Master Thesis Lorenzo Cantoni

RENE Master of Science,
KTH Royal Institute of Technology,

Stockholm, Sweden, Jun-Dec 2020.
Master Thesis Report

Loads Control Aerodynamic in
Offshore Wind Turbines
Lorenzo Cantoni
Authors
Lorenzo Cantoni <lorenzo.cantoni95@gmail.com>
School of Industrial Engineering and Management
KTH Royal Institute of Technology
Parties involved in the Project

Lorenzo Cantoni
KTH Royal Institute of Technology
Imperial College London
Vestas Wind Systems A/S
Examiners
Andrew Martin
Professor, Department of Energy Technology, KTH Royal Institute of Technology
Supervisor
Ricardo Vinuesa Motilva
Assistant Professor, KTH Royal Institute of Technology
ii
Abstract
Due to the increase of rotor size in horizontal axis wind turbine (HAWT) during the
past 25 years in order to achieve higher power output, all wind turbine components
and blades in particular, have to withstand higher structural loads. This upscaling
problem could be solved by applying technologies capable of reducing aerodynamic
loads the rotor has to withstand, either with passive or active control solutions.
These control devices and techniques can reduce the fatigue load upon the blades up
to 40% and therefore less maintenance is needed, resulting in an important money
savings for the wind farm manager. This project consists in a study of load control
techniques for offshore wind turbines from an aerodynamic and aeroelastic point of
view, with the aim to assess a cost effective, robust and reliable solution which could
operate maintenance free in quite hostile environments.
The first part of this study involves 2D and 3D aerodynamic and aeroelastic simulations
to validate the computational model with experimental data and to analyze the
interaction between the fluid and the structure. The second part of this study is
an assessment of the unsteady aerodynamic loads produced by a wind gust over the
blades and to verify how a trailing edge flap would influence the aerodynamic control
parameters for the selected wind turbine blade.
Keywords
Wind Turbine, Offshore, Blade, Rotor, Computational Fluid Dynamics, Airfoil,

Aerodynamic, Aeroelasticity, Simulation, Fluid, Structure, Interaction, Loads, Control,
Response, Harmonics, Vibration, Damping, Flutter, Trailing Edge, Flaps, Wind Gust,
Lift, Drag, Unsteady, Pressure, Velocity
iii
Acknowledgements
I would like to thank the following people, without whom I would not have been able
to complete this research project, and without whom I would not have made it through
my Masters Degree Programme with InnoEnergy!
The Aeroelasticity Group at Imperial College London, especially to my supervisor

Professor Mehdi Vahdati, whose insight and knowledge into the subject matter steered
me through this interesting research topic. And special thanks to Kirubakaran, whose
support as part of his PhD allowed my studies to go the extra mile (sorry for all the
extra work and countless questions Karan!).
The wind energy company Vestas, which gave me the opportunity to have access to a
lot of data useful for my purpose, especially with Dr. Tomas Vronsky who supported
me during the whole research period. Thanks also to Professor Ricardo Vinuesa to
be my supervisor at KTH and to guide me through the administrative process for the
completion of my Master Degree in Stockholm.
And my biggest thanks to my family for all the support you have shown me through
this research, the culmination of two years of distance learning. For my good friends
who have been by my side all my life, and for all the new friends I have made during
this fantastic InnoEnergy Master Program in Barcelona and Stockholm. Special thanks
mainly to Berto (Lini), Urielle, Beona, Martino, Anto, Albert and Filipe. Thanks a lot
also to my personal digital creator and old friend Luca Luke Martini to help me with
the creation of the digital contents and videos shown during the presentation of the
project, your help was fundamental for me man.
Finally, many thanks to all participants and places that took part in this study path and
enabled everything I have done to be possible.
iv
Acronyms
CFD Computational Fluid Dynamics
HAWT Horizontal Axis Wind Turbine
VAWT Vertical Axis Wind Turbine
LCOE Levelized Cost Of Energy
AFC Active Flow Control
DOF Degree of Freedom
UAE Unsteady Aerodynamics Experiment
BEM Blade Element Momentum
ABL Atmospheric Boundary Layer
LES Large Eddies Simulation
EOG Extreme Operating Gust
v
Contents
1 Introduction 1
1.1 Problem Overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.2 Scope of the Work and Project Outline . . . . . . . . . . . . . . . . . . 2
1.3 Wind Turbine Layout: HAWT vs VAWT . . . . . . . . . . . . . . . . . . 3
1.4 Load Control in a Wind Turbine . . . . . . . . . . . . . . . . . . . . . . . 5
1.4.1 Pitch Control and Stall Control . . . . . . . . . . . . . . . . . . . 6
1.4.2 Variable Blade Length . . . . . . . . . . . . . . . . . . . . . . . . 7
1.4.3 Aerodynamic Optimization . . . . . . . . . . . . . . . . . . . . . 8
2 Experiments and Literature Review 11

2.1 Experimental Testing for Wind Turbines . . . . . . . . . . . . . . . . . . 11
2.2 Literature Review of Previous Work . . . . . . . . . . . . . . . . . . . . 13
2.2.1 Aerodynamics of Wind Turbine . . . . . . . . . . . . . . . . . . . 13
2.2.2 Aeroelasticity of Wind Turbine . . . . . . . . . . . . . . . . . . . 13
2.2.3 Smart Rotors and Loads Control for Wind Turbines . . . . . . . 16
3 Aerodynamics and Aeroelasticy 20

3.1 Fundamental Dimensionless Quantities . . . . . . . . . . . . . . . . . . 20
3.2 Blade as Airfoil . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
3.3 Boundary Layer Theory . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
3.3.1 Concept of Boundary Layer . . . . . . . . . . . . . . . . . . . . . 25
3.3.2 Flow separation and Airfoil Stall . . . . . . . . . . . . . . . . . . 27
3.4 Flutter . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
3.4.1 Classical Flutter . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
3.4.2 Stall Flutter . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
3.4.3 Flutter in Airfoils . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
vi
CONTENTS
4 2-D Validation & Analysis 36

4.1 Geometric Model of the S809 Airfoil . . . . . . . . . . . . . . . . . . . . 36
4.2 Geometric Model of the NREL Wind Turbine . . . . . . . . . . . . . . . 37
4.3 Test Case Configuration . . . . . . . . . . . . . . . . . . . . . . . . . . . 40
4.3.1 S809 configuration . . . . . . . . . . . . . . . . . . . . . . . . . . 40
4.3.2 NREL Phase VI configuration . . . . . . . . . . . . . . . . . . . . 41
4.4 CFD Methodology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42
4.5 Results and Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45
4.5.1 2-D Grid of the S809 Airfoil . . . . . . . . . . . . . . . . . . . . . 45
4.5.2 Aerodynamic Results - Steady . . . . . . . . . . . . . . . . . . . 47
4.5.3 Aerodynamic Results - Unsteady . . . . . . . . . . . . . . . . . 56
4.5.4 Aerodynamic Results - Trailing Edge Flap . . . . . . . . . . . . 58
5 3-D Fluid Structure Interaction 64

5.1 CFD Case Configuration . . . . . . . . . . . . . . . . . . . . . . . . . . 64
5.2 CFD Methodology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 65
5.3.1 3-D Grid of the Blade . . . . . . . . . . . . . . . . . . . . . . . . 68
5.3.2 Aerodynamic Results . . . . . . . . . . . . . . . . . . . . . . . . 71
5.3.3 Aeroelastic Results . . . . . . . . . . . . . . . . . . . . . . . . . 75
6 Wind Gust Analysis with ATE 80

6.1 Gust Models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 81
6.2 IEC Standard . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 82
6.2.1 Extreme Operating Gust . . . . . . . . . . . . . . . . . . . . . . 83
6.3 CFD Case Configuration & Methodology . . . . . . . . . . . . . . . . . 84
6.4.1 2-D Grid & Gust Model . . . . . . . . . . . . . . . . . . . . . . . 86
6.4.2 Stationary Flap Response . . . . . . . . . . . . . . . . . . . . . 88
6.4.3 Quasi-Steady Response . . . . . . . . . . . . . . . . . . . . . . 90
6.4.4 Fully Transient Response . . . . . . . . . . . . . . . . . . . . . . 93
7 Conclusions 97
7.1 Future Work . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 98
References 100
vii
List of Figures
1.1.1 Continuous growth of rated power during the last years . . . . . . . . . 2

1.3.1 Components of an HAWT . . . . . . . . . . . . . . . . . . . . . . . . . . 4
1.3.2 Layout of a VAWT . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
1.4.1 Simplified power curve for commercial wind turbine . . . . . . . . . . . 5
1.4.2 Variable blade length concept . . . . . . . . . . . . . . . . . . . . . . . . 7
1.4.3 Schematic representation of existing AFC devices . . . . . . . . . . . . 8
1.4.4 Flexible trailing edge flap . . . . . . . . . . . . . . . . . . . . . . . . . . 9
1.4.5 Flexible leading edge flap . . . . . . . . . . . . . . . . . . . . . . . . . . 10
1.4.6 Active Gurney flap . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
3.2.1 Cross section of a modern wind turbine blade . . . . . . . . . . . . . . . 23

3.2.2Schematic of an airfoil . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
3.3.1 Profile of a typical boundary layer . . . . . . . . . . . . . . . . . . . . . 25
3.3.2Boundary layer on a flat plate . . . . . . . . . . . . . . . . . . . . . . . . 26
3.3.3 Flow separation schematics . . . . . . . . . . . . . . . . . . . . . . . . . 27
3.3.4Airfoil pressure distribution (left) and stall of the airfoil (right) . . . . . 28
3.4.1 Rotation and plunge motion for an airfoil experiencing flutter . . . . . 30
3.4.2Schematic of plunge (left) and twist (right) modes of an airfoil . . . . . 32
3.4.3Schematic of the phasing between force and velocity . . . . . . . . . . . 32
4.1.1 Schematic of the NREL S809 airfoil . . . . . . . . . . . . . . . . . . . . 36

4.2.1 UAE phase VI rotor in the NASA Ames wind tunnel . . . . . . . . . . . 37
4.2.2Blade platform dimensions . . . . . . . . . . . . . . . . . . . . . . . . . 38
4.2.3Blade surface pressure tap locations . . . . . . . . . . . . . . . . . . . . 39
4.3.1 Velocity and aerodynamic forces conventions . . . . . . . . . . . . . . . 40
4.5.1 Computational domain and far-field boundaries . . . . . . . . . . . . . 45
4.5.2 Detailed mesh around the airfoil . . . . . . . . . . . . . . . . . . . . . . 46
viii
LIST OF FIGURES
4.5.3 Detailed mesh around the leading edge . . . . . . . . . . . . . . . . . . 46

4.5.4 Lift coefficient vs. angle of attack . . . . . . . . . . . . . . . . . . . . . . 48
4.5.5 Drag coefficient vs. angle of attack . . . . . . . . . . . . . . . . . . . . . 49
4.5.6 Pressure distribution around the airfoil at various angles of attack . . . 51
4.5.7 Streamline plots of 5.13° (left) and 8.20° (right) angles of attack . . . . 53
4.5.8Steady contours of absolute velocity at AOA from 0° to 20.15° . . . . . 54
4.5.9 Flow separation at 14.24° (left) and 20.15° (right) angles of attack . . . 55
4.5.10Eddies formation for airfoil at full stall condition . . . . . . . . . . . . . 56
4.5.11Unsteady lift curve at 20.15° (Full stall) . . . . . . . . . . . . . . . . . . 57
4.5.12Lift spectrum in frequency domain . . . . . . . . . . . . . . . . . . . . . 57
4.5.13Cross sections for different flap configurations: +10°, -10°, 0° . . . . . 59
4.5.14Lift force for different flap configurations: +10°, 0°, -10° . . . . . . . . 60
4.5.15Drag force for different flap configurations: +10°, 0°, -10° . . . . . . . . 61
4.5.16Lift force for different wind speeds and flap configurations . . . . . . . 62
5.1.1 CAD model of the wind turbine blade . . . . . . . . . . . . . . . . . . . 64

5.1.2 Front view of the wind turbine blade . . . . . . . . . . . . . . . . . . . . 65
5.3.1 Computational domain . . . . . . . . . . . . . . . . . . . . . . . . . . . 68
5.3.2 Meshed computational domain . . . . . . . . . . . . . . . . . . . . . . . 69
5.3.3 Detailed mesh around the blade . . . . . . . . . . . . . . . . . . . . . . 69
5.3.4 Detailed mesh of the blade - 1 . . . . . . . . . . . . . . . . . . . . . . . . 70
5.3.5 Detailed mesh of the blade - 2 . . . . . . . . . . . . . . . . . . . . . . . 70
5.3.6 Velocity streamlines around the turbine . . . . . . . . . . . . . . . . . . 71
5.3.7 Blade velocity in standard frame for a 3-Blade rotor . . . . . . . . . . . 72
5.3.8Pressure contour on the blade front surface . . . . . . . . . . . . . . . . 73
5.3.9 Pressure contour on the blade back surface . . . . . . . . . . . . . . . . 73
5.3.10Pressure contour on the YZ plane . . . . . . . . . . . . . . . . . . . . . 74
5.3.11Blade deflection regions . . . . . . . . . . . . . . . . . . . . . . . . . . . 76
5.3.12Blade deflection compared to unloaded case . . . . . . . . . . . . . . . 76
5.3.13Equivalent stress - Blade front face . . . . . . . . . . . . . . . . . . . . . 77
5.3.14Equivalent stress - Blade back face . . . . . . . . . . . . . . . . . . . . . 77
5.3.15Root radial force . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 78
5.3.16Bending moment reaction . . . . . . . . . . . . . . . . . . . . . . . . . . 79
6.0.1 Characteristic parameters of an average wind gust . . . . . . . . . . . . 80
ix
LIST OF FIGURES
6.2.1 Canonical Mexican hat gust shape profile (z = zhub = 12.2m) . . . . . . 84

6.3.1 S809 airfoil with ATE and small gap . . . . . . . . . . . . . . . . . . . . 85
6.4.1 Detailed mesh around the airfoil . . . . . . . . . . . . . . . . . . . . . . 86
6.4.2Inlet wind speed - mexican hat shape . . . . . . . . . . . . . . . . . . . 87
6.4.3Lift force generated by the wind gust . . . . . . . . . . . . . . . . . . . . 88
6.4.4Drag force generated by the wind gust . . . . . . . . . . . . . . . . . . . 88
6.4.5 Velocity contours of wind gust evolution: t=0 s, t=2 s, t=6 s, t=12 s . . . 89
6.4.6Airfoil configuration with flap at -5° inclination angle . . . . . . . . . . 90
6.4.7 Lift force comparison between two flap configurations: 0° and -5° . . . 91
6.4.8Vel. contours comparison between first and second case at t=3.5 s . . . 92
6.4.9Pres. contours comparison between first and second case at t=3.5 s . . 92
6.4.10Lift comparison between two flap cases: 0° and movable (±10°) . . . . 93
6.4.11Drag comparison between two flap cases: 0° and movable (±10°) . . . 94
6.4.12Flap inclination during the wind gust . . . . . . . . . . . . . . . . . . . 95
6.4.13Velocity contours during the wind gust evolution (t=0 s → t=12 s) . . . 96
7.1.1 Deep reinforcement learning for aerodynamic optimization . . . . . . . 99
x
List of Tables
4.3.1 Test cases at 15 m/s free stream velocity . . . . . . . . . . . . . . . . . . 40

4.3.2Computational cases and parameters . . . . . . . . . . . . . . . . . . . 41
4.4.1 Wind speeds, densities and pressures . . . . . . . . . . . . . . . . . . . 44
4.5.1 Aerodynamic force coefficients prediction . . . . . . . . . . . . . . . . . 47
4.5.2 Calculation errors of aerodynamic coefficients . . . . . . . . . . . . . . 52
4.5.3 Aerodynamic force prediction . . . . . . . . . . . . . . . . . . . . . . . 60
4.5.4 Lift force for different wind speeds . . . . . . . . . . . . . . . . . . . . . 62
5.3.1 Blade material properties . . . . . . . . . . . . . . . . . . . . . . . . . . 75

5.3.2 Thickness specifications . . . . . . . . . . . . . . . . . . . . . . . . . . . 76
5.3.3 Force values in all directions . . . . . . . . . . . . . . . . . . . . . . . . 78
5.3.4 Moment values in all directions . . . . . . . . . . . . . . . . . . . . . . . 79
xi
Chapter 1
Introduction
Due to the uncertainties and risks associated with the price of fossil fuel, investment
in renewable energy sources such as wind power, with the nature of being a stable
and long term alternative, has been an increasingly favourable choice in the energy
sector with growing investment in both the European Union and the United States
[78]. Among all the renewable energy sources, both onshore and offshore wind power
are among the cleanest when the environmental impact of various renewable energy
sources are considered [82]. Therefore, there have been growing research interests
in assessing and optimising wind turbine designs, especially how to increase the rotor
size maintaining good safety structural requirements.
1.1 Problem Overview

During the past 25 years, the size of utility scale horizontal axis wind turbines grew
uninterrupted, constantly seeking a reduction of the cost of energy, as it is shown
in figure 1.1.1. As the rotor size increases, all wind turbine components, and the
blades in particular, have to withstand higher loads. In the past, the higher structural
requirements were mainly satisfied by increasing the weight of the structure, or by
employing more expensive compounds with an higher specific strength [9]. According
to Sieros et al. [67] such growing trend is likely to hit a limit, above which further
upscale of the turbine size would not yield any benefit in term of cost reduction,
unless technological breakthrough are introduced to allow for further upscale in a cost-
efficient way.
1
CHAPTER 1. INTRODUCTION
Figure 1.1.1: Continuous growth of rated power during the last years
The upscale problem could be solved by applying technologies capable of reducing the
aerodynamic loads the rotor has to withstand, either with passive or active load control
solutions. Passive load control could be achieved by bend-twist aeroelastic coupling of
the blades deformations: the blades would respond to an increase of the aerodynamic
forces with a torsion deformation that twist the blade toward lower angles of attack, and
therefore reduces aerodynamic loads [14]. Active load control is instead achieved by a
wind turbine rotor that, through a combination of sensor, control units, and actuators,
compensates for the variation in the wind field swept by the blades, and thus actively
reduces the aerodynamic loads it has to withstand.
1.2 Scope of the Work and Project Outline

The aim of this project is to calculate and reduce the loads upon an offshore wind
turbine blade with an aerodynamic simulation tool able to account for the presence of
an adaptive trailing edge flap. In this way it is possible to gain an insight on the effects
this solution has on the aerodynamic and aeroelastic forces, hence on the response
of the structure and loads experienced by the blades. With these insights it will be
possible to understand the blade loads reduction and, in particular, the changing in
lift due to input wind gust will be analysed. Thanks to this load reduction, it would be
possible to minimize the maintenance in the hostile environment of an offshore wind
turbine.
2
The starting point of this study is a literature review about aerodynamic, aeroelasticity
and structural mechanics of an offshore wind turbine, to gain a good problem overview
and to understand the state of the art of load control techniques. Moreover, a validation
of the numerical method will be done to assess its accuracy, comparing the results of
the Computational Fluid Dynamics (CFD) simulation with experimental data obtained
in the wind tunnel test.
Afterwards, a 3D fluid structure-interaction analysis will be performed to a specific

wind turbine blade, using the provided specifications. From the results it will be
possible to analyze the reaction of the blade to the incoming wind, studying the total
deformations, stresses and forces experienced by the blade.
Finally, the effectiveness of applying a trailing edge flap to the airfoil during a wind
gust will be assessed. The airfoil response with its standard configuration will be
tested and then compared to the configuration including the movable flap, in order
to understand the importance and the possible application of this devise in reducing
the loads produced by an incoming wind gust.
1.3 Wind Turbine Layout: HAWT vs VAWT

Depending on the axis relative to the ground plane that wind turbines rotate about,
they can be divided into two categories: Horizontal Axis Wind Turbine (HAWT) and
Vertical Axis Wind Turbine (VAWT), among which the former is more common and
better developed for commercial applications.
Wind turbines date back to 200 B.C. in the world’s history, and the most famous
HAWTs were evidently the Dutch windmills. The most important components of an
HAWT are located on the top of the tower, which capture the kinetic energy of the
wind and ultimately convert it in to electrical energy. The reception of mechanical
energy from the air stream is completed by the rotation of the turbine blades about the
rotor shaft due to the lift force generated on the blades. The torque induced drive then
goes through gear box and the generator does the rest of the job [27]. An horizontal
axis wind turbine must face directly to the wind in either upwind or downwind setup
in order to achieve the best efficiency and also avoid fatigue failure by cyclic loading.
Older and smaller HAWTs can be adjusted manually with the help of wind vanes, while
modern larger ones usually employ automated yaw drives [27].
3
Figure 1.3.1: Components of an HAWT
VAWT has quite different structural layout and operation conditions. The gearbox and
generator which sit on the top of the tower of a HAWT now can be placed directly on
the ground as the rotating shaft goes vertically to the bottom. Moreover, vertical axis
wind turbines can operate regardless of the incoming wind direction as they generally
have helical blades along the axis, however some may require a start motor at low wind
speed [4].
Figure 1.3.2: Layout of a VAWT
4
1.4 Load Control in a Wind Turbine
The key quantity when it comes to assess the competitiveness of a source of energy is
the Levelized Cost Of Energy (LCOE), given by Equation 1.1, which must be kept as low
as possible [25].
Sum of Costs Over Lifetime (Capital Costs + O&M Costs)

LCOE = (1.1)
Sum of Energy Produced Over Lifetime
Looking at this equation, there are several ways to lower the LCOE. The first one is to
decrease the capital cost. This can be achieved by decreasing the amount of materials or
improving manufacturing techniques. Another way would be to reduce the Operation
and Maintenance (O&M) costs by making more reliable turbines [25].
The last way, which is the most relevant in this project, is to increase the Sum of
Energy Produced Over Lifetime. There are two factors influencing this quantity for
a wind turbine: its size and its lifetime [25]. These two parameters are linked: by
increasing the rotor diameter, which is the current design trend, one would impose
higher loads to the structure and hence reduce its lifetime. Nowadays, it is thus a
prime objective to control the loads on a wind turbine and more and more techniques
are being investigated to this end. Of course, this load control cannot be done without
considering the power output of the wind turbine, the main quantity of interest as it
is directly related to the production of energy. Figure 1.4.1 presents the typical power
curve of a commercial wind turbine.
Figure 1.4.1: Simplified power curve for commercial wind turbine
5
The regions of interest when it comes to load control are regions 2 and 4. In region 2,
the turbine operates below design - or rated - power. Near region 4, the wind speed
is higher than the design wind speed and thus the power output is the same but with
higher loads which can hence be detrimental to the wind turbine. Several techniques
are currently being investigated for the purpose of controlling the loads on the blades.
They are better understood considering the general equation of the lift experienced by
a wind turbine blade.
∫ b
1 2
L= ρ[CLα (θpitch )Vwind + (2πnr)2 c]dr (1.2)
0 2
With: b the blade lenght, θpitch the blade incidence angle, Vwind the wind speed, n the
rotor speed and c the chord of the blade. From this equation, four control strategies
can be pointed out:
• Pitch Control and Stall Control
• Variable Speed Rotor
• Variable Blade Length
• Blade Section Aerodynamic Optimization
1.4.1 Pitch Control and Stall Control
The over working wind turbine can be controlled by adjusting the blade pitch angle,
which in turn reduces the angle of attack, so that the loading and thus the power output
is limited. This can be achieved by either employ a mechanical device which changes
the pitch by balancing the excessive centrifugal force, or use electronic controller to
alter the pitch angle according to measured power output [1].
Regarding stall control, in contrast with pitch control methods, wind turbine blades
are designed to exhibit an increase in angle of attack once the critical overload point
is reached. The rise of incidence stalls the blade by creating flow to separate from
the suction surface to become turbulent, which consequently reduces the lift and thus
the rotational speed of the blades. Similarly, two approaches are possible to achieve
this. Passive stall control depends on the aerodynamic and structural dynamic design
of the wind turbine blades to ensure the angle of attack reaches stall condition when
6
wind speed exceeds a safe limit. It usually comes with twisted blades along its span to
produce a gradual stall and avoid sudden change of load. The advantage of passive stall
control is the exclusion of moving parts, which decreases the possibility of component
failure.
The other approach is to control the stall of wind turbines actively. This method is
very much similar to pitch power control, except that the electronic controller increases
instead of decreases the angle of attack to create stall. Compared with the passive
method, active stall controlled wind turbines are generally able to maintain their rated
power at high wind speeds without a significant drop [1].
1.4.2 Variable Blade Length
The idea is the implementation of an extendable tip blade out of a root blade, in order
to increase the diameter in case of low wind speed to maximize the energy production
and decrease the diameter in high wind speed conditions to alleviate the loads on the
turbine [25]. Results of experiments showed a potential increase in power generation
up to 50%, but increase in extreme and fatigue loads in low wind speed conditions [55].
This concept is presented in figure 1.4.2.
Figure 1.4.2: Variable blade length concept
7
1.4.3 Aerodynamic Optimization
As previously mentioned, the purpose of this project is to control the loads on the
blades by means of Active Flow Control (AFC) devices. The term ”active” implies the
use of actuators and sensors along with integrated control theory able to adapt to any
change in the flow conditions that can have negative impact on the blades [25].
Figure 1.4.3 shows different passive/active devices on which experimental or numerical
researches have been conducted. It is worth noticing that, even if some of this devices
showed promising results, none of them have been tested on full-scale wind farms [25].
Moreover, not all of them have been investigated for wind turbines specifically, but for
other related fields (i.e. aeronautic/aerospace).
Figure 1.4.3: Schematic representation of existing AFC devices
Trailing Edge Flap
This is probably the most promising AFC device thanks to its proven success in
aircraft control and its principle is exactly the same, which is described in Figure 1.4.4:
deflection of the flap to the pressure surface generates an increase in aerodynamic load,
while a deflection to the suction surface decreases the aerodynamic load [25].
8
The implementation of the T.E. flap on wind turbines is the subject of numerous
investigations with emphasis on load alleviation purpose [2] [6], even if there has
been some interest also to rotor stall [51]. Thanks to wind tunnel measurements,
Pechlivanoglou and al. [51] proved the T.E. flap to have a significant lift variation
potential which may even eliminate the need for pitch systems. They also conducted
a parametric study to assess the optimal flap position and length and found out that
the optimal configuration leads to a load reduction potential characterized as the root
bending moment of 52%. Moreover, Buhl and al. [16] showed that a T.E flap can be
suitable when it comes to deal with the unsteady loads as well. Using a 2D model, they
found indeed that when a blade experiences a wind step of 10 to 12 m/s the standard
variation of the lift can be reduced of up to 95% and in case of a turbulent wind field, the
reduction can be of 81%. However, different studies conducted on the T.E. flap might
prevent it from being installed on commercial wind turbine in the near future due to
the complexity of its linkage with the remaining part of the blade, its expected slow
response and noise that the gap between the main blade and the flap might generate.
Figure 1.4.4: Flexible trailing edge flap
Leading Edge Flap
The leading leading edge flap (L.E. flap) works the same way as the flexible T.E. flap,
except that it is fixed on the leading edge of the blade, as it is shown in figure 1.4.5.
Some concepts have been suggested, but the development of the L.E. flap is far from
being as active as for the T.E. flap.
9
Nevertheless, Pechlivanoglou and al. [51] did test this solution in a wind tunnel and the
effect on the lift is interesting: the lift variation at very high or very low angle of attack
α is negligible. The control influence of this AFC solution is thus generally limited to
the near-stall region which could make it a suitable candidate for active stall control.
Figure 1.4.5: Flexible leading edge flap
Active Gurney Flap
This device consists of a simple flat plate (on the order of 1% of the chord length) as
shown in Figure 1.4.6. Different investigations reveal promising results: Van Dam and
al. [50] showed that it is suitable for load alleviation and it has been confirmed by the
results obtained in their wind tunnel by Pechlivanoglou and al. [51]. The reduction
of the root bending moment value reaches 35.8%. Nevertheless, they also showed that
these results were obtained with a constant high deflection of the Gurney Flap contrary
to the T.E. flap, which means the control authority of the Gurney Flap is much more
limited than the T.E flap. This is due to the small size of this device.
Figure 1.4.6: Active Gurney flap
10
Chapter 2
Experiments and Literature Review
2.1 Experimental Testing for Wind Turbines
The experimental reference test case was chosen to be the Unsteady Aerodynamics
Experiment of wind turbines conducted by the U.S. National Renewable Energy
Laboratory, which was comprehensively staged and has a well recorded database
stored in the International Energy Agency Annex programmes [40]. The founding
component of the turbine blades exclusively used in this experiment was the S809
airfoil designed by NREL and Airfoil Inc.
NREL UAE Phase VI
The U.S. Department of Energy’s National Renewable Energy Laboratory (NREL)

conducted the Unsteady Aerodynamics Experiment (UAE) testing wind turbines in the
world’s largest wind tunnel (24.4×36.6 m or 80×120 ft), of that time, located at NASA-
Ames Research Centre in Silicon Valley, California [68]. The testing was completed
in May 2000 and provided data for more than 1,700 different turbine test conditions.
The test consisted of 6 phases investigating 6 different wind turbines, where Phase II-V
were detailed in the IEA Annex XIV project and Phase VI is most studied by authors
all over the world.
The Phase VI experiment was carried out on a two bladed, tapered and twisted 10 m
diameter turbine inside the wind tunnel with thoroughly controlled inflow conditions,
which marks the first time in history of comprehensive in door testing of large wind
11
CHAPTER 2. EXPERIMENTS AND LITERATURE REVIEW
turbines [15]. The advantages of wind tunnel tests over field tests being more reliable
and accurate due to minimisation of inflow turbulence make the NREL Phase VI testing
more suitable as a study case. A ”blind comparison” was staged after the completion of
the wind tunnel testing in which worldwide wind turbine experts and institutes were
invited to predict the NREL wind turbine behaviour, either with the help of Blade
Element Momentum (BEM) method or CFD.
DUT wind tunnel testing of NREL S809 airfoil
The S809 airfoil was originally designed by Griffin [31] from Airfoils Incorporated
under the supervision of NREL in 1989. The purpose of designing this thick airfoil was
specifically for horizontal axis wind turbine applications, which was very different from
the designs of National Advisory Committee for Aeronautics (NACA) and National
Aeronautics and Space Administration (NASA) airfoils. Restrained maximum lift,
insensitive to roughness and low profile drag were the design objectives and had
been successfully achieved. Experimental verification was carried out in the low
turbulence wind tunnel of the Delft University of Technology (DUT) low speed
laboratory, the Netherlands. The testing was conducted at various Reynolds number
ranging from 1,000,000 to 3,000,000 and the effect of hysteresis was investigated.
Comparison between prior theoretical calculations and experimental results showed
good agreement.
OSU wind tunnel testing of NREL S809 airfoil
The Ohio State University (OSU) of the United States was contracted by National
Renewable Energy Laboratory (NREL) to conduct a wind tunnel test programme
to assess the performance of NREL wind turbine airfoils [59], among which the
S809 airfoil is the most recognised one due to NREL’s Unsteady Aerodynamics
Experiment Phase II-VI. Steady and unsteady angles of attack were tested during the
experiments.
12
2.2 Literature Review of Previous Work

A significant amount of work has been done by worldwide authors to study the accuracy
of CFD codes in predicting the aerodynamic behaviour of air flow at various speeds
over the NREL Phase VI two-bladed 10m diameter wind turbine [5] [58] [60] [70].
Moreover, studies of wind turbine aeroelasticity are also reviewed [35] [76].
2.2.1 Aerodynamics of Wind Turbine
Traditionally, due to relatively low tip flow speed in earlier wind turbine studies, CFD
works were performed primarily with incompressible solvers such as widely used code
EllipSys3D developed in cooperation between Technical University of Denmark (DTU)
and Risø National Laboratory [5] and CFD-ACE solver developed by ESI Group which
was used by Wolfe and Ochs to study the S809 airfoil [79]. However, as the size of
wind turbines increase, the compressibility effects need to be accounted for at near
blade tip regions. Thus, studies using compressible codes were also carried out by
many researchers, including OVERFLOW developed by Buning et al at NASA [20].
The grid construction was generally achieved using multi blocks structured mesh
among the literature, with sizes ranging from 1 to 12 million cells [30]. Most authors
considered only the isolated rotor [21] [30], however, some also investigated the
interactions between the tower and blades by modelling the whole assembly with a
tower [26] [81]. There are also studies about the effects of the wind tunnel walls on the
turbine performance, and specific concentration on aeroacoustic problems relating the
tip vortex region or the influence of root section [3].
2.2.2 Aeroelasticity of Wind Turbine
The current aeroelasticity design issues relating to wind turbines are well presented
and analysed in the overview paper by Madsen and Hansen [35] from Risø national
laboratory for sustainable energy of Technical University of Denmark (DTU), which
is the world leading wind energy research group. Some other overview papers of
wind turbine aeroelasticity published by researchers including Sepahy et al. [65] and
Vermeer et al. [76], besides Madsen and Hansen, were also reviewed.
13
Load concerns
Evolving from the old NACA type general airfoils, the current wind turbine blade
designs usually adopt specific tailored characteristics for particular application. This
involves various design parameters such as design lift, drag and noise characteristics
as well as structural stiffness. Robustness against varying load conditions is now
becoming more and more important as wind turbines becomes bigger and more
flexible. The load induced on wind turbine rotors mainly comes from the aerodynamic
forcing of the incoming free stream wind. However, other site parameters including
turbulent wake generated loads for turbines within wind farms and also the controller
load. This can be the unsteady loading on the blades when they are travelling through
turbulent eddies caused by the wake flow of other turbines, or the bending load on
the tower induced by the thrust on the turbine rotor. Even though the tower bending
mode by thrust along the wind flow is usually damped out by the structure at low wind
load, the bending mode perpendicular to the flow direction is generally left undamped
or lowly damped resulting in possible fatigue failure. Therefore, simulation of wind
turbine load conditions plays an important role in the design stage with the key tool to
solve it being aeroelastic analysis.
For the past decade, aeroelastic codes were developed and tested all over the world.
Among them, the most developed and widely used codes are HAWC2 from Risø DTU
[43], BLADED by the world’s largest renewable energy consultancy GL Garrad Hassan
[13] and FLEX4 by Stig Øye [53]. Multi body finite element method is used in HAWC2,
while the latter two codes are based on modal form computations. It is very worth
noting that all three aforementioned aeroelastic codes utilise BEM theory to simulate
aerodynamic performance, which holds some disadvantages when compared with CFD
codes.
Modes of Instability
The main objective of aeroelastic analysis is to capture the possible occurrence of

aeroelastic instability, which was detailed explained in the U.S. National Renewable
Energy Laboratory (NREL) research paper [11] and Hansen’s paper [38].
The aeroelastic instability originates from the unstable behaviour of wind turbine
modal dynamics, where two common types of instabilities are classical flutter and
stall induced vibration, or stall flutter. Classical flutter can be initiated by coupling
14
of several structural modes when flow is still attached. Once the rotor blade is stalled,
even the normally stable modes like flapping mode could become negatively damped
leading to stall flutter. Knowing the modal dynamic behaviour of a general wind
turbine that how the lower order modes can be involved in aeroelastic response,
specific aeroelastic flutter motions need to be studied, which will be further explained
in the next chapter.
Stall flutter is able to take place when the turbine blade is stalled, where the air
flow is separated from the blade surface causing fluctuation of aerodynamic forces.
Almost any type of vibration mode will be susceptible once stalled, especially for blades
designed with abrupt stall characteristics airfoils. The stall induced vibration can be
compensated by structural damping, but normally by a very limited amount. This
was further proved by the researches on aerodynamic stall flutter of wind turbines by
many authors [12] [23] [36] [38] [56] [57] [61] [62] [73]. On the other hand, classical
flutter can be initiated without the separation of flow but coupling of the first torsional
and flapwise bending modes. This flutter motion begins with a torsional deformation
leading to unfavourable phased force with flapwise bending, leading to highly negative
damping and thus instability. Researches on isolated blade flutter by Lobitz [44]
[45] had showed that unsteady computations result in higher flutter speed predictions
compared with quasi steady calculations caused by the decrease of the lift slope. Also,
as rotor gets bigger, the first torsional mode can couple with the second flapwise mode
to produce flutter. Moreover, studies of full turbine assembly flutter [37] also proved
Lobitz’s theory about larger bladed turbines stating that the aeroelastic damping will
be decreased with smaller natural frequency ratio between the two coupled modes.
Despite various studies and conclusions made on wind turbine aeroelasticity, there
are still a lot of researches undergoing currently, including stall induced vibrations
at standstill, dynamic stall, structural damping enhancement, flutter limits for yawed
flow and non linear coupling flutter.
15
2.2.3 Smart Rotors and Loads Control for Wind Turbines
This part is focused on available literature and research of interest for smart rotor
applications and progress related to wind turbine applications. Although the amount
of available progress in this specific field is quite new and can not be compared to the
available progress in rotor-craft, inspired by these similar applications and by the need
for more advanced active control, research programs investigate the possibilities for
such concepts in wind turbines.
DUWIND (TU Delft)
An important research work, concerning feasibility studies on smart dynamic rotor

control for wind turbine applications, has been conducted by Marrant, van Holten
and Kuik for the STW project ”Smart Dynamic Rotor Control for Large Offshore Wind
Turbines”.
The first report [47] deals with the inventory of rotor design options and possible load
reductions. The fluctuating loads on a wind turbine are described and possibilities
of influencing fatigue loads or structural loads are discussed. Active rotor control
concepts are presented, which include pitch control concepts (collective, cyclic, higher
harmonic), individual blade control (part-span pitch, aileron control, active twist)
and active damping of blade and tower vibrations. Also semi-active and passive
control options are discussed (passive tips, self-twisting blades, compliant blades).
Concluding, individual pitch control was considered as the most powerful strategy
(because of the ability to reduce aerodynamic loads both due to temporal and spatial
variances in inflow), but active structural damping was also considered interesting (for
reducing the loads due to axial tower mode and blade flapwise bending mode).
The second report [48] summarizes present techniques with regard to sensors,
actuators, aerodynamic constructions and control strategies and their application on
large offshore pitch regulated variable-speed wind turbines. Regarding sensors, strain
gauges, accelerometers and force sensors were analyzed. Piezoelectric force sensors at
the blade root were considered a feasible solution for the measurement of aerodynamic
loads. Optical fibers were considered expensive and not well established for measuring
strains on blades. Passive accelerometers were considered a good solution due to their
low bandwidth and low frequency limit.
16
Regarding control strategies, four control strategies that had been developed to
actively suppress vibrations in rotorcraft are analyzed: a feed forward adaptive
control algorithm (Broadband filtered X-LMS and tonal control), a Fourier synthesis
algorithm, a real-time adaptive neural network controller and an iterative learning
controller. Some important considerations regarding the connection between
controller design and wind turbine design are also pointed out. Regarding actuators,
many categories are analyzed: conventional (pneumatic, hydraulic, electro-motors),
smart materials (electrorheological, magnetorheological, shape memory alloys (SMA),
electrostrictive, piezoelectric, magnetostrictive).
The third report [49] summarizes theoretical and software tools, which are necessary
for the design of a smart rotor. Also hardware for testing of a smart rotor is mentioned.
It is assumed that existing quasi-steady BEM models are not enough for smart rotor
applications, vortex methods may be more reliable and unsteady aerodynamics are
very important for the simulation of the fluctuating inflow and loads. It is also
stated that new reliable dynamic models for smart materials and control algorithms
must be incorporated. Regarding software tools, WOBBE is mentioned for stability
calculations, and PHATAS/FOCUS, DAWIDUM and ADAMS for load calculations.
For controller requirements, a PC with Matlab/dSPACE is necessary. Concerning
hardware and testing, mainly wind tunnel and laboratory requirements are discussed.
An interesting first approach to the application of active control on wind turbine blades
was that of Aguirre at the TU Delft [52]. A blade section with a trailing edge flap was
used for the wind tunnel experiment. Strain gauges were implemented for measuring
the lift change on the blade, potentiometers for measuring the pitch angle and the flap
angle and servomotors as actuators for changing the pitch angle and the flap angle.
The experimental procedure was based on the idea of changing the pitch angle of the
blade and then, using a real time controller, activating the flap in order to reduce the
changes in lift. Steady measurements were firstly performed for different angles of
attack and flap. Unfortunately no model could be build because of the disturbances
in the signals. This experiment was a good simple approach to use active control on a
blade and showed the challenges and requirements for such efforts.
17
ADAPWING (RISØ)
A very important contribution to the research about smart rotor control for wind
turbines is made by the wind energy research group of Risø, the National Laboratory
in Denmark. Through the project Adapwing (Adaptive wing geometry for reduction
of wind turbine loads), the possibilities of eliminating the fast fluctuating loads on
wind turbines as a consequence of the turbulence in the wind and the tower shadow
through the use of adaptive wing geometry regulation are being investigated. By
cooperation between researchers and students from DTU many interesting reports
have been presented.
All cover computational studies on variable geometry airfoils using aero-servo-

elastic models. Firstly, in the MSc thesis of Basualdo, a two-dimensional model is
developed [8]. The aerodynamics is described using ”Thin Airfoil Theory” for attached,
irrotational, incompressible flow and solved numerically using a Panel Method for
steady and unsteady situations. The structural model consists of two degrees of
freedom (edgewise, flapwise), expressed by spring and damper models. For control,
a simple control algorithm is used, with a target control strategy to minimize the tip
deflection variation of the blade. The results show the potential of such a control. The
standard deviation of the airfoil displacements has been reduced to 25% of the value
corresponding to no control, during the 2 sec simulations. In all the other simulations
(e.g. 100 sec, gust) the decrease in the displacement amplitudes is evident. But, it must
be pointed out that the model consists of a simple control simulation, not taking into
account any delays caused by aerodynamics, actuator performance or noise.
In the work of Troldborg, as for his MSc thesis, a 2D computational study is being
performed [74]. The aerodynamic characteristics of the Risø-B1-18 airfoil equipped
with different compliant trailing edge flaps are calculated. The solver used was
Ellipsys2D (in-house RANS code) with a k-ω SST turbulence model for fully turbulent
flow, at Re=1.6 million. The calculations on the baseline airfoil were compared with
experimental data and showed perfect agreement. A comparison with the potential
theory solver developed by Gaunna [29] is also made. The influence of various key
parameters such as flap shape, flap size, and oscillating frequencies was investigated so
that an optimum design is suggested: a moderately curved flap with flap chord to airfoil
curve ratio between 0, 5 and 0, 10. Finally a study was conducted on an oscillating
18
airfoil fitted with an oscillating flap, with the objective to prescribe the motion of the
flap so that the lift remained constant.
Through the project Adapwing, as have been mentioned previously, computational

studies were performed in order to find general design directions for deformable
controllable airfoil geometry. The choice, as can been seen through the computational
studies, was deformable trailing edge flaps. An experimental setup for wind tunnel
tests was prepared. The actuator/control surface chosen was a piezoceramic uniform
bender actuator (Thunder TH-6R). So, no discrete servo or hinge flap was used but
just a piezo-bender device which serves as both a flap and actuator. The flap had 10%
chord length and could operate with deflection between -5 and +5 degrees. Firstly,
the actuator performance was measured [19]. Through static and dynamic tests,
the deflections of the bender were studied at various voltage inputs. An important
result was that the actuator response showed a clear hysteresis loop. Optimal voltage
operation regions were found. A blade section was constructed base on the Risø-B1-
18 Airfoil. A number of Thunder actuators were glued to the trailing edge. Static and
dynamic measurements were performed studying the pitch change and flap change and
their behaviour together at different frequencies of operation. No active controller was
implemented. Unfortunately, there are no official publications available regarding this
experiment yet.
All important results from the computational work in Adapwing have been included
in a series of publications [8] [17] [18] [74].
19
Chapter 3
Aerodynamics and Aeroelasticy
3.1 Fundamental Dimensionless Quantities
Reynolds Number (Re)
In fluid dynamics, the Reynolds number characterises the ratio of inertia to viscous
forces and is defined as
inertiaf orces ρU 2 /L ρU L
Re = = = (3.1)
viscousf orces µU /L2 µ
The terms ρ , U and µ are the density, velocity and dynamic viscosity of the fluid and L is
the characteristic length, usually taken as the chord length for an airfoil. The Reynolds
number is a dimensionless number which determines the flow regime of the system, i.e.
laminar or turbulent. Low Reynolds number indicates a dominance of viscous forces in
the flow and hence the flow is generally considered to be steady, smooth and laminar.
Turbulence is present at an high Reynolds number where the flow is dominated by
inertia forces with the appearance of eddies and vortices. Simulation of models can be
carried out providing that dynamic similarity is met between the model and the real
case, which implies a same Reynolds number.
20
CHAPTER 3. AERODYNAMICS AND AEROELASTICY
Mach Number (M)
For an object in air flow, the Mach number is the ratio of its relative moving velocity to
the speed of sound and is in the form of
U
M= (3.2)
a
where U is the relative speed and a is the environmental sound speed. Analyses based
on Mach number are usually found in researches of high speed fluid flow such as gas
turbine or flying airplane, where M is close to unity and phenomena like choking or
shock wave may take place. The relative air flow speed around a wind turbine blade is
generally very low compared with the sound speed so that Mach number is only used
as a non dimensional form of the flow velocity.
Pressure Coefficient (Cp)
The pressure coefficient is a non dimensional value of local pressure
P − P∞
Cp = 1 2
(3.3)
2
ρU∞
where P is the local pressure, P∞ is the freestream pressure, ρ is the fluid density
and U∞ is the free stream velocity. Pressure coefficient is widely used for analyses
of pressure distribution around airfoils of aircraft wings or wind turbine blades, where
pressure is normalised for easier comparison to assess the computation performance
and to compute other aerodynamic force coefficients.
Drag Coefficient (Cd)
The dimensionless quantity of drag force, which is the opposing force acting on an
object in a fluid flow, is used to measure its resistance in air environment and is defined
as
Drag F orce Fdrag
Cd = = 1 2 (3.4)
Ref erence F orce 2
ρU A
where Fdrag is the drag force, ρ is the fluid density, U is the relative velocity between the
object and the fluid, and A is the reference area which is usually replaced by the chord
length c to give a sectional drag coefficient.
21
The drag force is consisted of both pressure drag and shear drag due to skin friction.
Pressure drag is created by the pressure difference between the airfoil surfaces, caused
by the acceleration of flow on the suction side. Skin friction denotes the shear of
air on blade surfaces along air flow direction, which is commonly approximated with
respective to local Reynolds number in turbulent boundary layer by the ”1/7 Power
1/5
Law” derived by Theodore von Kármán as cf = 0.0576Rex for 5 · 105 < Rex < 107 .
The two components of drag can be calculated using
∑
Pressure Drag = Pi A i n i · k (3.5)
∑ 1
Frictional Drag = Cf,i Ai ρi Ui · k (3.6)
2
where i, P, A, n, ρ, U and Cf are the number, pressure, area, unit normal, density,
relative velocity and skin friction coefficient of the local grid element respectively; k is
the unit vector in the direction of drag, i.e. free stream flow direction. Thus, the front
of a blade will receive high pressure drag and the attached flow surfaces will contribute
most of the skin friction.
Lift Coefficient (Cl)
Similar to drag coefficient, the lift coefficient is also a non-dimensional value and is
defined with the same formula
Lif t F orce Flif t

Cl = = 1 2 (3.7)
Ref erence F orce 2
ρU A
The reference area A for lift coefficient calculation is usually the same as the one used
for drag coefficient , which is the chord projection area. The lift force can be evaluated
as
∑
Lift Force = Pi Ai ni · k (3.8)
where the unit vector k of lift is perpendicular to free stream velocity. Lift coefficient is
generally more important compared with drag coefficient for airfoils, mainly because
the amount of lift induced determines the power output of a wind turbine or the upward
force of an airplane.
22
Pitching Moment Coefficient (Cm)
The pitching moment coefficient is also an important parameter in determining both

the performance and the stability of an airfoil or a wind turbine blade. It is defined as
∑n
pi + pi+1 xi + xi+1 yi + yi+1
Cm = {( )[(xi+1 − xi )( − 0.25) + (yi+1 − yi )( ]} (3.9)
i=1
2 2 2
where x and y are grid coordinates and the moment is usually calculated about 0.25
chord.
3.2 Blade as Airfoil

The general design of a wind turbine blade usually contains two main components: the
outer aerodynamic shell for performance and inner main spar for maximum stiffness
at a relatively low structural weight. An example of wind turbine blade cross section
design is shown below in Figure 2.2.1
Figure 3.2.1: Cross section of a modern wind turbine blade
23
The outer shell of a turbine blade is normally manufactured using low density synthetic
fibre composites, while the inner load carrying D-spar is usually made of carbon-fibre
plastic composites, which has an extremely high strength to weight ratio in order to
ensure both high bending and torsional stiffness [68].
For a general airplane wing airfoil, the leading edge can be recognised as the very
front point facing the incoming flow, and the trailing edge at the very end. The line
connecting the leading and trailing edges is defined as the chord line and the camber
line is just a geometrical term to express the curvature of the airfoil.
Figure 3.2.2: Schematic of an airfoil
Similar to aircraft wings, wind turbine airfoils also possess aerodynamic characteristics
such as lift perpendicular to incoming wind direction and pressure drag along the wind
direction. A sketch of the general characteristics of a generic airfoil is shown above in
Figure 3.2.2. Wind turbine airfoils differ from aircraft wings since they have to rotate
about an axis, making the actual angle of attack dependent on the rotational speed.
The generation of aerodynamic forces is closely related to the pressure field around the
airfoil in airflow. As flow approaches the airfoil, it is directed to form a faster stream
on the upper surface than the lower surface. According to Bernoulli’s equation, faster
flow speed corresponds to lower pressure. Thus the pressure on the upper surface will
be lower than that on the lower surface, leading to an overall upward lifting force.
A drag force is also created by the shear stress of the flow on the surface and pressure
24
difference of the front and back of the blade. Therefore a good designed airfoil normally
has an optimised compromise of these characteristics.
3.3 Boundary Layer Theory
In fluid dynamics, boundary layer, represents a thin region very close to the body
surface in the flow path where the skin friction induced by the surface interferes with
the flow [10].
3.3.1 Concept of Boundary Layer
Outside of the layer, due to negligible effects of the viscous friction, flow is recognised
as inviscid, governing by potential theory where flow speed is more or less constant at
any position. Within the boundary layer, the flow adheres to the surface, getting slowed
down. Right on top of the surface, fluid stays relatively stationary which is referred to
as the no slip condition.
Figure 3.3.1: Profile of a typical boundary layer
The interaction between the fluid and the surface consequently produces a so called
wall friction, or wall shear stress τw [10] expressed as
∂µ
τw = µ |w (3.10)
∂y
25
where µ is the viscosity of the fluid, ∂µ

|
∂y w
states the velocity gradient normal to the
surface.
It is now clear that the fluid closer to the surface will be more affected due to a much
larger velocity gradient as shown in Figure 2.3.1 above , while in the outside inviscid
region the wall shear stress is negligible since the velocity difference is very small. This
viscous wall shear stress within the boundary layer eventually results in the globally
known skin friction drag.
The thickness of the boundary layer can be determined as the distance δ between
the surface and the point where the fluid speed has reached 99% of the free stream
velocity. It depends on the distance that the fluid has travelled from the leading edge
of the surface, and also the viscosity of the fluid where higher viscosity usually leads to
thicker boundary layer. It is worth noting that for very high viscosity, or small Reynolds
number, the thin layer approach of the boundary layer theory is no longer valid.
Figure 3.3.2: Boundary layer on a flat plate
As shown above in Figure 3.3.2, boundary layer can be categorised into three main
sections, namely laminar, transition and turbulent. Similar to flow regimes, the types
of boundary layer also depends on the Reynolds number. However instead, a local
Reynolds number Rex [10] is considered, which is expressed as
ρU x
Rex = (3.11)
µ
where symbols have their usual meanings except that the length scale is taken to be the
distance x advanced from the leading edge.
26
The transition region exists in between the laminar and turbulent boundary layers, it
is usually very short and corresponds to a certain critical local Reynolds number Recrit
x .
Below this value, the boundary layer is laminar where layers of constant velocity flow
are present. Above this value, fluctuation motions can be identified in the turbulent
boundary layer. Since the local Reynolds number increases as the flow advances, the
boundary layer transforms from laminar to turbulent.
3.3.2 Flow separation and Airfoil Stall
One of the most important limiting factor of the performance of an airfoil is stall, which
is closely related to flow separation. As flow separation starts inside the boundary layer,
it is of great importance for this study.
Figure 3.3.3: Flow separation schematics
Flow separation occurs when boundary layer gets detached from the surface: this is
mainly due to the change in pressure field as boundary layer travels along the surface.
At an earlier stage, boundary layer experiences favourable positive pressure gradient
dp
( dx < 0), which exhibits suction effect and keeps the flow attached. When it advances
far enough, the kinetic energy of the flow which is reduced due to frictional interaction
with the surface can no longer hold against the pressure field in front of the flow,
dp
producing an adverse pressure gradient ( dx > 0) [10]. After travelling in this adverse
pressure gradient for a certain distance, the boundary layer may eventually slow down
to zero velocity or even leading to reversed flow as shown in figure 3.3.3.
27
For a typical airfoil at zero incidence, the adverse pressure gradient is usually too small
to separate the boundary layer from the surface. Thus the flow is said to be attached
and the flow pattern is smooth. With a certain angle of attack, the air stream over
the upper (or suction) surface moves faster than that passing underneath (or pressure
surface) as shown below on the left of Figure 3.3.4. According to Bernoulli’s principle,
faster velocity produces lower pressure and vice versa. Therefore as the boundary layer
travels through the suction surface towards trailing edge, it will eventually encounter
a substantial adverse pressure gradient and may then separate from the surface.
Figure 3.3.4: Airfoil pressure distribution (left) and stall of the airfoil (right)
As the angle of attack increases, the suction effect magnifies causing the flow over the
upper surface to be even faster. Thus the pressure will be even lower according to
Bernoulli and more energy will be lost in friction due to higher wall shear stress, both
resulting in a weaker ability of the boundary layer to advance in the adverse pressure
gradient. At a certain point, the flow may stop moving relative to the airfoil or even
move back wards to lower pressure zones. The airfoil is said to be stalled at this point
shown on the right of Figure 3.3.4 [10].
Accompanying stall, eddies are formed due to the reversed flow, and will eventually
contribute to a low pressure wake. Since boundary layer no longer exists after flow
separation, the original structure which provides the lift of the airfoil experiences a
sudden decrease of upward force. This phenomenon of sudden loss of lift is one of
the most important consequences of stall. The other effects include the considerable
increase of pressure drag due to a higher pressure difference on both surfaces, and
the diminish of frictional drag because of the separation of the viscous boundary layer.
28
Therefore, avoiding flow separation will maintain a high level of lift and relatively low
amount of drag. In spite of many merits of avoiding stall, some passive stall controlled
wind turbines still rely on the stall of the turbine blades at a proper time in order to
protect them from potential aerodynamic induced vibration damage. However, the
stall delay effect (or rotational augmentation) which usually happens with rotor-crafts,
could bring problems to operating wind turbines where the rotation is found to have
a delaying effect on flow separation. This may potentially damage the wind turbines
by allowing excessive loads on blades with delayed stall and is well studied in Breton’s
thesis [15] and the paper by Schreck et al [64].
3.4 Flutter
Flutter is a dangerous vibration motion of flexible structures, as a result of the
interactions among aerodynamic, elastic and inertial forces first defined by Arthur
Roderick Collar in 1947. There may exist a point where the structural damping is
no longer sufficient to damp out the increasing vibration motion introduced by a
change of operating condition like a higher flow speed. This could eventually lead to a
catastrophic failure of the structure. There are many types of flutter for blades and the
most common two are introduced below.
3.4.1 Classical Flutter
Classical flutter is caused by coupling of modes, which indicates a flutter motion

initiated with the interference of blade mode shapes. As shown below in Figure 3.4.1, at
the start of a cycle (t=0), the airfoil is displaced with a rotation. This rotation, causing
a change in angle of attack, produces extra force that starts lifting the airfoil, while
the rotation is recovered to zero by its torsional stiffness (t=T/4). Due to the return
of the rotation, the extra lift no longer exists, and the bending stiffness tries to pull
the airfoil back to the neutral position with the airfoil rotating in a nose down manner
owing to inertia (t=T/2). This series of motions is carried out in a mirrored way with
the airfoil plunging down (t=3T/4) instead of rising up, and finally completes a full
cycle (t=T). This aerodynamic coupling of the torsional and bending vibration modes
of an airfoil is at its worst when the rotation mode leads the plunge mode by 90 degrees
[33]. The bending stiffness of the blade and aerodynamic forcing will eventually damp
out the plunge motion, however leaving the rotation undamped. The aerodynamic
29
damping on an airfoil will be explained in depth in the following section. The induced
rotational vibration motion cannot be compensated by normal structural damping and
may diverge leading to structure failure.
Figure 3.4.1: Rotation and plunge motion for an airfoil experiencing flutter
It is worth noting that the classical flutter is a two degree of freedom motion purely
caused by coupling of the first torsional and bending modes in an unfavourable phase,
causing large negative aerodynamic damping which cannot be compensated by the
mechanical structural damping. There is still no actual report of wind turbine flutter
failure, however it happens without the separation of flow from the wind turbine blades
which means wind turbines with high tip speeds and long, flexible blades are especially
susceptible and thus highly likely to happen in the future.
For a wind turbine to experience classical flutter, the air flow around the blade surface
has to be attached, i.e. stall is not reached. Furthermore, the turbine blade must have
high aspect ratio to produce large enough tip speed to ensure high energy air flow. The
bending and torsional modes’ natural frequencies also need to be low enough to be
well coupled to produce a flutter mode. Therefore, flutter is more likely to become a
problem in pitch controlled wind turbines since their blades are regulated to operate
constantly below stall. The risk increases if the wind turbine operates with a higher
tip speed, which is normally limited by noise and safety considerations, but could be
initiated by highly yawed inflow [24].
30
3.4.2 Stall Flutter
Unlike the classical flutter that can happen with steady flow, stall flutter is always found
with airfoils near or above the stall speeds where flow becomes unsteady. At this point,
the airfoils are usually under very high loads and the flow separates during stall. The
fluctuation of separated flow and the high loading force lead to a substantial increase
of the induced pitching moment, which corresponds to a single degree of freedom
torsional mode that may eventually cause flutter [72].
Fortunately, stall flutter can be compensated with a well aerodynamic design and
sufficient structural damping, since it is initiated by the separation of flow and it is
a single Degree of Freedom (DOF) motion. For wind turbine blades, the occurrence
of stall flutter usually comes with abrupt stall characteristics designed airfoil sections
and low structural damping . In contrast with classical flutter, stall flutter is a more
serious problem for stall regulated wind turbines as the rotor blades are either designed
or controlled to operate in stalled conditions to limit the power output, while pitch
regulated turbines generally do not operate within this range [24].
3.4.3 Flutter in Airfoils
Individual Modes
In general, flutter of airfoils can be typed into two main categories: aerodynamic
induced flutter and mechanical structure flutter, which essentially correspond to stall
and classical flutter respectively.
Critical stability, i.e. zero damping, happens at stall when the lift curve gradient
decreases to zero where theoretically the airfoil vibration will not be attenuated once
started. Beyond this point, the lift force induced on the airfoil begins to drop,
corresponding to a negative lift slope. Therefore a negative damping is present at this
post stall stage, where the aerodynamic forcing which is supposed to damp out the
airfoil motion becomes out of phase, causing an amplification of vibration and hence
negative damping effect. Although, as explained in the previous section, the major
possibility of stall flutter comes from the simple torsion mode due to sudden rise of
airfoil pitching moment after stall, the plunge mode is also highly likely to experience
flutter since the fluctuation of lift force is very high.
31
On the other hand, classical flutter is possible to occur without stalling the airfoil
through a mixing of simple structural modes, which in this case are the first plunge
and first twist mode. Now, the two modes are considered individually by consulting
their schematics shown below in Figure 3.4.2
Figure 3.4.2: Schematic of plunge (left) and twist (right) modes of an airfoil
From the energy point of view, a damped system is considered to lose energy to its
surrounding air flow while flutter means the system is gaining energy. Thus damping
can be related to the work done to the airfoil by the flow, which is
∫
Work = force · velocity dt (3.12)
period
where the phase difference between the force and the velocity will determine the sign
of the resulting work and thus the system stability. A positive air flow work means a
gaining of energy of the airfoil and therefore negative damping leading to instability,
and vice versa. The general relationship between the phase difference and the resulting
damping can be expressed as follows:
Figure 3.4.3: Schematic of the phasing between force and velocity
32
A simple schematic of the relationship between damping and force/displacement

phase is constructed above in Figure 3.4.3, based on the work integral. Since
velocity leads displacement by 90°, so if force is totally in phase or out of phase with
displacement, the work will always be an integral of the product of a sine and a cosine
which results in zero damping. As force leads displacement, aerodynamic damping
will become negative and reach a maximum when the lead is 90° (or lags 270°).
Consequently, damping will reach its positive maximum at 90° lagging displacement.
Therefore, by analysing the phasing between force and displacement response of a case,
the stability can be determined.
For the plunge motion, the lift force created can be treated as proportional to the angle
of attack, when the vibration amplitude is small. The additional induced angle of attack
due to the airfoil vertical motion can be expressed as − Uẋ where ẋ is the plunge velocity
and U is the free stream velocity. Thus the lift force can be expressed as
ẋ
Lift Force ∝ − (3.13)
U
Based on this relationship, it can be deduced that the real component of the force is in
phase with the airfoil velocity ẋ but opposite in sign, which creates plunge damping to
the system. The imaginary component is in phase with airfoil displacement x with
opposite sign, thus generating a plunge stiffening effect causing an increase in the
resulting vibration frequency, especially when frequency is low. It also has a potential
to destabilise the twist motion by enlarging the pitching moment, according to few
researches [72].
The twist motion can be analysed similarly: since the change in angle of attack for an
airfoil twisted about its mid chord is just the twist angle θ, it would follow the relation
of
Lift Force ∝ θ (3.14)
Correspondingly, the real component of the force is totally in phase with θ causing a
rotational de-stiffening effect, while the imaginary component is in phase with θ̇, but
opposite in sign providing rotational damping [72].
33
Generally, it is the real component of the force that determines the amount of damping
for a vibrating system. Therefore, the plunge motion is damped out eventually and the
twist motion will be left undamped, which proves the aero-damping theory of classical
flutter introduced in the previous section.
Mixed Modes
However, individual modes are only possible on two dimensional like’ objects with
unity hub tip ratio, meaning a uniform extension in the third dimension. For a common
wind turbine, the rotor blade is usually twisted and tapered, leading to a hub tip ratio
smaller than 1. In this case for the NREL Phase VI wind turbine, the ratio between the
root and the tip chord length is about 2.06. Therefore, the real vibration mode of a
wind turbine blade is always a mix of simple modes.
Despite the individual damping characteristics, the situation becomes very different
when the two modes are mixed together. If we let ϕ̄ stands for mode shape, or the
eigenvector of this case, then the mixed mode shape can be expressed as
ϕ̄m = ϕ̄P + ϕ̄T (3.15)
where ϕ̄m , ϕ̄P and ϕ̄T corresponds to the eigenvectors of the mixed, plunge and twist
modes respectively and α is the twist to plunge ratio controlling the relative amplitude
between plunge and twist motion.
It is known that certain mixing of the two modes would cause the structure to flutter
even though none of the two will experience instability by its own. This can be
understood as the result of the coupling effect where the stable force induced by one
mode motion contributes to the instability of the other mode. Specifically speaking,
the aerodynamic force created by plunge mode denoted as Fp is a positive damping
force contributing to stability, while the twist mode force Fp is negative in sign leading
to instability. Therefore theoretically the blade will flutter if the sum of the two forces
becomes negative, i.e.
|Ft | > |Fp | (3.16)
Based on the above theory, the relation can be further derived as
Fp k
∝ (3.17)
Ft α
34
where Fp and Fp stand for the induced aerodynamic forces by these two modes and k is
the reduced frequency and is in a form of the Strouhal number. The reduced frequency
k is a non dimensionalised frequency term characterising the variation of flow with
time defined as
fL
k= (3.18)
u
where f is the cyclic frequency in Hz, L is the reference length taken as the full chord
in this study and u is free stream velocity.
It is clear that flutter will become a problem once the frequency is reduced to very low.
It is worth noting that this relation only applies to small amplitude motions where a
quasi linear relationship between the aerodynamic force and angle of attack is valid.
Once the airfoil is stalled, classical flutter by mixing of modes no longer exists and stall
flutter will be present.
35
Chapter 4
2-D Validation & Analysis
4.1 Geometric Model of the S809 Airfoil

The airfoil selected for 2D validation is the S809 airfoil designed specifically for HAWT
applications [69]. It has a thickness of 21% chord length which belongs to the NREL
thick airfoil family optimised to compete with the widely used NACA-HAWT airfoil
series with primary objectives of restrained maximum lift, insensitive to leading edge
roughness and low profile drag [69]. A sketch of the S809 airfoil is shown below
in Figure 4.1.1 and the detailed S809 airfoil profile characteristics are included in
APPENDIX A: S809 Airfoil Profile Coordinates.
Figure 4.1.1: Schematic of the NREL S809 airfoil
36
CHAPTER 4. 2-D VALIDATION & ANALYSIS
The airfoil section model tested in Netherlands Delft University of Technology (DUT)
Low Speed Laboratory has a chord length of 600 mm. The experiments conducted in
the low turbulence wind tunnel, which is 1.8 m by 1.25 m, recorded basic, low speed,
two dimensional aerodynamic characteristics of the S809 airfoil [79].
A 457 mm constant chord model of the airfoil was further tested by Aeronautical
and Astronautical Research Laboratories (AARL) of Ohio State University (OSU),
concentrated on the effects of leading edge roughness, which was one of the primary
design objective [39]. The experimental wind tunnel data, which includes pressure
distributions and aerodynamic force coefficients, was documented in the final reports
by Somers et al [69] and Reuss et al [39] for the DUT and OSU tests respectively.
4.2 Geometric Model of the NREL Wind Turbine

The test turbine of the NREL NASA-Ames Unsteady Aerodynamics Experiment was
a two bladed, 10 meters rotor diameter stall-regulated wind turbine with rated power
of 19.8 kW. It has a tower height of 12.2 m and was rotating at 72 rpm synchronous
speed. Both upwind and downwind configurations were tested, as well as various yaw
and blade tip pitch angles. A photo of the actual test turbine inside the wind tunnel is
shown below in Figure 4.2.1.
Figure 4.2.1: UAE phase VI rotor in the NASA Ames wind tunnel
37
The two blades of the turbine were non-linearly twisted and linearly tapered, using
exclusively the aforementioned S809 airfoil developed by Airfoils Incorporated under
contract by NREL, which was optimised for better wind power production [34]. The
blade dimensions are illustrated below in Figure 4.2.2 and the blade chord and twist
distribution along the span are shown in APPENDIX B: NREL Phase VI Turbine Chord
and Twist Distributions.
Figure 4.2.2: Blade platform dimensions
Apart from the specified geometry where the S809 airfoil was used exclusively from
25% span to the tip, where the blades were tapered with the chord length decreasing
from 0.737 m to 0.356 m , the root and hub adapter part shows a transition in cross
section from the airfoil shape to a circle towards the hub at 12% span, and it is shown
in details in APPENDIX C: NREL Phase VI Turbine Blade Root Surface Depiction.
However, the near hub root section and the tip section were not specifically defined
in the technical report [34], therefore a simple interpolation with linear transition
was applied to the root section and the tip was treated as a flat surface, which would
certainly have effects on the accuracy of computational predictions.
With pressure taps along the blade span and various types of sensors mounting on
the turbine body, pressures distribution on the blades and the hub, shaft torque, and
the root flapwise and edgewise bending moments were measured and recorded. At
5 nominated span-wise sections shown as black solid lines in Figure 4.2.3 (i.e. 0.30,
0.47, 0.63, 0.80 and 0.95 r/R), full distribution of pressure around the sections were
38
captured by 22 taps, allowing the corresponding aerodynamic force coefficients to be

calculated and compared with computation results [34].
Figure 4.2.3: Blade surface pressure tap locations
As the NREL Phase VI wind turbine was stall regulated instead of pitch regulated,
its maximum power generation depends entirely on the extreme condition at stall.
Therefore, the accurate prediction of stall initiation on the turbine blades becomes
crucial in controlling the turbine rotation, which remains a rather challenging task for
CFD modelling.
39
4.3 Test Case Configuration
4.3.1 S809 configuration
The model was tested experimentally by DUT at Reynolds number ranging from
1.0×106 to 3.0×106 based on the model chord. Tests started with increasing angle of
attack from 0° until a full separation of flow on the upper surface at about 20° and then
decreased incidence angle to determine hysteresis. Based on the experiment, detailed
computational case configurations were drawn and listed below.
Figure 4.3.1: Velocity and aerodynamic forces conventions
Cases AOA Ux (m/s) Uy (m/s)

1 0.00° 15.00 0.00
2 5.13° 14.94 1.34
3 7.17° 14.88 1.87
4 9.22° 14.81 2.40
5 12.22° 14.66 3.175
6 14.24° 14.54 3.69
7 20.15° 14.08 5.17
Table 4.3.1: Test cases at 15 m/s free stream velocity
40
Various angles of attack were investigated with the maximum being 20.15°. Since one
of the major tasks was to capture the initiation of stall, which was found to be around
9° from NREL experimental data, test cases were concentrated around this angle. The
free stream velocity of M=0.044, i.e. 15 m/s, was chosen to give a Reynolds number of
1.0×106 and was decomposed into the horizontal and vertical components with respect
to the angle of attack.
4.3.2 NREL Phase VI configuration
Numerous data based on different test sequences were captured during the
experiment. Sequence ’S’ is chosen to be the candidate in the following analysis, which
was also used in the blind comparison. This sequence included a series of upwind tests
with 3 blade tip pitch and slow yaw sweep. The inlet wind speed was varied from 5 m/s
to 25 m/s, while the turbine was maintained at a 72 rpm rotational speed. Structurally,
all connection were rigid, cone angle was set to 0 and no probes were attached, allowing
free transition to occur [34].
Case Tip Pitch Angle (°) Yaw Angle (°) Air T (°C) Wind Speed (m/s)
S0700000 3.0 0.0 11.1 7.0
S1000000 3.0 0.0 11.0 10.0
S1300000 3.0 0.0 13.7 13.0
S1500000 3.0 0.0 14.2 15.0
Table 4.3.2: Computational cases and parameters
It is worth noting that standard Spalart-Allmaras turbulence model was used

exclusively for the 3D cases.
41
4.4 CFD Methodology

The computational fluid dynamic package used for this project is ANSYS, which
is a platform of integrated engineering simulation programs capable of complex
multi-physics solutions. The core code solver used is FLUENT, that is an implicit,
unstructured, time accurate flow solver for both inviscid and viscous compressible
flows. Apart from the main solver, the platform also contains programs for
mesh generation, finite element based mode shape interpolation, pre- and post-
processing.
Governing Equations
The point of CFD analysis is discretization. During the simulation the flow motion is
not solved in any position of the domain, but in the center of all the fluid elements of the
computational grid. The basis of the numerical method applied stands in the Navier-
Stokes equations set, which include some pillar principles of physics: the conservation
of mass, momentum and energy. The incompressible fluid approximation has been
employed, as the involved Mach number values are low enough. The conventional
limit is considered at 0.3, and for the highest wind speed simulated in this work a
value around 0.1 has been computed. Therefore, the flow velocity doesn’t introduce
significant changes in air temperature or in its density ρ . This is a common
approach not only for liquids, but even for air flowing at moderate speeds. The
next two equations are the aforementioned conservation equations considering the
incompressibility hypothesis: mass (4.1) and momentum (4.2)
∇ · ⃗u = 0 (4.1)
∂ρ⃗u
+ ρ(⃗u · ∇)⃗u + ∇p = µ∇2⃗u + ρ⃗g + F⃗ (4.2)
∂t
Also, if ρ is constant, energy equations are an independent set, not correlated to

continuity and momentum. As the analysed situation doesn’t treat heat transfer, these
equations are not considered in the model.
42
Since the flow to be solved is characterized by turbulence, the variables experience a

continuous random fluctuation in time. Any of these variables ϕ can be represented by
an average value ϕ plus an oscillating term ϕ’. The ϕ term is a time-averaged value and
if this method is applied to the previous relations, Reynolds averaged Navier-Stokes
(RANS) equations are obtained.
The solutions are average values in the time interval ∆t, which is big enough
compared to molecular scales causing the momentum transfer, but much smaller
than the characteristic scales of the analysed problem. This procedure of continuum
approximation strongly reduces the computational effort when the detailed solution
of turbulence structures is not of first importance, and when the mean flow has a
dominant effect on the anisotropy of turbulence. Moreover, the direct calculations
of turbulent fluctuation for smaller scales, like Large Eddy Simulations (LES) and
Direct Numerical Simulations (DNS), is still prohibitive for most of the practical
applications.
Turbulence
The turbulence model choice is a fundamental step that significantly influences

numerical results, because they are employed to obtain an estimation of the Reynolds
stresses and close the RANS system of equations. They can be roughly divided in
one-equation or two-equations models: in one equation model, like Spalart-Allmaras,
the differential transport equation for turbulent kinetic energy k is solved. In two-
equations models, like k −ϵ or k −ω, in addition to the k equation another one is solved,
determining either the rate of k dissipation, ϵ, or the specific dissipation ω = k/ϵ.
Another widespread turbulence models employed in wind turbine simulations is the

k − ω SST model. It is a combination between the two previously mentioned models: a
blending function ensures the gradual variation from k − ω to k − ϵ moving away from
the wall, such that advantages are taken from both. In this project, three turbulence
models will be investigated, in order to understand which one is the best for the
considered case.
43
Boundary Conditions
The computational mesh was imported in Fluent pre-processor. Air was considered as
an incompressible fluid, as already stated, and its density was set to the experimental
value for each wind speed, as reported in Table 4.4.1. No slip wall condition was set for
the blade, velocity inlet condition was applied for the front section, with wind direction
normal to the surface. Turbulent intensity was set to 5%, which is Fluent standard
value, even if a lower value could be suggested to simulate the wind tunnel controlled
inflow conditions.
Zero gauge pressure was set to the rear section of the control domain with the pressure
outlet condition, and the same turbulence parameters of the inlet section were applied.
Discretization for pressure, momentum, k and ω was imposed to second order. A
maximum allowable Courant number of 40 was chosen, accordingly to Fluent advice
and explicit relaxation factors of 0.5 and 10−6 residuals were used. A lift and drag
monitor were set to check the convergence trends during the simulation and then
10000 iterations were launched.
Table 4.4.1: Wind speeds, densities and pressures
Speed [m/s] Density [Kg/m3] Pressure [P a]

7.016 1.246 101959.328
10.047 1.246 101894.813
13.069 1.227 101410.336
15.098 1.224 101365.945
20.131 1.221 101204.07
25.109 1.22 101025.648
44
4.5 Results and Analysis
4.5.1 2-D Grid of the S809 Airfoil
For grid generation, the integrated mesh generator in ANSYS was used, due to its
extensive mesh functions and ease of use. The S809 airfoil has a chord of 600 mm
and 21% thickness and it is to be tested with a Reynolds number of 1.0×106 . The chord
length of the airfoil geometrical model was set to 1 m in ANSYS, which resulted a free
stream air flow of about 15 m/s due to dynamic similarity. To avoid possible effects of
far field boundary on the flow around the airfoil, the grid domain was set to be 30 chord
in horizontal and 30 chord in vertical direction away from the blade, as it is possible to
see in figure 4.5.1.
Figure 4.5.1: Computational domain and far-field boundaries
A structured quadrilateral mesh was applied to the geometrical model, with a total
number of 376500 elements along with a set of structured quadrilateral boundary
layers (inflation layers). The boundary layer was attached on the airfoil surface with
15 layers and a relatively slow growth rate of 1.1, starting with a row depth 0.0002
m. Size function was defined for refinement around the leading and trailing edge to
capture the stagnation point and wake flow respectively. This grid density was tested to
produce good spatial and temporal resolution for simple aerodynamic and aeroelastic
computations.
45
Figure 4.5.2: Detailed mesh around the airfoil
Figure 4.5.3: Detailed mesh around the leading edge
Structured grids have the possibility to reduce the computational time respect to
unstructured grids due to the explicit storage of inter cell connectivity, maintaining an
high quality to achieve the desired results. On the other hand, it is very complicated to
generate a structured mesh for complex body construction, such as a complete aircraft.
In this specific case, due to the relative simplicity of the considered geometry, it was
decided to use a structured mesh for the advantages described before.
46
4.5.2 Aerodynamic Results - Steady
Aerodynamic Coefficients
The results for aerodynamic force coefficients, Lift and Drag coefficients, were
obtained and shown below in Table 4.5.1. Three different turbulence models were
tested in this calculation: Spalart - Allmaras, k − ϵ and SST k − ω turbulence model.
Table 4.5.1: Aerodynamic force coefficients prediction
Cl Cd
AOA (°)
S-A k−ω k−ϵ Exp S-A k−ω k−ϵ Exp
0.00 0.112 0.096 0.111 0.139 0.0131 0.0141 0.0147 0.0094
5.13 0.665 0.635 0.6632 0.736 0.0163 0.0171 0.0177 0.0097
7.17 0.862 0.823 0.863 0.913 0.0196 0.0205 0.021 0.0127
8.20 0.952 0.906 0.9545 0.952 0.022 0.023 0.0233 0.0169
9.22 1.031 0.976 1.0375 0.973 0.0249 0.0265 0.0262 0.0247
10.21 1.0935 1.0198 1.108 0.952 0.0289 0.0318 0.0297 0.0375
11.21 1.1326 1.032 1.161 0.947 0.035 0.0402 0.0344 0.0725
12.22 1.1379 1.033 1.2 1.007 0.04435 0.0501 0.0403 0.0636
14.24 1.141 1.055 1.221 1.055 0.0662 0.0697 0.0562 0.0828
20.15 0.86775 1 0.95 0.923 0.197 0.20 0.22 0.1853
It is worth mentioning that both steady and unsteady calculations were performed.
For high angles of attack (post stall), no convergence could be achieved in steady
computation due to a substantial amount of flow separation, thus unsteady time
accurate computations were necessary and the results are shown as circled data points
in the following plots. Since the k − ϵ turbulence model has generated coefficients
predictions far from the experimental data, with an average difference greater than
10%, a comparison between the other two more promising methods was performed.
According to the plots of lift and drag coefficients shown in Figure 4.5.4 and Figure
4.5.5, it is evident that the k − ω turbulence model predicts the flow condition
much more accurately than the Spalart-Allmaras one, thanks to its accuracy for
adverse pressure gradient prediction in simulating near stall flow. Nevertheless, the
standard Spalart-Allmaras model actually predicts the trends rather well, despite being
inaccurate about the exact values at certain conditions.
47
Figure 4.5.4: Lift coefficient vs. angle of attack
It can be seen from the lift coefficient plot shown in Figure 4.5.4 that the lift predicted
by the SST k − ω model is very similar to the DUT experimental measurements.
Unlike the experimental results or CFD predictions performed by other authors [79],
no apparent point of stall was detected which is supposed to be around 9°. However,
the lift curve slope does show a significant decrease between 10.21° and 14.24°, which
indicates possible flow separation and the occurrence of stall, even though strictly
speaking a decrease of lift coefficient is not detected. This is further proved by
inspecting the pressure coefficient and streamline plots at these angles of attack later.
Moreover, it is clear that the lift force is under predicted at all incidents compared with
DUT measurements until stall takes place (around AOA=9°). The unclear detection of
stall point and the over prediction of lift after stall both indicate a lack of turbulence
level compared with realistic flow, showing that the separation process takes place a lot
slower. The calculated high lift coefficient at 20.15° is on the other hand due to the large
amount of unsteady fluctuation, since at this incidence the flow is fully separated from
the upper suction surface. Thus, only a cycle averaged value is chosen to be shown on
the plot and is very hard to compare with the experimental data since the experiment
itself experienced unsteadiness as well at this flow incidence. The general trend of lift
prediction by the two codes are very good comparing to some popular previous studies
by Wolfe and Ochs [79] and Guerri et al [32]: substantial under- and over-predictions
for the same SST k − ω turbulence model were present.
48
Figure 4.5.5: Drag coefficient vs. angle of attack
In spite of the slightly difference in predicting the lift coefficient by the two models, the
drag force is similarly predicted with good agreement with experimental data shown in
Figure 4.5.5. It is important to know that only the pressure drag is accounted for in this
plot, meaning that the viscous shear drag is ignored since it is considered to be orders
of magnitude smaller. The SST k − ω model experiences an over-prediction until the
airfoil reaches stall condition, while after stall an under-prediction is experienced, but
always following the parabolic shape of the experimental data. The peak at 11.21° was
not captured in the computation. However, by considering that none of the previous
authors [32] [79] had successfully predicted the drag peak, it is fair to assume that the
peak could be experimentally induced. The slight over-prediction at 20.15° is probably
due to the same reason as the over-prediction of lift at the same point.
Predictions of the Spalart-Allmaras model show very similar results with the k − ω
one, except for slightly smaller post stall values. This is sensible as the flow around the
airfoil in SST k − ω computation experiences a higher turbulence level on the suction
surface after flow separation, leading to a slightly higher drag force.
49
Pressure distributions
Pressure coefficients around the airfoil were calculated to further investigate the solver
performance. Plots of results using the SST k − ω turbulence model at angles of attack
of 0°, 5.13°, 9.22°, 14.24° and 20.15° are presented below in Figure 4.5.6. The range
of the angle of attack in the plots completes a full coverage from zero incidence to fully
stalled condition (20.15°), thus should give a general idea of how the code is able to
cope with different flow regimes.
50
Figure 4.5.6: Pressure distribution around the airfoil at various angles of attack
At zero incidence, the overall prediction shows good agreement with the experimental
data, except for mid chord region on the upper surface and the trailing edge region.
In the experiment conducted by DUT, laminar separation bubbles were observed on
both the pressure and suction surfaces just after the mid chord, causing a nearly flat
distribution of pressure with respect to x/c [69], which is not captured by the fully
turbulent simulation. The following sudden increase in pressure due to turbulent re-
attachment also can not be identified on the plot. The computation error at the trailing
edge on the other hand is possibly due to the fact that the thickening of boundary layer
as flow travels along airfoil surface is not well captured. This might be the result of a
slightly thin boundary layer construction during grid generation phase. Moreover, the
situation is likely to deteriorate when the solver is over allowing flow separation due to
the numerical setup, which shows a low pressure distribution on the pressure surface
at the last 40% of the chord. This phenomenon can be identified on higher angles of
attack as well.
For AOA = 5.13°, the general trend of prediction is similar to that of the 0° case.
Laminar separation bubbles are not captured, and the rear of the airfoil experiences
more separated like flow which is probably due to the same reason as the zero incidence
case. The pressure distribution on the forward half of the lower surface is constantly
under predicted, but only by a very little amount. However for the forward half of the
51
upper surface, the suction peak is never reached and the pressure distribution is not
well captured. A similar discrepancy of the prediction can be found in many works done
on the S809 airfoil such as Wolfe and Ochs [79], Guerri et al [32] and Chang et al [22]
although smaller inconsistencies were generally achieved. Wolfe and Ochs [79] further
proved the problem for 5.13° angle of attack to be the incapable simulation of laminar
to turbulent transition. Therefore, the reason for the major discrepancy between the
computation results with the SST k − ω turbulence model and the experimental data
is considered to be the combined effect of a lack of leading edge refinement, which
contributes to the failure in capturing the suction peak close to the nose and a lack of
transition model, which causes the following over prediction.
At 9.22° angle of attack and afterwards when the angle of attack was increased to
14.24°, the experimental airfoil entered the stall phase with very sharp suction peak
(extremely low pressure) due to an exceedingly fast flow around the nose region. The
pressure coefficient prediction turns out to be exceptionally good at these incidences,
with identical data on the pressure side and on the suction side. The suction peak and
the forward half of the suction surface pressure distribution are not perfectly captured,
due to similar reasons for the previous cases. Nonetheless, the lift and drag coefficient
are well calculated with only 1% and -8% errors respectively, as it is shown in table
4.5.2. Stall is detected while many other authors failed to achieve [79].
Table 4.5.2: Calculation errors of aerodynamic coefficients
Cl % error Cd % error
AOA (°)
S-A k-omega k-epsilon S-A k-omega k-epsilon
0.00 -19% -21% -20% 29% 20% 36%
5.13 -10% -14% -10% 28% 26% 42%
7.17 -6% -10% -5% 34% 21% 45%
8.20 0% -5% 0% 30% 26% 38%
9.22 6% 0% 7% 1% 7% 6%
10.21 15% 7% 16% -23% -15% -21%
11.21 20% 9% 23% -22% -15% -33%
12.22 13% 3% 19% -20% -11% -27%
14.24 8% 1% 16% -13% -8% -22%
20.15 -6% 8% 3% 6% 8% 19%
52
At AOA=20.15°, the flow essentially detaches from the entire suction surface as the
transition shifts to the point just after the leading edge. As can be seen on the pressure
coefficient plot, the over prediction of pressures on the upper suction surface just
behind the leading edge is due to a deeper stall of the airfoil, since it experiences stall
earlier than the experimental case. Nevertheless, the data of other regions is in good
agreement with experimental measurements, which is better than a lot of other authors
calculations with various turbulence models [79] [32]. It is worth noting that the SST
k −ω turbulence model used by Guerri [32] and Le Pape and Lecanu [54] also provided
very good results for nearly all angles of attack. This is mainly because of its speciality
in predicting adverse pressure gradient, leading to a better simulation of separated flow
on the upper surface.
Velocity contour and streamline plots
In order to investigate the behaviour of flow at certain incidences compared with the
experimental case, velocity contour and streamline plots are presented with angles
of attack 0°, 5.13°, 7.17°, 8.20°, 9.22°, 10.21°, 12.22°, 14.24° and 20.15° for the SST
k − ω turbulence model. According to the contour plots in figure 4.5.8, the growth of
flow separation with increasing angle of attack is clearly identified. It can be seen that
starting from 7.17°, or even 5.13°, the flow over the upper surface shows a tendency to
separate with the boundary layer significantly thickens. To prove this, the streamline
plots are constructed below in Figure 4.5.7.
Figure 4.5.7: Streamline plots of 5.13° (left) and 8.20° (right) angles of attack
53
Figure 4.5.8: Steady contours of absolute velocity at AOA from 0° to 20.15°
The left plot of Figure 4.5.7 shows a clear separation very close to the rear end at about
98% chord, which then reattaches to the surface just before leaving the trailing edge.
This is a premature separation prediction based on the experiment results, and it is
believed to be one of the reasons causing the inaccurate prediction of the pressure
distribution near the trailing edge in Figure 4.5.6. With this small separation initiating
at 5.13°, the flow becomes more and more detached as the angle of attack increases. It
can be seen that at angle of 8.20°, which is plotted on the right of Figure 4.5.7, there is
the guessed point of stall initiation in previous sections, the rear half of the airfoil upper
surface experiences separated flow. The situation here is far more developed than
that measured in the experiment, where a small separation region was detected only
54
until 9.22°. Considering the sudden decrease of lift slope in the lift coefficient against
angle of attack plot (Figure 4.5.4), it is evident that at around 8.20° the separated flow
occupying 50% of the chord as seen above, has become so dominant that a further
increase of angle of attack will slow down the lift growth.
Figure 4.5.9: Flow separation at 14.24° (left) and 20.15° (right) angles of attack
By investigating the streamline plots at even higher angles of attack shown in Figure
4.5.9, the rapid growth of separation region can be identified. At AOA=14.24°, two
large vortices are present and the flow separates the upper surface at about x/c=0.4,
where in experiment it occurred at x/c=0.5. The close prediction of the separation
point at this incidence is one of the main reasons leading to accurate pressure
distribution prediction in Figure 4.5.6. Since the flow becomes very unsteady at a high
angle of attack of 20.15°, only a stationary snapshot is chosen to be shown on the right
of Figure 4.5.9. In this case, a full separation from the upper surface is seen with two
huge vortices, causing the predicted lift to drop substantially. It is also interesting
to notice that the flow arriving at the stagnation point have to travel backwards to
pass around the leading edge which generates very high speed streams. Based on the
streamline plot, it is clear why the pressure coefficient plot at 20.15° in Figure 4.5.6
matches very well with experimental data, as the experimental case also had separated
flow for the entire upper surface. However, many other authors only got half chord
separation from their codes’ simulations due to delayed stall.
55
4.5.3 Aerodynamic Results - Unsteady
Regarding the unsteady behavior of the flow, the case of the airfoil with an angle of
attack of 20.15° is considered. In this configuration, full stall is present, as explained
in the previous sections. After a short period of transition, where an increasing flow
separation from the suction surface is present, the flow starts oscillating with a periodic
behaviour, bringing to the consequence formation of eddies in the airfoil wake (Figure
4.5.10).
Figure 4.5.10: Eddies formation for airfoil at full stall condition
From this Figure, it is possible to clearly see the Kármán1 vortex street, which is a
repeating pattern of swirling vortices, caused by a process known as vortex shedding,
which is responsible for the unsteady separation of fluid flows around bodies. As
a consequence of the formation of a vortex, there is a modification of the pressure
distribution around the body. Consequently, an alternating formation of vortices
generates periodically variable forces and therefore a vibration of the body [80]. If the
frequency of formation of the vortices approaches the natural frequency of vibration
of the body, the phenomenon of resonance can occur, which can lead to flutter, as
explained in Chapter 3, vibrating with harmonic oscillations driven by the energy of
the flow with possible failures of the structure.
1
From T. Von Kármán, the Hungarian engineer and physicist who discovered this phenomenon
56
This is the mainly reason because it’s crucial to analyze the frequency of formation of
the vortices and, for this scope, it has been studied the unsteady behaviour of the lift
during the transient simulation, which is shown in Figure 4.5.11.
Figure 4.5.11: Unsteady lift curve at 20.15° (Full stall)
It is possible to see the oscillating behaviour due to the formation of the vortices, after
the initial transition period of few seconds. This function was analyzed during the
oscillation period with a Fast Fourier Transform (FFT), thus the signal was converted
in MATLAB through a specific code from its original time domain to a representation
in the frequency domain, as shown in figure 4.5.12.
Figure 4.5.12: Lift spectrum in frequency domain
57
From this plot it can be seen that the first and principal harmonic of the oscillating
signal has a frequency f = 7.714 Hz ≈ 48.5 rad/s. From this value, it’s possible to
calculate a fundamental parameter for this aeroelastic analysis: The Strouhal2 number,
which is a dimensionless number describing oscillating flow mechanisms.
The Strouhal number is given as
fL L/u time for the flow to travel the entire airfoil

St = = = (4.3)
u 1/f time for airfoil to complete a vibration cycle
where f is the frequency of vortex shedding, L is the characteristic length of the airfoil
and u is the flow velocity.
Normally, a Strouhal number of 1 indicates by the time a disturbance in the flow

passes downstream the chord length, the airfoil has just finished its oscillation cycle.
According to Von Kármán [77], this also means that the characteristic length L of the
airfoil is equal to the wavelength of the disturbance.
It is always quite dangerous for operations with reduced frequency close to unity, which
is introduced by Fung [28] that some experimental and empirical results showed a
tendency of flutter around St=1 for large aircraft wings. In this particular case, the
characteristic length of the airfoil is L = 1 m and the flow velocity is u = 15 m/s, which
lead to a Strouhal number St = 0.514, which implies very low possibility of flutter for
this airfoil geometry and specific boundary conditions.
4.5.4 Aerodynamic Results - Trailing Edge Flap
Due to the proven results in aircraft applications and described as the most promising
device for loads control in large wind turbines in many publications, it was decided to
analyze the effect on applying a Trailing Edge Flap to the 2D S809 airfoil to compare
the results with the standard configuration. Same CFD methodology and boundary
conditions have been applied to the computational model, while the only different from
the previous analysis is indeed in the geometrical model of the airfoil, equipped with a
movable trailing edge flap.
2
Named in honor of the Czech physicist Vincent Strouhal
58
The trailing edge flap, with a chord length of 20%, was designed using the same CAD
software mentioned for the previous analysis and it is presented in Figure 4.5.13, with
different flap angle configurations. The length of the trailing edge flap was chosen
thanks to the promising results described in important publications like Tsiantas et al.
[75] and Lutz et al. [46].
Figure 4.5.13: Cross sections for different flap configurations: +10°, -10°, 0°
The results for aerodynamic Lift force and Drag force were obtained and shown in
Table 4.5.3. For this calculations the SST k − ω turbulence model was used. It is worth
mentioning that both steady and unsteady calculations were performed.
For high angles of attack (post stall), no convergence could be achieved in steady
computation due to a substantial amount of flow separation, thus unsteady time
accurate computations were necessary and the results are shown as circled data points
in the following plots.
59
Table 4.5.3: Aerodynamic force prediction
Lift Force Drag Force

AOA (°)
Flap 10° Flap 0° Flap -10° Flap 10° Flap 0° Flap -10°
0 64.45 13 -70.12 2.91 1.94 4.98
5.13 123.84 83.78 5.14 4.39 2.43 4.68
7.17 138.45 106.93 34.06 6.03 3.03 5.35
9.22 148.25 124.28 60.23 8.25 4.1 6.58
10.21 153.72 130.23 71.85 9.63 5 7.44
12.22 153.97 146.53 91.93 13.41 6.88 9.58
14.24 155.14 151.59 114.11 17.77 9.62 12.7
20.15 139.01 125.22 95.91 52.73 39.7 30.11
It can be noted, as shown in Figure 4.5.14, that the influence of the flap is significant.
The deflection of the flap to the pressure surface (+10°) generates an increase in the
aerodynamic loads, with a resulting increase of lift; while a deflection to the suction
surface (-10°) decreases the aerodynamic loads generating a lift decrease. This occurs
because, thanks to a movable flap, it’s possible to modify the useful area of the pressure
or suction side, with a consequently increase or decrease or lift.
Figure 4.5.14: Lift force for different flap configurations: +10°, 0°, -10°
60
It’s interesting to note that this difference is very high for small angles of attack, while
it reduces as the AOA increases, reaching a point around stall where the difference is
almost negligible. Again, this is due to the large amount of unsteady fluctuation in the
flow, since at this incidence the flow is fully separated from the suction surface.
Two more observations can be done regarding the lift curve for the flap in upstream
configuration (grey curve): at incidence of 0°, the lift force is negative, thus producing a
rotation in the opposite direction of the desired one. For this reason, this configuration
is mainly used for higher angles of attack, when the freestream velocity reaches high
speed and the main purpose of the flap is to have a beneficial impact on blade loading,
as it will be further explored in Chapter 6. It can be also noted that in this particular
configuration, the stall condition starts some degrees of incidence after the standard
and +10° configuration, so it can be said that some degrees of incidences are actually
gained in this configuration. This is due to the particular geometry of this case, which
delays the separation point in the upper surface of the airfoil, resulting in a delay of
stall condition.
Figure 4.5.15: Drag force for different flap configurations: +10°, 0°, -10°
Regarding the drag force prediction, it’s possible to see in Figure 4.5.15 that the
behaviour of the drag function for the different configurations is the opposite of the
lift function. The different between the configurations is very low for small AOA,
while it increases with the increase of the angle of attack. This could be caused by the
formation of Form Drag, which is due to the separation of the boundary layers from the
61
airfoil surface and the consequent lack of pressure recovery. This drag component can
be larger than the frictional one when massive separation is involved and therefore
it is important to reduce it as much as possible by delaying the separation. In this
particular case, the separation is much higher for the configuration with the flap moved
downwards (+10°), due to the particular geometrical model of this case.
Table 4.5.4: Lift force for different wind speeds
Lift Force
Wind Speed (m/s)
Flap 10° Flap 0° Flap -10°
3 5.94 3.23 1.36
5 16.61 7.52 3.81
8 42.6 29.34 9.79
10 67.07 45.99 15.29
13 114.4 78.38 25.84
15 153.3 104.24 34.37
18 221.95 152.28 49.42
20 274.66 188.66 60.96
Figure 4.5.16: Lift force for different wind speeds and flap configurations
62
Lastly, the lift force for different flap configurations at different wind speeds can be
seen in Figure 4.5.16. As expected, the lift increases proportionally with the increase in
wind speed velocity, mostly for the configuration with flap moved to the pressure side.
At the same time, it’s very interesting to see that the lift curve for the -10° configuration
is very flattened, with a relatively small increase of lift considering the high increase in
wind speed. This observation will be an important insight for a further analysis which
will be developed in Chapter 6.
63
Chapter 5
3-D Fluid Structure Interaction
5.1 CFD Case Configuration

For this analysis, a wind turbine blade (Figure 5.1.1) was modelled following the
specification of the geometric model of the NREL wind turbine, as shown in details
in APPENDIX C. The geometric CAD model of the blade was constructed using a
commercial CAD software and then imported into ANSYS. The twisted and tapered
blade body was constructed according to APPENDIX C, but since there is no clear
record of the exact dimension of the blade root and tip sections, simple geometry was
adopted. The blade tip was set to be a flat surface for simplicity and the root connection
section was constructed using linear interpolation.
Figure 5.1.1: CAD model of the wind turbine blade
64
CHAPTER 5. 3-D FLUID STRUCTURE INTERACTION
The blade is made out of an orthotropic composite material, it has a varying thickness
and it also has a spar inside the blade for structural rigidity. The turbulent wind flows
towards the negative z-direction at 15 m/s, which is a typical rated wind speed for a
turbine of this size, as shown in figure 5.1.2. This incoming flow is assumed to make
the blade rotate at an angular velocity of -3.22 rad/s about the z-axis (the blade is thus
spinning clockwise when looking at it from the front, like most real wind turbines).
The tip speed ratio (the ratio of the blade tip velocity to the incoming wind velocity) is
therefore equal to 5, which is a reasonable value for a wind turbine of this size.
Figure 5.1.2: Front view of the wind turbine blade
5.2 CFD Methodology

The computational fluid dynamic and finite element method package used for
this analysis is again ANSYS, thanks to its integrated platform of engineering
simulation programs both for aerodynamic and structural analysis. In this case,
the two solvers used are FLUENT regarding the CFD analysis in order to obtain
the aerodynamic loading, and Mechanical APDL for the FEA structural analysis to
determine stresses and deformations from the previous results. Apart from the main
solver, also integrated programs for mesh generation, finte element based mode shape
interpolation, pre- and post- processing were used.
Governing Equations
For this analysis, the governing equations are divided considering the two different
simulations which will be performed. In the CFD analysis, for the case of a 3D
rotating blade, the governing equations are the continuity equation or conservation
65
of mass (5.1) and Navier-Stokes equations or conservation of momentum (5.2). These

equations are written in a frame of reference rotating with the blade. This has the
advantage of making the simulation not require a moving mesh to account for the
rotation of the blade.
∂ρ
+ ∇ · ρ⃗vr = 0 (5.1)
∂t
∇ · (ρ⃗vr⃗vr ) + ρ(2⃗ω × ⃗vr + ω

⃗ ×ω
⃗ × ⃗r) = −∇p + ∇τr (5.2)
Where ⃗vr is the relative velocity (the velocity viewed from the moving frame) and ω
⃗
is the angular velocity. Interesting to note the additional terms for the Coriolis force
(2⃗ω × ⃗vr ) and the centripetal acceleration (⃗ω × ω
⃗ × ⃗r) in the Navier-Stokes equations.
Also in this case, the flow to be solved is characterized by turbulence and the variables
experience a continuous random fluctuation in time. For this reason, again the method
of Reynolds averaged Navier-Stokes (RANS) equations is used to solve the problem and
the k − ω SST turbulence model was selected to close the set of equations.
Regarding the FEA analysis, the governing equations are based on shell theory, which
are an extension of the Euler-Bernoulli beam theory and plate theory. Shell theory
takes the idea form the two previous mentioned theory and extend it to 3-dimensional
and curved surface. In shell theory, the main focus is on the mid-surface of the
structure and how that mid-surface is going to deform under the influence of a load,
assuming that normals will remain normal at every point on the mid-surface. Without
exploring in deep the mathematical details of shell theory, which are really complex
from an analytical point of view, the main idea is that the solver finds the displacements
and rotations at selected point in the mid-surface of the structure where the potential
energy is minimized (point of equilibrium), after the influence of the specific load. In
order to find the displacements anywhere in the structure, an interpolation process
between the points is applied and the stresses and deformations are calculated through
the potential energy of the mid-surface, knowing the materials characteristics and
surface geometries.
66
Boundary Conditions
For the aerodynamic simulation, the computational mesh was imported in Fluent
pre-processor. Air density was set to the experimental value for each wind speed,
as reported in Chapter 4, Table 4.4.1. No slip wall condition was set for the blade,
velocity inlet condition was applied for the front section, with wind direction normal
to the surface. Turbulent intensity was set to 5% and turbulent viscosity ratio equal
to 10. Zero gauge pressure was set to the rear section of the control domain with
the pressure outlet condition, and the same turbulence parameters of the inlet section
were applied. Discretization for pressure, momentum, k and ω was imposed to second
order. A maximum allowable Courant number of 40 was chosen and an high explicit
relaxation factors and 10−6 residuals were used. The simulation is performed in the
rotating frame of reference (frame motion) and the periodic side boundary was applied
to the model: only 1/3 of the full domain is modelled (one blade) using periodicity
assumptions (⃗v (r1 , θ) = ⃗v (r1 , θ) − 120n). Pseudo-transient simulation was applied and
then 1500 iterations were launched.
Regarding the mechanical simulation, all the material properties reported in Table
5.3.1 are inserted and the pressure loads from the aerodynamic computation are
uploaded. A remote rigid point was hold at the center of rotation, in order to simulate
the effect of having the blade connected to hub to calculate forces and moment
reactions. Stress stiffening effect was applied to the simulation: the faster the blade
is spinning, the stiffer the response is going to be. This is very important because
natural frequencies of an unstiffened and stiffened blade are very different. After this
preliminary set-up, the simulation was launched.
67
5.3.1 3-D Grid of the Blade
For grid generation, again the integrated mesh generator in ANSYS was used, due to its
extensive mesh functions and ease of use. The length of the blade geometrical model
was set to 21.5 m in ANSYS, which resulted a free stream air flow of about 15 m/s due
to dynamic similarity. In Figure 5.3.1, it is possible to see the geometry of the fluid
surrounding the wind turbine blade and the outer boundaries of the fluid volume. In
the middle, the blade is represented as a void, so an empty space inside the volume.
To do this, the blade geometry was imported, translated and finally oriented, in order
to face the incoming wind: indeed, the blade lays along the negative x-direction while
the wind flows towards the negative z-direction.
Figure 5.3.1: Computational domain
An unstructured tetrahedral mesh was applied to the geometrical model, with a mesh
refinement around the blade in order to obtain more accurate results around this
region, for a total number of 369312 elements. The boundary layer was attached on
the airfoil surface with 10 layers, with a relatively slow growth rate of 1.2 and an high
density around the leading edge and trailing edge. There is no deliberate cluster of
elements at the tips or their trails since tip vortices are not the investigation objectives
in this study.
68
Figure 5.3.2: Meshed computational domain
Figure 5.3.3: Detailed mesh around the blade
To avoid possible effects of far field boundary on the flow around the blade, the grid
domain was set to be 25 blade lengths in horizontal and 15 blade lengths in vertical
direction away from the blade.
Regarding the mesh of the blade for the structural analysis, a mapped face mashing
was used for creating about 5000 quadrilateral elements, which are suitable for this
kind of model, as it can be seen in Figure 5.3.4 and Figure 5.3.5.
69
Figure 5.3.4: Detailed mesh of the blade - 1
Figure 5.3.5: Detailed mesh of the blade - 2
70
5.3.2 Aerodynamic Results
In this section it’s possible to analyze the results from the first part of this analysis,
where the aerodynamics loading on the blade are detected. In the second part,
the pressures on certain areas of the blade are passed as pressure loads to ANSYS
Mechanical to determine stresses and deformations on the blade.
Velocity Streamlines
The first result which is shown is the velocity streamlines plot. In this plot, it’s very
interesting to see the clear drop in velocity behind the turbine, which is definitely the
correct behaviour, as it’s showing the wake behind the wind turbine. Moreover, it can
be seen a clear acceleration of the flow around the wake, which is expected from mass
and momentum balance. It can be proved that this behaviour matches quite well the
actuator disk theory for wind turbines.
Figure 5.3.6: Velocity streamlines around the turbine
71
Blade Velocity
The second result which is shown is the blade velocity in standard frame, which is the
frame of reference from the ground. As expected, the local blade velocity increases with
radius and the velocity at the tip is the highest velocity. In this case, as it can be seen
well in figure 5.3.7, the value of the velocity at the tip is about 98 m/s.
Figure 5.3.7: Blade velocity in standard frame for a 3-Blade rotor
Pressure Contours
Finally, the pressure contours on the surface of the blade are presented. The main
understating from these plots is that the pressure is lower on the back surface of the
blade, compared to the front surface of the blade. From the color scheme it can be seen
that the red regions, which show positive pressures, have about the same magnitude as
the green regions that show negative pressures. The blue regions, however, are much
higher in magnitude and therefore have more weight towards the negative pressures.
So, this pressure difference between the front and the back surface creates a lift force
and it points it normal to the back surface of the blade. Because the blade is not fully
perpendicular to the ground, there is a component of the lift in the direction of rotation,
72
which is the XY plane. The component of the lift in the negative Z direction has a large
effect on blade deflection, which will be analyzed in the next section of this analysis.
Figure 5.3.8: Pressure contour on the blade front surface
Figure 5.3.9: Pressure contour on the blade back surface
73
In the next Figure 5.3.10, it’s possible to see the pressure contours in the Y-Z plane
cutting through the blade. With this view, it can be better seen the pressure distribution
around the airfoil: there is an high pressure area (red area) near the leading edge on
the pressure surface, while there is a low pressure zone at the mid-span on the suction
surface (blue zone). From this pressure difference between the pressure and suction
surfaces, as seen before also in the 3D plot, a lift force is generated through the direction
of rotation, so the X-Y plane.
Figure 5.3.10: Pressure contour on the YZ plane
74
5.3.3 Aeroelastic Results
In this section it’s possible to analyze the results from the second part of this analysis, in
which the solid mechanics aspects of this wind turbine blade are involved. The pressure
load found in the first part are imported in ANSYS Mechanical and the stresses and
deformations on the blade are subsequently determined.
Wind turbine blades are now made of composite materials to reduce the weight of
these massive machines. In this case, the structural analysis is simplified by assuming
that the composite material can be approximated by the following orthotropic material
properties (Table 5.3.1).
Table 5.3.1: Blade material properties
Density (Kg/m3) 1550

Young’s Modulus - X (Pa) 1.14e+11
Young’s Modulus - Y (Pa) 7.58e+09
Young’s Modulus - Z (Pa) 7.58e+09
Poisson’s Ratio - XY 0.32
Poisson’s Ratio - YZ 0.37
Poisson’s Ratio - XZ 0.35
Shear Modulus - XY (Pa) 5.45e+09
Shear Modulus - YZ (Pa) 2.96e+09
Shear Modulus - XZ (Pa) 2.96e+09
The blade is composed of an outer surface and an inner spar. The thickness of the
outside surface linearly decreases from 0.1 m at the root to 0.005 m at the tip. The spar
has a similar thickness behavior with 0.1 m at its closest point to the root and 0.03 m
at the tip. In the following table, there are the thickness specifications needed along
with their location with respect to the global coordinate system (which represents the
center of an imaginary hub and thus the center of rotation).
75
Table 5.3.2: Thickness specifications
X(m) Thickness (m)

-1 0.1
Surface
-44.2 0.005
-3 0.1
Spar
-44.2 0.03
Total Deformation
In Figure 5.3.11 it’s possible to see the deflection of the blade due to the pressure
loads. As expected, the blade deflection is away from the incoming wind, like actual
wind turbines do. In Figure 5.3.12 it’s also possible to see the undeformed wire frame,
to better understand the blade deflection from the standard unloaded configuration.
Figure 5.3.11: Blade deflection regions
Figure 5.3.12: Blade deflection compared to unloaded case
76
Von Mises Equivalent Stress
Regarding the equivalent stresses, it can be seen that the maximum stress is located
around the mid-span of the blade and is about 32 MegaPascal. This stress is tensile
due to the deflection of the blade away from the incoming wind.
Figure 5.3.13: Equivalent stress - Blade front face
Figure 5.3.14: Equivalent stress - Blade back face
It is worth noting that in the back face of the blade, as shown in Figure 5.3.14, the
stresses are lower than in the front face of the blade. Again, this is due to the direction
of the deflection of the blade from the pressure loads generated by the incoming
wind.
77
Force Reaction
Regarding the root radial force, it can be seen that almost all the force reaction is in the
x-direction (more than 95% of the total force). This is absolutely normal, considering
the centripetal acceleration acting on the blade. In table 5.3.3, the values of the force
in all directions are computed.
Figure 5.3.15: Root radial force
Table 5.3.3: Force values in all directions
Maximum Value Over Time

X - Axis 1.582e+06 N
Y - Axis 16633 N
Z - Axis 64911 N
Total 1.663e+06 N
78
Moment Reaction
Finally, the bending moment acting on the blade is shown. In this case, the majority
(almost 94%) of the reaction bending moment is about the y-axis, due to the loading
on the blade and its consequently deformation.
Figure 5.3.16: Bending moment reaction
Table 5.3.4: Moment values in all directions
Maximum Value Over Time

X - Axis 23829 Nm
Y - Axis 2.26E+06 Nm
Z - Axis 1.31E+05 Nm
Total 2.413E+06 Nm
It’s important to say that much more results were obtained from this structural analysis
(i.e shear stresses or elastic strain), but it was decided to show just the most important
results for the scope of this project.
79
Chapter 6
Wind Gust Analysis with ATE
The technological advancement in wind turbines is not only centred in a higher energy
efficiency, but also in a more safe and economical design of the machine. From
this point of view, a deeper knowledge of the environmental conditions in which the
turbine has to operate is of utmost importance. Hence, methods for the prediction
and simulation of a possible extreme load event have been developed. A wind gust is
defined as a short term speed variation within a turbulent wind field. Its typical shape
can be characterized by some parameters, as clarified in Figure 6.0.1.
Figure 6.0.1: Characteristic parameters of an average wind gust
80
CHAPTER 6. WIND GUST ANALYSIS WITH ATE
The flow field, incoming with an average speed u , is followed by a negative fluctuation,
a positive peak and another negative pulse before stabilizing again. The gust relative
amplitude a is the speed difference u−umax . The gust rise time b is defined as the
period between the beginning of the gust and the reaching of the maximum speed umax ,
whereas the maximum gust variation c is the absolute difference umax − umin . The
lapse time d is defined as the period between the max and min velocity occurrences. A
parameter to evaluate the gust magnitude is the gust factor G = umax /u.
6.1 Gust Models

The importance of gusts for structural issues and loading assessments, combined
with their inherent random nature, promoted the development of various gust
models. Average gust shapes, probability distributions and simulated wind fields were
investigated. The simplest gust models derive from plain mathematical functions.
These are Gaussian profiles or sinusoidal profiles, as one-minus-cosine, sine and
cosine [7]. The one-minus-cosine is one of the most popular average gust shape
model, due to its simplicity, and is currently adopted for load assessment in aerospace
industry and international standards. However a good matching with experimental
gust measurements is not always present.
Therefore more complex descriptions were investigated. Many approaches are based
on the statistical approach of averaging a suitably large number of measured gusts
to define a gust shape model. Larsen et al. [42] discuss a model to extract a
mean speed gust shape based on the peak over threshold detection procedure. The
benefit of including realistic gusts in the stochastic turbulence of the Atmospheric
Boundary Layer (ABL) is explained. The original stochastic time series is transformed
into a series of Dirac delta functions located at the detected gusts intervals, more
precisely at the peak instant. The mean gust shape is then extracted as a function
turbulence standard deviation, gust amplitude and time. The spatial shape of the
gust is investigated too, by applying the previous procedure also to a second position.
The produced gust shapes are then compared both with measured wind data sets and
numerically simulated wind field, with good results. Various terrain conditions and
atmospheric stability situations are tested too.
81
The availability of useful measurements for the gust assessment is limited in space and
time though. Knigge and Raasch [41] performed a LES simulations to provide a virtual
multi-dimensional data sets of atmospheric turbulence with all the ABL characteristics.
Both 1D and 2D gust profiles were extracted by means of peak over threshold method.
A comparison between the obtained 1D gust profile, the one-cosine gust shape and
the measured data is performed. A noticeable difference is found between the 1D
profile from Large Eddies Simulation (LES) calculation and the one-cosine shape,
mostly in the steep increase and decrease of the wind speed. Therefore, an adjustment
is proposed for the one-cosine formulation, parametric for each component of the
velocity vector. The accuracy of the method is underlined for low altitudes, suitable
for wind turbine applications.
Seregina et al. [66] demonstrate the extrapolation of wind gust velocities from
common hourly measurements. The hypothesis made is that both wind and gust
velocities follow Weibull distributions. Then the gust model is build as a transfer
function between shape and scale parameters of the two different distributions. As gust
are fluctuations of wind, the two distributions are thought to have similar shape and
different scale. Therefore, the estimation of the gust speed by the combination of the
two Weibull functions is explained and validated with meteorological data. Mann [63]
elaborated an algorithm to simulate a complete 3D turbulent wind field with realistic
ABL behaviour, including gusts and shear layer. The model is based on the spectral
tensor for atmospheric turbulence at high wind speeds. The wind turbine company
Vestas developed an appropriate software which generates a wind field based on Mann
model and gives a 3D velocity components as output. The field is divergence-free, so
mass is conserved and the use in CFD simulations is possible.
6.2 IEC Standard
Besides the deep research about wind gusts, the urgency of a safe turbine operation
claims for straightforward specifications. As other works about wind turbines and
gusts, the approach adopted in this analysis follows the International Electrotechnical
Commission (IEC) guidelines [71]. A series of well-defined wind gust models with a set
of extreme loading cases are listed in IEC 61400-2. Even if the standard is meant for
design purposes, in the present work it has been used as an international reference for
82
the definition of a gust shape and for an immediate structural verification. For these
reasons, the case which will be analyzed in this Chapter doesn’t follow exactly the load
cases of the standard, as the main goal is to study the aerodynamic response of the
turbine in a transient operating condition.
Larger machines are provided with control systems to prevent rotor over-speed and
blade overload, but the turbine has to withstand even these loads for safety and
continuity reasons. In fact, a system fault is always possible. The most commonly
followed procedure consists in the definition of several load cases which may represent
the turbine lifetime. In the following analysis, it is considered an Extreme Operating
Gust (EOG) for the case of an operative load without fault, which is the most common
case, an operative load defined during the turbine normal productive operation.
6.2.1 Extreme Operating Gust
This is the most representative gust profile, a symmetric so called Mexican hat shape.
Before and after the rather steep peak, a brief speed dip is included. Gust amplitude
and duration vary with the return period. The analytical form is expressed by equation
6.1, for zhub < 30 m, and the resulting velocity fluctuation at hub height in time is
reported in Figure 6.2.1. A normal stationary profile u(z) is supposed to persist up
to t = 0 s, when the gust begins.




 u(z) t<0

u(z, t) = u(z) − 0.37ugust (z)sin(3π t )(1 − cos(2π t )) 0<t<T (6.1)

 T T


 u(z) t>T
The gust period T is the total event duration, considered 12 s. After that time, the
velocity profile goes back to u(z). Hence, the EOG vertical profile shape in time
corresponds to a pulsating normal profile.
83
Figure 6.2.1: Canonical Mexican hat gust shape profile (z = zhub = 12.2m)
6.3 CFD Case Configuration & Methodology

One of the main reasons for structural issues in offshore wind turbines is the presence
of sudden wind gusts during a normal operation, with a resulting increase in loads
the blades have to withstand. From the results in Chapter 4, it has been proved
how the presence of an adaptive trailing edge flap could provide significant variations
in the aerodynamic forces upon the airfoil. In particular, from Figure 4.5.16, it has
been shown that the lift force in function of an increasing wind speed, has a relatively
small increase when the flap is moved upwards, with a -10° configuration. From this
interesting remark, it was thought to assess the influence of a movable trailing edge flap
during the presence of a wind gust, in order to understand if this solution can reduce
the loads upon the blades.
The same S809 airfoil used in Chapter 4 (flap length = 20% chord) was considered
for this analysis, with the only difference of a small gap (1% chord) between the airfoil
and the flap, as it is possible to see in Figure 6.3.1. This gap was made for the sake
of the simulation, in order to allow a transient simulation with a rotating flap without
the risk of a collapsing mesh or formation of negative element volumes, which would
compromise or abort the computation. A Mexican hat canonical shape for the gust
model was used thanks to its simple, but realistic shape, as explained in the previous
sections.
84
Figure 6.3.1: S809 airfoil with ATE and small gap
To assess the effectiveness of the adaptive trailing edge flap during the wind gust, it was
tried to maintain the lift curve as flat as possible, like during normal operational time
at rated power, with a constant free stream velocity which would generates a steady
and not dangerous load on the blades.
Same CFD methodology of Chapter 4 was applied, with the only difference that in
this case the simulation is transient. Time step size was set to 0.01 s with a maximum
number of iterations for time step equal to 70, in order to smoothly reach convergence.
A number of time steps equal to 1200 was set, so to cover all the 12 s of gust period.
For the last case, with the fully transient movement of the flap, the mesh morphing
technique was used in FLUENT and the mesh stiffness was improved in order to avoid
mesh collapsing or formation of negative volumes during the simulation which would
lead to the abortion of the computation. Inlet velocity boundary condition was again
applied, but in this case it’s not constant, but it varies following the Mexican hat gust
shape. After this preliminary set up, 84000 iterations were launched.
85
6.4.1 2-D Grid & Gust Model
For grid generation, again the integrated mesh generator in ANSYS was used, due to
its extensive mesh functions and ease of use. The S809 airfoil has a chord of 600 mm
and 21% thickness and it is to be tested with a Reynolds number of 1.0×106 . The chord
length of the airfoil geometrical model was set to 1 m in ANSYS, which resulted a free
stream air flow of about 15 m/s due to dynamic similarity. The flap length was set at
20% of the airfoil chord, so 200 mm of flap length with a gap between the airfoil and
the flap of 10 mm for computational reasons.
To avoid possible effects of far field boundary on the flow around the airfoil, the
grid domain was set to be 30 chord in horizontal and 30 chord in vertical direction
away from the blade, as shown previously in Chapter 4, Figure 4.5.1. An unstructured
triangular mesh was applied in this case to the geometrical model, with a total number
of 44554 elements along with a set of structured boundary layers with quadrilateral
elements (inflation layers). The boundary layer was attached on the airfoil surface
with 15 layers and a relatively slow growth rate of 1.1 starting with a row depth 0.0002
m. Size function was defined for refinement around the leading and trailing edge to
capture the stagnation point and wake flow respectively.
Figure 6.4.1: Detailed mesh around the airfoil
86
Unstructured grid is known to be more capable of conforming to desired geometries

due to flexible element shapes and sizes, while compared with structured grid the
computation time is expected to be longer due to the explicit storage of inter cell
connectivity. In case of complex geometries or moving mesh, an unstructured mesh is
suggested due to its simplicity compared to a structured one. The inner mesh around
the airfoil was constructed with triangles except for the boundary layer for better
geometry conforming.
Regarding the wing gust, it was modelled following the canonical Mexican hat shape
from IEC, in order to consider a real model used for blade design and load forecasting.
It can be seen in Figure 6.4.2 the wind gust modelled with the aforementioned standard
shape, which will be also the boundary condition for the inlet wind speed considered
in this simulation.
Figure 6.4.2: Inlet wind speed - mexican hat shape
The gust modelled in this Figure, follows perfectly the standard in Figure 6.2.1: before
the peak, a brief speed deep is included and the peak can reach a maximum wind
speed around 37 m/s. The duration of this phenomenon was considered to be around
12 s, which is a good time-scale for a gust fitting well the Mexican hat shape from
IEC. In order to asses the effectiveness of the flap, it was decided to compare three
different cases: initially the forces upon the airfoil are calculated without any flap
control applied, later on a quasi-steady state simulation is performed in order to move
the flap just during the peak of the gust, and finally a fully transient simulation with
the flap moving in a continuous fashion is done.
87
6.4.2 Stationary Flap Response
In this first case, the forces generated by the modelled wind gust on a standard S809
airfoil are determined. It’s important to underline that the geometric model is the
same one of Chapter 4, but in this case the inlet velocity is changing in function of
time, following the curve shown in Figure 6.4.2. For this reason, the simulation is
considered to be a transient simulation due to the variable inlet wind speed and also
the response of the airfoil, as expected, will change in function of the inlet wind speed.
The difference between the next cases, it’s in the stationary behaviour of the flap, which
will be always in its standard configuration, as shown in Figure 6.4.1.
Figure 6.4.3: Lift force generated by the wind gust
Figure 6.4.4: Drag force generated by the wind gust
88
The lift force L(t) and drag force D(t) monitored during the simulation are reported
in Figure 6.4.3 and Figure 6.4.4. It can be seen for the lift plot that the trend reflects
well the wind gust profile: coming from 25 m/s, the first value for t=0 s agrees with
the lift result of the steady case with the same speed. The effect lowers, and after the
peak the trend continues almost in a symmetrical way, restoring the initial value at
t=12 s. Within 3 seconds from t=3 s the lift experiences a steep enhancement, with
an increase of about 70% from the beginning. This strong gradient is of interest for
structural ultimate and fatigue analyses of both the blade and the shaft.
Same consideration can be done also for the drag plot: the trend reflects well the wind
gust profile and it’s very similar to the lift plot, indeed the L/D ratio is almost constant
during the whole gust (11 ÷ 12). This is due because in this aerodynamic analysis these
variables depend only on the flow kinetic energy and due to their purely aerodynamic
cause, the curves reflect the gust shape, with the slow oscillation period of the gust. A
real blade would bend in downwind direction, so even gravitational and inertial forces
would influence the result.
Figure 6.4.5: Velocity contours of wind gust evolution: t=0 s, t=2 s, t=6 s, t=12 s
89
In Figure 6.4.5 it’s possible to see the evolution of the velocity contours during the
analyzed wind gust with a total period t=12 seconds. In the first picture, at time t=0 s,
a stable situation is shown, normally with a constant free stream velocity. In the second
picture, at t=2 s, the wind speed is decreased from the stable configuration, showing
the brief speed deep before the gust. In the third picture, at t=6 s, the peak with the
maximum wind speed reached is shown: a big velocity enhancement is present on the
suction surface of the airfoil and for Bernoulli equation, it means that a big pressure
gradient is generated from the suction surface and the pressure surface, enhancing the
lift force. For this reason, during the peak, the lift force experiences a steep increase
as shown in Figure 6.4.3. In the last picture, at t=12 s, a stable situation after the wind
gust is again reached.
6.4.3 Quasi-Steady Response
In this second intermediate analysis, it was decided to move the flap upwards by 5°
(Figure 6.4.6) just during the peak, to check whether it has a positive influence on the
blade loads. From a timeline point of view, the flap was moved from its stationary
position at time t=0 s directly to the sloped position at t=3.5 s and finally again to the
stationary one at time t=7.5 s. For this reason, the behaviour of the flap was considered
to be quasi-steady respect to the changing wind speed velocity during the simulation.
Figure 6.4.6: Airfoil configuration with flap at -5° inclination angle
90
The lift force L(t) monitored during the simulation is reported in the following Figure.
It can be seen also in this case that the trend of the lift plot reflects the shape of the wind
gust profile, but for this configuration the benefits of the flap are evident. In Figure
6.4.7 are compared the lift force for the first standard configuration (blue curve) and
the second configuration with the flap moved upwards (orange curve). From t=3.5 s
when the steep enhancement in wind speed starts, until t=7.5 s when the velocity starts
decreasing, the peak was flatted about 40% of its original value, which means a serious
and positive effect of the flap on the loads generated by this wind gust.
Figure 6.4.7: Lift force comparison between two flap configurations: 0° and -5°
It’s possible to see that also in this case the brief speed deep before and after the
peak are still present, as well as the peak in lift force is not completely flattened. The
outcome for this intermediate analysis was to understand if moving the flap could give
some important results from a load point of view and what was understood is that
the movable flap can modify significantly the lift force generated by the gust, if it is
controlled properly. The next step for this analysis, which will be shown in the next
section, is to see if moving the flap in a continuous fashion during the whole wind gust
can reduce completely the peak in lift and flatten the curve for the whole period of the
phenomenon, like if no gust was come and the velocity remains constant, producing a
constant value of lift.
91
Figure 6.4.8: Vel. contours comparison between first and second case at t=3.5 s
Figure 6.4.9: Pres. contours comparison between first and second case at t=3.5 s
In Figure 6.4.8 and Figure 6.4.9 it’s possible to see respectively the velocity contours
and pressure contours for the two configurations analyzed so far. It can be well
observed that in the first configuration an higher pressure drop (velocity increment)
is present in the upper surface of the airfoil. From this higher pressure difference
between the suction surface and pressure surface, an higher lift force is generated by
the airfoil, always perpendicular to the incoming wind speed for definition. In this
condition, the flap has a fundamental contribution to decrease the loads on the blade:
indeed it modifies the useful area facing the incoming wind on the pressure surface,
decreasing the pressure difference between the upper and lower face of the airfoil.
Furthermore, it’s interesting to see that a slight increase of pressure is created on the
suction surface directly on the flap when it’s moved upwards, hence it’s very important
to design it properly to guarantee an efficient utilization. It’s important to underline
that in this project all the issues related with the actuator system and sensors are not
considered, due to the limited importance from an aerodynamic and aeroelastic point
of view.
92
6.4.4 Fully Transient Response
In this last analysis, a fully transient simulation is performed. The airfoil flap is moving
upwards and downwards according to the inlet velocity in a continuous fashion, in
order to maintain the lift curve function as flat as possible during the wind gust. Thanks
to this continuous motion of the flap and flattened lift force function, it would be
possible to reduce or theoretically avoid any dynamic structural load caused by the
wind gust. As explained in Chapter 4, in order to increase the lift, the flap has to be
moved downwards, while to decrease the lift it has to be moved upwards. It follows that
during the speed deep before and after the peak the flap inclination will have positive
values, while during the peak it will have negative values.
For this computation, a fixed range of flap inclination is considered for simplicity,
which is approximately between -10° and +10° from the chord line. The limits of this
range were set according to several studies about trailing edge flaps applications in the
wind energy sector: with this range boundaries and flap length, the stresses upon the
mechanical connections between flap and airfoil are considered almost neglectable.
The lift force L(t) and drag force D(t) monitored during the simulation are reported
respectively in Figure 6.4.10 and 6.4.11.
Figure 6.4.10: Lift comparison between two flap cases: 0° and movable (±10°)
93
Figure 6.4.11: Drag comparison between two flap cases: 0° and movable (±10°)
It can be well observed how the continuous motion of the flap had a significant impact
on the lift and drag force experienced by the airfoil. In Figure 6.4.10 the lift force for
the ”No Flap” case (grey curve) has the same shape of the Mexican hat wind gust, as it
was explained in the previous sections. In this case, the steep lift enhancement during
the peak produces high loads the blade has to withstand, with serious problems in
maintenance and reliability. In the ”Flap” case instead (blue curve), the lift force was
almost completely flattened thanks to the motion of the flap, which increased the lift
during the speed drop and decreased the lift during the peak.
Thanks to a smart use of the flap, it was possible to reduce significantly the lift
produced by a sudden wind gust, with a resulting enormous benefit from the loads
applied to the rotor blade. This result is of extreme importance, because it shows an
alternative possible solution for loads control besides traditional pitch control. The
main advantage of flap controls respect to pitch control, mainly during a wind gust, is
that flap control has a much faster reaction than pitch control, if designed properly.
Furthermore, it allows the possibility to control the loads in a local scale, without the
need to pitch the whole blade if not necessary. This is also very important, because
frequently high dynamic loads are applied only to a local section on the blade and
pitching the entire blade would signify some losses of useful power as well.
94
In Figure 6.4.11 the drag force for the two cases is shown. It can be seen also in this
case the positive influence of the flap, which didn’t increase the drag during the gust,
but instead did decreased it compared to the standard case. The L/D ratio (Glide ratio)
is almost constant during the whole gust (12 ÷ 15), which is quite similar for the ratio
of the first case. This means that the flap does not increase the aerodynamic efficiency
of the airfoil, but it allows an important management of the loads applied to the airfoil,
reducing significantly the experienced lift.
In the following Figure it’s shown the flap inclination during the time period of the wind
gust. As previously anticipated, during the wind speed decrease the flap was moved
towards the pressure surface in order to increase the lift and maintain the curve as flat
as possible, while during the gust peak it was moved to the suction surface in order
to decrease the lift and reduce the aerodynamic loads, as seen before. It’s interesting
to observe how the flap inclination function has approximately the opposite shape of
the Mexican hat gust shape, with a velocity raise before and after the central velocity
drop. Knowing the behaviour of the flap in response to the inlet wind speed, through a
smart sensor system it would be possible to design an ideal flap motion for every type
of incoming wind gust and boundary condition.
Figure 6.4.12: Flap inclination during the wind gust
95
Figure 6.4.13: Velocity contours during the wind gust evolution (t=0 s → t=12 s)
In Figure 6.4.13 it’s possible to see the evolution of the velocity contours during the
analyzed wind gust, with a total period t=12 seconds. In the first and third row,
respectively from t=0 s to t=3.5 s and from t=8.5 s to t=12 s, the brief velocity deeps
before and after the peak are shown. It can be seen during the velocity decrease
the motion of the flap towards the pressure surface and then its return to the initial
configuration. It can be also observed that when the flap is down, a low speed (high
pressure) zone is created at the end of the flap on the suction surface, which explains
well the increase in lift compared to the ”No flap” configuration.
In the second row, from t=3.5 s to t=8.5 s, the steep enhancement of wind velocity is
reported. In this case, during the velocity increase the motion of the flap is towards the
suction surface, producing an important lift reduction during the peak, as explained in
the previous section.
96
Chapter 7
Conclusions
Since the current design trend is to have longer blades for wind turbines in order
to achieve higher power output, blades are going to be less stiff and exposed to
higher loads. The first objective of this project was to study the aerodynamic and
aeroelastic behaviour of a wind turbine blade by investigating the interaction between
the structure and the fluid, both in 2D and 3D configuration. The second objective of
the project was to assess the effectiveness of a trailing edge flap applied to the studied
airfoil, in order to analyzed its influence in reducing the dynamic loads applied to the
blade, mainly produced during a wind gust event.
Initially, the computational model applied to a wind turbine airfoil has been
successfully validated with a comparison between numerical CFD results and
experimental data. The results of aerodynamic coefficients, pressure distributions and
velocity contours have been confirmed. Then the behaviour of the turbine operating
in unsteady conditions has been presented together with the results of the transient
simulation. Through an analysis in the frequency domain of the signal generated by the
oscillating airfoil, it was evaluated the low possibility of flutter for the studied geometry
and boundary conditions.
Afterwards, a 3D fluid structure-interaction analysis was performed to a specific wind

turbine blade, using the provided specifications. The blade model was designed with
a commercial CAD software, and an aerodynamic analysis was done at the beginning.
Blade velocity and pressure loads were determined in this stage, in order to have the
input for the following aeroelastic analysis. The total deformation, the Von Mises
97
CHAPTER 7. CONCLUSIONS
equivalent stress, the root radial force and the bending moment applied to the blade
were finally calculated, in order to understand the behaviour of this blade interacting
with the incoming wind.
Finally, the effectiveness of applying a trailing edge flap to the airfoil during a wind
gust was assessed. Initially, the airfoil response with its standard configuration was
tested: lift and drag force in function of time are reported. Afterwards, a quasi-steady
computation was performed in order to observe the initial response of the flap, with
positive benefits during the gust peak in reducing the lift. Lastly, a fully transient
simulation was done during the entire wind gust time: the flap played a fundamental
role in flattening almost completely the lift function, with positive effects in reducing
the aerodynamic loads produced by a sudden wind gust.
7.1 Future Work

Based on the difficulties and uncertainties faced in this project, the following ideas
are recommended to be attempted in future works so that the methods established in
this project could be further validated and enhanced in term of accuracy and scope of
potential applications.
Firstly, alternative sources of unsteady aerodynamic loading should be included in

the analysis. Alternative sources such as variation of wind speed with height (i.e to
model the atmospheric boundary layer) should be included in future analysis. This
would introduce an additional dimension to the project which will allow the assessment
method to be applied to assess the feasibility of constructing wind turbines in different
and more realistic wind climate conditions.
Secondly, due to a limited amount of time, studies such as grid density effects and
yaw angle for the 3D wind turbine were not performed, and is needed for further
research. Improvement on stall prediction can also be investigated, maybe with the
utilization of more complicated turbulence models like unsteady-Reynolds-Averaged
Navier–Stokes (URANS), detached eddy simulations (DES) or Large Eddy Simulation
(LES), in order the assess the aerodynamic performance of the wind turbine.
98
CHAPTER 7. CONCLUSIONS
Thirdly, a technical and economical feasibility assessment should be done for the
utilization of trailing edge flaps as load control technique for offshore wind turbines.
Also a technical and economical comparison between pitch control and flap control in
different unsteady conditions should be done, in order to understand the drawbacks
and benefits of these control solutions and when one is better than the other. A huge
study is also needed regarding the mechanical linkage between the flaps and the blade,
the actuator system which controls the flap movement and all the sensors needed for
a correct utilization of the trailing edge flap.
Lastly, a natural development and possible next stage for this project is to make an
active wind gust control using machine learning and deep reinforcement learning
techniques, in a such a way to have the flap moving automatically according to the inlet
wind speed condition without any external control. In this way, it would be possible to
have an optimal control of this complex dynamic system maintenance free.
Figure 7.1.1: Deep reinforcement learning for aerodynamic optimization
99
Bibliography
[1] Abdel Gawad, Ahmed. “New, Simple Blade-Pitch Control Mechanism for Small-
Size, Horizontal-Axis Wind Turbines”. In: Journal of Energy and Power
Engineering 7 (Dec. 2013), pp. 2237–2248. DOI: 10.17265/1934-8975/2013.
12.004.
[2] Andersen, Peter, Gaunaa, Mac, Bak, Christian, and Buhl, Thomas. “Load
alleviation on wind turbine blades using variable airfoil geometry”. In: (Jan.
2006).
[3] Arakawa, C., Fleig, O., Iida, M., and Shimooka, M. “Numerical Approach for
Noise Reduction of Wind Turbine Blade Tip with Earth Simulator”. In: 2005.
[4] Aslam Bhutta, Muhammad Mahmood, Hayat, Nasir, Farooq, Ahmed Uzair, Ali,
Zain, Jamil, Sh. Rehan, and Hussain, Zahid. “Vertical axis wind turbine – A
review of various configurations and design techniques”. In: Renewable and
Sustainable Energy Reviews 16.4 (2012), pp. 1926–1939. ISSN: 1364-0321.
DOI: https : / / doi . org / 10 . 1016 / j . rser . 2011 . 12 . 004. URL: http : / /
www.sciencedirect.com/science/article/pii/S136403211100596X.
[5] Bangga, Galih, Lutz, Thorsten, Jost, Eva, and Krämer, Ewald. “CFD studies on
rotational augmentation at the inboard sections of a 10 MW wind turbine rotor”.
In: Journal of Renewable and Sustainable Energy 9.2 (2017), p. 023304. DOI:
10.1063/1.4978681.
[6] Barlas, T K and Kuik, G A M van. “State of the art and prospectives of smart
rotor control for wind turbines”. In: Journal of Physics: Conference Series 75
(July 2007), p. 012080. DOI: 10.1088/1742-6596/75/1/012080. URL: https:
//doi.org/10.1088%2F1742-6596%2F75%2F1%2F012080.
[7] Bartels, Robert E. “Development, verification and use of gust modeling in the
nasa computational fluid dynamics code fun3d”. In: (2012).
100
BIBLIOGRAPHY
[8] Basualdo, Santiago. “Load alleviation on wind turbines using variable airfoil
geometry (a two-dimensional analysis)”. In: Fluid Mechanics Section. Master
of Science, Technical University of Denmark (2004).
[9] Bergami, Leonardo. “Adaptive Trailing Edge Flaps for Active Load Alleviation
in a Smart Rotor Configuration”. English. PhD thesis. Denmark, 2013.
[10] Binnie, A. M. and Easterling, H. J. “Hydrodynamics and Hydraulics”. In:

Journal of Fluid Mechanics 38.4 (1969), pp. 855–856. DOI: 10 . 1017 /
S0022112069212655.
[11] Bir, Gunjit and Jonkman, Jason. “Aeroelastic Instabilities of Large Offshore and
Onshore Wind Turbines”. In: Journal of Physics: Conference Series 75 (July
2007), p. 012069. DOI: 10.1088/1742-6596/75/1/012069.
[12] Björck, A, Dahlberg, J Aa, Östman, A, and Ganander, H. “Computations of

aerodynamic damping for blade vibrations in stall”. In: EWEC-CONFERENCE-.
BOOKSHOP FOR SCIENTIFIC PUBLICATIONS. 1997, pp. 503–507.
[13] Bossanyi, EA. “GH bladed theory manual”. In: GH & Partners Ltd 2 (2003),
pp. 56–58.
[14] Bottasso, C.L., Campagnolo, F., Croce, A., and Tibaldi, C. “Optimization-
based study of bend–twist coupled rotor blades for passive and integrated
passive/active load alleviation”. In: Wind Energy 16.8 (2013), pp. 1149–1166.
DOI: 10.1002/we.1543. eprint: https://onlinelibrary.wiley.com/doi/pdf/
10.1002/we.1543. URL: https://onlinelibrary.wiley.com/doi/abs/10.
1002/we.1543.
[15] Breton, Simon-Philippe. “Study of the stall delay phenomenon and of wind
turbine blade dynamics using numerical approaches and NREL’s wind tunnel
tests”. In: (June 2008).
[16] Buhl, Thomas and Bak, Christian. “Erratum: “Potential Load Reduction Using
Airfoils with Variable Trailing Edge Geometry””. In: Journal of Solar Energy
Engineering-transactions of The Asme - J SOL ENERGY ENG 128 (Nov. 2006).
DOI: 10.1115/1.2344829.
[17] Buhl, Thomas, Gaunaa, Mac, and Bak, Christian. “Load reduction potential
using airfoils with variable trailing edge geometry”. In: 43rd AIAA Aerospace
Sciences Meeting and Exhibit. 2005, p. 1183.
101
BIBLIOGRAPHY
[18] Buhl, Thomas, Gaunaa, Mac, and Bak, Christian. “Potential load reduction using
airfoils with variable trailing edge geometry”. In: (2005).
[19] Buhl, Thomas, Gaunaa, Mac, Bak, Christian, Hansen, Per, and Clemmensen,
Kasper. “Measurements on the Thunder TH-6R actuator”. In: Proceedings of
the ASME2007 Fluids Engineering Division Summer Meeting, Riso. 2005.
[20] Buning, P., Jespersen, D., Pulliam, Tom, Chan, William, Slotnick, J., Krist,
Steven, and Renze, K. “OVERFLOW User”s Manual, Version 1. 8b”. In: (Jan.
1998).
[21] Carrión, M., Steijl, Rene, Woodgate, Mark, Barakos, George, Munduate, Xabier,
and Gomez-Iradi, Sugoi. “Aeroelastic analysis of wind turbines using a tightly
coupled CFD–CSD method”. In: Journal of Fluids and Structures 50 (Oct.
2014). DOI: 10.1016/j.jfluidstructs.2014.06.029.
[22] Chang, YL, Yang, SL, and Arici, O. Flow field computation of the NREL
S809 airfoil using various turbulence models. Tech. rep. American Society of
Mechanical Engineers, New York, NY (United States), 1996.
[23] Chaviaropoulos, PK. “Flap/lead–lag aeroelastic stability of wind turbine

blades”. In: Wind Energy: An International Journal for Progress and
Applications in Wind Power Conversion Technology 4.4 (2001), pp. 183–200.
[24] Chiu, Phillip. “Aerodynamics and Optimal Design of Biplane Wind Turbine
Blades”. PhD thesis. University of California, Los Angeles, Jan. 2017.
[25] Dam, C., Berg, D., and Johnson, S. J. “Active load control techniques for wind
turbines.” In: 2008.
[26] Duque, Earl and Johnson, Wayne. “Navier-Stokes and Comprehensive Analysis
Performance Predictions of the NREL Phase VI Experiment”. In: Journal of
Solar Energy Engineering-transactions of The Asme - J SOL ENERGY ENG
125 (Jan. 2003). DOI: 10.1115/1.1624088.
[27] Echjijem, Imane and Djebli, Abdelouahed. “Design and Optimization of Wind
Turbine with Axial Induction Factor and Tip Loss Corrections”. In: Procedia
Manufacturing 46 (2020). 13th International Conference Interdisciplinarity in
Engineering, INTER-ENG 2019, 3–4 October 2019, Targu Mures, Romania,
pp. 708–714. ISSN: 2351-9789. DOI: https://doi.org/10.1016/j.promfg.
2020.03.100. URL: http://www.sciencedirect.com/science/article/pii/
S2351978920309781.
102
BIBLIOGRAPHY
[28] Fung, Yuan Cheng. An introduction to the theory of aeroelasticity. Courier

Dover Publications, 2008.
[29] Gaunaa, Mac. “Unsteady 2D potential-flow forces on a thin variable geometry

airfoil undergoing arbitrary motion”. In: (2006).
[30] Gómez-Iradi, S and Barakos, G N. “Computational fluid dynamics investigation

of some wind turbine rotor design parameters”. In: Proceedings of the
Institution of Mechanical Engineers, Part A: Journal of Power and Energy
222.5 (2008), pp. 455–470. DOI: 10.1243/09576509JPE526.
[31] Griffin, D A. “NREL Advanced Research Turbine (ART) Aerodynamic Design of

ART-2B Rotor Blades”. In: (Sept. 2000). DOI: 10.2172/763408.
[32] Guerri, Ouahiba, Bouhadef, Khadidja, and Harhad, Ameziane. “Turbulent Flow
Simulation of the NREL S809 Airfoil”. In: Wind Engineering 30.4 (2006),
pp. 287–301. DOI: 10.1260/030952406779295471.
[33] Haidar, Muhmmad, Kamel, M., Shabka, A., and Negm, Hanan. “Flutter
Investigation of Isotropic 3-D Wings”. In: International Conference on
Aerospace Sciences and Aviation Technology 15 (May 2013), pp. 1–13. DOI:
10.21608/asat.2013.22192.
[34] Hand, MM, Simms, DA, Fingersh, LJ, Jager, DW, Cotrell, JR, Schreck, S, and
Larwood, SM. Unsteady aerodynamics experiment phase VI: wind tunnel test
configurations and available data campaigns. Tech. rep. National Renewable
Energy Lab., Golden, CO.(US), 2001.
[35] Hansen, M.O.L., Sørensen, J.N., Voutsinas, S., Sørensen, N., and Madsen,
H.Aa. “State of the art in wind turbine aerodynamics and aeroelasticity”. In:
Progress in Aerospace Sciences 42.4 (2006), pp. 285–330. ISSN: 0376-0421.
DOI: https : / / doi . org / 10 . 1016 / j . paerosci . 2006 . 10 . 002. URL: http :
//www.sciencedirect.com/science/article/pii/S0376042106000649.
[36] Hansen, MH. “Aeroelastic stability analysis of wind turbines using an eigenvalue
approach”. In: Wind Energy: An International Journal for Progress and
[37] Hansen, Morten. “Stability analysis of three-bladed turbines using an

eigenvalue approach”. In: 42nd AIAA Aerospace Sciences Meeting and Exhibit.
2004, p. 505.
103
BIBLIOGRAPHY
[38] Hansen, Morten Hartvig. “Aeroelastic instability problems for wind turbines”.
In: Wind Energy: An International Journal for Progress and Applications in
Wind Power Conversion Technology 10.6 (2007), pp. 551–577.
[39] Hoffmann, MJ, Reuss Ramsay, R, and Gregorek, GM. Effects of grit roughness
and pitch oscillations on the NACA 4415 airfoil. Tech. rep. National Renewable
Energy Lab., Golden, CO (United States); The Ohio State …, 1996.
[40] J.G. Shepers A. J. Brand, M. M. Hand. “Enhanced Field Rotor Aerodynamics

Database”. In: IEA RD Wind Executive Committee (1997), pp. 1–104.
[41] Knigge, Christoph and Raasch, Siegfried. “Improvement and development of

one-and two-dimensional discrete gust models using a large-eddy simulation
model”. In: Journal of Wind Engineering and Industrial Aerodynamics 153
(2016), pp. 46–59.
[42] Larsen, Gunner C and Hansen, Kurt S. “Database on Wind Characteristics–

Analyses of Wind Turbine Design Loads”. In: Risø National Laboratory,
Roskilde, Denmark. Feb (2004).
[43] Larsen, Torben J., Aagaard Madsen, Helge, and Thomsen, K. “Investigation of
stability effects of an offshore wind turbine. The new aeroelastic code HAWC2”.
eng. In: Windtech International 2.March (2006), pp. 33–35. ISSN: 15742415.
[44] Lobitz, Don W. “Aeroelastic stability predictions for a MW-sized blade”. In:
Wind Energy: An International Journal for Progress and Applications in Wind
Power Conversion Technology 7.3 (2004), pp. 211–224.
[45] Lobitz, Don W. “Parameter sensitivities affecting the flutter speed of a MW-sized
blade”. In: (2005).
[46] Lutz, T, Wolf, A, Wiirz, W, and Jeremiasz, JG. “Design and verification of an
airfoil with trailing edge flap and unsteady wind tunnel tests”. In: Stuttgart,
Germany (2011).
[47] Marrant, BAH, Van Holten, Th, and Kuik, GAM van. “Smart Dynamic Rotor
Control of Large Offshore Wind Turbines”. In: Inventory of Rotor Design
Options and Possible Load Reductions, Duwind (2002).
Control of Large Offshore Wind Turbines”. In: Inventory of Present Techniques
(2002).
104
BIBLIOGRAPHY
Control of Large Offshore Wind Turbines”. In: Inventory and assessment of
available tools (2002).
[50] Mayda, Edward, Dam, C.P., and Nakafuji, Dora. “Computational Investigation
of Finite Width Microtabs for Aerodynamic Load Control”. In: (Jan. 2005). DOI:
10.2514/6.2005-1185.
[51] Nayeri Pechlivanoglu, Paschereit. “Performance Optimization of Wind Turbine

Rotors With Active Flow Control”. In: Proceedings of the ASME 2011 Turbo
Expo: Turbine Technical Conference and Exposition 1 (2011), pp. 763–775.
[52] Obeid, Sohaib, Ahmadi, Goodarz, and Jha, Ratneshwar.

“NARMAX Identification Based Closed-Loop Control of Flow Separation over
NACA 0015 Airfoil”. In: Fluids 5.3 (2020), p. 100.
[53] Øye, Stig. “FLEX4 simulation of wind turbine dynamics”. In: Proceedings of the
28th IEA Meeting of Experts Concerning State of the Art of Aeroelastic Codes
for Wind Turbine Calculations. 1996, pp. 129–135.
[54] Pape, A Le and Lecanu, J. “3D Navier–Stokes computations of a stall-regulated

wind turbine”. In: Wind Energy: An International Journal for Progress and
[55] Peeters, Mathijs, Santo, Gilberto, Degroote, Joris, and Van Paepegem, Wim.
“The Concept of Segmented Wind Turbine Blades: A Review”. In: Energies 10
(July 2017), p. 1112. DOI: 10.3390/en10081112.
[56] Petersen, J Thirstrup, Madsen, Helge Aagaard, Björck, Anders, Enevoldsen,

Peder, Øye, Stig, Ganander, Hans, and Winkelaar, Danny. “Prediction of
dynamic loads and induced vibrations in stall”. In: (1998).
[57] Petersen, J Thirstrup, Thomsen, Kenneth, and Madsen, Helge Aagaard. Local
blade whirl and global rotor whirl interaction. 1998.
[58] Potsdam, Mark and Mavriplis, Dimitri. “Unstructured Mesh CFD Aerodynamic
Analysis of the NREL Phase VI Rotor”. In: 47th AIAA Aerospace Sciences
Meeting including The New Horizons Forum and Aerospace Exposition. DOI:
10.2514/6.2009-1221. eprint: https://arc.aiaa.org/doi/pdf/10.2514/6.
2009-1221. URL: https://arc.aiaa.org/doi/abs/10.2514/6.2009-1221.
105
BIBLIOGRAPHY
[59] R. Reuss, M. J. Hoffmann and Gregorek, G. M. “Effects of Grit Roughness and

Pitch Oscillations on the S809 Airfoil”. In: (1995), pp. 1–158.
[60] Radmanesh, Amir Reza, Abbaspour, Madjid, and Soltani, Mohamad. “Effect of
Unsteady Flow over Aerodynamic Fluctuation of Different Multi MW Horizontal
Axis Wind Turbine Blade Profiles on SST-K-ω Model”. In: June 2015.
[61] Rasmussen, F, Petersen, Jørgen Thirstrup, Winkelaar, D, and Rawlinson-Smith,

R. “Response of stall regulated wind turbines-stall induced vibrations”. In: Final
Report on JOULE 1 (1993).
[62] Rasmussen, Flemming, Petersen, J Thirstrup, and Madsen, H Aagaard.

“Dynamic stall and aerodynamic damping”. In: (1999).
[63] Sathe, Ameya, Mann, Jakob, Barlas, Thanasis, Bierbooms, WAAM, and Van
Bussel, GJW. “Influence of atmospheric stability on wind turbine loads”. In:
Wind Energy 16.7 (2013), pp. 1013–1032.
[64] Schreck, S, Sant, T, and Micallef, D. “Rotational Augmentation Disparities in

the MEXICO and UAE Phase VI Experiments: Preprint”. In: (May 2010).
[65] Sepahy, D. “Advances in Aeroelasticity”. In: Numerical Techniques for

Engineering Analysis and Design. Ed. by G. N. Pande and J. Middleton.
Dordrecht: Springer Netherlands, 1987, pp. 153–161.
[66] Seregina, Larisa S,

Haas, Rabea, Born, Kai, and Pinto, Joaquim G. “Development of a wind gust
model to estimate gust speeds and their return periods”. In: Tellus A: Dynamic
Meteorology and Oceanography 66.1 (2014), p. 22905.
[67] Sieros, G., Chaviaropoulos, P., Sørensen, J. D., Bulder, B. H., and Jamieson, P.
“Upscaling wind turbines: theoretical and practical aspects and their impact on
the cost of energy”. In: Wind Energy 15.1 (), pp. 3–17. DOI: 10.1002/we.527.
URL: https://onlinelibrary.wiley.com/doi/abs/10.1002/we.527.
[68] Simms, David, Schreck, S., MM, Hand, and Fingersh, L. “NREL Unsteady
Aerodynamics Experiment in the NASA-Ames Wind Tunnel: a Comparison of
Predictions to Measurements”. In: (Jan. 2001). DOI: 10.2172/783409.
[69] Somers, Dan M. Design and experimental results for the S809 airfoil. Tech. rep.
National Renewable Energy Lab., Golden, CO (United States), 1997.
106
BIBLIOGRAPHY
[70] Sørensen, N. N., Michelsen, J. A., and Schreck, S. “Navier–Stokes predictions of

the NREL phase VI rotor in the NASA Ames 80 ft × 120 ft wind tunnel”. In: Wind
Energy 5.2�3 (2002), pp. 151–169. DOI: https://doi.org/10.1002/we.64.
eprint: https://onlinelibrary.wiley.com/doi/pdf/10.1002/we.64. URL:
https://onlinelibrary.wiley.com/doi/abs/10.1002/we.64.
[71] Standard, IEC. “61400-2,(2006) Design requirements for small wind turbines”.
In: International Electrotechnical Commission (2006).
[72] THEODORSEN, T. “General Theory of Aerodynamic Instability and the

Mechanism of Flutter”. In: N. A. C. A. Thecnical Report 496 (1935). URL: https:
//ci.nii.ac.jp/naid/10003492039/en/.
[73] Thomsen, K, Petersen, J Thirstrup, Nim, E, Øye, S, and Petersen, B. “A

method for determination of damping for edgewise blade vibrations”. In: Wind
Energy: An International Journal for Progress and Applications in Wind
Power Conversion Technology 3.4 (2000), pp. 233–246.
[74] Troldborg, Niels. “Computational study of the risø-b1-18 airfoil equipped with
actively controlled trailing edge flaps”. In: Technical University of Danmark
(2004).
[75] Tsiantas, Theofanis, Manolas, Dimitris I, Machairas, Theodore, Karakalas,

Anargyros, Riziotis, Vasilis A, Saravanos, Dimitrios, and Voutsinas, Spyros G.
“Assessment of fatigue load alleviation potential through blade trailing edge
morphing”. In: Journal of Physics: Conference Series. Vol. 753. 4. 2016.
[76] Vermeer,
L.J., Sørensen, J.N., and Crespo, A. “Wind turbine wake aerodynamics”. In:
Progress in Aerospace Sciences 39.6 (2003), pp. 467–510. ISSN: 0376-0421.
DOI: https : / / doi . org / 10 . 1016 / S0376 - 0421(03 ) 00078 - 2. URL: http :
//www.sciencedirect.com/science/article/pii/S0376042103000782.
[77] Von Karman, Th. “Turbulence and skin friction”. In: Journal of the
Aeronautical Sciences 1.1 (1934), pp. 1–20.
[78] “Will communities “open-up” to offshore wind? Lessons learned from New
England islands in the United States”. In: Energy Research Social Science 34
(2017), pp. 13–26. ISSN: 2214-6296. DOI: https://doi.org/10.1016/j.erss.
2017.05.009.
107
BIBLIOGRAPHY
[79] Wolfe, Walter, Ochs, Stuart, Wolfe, Walter, and Ochs, Stuart. “CFD calculations
of S809 aerodynamic characteristics”. In: 35th Aerospace Sciences Meeting and
Exhibit. 1997, p. 973.
[80] Yarusevych, Serhiy, Sullivan, Pierre E, and Kawall, John G. “On vortex shedding
from an airfoil in low-Reynolds-number flows”. In: Journal of Fluid Mechanics
632 (2009), p. 245.
[81] Zahle, Frederik, Johansen, Jeppe, Sørensen, Niels, and Graham, J. Michael R.
“Wind Turbine Rotor-Tower Interaction Using an Incompressible Overset Grid
Method”. English. In: The Proceedings of the 45th AIAA Aerospace Sciences
Meeting and Exhibit. See also http://www.aiaa.org/; null ; Conference date: 08-
01-2007 Through 11-01-2007. United States: American Institute of Aeronautics
and Astronautics, 2007.
[82] Zhang, Zhenyou. “Automatic Fault Prediction of Wind Turbine Main Bearing
Based on SCADA Data and Artificial Neural Network”. In: Open Journal of
Applied Sciences, Vol.8 No.6, June 28, 2018 ().
108
Appendix - Contents
A S809 Airfoil Profile Coordinates 110
B NREL Phase VI Turbine Blade Root Surface Depiction 111
C Data for NREL Phase VI Turbine 112
109
Appendix A
S809 Airfoil Profile Coordinates
110
Appendix B
NREL Phase VI Turbine Blade Root

Surface Depiction
111
Appendix C
Data for NREL Phase VI Turbine
• Number of blades: 2
– Rotor diameter: varies with tip attachment
– 10.058 m with standard tip or smoke tip (all sequences except V and W)
– 9.886 m with tip plate (Sequence V)
– 11.064 m with tip extension (Sequence W)
• Hub height: 12.192 m
• Type of rotor: teetered (Sequences B, C, D, G, and 4) or rigid
• Rotational speed: 71.63 RPM synchronous speed, 90 RPM (Sequence X only)
• Cut in wind speed: 6 m/s some tests were run at 5 m/s
• Power regulation: stall
• Rated power: 19.8 kW
• Tilt: 0°
• Cone angle: 0° , 3.4° , or 18°
• Location of rotor: upwind or downwind
• Rotational direction: counter clockwise (viewed from upwind)
• Rotor overhang: 1.401 m (yaw axis to blade axis); 1.469 m (yaw axis to teeter pin)
(Source: Hand et al, 2001)
112

Master Thesis Lorenzo Cantoni

Uploaded by

Copyright:

Available Formats

Master Thesis Lorenzo Cantoni

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Master Thesis Lorenzo Cantoni

Uploaded by

Copyright:

Available Formats

RENE Master of Science,

KTH Royal Institute of Technology,

Master Thesis Report

Parties involved in the Project

Wind Turbine, Offshore, Blade, Rotor, Computational Fluid Dynamics, Airfoil,

The Aeroelasticity Group at Imperial College London, especially to my supervisor

CFD Computational Fluid Dynamics

HAWT Horizontal Axis Wind Turbine

VAWT Vertical Axis Wind Turbine

LCOE Levelized Cost Of Energy

AFC Active Flow Control

DOF Degree of Freedom

UAE Unsteady Aerodynamics Experiment

BEM Blade Element Momentum

ABL Atmospheric Boundary Layer

LES Large Eddies Simulation

EOG Extreme Operating Gust

2 Experiments and Literature Review 11

3 Aerodynamics and Aeroelasticy 20

4 2-D Validation & Analysis 36

5 3-D Fluid Structure Interaction 64

6 Wind Gust Analysis with ATE 80

1.1.1 Continuous growth of rated power during the last years . . . . . . . . . 2

3.2.1 Cross section of a modern wind turbine blade . . . . . . . . . . . . . . . 23

4.1.1 Schematic of the NREL S809 airfoil . . . . . . . . . . . . . . . . . . . . 36

4.5.3 Detailed mesh around the leading edge . . . . . . . . . . . . . . . . . . 46

5.1.1 CAD model of the wind turbine blade . . . . . . . . . . . . . . . . . . . 64

6.0.1 Characteristic parameters of an average wind gust . . . . . . . . . . . . 80

6.2.1 Canonical Mexican hat gust shape profile (z = zhub = 12.2m) . . . . . . 84

7.1.1 Deep reinforcement learning for aerodynamic optimization . . . . . . . 99

4.3.1 Test cases at 15 m/s free stream velocity . . . . . . . . . . . . . . . . . . 40

5.3.1 Blade material properties . . . . . . . . . . . . . . . . . . . . . . . . . . 75

1.1 Problem Overview

1.2 Scope of the Work and Project Outline

Afterwards, a 3D fluid structure-interaction analysis will be performed to a specific

1.3 Wind Turbine Layout: HAWT vs VAWT

Figure 1.3.1: Components of an HAWT

Figure 1.3.2: Layout of a VAWT

1.4 Load Control in a Wind Turbine

Sum of Costs Over Lifetime (Capital Costs + O&M Costs)

Figure 1.4.1: Simplified power curve for commercial wind turbine

• Pitch Control and Stall Control

• Variable Speed Rotor

• Variable Blade Length

• Blade Section Aerodynamic Optimization

1.4.1 Pitch Control and Stall Control

1.4.2 Variable Blade Length

Figure 1.4.2: Variable blade length concept

1.4.3 Aerodynamic Optimization

Figure 1.4.3: Schematic representation of existing AFC devices

Trailing Edge Flap

Figure 1.4.4: Flexible trailing edge flap

Leading Edge Flap

Figure 1.4.5: Flexible leading edge flap

Active Gurney Flap

Figure 1.4.6: Active Gurney flap

Experiments and Literature Review

2.1 Experimental Testing for Wind Turbines

NREL UAE Phase VI