Thesis

Erasmus University College
Faculty of Industrial Sciences

Comparison of Dierent Flutter Speed
Prediction Algorithms
by
Thomas De Kerf
Promotor: Prof. Dr. Ing. T. De Troyer
A thesis submitted in partial fulllment of the requirements for the degree of Master in
Applied Engineering - Electromechanics
Academic year 20112012
Acknowledgements
I would like to thank Mr. De Troyer for the excellent guidance during my two years at Erasmus.
Through his rigorous demands, I am proud to present this thesis.
This thesis would not be possible without the support of my girlfriend. She has stood by
me, even when I quit my steady job and told her I wanted to get my Masters degree.
Lastly, I oer my regards and a sincere thank you to all of those who supported me in any
aspect (from help with INVENTOR to allowing me to borrow their car to go to school) during the
completion of my thesis and Erasmus years in general.
Antwerp, June 1, 2012
De Kerf Thomas
i
Contents
Preface i
Contents iii
Abbreviations v
Nederlandse samenvatting vii
English abstract ix
List of Figures xi
List of Tables xiii
1 Introduction 1
1.1 What is utter? . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
1.2 Flutter in the design process of an aircraft model . . . . . . . . . . . . . . . . . . 2
1.2.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
1.2.2 Ground Vibration Testing . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
1.2.3 Flight Flutter Testing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
1.3 Quasi-steady Aerodynamics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
1.4 Unsteady Aerodynamics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
1.4.1 Theodorsen Unsteady Aerodynamics . . . . . . . . . . . . . . . . . . . . . 6
1.5 Mathematical Aeroelastic Model . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
1.6 pk-Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
1.7 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
2 Flutter Prediction Methods 11
2.1 Damping Extrapolation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
2.2 Flutter Margin . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
2.3 Envelope Function . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
2.4 Frequency Domain Prediction Algorithm . . . . . . . . . . . . . . . . . . . . . . . 14
3 Numerical Aspect 17
3.1 Matlab simulation model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
3.1.1 Spring Constants . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
3.1.2 PK-analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
3.1.3 Eect of Noise . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
3.2 Damping Extrapolation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
3.2.1 Eect of noise . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
iii
3.3 Flutter Margin . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
3.3.1 Eect of noise . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
3.4 Envelope Function . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
3.4.1 Eect of noise . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
3.5 Frequency Domain Prediction Algorithm . . . . . . . . . . . . . . . . . . . . . . . 28
3.5.1 Eect of noise . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
3.6 Summary of results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30
4 Experimental Aspect 31
4.1 Design of the Test Setup . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32
4.2 First Windtunnel Test . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32
4.3 Second Windtunnel Test . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32
4.3.1 Damping Extrapolation . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
4.3.2 Flutter Margin Method . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34
4.3.3 Envelope Function . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
4.3.4 Frequency Domain Prediction Algorithm . . . . . . . . . . . . . . . . . . . 36
4.4 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37
5 Experimental vs Numerical results 39
5.1 Flutter Speed . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40
5.2 Simulated model vs Experimental setup . . . . . . . . . . . . . . . . . . . . . . . 40
5.3 Prediction methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40
6 Conclusions 41
6.1 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42
6.2 To Conclude . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42
6.3 Further Research . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42
Bibliography 45
A Derivation of Routh-Hurwitz Stability Criterion 49
B Contents of the CD 51
C JAR 25.629 Flutter, Deformation and failsafe criteria 53
D Rational Fraction Polynomial 55
E Numerical results: tables 57
F Experimental results: tables 59
Abbreviations
GVT Ground Vibration Test
PRM Phase Resonance Method
FRF Frequency Response Function
SDOF Single Degree Of Freedom
MDOF Multiple Degrees Of Freedom
RFP Rational Fraction Polynomial
NACA National Advisory Committee for Aeronautics
LCO Limit Cycle Oscillations
v
Nederlandse samenvatting
Flutter is een vorm van dynamische aeroelasticiteit die zich uit in een onstabiele,
zelfgenduceerde trilling waarbij de structuur onderhevig is aan extreme belastingen. Het is
voor ieder nieuw ontwikkeld type vliegtuig nodig om een luchtwaardigheidscerticaat te beha-
len. Een van de testen is een ight utter test maar omdat deze testen voor een reeel risico
zorgen, is het nodig om een methode te vinden die het proces veiliger maakt.
In deze thesis is een vergelijkende studie van de verschillende methodes om utter te voor-
spellen uitgevoerd. Dit zowel op een experimentele als analytische manier. De verschillende
methodes zijn: de klassieke demping extrapolatie, de envelope functie, de Flutter Margin en een
nieuw voorspel algoritme in het frequentiedomein. Met betrekking tot het analytische aspect is
een simulatiemodel gemaakt in MATLAB, vertrekkende van een tweevrijheidsgraden model. Ver-
trekkende van dit simulatiemodel genereren we vervolgens alle parameters die nodig zijn voor
de verschillende utter voorspelmethodes. Voor de experimentele methode vertrekken we vanaf
de thesis van Ertveldt (2011) en houden we rekening met de conclusies en aanbevelingen om een
nieuwe opstelling op te bouwen.
De analytische resultaten tonen aan dat het voorspel algoritme in het frequentiedomein
de beste resultaten geeft, zelfs na het toevoegen van ruis. Alle methodes geven aanvaardbare
resultaten maar er zijn toch een aantal factoren die de resultaten benvloeden zoals: het simu-
latiemodel is een aangepaste versie van een quasi-steady aerodynamisch model en het nieuwe
frequentie domein algoritme maakt daar optimaal gebruikt van, de envelope functie ondergaat
een bewerking meer (inverse fourier transformatie) die extra fouten gaat opleveren, de envelope
functie en demping extrapolatie zijn in het voordeel omdat we op voorhand de utter snelheid
weten en dus de vrij de beste polynoom kunnen kiezen.
Om de experimentele resultaten te verkrijgen is een opstelling gemaakt waarbij we utter
kunnen opwekken op een repetitieve manier. De opstelling is aanpasbaar zodat we zonder extra
materiaal verschillende uttersnelheden kunnen bereiken. De in dit werk verkregen experimen-
tele data vertonen een grote spreiding wat het moeilijk maakt om de verschillende methodes
met elkaar te vergelijken. De beschikbare data tonen aan dat de envelope functie samen met
de demping extrapolatie methode de beste resultaten geeft, gevolgd door het frequentie domein
voorspel algoritme. De utter margin geeft de slechtste resultaten.
vii
English abstract
Flutter is a phenomenon in the eld of dynamic aeroelasticity. A formal denition of utter is:
an unstable self-excited vibration in which the structure extracts energy from the air stream and
often results in catastrophic failure. Every new type of aircraft has to obtain an airworthiness
certicate and one of the tests to obtain this certicate is a ight utter test. Due to the fact
that these tests cause a real risk, it is advantageous to increase the safety of these ight tests.
This can be done by having a more accurate prediction of the utter speed.
The objective of this thesis is to compare dierent methods to predict the onset of utter
in an experimental and analytical manner. The dierent methods are: the classic damping ex-
trapolation, the envelope function, the utter Margin method and a newly developed frequency
domain parametric estimation method. A theoretical model is programmed using mathematical
software (MATLAB) to provide each method with the necessary input data. Regarding the exper-
imental part, we depart from the thesis of Ertveldt (2011) and based on hist recommendations
a new setup is built.
When considering the numerical results, we nd that the frequency domain prediction
algorithm has the best results, even afted adding extra noise. All of the methods provide
acceptable results but there are a couple of factors that will inuence the results such as: the
simulation model uses a modied quasi-steady aerodynamic model and so does the frequency
domain prediction algorithm, the envelope function has to undergo an extra operation (inverse
fourier transform) that introduces an error, the envelope function and damping extrapolation
benet from the fact that we already know the utter speed so we freely choose the best result
and polynomial.
To obtain the experimental results, a setup is built where we can induce utter in a
repetitive manner. This setup is adjustable, without the need of extra of dierent components,
to achieve dierent utter speeds. The obtained experimental results have a large variation of
modal parameters, which makes it harder for the dierent prediction methods to reach a good
prediction. The results show that the damping extrapolation method and the envelope function
get the best results, followed by the frequency domain prediction algorithm. The utter margin
method proves to have the worst results.
ix
List of Figures
1.1 Collars aeroelastic triangle. Reprint of Figure 1 from Wright and Cooper (2008) 2
1.2 Flutter in the Design process ow of an aircraft. Reprint from Debille (2011) . . 3
1.3 Ground vibration testing strategy. Reprint from Goge et al. (2007). . . . . . . . . 4
1.4 Typical Flight Clearance Envelope. Reprint from Cooper (2003) . . . . . . . . . 5
1.5 Unsteady lift for an airfoil with constant k. Reprinted from Cooper (2003) . . . . 7
1.6 classic pitch-plunge aeroelastic model. Reprint from Hodges and Pierce (2002) . 8
2.1 Comparison of Matlab Envelope vs Hilbert Transform Envelope . . . . . . . . . . 15
3.1 Pitch movement of wing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
3.2 Variation of k
with . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
3.3 Visualisation of the characteristic wing parameters . . . . . . . . . . . . . . . . . 19
3.4 pk-analysis of the wing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
3.5 Pk analysis for dierent values of k
. . . . . . . . . . . . . . . . . . . . . . . . . 20
3.6 True damping values of the simulated model with d
= 0.128 . . . . . . . . . . . 21
3.7 Eect of noise on the FRF . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
3.8 Example of damping extrapolation . . . . . . . . . . . . . . . . . . . . . . . . . . 22
3.9 Example of the Flutter margin method . . . . . . . . . . . . . . . . . . . . . . . . 24
3.10 Example of the Envelope Function - polynomial curve tting . . . . . . . . . . . 26
3.11 Example of the Envelope Function - time domain . . . . . . . . . . . . . . . . . . 27
3.12 Example of Freq. domain prediction algorithm . . . . . . . . . . . . . . . . . . . 28
4.1 Old and new setup . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32
4.2 Flutter margin method using experimental results . . . . . . . . . . . . . . . . . 34
4.3 Time signal of 2 dierent sensors . . . . . . . . . . . . . . . . . . . . . . . . . . . 36
4.4 Example of frequency domain utter prediction algorithm . . . . . . . . . . . . . 36
5.1 PK-analysis with d
= 0.09m/s . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40
xi
List of Tables
3.1 Parameters for the pk-analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
3.2 Numerical results using the damping extrapolation . . . . . . . . . . . . . . . . . 23
3.3 Numerical results using the damping extrapolation, including noise . . . . . . . 23
3.4 Numerical results using the utter margin . . . . . . . . . . . . . . . . . . . . . . 24
3.5 Numerical results using the utter margin, including noise . . . . . . . . . . . . 25
3.6 Numerical results using the envelope function . . . . . . . . . . . . . . . . . . . . 27
3.7 Numerical results using the envelope function, including noise . . . . . . . . . . 28
3.8 Numerical results using the frequency domain prediction algorithm . . . . . . . . 29
3.9 Numerical results using the frequency domain prediction algorithm, including noise 29
4.1 Experimental results using the damping extrapolation . . . . . . . . . . . . . . . 33
4.2 Experimental results using the utter margin . . . . . . . . . . . . . . . . . . . . 34
4.3 Experimental results using the envelope function and sensor 1 . . . . . . . . . . 35
4.4 Experimental results using the envelope function and sensor 2 . . . . . . . . . . 35
4.5 Experimental results using frequency domain prediction algorithm . . . . . . . . 36
A.1 The general Routh-Hurwitz array structure . . . . . . . . . . . . . . . . . . . . . 49
E.1 Table with errors for every polynomial and range generated using damping ex-
trapolation (d
= 0.128) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 57
E.2 Table with errors for every polynomial and range generated using damping ex-
trapolation (d
= 0.128) with noise . . . . . . . . . . . . . . . . . . . . . . . . 58

E.3 Full results of the damping extrapolation (d
= 0.128) method on the experi-

mental data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 58
xiii
CHAPTER 1
Introduction
Chapter Abstract This chapter is the starting point of the thesis. The most essential topics
are explained so that the reader is aware of where this thesis is situated in the eld of aeronautical
engineering. Concepts necessary to understanding the dierent utter methods are explained in
this chapter.
1
Chapter 1. Introduction
1.1 What is utter?
To discuss utter we need to set the background rst. Flutter is a phenomenon in the eld of
aeroelasticity. Aeroelasticity is the study of three forces that are intertwined for a exible struc-
ture and the phenomena that can result (Wright and Cooper, 2008). These three forces consist
of Inertial forces, Aerodynamic forces and Elastic forces. Aeroelasticity can be classied into
two categories (Hodges and Pierce, 2002). We have static aeroelasticity where the interactions
of aerodynamic and elastic forces are considered. There is also dynamic aeroelasticity, where all
three forces are taken into account. One of the phenomena that can arise because of dynamic
aeroelastic instabilities is utter. A formal denition of utter given by Wright and Cooper
(2008) is: an unstable self-excited vibration in which the structure extracts energy from the air
stream and often results in catastrophic structural failure.
Figure 1.1: Collars aeroelastic triangle. Reprint of Figure 1 from Wright and Cooper (2008)
1.2 Flutter in the design process of an aircraft model
1.2.1 Introduction
During the design process of an aircraft the manufacturer tries to estimate the aeroelastic be-
havior. This is accomplished by performing wind tunnel tests on scale models and develop-
ing mathematical models. These models do not exclude each other, they are complementary.
The computational methods are updated using results obtained from wind tunnel experiments.
Computational methods combine structural models and aerodynamic models into an aeroelas-
tic model that can be solved (Cooper, 2003). These dierent techniques are used to test the
design of the aircraft and if necessary adjust certain parameters early in the design process. A
physical prototype is then built and a ground vibration test (GVT) is performed. The objective
of this test is to obtain modal parameters of the aircraft. These parameters can then be used
to analytically predict the utter speed. They can also be used to update existing aeroelastic
models (Peeters et al., 2009). The results of the GVT are necessary to advance to the next
stage, namely the ight tests. These tests are necessary to obtain an airworthiness certicate
accredited by the appropriate aviation authorities. An example of a typical design process for
an aircraft and where utter testing is done, can be seen in gure 1.2.
An example specically for utter the Joint Aviation Authorities (JAR) (2007), section 25.629:
Aeroelastic stability, requirements states:
The airplane must be designed to be free from aeroelastic instability for all congura-
tions and design conditions within the aeroelastic stability envelopes as follows: For
normal conditions without failures, malfunctions, or adverse conditions, all combi-
nations of altitudes and speeds encompassed by the V
D
/M
D
versus altitude envelope
2
Section 1.2 Flutter in the design process of an aircraft model
enlarged at all points by an increase of 15 percent in equivalent airspeed at both con-
stant Mach number and constant altitude. In addition, a proper margin of stability
must exist at all speeds up to V
D
/M
D
and, there must be no large and rapid reduc-
tion in stability as V
D
/M
D
is approached. The enlarged envelope may be limited to
Mach 1.0 when M
D
is less than 1.0 at all design altitudes.
The full JAR 25.629 specication regarding to utter is in appendix C.
Since utter testing is done at the end of the design process of the aircraft, it would be a
enormous nancial setback if the utter test indicates that a substantial redesign needs to be
done. That is why a lot of eort is taken to predict the dynamic behaviour of the aircraft. This
also shows that it is important to have dierent methods of predicting the utter speed, so that
you do not depend on one opinion .
Now that there is a step by step overview in the design of an aircraft in regards to the aeroelastic
behaviour, we will go into more detail on the subjects that are of importance to the thesis.
Figure 1.2: Flutter in the Design process ow of an aircraft. Reprint from Debille (2011)
1.2.2 Ground Vibration Testing
A ground vibration test (GVT) on the prototype of a new aircraft is often regarded as necessary
to update the mathematical model of the aircraft in order to make reliable utter predictions for
ight tests (Goge et al., 2007). The experimental modal analysis of a prototype aircraft is very
dierent from vibration tests on smaller objects. For instance; in testing a large aircraft it is no
longer possible to work with a small number of sensors. Furthermore, the excitation can not be
realized with simple impact hammers. Several large electrodynamic exciters have to be operated
simultaneously in order to excite all modes of vibration with sucient amount of energy but
an excitation with too much energy will result in a non-linear response of the structure. In
addition, for new aircraft prototypes the regulations of the Airworthiness Authorities require
such vibration tests, the results being used in the certication process. Therefore, it has to be
assured that all modes in the requested frequency range are identied and that the accuracy of
the modal parameters is as high as possible. The test programmes are repeated for a number
of dierent congurations such as full and empty tanks, ...
The use of the Phase Resonance Method (PRM) or broadband testing has been almost
exclusively used for GVT on large aircraft because this method is generally well suited for the
separation of closely spaced modes. In PRM, essentially single sine excitation at a natural
frequency is used. By carefully selecting the shaker locations and the phase relation between
the sine excitation signals, the aircraft is forced to act as a single degree of freedom system
and the vibration response only contains a contribution from the mode of interest. The main
disadvantage of this method is that it is very time consuming. Therefore, the phase resonance
method is complemented and partially substituted by so-called Phase Separation Method
3
Figure 1.3: Actual Aircraft Ground Vibration Testing Strategy. Reprint of gure 2 in Goge
et al. (2007)
(PSM), which nds the aircraft modes by evaluating frequency response functions (FRFs). The
idea is that most modes are extracted from a fast PSM, but that critical modes are identied
based on PRM (Peeters et al., 2009). Modes are dened as critical when: they show non-
linear behaviour; they are important for utter calculations; or they dier signicantly from the
predictions.
A large variety of shaker excitation signals can be used to experimentally determine the
aircraft broadband FRFs, which are required in phase separation testing. These excitation
signals include harmonic signals like discrete stepped sines, periodic signals like multi-sines or a
periodic chirp (i.e. very fast swept-sine), and transient signals like an impulse, a (burst) random,
or a swept-sine signal.
In order to obtain a free-free boundary condition there are several techniques that can be used:
attening the tyres, hanging the plane on elastic wired or an airbag system supporting the
landing gear (Kehoe, 1987).
1.2.3 Flight Flutter Testing
In the early years of aviation, no formal utter testing of full-scale aircraft was carried out. The
aircraft was simply own to its maximum speed to demonstrate the aeroelastic stability of the
vehicle. The rst formal utter test was carried out by Von Schlippe (1936) in Germany. His
approach was to vibrate the aircraft at resonant frequencies at progressively higher speeds and
plot amplitude as a function of airspeed. A rise in amplitude would suggest reduced damping
with utter occurring at the asymptote of theoretically innite amplitude (Kehoe et al., 1995).
The classical approach to utter testing has not changed dramatically during the last 60 years,
as can be seen in gure 4. At each ight test point, three separate procedures are performed
(Cooper, 2003):
1. The aircraft is excited in some manner and the vibration response measured.
2. The data are curve-tted using system identication methods and the modal parameters
estimated.
4
Section 1.3 Quasi-steady Aerodynamics
3. The decision is made to progress to the next ight test point.
Figure 1.4: Typical Flight Clearance Envelope. Reprint from Cooper (2003)
Despite the numerous advances over the past 60 years, a ight utter test is still a dangerous
operation because of a number of reasons:
one must y fairly close to the actual utter speed to detect imminent instabilities.
Subcritical damping trends are hard to extrapolate to predict stability at higher speeds.
Aeroelastic stability may change abruptly from stable to an unstable condition with very
small speed increase.
Not only is a utter test dangerous, it is also time consuming and therefore very expensive. The
key research objective should be to increase safety by reducing the number of required ight
tests or test time during the ight test.
1.3 Quasi-steady Aerodynamics
Since we are dealing with dynamic aeroelasticity we cannot use steady aerodynamic theory.
When we consider an airfoil in an airow, the airfoil will undergo a motion relative to the
airow. This causes a change in aerodynamic moments and forces acting on the airfoil. One
(simplied) way to consider these changing forces and moments is to use the quasi-steady aero-
dynamics theory. This theory assumes that at any instant of time the aerofoil behaves with the
characteristics of the same aerofoil moving with constant heave and/or pitch velocities equal to
the instantaneous values. Dimitriadis (2010) indicates that quasi-steady models only account
for following factors to account for aerodynamic forces:
Horizontal airspeed U at angle of attack () to airfoil
Airfoil plunge speed
Normal component of pitch speed
Local velocity induced by the vorticity around the airfoil
The eect of the wake is however ignored.
1.4 Unsteady Aerodynamics
According to Hodges and Pierce (2002) the ow is unsteady because of three reasons:
The relative wind vector is not xed in space, due to the unsteady motion of the airfoil
5
The airfoil motion disturbs the ow, shedding a vortex at the trailing edge. The downwash
from this vortex, in turn, changes the ow that impinges on the airfoil.
The motion of the airfoil accelerates air particles near the airfoil surface, thus creating the
need to account for the resulting inertial forces
Thus, in order to obtain a more accurate prediction of the utter speed, it is necessary to include
unsteadiness in the aerodynamic theory. This demands a far more sophisticated aerodynamic
theory. Unfortunately, development of unsteady aerodynamic theories is no small undertaking.
An unsteady aerodynamic theory consists of two dierent phenomena: A circulatory part and
a non-circulatory part. The circulatory part is caused by the vorticity of the air owing over
the airfoil. When placed at an angle of attack, the velocity proles above and below the airfoil
will be dierent. The resulting ow eld can be represented by a constant velocity ow plus a
vortex. The strength of the vortex changes with time because we deal with a dynamic system.
Considering a two dimensional potential ow, the Helmholtz theorem states that if a clockwise
vortex develops about the airfoil, there must be a counterclockwise vortex of the same strength
shed into the ow. As this counterclockwise vector moves along, it changes the ow eld. The
rate of change in the ow eld is a function of the strength of the vortex and the distance to the
airfoil. It should become clear that calculating an unsteady aerodynamic ow is not that easy
and would require that we know the strength of each vortex. One of the assumptions we can
make is that the shed vortex moves along with the ow, so we can estimate the eect of these
vortices. The non-circulatory part is caused by apparent mass and inertia eects. When the
airfoil has a non-zero acceleration, it carries the surrounding air with it. This air has a mass so
there is an inertial force opposing the acceleration of the airfoil.
1.4.1 Theodorsen Unsteady Aerodynamics
The unsteady aerodynamics theory used in this thesis is the theory developed by Theodorsen
and Center (1935). There are three major assumptions made:
Flow is always attached, i.e. , the motion amplitude is small
The airfoil is a at plate
The wake is at
If the motion is small then the wake has little inuence. If we want to treat asymmetrical wings,
it is possible to do so by introducing an airfoil with control surface (3DOF).
To fully understand Theodorsens theory, a new parameter, called the reduced frequency (k) is
introduced. This value of k is a function of the phase shift and amplitude attenuation between
the unsteady and quasi-steady lift values. In other words, it represents the unsteadiness of
the ow.
k =
b
V
(1.1)
with V as the speed, is the frequency of the airfoil (in rad/s) and b = c/2 as the semi-chord.
k can be interpreted as the number of oscillations that the airfoil endures during the time
the ow passes over the full chord, in radians. Figure 1.5 illustrates the dierence in phase and
amplitude for a xed value of k between quasi-steady aerodynamics and unsteady aerodynamics.
Theodorsens function (equation 1.2) is used to model the changes in amplitude and phase of the
sinusoidal unsteady aerodynamic forces relative to the quasi-steady forces for dierent reduced
frequencies. Another way of describing theodorsens function is to think of it as if it were a
lter that modies the input to a system (i.e. the quasi-steady lift for aerofoil oscillations at
some frequency) to give an output (i.e. the unsteady air forces) depending upon the reduced
6
Section 1.5 Mathematical Aeroelastic Model
Figure 1.5: Unsteady lift for an airfoil at a xed reduced frequency. Reprinted from Cooper
(2003)
frequency. The function C(k) is a complex valued function (since both amplitude and phase
must change) of the reduced frequency parameter k.
C(k) = F(k) +iG(k) =
H
2
1
(k)
H
2
1
(k) +iH
2
0
(k)
(1.2)
where H
2
n
(k) are Hankel functions of the second kind and i is the imaginary unit. An approximate
expression, found by Jones (1940) is
C(k) = 1
0.165
1
0.045
k
i
0.335
1
0.03
k
i
, k 0.5
= 1
0.165
1
0.041
k
i
0.335
1
0.032
k
i
, k > 0.5 (1.3)
The function C(k) is sometimes called the lift-deciency function because it reduces the magni-
tude of unsteady lift relative to steady lift. According to Theodorsens theory lift and moment
are given by
1
L = 2
UbC (k)
_
h +U +b
_
1
2
a
_

_
. .
Circulatory
+
b
2
_
h +U

ba
_
. .
NonCirculatory
(1.4)
M1
4
=
b
3
_
1
2
h +U

+b
_
1
8

a
2
_
_
(1.5)
The equations for the lift (equation 1.4) consist of 2 parts: a circulatory part (dependent on
C(k)) and a non-circulatory part.
1.5 Mathematical Aeroelastic Model
In this thesis we treat classical binary utter as dened in Fung (1994); Dowell et al. (1995);
Bisplingho et al. (1996); Hodges and Pierce (2002). Classical binary utter occurs when the
aerodynamic forces associated with motion in two modes of vibration cause the modes to couple
in an unfavourable manner. Since the aerodynamic forces depend on the speed of the structure
(or the wind speed of the wind tunnel), there is a critical utter speed. Above this critical utter
speed the oscillations are damped, below this speed the oscillations are negatively damped and
unstable oscillations occur. The commonly used binary aeroelastic model is a pitch plunge two
degree of freedom (DOF) model as displayed in gure 1.6
In order to simulate a 2 DOF model we depart from the equations of motion with no damping
2
M q +K = 0 (1.6)
1
A full deriviation of the theodorsen function can be found in Dimitriadis (2010) or Theodorsen and Center
(1935)
2
The following equations are based on Dimitriadis (2010)
7
Figure 1.6: classic pitch-plunge aeroelastic model. Reprint from Hodges and Pierce (2002)
with q =
. We have to include an aerodynamic force since this is an aeroelastic problem
m S
S I
K
h
0
0 K
L
M1
4
(1.7)
for sinusoidal motion,we can assume

h =

h
0
e
iwt
qnd =
0
e
iwt
so that the equation becomes
2
m+K
h

2
S
S
2
I
+K
h
0
e
iwt
=
L
M1
4
(1.8)
When we substitute L and M1
4
using Theodorsens equations (equations 1.4 and 1.5)
K
h
2
m+UcC (h) i
2
b
2
2
S +b
2
Ui +b
2
_
x
f

c
2
_
2
+UcC (h)
_
3
4
c x
f
_
i
2
S Uec
2
C (h) i
+
_
x
f

c
2
_
b
2
2
K
2
I
+
_
3
4
c x
f
_
b
2
Ui
Uec
2
C (h)
_
U +
_
3
4
c x
f
_
i
_
_
x
f

c
2
_
2
b
2
b
4
8

2
. .
D
h
0
= 0 (1.9)
This equation (1.9) is only valid when the motion of the wing is sinusoidal. That type of
oscillation is only possible when either:
The airspeed is zero and there is no structural damping
The wing is forced into sinusoidal movements
The wing is at a critical utter condition
This last case is the only one we are interested in. It means that when we solve the D matrix
(in equation 1.9) to nd the utter speed U and frequency . Since we know that Re(D) = 0
and Im(D) = 0, we have two equations with two unknowns.
8
Section 1.6 pk-Analysis
1.6 pk-Analysis
One of the assumptions made in the Theodorsens theory is a sinusoidal motion. This means that
the Theodorsen equations are only valid when there is no damping present. When there is no
damping in the system it means that the critical utter point is reached. Therefore variation of
the damping with speed is only correct for one value, namely where the damping equals to zero
. However it can be useful for the aeroelastic engineers to predict this variation. For example,
it is essential to know if there is hard or soft utter present. Hard utter is a sudden decrease
in damping whereas soft utter has a steadily decreasing damping. Keeping this in mind, there
are a few engineering solutions to track the damping versus the airspeed. In this thesis, I have
chosen to use the pk-method since it is widely known and often used in aeroelasticity text books.
The pk method uses the standard equations of motion as seen in equation 1.8. Coupled with
Theodorsens aerodynamic forces 1.4. Then we dene
q =
h
a
(1.10)
When the equations of motion (EOM) are transformed to the laplace domain and formula 1.1
is substituted into the previous equation, we obtain
_
p
2
M+K
1
2
U
2
Q(ik)
_
q = 0 (1.11)
With Q(ik) as the aerodynamic matrix and p
3
as a complex number.
Q(ik) =
_
_
4C(k)ik + 2k
2
2C(k) 2bik
4C(k)
_
3
4
c x
f
_
ik + 2b
2
k
2
4ecC(k)ik 2
_
x
f

c
2
c
_
k
2
2
_
3
4
c x
f
_
bik + 4ecC(k)
_
3
4
c x
f
_
ik
+2
_
x
f

c
2
c
_
2
k
2
+
b
2
4
k
2
+ 2ec
2
C(k)
_
_
(1.12)
Procedure for solving the pk-analysis
1. Choose an airspeed (U).
2. Choose a degree of freedom.
3. Choose an initial value of and thus k. As an initial value can be
_
K
h
/m for the
plunge mode and
_
K
/I
for the pitch mode.

4. Calculate the eigenvalues of p of equation 1.11.
5. Sort the eigenvalues in ascending order of imaginary part.
6. Choose a new value of equal to the imaginary part of the sorted eigenvalue and repeat
step 4.
7. Stop this iteration when the value of converges.
8. Repeat from step 2 for all the degrees of freedom.
9. Repeat from step 1 for all the airspeeds.
3
The Laplace transform of a function f(t) is the function F(p)
9
1.7 Conclusion
In this chapter we introduce the concepts of aeroelasticity and situate the research objective,
namely to compare dierent types of utter prediction algorithms. We touched upon the im-
portance of utter testing and how utter tests are performed. After that, some key subjects in
the eld of aerodynamics (quasi-steady , theodorsen and unsteady) that we will be using later
on in this work, are explained . We end with a detailed explanation of the pk-analysis.
10
CHAPTER 2
Flutter Prediction Methods
Chapter Abstract In the next chapter we get to the heart of the thesis namely the ight
utter prediction algorithms. Each of the dierent methods to determine utter will be treated
in detail.
11
Chapter 2. Flutter Prediction Methods
2.1 Damping Extrapolation
This is the most straightforward method to predict the utter speed. Simply stated, if the
damping of at least one mode becomes zero, there will be an onset of utter. In a typical ight
utter test, the damping ratios are evaluated at a number of sub critical speeds using system
identication methods. To calculate the utter speed we extrapolate the damping curve to see
where the damping becomes zero, this speed is then known as the utter speed. Care must be
taken when choosing the order of the polynomial tting curve because the possible occurrence
of hard utter. Hard utter is a form of utter where the damping quickly decreases to zero
at a certain velocity.
An example of an experiment gone wrong is found in Bennett and Abel (1981). Where a
drone was lost despite the online utter monitoring. The accuracy of predicting the utter speed
depends on how well the damping ratios are estimated. In Cooper (1995) there are a number of
parametric estimation techniques evaluated for their use in a ight utter test.
2.2 Flutter Margin
This method was rst presented by Zimmerman (1964). It introduces a more fundamental
stability criterion than just tracking the damping present in the system. The utter margin
employs the Routh-Stability criterion (Dorf and Bishop, 2008) applied to the characteristic
equation of a two degree of freedom system in the laplace domain.
The equations of motion for an unforced aeroelastic system are in standard form
M q+C q+Kq=0 (2.1)
with M, C, K as the generalised mass, damping and stiness matrix. If we choose solutions of
the form q = q
0
e
t
, with as eigenvalues of the system and q
0
as an arbitrary constant, then
the characteristic equation becomes
a
4
s
4
+a
3
s
3
+a
2
s
2
+a
1
s +a
0
= 0 (2.2)
or
(s
1
)(s
2
)(s
3
)(s
4
) = 0 (2.3)
(2.4)
with
1,2
=
1
i
1
(2.5)
3,4
=
2
i
2
(2.6)
We can introduce a criterion in which a two degree of freedom system is stable by using the
Routh-stability criterion. In appendix A these equations are derived. We consider a
4
= 1 and
divide by a
2
3
_
a
2
(
a
1
a
3
) (
a
1
a
3
)
2
a
0
_
> 0 (2.7)
The equation above has to be positive for the system to be stable. We can derive the equation
in such a way that it becomes an indication of the utter stability. This quantity is called the
12
Section 2.3 Envelope Function
utter margin and can be expressed by combining equations 2.2 trough 2.7 as
1
F =
__
2
2
2
1
2
_
+
_
2
2

2
1
2
__
2
+ 4
1
2
_
_
2
2
+
2
1
2
_
+ 2
_
2
2

2
1
2
_
2
_
_
_
2
+1
__
2
2
2
1
2
_
+
_
2
2

2
1
2
_
2
_
2
(2.8)
The application of the method is quite straightforward. For a two degree of freedom system,
the response to a known input at a particular airspeed is recorded and the eigenvalues of the
system are calculated. In Zimmerman (1964) it is shown that the utter margin formula can be
derived further to show that the utter margin is a quadratic function of the dynamic system:
F = B
2
q
2
+B
1
q +B
0
(2.9)
with B
2
, B
1
and B
0
as coecients to be evaluated. This means that if the utter margin is
determined at three dierent speeds, it can be tted by a second order polynomial to predict the
utter speed. However, in practice we need to counteract the eects of experimental uncertainty
so the utter margin in estimated at a wider range of speeds and tted in a least squares way.
On experimental data obtained by tests on the CF-18 (Dickinson (1995)) it is shown that the
Flutter Margin method can be very sensitive to errors or uncertainty in the data.
The original method was for a two mode utter (pitch, plunge) but there are studies (Price and
Lee (1993, 1992)) involving one and three modes of instability. If it is know beforehand which
modes are to cause utter, the utter margin method can be applied with success. If this is not
known, all possible pairs of modes must be examined.
2.3 Envelope Function
The aim of this method, as described in Cooper et al. (1993); Lind (2003), is to compare the
envelopes of the impulse response function (IRF) as speed or Mach number is increased by
inspection and through a simple stability parameter. This method is intended to be comple-
mentary to other prediction methods and can give a quick impression for a change in overall
stability.
Obtaining the decay envelope The normal way of obtaining the envelope is to search
for successive maxima and minima and connect the peaks and troughs. This method can be
disadvantageous if there are only a small number of cycles, also a lot of small peaks due to the
presence of noise can be highly inuential. Therefore an alternative way is proposed by using
the Hilbert transform. This will allow us to generate a signal for which every component is
in quadrature with the measured decay at all frequencies within the measurement range. The
Hilbert transform is calculated by
y
H
(t) = F
1
{Im[Y ()] jRe [Y ()]} (2.10)
with F
1
being the inverse fourier transform. Im[Y ()] and Re [Y ()] are obtained by the
fourier transform of y(t)
F [y(t)] = Re() +jIm() (2.11)
The envelope can be calculated as follows
env(t) =
_
y(t)
2
+y
H
(t)
2
(2.12)
1
There are many dierent versions available of the utter margin but this is the equation as it is rst printed
by Zimmerman (1964)
13
The envelope is easier to evaluate than the entire oscillatory decay of a sensor. One way of
observing the change in stability from one speed to another is to compare the envelope functions
for particular transducers. The envelope decay will have the tendency to be extended when there
is a decrease in damping. One problem with a simple comparison is that the amplitudes can
change for dierent airspeeds due to aerodynamic eects. This can be countered by considering
the shape
2
of the envelope independently of the overall amplitude. An extended decay i.e. lower
damping, would be indicated by an increase in the distance of the center of the area from the
amplitude axis. The time center can be dened as
t =
_
t
max
0
env(t)tdt
_
t
max
0
env(t)dt
(2.13)
where t
max
is the maximum value of time that is the same for all tests for each transducer.

t
will increase as damping decreases. A shape parameter S is introduced
S =
1
t
(2.14)
S will decrease as damping increases. For a SDOF function it may be shown that when
the damping becomes zero, S becomes 2/t
max
. For a MDOF system this is less obvious. If the
damping decreases to zero, the value of S will also decrease rapidly but not completely to the
value of 2/t
max
because of the eect of other damped modes.
In practice the envelope function is most likely to be spoiled by the eects of the response to
the atmospheric turbulence on the decay data. The value of t
max
must be chosen very carefully.
If the value of t
max
is too high or too low it will atten the shape parameter variation.
The question that arises is: why do we need this complicated Hilbert transform? It is easy
to detect the peaks and troughs for a clean (no noise) function but it would require a lot more
code to do this for a signal with noise. This is demonstrated in gure 2.1.
2.4 Frequency Domain Prediction Algorithm
This method, published by De Troyer et al. (2008), identies and extrapolates the coecients of
the characteristic polynomial of the equations of motion. This extrapolation is physically more
justied than the extrapolation of the damping ratios.
Theoretical background Consider a pitch plunge wing section. This system has the follow-
ing characteristic equation in the Laplace domain
_
Ms
2
+ (CUE)s + (KU
2
D)
X(s) = 0 (2.15)
M,K and C are the mass, stiness and structural damping matrices respectively. E and D are
the aerodynamic damping and stiness inuence matrices. D and E are only dependent on the
airspeed and geometric properties of the wing. The eigenfrequencies of this system are the roots
of the characteristic equation
det(
_
Ms
2
+ (CUE)s + (KU
2
D)
) = 0 (2.16)
This determinant is a polynomial in s and U of degree 2N where N is the number of degrees of
freedom of the system.
a
2N
(U)s
2N
+a
2N1
(U)s
2N1
+. . . +a
1
(U)s +a
0N
(U) =
2N
i=0
a
i
(U)s
i
= 0 (2.17)
2
The shape of the envelope is the area between it and the time axis.
14
Section 2.4 Frequency Domain Prediction Algorithm
0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5
1.5
1
0.5
0
0.5
1
1.5
Signal with no noise
time in seconds
A
m
p
l
i
t
u
d
e

IRF with no noise
Matlab Envelope
Hilbert Transform Envelope
0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5
1.5
1
0.5
0
0.5
1
1.5
Signal with 2 % noise
time in seconds
A
m
p
l
i
t
u
d
e

IRF with no noise
Matlab Envelope
Hilbert Transform Envelope
Figure 2.1: Comparison of Matlab Envelope vs Hilbert Transform Envelope
15
Every coecient in a
i
is itself a polynomial function of U of degree 2N i
a
i
= c
2Ni,i
U
2Ni
+c
2Ni1,i
U
2Ni1
+. . . +c
1,i
U +c
0,i
=
2Ni
i=0
c
k,i
(U)
k
(2.18)
In matrix notation
_
U
0
U
1
. . . U
2N
_
c
0,2N
c
0,2N1
. . . c
0,0
0 c
1,2N1
. . . c
1,0
.
.
. 0
.
.
.
.
.
.
0 0 . . . c
2N,0
_
_
=
_
a
2N
a
2N1
. . . a
0
(2.19)
The constant c
k,i
coecients link the coecients of the characteristic polynomial a
i,j
to its
corresponding speed U
j
_
_
U
0
1
U
1
1
. . . U
2N
1
U
0
2
U
2
2
. . . U
2N
2
.
.
.
.
.
.
.
.
.
.
.
.
U
0
j1
U
0
j
. . . U
2N
j
_
_
_
_
c
0,2N
c
0,2N1
. . . c
0,0
0 c
1,2N1
. . . c
1,0
.
.
. 0
.
.
.
.
.
.
0 0 . . . c
2N,0
_
_
=
_
_
a
2N,1
a
2N1,1
. . . a
0,1
a
2N,2
a
2N1,2
. . . a
0,2
.
.
.
.
.
.
.
.
.
.
.
.
a
2N,j
a
2N1,j
. . . a
0,j
_
_
(2.20)
UB = A
If the characteristics equation is identied for dierent ight test speeds, the coecients c
k,i
can
be determined in a least-squares sense.
B = U
+
A (2.21)
Where ()
+
denotes the pseudo inverse.
At least 2N+1 measurements are needed to obtain the (2N+1)(2N+2)/2 unknown coecients
c
k,i
. Once the coecients are known, the characteristic polynomial can be determined for every
speed U
test
. Next step is to obtain the roots of the polynomial and thus the resonance frequencies
and damping ratios.
The results can be presented in multiple ways to examine the stability. The most straight-
forward way is the damping ratio plot but a root locus curve, a utter margin method or a
statistical detection test such as CUSUM (Zouari et al., 2008) is possible.
Frequency-domain modal parameter estimators Verboven et al. (2004) are ideally suited to
identify the coecients of the transfer function. Wind tunnel tests show that the error in utter
prediction are similar to a damping t estimation method Van De Walle et al. (2011). But with
this model the user does not have to choose a polynomial degree for the estimation.
16
CHAPTER 3
Numerical Aspect
Chapter Abstract The next chapter contains the results of the analytical comparison of
the dierent utter prediction methods. We begin by explaining the simulated model used to
generate input data for the dierent utter prediction algorithms. Then we asses every algorithm
individually and present a summary of the obtained results.
17
Chapter 3. Numerical Aspect
3.1 Matlab simulation model
To implement the model in Matlab we have to program equation 1.11 with full Q(ik) matrix
as seen in equation 1.12. Since these equations are in the frequency domain, we can construct
a frequency response function (FRF) for every airspeed U. Once we have this FRF, we can use
it as an input for the dierent solution methods. In order to have comparable results between
the theoretical matlab simulation and the experimental setup, we have to make sure that all the
parameters of both models are the same. Since we will be using the wing, described in Ertveldt
(2011), most of the parameters are xed. The values that we still have to choose are the spring
constants.
3.1.1 Spring Constants
The experimental setup is devised in such a way that the pitch and the plunge motion both use
the same springs (a detailed explanation of the real setup can be found in chapter 4.1). This is
shown in gure 3.1.
Flexural axis
x
h
Figure 3.1: Pitch movement of wing
with d
as the distance between the springs, as the angle of attack and x

h
denotes the dif-
ference between the length of the compressed (or extended spring) and the length of the spring
in uncompressed (or unextended) condition. As a convention a subscript
h
denotes a plunge
movement and subscript

denotes pitch movement. The relation between pitch and plunge
stiness is found to be
F
h
= k
h
x
h
(3.1)
M
= k
(3.2)
k
=
M
=
d
k
h
x
h
tan
_
x
h
d
_ (3.3)
In order to change the pitch spring constant, we have to change the value of d
. We see in
equation 3.3 that the relationship between the and k
is non linear. This means that we have

to choose a representative value for k
, this is done by calculating the surface below the function

and dividing it by the range of x
h
. The relationship between d
and k
is shown in gure 3.2.

The method used for calculation is as follows:
1. A value for k
h
is chosen.
2. We calculate the maximum extension of the springs x
h(max)
. In this case the maximum
extension is limited to the where the springs stop behaving in a linear manner.
3. Using equation 3.3, we calculate k
for every angle of from 0 to 11.

4. The surface below the curve is calculated (using the trapz function) and divided by the
range of x
h
to obtain a representative value of k
18
Section 3.1 Matlab simulation model
0 2 4 6 8 10
64.05
64.1
64.15
64.2
64.25
64.3
64.35
in degrees
T
o
r
s
i
o
n
a
l

s
p
r
i
n
g

c
o
n
s
t
a
n
t

(
k
)

i
n

N
m
/
r
a
d
Variation of k
with (d
is 0.4m)
Figure 3.2: Variation of k
with
3.1.2 PK-analysis
Using the PK-analysis it is possible to predict the variation of damping with the airspeed, when
all the variables are known. But when one variable (spring constant) still has to be chosen, it
is possible to use a trial and error method to determine an optimal value for that particular
parameter. As a result of this approach, a spring constant of 400 N/m with a d
of 0.128 m is
chosen. One of the key aspects of choosing this spring constant and d
is that the system would

not have a hard utter characteristic. Research has shown that when there is truly hard utter,
most of the methods are useless. Another important aspect was the choice for a system where
we can inuence the utter speed by adjusting d
.
The parameters used for the calculation are displayed in table 3.1 and on gure 3.3. Figure 3.4
shows the outcome of the PK-analysis. The utter speed is identied as 30.17 m/s.
Parameter Value
Airfoil: NACA 0012
c 0.15 m
x
f
0.065 m
ec 0.03 m
1.225
kg
m
3
m 3.2 kg
k
h
400
N
m
d
0.01 - 0.4 m
k
0.04 - 64.13
Nm
rad
Table 3.1: Parameters for the pk-analysis
Flexural axis
c/4 ec
c
Quarter chord point
xf
Half chord point
Figure 3.3: Visualisation of the characteristic wing parameters
In equation 3.3 we can see that even when we choose a spring stiness, there is still some margin
to adjust the pitch stiness and thus the utter speed. So the utter speed of 30.17 m/s is only
19
5 10 15 20 25 30 35 40 45
0.1
0
0.1
0.2

X: 17.65
Y: 0
D
a
m
p
i
n
g

r
a
t
i
o
Airspeed (m/s)
5 10 15 20 25 30 35 40 45
0
2
4
6
8
10
N
a
t
u
r
a
l

f
r
e
q
u
e
n
c
y

(
H
z
)
Airspeed (m/s)

Plunge
Pitch
Plunge
Pitch
Figure 3.4: pk-analysis of the wing
valid for one value of d
. It is more useful to perform a pk-analysis on a variety of dierent

values of d
. This is done in gure 3.5 where the y-axis presents the value of d
, which is a
representation of k
, and the x-axis is the wind velocity. A dark grey area is an indication of
negative damping for a corresponding value of V and d
, a light grey area is an indication of

positive damping.
mode1
d
in cm
V
e
l
o
c
i
t
y

(
m
/
s
)

5 10 15 20 25 30
10
20
30
40 Damping (%)
mode2
d
in cm
V
e
l
o
c
i
t
y

(
m
/
s
)

0.02 0.04 0.06 0.08 0.1 0.12 0.14 0.16
10
20
30
40 Damping (%)
20
10
0
10
20
20
10
0
10
20
Figure 3.5: Pk analysis for dierent values of k
Flutter speed In gure 3.6 the true damping of every mode is displayed and we can see clearly
that one mode drops below zero. The speed where the damping becomes zero is called the utter
speed. For this setup it is evaluated at 30.17m/s. The value of d
= 0.128 is specically chosen

because this case had the most soft utter
1
.
1
Soft utter is a type of utter where the damping of the structure decreases gradually.
20
Section 3.1 Matlab simulation model
0 5 10 15 20 25 30 35 40 45
0.01
0
0.01
0.02
0.03
0.04
0.05
0.06
Speed in m/s
d
a
m
p
i
n
g
True damping values with d =0.128

Mode 1
Mode 2
Figure 3.6: True damping values of the simulated model with d
= 0.128
3.1.3 Eect of Noise
In order to have a more realistic system, each of the methods is evaluated with random noise.
This noise is added after the FRF. In equation 3.4 the formula is presented to add noise.
H(i)
noise
= H(i)
original
+
_

2
(randn(n, m) +i(randn(n, m)))
_
(3.4)
with randn as a matlab command to generate normal distributed random matrix and as the
standard deviation. To asses the eect of this noise, the methods are evaluated with and without
noise. An illustration of noise can be found in gure 3.7.
10
0
10
1
10
2
10
3
140
130
120
110
100
90
80
70
60
50
40
frequency (rad/s)
A
m
p
l
i
t
u
d
e

signal with noise
signal without noise
Figure 3.7: Eect of noise on the FRF
21
3.2 Damping Extrapolation
The procedure used in this thesis to determine the utter speed using the damping extrapolation
method is as follows:
1. For every speed of interest we generate the FRFs (4 in this case) using the simulated
model.
2. Using the RFP method we estimate the damping for every mode.
3. The datapoints are curve-tted with a polynomial.
4. For every degree of polynomial used, the roots are determined.
An example of step three can be found in gure 3.8.
0 5 10 15 20 25 30 35
0.01
0.005
0
0.005
0.01
0.015
0.02
Speed in m/s
d
a
m
p
i
n
g
Damping extrapolation d
=0.128 Ustart=7.5475 Ustop=27.171

2nd Order
3rd Order
4th Order
5th Order
6th Order
7th Order
8th Order
True Damping
Figure 3.8: Example of damping extrapolation
As described in chapter 2.1, the order of the chosen degree of polynomial is crucial to de-
termine the utter speed. This can be seen in table 3.2 where dierent ranges of speeds are
simulated and the procentual error is displayed. To obtain the best results the polynomial order
should be 8. The accuracy increases as the degree of the polynomial increases. This is a result
of the type of utter, in this case it is quite hard utter which means that more sudden changes
in the polynomial will provide a better estimate. This can be seen in gure 3.8. When the range
has a higher end point the error will decrease. The full table with all the errors can be found
in table E.1 located in appendix E. Another observation is that all the estimates are positive,
which means that the predicted speed is higher than the true utter speed.
3.2.1 Eect of noise
Because the noise is added after the FRF is generated, the RFP method has a harder time of
predicting the right damping. The estimated damping values will have a larger variation. The
results of the damping extrapolation method with noise can be viewed in table 3.3. Comparing
this table to the no noise one (3.2), we notice a couple of dierences. The best results are not
the result of a polynomial of 8
th
order but of a wider range of degrees. The extra noise will cause
the method to give conservative and over estimate utter speeds, in contrast to the non-noise.
We see that the accuracy of the prediction rises when the end of the range is more towards the
utter speed. The overall impression is that the results are more inaccurate but still acceptable.
22
Section 3.2 Damping Extrapolation
Range (% of
utter speed)
Best Degree Error (in %)
10 - 75 8 1,40
10 - 80 8 1,05
10 - 85 8 0,73
10 - 90 8 0,45
25 - 75 8 1,02
25 - 80 8 0,74
25 - 85 8 0,51
25 - 90 8 0,31
50 - 75 8 0,47
50 - 80 8 0,33
50 - 85 8 0,23
50 - 90 8 0,14
Table 3.2: Numerical results using the damping extrapolation
Range (% of
utter speed)
10 - 80 7 0,73
10 - 85 8 1,16
10 - 90 8 0,49
25 - 75 6 5,95
25 - 80 7 -1,60
25 - 85 6 2,03
25 - 90 7 0,80
50 - 75 4 3,97
50 - 80 6 -1,48
50 - 85 6 0,50
50 - 90 6 0,17
Table 3.3: Numerical results using the damping extrapolation, including noise
23
3.3 Flutter Margin
To use the utter margin method, most of the steps are the same as the damping extrapo-
lation method. For every speed we generate an FRF and then the damping and frequencies
are estimated using the RFP method. A utter margin parameter is then calculated for ev-
ery speed using equation 2.8. This parameter is then plotted with speed and using a second
degree polynomial, we predict the utter speed. An example of this step can be seen in gure 3.9.
When looking at table 3.4 a couple of things stand out. One is that all the errors are negative,
which means that the predicted utter speed is a conservative estimate. Considering we do not
have to choose for a certain polynomial, the results are close to the actual utter speed. The
results improve when the range has a higher end point.
0 5 10 15 20 25 30
1
0
1
2
3
4
5
6
7
8
x 10
5
Flutter margin method
Speed in m/s
D
a
m
p
i
n
g
Figure 3.9: Example of the Flutter margin method
Range (% of
utter speed)
Error (in %)
10 - 75 -2,18
10 - 80 -2,16
10 - 85 -2,00
10 - 90 -1,75
25 - 75 -3,30
25 - 80 -2,86
25 - 85 -2,40
25 - 90 -1,97
50 - 75 -2,34
50 - 80 -1,96
50 - 85 -1,63
50 - 90 -1,36
Table 3.4: Numerical results using the utter margin
24
Section 3.3 Flutter Margin
3.3.1 Eect of noise
When we look at table 3.5, we see that the accuracy decreases in a signicant way. The damping
extrapolation method uses four values that it receives from the FRF through the RFP method.
When combining the four errors, we get a larger error on the end result. The predicted utter
speed is still a conservative value for every range. The procentual errors reduces when the end
of the range is more towards the utter speed.
Range (% of
utter speed)
Error (in %)
10 - 75 -6,35
10 - 80 -5,58
10 - 85 -5,29
10 - 90 -4,56
25 - 75 -7,31
25 - 80 -6,42
25 - 85 -5,64
25 - 90 -4,86
50 - 75 -5,70
50 - 80 -5,45
50 - 85 -4,62
50 - 90 -3,92
Table 3.5: Numerical results using the utter margin, including noise
25
3.4 Envelope Function
The envelope function is the only method that uses time domain measurements so an additional
step has to be performed in comparison to the other methods. The approach used in this work
is
For every speed of interest we generate the FRFs (4 in this case) using the simulated
model.
A FRF is transformed from the frequency domain to the time domain using the fast fourier
transform.
Using this time domain signal, the envelope is calculated.
An S-parameter is calculated using the equations 2.12, 2.13 and 2.13 as seen in gure 3.11.
This S-parameter is plotted against the speed.
The data points are used to curve t and predict the utter speed, with polynomial degrees
ranging from 2 to 8. An example of this can be seen in gure 3.10.
0 5 10 15 20 25 30
0
0.2
0.4
0.6
0.8
1
Speed in m/s
S
v
a
l
u
e
Envelope function d
=0.128 Ustart=15.085 Ustop=27.153

2nd order
3rd order
4th order
5th order
SCentroid points
Figure 3.10: Example of the Envelope Function - polynomial curve tting
When looking at table 3.6 we see that when we change the value of t
max
there is a change in
accuracy but there is no clear pattern involved. By that we mean that the accuracy depends on
the range, not so much on the t
max
. Just as with previous methods, the accuracy improves when
the range evolves towards the utter speed. This method has an even distribution between too
low utter speed predictions and too high utter speed prediction.
3.4.1 Eect of noise
The results of the envelope function can be seen in table 3.7. From this table we see that
the accuracy decreases severely when compared to the no noise results. This could be due to
the extra operation that the method has to endure, namely the inverse fourier transform. The
accuracy does not increase when the end of the range gets closer to the utter speed. The best
value of t
max
seems to be 15 seconds.
26
Section 3.4 Envelope Function
Range (% of
utter speed)
t=30s t=20s t=15s
Best
degree
Error
(%)
Best
degree
Error
(%)
Best
degree
Error
(%)
10 - 75 3 1,69 3 2,47 3 3,84
10 - 80 3 9,80 3 8,61 3 8,94
10 - 85 6 -4,25 6 -5,52 6 -5,91
10 - 90 6 2,03 6 -0,48 6 -1,24
25 - 75 5 -14,40 5 -14,83 5 -14,83
25 - 80 5 -8,56 5 -9,37 5 -9,58
25 - 85 5 -1,86 5 -3,59 5 -4,19
25 - 90 7 -2,97 5 0,38 5 -0,54
50 - 75 6 -8,24 4 13,62 4 8,16
50 - 80 6 -6,30 6 -2,63 6 5,45
50 - 85 6 3,45 6 4,43 6 5,88
50 - 90 7 1,63 7 -0,29 6 0,13
Table 3.6: Numerical results using the envelope function
0 5 10 15 20 25 30 35 40
8
6
4
2
0
2
4
6
8
x 10
4
Svalue=0.93136 at a speed of 27.153m/s
tijd(in seconden)
A
m
p
l
i
t
u
d
e

time signal
Envelope
SValue
Figure 3.11: Example of the Envelope Function - time domain
27
Range (% of
utter speed)
t=30s t=20s t=15s
Best
degree
Error
(%)
Best
degree
Error
(%)
Best
degree
Error
(%)
10 - 75 3 6,51 3 5,06 3 4,34
10 - 80 3 12,10 3 9,07 3 8,43
10 - 85 8 -15,14 3 12,11 3 10,55
10 - 90 8 -10,12 8 -10,10 8 -10,08
25 - 75 5 5,00 5 -4,45 5 -6,37
25 - 80 5 -1,47 4 1,78 4 1,22
25 - 85 5 3,60 5 -2,13 4 2,67
25 - 90 6 1,00 5 1,30 5 -0,01
50 - 75 5 2,54 3 14,46 5 3,17
50 - 80 3 31,86 3 13,46 4 7,96
50 - 85 3 71,72 3 15,15 3 11,34
50 - 90 3 37,78 4 11,64 4 7,92
Table 3.7: Numerical results using the envelope function, including noise
3.5 Frequency Domain Prediction Algorithm
To implement the frequency domain prediction algorithm in matlab, starting from the numerical
model, following steps are taken
1. Starting from the FRF, generated from the analytical model we apply the invfreqs Matlab
function. In order to create the denominator coecient matrix A
2. The speed matrix U is constructed
3. Step 1 and step 2 are repeated for (2N + 1)(2N + 2)/2 times. With N as the number of
DOF.
4. The speed matrix U is inverted and multiplied by the denominator coecient matrix A
to get the coecients matrix B.
5. Using the coecients matrix B, we simulate every speed and see where the damping
becomes zero. An example of this step can be found in gure 3.12.
0 5 10 15 20 25 30
0.1
0.05
0
0.05
0.1
0.15
0.2
0.25
0.3
range 10 90% of flutter speed
D
a
m
p
i
n
g
Speed (in m/s)
Figure 3.12: Example of Freq. domain prediction algorithm
28
Section 3.5 Frequency Domain Prediction Algorithm
Range (% of
utter speed)
Error (in %)
10 - 75 -0,07
10 - 80 -0,04
10 - 85 -0,01
10 - 90 0,04
25 - 75 0,06
25 - 80 0,01
25 - 85 0,01
25 - 90 0,05
50 - 75 -0,05
50 - 80 -0,11
50 - 85 0,00
50 - 90 0,11
Table 3.8: Numerical results using the frequency domain prediction algorithm
The results are shown in table 3.8 where we can see that this seems to be the method with
the smallest procentual errors. In contrast to other methods, there is no improvement when the
end of the range is shifted more towards the utter speed.
3.5.1 Eect of noise
The results of de Frequency Domain Prediction Algorithm with added noise is presented in table
3.9. In this table it is clear that the accuracy decreases when noise is added to the FRF. Same
as for this method without noise, the results seem to be the most accurate of all the methods.
With a maximum error of 2% this can be considered as a very successful method.
Range (% of
utter speed)
Error (in %)
10 - 75 -2,03
10 - 80 -0,17
10 - 85 1,11
10 - 90 -0,10
25 - 75 -0,98
25 - 80 -0,44
25 - 85 -0,46
25 - 90 0,48
50 - 75 0,48
50 - 80 0,48
50 - 85 1,43
50 - 90 0,42
Table 3.9: Numerical results using the frequency domain prediction algorithm, including noise
29
3.6 Summary of results
In this chapter we evaluated the utter prediction methods using the simulated model. First of
all, a pk-analysis was performed to see whether utter would occur inside the range of speeds
that the VUB windtunnel can achieve. Using a trial and error method, the spring constants
were chosen. Furthermore, a graph is made to asses the utter speed with dierent values of the
spring constants for the pitch motion (this by changing the distance between the springs). Once
all the parameters were known, the 2 degree of freedom model was implemented in MATLAB and
we evaluated the damping (with a speed interval of 0.1 m/s), to get an impression of the type
of utter.
Once the model was functional, the dierent methods were programmed in MATLAB and evalu-
ated for dierent ranges of speeds. When comparing these predictions we see that the frequency
domain prediction algorithm has the best results, followed by the damping extrapolation. The
envelope function and utter margin have comparable errors and can be ranked 3rd and 4th.
The frequency domain utter prediction algorithm has an advantage because it tries to model
the system using the quasi-steady formulations and the system is based upon a modied quasi-
steady aerodynamics.
The damping extrapolation, utter margin and frequency domain prediction algorithm ben-
et from the fact that the model is built in the frequency domain and does not have to undergo
an inverse fourier transform (unlike the envelope function) witch introduces errors.
When noise is added, all of the methods become more inaccurate. The eect of noise has
the most impact on the envelope function. The reason can be found in the fact that the other
methods get their input via the frequency domain but the envelope function needs time domain
input. To get from frequency domain to time domain, we use the inverse fourier transform witch
introduces another error. The utter margin sees its error double when adding noise. This could
be due to using four values to predict the utter speed and those four values are each corrupted
with noise.
30
CHAPTER 4
Experimental Aspect
Chapter Abstract In the next chapter the design of the experimental set-up is described.
Using experimental modal analysis techniques, the necessary parameters are extracted from the
airfoil and used for the dierent utter prediction methods. The results of each method are
discussed and a comparison is made between the dierent methods.
31
Chapter 4. Experimental Aspect
4.1 Design of the Test Setup
This thesis is a continuation of previous work done by Ertveldt (2011) where an experimental
setup was built using springs and a guidance system. Despite the excellent design of that setup,
there were problems with the amount of damping present (possibly due to the guidance system)
and the fact that the springs were easily overstretched into a non-linear area. In the thesis of
Ertveldt (2011) an alternative to the setup is suggested based on the work of Razak et al. (2011).
The focus of this paper is shifted towards stall-utter and LCO and it is shown that this setup
can be used to induce utter in a repetitive manner in the Ulg wind tunnel.
(a) Photo of setup used in
Ertveldt (2011)
(b) Photo of setup based on Razak et al. (2011)
Figure 4.1: Old and new setup
4.2 First Windtunnel Test
The rst series of windtunnel tests was performed on 2 and 3 april 2012. The rst test was
with a d
at maximum. the second test was with the same d
but at a dierent angle of attack.

The third test was at an d
of 0.128m and the fourth test was with the same d
as test number
three,but at a dierent angle of attack. During the fth test (same conditions as test number
3), the connection at the side of the airfoil broke o. When analyzing the obtained results, it
came apparent that there were not enough measurements to perform a usefull analysis. Another
problem was that the utter speed could not be determined exactly, because utter arose at
dierent speeds.
4.3 Second Windtunnel Test
The second series of windtunnel tests was performed on 14 may 2012. d
was set at 0.09 m and

held constant for all the measurements. During this test, the utter speed was evaluated at 21
m/s, with an accuracy of 0.5 m/s. In total there are 58 measurements at speeds between 3.9 and
20.5m/s executed. Excitation of the wing is performed through applying an impulse on the wing.
During the windtunnel tests, a couple of faults such as (no excitation applied to the wing,
measurement time too short,...) were made. Unfortunately, when these faults were discov-
ered, there was no more time to perform additional windtunnel experiments. As a result, the
experimental data suers from a large variation.
32
Section 4.3 Second Windtunnel Test
4.3.1 Damping Extrapolation
In order to use the damping extrapolation method, we need to examine the damping of a mode
where the damping goes to zero. For the numerical results, only the pitch and plunge mode were
accounted for but since the experimental has more degrees of freedom there are other modes
that come into play. When the results are imported into the MATLAB script, we obtain the results
that are portrayed in table 4.1. In this table we see that the predicted speeds are fairly close
to the actual speed for 6 of the 9 ranges. The accuracy improves as the top range is closer to
the actual utter speed. Unlike the experimental noise free results, the best polynomial degree
is not the highest. There is no clear pattern visible that suggest which degree is best suited for
the dierent ranges. The highly scattered damping values causes the bad results for the 3 bad
predictions.
Range (% of
utter speed)
20,95 - 83,33 5 -2,57
20,95 - 87,14 3 75,59
20,95 - 91,43 8 -2,23
24,76 - 83,33 4 -4,65
24,76 - 87,14 3 67,67
24,76 - 91,43 7 2,02
29,05 - 83,33 6 -1,57
29,05 - 87,14 3 59,81
29,05 - 91,43 7 0,25
Table 4.1: Experimental results using the damping extrapolation
33
4.3.2 Flutter Margin Method
To implement the utter margin method, we need the damping and frequencies of the pitch
and plunge mode at dierent speeds. As we can see in table 4.2, the results are not good.
Figure 4.2(a) shows an example of an unsuccessful attempt to predict the utter margin, gure
4.2(b) on the other hand shows a successful prediction. The damping for both modes are highly
scattered. This problem can be prevented by performing the same measurements again at the
same speed and averaging the results afterwards.
Range (% of
utter speed)
Error (in %)
20,95 - 83,33 -81,8003
20,95 - 87,14 -100
20,95 - 91,43 -100
24,76 - 83,33 -78,1283
24,76 - 87,14 -100
24,76 - 91,43 -100
29,05 - 83,33 -74,2517
29,05 - 87,14 -100
29,05 - 91,43 -100
Table 4.2: Experimental results using the utter margin
0 2 4 6 8 10 12 14 16 18 20 22
0
20
40
60
80
100
120
Flutter margin for experimental results
Speed in m/s
F
lu
t
t
e
r

m
a
r
g
in

v
a
lu
e

Flutter margin values
Flutter Speed
2nd degree polynomial fit
(a) Unable to predict utter speed
0 2 4 6 8 10 12 14 16 18 20 22
0
20
40
60
80
100
120
Flutter margin for experimental results
Speed in m/s
F
lu
t
t
e
r

m
a
r
g
in

v
a
lu
e

Flutter margin values
Flutter Speed
2nd degree polynomial fit
(b) Succesful prediciton of utter speed
Figure 4.2: Flutter margin method using experimental results
34
Section 4.3 Second Windtunnel Test
4.3.3 Envelope Function
In contradiction to the other methods, the envelope function uses time domain measurements
as opposed to frequency domain measurements. In total 19 measurements were executed in
the time domain. The enevelope function requires some sort of excitation so at each speed an
impulse has been given to the wing. Each measurement delivered 4 datasets because we used
4 sensors. In table 4.3 and 4.4, the results are shown for sensor 1 (at the leading edge) and
sensor 2 (at the trailing edge) respectively. Figure 4.3 displays the dierent time signals for
both sensors. It is noticable that sensor 2 seems to move more than sensor 1. This has no great
eect on the results (the dierence between the average results of sensor 1 and sensor 2 is 0.28
%) . The parameter that is of inuence is t
max
, changing this parameter will have inuence on
the results. But there is a dierence between the most accurate t
max
; sensor 1 has an optimal
t
max
of 5 seconds while the optimal t
m
ax for sensor 2 is 15 seconds.
Range (% of
utter speed)
t=5s t=10s t=15s t=20s
Best
degree
Error
(%)
Best
degree
Error
(%)
Best
degree
Error
(%)
Best
degree
Error
(%)
20,95 - 83,33 5 -2,18 6 -3,49 4 -2,25 4 -6,20
20,95 - 91,43 8 -3,81 8 -4,58 6 2,61 6 -1,22
20,95 - 95,71 8 1,17 8 -0,15 8 0,66 8 0,29
29,05 - 83,33 8 -11,43 5 -1,05 4 -2,39 4 -7,63
29,05 - 91,43 6 0,06 6 4,35 6 -1,15 4 1,75
29,05 - 95,71 8 16,90 8 0,68 8 0,62 8 1,41
54,29 - 83,33 3 -2,21 2 1,54 2 -1,80 2 -2,38
54,29 - 91,43 6 4,83 6 -2,89 6 -3,04 4 1,45
54,29 - 95,71 5 1,28 5 1,66 7 -1,54 4 1,46
Table 4.3: Experimental results using the envelope function and sensor 1
Range (% of
utter speed)
t=5s t=10s t=15s t=20s
Best
degree
Error
(%)
Best
degree
Error
(%)
Best
degree
Error
(%)
Best
degree
Error
(%)
20,95 - 83,33 4 -0,25 4 -4,92 4 -5,66 4 -6,29
20,95 - 91,43 8 0,82 8 -3,64 8 -1,88 6 -0,17
20,95 - 95,71 8 -0,68 8 -0,35 8 1,58 8 0,88
29,05 - 83,33 5 -5,32 4 3,77 4 -4,87 4 -7,38
29,05 - 91,43 5 -0,36 5 0,78 6 1,06 4 1,75
29,05 - 95,71 8 2,27 8 1,83 8 0,67 8 0,89
54,29 - 83,33 3 -2,88 2 -2,56 2 -3,53 2 0,75
54,29 - 91,43 8 -5,94 8 3,78 7 -4,42 7 -4,76
54,29 - 95,71 6 -1,20 6 -2,01 6 -2,03 6 -1,95
Table 4.4: Experimental results using the envelope function and sensor 2
35
0 5 10 15 20 25 30
4
3
2
1
0
1
2
3
4
Time (in seconds)
A
m
p
l
i
t
u
d
e

Sensor 4
Sensor 1
Figure 4.3: Time signal of 2 dierent sensors
4.3.4 Frequency Domain Prediction Algorithm
As the name suggests, the frequency domain prediction algorithm uses the measurements in the
frequency domain. Using this method only one range was really successful (see gure 4.4). When
we look at table 4.5, the errors are quite bad. This can be accounted to bad measurements and
a large variation of the experimental data. That being said, the minimum error is 6 % which is
not bad.
Range (% of
utter speed)
Error (in %)
29,05 - 50,00 -28,5
50,00 - 66,67 6,3
66,67 - 83,33 -13,2
Table 4.5: Experimental results using frequency domain prediction algorithm
0 5 10 15 20 25 30
0.4
0.3
0.2
0.1
0
0.1
0.2
0.3
Frequency domain flutter prediction algorithm
Speed (m/s)
D
a
m
p
i
n
g
Figure 4.4: Example of frequency domain utter prediction algorithm
36
Section 4.4 Conclusion
4.4 Conclusion
In this chapter we built an experimental setup and were able to induce utter in a repetitive
manner. Using this wing it was possible to extract all the necessary data needed for the dierent
utter prediction methods.
When examining the results, we nd that the damping extrapolation and envelope function
both have good results. This cannot be said for the frequency domain prediction algorithm and
especially not for the utter margin method.
The damping extrapolation and envelope function do have an undeniable advantage because
the utter speed is already known. It is therefore possible to choose a polynomial that comes
close to this utter speed. But it is impossible to know what degree of polynomial should be
used. The utter margin method also suers from the fact that there is such a large variance in
the experimental data that it is hard to predict a correct speed with a limited set of measure-
ments.
Although the Frequency domain prediction algorithm does not provide great results, they are
still better than the utter margin method. We have to compare the utter margin and fre-
quency domain prediction algorithm because these are the only 2 methods where we do not
have to choose a polynomial degree. Another advantage that the frequency domain prediction
algorithm has, is that the end speed is only 66 % of the utter speed. When we consider ight
tests, this is a denite improvement towards safety.
37
CHAPTER 5
Experimental vs Numerical results
Chapter Abstract The next chapter examines the dierences between the numerical and
experimental results.
39
Chapter 5. Experimental vs Numerical results
5.1 Flutter Speed
When comparing the utter speed that is determined experimentally (21 m/s) to the speed
that is calculated using the pk-analysis (14.04 m/s, as seen in gure 5.1), there is quite a large
dierence. One of the reasons why the theoretical pk-analysis has a large error is because it uses
a theoretical model that makes certain assumptions. These assumptions are listed in section
1.4.1. The mathematical model used for the pk-analysis does not lter in the eect of the wake
or the fact that it is an airfoil and not a at plate. Other errors can be found in the experimental
setup where there are more than 2 degrees of freedom. The eect that the frame has on the
airow around the wing is also not accounted for.
2 4 6 8 10 12 14 16 18 20
0.01
0.005
0
0.005
0.01
0.015
0.02
D
a
m
p
i
n
g

r
a
t
i
o
Airspeed (m/s)

Plunge
Pitch
Figure 5.1: PK-analysis with d
= 0.09m/s
5.2 Simulated model vs Experimental setup
Although the utter speed of the simulated model does not match with the experimental setup,
there are some similarities. When we calculate the frequency using a PK-Analysis or by re-
trieving the frequency from the generated FRF, we nd the same frequencies that are present
in the experimental setup for the plunge mode (around 5.1 Hz). The plunge mode does dier:
simulated mode 6.3 Hz and experimental setup 7.1 Hz.
5.3 Prediction methods
The only valid comparison that can be made is the one between the simulated model with noise
and the experimental setup. We can say that all but one method perform worse using the
experimental data and that is the envelope function. The reason why this method performs so
good is that it does not rely on frequency domain measurements. As mentioned in section 4.3,
the frequency domain measurements were not entirely successful. Another reason is that is uses
the best polynomial approach.
40
CHAPTER 6
Conclusions
Chapter Abstract In the next chapter a summary is given of the work done and conclusions
are drawn. Furthermore a section is added on the possibilities of further research.
41
Chapter 6. Conclusions
6.1 Summary
In this thesis 4 dierent utter prediction methods are evaluated in an experimental and ana-
lytical manner.
To compare the prediction methods in an analytic way, a mathematical model is programmed
in MATLAB and used to obtain the necessary input needed for the dierent prediction methods.
The theoretical utter speed of this simulated model is calculated using the pk-analysis. The
results of this and the comparison between methods can be found in chapter 3.
A new setup is built based on the conclusions of (Ertveldt, 2011). This setup is able to induce
utter in a repetitive manner. The utter speed can be changed without replacing components.
Using modal analysis, the input for the dierent prediction methods are obtained. A comparison
of these wind tunnel experiments can be found in chapter 4.
Furthermore the comparison between the experimental results and analytical results is made.
This is presented in chapter 5.
6.2 To Conclude
The frequency domain utter prediction algorithm proves to be the most accurate in respect to
the simulated MATLAB model. This applies to the noise free and noise polluted data. This method
and the utter margin have the advantage that they do not have an extra unknown parameter
(order of the polynomial t) that has to be chosen. A remark has to be made that the simulated
model is constructed via a modied quasi-steady aerodynamic theory. Because the frequency do-
main prediction algorithm uses a quasi-steady aerodynamic approach, it does have an advantage.
The experimental data proved to have a large variation, which makes comparison between meth-
ods dicult. A solution to this problem is to perform more measurements to obtain a statistically
better estimate of the modal parameters. Unfortunately due to external circumstances (missing
parts, breaking of the wing and late deliveries), there was no time left to perform additional
measurements. But thanks to this thesis, when extra measurements are performed the MATLAB
scripts, provided in the appendix, take care of all the post processing data to compare the dif-
ferent methods.
The Experimental utter speed is about 43 % higher than the utter speed predicted with
the simulated model, using comparable parameters (same pitch and plunge stiness). This high
dierence proves that the theodorsen aerodynamic model does have limitations. The fact that
the experimental model has more than 2 degrees of freedom is a source of error. Other possible
errors arise from the assumptions made in the theodorsen aerodynamics, such as; not accounting
the wake or assuming a at plate instead of an airfoil.
6.3 Further Research
If there is a continuation of this work, it should start with performing the windtunnel tests
again. More measurement data will make sure that there is a greater condence bound when
comparing the dierent prediction methods.
One of the challenges of the experimental aspect is to dene at what speed we can dene
utter. This is not always ease because sometimes there are severe vibrations but there is still
damping present. A possible improvement to the experimental set up can be to develop a system
that detects when the springs become non-linear.
42
Section 6.3 Further Research
Further research could involve the inuence of dierent NACA proles, the eect of added
mass or adjusting the exural point of the airfoil. All these inuences can be investigated in a
experimental and numerical manner.
Using the experimental setup and analytical code, it is possible to examine the dierent physical
parameters such as weight, type and thickness of airfoil, position of exural axis,... A compari-
son can be made to see if the analytical model and the experimental results match up, e.g. will
the utter speed increase when we add mass and will the dierence is utter speed be the same
for the experimental model and analytical model?
Since there is an experimental setup that successfully can induce a state of utter, it might
be interesting to further investigate how we can prevent utter. One can use the aileron on the
wing or an eccentric mass inside the wing that counteracts the motion.
43
Bibliography
Bennett, R. and Abel, I. (1981). Application of a ight test and data analysis technique to
utter of a drone aircraft. In Structures, Structural Dynamics and Materials Conference, 22
nd, Atlanta, Ga, pages 811820.
Bisplingho, R. L., Ashley, H., and Halfman, R. L. (1996). Aeroelasticity. Dover Publications.
Cooper, J. (1995). Parameter estimation methods for ight utter testing. AGARD CP-566,
Paper No 10.
Cooper, J. (2003). Towards faster and safer ight utter testing. Technical report, DTIC
Document.
Cooper, J., Emmet, P., Wright, J., and Schoeld, M. (1993). Envelope Function - A Tool for
Analyzing Flutter Data. Journal of Aircraft, 30(5):785790.
De Troyer, T., Zouari, R., Guillaume, P., and Mevel, L. (2008). A new frequency-domain utter
speed prediction algorithm using a simplied linear aeroelastic model. Proceedings of ISMA
2008: International Conference on Noise and Vibration, pages 11971206.
Debille, J. (2011). Industrial solutions for in-ight & oine experimental utter analysis. Pre-
sentation given at Aerospace testing 2011 conference.
Dickinson, M. (1995). Cf-18 ight utter testing (t) techniques. AGARD CP-566.
Dimitriadis (2010). Lecture notes on aeroelasticity and uid-structure interaction course.
http://www.ltas-aea.ulg.ac.be/cms/index.php?page=aeroelasticity-course.
Dimitriadis, G. and Cooper, J. (2001). Flutter prediction from ight utter test data. Journal
of Aircraft, 38(2):355367.
Dorf, R. and Bishop, R. (2008). Modern control systems. Pearson Prentice Hall.
Dowell, E., Crawley, E. F., Curtiss Jr, H. C., Peters, D. A., Scanlan, R. H., and Sisto, F. (1995).
A Modern Course in Aeroelasticity. Springer, 3rd rev. and enlarged ed. edition.
Ertveldt, J. (2011). Modal analysis for utter speed prediction: Wind tunnel application. Mas-
ters thesis, Erasmus University College Brussels Department IWT, Nijverheidskaai 170, An-
derlecht.
Fung, Y. C. (1994). An Introduction to the Theory of Aeroelasticity. Dover Publications.
Goge, D., Boswald, M., Fullekrug, U., and Lubrina, P. (2007). Ground vibration testing of large
aircraftstate-of-the-art and future perspectives. In Proceedings of the 25th International
Modal Analysis Conference (IMAC), Paper, pages 113.
45
Bibliography
He, J. and Fu, Z. (2001). Modal Analysis. Butterworth-Heinemann.
Hodges, D. H. and Pierce, G. A. (2002). Introduction to Structural Dynamics and Aeroelasticity.
Cambridge University Press.
Joint Aviation Authorities (JAR) (2007). JAR-25 Large Aeroplanes.
Jones, R. (1940). The unsteady lift of a wing of nite aspect ratio. NACA report, 681.
Kehoe, M. (1987). Aircraft ground vibration testing at nasa ames-dryden ight research facility.
In International Modal Analysis Conference, 5 th, London, England, Proceedings., volume 1,
pages 728736.
Kehoe, M. W. et al. (1995). A historical overview of ight utter testing. Nasa Technical
Memorandum, TM-4720.
Lind, R. (2003). Flight-test evaluation of utter prediction methods. Journal of Aircraft,
40(5):964970.
Newman, N., Friswell, M., and Penny, J. (1993). The parallel implementation of the rational
fraction polynomial method. Proceedings of the International Modal Analysis Conference,
pages 318318.
Nissim, E. and Gilyard, G. B. (1989). Nasa technical paper 2923: Method for experimental
determination of utter speed by parameter identication. Periodical; periodical/journal,
magazine, other, NASA. Description based on: 1823.
Peeters, B., Hendricx, W., Debille, J., and Climent, H. (2009). Modern solutions for ground
vibration testing of large aircraft. Sound and Vibration, 43(1):815.
Price, S. and Lee, B. (1992). Development and analysis of ight utter prediciton methods.
AIAA Paper, CP-2101:188200.
Price, S. and Lee, B. (1993). Evaluation And Extension Of The Flutter-margin Method For
Flight Flutter Prediction. Journal of Aircraft, 30(3):395402.
Razak, N. A., Andrianne, T., and Dimitriadis, G. (2011). Flutter and stall utter of a rectangular
wing in a wind tunnel. AIAA Journal, 49:22582271.
Routh, E. (1877). A treatise on the stability of a given state of motion, particularly steady
motion. London, Macmillan and Co.
Theodorsen, T. and Center, N. L. R. (1935). General theory of aerodynamic instability and the
mechanism of utter. National Advisory Committee for Aeronautics.
Van De Walle, M., De Troyer, T., and Schoukens, J. (2011). Experimental validation of a new
frequency-domain utter speed prediction algorithm using a simplied linear aerolastic model.
In IFASD.
Verboven, P., Cauberghe, B., Guillaume, P., Vanlanduit, S., and Parloo, E. (2004). Modal
parameter estimation and monitoring for on-line ight utter analysis. Mechanical systems
and signal processing, 18(3):587610.
Von Schlippe, B. (1936). The question of spontaneous wing oscillations (determination of critical
velocity trough ight vibration tests). NACA technical memorandum, TM-806. Translation.
Wright, J. R. and Cooper, J. E. (2008). Introduction to Aircraft Aeroelasticity and Loads. Wiley,
1 edition.
Zimmerman, N. W. J. (1964). Prediction of utter onset speed based on ight testing at
subcritical speeds. Journal of Aircraft, Vol. 1(No. 4):190202.
46
Bibliography
Zouari, R., De Troyer, T., Mevel, L., Basseville, M., and Guillaume, P. (2008). Flutter mon-
itoring using a mixed model-based and data-based approach. Proceedings Of ISMA 2008:
International Conference On Noise And Vibration, 2008:12651274.
47
APPENDIX A
Derivation of Routh-Hurwitz Stability Criterion
Joint Aviation Authorities (JAR) (2007) The equations to form a Routh-Hurwitz array are
found in Routh (1877). They are reprinted below. Starting from the following characteristic
equation
P(s) = a
n
s
n
+a
n1
s
n1
+ +a
0
(A.1)
The coecients in the table are calculated using following formulas
s
n
a
n
a
n2
a
n4

s
n1
a
n1
a
n3
a
n3

s
n2
b
1
b
2
b
3

s
n3
c
1
c
2
c
3

s
n4
d
1
d
2
d
3

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
Table A.1: The general Routh-Hurwitz array structure
b
i
=
a
n1
a
n2i
a
n
a
n2i1
a
n1
(A.2)
c
i
=
b
1
a
n2i1
a
n1
b
i+1
b
1
(A.3)
d
i
=
b
i
c
i+1
b
i+1
c
i
c
i
(A.4)
The Routh-Hurwitz theorem states that all the elements in the rst column of the array must
be positive. When we apply this to a 4th order polynomial
P(s) = a
4
s
4
+a
3
s
3
+a
2
s
2
+a
1
s +a
0
(A.5)
we get following coecients
b
1
=
a
3
a
2
a
4
a
1
a
3
(A.6)
b
2
=
a
3
a
0
a
4
0
a
3
= a
0
(A.7)
c
1
=
a
3
a
2
a
4
a
1
a
3
a
1
a
3
a
0
a
3
a
2
a
4
a
1
a
3
=
(a
3
a
2
a
4
a
1
)a
1
a
2
3
a
0
a
3
a
2
a
4
a
1
(A.8)
d
1
=
b
1
0 b
2
c
1
c
1
= b
2
= a
0
(A.9)
49
Chapter A. Derivation of Routh-Hurwitz Stability Criterion
s
4
a
4
a
2
a
0
s
3
a
3
a
1
s
2 a
3
a
2
a
4
a
1
a
3
a
0
s
1
(a
3
a
2
a
4
a
1
)a
1
a
2
3
a
0
a
3
a
2
a
4
a
1
s
0
a
0
Which results in following array Considering that for a stable system all the all the elements in
the rst column must be positive, we get the following stability criteria
a
4
> 0 (A.10)
a
3
> 0 (A.11)
a
0
> 0 (A.12)
(a
3
a
2
a
4
a
1
)a
1
a
2
3
a
0
> 0 (A.13)
a
3
a
2
a
4
a
1
> 0 (A.14)
50
APPENDIX B
Contents of the CD
MATLAB programs
numerical.m is the matlab program used to generate all of the numerical results.
experimental.m is the matlab program used to process all of the experimental data.
kphicalc.m is the matlab program used to calculate the pitch spring constant.
51
Chapter B. Contents of the CD
52
APPENDIX C
JAR 25.629 Flutter, Deformation and failsafe criteria
This text is an excerpt of the conditions to grant an airworthiness certicate. Only the section
that is important to utter is reprinted here
a General. Compliance with this paragraph must be shown by calculations, resonance tests,
or other tests found necessary by the Authority. Full scale ight utter tests at speeds up
to V
DF
/M
DF
for the critical aeroplane utter modes must be conducted when
1 M
D
is equal to or greater than 0.8 M
2 The adequacy of utter analysis and wind tunnel tests have not been established by
previous experience with aircraft having similar design features; or
3 The conditions specied in subparagraph (a)(1) or (2) of this paragraph exist, and
modications to the type design have a signicant eect on the critical utter modes.
b Flutter and divergence prevention. The dynamic evaluation of the aeroplane must
include an investigation of the signicant elastic, inertia, and aerodynamic forces associ-
ated with the rotations and displacements of the plane of the propeller. In addition, the
following apply:
1 The aeroplane must be designed to be free from utter and divergence (unstable
structural distortion due to aerodynamic loading) for all combinations of altitude and
speed encompassed by the V
D
/M
D
versus altitude envelope enlarged at all points by
an increase of 20% in equivalent air-speed at both constant Mach number and constant
altitude, except that the envelope may be limited to a maximum Mach number of 1.0
when M D is less than 1.0 at all design altitudes a n d - s following is established:
i A proper margin of damping exists at all-speeds up to M
D
; and
ii There is no large and rapid reduction in damping as M
D
is approached.
2 If concentrated balance weights are used on control surfaces, their eectivenes and
strength, including supporting structure, must be substantiated.
c Loss of control due to structural deformation. The aeroplane must be designed
to be free from control reversal and from undue loss of longitudinal, lateral, and direc-
tional stability and control, as a result of structural deformation (including that of the
control surface covering) at speeds up to the speed prescribed in subparagraph (b) of this
paragraph for utter prevention
d Fail safe criteria. The following fail-safe criteria must be met:
53
Chapter C. JAR 25.629 Flutter, Deformation and failsafe criteria
1 It must be shown, by analysis or tests, that the aeroplane is free from such utter or
divergence that would preclude safe ight, at any speed up to V
D
, after each of the
following:
i Each of the failures, malfunctions, or adverse conditions listed in sub-paragraph
(d)(4) of this paragraph.
ii Any other combination of failures, malfunctions, or adverse conditions not shown
to be extremely improbable.
2 If a failure, malfunction, or adverse condition described in sub-paragraph (d)(4) of this
paragraph is simulated during a ight test in showing compliance with this paragraph,
the maximum speed investigated need not exceed V
FC
if it is shown, by correlation
of the ight test data with other test data or analyses, that hazardous utter or
divergence will not occur at any speed up to V
D
3 The structural failures described in sub-paragraphs (d)(4)(i) and (ii) of this paragraph
need not be considered in showing compliance with this paragraph if engineering data
substantiate that the probability of their occurrence is negligible by showing that the
structural element is designed with:
i Conservative static strength margins for each ground and ight loading conditions
specied in this JAR-25; or
ii Sucient fatigue strength for the loading spectrum expected in operation.
4 The failures, malfunctions, or adverse conditions used to show compliance with this
paragraph are as follows:
i Failure of any single element of the structure supporting any engine, indepen-
dently mounted propeller shaft, large auxiliary power unit, or large externally
mounted aerodynamic body (such as an external fuel tank).
ii Any single failure of the engine structure, on turbo-propeller aeroplanes, that
would reduce the yaw or pitch rigidity of the propeller rotational axis.
iii Absence of propeller aerodynamic forces resulting om the feathering of any single
propeller, and, for aeroplanes with four or more engines, the feathering of the
critical combination of two propellers. In addition, any single feathered propeller
must be paired with the failures, specied in (d)(4)(i) of this sub-paragraph,
involving failure of any single element of the structure supporting any engine or
independently mounted propeller shaft, and the failures specied in (d)(4)(ii) of
this sub-paragraph.
iv Any single propeller rotating at the highest likely overspeed. Failure of each
principal structural element selected for compliance with JAR 25.571 (b). Safety
following a failure may be substantiated by showing that losses in rigidity or
changes in frequency, mode shape, or damping are within the parameter varia-
tions shown to be satisfactory in the utter and divergence investigations.
v Any single failure or malfunction, or combinations thereof, in the - ight control
system considered under JAR 25.671, 25.672 and 25.1309, and any single failure
in any utter damper system. Investigation of forced structural vibration other
than utter, resulting from failures, malfunctions, or adverse conditions in the
automatic ight control system may be limited to airspeed up to V
C
.
54
APPENDIX D
Rational Fraction Polynomial
The Rational Fraction Polynomial method (RFP) is a way to derive modal data, such as damping
and frequency, from an FRF for a MDOF system. A brief summary of this method is presented,
based on the article by (Newman et al., 1993). For a more detailed explanation, He and Fu
(2001) can be consulted.
ij
(s) =
N(s)
D(s)
=
a
0
+a
1
s +a
2
s
2
+ +a
m
s
m
b
0
+b
1
s +b
2
s
2
+ +s
n
(D.1)
For the sake of simplicity following notations are adopted
p
0
(s) = 1, p
1
(s) = s, p
2
(s) = s
2
, (D.2)
q
0
(s) = 1, q
1
(s) = s, q
2
(s) = s
2
, (D.3)
When implementing these notations and considering a frequency response function, we get
(j) =
m
k=0
a
k
p
k
(j)
n
k=0
b
k
q
k
(j)
(D.4)
with a
k
, b
k
as the coecients to be determined and p
k
(j), q
k
(j) are polynomials for positive
frequencies. The FRF at the negative frequencies can be shown as
(j
i
) = (j
i
) =
(j
i
) i = 1, 2, , p (D.5)
We can express the error at all measured frequencies as
e
ij
() = ()
ij
() (D.6)
For the ease of analysis and more accurate results, a new error function can be introduced
e(j) = e(j)
n
k=0
b
k
q
k
(j) =
m
k=0
a
k
p
k
(j) (j)
_
n1
k=0
b
k
q
k
(j) +q
n
(j
_
(D.7)
The total error for the whole column of FRF becomes
|E| =
e
ij
(w
p
)
.
.
.
e
ij
(w
1
)
e
ij
(w
1
)
.
.
.
e
ij
(w
p
)
(D.8)
55
Chapter D. Rational Fraction Polynomial
|E| = |U| |A| |V | |B| |W| (D.9)
with
|U| =
p
0
(j
p
) p
1
(j
p
) p
m
(j
p
)

p
0
(j
1
) p
1
(j
1
) p
m
(j
1
)
p
0
(j
1
) p
1
(j
1
) p
m
(j
1
)

p
0
(j
p
) p
1
(j
p
) p
m
(j
p
)
(D.10)
|V | =
a(j
p
)q
0
(j
p
) a(j
p
)q
1
(j
p
) a(j
p
)q
n1
(j
p
)

a(j
1
)q
0
(j
1
) a(j
1
)q
1
(j
1
) a(j
1
)q
n1
(j
1
)
a(j
1
)q
0
(j
1
) a(j
1
)q
1
(j
1
) a(j
1
)q
n1
(j
1
)

a(j
p
)q
0
(j
p
) a(j
p
)q
1
(j
p
) a(j
p
)q
n1
(j
p
)
(D.11)
|A| =
a
0
a
1
.
.
.
a
m
, |B| =
b
0
b
1
.
.
.
b
n1
, |B| =
a(j
p
)q
n
(j
p
)
.
.
.
a(j
1
)q
1
(j
p
)
a(j
1
)q
n1
(j
p
)
.
.
.
a(j
p
)q
n
(j
p
)
(D.12)
56
APPENDIX E
Numerical results: tables
Range (% of
utter speed)
Degree of the polynomial
2 3 4 5 6 7 8
E
r
r
o
r
(
i
n
%
)
-92,67 37,47 12,81 7,37 3,53 2,43 1,40 10 - 75
-91,28 27,63 9,89 5,35 2,71 1,76 1,05 10 - 80
-89,86 19,26 7,06 3,65 1,90 1,19 0,73 10 - 85
-88,40 12,18 4,47 2,22 1,17 0,71 0,45 10 - 90
-85,39 26,05 10,10 5,15 2,73 1,67 1,02 25 - 75
-83,54 19,49 7,62 3,80 2,03 1,21 0,74 25 - 80
-81,61 13,71 5,35 2,61 1,39 0,82 0,51 25 - 85
53,83 8,70 3,33 1,58 0,84 0,49 0,31 25 - 90
-72,72 13,36 5,41 2,62 1,38 0,78 0,47 50 - 75
49,19 9,98 4,02 1,91 1,00 0,56 0,33 50 - 80
31,05 6,99 2,77 1,29 0,66 0,37 0,23 50 - 85
18,48 4,38 1,68 0,76 0,39 0,22 0,14 50 - 90
Table E.1: Table with errors for every polynomial and range generated using damping extrap-
olation (d
= 0.128)
57
Chapter E. Numerical results: tables
Range (% of
utter speed)
2 3 4 5 6 7 8
E
r
r
o
r
(
i
n
%
)
-92,70 37,85 12,90 7,34 4,34 1,82 5,83 10 - 75
-91,30 27,90 9,98 5,41 2,70 0,73 -0,75 10 - 80
-89,86 19,36 7,16 3,65 1,83 1,46 1,16 10 - 85
-88,41 12,30 4,54 2,27 1,23 0,79 0,49 10 - 90
-85,41 26,58 10,75 6,00 5,95 -86,75 -100,00 25 - 75
-83,54 19,41 7,57 4,11 1,95 -1,60 -3,84 25 - 80
-81,63 13,85 5,50 3,02 2,03 2,23 -100,00 25 - 85
54,40 8,88 3,46 1,71 1,05 0,80 1,23 25 - 90
-72,67 12,52 3,97 -69,62 -100,00 -12,46 -100,00 50 - 75
49,28 10,49 4,30 1,49 -1,48 -60,54 -9,90 50 - 80
31,37 7,12 3,06 1,68 0,50 -60,14 -4,17 50 - 85
18,75 4,51 1,71 0,72 0,17 -0,64 -0,18 50 - 90
Table E.2: Table with errors for every polynomial and range generated using damping extrap-
olation (d
= 0.128) with noise

Range (% of
utter speed)
2 3 4 5 6 7 8
E
r
r
o
r
(
i
n
%
)
-92,70 37,85 12,90 7,34 4,34 1,82 5,83 10 - 75
9,867406 -5,49656 -6,79182783 -7,72281 -7,84467 -7,95976 -6,68647 60,48 - 91,90
3,287624 -5,35666 -6,140492413 -6,81515 -6,72788 1,612416 -100 60,48 - 92,86
-51,2212 -41,6368 17,73900807 -39,0449 -8,55046 -37,3171 -34,7055 62,86 - 89,52
1,921622 -5,14828 -7,285188644 -7,70003 -7,9471 -2,76277 -7,89193 62,86 - 91,90
-0,59443 -5,13101 -6,517388448 -6,71306 -6,88748 -37,3438 -37,6259 62,86 - 92,86
-100 40,23395 -100 -7,97316 -100 -33,846 -29,4329 67,14 - 89,52
0,999024 -6,37395 -7,351637808 -7,91157 -7,87339 -7,87673 -3,97933 67,14 - 91,90
-1,41048 -5,91573 -6,486303796 -6,93699 -100 -34,8332 -30,9541 67,14 - 92,86
Table E.3: Full results of the damping extrapolation (d
= 0.128) method on the experimental

data
58
APPENDIX F
Experimental results: tables
59

Thesis

Încărcat de

Informații document

Descriere originală:

Drepturi de autor

Formate disponibile

Partajați acest document

Partajați sau inserați document

Opțiuni de partajare

Vi se pare util acest document?

Este necorespunzător acest conținut?

Drepturi de autor:

Formate disponibile

Thesis

Încărcat de

Drepturi de autor:

Formate disponibile

Erasmus University College

Faculty of Industrial Sciences

= 0.128) with noise . . . . . . . . . . . . . . . . . . . . . . . . 58

= 0.128) method on the experi-

. We have to include an aerodynamic force since this is an aeroelastic problem

for the pitch mode.

as the distance between the springs, as the angle of attack and x

is non linear. This means that we have

, this is done by calculating the surface below the function

is shown in gure 3.2.

for every angle of from 0 to 11.

is that the system would

. It is more useful to perform a pk-analysis on a variety of dierent

, a light grey area is an indication of

= 0.128 is specically chosen

=0.128 Ustart=7.5475 Ustop=27.171

=0.128 Ustart=15.085 Ustop=27.153

at maximum. the second test was with the same d

but at a dierent angle of attack.

of 0.128m and the fourth test was with the same d

was set at 0.09 m and

= 0.128) with noise

= 0.128) method on the experimental

S-ar putea să vă placă și