as a singularity element. The choice is
depending on the eects which would be studied and on the provided calculation
time. For the Vortex Lattice Method implemented in this thesis, the horseshoe
vortex is chosen as singularity element.
Mesh and Vortex Arrangement The wing is divided into several el-
ements in wingspan direction as well as in chord direction. The result is a
rectangular mesh over the whole wing surface. In each of these rectangular
panels a horseshoe vortex is laid. The arrangement is shown in Figure 7. It is
important for the trailing vortices, that they are laid on the surface of the wing.
If they are not laid on the surface, eects such as twist do not have a suciently
strong inuence. The following Figures 8 and 9 show more geometrical details
of the horseshoe vortex placement.
14
3 METHODS FOR AERODYNAMIC FORCE CALCULATION
Collocation Point
Trailing
Vortex
Bound Vortex
c/2
Wake
Figure 7: Vortex arrangement of the Vortex Lattice Method
Figure 8: Detailed horseshoe vortex and alignment in a panel, copied from [8]
15
3 METHODS FOR AERODYNAMIC FORCE CALCULATION
Tomas Melin. KTH, Department of Aeronautics. Page 22 (45)
A Vortex Lattice MATLAB Implementation for Linear Aerodynamic Wing Applications.
4.3.2 Twist and the skewed vortex loop.
When adding twist to the layout, it implies that the geometric angle of attack
varies with span, the design is no longer a flat plate but a mildly skewed surface.
The twist will cause the two outgoing vortex legs from a panel are no longer
parallel, see figure 11. This is the source of the vortex-sling arrangement in
Tornado, which is used instead of the more commonplace horseshoe vortex.
4.3.3 Camber and thin airfoil boundary application.
To extend the geometry even more, the wing could also be cambered. In
Tornado the wing is still regarded as flat with a thin wing approximation where
the boundary conditions are shifted. That is, the normal of the cambered surface
is calculated and the non-flow-through boundary condition is employed at the
chord line (see figure 12). This approximation is common and used in a variety
of methods.
Fig 11: Twisted vortex sling. Figure 9: Trailing vortices and twist, the trailing vortices are not parallel, copied
from [13]
Wake The wake modelling in this simple model does not take the wake
roll up and the unsteady ow into account. For more precise results, the wake
would be modelled by vortex panels with time variant circulation strength. The
wake in this implementation contains all trailing vortices which are following
the local chord direction.
Boundary Condition and Prole Information To take care of prole
information, the boundary condition must be improved. The boundary condi-
tion is made on the skeleton line, so that the cambering has an inuence on
the circulation distribution. So the normal direction on the skeleton line is cal-
culated, and in this direction the total ow must be zero. This simply means,
that no ow can go through the skeleton line. This is an approximation of the
prole, it is assumed that the prole is thin (see Figure 10). This approximation
is commonly used. See in [6, 18, 8, 12, 21, 11]
Tomas Melin. KTH, Department of Aeronautics. Page 23 (45)
A Vortex Lattice MATLAB Implementation for Linear Aerodynamic Wing Applications.
4.3.4 The polyhedral wing.
The geometry may be even more intricate when we allow cranked wings, i.e.
wings that are polyhedral, like the F-16 main wing, see figure 13. However from
the geometric layout and meshing point of view, this is not a big problem as
every polyhedral wing may be broken down into quadrilateral partitions. In
Tornado, this partitioning takes place early in the user input of geometry
definitions.
Fig 12: Camber and shifted boundary condition.
Fig 13: Cranked wing on a F16 type of aircraft.
Figure 10: Inuence of the cambering in the boundary condition.
The Figure is copied from [13]
System of Equations The resulting equation for the calculation of the
circulation strength in a collocation point is the same as equation 6. This
equation is solved for each panel and the resulting system of equations can be
16
3 METHODS FOR AERODYNAMIC FORCE CALCULATION
solved for the circulation strength in each panel. The system of equations can
be written in a matrix form, similar to equation 7. To solve the system of
equations, only a matrix inversion must be done.
Advantages and Disadvantages of the VL-Method in general
Advantages
A force distribution over the chord and the span is made. This provides
a better momentum distribution.
The centre of gravity can be calculated more exactly.
Disadvantages
The wake arrangement with innite trailing vortices is an enormous sim-
plication.
The Kutta condition is a steady Kutta condition and neglects dynamic
ow behaviour.
Eects of ow separation and transition are neglected in the potential
ow, but they are not negligible in the airfoil viscous drag.
17
4 IMPLEMENTATION OF A SUITABLE METHOD
4 Implementation of a Suitable Method
There are various existing programs, which can calculate aerodynamic forces.
Tornado (MATLAB code) [21]
AVL [11]
XFLR [10]
Miarex (MATLAB code) [23]
The best results can be obtained with panel methods. There is an Open Source
tool called Tornado [21]. It was tested, if this software could be used for a
dynamic simulation. Unfortunately the calculation times are too long and a
vast look up table would be necessary, because the program has to calculate a
new mesh and a new inuence matrix for each ap deection.
There is another Open Source tool from Mark Drela [11], which is a Vortex
Lattice Method as well, but the calculation time is also too long, similarly to
Tornado. Flap deections must always be meshed again.
Another problem is, that these two programs do not care about the friction
of the prole. Therefore the friction force is only the induced drag. For the
whole drag force it is necessary to integrate the lift distribution with prole
data. This has to be calculated separately and again increases the calculation
time.
XFLR is another program with great potential which calculates with dier-
ent methods. Often comparisons with XFLR are made in this thesis. Integrating
XFLR in a dynamic simulation would not pe possible either, the geometry for
each ap deection would have to be changed. In XFLR ap deections are
dened as a new prole. Therefore for each change of the prole, the prole
coecients have to be recalculated for the interpolation. Miarex is a kind of
Nonlinear Lifting Line Method combined with Xfoil, but not ecient enough
either for dynamical solutions. Miarex is also limited in the wing geometries.
Some requirements, which do not allow to generate a look up table, are:
asymmetrical ap deections
sideslip angles
angular rates for damping the movement in yaw-, pitch- and roll-axes
These variables generate too many ight states.
After checking the qualities and limitations of the dierent programs, a Lift-
ing Line Method was chosen, which has the great advantage that ap deections
are not cared about in the inuence matrix of the vortices. The ap deections
are only considered in the iteration. A reduced look up table would be gener-
ated with the only parameter , the sideslip angle. Unfortunately the results
were insucient, because the method does not care about a force distribution
in chord direction.
Finally a Vortex Lattice Method was implemented. The method was opti-
mised for a fast calculation, so that the method can be used for dynamical simu-
lation. Some simplications with ap deections were made. In both methods,
simplications are made in the wake modelling. Otherwise, the simulation time
would be much longer.
18
4 IMPLEMENTATION OF A SUITABLE METHOD
4.1 Nonlinear Lifting Line Method
The main idea of this implementation is, to split up the calculations. The
meaning is to calculate as much as possible before the simulation starts so that
during the simulation only the most important calculations must be carried out.
A simple structure of the program is shown in Figure 11, listing the most
important functions and procedures.
load all input
data
geometry
denition
initial state
denition
prole polar
denition
Xfoil
polars
generate
geometry and
vortex mesh
calculate
inuence
coefcient
Nonlinear Lifting Line Method
Process Diagramm
make initial
guess for
gamma start
distribution
iterate
calculate
forces and
moments
geometry
mesh
inuence
matrix K
gamma start and
ap deection
gamma
Manual input
Figure 11:
The following chapter gives more detailed information about the most im-
portant functions.
4.1.1 Inputs
The geometry can be a 3D wing geometry with aps, dihedral, twist and sweep
angle. Even discontinuous functions of the twist angle and the geometry are
possible. As inow information, the ow velocity must be dened in three
components, and the angular rates have to be initialised.
4.1.2 Mesh Generation
The coordinate system is dened as commonly used in aircraft design. The
y-axis is in the spanwise direction, the x-axis goes backwards of the wing and
19
4 IMPLEMENTATION OF A SUITABLE METHOD
the z-axis is looking away from the earth. This denition is dierent than the
denition of the dynamics.
As singularity element, the horseshoe vortex is chosen. The trailing vortices
are aligned in x-direction in this implementation .
The wing geometry can be devided into several partitions. Each partition is
a trapezoid with geometry data listed below. The geometry of the wing is only
dened for a wing half, then the geometry is mirrored.
b_root root chord, skalar
b_tip tip chord skalar
alpha_g geometrical angle to the body x-axis, this is a vector and contains
the root angle and the tip angle
s span of the partition
n number of single panels
phi sweep angle at leading edge
x0 reference point, where the partition has to be placed, it is dened at
the tip of the root chord for each partition and therefore is a matrix.
With this parameters the function generate_Mesh.m gives as output the mesh
data for the vortex placements and some other geometry data, which are used
for the force calculation. The most important outputs are:
T_left all left horseshoe points
T_right all right horseshoe points
A all collocation points
The trailing vortices are leaving the wing in x-direction. This is a small angle
approximation of the wake. Physically they have to leave the wing in the ow
direction behind the wing. Therefore twist eects are neglected in the vortex
arrangement. The eect of the wings geometrical angles are taken into account
in the calculation of the new Gamma distribution (), see Algorithm 1 in Ap-
pendix A . The new Gamma circulation is calculated with the geometrical and
the induced angle and then damped with the old circulation.
4.1.3 Sideslip Angle
The whole geometry is rotated with an angle around the z-axis. The sideslip
angle is calculated:
1 betha_in=atan(global_flow_speed (2)/global_flow_speed (1));
However, the trailing vortices are still pointing into the x-direction.
20
4 IMPLEMENTATION OF A SUITABLE METHOD
4.1.4 Inuence Coecients
The Kutta condition is fullled at each collocation point. The induced velocities
of all horseshoe vortices must be calculated at the collocation point i, and then be
compared with the incident ow V

projected on the local geometrical normal


vector. For all collocation points the system has the following form:
Coll. Points Matrix K Gamma Boundary Cond.
1 a
1,1
a
1,2
a
1,3
... a
1,N

1
RHS
1
2 a
2,1
a
2,2
a
2,3
... a
2,N

2
RHS
2
3 a
3,1
a
3,2
a
3,3
... a
3,N

3
= RHS
3
4 a
4,1
a
4,2
a
4,3
... a
4,N

4
RHS
4
... ... ... ... ... ... ... ...
N ... ... ... ... ...
N
RHS
N
Table 2: System of linear equations of the Nonlinear Lifting Line Method
The inuence coecients a
i,j
are calculated with the function vortex.m, vor-
tex_left.m, vortex_right.m. They are principally vectors, projected on each
normal vector in the collocation point. The function is plotted in Algorithm 2
Appendix A .
The circulation Gamma can be excluded from the calculation of the inuence
coecients, because it is linear in the equation.
The right hand side of the equation is the free stream ow projected on the
normal vector.
RHS
i
= (U

, V

, W

) n
i
(11)
The inuence coecients are only depending on the geometry and the sideslip
angle . The assumption is, that the wake is stationary, so that the vortex ar-
rangement, despite a variation in the angle of attack, rests always the same.
This inuence matrix is saved for several sideslip angles and later interpo-
lated at the desired sideslip angle.
4.1.5 Initial Guess
The initial function for the circulation strength is chosen in this implementation
as:
1 Gamma_start=-inv(K)*RHS
This is a linear approach which later is iteratet with nonlinear prole infor-
maion.
The RHS is:
1 RHS=v_abs.*sin(alpha_g)
v_abs norm of the velocity, v_abs is a vector
alpha_g geometrical angle, a vector
The twist eect is neglected in the RHS, because the inuence of twist on the
normal vector in the collocation points is neglibible.
21
4 IMPLEMENTATION OF A SUITABLE METHOD
4.1.6 Iterative Process / Coupling the Prole Information
The iteration is started with the following commonly used parameters.
1 %Damping Factor
D=0.05;
3 %Stop criteria
min =0.001;
The iteration function is the core of the total program, it is plotted in Algo-
rithm 1 on page 60 Appendix A .
A short description of what the function does:
The induced velocity w_i here is calculated with the Tretz analysis. It is
calculated with the following formula:
w_i =1/2*( K_far*Gamma_alt);
K_far is the inuence matrix for the Tretz analysis. The eective angle of
attack is calculated:
1 alpha_i=-atan(w_i ./(v_abs ));
The induced angle alpha_i could also be simplied to: alpha_i=-w_i./v_abs.
The new circulation Gamma_new is calculated with the following two for-
mulas:
Kutta-Joukovsky
L = v y (12)
Prole_CL
L = CL
v
2
2
y b (13)
Therefore the circulation Gamma is:
= CL
v
2
b (14)
The CL value is calculated with an interpolation function between several
polars. In the interpolation, the ap angle and the angle of attack are the input
values. With the CL distribution, the viscous drag CD is interpolated from the
prole polars.
The convergence process of the iteration is not a robust process. So the
damping factor and the stop criteria have to be chosen carefully. The practice
has shown, that the values given here mostly provide good results.
4.1.7 Force and Moment Calculation
The lift force is calculated with the CL distribution. It is projected on the
speed normal direction, vertical to the ow speed vector and vertical to the
bound vortex. The induced drag force is calculated with the formula:
D
induced
(i) = (i) w
induced
y (15)
The viscous drag is calculated with the integration of the local viscous drag
over the span of the wing which is calculated with prole polar information.
The drag force is projected in ow speed direction. The point of attack of the
22
4 IMPLEMENTATION OF A SUITABLE METHOD
forces is on the bound vortex. The resulting moment is evaluated in the origin
of the wing. To the resulting moment the local momentum coecient of the
prole integrated over the span of the wing is added.
4.1.8 Evaluation of the Nonlinenar Lifting Line Method
Unfortunately in the Tretz analysis sweep angles are neglected. So a variation
of the sweep angle does not change the induced velocities in the Tretz plane.
For swept wings, it is important to include the inuence of the sweep angle, so
a solution is searched to include the sweep angle. This is done by calculating
the induced velocitities on the bound vortex. The Tretz plane only recognises
the innite vortex lines, the dierence of the starting points x-component is
neglected because it is too far away. On the bound vortex, the inuence of
dierent x-components of the trailing vortex starting point is considered. This
causes an inuence on the sweep. The implementation results showed, that the
inuence is too strong and gives a qualitatively incorrect distribution of lift.
To illustrate the eect of Tretz and bound analysis, a Figure with the
dierence of the two analysis types is added, see Figure 12. The wing is a swept
wing without dihedral. The geometry denition is listed below.
Wing denition of the test wing:
1 p=3; %number of Partitions
ny=[0,0,0]*pi /180; %Dihedral of the Partitions
3
n=[10 ,10 ,7]; %Number of collocation points
5 s=[0.5;0.3;0.2]; %Span of partition
b_root =[0.2;0.2;0.2]; %root chord
7 b_tip =[0.2;0.2;0.2]; %tip chord
phi =[20 ,20 ,20]*pi /180; %Sweep angle , back is
positive
9
%geometrical twist angles
11 alpha =[0 -2 -2 -3 -3 -3.8]*pi /180;
!1 !0.8 !0.6 !0.4 !0.2 0 0.2 0.4 0.6 0.8 1
0.1
0.2
0.3
0.4
0.5
0.6
y
G
a
m
m
a


!1 !0.8 !0.6 !0.4 !0.2 0 0.2 0.4 0.6 0.8 1
0.1
0.2
0.3
0.4
0.5
0.6
y
C
L


Gamma
Gamma with Trefftz Analysis
CL Distribution
CL Distribution with Trefftz Analysis
Figure 12: Dierences between alternative calculations of the induced angle
Angle of attack: 6 degrees
Calculations made with Nonlinear Lifting Line Method Gohl
The test wing from above is compared with other results. Here compared
are the CL distributions, the circulation distribution curve is qualitatively the
23
4 IMPLEMENTATION OF A SUITABLE METHOD
same because the chord is constant and therefore not plotted, see Figure 13.
!1 !0.8 !0.6 !0.4 !0.2 0 0.2 0.4 0.6 0.8 1
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
Y
C
L


Nonlinear Lifting Line Method, Trefftz analysis Gohl
Nonlinear Lifting Line Method XFLR
Panel Method XFLR
Nonlinear Lifting Line Method, bound analysis Gohl
Figure 13: Comparison of the Lifting Line Method with a Panel calculation for
the swept wing with twist
angle of attack: 5
Advantages of the Nonlinear Lifting Line Implementation
Nonlinear behaviour of prole information inuences the circulation dis-
tribution.
Flap deections do not need to be regarded in mesh generation. The ap
inuence is regarded in the iterative function where the local lift coecient
is interpolated.
The calculation is very fast.
Disadvantages of the Nonlinear Lifting Line Implementation
The twist eect is not modelled well, because all trailing vortices are leav-
ing the wing and not following the local chord direction.
This method does not provide accurate results for wings with low aspect
ratios.
The results of swept wings are not usable, because sweep angle eects are
neglected.
An interpolation between Reynolds numbers is not done in this imple-
mentation. When prole polars are made, the Reynolds number must be
estimated.
24
4 IMPLEMENTATION OF A SUITABLE METHOD
4.2 Vortex Lattice Method
The Vortex Lattice Method is implemented as a consequence of the unsatisfac-
tory results provided by the Nonlinear Lifting Line Method and the limitations
of the method (see section 3.1.4 on page 12 and section 3.1.4 on page 13).
The Vortex Lattice implementation is similar to the Nonlinear Lifting Line
Method. However there are some signicant changes in the mesh composition
and in the structure of the linear system of equation, and there are other mod-
ications, which are shown in this section. The basic problem of the Vortex
Lattice method is, that the dimension of the inuence matrix is much higher.
If a vortex lattice method should be implemented in a dynamic simulation, the
calculation time is a severe problem. To reduce the computing time, some im-
portant modications in saving the data and in calculating the inuence matrix
are done. The most important steps in the program are shown in a ow diagram,
see Figure 14
load all input
data
geometry
denition
inow
information
prole polar
denition
Xfoil
polars
generate
geometry and
vortex mesh
calculate
inuence
coefcient
Vortex Lattice Method
Process Diagramm
generate RHS,
insert ap deection
inverting Matrix K
and calculate
Gamma
calculate
forces and
moments
geometry
mesh
Inuence
matrix K
Gamma
Manual input
Figure 14: Process diagram of the Vortex Lattice Implementation
25
4 IMPLEMENTATION OF A SUITABLE METHOD
4.2.1 Inclusion of the Prole
The prole coordinates can directly be included in the implementation. Then
the mean line is calculated from the prole data, as an example see the prole
MH45 in Figure 15:
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
!0.3
!0.2
!0.1
0
0.1
0.2
0.3
x!axis
z
!
a
x
i
s


Figure 15: Mean line of the prole MH45 calculated with prole.m
4.2.2 Mesh Generating
The wing is divided into rectangular elements in spanwise direction and in chord
direction. In each of the panel a horseshoe vortex is laid. The trailing vortices in
this implementation are following the chord direction, otherwise the twist eect
would be neglected. The arrangement in top view is shown in the theoretical
part (see Figure 7 on page 15). For details about dierent mesh arrangements
and results, see in section 4.2.8 on page 31. A detailed description of a panel is
shown in Figure 16
T_right
T_left
P
bound
vortex Trailing
vortex
!
!
!
Panel
V"
r1
r2
n
T_left
behind
T_right
behind
P
Figure 16: Detailed description of a single panel
26
4 IMPLEMENTATION OF A SUITABLE METHOD
The mesh points are saved in matrices:
T
left
X =
T
left
X
1,1
T
left
X
1,2
T
left
X
1,3
...
T
left
X
2,1
T
left
X
2,2
... ...
T
left
X
3,1
... ... ...
... ... ... T
left
X
m,n
The rst index is the number of the element in chord direction, the second
indices is the number of element in spanwise direction. The same is applied for
all other coordinates and points. The organisation of the elements is shown in
the following Figure:
1,1
1,2
1,3
2,1
3,1
3,2
2,2
2,3
2,4
3,3
3,4
3,5
. . .
. . .
. . .
Figure 17: Organisation of the indices of the mesh points
With this matrix structure, MATLAB can calculate faster, because the
basic vector calculations can be calculated much faster.
The normal vector is now calculated at the mean line of the prole, see
Figure 16. It is calculated with the cross product:
r
1
r
2
r
1
r
2

=

n (16)
For the calculation of the normal direction, the collocation point must lie on
the prole mean line.
4.2.3 Sideslip Angle
If the ow has a y-component, the mesh must be changed. In this implemen-
tation, the trailing vortices only are rotated around the z-axis. The inuence
coecients are saved for a range of sideslip angles and then the desired sideslip
angle is interpolated between this data.
27
4 IMPLEMENTATION OF A SUITABLE METHOD
4.2.4 Inuence Coecients
The inuence coecients are organised as the following Table shows:
Coll. Points Matrix K
p
1,1
: a
1,1
a
1,2
a
1,3
... a
2,1
a
2,2
a
2,3
...
1,1
RHS
1,1
p
2,1
: a
1,1
a
1,2
a
1,3
... a
2,1
a
2,2
a
2,3
...
1,2
RHS
2,1
p
3,1
: a
1,1
a
1,2
a
1,3
... a
2,1
a
2,2
a
2,3
...
1,3
RHS
3,1
... a
1,1
a
1,2
a
1,3
... a
2,1
a
2,2
a
2,3
... ... RHS
1,2
p
1,2
: ... ... ... ... ... ... ... ...
2,1
= RHS
2,2
p
2,2
: ... ... ... ... ... ... ... ...
2,2
RHS
3,2
p
3,2
: ... ... ... ... ... ... ... ...
2,3
RHS
1,3
... ... ... ... ... ... ... ... ... ... RHS
2,3
p
1,3
: ... ... ... ... ... ... ... ... ... RHS
3,3
... ... ... ... ... ... ... ... ... ... ...
Table 3: System of linear equations
The boundary condition is evaluated at each collocation point, like for ex-
ample, the equation for the collocation point 1,1:
a
1,1

1,1
+a
1,2

1,2
+a
1,3

1,3
+...+a
2,1

2,1
+...+a
m,n

m,n
=

n (17)
Attention, many errors can occur if the numbering of the circulation vec-
tor and the numbering of the RHS vector are exchanged. There are writ-
ten functions, which can rewrite the numbering of these vectors (rewrite.m,
rewrite_RHS.m).
The calculation of inuence coecients is basically similar to the calcula-
tion in the nonlinear implementation. The principle to calculate the inuence
coecients of a vortex line is described below.
The basic formula to calculate the induced velocity of a straight vortex
lament is given in the theoretical part. So the cosines of the angles between
R_0 and R_1, R_1 and R_0 must be found. These angles are calculated with
the help of the dot product. The geometrical illustration is shown in Figure 18.
cos() =
R0 R1
R0 R1
(18)
The distance r is calculated with the help of the cross product:
r =
R0 R1
R0
(19)
The resulting induced velocity is calculated with the principle of equation
18 and 19:
a =
R1 R2
R1 R2
2
1
4
(
R0 R1
R1

R0 R1
R2
) (20)
The value R0 in equation 20 can be cancelled.
28
4 IMPLEMENTATION OF A SUITABLE METHOD
!
L(x,y,z)
R(x,y,z)
P(x,y,z)
R_0
R_1
R_2
r
Figure 18: Vortex inuence coecient
A numerical problem in calculating the inuence coecients can occur, if a
collocation point lies too close to the vortex line. The induced velocity then rises
to innite. To avoid this problem, a region is dened in which the velocities are
set to zero. The problem occurs if:
R1 < (21)
R2 < (22)
R1 R2
2
< (23)
If one of these three equations is fullled, then the value of a is set to zero.
So the equation 20 for all values of the denumerators is always dened.
The idea is, to calculate the induced velocity from all bound vortices, left
trailing vortices and right trailing vortices for one collocation point i at once.
This is much faster. All the operations of the equations 18, 19 and 20 can
be calculated with matrices. The function vortex_panel.m needs as input all
points in matrix structures and gives out the coecients. The structure of a
(see equation 20) is therefore a matrix. The indices are organised in the way,
that the coecient a
m,n
is the inuence of the panel m,n on the collocation
point i. These coecients are directly rewritten as vectors, so that they are in
the structure of a row of the inuence matrix, see the system of equation in
Table 3
The function is added in Algorithm 3, Appendix A
The calculation of the inuence matrix for one collocation point is made in
a loop for each collocation point. See the function in Algorithm 4, Appendix A
.
29
4 IMPLEMENTATION OF A SUITABLE METHOD
For lift forces, the calculation of the inuence coecients is done with the
collocation points on the 3/4 line of the panel. For induced drag forces, the
calculation of inuence coecients is done on collocation points lying on the
bound vortex. Then the K matrix is saved as the inverted matrix, so that
during the simulation no matrix inversion is needed.
4.2.5 Flap deections
Flap deections in this implementation are made with the boundary condition.
All surface normals which lie in the ap region are rotated around the y-axis
with the angle of the ap deection.
4.2.6 Force and Moment Calculation
The lift forces act on the bound vortices. They are calculated for each panel:
Liftforce =
n

i=1
m

j=1

j,i
y
j,i
(24)
The induced drag is calculated with the following formula:
Drag
induced
=
n

i=1
m

j=1
sign(
j,i
)
j,i
w
indj,i
y
j,i
(25)
The term w
ind
is the induced velocity from the trailing vortices. It is calcu-
lated with the following formula
w
ind
= K
bound
(26)
K
bound
is the inuence matrix of the trailing vortices on the bound vortices.
It is calculated with the following formula:
K
bound
= K
bound
X U
normalvel
+K
bound
Y V
normalvel
+K
bound
Z W
normalvel
(27)
K
bound
X, K
bound
Y , K
bound
Z are the inuence coecient matrices in the
directions X, Y, Z.
U
normalvel
, V
normalvel
, W
normalvel
are the components of the vectors ver-
tical to the ow speed and the bound vortex vector.
The viscous drag is interpolated with the prole polars.
30
4 IMPLEMENTATION OF A SUITABLE METHOD
4.2.7 Stall
In the simulation high eective angles might occur. It could therefore be, that
CL values of 2.0 or more appear (as a consequence of the linear behaviour of the
method). Values of CL over 1.5 for normal proles do not exist physically. In
this case the interpolation function for calculating the CD values has a problem,
because in the polar table no CD value at this too high CL value exists. In order
to enable the simulation to calculate stall situations, CL values which are not
found in the table are assumed as the last element in the table. The results of
this method are not signigicant, therefore the Vortex Lattice Method does not
give accurate results for stall behaviour.
4.2.8 Evaluation of the Vortex Lattice Method
Advantages
Any desired wing geometry with sweep, dihedral, twist, winglets and sev-
eral wing parts can be calculated. Even elevator and n can be added to
the geometry, but the mesh must be looked at carefully due to singulari-
ties.
Airfoil cambering is taken in account.
Disadvantages
Non-linearities of the airs viscousity and dependencies of speed are ne-
glected in the lift force and the induced drag force.
In this implementation a strong wake simplication is made, see in sec-
tion 4.2.8.
An interpolation between Reynolds numbers is not done. Inuences of
speed and the chord distribution of the wing are approximated to the
same Reynolds numbers.
Inuence of Dierent Mesh Types
In this part the dierences between a at mesh and a mesh on skeleton line of the
wing are shown. It elucidates again the simplication of the mesh arrangement
of the implemented method. A comparison of dierent mesh types is done and
analysed qualitatively and quantitatively.
The boundary conditions are still evaluated at the skeleton line. The great
dierence is in direction of the trailing vortex. It follows the skeleton direction
or the chord direction.
31
4 IMPLEMENTATION OF A SUITABLE METHOD
0 5 10 15
!10
0
10
20
30
CL/CD
angle of attack
C
L
/
C
D


0.01 0.02 0.03 0.04
!0.2
0
0.2
0.4
0.6
0.8
CD and CL
CD
C
L

0 5 10 15
!0.2
0
0.2
0.4
0.6
0.8
CL and angle of attack
angle of attack
C
L
0 5 10 15
!10
0
10
20
30
angle of attack
CL^(3/2)/CD
collocation points on chord VLM Goh
coll. points on skeleton VLM Gohl
VLM with wake modeling XFLR
Figure 19: Dierent mesh types
squares: collocation points and trailing vortices are liyng on the skeleton line of
the wing,
circles: collocation points and trailing vortices are laying on the chord line of
the wing
triangles: a calculation from XFLR with VLM classic method, induced drag
with Tretz analysis
The greatest dierence is visible in the induced drag. This is also the value
which causes most of the problems in this method. The wake modeling is
not physical. The trailing vortices are not leaving the wing in the ow speed
direction. In this implementation, the twist has a too strong inuence on the
direction of the trailing vortices. But for exact lift distribution and center of
lift calculations, the vortices must follow the prole chord or the local skeleton
line.
For calculating exact glide numbers, the method provides too high values
caused by too small induced drag values (see Figure 19). The dierence of
the induced drag is shown in Figure 20, where the wing is the model glider
FG-WingX-02.
32
4 IMPLEMENTATION OF A SUITABLE METHOD
0 2 4 6 8 10 12
!4
!2
0
2
4
6
8
10
12
x 10
!3
ange of attack
C
d

i
n
d
u
c
e
d
Induced Drag


VLM Points on skeleton line
VLM with flat points
Figure 20: Induced drag calculated with dierent meshes
Why not let the trailing vortices follow the ow speed direction? For a static
calculation, there would not be any problems, but for a dynamic simulation there
are obstacles. The vortex arrangement for dynamic analysis should not change.
If it changes, the following steps have to be calculated:
1. The vortex arrangement must be corrected at each time step.
2. All the inuence coecients in the Biot-Savart inuence matrix K change
and must be recalculated.
3. The matrix inversion must be recalculated. The matrix can have dimen-
sions of 500x500.
4. An interpolation with the sideslip angle is not possible, because the mesh
is not always the same. So the arrangement has also to be corrected with
sideslip angle. But this step 4 could be directly done in step 1.
If all these calculations had to be done at each time step, this would cost much
more simulation time.
33
4 IMPLEMENTATION OF A SUITABLE METHOD
4.3 Integrating the Methods in the Simulation
Now that the aerodynamic forces may be calculated as a function of incident
ow direction, ap deection and angular rates, the airplane dynamics may
be calculated by applying rigid body dynamics. For the rigid body motion
simulation, the existing MATLAB/Simulink model by Andr Noth is used.
Therefore, only the mass, centre of gravity and the inertia must be added.
The inertia is calculated in the function inertia_calc.m. Some simplication
are made:
homogeneous mass distribution
constant thickness of the wing, with thickness c dened in wingdef.m
wing, modelled with many rectangular boxes
neglected dihedral
Vortex Lattice Method
The input data from the Euler-Lagrange dynamics and the output data of the
force calculating block is shown in the following ow Figure 21:
Aerodynamic Forces
Fx, Fy, Fz
Mx, My, Mz
ap deections
speed in body system
angluar rates
in earth system
phi, theta, psi
Figure 21: Flow diagram of the aerodynamic force block
The data must be transformed into the wings body system. Moreover it
is essential to calculate the angular rates p, q, r around the body axis. The
interpolation of the inuence matrices must be done. Then the velocity vector
is calculated. The angular rates are taken in account in the velocity vector. It
is calculated with following principle:
v
local
=

Omega

R +

V

(28)
Omega is a vector with the angular rates in the body system, R is the
distance between the center of gravity and the local panel.
All the vectors have to be reorganised for the RHS calculation, even the
surface normal vectors. The ap deection is calculated with changing the
RHS. The mesh arrangement is not changed for the ap deection. This is a
simplication in favour of faster calculation, otherwise the whole mesh would
have to be recalculated. Finally the circulation distribution is calculated with
the interpolated inuence matrices at the desired sideslip angle.
Then the forces and moments are calculated and delivered to the output.
34
4 IMPLEMENTATION OF A SUITABLE METHOD
Nonlinear Lifting Line Method
The inclusion of the Nonlinear Lifting Line Method in the dynamics is similar to
the Vortex Lattice Method. Some important dierences occur in the calculation
of the circulation. The circulation is calculated with the initial circulation at
the rst time step. For the next time step, the circulation distribution from the
old time step is used for starting the iteration. The changes of the circulation
distributions are always small, so that only a few iterations are necessary.
35
5 EVALUATION OF THE SIMULATION
5 Evaluation of the Simulation
The simulation is evaluated with real ight tests. In order to at least qualita-
tively compare the behaviour of the simulation to real ight, ight test are made
and taken on video. The experimental data are collected with a self built model
ying wing. More information about the wing can be found in section 6.1 on
page 39 .
5.1 Flight Test Results
For the ight test, a well dened ight attitude must be found, as the simulation
needs the same initial condition as in the ight test.
This attitude can be reached with the following scenario. At rst the wing is
accelerated and then, at high speed, the aileron are set to the initial condition.
The wing slows down until the velocity is zero and then the wing tilts forward.
It begins to oscillate periodically as it stabilises on its own. The period time
can be measured by a clock. Then the period times can be compared with
simulated results. For visualisation, the course of the ight is shown in the
following Figure:
acceleration
with ap
deection
aps set to
the initial state
velocity: zero angle theta: -90
rst period second period
comparison with simulation
no action by the pilot
Figure 22: Path for ight tests
Measured times are written in Table 4. The measurements were made in
good weather conditions, early in the morning, so that weather eects could be
neglected. Two ights were evaluated. Flight one: Only one period time was
measurable. Flight two: Two period times were measurable.
Phygoid Nr. 1 2 Tolerance
Flight 1 [s] 5.8 not measurable +/- 0.3
Flight 2 [s] 5.5 5.4 +/- 0.3
Table 4: Measured times of the phygoids
36
5 EVALUATION OF THE SIMULATION
5.2 Simulation Results
The following attitude is chosen for the evaluation:
-85 real ight data
speed 1 [m/s] a realistic value is chosen
centre of gravity 0.21 m data from VLM Gohl
ap deection 4 (+/- 0.5 degrees) real ight data
angle of attack 2 a realistic value is chosen
Table 5: Initial Condition
The simulation results with the initial condition in Table 5 are shown in
Figure 23. In Table 6 the measured period times are shown. The polars in Xfoil
are made with a turbulence model of N
crit
=9, details about the turbulence
model see in Xfoil manual, in the section Viscous Formulation [4].
Period Nr. 1 2 3
Period time [s] 5.76 5.88 5.76
Table 6: Simulation Results, times of the rst three periods of the phygoid
oszillation
There are several reasons for the slight variation of the simulated and mea-
sured values:
The model is not absolutely perfect, it is a simplication of the reality.
Values such as the induced drag are inexact and can cause small errors.
The calculation of the inertia is another simplication. There could also
occur 3D ow eects, which are neglected.
The airfoil polars are generated with Xfoil. The turbulence model of
N
crit
= 9 in Xfoil might be incorrect.
The model wing is not perfect. The trailing edge is not as sharp as the
simulated prole is. In addition the leading edge is less rounded than the
original airfoil, due to the manufacturing method.
The initial condition is not absolutely correct as it is estimated. The initial
condition of
init
=

2
causes some problems with the Tait-Bryan angles
(more details see in [16]) so that the angle must be approximated.
37
5 EVALUATION OF THE SIMULATION
0
2
0
4
0
6
0
8
0
1
0
0
1
2
0
1
4
0
1
6
0
1
8
0
2
0
0
0 5 1
0
1
5
2
0
2
5
3
0
X

[
m
]
Z [m]
0
5
1
0
1
5
2
0
0
5
0
1
0
0
1
5
0
2
0
0
2
5
0
X

p
o
s
i
t
i
o
n

[
m
]
0
5
1
0
1
5
2
0
!
1
!
0
.5 0
0
.5 1
Y

p
o
s
i
t
i
o
n

[
m
]
0
5
1
0
1
5
2
0
0
1
0
2
0
3
0
4
0
M
i
n
u
s

Z

p
o
s
i
t
i
o
n

[
m
]
0
5
1
0
1
5
2
0
!
1
!
0
.5 0
0
.5 1
P
h
i

[
r
a
d
]
0
5
1
0
1
5
2
0
!
1
.5
!
1
!
0
.5 0
0
.5 1
T
h
e
t
a


[
r
a
d
]
0
5
1
0
1
5
2
0
!
1
!
0
.5 0
0
.5 1
P
s
i

[
r
a
d
]
Figure 23: Simulation results of the periodic oscillation
38
6 RESULTS OF THE SIMULATION
6 Results of the Simulation
6.1 Flying Wing, FG-WingX-02
In the next three sections three dierent wings are presented, simulated and
analysed. The ying wings are tuned for high ight performance and are com-
pared in their stability and performance. The most important design rules about
ying wings are written in Appendix C.
Before this thesis was written, a model glider, FG-WingX-02, was built.
The glider was calculated with [20]. The design contains nine dierent chord
widths and furthermore the twist is not only linear. The chord increases with
the spanwise coordinate, which is unusual for a conventional wing. This eect
improves the stability and increases the Reynolds numbers at the wing tips.
The wing is designed for best glide ratio with ap deection and for good fast
ight without ap deection.
Originally the wing was designed without winglets. The side stability should
have been generated with the high sweep angle. However practical tests showed,
that the side stability was not adequate. An improved design with winglets is
described in 6.2. Basically it is possible to attain side stability only with a sweep
angle. For more details see the Horten wings.
Figure 24: My Wing
Some data of the wing:
Wing span: 2 m
Wing span with winglets: 2.32 m
Wing load: 25.945 g/dm
2
Taper Ratio: 2.0
With ap deections of -3 degrees the XFLR polar calculation generates the
following results (see Figure 6.1):
39
6 RESULTS OF THE SIMULATION
0
2
4
6
8
1
0
!
0
.
2 0
0
.
2
0
.
4
0
.
6
0
.
8
C
L

a
n
d

a
n
g
l
e

o
f

a
t
t
a
c
k
a
n
g
l
e

o
f

a
t
t
a
c
k
CL
0
.
0
1
0
.
0
1
5
0
.
0
2
0
.
0
2
5
0
.
0
3
!
0
.
2 0
0
.
2
0
.
4
0
.
6
0
.
8
C
D

a
n
d

C
L
C
D
CL
0
2
4
6
8
1
0
!
1
0
!
5 0 5
1
0
1
5
2
0
2
5
G
l
i
d
e

R
a
t
i
o
a
n
g
l
e

o
f

a
t
t
a
c
k
Glide Ratio
0
2
4
6
8
1
0
0 5
1
0
1
5
2
0
c
L
3
/
2
c
D
a
n
g
l
e

o
f

a
t
t
a
c
k
cL3/2
cD


V
L
M

X
F
L
R
Figure 25: XFLR Polar Calculations
The best glide ratio is about 23 at an angle of attack of 9. The best sink
rate is at an angle above 10. Angles above 10 are in a critical region of stall.
40
6 RESULTS OF THE SIMULATION
6.1.1 Stability Results
The wing is simulated for the the initial condition shown in Table 7. In Appendix
D is shown, how the equilibrium point is calculated. The wing is dynamically
stable, which shows Figure 26.
v_initial 7 m/s
center of gravity 0.21 m
ap deection -4
sideslip 0
angle of attack 6

init
-85
Table 7: Initial point for simulation
The result is plotted in Figure 26.
41
6 RESULTS OF THE SIMULATION
0
2
0
0
4
0
0
6
0
0
8
0
0
1
0
0
0
1
2
0
0
0 2
0
4
0
6
0
8
0
X

[
m
]
Z [m]
0
2
0
4
0
6
0
8
0
1
0
0
1
2
0
0
5
0
0
1
0
0
0
1
5
0
0
X

p
o
s
it
io
n

[
m
]
tim
e
[s
]
0
2
0
4
0
6
0
8
0
1
0
0
1
2
0
!
1
!
0
.5 0
0
.5 1
Y

p
o
s
it
io
n

[
m
]
0
2
0
4
0
6
0
8
0
1
0
0
1
2
0
0
2
0
4
0
6
0
8
0
1
0
0
Z

p
o
s
it
io
n

[
m
]
0
2
0
4
0
6
0
8
0
1
0
0
1
2
0
!
1
!
0
.5 0
0
.5 1
P
h
i
[
r
a
d
]
0
2
0
4
0
6
0
8
0
1
0
0
1
2
0
!
2
!
1 0 1 2
T
h
e
t
a


[
r
a
d
]
0
2
0
4
0
6
0
8
0
1
0
0
1
2
0
!
1
!
0
.5 0
0
.5 1
P
s
i
[
r
a
d
]
Figure 26: Simulation Results of FG-WingX-02
The simulation of the wing FG-WingX-02 for longer times has shown, that
the wing is spiral mode unstable (similar behaviour as shoen in Figure 29). This
behaviour is noticeable in practical ight test. If the wing ees once into a curve
then he would tilt more and more into the curve. This dynamic behaviour is
slow and can be stabilised by the pilot
42
6 RESULTS OF THE SIMULATION
6.2 Optimised Stable Wing
Originally the wing FG-WingX-02 was designed without winglets. As the winglets
were added later, the circulation distribution is not elliptic anymore.
The optimised wing has the same outline but a dierent twist. The twist was
increased the most at the tip of the wing, where the winglets inuence is the
strongest. The wing is optimised at best glide ratio, not at best sink rate for a
better stall behaviour because the angle of attack at best sink rate is very high.
This modication is done, to create a self stable wing with high performance.
The polars calculated in XFLR are shown in Figure 28.
!1.5 !1 !0.5 0 0.5 1 1.5
0
0.2
0.4
0.6
0.8
Y
C
L
CL Distribution


!1.5 !1 !0.5 0 0.5 1 1.5
0
0.05
0.1
CL*t Distribution
Y
C
L
*
t
CL distribution FG!WingX!02
CL distribution modificated wing
Figure 27: Lift (CLt) and CL distribution at operating point
43
6 RESULTS OF THE SIMULATION
0
2
4
6
8
1
0
1
2
!
0
.
2 0
0
.
2
0
.
4
0
.
6
0
.
8 1
1
.
2
C
L

a
n
d

a
n
g
l
e

o
f

a
t
t
a
c
k
a
n
g
l
e

o
f

a
t
t
a
c
k
CL
0
.
0
1
0
.
0
1
5
0
.
0
2
0
.
0
2
5
0
.
0
3
0
.
0
3
5
0
.
0
4
!
0
.
2 0
0
.
2
0
.
4
0
.
6
0
.
8 1
1
.
2
C
D

a
n
d

C
L
C
D
CL
0
2
4
6
8
1
0
1
2
!
5 0 5
1
0
1
5
2
0
2
5
G
l
i
d
e

R
a
t
i
o
a
n
g
l
e

o
f

a
t
t
a
c
k
Glide Ratio
0
2
4
6
8
1
0
1
2
0 5
1
0
1
5
2
0
c
L
3
/
2
c
D
a
n
g
l
e

o
f

a
t
t
a
c
k
cL3/2
cD
Figure 28: Polars calculated in XFLR
The main dierence between the optimised stable wing and the original wing
is the angle of attack, when the best glide ratio occurs. The optimised stable
wing is fundamentally better manoeuvrable at best glide ratio. The angle of
attack at best glide ratio is further from stall and even ying at best sink rate
would be possible. In the middle of the wing, the CL value is increased for
better stall behaviour, but the performance hardly improved.
The lift distribution at the operating point is shown in Figure 27. For com-
parison only the lift distribution of the real model glider and the model glider
without winglets are plotted in the same Figure. The spanwidth coordinate is
folded on the y-axis for better visualisation.
44
6 RESULTS OF THE SIMULATION
6.2.1 Stability Results
Simulations are done, to analyse the stability behaviour of the modied wing.
The operating point is dened, as shown in Table 8.
equilibrium point initial point
v_initial 8.4437 m/s data from VLM Gohl 8.4437
angle of attack 8 5

Glide
2.16
centre of gravity 0.2461 m data from VLM Gohl 0.2461
ap deection -2 -2
sideslip 3 3
Table 8: Operating point
Then the wing is simulated for stability analysis. The wing should be stable.
The simsulation results with the conguration listed in Table 8 are shown in
Figure 29
45
6 RESULTS OF THE SIMULATION
0
1
0
0
2
0
0
3
0
0
4
0
0
5
0
0
6
0
0
7
0
0
8
0
0
9
0
0
1
0
0
0
!
0
2
0
4
0
X

[
m
]
Z [m]
0
5
0
1
0
0
1
5
0
0
5
0
0
1
0
0
0
1
5
0
0
X

p
o
s
i
t
i
o
n

[
m
]
t
i
m
e

[
s
]
0
5
0
1
0
0
1
5
0
!
1
5
0
!
1
0
0
!
5
0 0
Y

p
o
s
i
t
i
o
n

[
m
]
0
5
0
1
0
0
1
5
0
0
2
0
4
0
M
i
n
u
s

Z

p
o
s
i
t
i
o
n

[
m
]
0
5
0
1
0
0
1
5
0
!
0
.
4
!
0
.
2 0
P
h
i

[
r
a
d
]
0
5
0
1
0
0
1
5
0
!
0
.
2 0
0
.
2
T
h
e
t
a


[
r
a
d
]
0
5
0
1
0
0
1
5
0
!
4
!
2 0
P
s
i

[
r
a
d
]
Figure 29: Simulation results with initial state not in equilibrium point (initial
point: Table 8)
The time the wing needs to balance itself is about 50 seconds. The dimension
of the momentum in y-axis after 30 seconds is 1 10

4Nm. The dynamical


behaviour is similar to the wing FG-WingX-02.
After 100 seconds the wing begins to instabilise itself around yaw and roll
axes caused by small numerical noise. These eects can only be canceled with a
complex optimisation of the geometry (dihedral and winglet, see section 6.3 on
the next page). These modications are not done here, they are too complex.
The modied wing does not garantee the spiral stability and dutch roll mode
stability, but the dynamic of these instabilities are slow and can be stabilised
by the pilot.
46
6 RESULTS OF THE SIMULATION
6.3 Dutch Roll Mode (germ. Taumelschwingung) and
Spiral Mode
A problem with ying wing, which is dicult to handle, is the dutch roll and
spiral mode stability. Here some of information about yaw stability coupled
with dutch roll and spiral mode is added. For ying wings these problems often
appear.
Eects of dihedral and sweep angle are qualitatively explained. This section
does not give detailed theoretical information, it only helps to understand the
simulated results and enlightens the problem of yaw stability coupled with dutch
roll mode and spiral mode. More detailed theoretical information can be found
in [3], page 180 and in [9]
To reach yaw stability the most simple solution is to add large enough
winglets. Another way is to add a high sweep angle, which stabilises the wing by
drag. The dihedral is basically added for better roll stability, but if it was added
on the entire wing it would not stabilise the wing on the yaw axis. Dihedral and
sweep angle are quasi similar roll angle stabilising elements.
Dutch roll means, that the wing oscillates in yaw axis coupled with an oscil-
lation in roll axis. This is caused by too much dihedral and not enough inuence
of the n (If the winglets are attached behind the centre of gravity, they act as
ns.). The opposite problem which occurs with too strong inuence of the n
and too little dihedral, is the spiral mode. The wing tends to turn more and
more into the curve until its ight path is a spiral.
To design a wing which fulls both stability criteria (roll and yaw) is chal-
lenging. Winglets are often a good solution, they do not have a dihedral eect.
Another solution could be to add a negative dihedral. It eliminates the eect
of dihedral on the sweep, so that dutch roll and spiral stability is sucient. On
the other hand, negative dihedral would also cause negative eects in fast ight.
An important design rule for ying wing is, that dihedral should not be
added to the wing. If dihedral is added for roll stability, dutch roll and spiral
stability must be studied carefully. For the two stable built wings (see above),
the spiral mode is slow enough to be controlled by the pilot and therefore is not
a hazardous instability.
6.4 Unstable Wing
A wing is designed which is unstable and has a high performance. The design
criteria is to reach a high glide ratio. This is done by a wing without sweep angle
and twist angle, as these elements create a loss of performance. To minimise
the induced drag an elliptical circulation distribution is aimed. Therefore the
resulting ideal geometry would be a perfect ellipse. Due to the low Reynolds
numbers at the wing tip, the chord at the wing tip is increased and a small
twist is added. Therefore the shape of the wing is not a perfect ellipse, but the
circulation distribution is a perfect ellipse. Winglets are added to garantee the
side stability. The are not added to create an inuence on the ow on the wing,
but only to garantee the side stability of the wing. They act similarly to ns.
Nevertheless supplementary winglets could be added in order to improve the
performance of the wing, but this is not done here.
The centre of lift is not lying in front of the neutral point, so the wing is
absolutely pitch unstable. As the wing is unstable in the pitch axis, a controller
47
6 RESULTS OF THE SIMULATION
must be designed to garantee the pitch stability. Otherwise, without a controller,
a simulation would not be possible.
Consequently the yaw stability is tested in simulation. With the controller
the wing should be stable on the pitch axis and the stability in yaw and roll
axes must be fullled by the self-stability of the wing. It must be tested as well,
if the wing is manoeuvrable. In other words, can the side stability of the wing
compensate the negative momentum from rudder deections?
6.4.1 Geometry
The geometry denition is shown in Appendix A, Algorithm 5 on page 63.
For visualisation a mesh plot is shown in Figure 30.
0.2
0.4
0.6
!1
!0.5
0
0.5
1
0
0.05
0.1


Z
X
Y
Mesh
0
0.2
0.4
0.6
0.8
!1.5 !1 !0.5 0 0.5 1 1.5

Lift Distribution [N]

0
0.02
0.04
0.06
0.08
0.1
0.12
Figure 30: Mesh and force distribution with initial state as shown in Table 10
6.4.2 Static Results
The polars are calculated with a Vortex Lattice Method in XFLR. A calculation
with a panel method is added. In XFLR, the panel method is a 3D method which
places the singularity elements on the top and the down surface of the wing.
Detailed informations about the XFLR calculation methods can be found in [1].
The best glide ratio can be reached at an angle of attack of 5 to 6, the
maximum value with the VLM calculation is 26.
48
6 RESULTS OF THE SIMULATION
0
2
4
6
8
1
0
0
0
.
2
0
.
4
0
.
6
0
.
8 1
C
L

a
n
d

a
n
g
l
e

o
f

a
t
t
a
c
k
a
n
g
l
e

o
f

a
t
t
a
c
k
CL
0
0
.
0
2
0
.
0
4
0
.
0
6
0
.
0
8
0
0
.
2
0
.
4
0
.
6
0
.
8 1
C
D

a
n
d

C
L
C
D
CL
0
2
4
6
8
1
0
1
2
1
4
1
6
1
8
2
0
2
2
2
4
2
6
2
8
G
l
i
d
e

R
a
t
i
o
a
n
g
l
e

o
f

a
t
t
a
c
k
Glide Ratio
0
2
4
6
8
1
0
5
1
0
1
5
2
0
2
5
c
L
3
/
2
c
D
a
n
g
l
e

o
f

a
t
t
a
c
k
cL3/2
cD


P
a
n
e
l

M
e
t
h
o
d
V
L
M

M
e
t
h
o
d
Figure 31: polars of the unstable optimal wing, caluculated in XFLR with a 3D
panel method and a VL Method.
6.4.3 Controller
The controller is a PID controller. The controller parameters are calculated
with the Ziegler Nichols Method [5]. The parameters are listed in Appendix A.
An actuator saturation was added, because the polars are only made for ap
deections of -15 ... 15. The dierence of
init

out
(pitch angle is
out
)
must be multiplied with -1, because the ap deection is positive if the ap goes
down (Xfoil denition). The controller is shown in Figure 32.
49
6 RESULTS OF THE SIMULATION
Figure 32: Simulator with controller
Attention: This controller is only designed for controlling the pitch
axis. The control design in this way is not meant to be realised. But
for studying the instabilities of the other axes it is only essential, that
the pitch axis is stable.
6.4.4 Simulation results
The simulation is started with the following initial condition, see Table 9.
equilibrum point initial point
v_initial 7.25 m/s data from VLM Gohl 7.25 m/s
angle of attack 6 at best glide ratio 4

Glide
(glide angle) 2.72
centre of gravity 0.0883 m data from VLM Gohl 0.0883 m
ap deection 0 0
sideslip 3
Table 9: Initial point
The simulation results are shown in Figure 33.
50
6 RESULTS OF THE SIMULATION
0
1
0
0
2
0
0
3
0
0
4
0
0
5
0
0
6
0
0
0
1
0
2
0
X

[
m
]
Minus Z [m]
0
2
0
4
0
6
0
8
0
1
0
0
0
5
0
0
1
0
0
0
X

p
o
s
i
t
i
o
n

[
m
]
t
i
m
e

[
s
]
0
2
0
4
0
6
0
8
0
1
0
0
0 2 4
Y

p
o
s
i
t
i
o
n

[
m
]
0
2
0
4
0
6
0
8
0
1
0
0
!
1
0 0
1
0
2
0
3
0
M
i
n
u
s

Z

p
o
s
i
t
i
o
n

[
m
]
0
2
0
4
0
6
0
8
0
1
0
0
0
0
.
0
5
0
.
1
P
h
i

[
r
a
d
]
0
2
0
4
0
6
0
8
0
1
0
0
0
.
0
6
4
0
.
0
6
6
T
h
e
t
a


[
r
a
d
]
0
2
0
4
0
6
0
8
0
1
0
0
0
0
.
0
5
0
.
1
P
s
i

[
r
a
d
]
Figure 33: Simulation Results
In the consequence it is tested, what happens with dierent ap deections in
order to y curves, with a closer look at the angle of sideslip. This is important,
because rudder deections are coupled with an unpleasant eect, the momentum
in the yaw axis. It causes a sideslip angle in the opposite direction.
First simulation results showed, that with this wing it is not possible to y
51
6 RESULTS OF THE SIMULATION
a curve! The sideslip increases too much. Therefore the wing must be stabilised
more in the yaw axis.
A positive dihedral or larger winglets cause the best eect on the side stabil-
ity. It is important to place the winglets far enough behind the physical centre
of gravity. To improve the side stability, the winglets are placed 50 cm behind
the wing (see Figure 30). After that, the simulation results are much better and
ying curves has become possible. Simulation results are shown in Figure 34
with the aileron deection shown in Table 11. The simulation is started with
the following initial point (see Table 10). In Figure 35 the relative sideslip angle
is plotted versus the time.
The characteristic of
in
shows clearly that the wing is stable, even with ap
deections.
The speed is plotted in Figure 6.4.4
Initial point
v_initial 7 m/s arbitrary
angle of attack 5 arbitrary

Glide
2.72 calculated from glide ratio
centre of gravity 0.0883 m data from VLM Gohl
ap deection 0
Table 10: Initial point
time [s] 3 - 3.5 3.5 - 6 6 - 9s 9 - 60
left aileron 1 0 -1 0 + / - controller deection added
right aileron -1 0 1 0 + / - controller deection added
Table 11: Flap deectioins
52
6 RESULTS OF THE SIMULATION
0
2
0
4
0
6
0
8
0
1
0
0
!
1
5
!
1
0
!
5
0
0
1
0
2
0
3
0
X

[
m
]
Y

[
m
]
Z [m]
0
2
0
4
0
6
0
0
5
0
1
0
0
1
5
0
X

p
o
s
i
t
i
o
n

[
m
]
t
i
m
e

[
s
]
0
2
0
4
0
6
0
!
2
0
!
1
0 0
1
0
Y

p
o
s
i
t
i
o
n

[
m
]
0
2
0
4
0
6
0
0
2
0
4
0
M
i
n
u
s

Z

p
o
s
i
t
i
o
n

[
m
]
0
2
0
4
0
6
0
!
0
.
2 0
0
.
2
0
.
4
0
.
6
P
h
i

[
r
a
d
]
0
2
0
4
0
6
0
0
.
0
5
5
0
.
0
6
0
.
0
6
5
T
h
e
t
a


[
r
a
d
]
0
2
0
4
0
6
0
!
5
0 0
5
0
P
s
i

[
r
a
d
]
Figure 34: Simulation results with ap deections. At rst a left curve, then a
right curve is own.
53
6 RESULTS OF THE SIMULATION
0 10 20 30 40 50 60
!0.06
!0.04
!0.02
0
0.02
0.04
0.06
0.08
0.1
0.12
0.14
Time
B
e
t
a

(
A
n
g
l
e

o
f

S
i
d
e
s
l
i
p
)
Figure 35: Sideslip angle of the simulation with ap deections, as shown in
Figure 34
0 10 20 30 40 50 60
6.4
6.5
6.6
6.7
6.8
6.9
7
7.1
7.2
Time
S
p
e
e
d

Figure 36: Speed of the ying wing
54
7 Discussion
The simulation results show that the performance of an accurately optimised
stable wing can be topped by a pitch controlled unstable wing. However, the
fact is, that even a well designed self stable wing with an elliptic circulation
distribution can have an excellent performance and full the stability criteria.
Therefore, the question whether it is worth to design a pitch unstable wing with
a controller can not be answered here.
The simulated unstable wing has calculated glide number of about 26 in the
area of Reynolds numbers of 100000, whereas comparison the optimised stable
wing has the best glide ratio at 24, with the same Reynolds number. The self
stable ying wing FG-WingX-02 has best glide ratios about 23. These glide
ratios are calculated with a vortex lattice method with the panels lying on the
prole mean line. For this comparison only glide ratios are considered, there
are many other criterias to compare (as example: stall behaviour or curve ying
behaviour). The statement, that the controlled pitch unstable wing has the best
glide ratio does not necessarily mean, that it is the best wing in every respect .
A comparison of the calculated glide ratios with real measured glide ratios
of contest wing airplanes is dicult to nd, and most gliders are not tuned for
best glide ratio. Some ight data of succeeded F3B competition models are
presented in [19], their glide ratios are around 15 - 17.
For more accurate statements wind tunnel tests or exact measured ight
paths would have to be done.
The simulation results show clearly, that the time until the wing stabilises
itself from a disturbance is long (duration in the tests: about 50 seconds at an
initial angle of -90). The reason is the low damping around the pitch axis.
The damping basically depends on the induced velocity of the angular rate p
on the wing. The further the wing elements are away from the pitch-axis, the
higher the damping velocity is, as written in the following formula: v = R.
All the simulated wings tend to y a curve. If the wing once has slightly
tilted around the roll axis, the wing does not act with a moment against this
tendency. Therefore, the wing tilts into a curve, until an equilibrium point in
the curve is reached. The reason for this behaviour is, that a numerical noise
always acts on the calculated moments. Caused by these small disturbances,
the wing passes into a dutch roll instability.
55
8 Conclusion
The most important tasks and limitations for a dynamical method were dened,
and consequently numerical methods which full the tasks were presented. Ex-
isting statical and dynamical applications were introduced and discussed. The-
oretical information about vortex methods was given.
A complex method for calculating aerodynamic forces with regard to com-
plex three-dimensional wing geometries, angular rates and asymmetrical ap
deections was implemented in MATLAB. It turned out, that a model, where
all ow eects would have been included consumed a lot of computer process
time. Therefore the implemented vortex lattice method was modied for short
calculation times and some simplications in the wake arrangement were made.
The wake was modelled statically with vortex laments. Horseshoe vortices as
singularity elements in chord and spanwise direction were chosen. This Vortex
Lattice Method was integrated in a rigid body motion simulation implemented
in Simulink.
With the simulation tool, it is possible to study dynamical and statical
results for arbitrary three-dimensional ying wing congurations. Statements
about stability in all axis and manoeuvrability with ap deections can be made.
Complex dynamical eect such as spiral mode, dutch mode and phygoid mode
can be simulated and analysed.
To answer the question about the unstable wings gain of performance, a
pitch unstable wing was designed and compared with two self stable ying wings.
For testing the ight characteristics and the dynamic stability, the ying wings
were simulated in the dynamic simulation. The unstable wing was stabilised by
a pitch controller to garantee the pitch stability.
The results show, that ight performance with an unstable wing can be won
and high calculated glide ratios can be reached. The unstable wing, simulated
in this thesis reaches glide ratios about 26. For comparison, the stable modied
ying wing reaches a glide ratio about 24 at Reynolds numbers of 100000. The
interesting result is, that the pitch unstable wing does not have a signicantly
better glide ratio in comparison to a well designed self stable ying wing. But
the unstable wing has a much better stall behaviour, because the angle of attack
at best glide ratio is smaller than than the angle of attack at best glide ratio
of the stable wings . So the question whether it is worth to design a controller
and achieve better results of the glide ratio must be answered in consideration
of other criteria needed.
56
9 Future Work
For a continuative work there are many modications to do. In general three
optimisation strategies might be useful:
Increasing the simulation speed to improve the vortex model. A high im-
provement of the calculation speed could allow automatic geometry variation
to nd best geometry parameters. An improved model would be necessary to
combine more aerodynamic eects and therefore give more accurate results. A
better evaluation of the simulation would elucidate, if even a improvement of
the model is necessary.
Increasing the Simulation Speed
With a modied 6DOF simulator the computing time could be improved. Up-
grading the computing time of the calculation methods would also be necessary
for a faster simulation.
Improving the Model
Many improvements in wake modelling and ow separation could be done. A
simple improvement in this implementation would be, that Xfoil is automati-
cally coupled with the vortex methods. An interpolation for dierent Reynolds
numbers could also improve the physical model without expending a lot of com-
puting time. To achieve better results for induced, a better wake modelling
would be necessary. For even better results, a vortex lattice method with a
lattice of vortices over the top and bottom surface of the wing could be laid.
Evaluating the Simulation
The simulation results could be evaluated more precisely with ight path mea-
surements or even with wind tunnel measurements.
57
References
[1] Author unknown, guidelines for xr5 v4.16. Analysis of foils and wings
operating at low Reynolds numbers, April, 2009.
[2] Jon S. Berndt. jsbsim.sourceforge.net/. Manual.
[3] Michael Cook. Flight Dynamic Principles. Number ISBN-13: 978-0-7506-
6927-6 in second edition. BUTTERWORTH HEINEMANN, 2007.
[4] Mark Drela and Harold Youngren. Xfoil 6.9 user primer. 30 Nov 2001.
[5] Lino Guzzella. Analysis and Synthesis of Single-Input Single-Output Con-
trol Systems. vdf, 2007.
[6] Karl-Heinz Horstmann. Ein Mehrfach - Traglinienverfahren und seine Ver-
wendung fr Entwurf und Nachrechnung nichtplanarer Flgelanordnungen.
1988.
[7] Michael Wohlfahrt Karl Nickel. Schwanzlose Flugzeuge. Birkhuser Verlag
(Flugtechnische Reihe Band 3), 1990.
[8] Joseph Katz and Allen Plotkin. Low-Speed Aerodynamics. Cambridge Uni-
versity Press, 2001.
[9] Bill Kuhlman and Bunny Kuhlman. Swept wings and eective dihedral.
RC Soaring Digest, March 2000.
[10] GNU General Public License. Xr5 is an analysis tool for airfoils, wings
and planes operating at low reynolds numbers.
[11] Harold Youngren Mark Drela. Avl,
http://web.mit.edu/drela/public/web/avl/.
[12] W. H. Mason. Applied computational aerodynamics text/notes aerodynam-
ics of 3d lifting surfaces http://www.aoe.vt.edu/. Department of Aerospace
and Ocean Engineering Virginia Polytechnic Institute and State University,
Mar 11, 1998.
[13] Tomas Melin. A vortex lattice matlab implementation for linear aerody-
namic wing applications. Masters thesis, Royal Institute of Technology
(KTH), 2000.
[14] Austin Meyer. www.x-plane.com/.
[15] Stuart Buchanan Jon Berndt Bernhard Buckel Cameron Moore Curt Olson
Dave Perry Michael Selig Darrell Walisser Michael Basler, Martin Spott
et al. www.ightgear.org/. The Flight Gear Manual 1.9.0.
[16] A. Noth, S. Bouabdallah, and R. Siegwart. Dynamic Modelling of Fixed-
Wing UAVs. ETH Zrich, 2008, Version 2.0.
[17] D. Bruce Owens. Weissingers model of the nonlinear lifting-line method
for aircraft design. AIAA 98-0597, 1979.
[18] Ion Paraschivoiu. Aerodynamique Subsonique. 1998.
58
[19] Helmut Quabeck. Design, Leistung und Dynamik von Segelugmodellen.
HQ Modellugliteratur, 1994.
[20] Frank Ranis and Herbert Stammler. Nurgel v2.17 (freeware).
[21] redhammer project. Tornado 1.0. http://www.redhammer.se/tornado/,
RELEASE 2.3 2001-01-31.
[22] P. SANTINI and P. GASBARRI. Lifting surface in subsonic unsteady
regime. Universit a di Roma La Sapienza, Dipartimento di Aerospaziale,
via Eudossiana 18; 00184 Roma, Italy, 5 May 1998).
[23] Matthieu Scherrer. Mthode dintgration sur une aile de rsultat expri-
mentaux et xfoil.
[24] Herrmann Schlichting and Erich Truckenbrodt. Aerodynamik des
Flugzeuges, Band II. 1969.
[25] James C. Sivellis and Rober H. Neely. Method for calculating wing char-
acteristics by lifting-line theory using nonlinear section lift data. National
Advisory Commitee for Aeronautics Langley Field Report No. 865, 1946.
[26] James C. Sivellis and Robert H. Neely. Method for calculating wing char-
acteristics by lifting-line theory using nonlinear section lift data. Technical
Note no. 1269, April 1947.
59
Appendix
Appendix A) Most Important Code Fragments
Implementation Nonlinear Lifting Line Method
Algorithm 1 Iteration function (Gammait_nonlin.m) of the Nonlinear Lifting
Line Method
function[Gamma_alt ,alpha_i ,w_i]= Gammait_nonlin(K,K_far ,Gamma_start ,D,min ,b,
v_abs ,angle_of_attack ,y,alpha_p ,CL_p ,flap_angle ,poldef)
2 Gamma_alt=Gamma_start;
Max =1;
4 while Max >min
w_i =1/2*( K_far*Gamma_alt); %here: Trefftz Analysis , with factor
1/2, else 1!
6 alpha_i=-atan(w_i ./(v_abs ));
Gamma_new =(1/2* CL(alpha_p ,CL_p ,flap_angle ,alpha_g -alpha_i ,
poldef) .*b).*v_abs ;
8 Gamma =(1-D)*Gamma_alt+D*Gamma_new;
Max=abs(max(Gamma -Gamma_new));
10 Gamma_alt=Gamma;
12 end
end
Algorithm 2 Generate Vortices (generate_vorices.m)of the Nonlinear Lifting
Line Method
1 function[K]= generate_Vortices_VarIII(A,T_left ,T_right ,n,N)
for i=1:2* sum(n)
3 for j=1:2* sum(n)
%Collocation Point , where the Velocity will be calculated: i
5 %this velocity gets summated from element 0 till n: j
%Velocity of the upper Vortex
7 w_up=vortex(T_left(:,j),T_right(:,j),A(:,i));
%Velocity of the left Vortex
9 w_left=vortex_left(T_left(:,j),A(:,i));
%Velocity of the right Vortex
11 w_right=vortex_right(T_right(:,j),A(:,i));
13 w=w_up+w_left+w_right;
%Projection in the normal direction
15 K(i,j)=w(1)*N(1,i)+w(2)*N(2,i)+w(3)*N(3,i);
end
17 end
end
60
Appendix
Implementation: VLM Method
Algorithm 3 Function vortex_panel.m of the Vortex Lattice Method
1 %This Program calculates the velocity divided by the Zirculation
%written and tested: 20. Februar 2009 by Flavio Gohl
3 %The Code is modified to the code in book low speed Aerodynamics page 255
from Joseph Katz.
%Attention , the code is not the same!
5 function[a_i_X_vec ,a_i_Y_vec ,a_i_Z_vec ]= vortex_panel(L_X ,L_Y ,L_Z ,R_X ,R_Y ,
R_Z ,P_X ,P_Y ,P_Z ,m,n,epsilon)
einsmatrix=ones(m(1),sum(n)*2);
7 %define the three vectors r_0 , r_1 ,r_2
r_0_X=R_X -L_X;
9 r_0_Y=R_Y -L_Y;
r_0_Z=R_Z -L_Z;
11
r_1_X=P_X*einsmatrix -L_X;
13 r_1_Y=P_Y*einsmatrix -L_Y;
r_1_Z=P_Z*einsmatrix -L_Z;
15
r_2_X=P_X*einsmatrix -R_X;
17 r_2_Y=P_Y*einsmatrix -R_Y;
r_2_Z=P_Z*einsmatrix -R_Z;
19
%Calculate r_1xr_2=vprod
21 vprod_X=r_1_Y.*r_2_Z -r_1_Z.* r_2_Y;
vprod_Y=r_1_Z.*r_2_X -r_1_X.* r_2_Z;
23 vprod_Z=r_1_X.*r_2_Y -r_1_Y.* r_2_X;
25 %Calculate norm
norm_vprodsquare=vprod_X .^2+ vprod_Y .^2+ vprod_Z .^2;
27 R1=sqrt(r_1_X .^2+ r_1_Y .^2+ r_1_Z .^2);
R2=sqrt(r_2_X .^2+ r_2_Y .^2+ r_2_Z .^2);
29
%!!!!! Gamma is linear with K, so it comes later in the equation !!!!!! a is
a velocity/Gamma
31 a_i =(1/(4* pi)*1./ norm_vprodsquare).*(( r_0_X.* r_1_X+r_0_Y .*r_1_Y+r_0_Z.*
r_1_Z)./R1 -(r_0_X .* r_2_X+r_0_Y .*r_2_Y+r_0_Z.*r_2_Z)./R2);
33 %Check singularities
[zeile ,colonne ]=find(R1 <epsilon);
35 a_i(zeile ,colonne)=0;
[zeile ,colonne ]=find(R2 <epsilon);
37 a_i(zeile ,colonne)=0;
[zeile ,colonne ]=find(norm_vprodsquare .^2< epsilon);
39 a_i(zeile ,colonne)=0;
41 %influence coefficients
a_i_X=a_i.* vprod_X;
43 a_i_Y=a_i.* vprod_Y;
a_i_Z=a_i.* vprod_Z;
45
%write a_i in a row , in the structure of K
47 for i=1:m(1)
a_i_X_vec (1,[(i-1)*sum(n)*2+1:1:(i)*sum(n)*2])=a_i_X(i,:);
49 a_i_Y_vec (1,[(i-1)*sum(n)*2+1:1:(i)*sum(n)*2])=a_i_Y(i,:);
a_i_Z_vec (1,[(i-1)*sum(n)*2+1:1:(i)*sum(n)*2])=a_i_Z(i,:);
51 end
53 end
61
Appendix
Algorithm 4 Function generate_vortices_panel.m of the Vortex Lattice
Method
1 function[K]= generate_Vortices_panel(A_X ,A_Y ,A_Z ,T_right_X ,T_right_Y ,
T_right_Z ,T_left_X ,T_left_Y ,T_left_Z ,T_right_behind_X ,T_right_behind_Y
,T_right_behind_Z ,T_left_behind_X ,T_left_behind_Y ,T_left_behind_Z ,n,m,
U,V,W)
epsilon =1*10^ -20;
3 lenj =0;
%Rewrite the surface normals
5 for i=1:m(1)
U_vec (1,[(i-1)*sum(n)*2+1:1:(i)*sum(n)*2])=U(i,:);
7 V_vec (1,[(i-1)*sum(n)*2+1:1:(i)*sum(n)*2])=V(i,:);
W_vec (1,[(i-1)*sum(n)*2+1:1:(i)*sum(n)*2])=W(i,:);
9 end
11 for i=1:2* sum(n)
for j=1:m(1)
13 %velocity of the bound vortex
[a_i_X ,a_i_Y ,a_i_Z] = vortex_panel(T_left_X ,
T_left_Y ,T_left_Z ,T_right_X ,T_right_Y ,T_right_Z ,A_X(j,i),
A_Y(j,i), A_Z(j,i),m,n,epsilon);
15
%Velocity of the left Vortex
17 [La_i_X ,La_i_Y ,La_i_Z] = vortex_panel_l(
T_left_behind_X ,T_left_behind_Y ,T_left_behind_Z ,T_left_X ,
T_left_Y ,T_left_Z ,A_X(j,i), A_Y(j,i), A_Z(j,i),m,n,epsilon
);
19 %Velocity of the right Vortex
[Ra_i_X ,Ra_i_Y ,Ra_i_Z] = vortex_panel_r(T_right_X ,
T_right_Y ,T_right_Z ,T_right_behind_X ,T_right_behind_Y ,
T_right_behind_Z ,A_X(j,i), A_Y(j,i), A_Z(j,i),m,n,epsilon)
;
21
wx=a_i_X+La_i_X+Ra_i_X;
23 wy=a_i_Y+La_i_Y+Ra_i_Y;
wz=a_i_Z+La_i_Z+Ra_i_Z;
25
%Projection in the normal direction
27 K(j+lenj ,:)=wx.*U_vec+wy.*V_vec+wz.*W_vec;
29 end
lenj=j+lenj;
31 end
end
62
Appendix
Simulations
Algorithm 5 Wing geometry of the unstable ying wing
%===================
2 %== Wing Geometry
%===================
4 betha_end =1;%betha_end /10 is the real angle!
6 p=5;%number of partitions
ny=[0 0 0 0 90]*pi /180;%Dihedral of the Partitions
8 n=[6 5 4 3 3];%number of collocation points in y-Direction
m=[4 4 4 4 4];%number of collocation points in x-Direction
10 s=[0.5 0.25 0.9 -0.75 0.1 0.15];%Spanwidth of partition
b_root =[0.24 0.215 0.18 0.155 0.14];
12 b_tip =[0.215 0.18 0.155 0.125 0.06];
phi =[2 6 8 9 10]/360*2* pi; %sweep angle(back is positiv !)
14 alfa =[0 0 0 0 0 0 0 0 0 0]*pi /180; %Twist angle
x_0_rel =[0 0 0
16 tan(phi (1))*s(1) s(1) sin(ny(1))*s(1)
tan(phi (2))*s(2) s(2) sin(ny(2))*s(2)
18 tan(phi (3))*s(3) s(3) sin(ny(3))*s(3)
tan(phi (4))*s(4) +0.5 s(4) sin(ny(4))*s(4)
20 tan(phi (5))*s(5) s(5) sin(ny(5))*s(5)];
22 x_0(:,1)=[0 0 0];
for i=2: length(x_0_rel)
24 x_0(:,i)=x_0(:,i-1)+x_0_rel(:,i);
end
26 cog = [0.0883 0 0]; % Correct centre of gravity
c=0.012 ; %Thickness of the Wing for Inertia
28 flap_hinge =0.75;
Algorithm 6 controller.m, parameters used for simulating the unstable ying
wing
1 %Controller Data
kp_crit =50;
3 T_crit =0.14;
kp=0.6* kp_crit;
5 Ti=0.5* T_crit;
Td =0.125* T_crit;
7 Num=kp*[Ti*Td Ti 1];
Denom =[0 Ti 0]; P=kp;
9 TI=kp/Ti;
TD=kp*Td;
63
Appendix
Appendix B) Manual
Manual of the Vortex Lattice Method
Statical analysis
Copy the m-les (on CD-Rom) and the simulink model into a folder, in the
current directory of the m-les should be made a folder named polars, where
Xfoil polars are saved.
1. Calculate in Xfoil the prole polar for dierent ap angles.
User Guide for creating polars:
(a) open Xfoil
(b) dene the point, where the momentum is calculated
(c) load mh45.dat or any other prole
(d) oper
(e) visc tipe Re
(f) vpar (enter viscous parameter menu) N 9.0 (set new lower turbulence
level) <return> (back to oper)
(g) optionally type iter for more iterations, so that the calculation con-
verges
(h) write the polars:
i. set the initial Value of alfa bsp: alfa -5
ii. pacc enter the le name as ex.: mh45.pol
iii. enter
iv. aseq
v. if it doesnt work, increase the iterations (for example iter 400)
(i) Flap deection:
i. gdes
ii. ap
iii. then set the values
iv. exec
(j) if not converged, change the panelling! ppar and pane, then save the
prol and try again
2. Save all polars in *.dat les in the folder polars.
64
Appendix
3. In the function create_polar_table.m, dene and include the polar:
poldef =[15 7.5 0 -7.5 -15]*2*pi /360;
2
[alpha_p (:,1,:),CL_p(:,1,:),CD_p(:,1,:),CM_p(:,1,:)]= import_polar(
mh45f15.dat);
4 [alpha_p (:,2,:),CL_p(:,2,:),CD_p(:,2,:),CM_p(:,2,:)]= import_polar(
mh45f7_5.dat);
[alpha_p (:,3,:),CL_p(:,3,:),CD_p(:,3,:),CM_p(:,3,:)]= import_polar(
mh45pol.dat);
6 [alpha_p (:,4,:),CL_p(:,4,:),CD_p(:,4,:),CM_p(:,4,:)]= import_polar(
mh45fm7_5.dat);
[alpha_p (:,5,:),CL_p(:,5,:),CD_p(:,5,:),CM_p(:,5,:)]= import_polar(
mh45fm15.dat);
For example, ve ap deections are saved at the angles dened in poldef.
4. Dene your wing. Attention, all partitions must have the same number of
panels in chord direction! Otherwise, the mesh matrices can not be built.
Take as example:
1 betha_end =40;
3 p=2; %number of Partitions
ny=[0 90]*pi /180; %Neigungswinkel of the Partitions
5
n=[4 2]; %anzahl Aufpunkte fr die
einzelne Partition / Flgelhlfte
7 m=[5 5];
s=[1 0.6]; %Spannweite jeder
Parition
9 b_root =[0.2 0.2]; %Breite , Symmetrische Eingabe
b_tip =[0.2 0.2];
11 phi =[20 29]/360*2* pi; %Pfeilungswinkel(gegen hinten
positiv !)
alpha =[0 0 0 0]*pi /180;
13 x_0 =[0 0 0
tan(phi (1))*s(1) s(1) sin(ny(1))*s(1)];
15
x_0_Trefftz =[0 0 0;tan(phi (1))*s(1) s(1) sin(ny(1))*s(1)+sin(alpha (1)
)*100] ;
17
19 cog = [0.212690364338495 0 0]; % Correct center of
gravity
21 c=0.012 ; %Thickness of the Wing for
Inertia
23 flap_hinge =0.75;
If results with sideslip are analysed, check that the value of betha_in
contains the desired sideslip angle. The range of sideslip angles are dened:
[-betha_end/10 : 0.1 : betha_end/10]
5. Create an initial.m le and write the initial values of the ow. In this
le, the ap vector is initialized. You must dene for each partition if it
is apped or not, and if it is the left or right aileron. For the wing above,
the ap vector would have the following form:
1 flap_angle_part =-[0 left_ail right_ail 0];
In this vector, the winglets do not have aps.
6. Include your wingdef.m le in the code main_panel.m the name is free,
however you could dene it as wingdef_myrstwing.m. The initial.m le
65
Appendix
should not be unnamed, otherwise the name has to be changed in other
functions.
7. type main_panel.m in MATLAB workspace for running the application.
(a) Type CA, CW, CW_ind, CW_visc, CM, CL_distribution, CD_distribution,
Lift_Matrix, xg_X (Centre of pressure), and many others
(b) Mesh plots and force distributions over the wing are shown in gures.
Check in the mesh plot if the geometry is panelled correctly.
Numerical Problems It might occur, that some collocation points are badly
positioned. So change the geometry or change the panel number in chord
or spanwise direction, then try again. The value of epsilon can also be
changed in the function generate_vortice_panel.m.
Dynamic analysis
1. First make a static analysis for initial values (see above).
2. Calculate with inertia_calc.m the inertia Ixx, Iyy, Izz or put the inertias
from CAD data into the le initial.m.
3. Write your correct centre of gravity in the initial.m le, take it from the
static analysis or from experimental data.
4. Run the main_panel.m le for a range of betha_end=50 (dened in
wingdef.m). Then, tables for sideslip angles from -5... 5 are made in
0.1 steps. This may take some time (one or two minutes).
5. In the function calculate_force_and_moment_avio_panel.m write the
correct ap vector.
6. Start the simulator by Andr Noth modied by Flavio Gohl for VLM
calculations. In the solver preferences of the simulator, set the step time.
7. Start
8. The values can be visualised with running show_uav_nal.m. The sideslip
angles are saved in a vector betha_in, it can be plotted versus the time if
desired.
Errors which occur during simulation:
The sideslip angle is too high, so that no interpolation is possible
increase the range of betha_end in wingdef.m and try again, or mod-
ify your geometry.
The ap vector is not dened correctly in the le calculate_force_and_moment_avio_panel.m
66
Appendix
Appendix C) Design Criteria
The most important design rules for ying wings, recommended by the author.
1. Choice of a prole type. For ying wings it is recommended to choose a
prole with a constant pressure point. This means that the momentum
coecient at the 1/4 line should be zero or slightly positiv. Good proles
with a high CL
max
at low reynolds numbers are the MH45 and MH60.
2. For a pitch stable wing, at rst it is important to have the wings centre
of gravity in front of the aerodynamic neutral point.
3. With a simple apped wing, it must be decided at which C
L
the airplane
has to y in trimmed ight. It is recommended to choose a high C
L
value
with a ap deection, so that the wing without ap deection is in fast
ight. The reason is, that at slow speed the induced drag is very high
(Cd
ind
= const C
L
2
) and the viscous drag small (because the dynamic
pressure is small), and therefore a ap deection is not too expensive. In
fast ight, the viscous drag must be extremely small, and consequently it
is best to y without ap deection!
4. At the chosen C
L
it must be decided, what the shape of the circulation
distribution should look like. For minimal induced drag, it is elliptic. This
could cause a problem in combination with point 1, as the center of gravity
is not necessarily in front of the neutral point. In this case it is an iterative
process to nd a geometry which can fullll both requirements of point 1
and 3. For an ecient design it is recomended, to have an aspect ratio of
0.4 to 1.0. Only in this range both requirements can be fulllled. [7]page
107
5. For good stall behaviour, it is advantageous, to place the center of gravity
in front of the the C-Point. The C-Point is the point of lift distribution
which is proportional to the local chord. This setting is also possible if
the aspect ratio is between 0.4 and 1.0 [7]page 105
6. Reynolds numbers: It is important to have a closed look at the local
reynolds numbers, as they may decrease severely at the tip of the wing.
However, even at te wing tip, the reynolds number must be suciently
high. In case of this problem it is better to twist the prole a little and
give more chord length for better reynolds numbers.
67
Appendix
Appendix D) Estimation of Initial Condition
In this part, a method to nd the speed and angle theta in equilibrium at a
desired angle of attack is described. The method presented here saves a lot of
simulation time for nding the equilibrium point. The following points must be
calculated, and written in the les initial.m and wingdef.m.
1. A desired angle of attack must be chosen (for example at best glide ratio
or at best sink rate)
2. The correct centre of gravity at the desired angle of attack must be cal-
culated.
3. With the CL value (which is estimated or calculated) at the desired angle
of attack, the velocity can be calculated:
v
norm
=

2 m g
CL A
(29)
v
initial
=

cos() v
norm
0
sin() v
norm

(30)
4. From the glide ratio the glide angle can be calculated:

Glide
= arctan

CD
CL

(31)
5. The initial angle Theta can be calculated:
= (
Glide

Attack
) (32)
68

Vous aimerez peut-être aussi