Modelling and Control of Large Wind

Modelling and Control of Large Wind
Turbine
Modellering och reglering av stora vindkraftverk
Syed Hammad Zafar
Faculty of Health, Science and Technology

Master’s Program in Electrical Engineering
Degree Project of 30 credit points
External Supervisor : Wouter Engels, Energy Research Center of Netherlands, The Netherlands
Internal Supervisor : Jorge Solis, Karlstad University, Sweden
Examiner : Arild Moldsvor, Karlstad University, Sweden
Date : 09.10.2013
Serial number
I hereby declare that all work, submitted as my final thesis report, is com-
pleted on my own with the supervision of both external and internal super-
visors and used no other sources or tools than the ones listed. All references
and citations are made accordingly.
Karlstad, October2013
Syed Hammad Zafar
Master’s Thesis Report
Abstract
In order to make the wind energy an economical alternative for energy pro-
duction, upscaling of turbine to 10 − 15M W maybe necessary to reduce
the overall cost of energy production. This production target requires a
considerable increase in the turbine size and placing the turbines at high
wind speed locations. But increase in turbine size also increases the uneven
load distribution across the turbine structure. Therefore an efficient load
reduction technique is necessary to increase the turbine reliability in high
wind speed locations. Variable speed wind turbine offers most desirable load
reduction through actively pitch angle control of turbine blades. Research
has shown that the Individual Pitch Control (IPC) is most promising option
for turbine load reduction.
This thesis work is focused on modelling of a large wind turbine and imple-
mentation of a new mutlivariable control concept for turbine load reduction.
A detailed mathematical model is designed which includes turbine blade
and tower dynamics and a proposed Linear Quadratic Gaussian (LQG) al-
gorithm is implemented for Individual Pitch Control (IPC) loop of wind
turbine. Proposed model in this thesis work is derived from the previous
turbine model used in ECN with additional tower dynamics. My contri-
bution in turbine modelling portion is to linearise the equations of motion
to form a statespace model and to implement LQG algorithm for turbine
active load reduction. This proposed method is compared with the previous
control technique used in ECN for turbine fatigue load reduction to measure
the overall efficiency of the proposed technique.
Fatigue load has major effect on the turbine working age. In quantitative
way, proposed LGQ design offers 8−10% approx more fatigue load reduction
in comparison with the previous design. In simple convention, decrease in
turbine fatigue load increases the turbine age. This 8 − 10% fatigue load
reduction offers 8 − 10% minimum increase in turbine working age which
means that if a turbine works for 20 years in total for energy production,
this proposed technique will add 2 extra years into the turbine working life.
This age increase has major economic impact to make the wind turbine a
viable alternative for energy production.
Syed Hammad Zafar i

Acknowledgements
This thesis work has been carried out to fulfil the requirement for Mas-
ter’s degree in Electrical Engineering. First of all, I would like to thank
ALMIGHTY ALLAH for giving me this opportunity and also the strength
to achieve this milestone.
Secondly, my sincere thanks to the government of Sweden and Karlstad

university for providing me this wonderful opportunity to pursue higher
education in their university.
I would like to express the deepest appreciation to my external supervisor

Dr. Wouter Engels from Energy Research Centre of the Netherlands (ECN).
Without his support and cooperation, I would not be able to deal with this
challenging project. I am also very grateful to Dr. Stoyan Kanev for giving
me his precious time and guidance throughout this project.
I would like to thank my internal supervisor Dr. Jorge Solis, Associate

Professor at Karlstad University. His excellent support and guidance has
always been an encouragement to achieve the best results.
Finally, I would like to dedicate this work to my parents and family specially
to my uncle Mr. Syed Arshad Ali for his unconditional love and support
throughout the life.
Syed Hammad Zafar ii

Contents
1 Introduction 1
1.1 Energy Research Centre of the Netherlands (ECN) . . . . . . 1
1.2 Historical Development of Wind Energy . . . . . . . . . . . . 2
1.3 Motivation . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
1.4 Background . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
1.5 Problem Definition . . . . . . . . . . . . . . . . . . . . . . . . 5
1.6 Research Objectives . . . . . . . . . . . . . . . . . . . . . . . 5
1.7 Thesis Outline . . . . . . . . . . . . . . . . . . . . . . . . . . 6
2 Literature Study 8
2.1 Architectural Design Of Wind Turbine . . . . . . . . . . . . . 8
2.2 How Turbine Blades Capture Wind Energy? . . . . . . . . . . 9
2.3 Tip Speed Ratio . . . . . . . . . . . . . . . . . . . . . . . . . 10
2.4 Turbine Power Output . . . . . . . . . . . . . . . . . . . . . . 11
2.5 Maximum Power Captured By Wind Turbine . . . . . . . . . 12
2.6 Loads Of Wind Turbine . . . . . . . . . . . . . . . . . . . . . 13
2.7 Sustainable Control Concept . . . . . . . . . . . . . . . . . . 14
2.8 Cyclic Pitch Control . . . . . . . . . . . . . . . . . . . . . . . 15
2.9 Individual Pitch Control (IPC) . . . . . . . . . . . . . . . . . 15
2.9.1 Load Reduction Through IPC . . . . . . . . . . . . . . 16
2.9.2 Higher Harmonics Control . . . . . . . . . . . . . . . . 16
2.10 Design Filters and IPC Gain . . . . . . . . . . . . . . . . . . 17
3 Modelling of Wind Turbine 19

3.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
3.2 Multibody System . . . . . . . . . . . . . . . . . . . . . . . . 19
3.3 Mass Spring Damper Configuration . . . . . . . . . . . . . . . 21
3.3.1 Blade Motion . . . . . . . . . . . . . . . . . . . . . . . 23
3.3.2 Tower Motion . . . . . . . . . . . . . . . . . . . . . . . 23
3.3.3 Blade Velocities . . . . . . . . . . . . . . . . . . . . . . 23
3.3.4 Blade Flap-wise Bending Moments . . . . . . . . . . . 24
3.3.5 Boundary Condition . . . . . . . . . . . . . . . . . . . 24
3.3.6 Joint Force . . . . . . . . . . . . . . . . . . . . . . . . 24
Syed Hammad Zafar iii

CONTENTS
3.4 Linearised Model Calculations . . . . . . . . . . . . . . . . . . 25

3.4.1 Coleman Transformation . . . . . . . . . . . . . . . . 25
3.4.2 Linearisation . . . . . . . . . . . . . . . . . . . . . . . 26
3.5 Statespace Representation . . . . . . . . . . . . . . . . . . . . 31
3.5.1 Model Description . . . . . . . . . . . . . . . . . . . . 31
3.5.2 Matrix Representation . . . . . . . . . . . . . . . . . . 33
3.6 Model Constraints . . . . . . . . . . . . . . . . . . . . . . . . 35
4 Model Analysis 36
4.1 Stability Analysis . . . . . . . . . . . . . . . . . . . . . . . . . 36
4.1.1 Open Loop System . . . . . . . . . . . . . . . . . . . . 36
4.1.2 Velocities Feedback Loop . . . . . . . . . . . . . . . . 37
4.1.3 System With Blade Velocities and Flapwise Bending
Moments Feedback Loops . . . . . . . . . . . . . . . . 39
4.1.4 Why To Make Analysis Based On Model Configura-
tion With and Without Design Filters ? . . . . . . . . 41
4.2 Time Domain Analysis . . . . . . . . . . . . . . . . . . . . . . 43
4.3 Frequency Domain Analysis . . . . . . . . . . . . . . . . . . . 46
4.3.1 MIMO Nyquist Stability Criteria . . . . . . . . . . . . 46
4.4 Model Fatigue and Pitch Activity Calculations . . . . . . . . 50
4.4.1 Fatigue Calculation . . . . . . . . . . . . . . . . . . . 50
4.4.2 Pitch Activity Calculation . . . . . . . . . . . . . . . . 52
5 Controller Design 54
5.1 Multivariable Control . . . . . . . . . . . . . . . . . . . . . . 54
5.1.1 PID Control . . . . . . . . . . . . . . . . . . . . . . . 54
5.1.2 Model Predictive Control . . . . . . . . . . . . . . . . 55
5.2 Linear Quadratic Gaussian Control . . . . . . . . . . . . . . . 55
5.2.1 Kalman Filter Design . . . . . . . . . . . . . . . . . . 56
5.2.2 Linear Quadratic Regulator Design . . . . . . . . . . . 58
5.3 Fatigue Calculation With LQG Design . . . . . . . . . . . . . 60
5.4 Pitch Activity Calculation With LQG Design . . . . . . . . . 61
5.5 Fatigue Reduction VS Control Effort . . . . . . . . . . . . . . 62
5.6 Economic Aspect of Project . . . . . . . . . . . . . . . . . . . 64
6 Conclusions and Future Work 66

6.1 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . 66
6.2 Future Work . . . . . . . . . . . . . . . . . . . . . . . . . . . 67
A Mathematical Model Calculations 69
B Coleman’s Transformation Calculation 94
C Matlab Script 100
Syed Hammad Zafar iv

CONTENTS
Bibliography 110
Syed Hammad Zafar v

List of Figures
1.1 Energy Research Centre of the Netherlands . . . . . . . . . . 1
2.1 Horizontal axis upwind wind turbine . . . . . . . . . . . . . . 8

2.2 Forces experienced by rotor blade in airflow . . . . . . . . . . 9
2.3 Rotor blade at AoA = α . . . . . . . . . . . . . . . . . . . . 10
2.4 Power output with steady wind speed . . . . . . . . . . . . . 11
2.5 Tilt and Yaw movement of turbine . . . . . . . . . . . . . . . 13
2.6 Schematic diagram of Sustainable Control concept . . . . . . 14
2.7 Turbine Configuration . . . . . . . . . . . . . . . . . . . . . . 17
2.8 Turbine Configuration . . . . . . . . . . . . . . . . . . . . . . 18
3.1 Turbine with blade and tower hinge points . . . . . . . . . . 20

3.2 Schematic diagram of blade with hinge point and complete
layout . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
3.3 Schematic diagram of MSD configuration of turbine . . . . . 22
3.4 Turbine Model . . . . . . . . . . . . . . . . . . . . . . . . . . 32
3.5 Wind Model Designed at ECN . . . . . . . . . . . . . . . . . 35
4.1 Open Loop System . . . . . . . . . . . . . . . . . . . . . . . . 36

4.2 Pole Zero Plot for Open Loop System . . . . . . . . . . . . . 37
4.3 Blade Velocities Feedback Loop . . . . . . . . . . . . . . . . . 38
4.4 Pole Zero Plot for System with Blade Velocities Feedback Loop 38
4.5 Turbine Model . . . . . . . . . . . . . . . . . . . . . . . . . . 39
4.6 Pole Zero plot for Complete Model with and without Design
Filters . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40
4.7 Varying Gain with and without Design Filters . . . . . . . . . 42
4.8 Poles Sorting . . . . . . . . . . . . . . . . . . . . . . . . . . . 43
4.9 Impulse Response with and without Design Filters . . . . . . 44
4.10 Unit Step Response with and without Design Filters . . . . . 45
4.11 Model Configuration for MIMO Nyquist Stability Analysis . . 46
4.12 Nyquist Plot of Complete Model with and without Design
Filters . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49
4.13 Fatigue for Blade Flapwise Bending Moments . . . . . . . . 50
4.14 Fatigue for Tower Bending Moments . . . . . . . . . . . . . . 51
Syed Hammad Zafar vi

LIST OF FIGURES
4.15 Pitch Activity . . . . . . . . . . . . . . . . . . . . . . . . . . . 52

4.16 Pitch Activity (PSD) . . . . . . . . . . . . . . . . . . . . . . . 53
5.1 Schematic diagram of Kalman Filter implementation . . . . 56

5.2 Individual solution for each node . . . . . . . . . . . . . . . . 57
5.3 Kalman Filter Estimates for Blade Bending Moments . . . . 58
5.4 Schematic diagram of LQG design . . . . . . . . . . . . . . . 59
5.5 System Response with Varying Weighting Factor . . . . . . . 60
5.6 Pitch Activity Comparison . . . . . . . . . . . . . . . . . . . 61
5.7 Pitch Activity Comparison . . . . . . . . . . . . . . . . . . . 62
5.8 Fatigue VS Effort . . . . . . . . . . . . . . . . . . . . . . . . . 63
5.9 Fatigue VS Effort . . . . . . . . . . . . . . . . . . . . . . . . . 65
6.1 Blade layout with two hinge points . . . . . . . . . . . . . . . 68
B.1 Coleman Transformation & Inverse Coleman Transformation

eaxmple using Matlab script . . . . . . . . . . . . . . . . . . . 97
Syed Hammad Zafar vii

List of Abbreviation
HAWT Horizontal Axis Wind Turbine

IPC Individual Pitch Control
CPC Collective Pitch Control
MSD Mass Spring Damper
AoA Angle of Attack
λ Tip Speed Ratio
Myaw Moment of Yaw
Mtilt Moment of Tilt
1p,2p,3p Once Per, Twice Per, Thrice Per Revolution
θ Pitch Angle
EOM Equation of Motion
CM Coleman Transformation
ψ Azimuth Angle
SISO Single Input Single Output
MIMO Multi Input Multi Output
PSD Power Spectral Density
PID Proportional Integral Derivative
LQG Linear Quadratic Gaussian
Syed Hammad Zafar viii

Chapter 1
Introduction
1.1 Energy Research Centre of the Netherlands

(ECN)
Energy Research Centre of the Netherlands (ECN) is the one of the leading
research institutes for energy innovation in European region. ECN holds a
strong international position for the development of knowledge and technol-
ogy that enable a transition to a sustainable energy system.
Figure 1.1: Energy Research Centre of the Netherlands
ECN has six core activities in which they work together with industry, the
business sector, government authorities and knowledge institutes.
• Wind energy: Focused on reducing offshore wind energy cost, through
improvement in turbine design, turbine parts, through proper control
Syed Hammad Zafar 1

Introduction
design programming for more efficient and reliable turbine working

and by integrated application of wind farm system expertise.
• Solar power: Latest scientific research is applied in collaboration with

solar panel manufacturers to make solar power cheap and to make it
more easily deployable.
• Biomass: Focus on thermochemical conversion of biomass and biobased

economy. Research areas include pretreatment, gasification, purifica-
tion and biorefining.
• Energy efficiency: Focused on making industrial process more efficient

using heat pumps, separation technology and revolutionary chemical
reactors in collaboration with different industries.
• Environment & Energy Engineering: Focused on innovative businesses

by creating experimental installations, prototypes and high tech com-
ponents.
• Policy studies: Strategic advices are developed by combined efforts

of engineers, economists, social science experts and environmental ex-
perts with technical departments.
1.2 Historical Development of Wind Energy

Fact that wind is a power house, was realized thousands of years ago. Sail-
ing boats, ships and wind-driven building ventilations have been using wind
power since ancient times. In 1st century AD, a Greek engineer Heron of
Alexandria designed wind-driven wheel to power a machine. Later on this
idea prevails and a new era of windmills came into account in late 9th cen-
tury. These windmills were primarily used for water pumping, flour grinding
and for construction purposes. A Scottish Professor James Blyth build the
first cloth-sailed wind turbine for electricity production in July 1887 [1].
After that horizons of this unlimited renewable power house expanded grad-
ually.
Today modern world energy consumptions is much more than the collective
amount of energy consumed in previous centuries. Thirst for energy have
increased significantly in recent times because of the technological evolution
in the modern world. This increase in energy demands has not only made
electrical energy very expensive but also increased the usage of fossil fuels,
like coal and gas, which are limited in quantity and also the prime source
of increase in CO2 emission in environment. These challenges made the
world to lean more onto the sustainable energy resources which are not only
economically viable alternative but also offers opportunities for business as
well.
Syed Hammad Zafar 2

Introduction
Among all the sustainable energy sources available today, wind power has
gained extensive popularity. It is one of the fastest globally growing renew-
able electrical power industries. Annual growth rate of installed wind energy
capacity is 27% on average throughout the last decade. According to the
latest report released by the Global Wind Energy Council (GWEC) claims
that global installed capacity for wind power is now at 282.5GW. Prime
reason for this growth is that most of the countries around the globe are
trying to minimize their oil dependency and also finding ways to minimize
the global warming effect.
1.3 Motivation
Wind energy, in simplest words, is the use of kinetic energy of the wind to
generate electricity. Renewable energy industries are trying to transform
this unlimited resource of wind into a renewable energy powerhouse [2].
Two main criteria have been followed by the wind industry to make the
wind energy a viable alternative for energy production.
• Increase in turbine size.
• Placing turbine at high wind speed location.
Commercial turbine manufacturing companies are focused on large wind tur-

bines due to the economic reasons. These typical turbines have significantly
large blade spans (70 − 100m in diameter) to increase the blade swept area,
sky-high towers (70 − 100m in height) with power rating around 750kW
to 7.5M W . Placing these large wind turbines at high wind speed location
(offshore locations) gives many advantages like high speed wind availability,
turbulence in wind is low at offshore locations because of no hurdles in wind
path, increase in rotor size doesn’t effect living or agricultural spaces. A
bundle of different technologies work together to extract maximum energy
from these mechanical structures.
However offshore wind energy is viable alternative if energy price drops and
reliability of wind turbines is improved. Increase in turbine size allows tur-
bine blades to experience as much wind as possible. But this increase in
blade size also increases the fatigue and asymmetrical loads [3] on wind
turbine. Due to the turbulent nature of wind flow, all three blades expe-
rience uneven load distribution and at offshore locations this difference in
load distribution is significantly high [4]. This asymmetrical loading further
increases the supervision and maintenance cost of off-shore wind turbines
which means that detailed modelling and efficient control mechanism is re-
quired for these load reduction.
Syed Hammad Zafar 3

Introduction
1.4 Background
This thesis work deals with the development of the turbine model and im-
plementation of a multivariable control technique for the turbine load re-
duction. Wind turbine, in general, is a complex system which is composed
of interconnected flexible bodies which undergo translational or rotational
displacement due to the wind force. Multibody system approach [5, 6] is fol-
lowed for the modelling of dynamic behaviour of wind turbine. Formulation
of multibody system characterise the mathematical appearance of turbine
dynamics in terms of equations of motion. Basically these equations of mo-
tion of turbine dynamics are the differential equations based on the design
specification provided by the manufacturer. Two major configurations are:
• Mass Spring Damper Configuration (MSD)
• Tuned Circuit Configuration (LC circuit)
Both these two configurations are used to characterise the differential equa-
tion for the turbine dynamics. In our case Mass Spring Damper configura-
tion had been utilized because of the previous work and information provided
regarding the design specification of the wind turbine mechanical structure.
To deal with the problem of turbine load reduction, almost all the turbine
manufacturers employ “Control Turbine Concept” which usually is a com-
bination of different controllers one of them is to actively adjust the turbine
blades to vary the rotational speed [7]. This technique offers two main ad-
vantages:
• Extreme and fatigue load reduction of wind turbine.
• Enables the turbine operation in changing and extreme weather.
Number of different techniques are available for this problem including Col-
lective Pitch Control (CPC), Stall Control and Individual Pitch Control
technique. In Collective Pitch Control technique all the three turbine blades
are pitched with the same angle at the same time. Major advantage of Col-
lective Pitch Control technique is that the controller design is less complex in
comparison with other technique. Load reduction through Collective Pitch
Control gives satisfactory results for the small wind turbines with power
rating within 200kW ←→ 800kW where the turbine blade size is not that
big and all blades experience almost the same wind load distribution.
However in large wind turbines with power rating within 800kW ←→ 7.5M W ,
load distribution is significantly different because of the size of blades and
pitching all blades with the same angle doesn’t work for the load reduction.
There comes the concept of Individual Pitch Control technique (IPC) in
Syed Hammad Zafar 4

Introduction
which actively adjusting the pitch of each blade individually based on the
individual load of each blade in real time. Individual pitch control (IPC)
techniques offers better load reduction due to many reasons [8, 9, 10, 11]:
• Individual pitch control technique improves the trade-off between en-

ergy outcome, blade loads and actuator system.
• Individual pitch control technique implementation requires limited
physical adjustment as almost all commercial wind turbines are equipped
with individual pitch actuation system for each blade. Load reduction
through IPC is an attractive option instead of a mechanical system
design to deal with the same problem.
• Individual pitch control technique relies on sensors output for asym-
metrical loads of the turbine and these load sensors with the sufficient
level of reliability are now available.
• Individual pitch control technique offers the reduction of asymmetrical
loads across the rotor as the turbine size increases, these asymmetrical
loads becomes more critical for performance and reliability factor.
1.5 Problem Definition

In order to make the wind energy a viable alternative for energy production,
increase in turbine size and placing turbines at high wind speed locations
is necessary to reduce the overall cost and to achieve the maximum out-
put. But these requirements also bring the problems of turbine fatigue and
extreme loads due to the uneven wind load distribution across the turbine
structure. To meet these requirements, detailed turbine modelling and a
proper control mechanism is essential to reduce the turbine fatigue and ex-
treme loads that will lead to the improvement in reliability and efficiency for
the energy production of wind turbine. Therefore the problem definition for
this thesis work is to provide a comprehensive briefing on current turbine
model and control design technique used for turbine load reduction used in
ECN. Then to design a new detailed turbine model which includes turbine
blade and tower dynamics and implementation of a new proposed controller
design through Individual Pitch Control technique to achieve the maximum
load reduction. Finally a comparison between the new designed model and
previous model to analyse the overall improvement in terms of turbine load
reduction.
1.6 Research Objectives

The main objectives of this thesis work is to improve the reliability and effi-
ciency of a large wind turbine through fatigue and extreme load reduction.
Syed Hammad Zafar 5

Introduction
From academic point of view, previous knowledge and experience gained

in the field of mathematics / numerical analysis, optimal and model base
control system approach will be utilized for modelling and control design
respectively. Simulations will be carried out on Matlab / Simulink. These
simulations will show important problem images / blocks like sensors, actu-
ators, controller etc. This overall objective of thesis work is further divided
into the four subtask to organize the complete work.
• Investigate Current Control Design: Understanding and analysis

of current control design technique for turbine load reduction used in
ECN.
• Turbine Modelling: Detailed model designing for the large wind

turbine including turbine blade and tower dynamics.
• Model Analysis: Model analysis using the current control design

technique to check system stability and performance.
• Multivariable Control Implementation: New proposed multivari-

able control design technique implementation and investigation of ef-
fects on turbine fatigue load reduction in comparison with current
control design.
1.7 Thesis Outline

This thesis report consist of 6 chapters in total. Brief overview of each
chapter is presented here.
Chapter 2 describes the research study of working and control of a wind

turbine using Individual Pitch Control technique and current control design
technique implemented at ECN for turbine load reduction.
Chapter 3 describes the development of a detailed model design for large

wind turbine which includes turbine blade and tower dynamics and devel-
opment of a statespace model which includes all the feedback loops of the
system.
Chapter 4 discusses all the analysis carried out on the designed model using
current control technique.
Chapter 5 discusses the new proposed control design technique using LQG
algorithm and it’s comparison with the current design on the basis of tur-
bine fatigue load reduction.
Syed Hammad Zafar 6

Introduction
Chapter 6 describes the final conclusions of the work based on all the anal-
ysis and comparison and future work.
Syed Hammad Zafar 7

Chapter 2
Literature Study
2.1 Architectural Design Of Wind Turbine

Wind turbine is a machine that converts the wind kinetic energy in electrical
energy. There are two main architectures of wind turbines used for the
electrical power generation world wide.
• Horizontal Axis Wind Turbine
• Vertical Axis Wind Turbine
This thesis work is focused on large wind turbine with three blades (120◦
apart from each other) and Horizontal axis configuration.
Figure 2.1: Horizontal axis upwind wind turbine
Syed Hammad Zafar 8

Literature Study
This is the most popular architecture of commercial wind turbines and rep-
resents more than 90% of the installed capacity. In horizontal axis wind tur-
bines, rotor shaft and electrical generator are placed at the top of the tower.
Generator is supplied with the rotational movement of blades through a gear
box. Turbine under consideration is a variable speed wind turbine which in
other words, generator and rotor speed of turbine are not strictly coupled
to the grid frequency which is the case for fixed-speed wind turbines [12].
2.2 How Turbine Blades Capture Wind Energy?

Modern wind turbine works like an aircraft. Blades works on lift and drag
principle. When blades are exposed to the air they experience life force due
to its shape. More curved side experience the low pressure of air while high
pressure on other side. This pressure difference on both side pushes the
airfoil and net result is the lift force which is perpendicular to the direction
of wind flow [4].
Figure 2.2: Forces experienced by rotor blade in airflow
Lift forces will increase when rotor blade is turned in a manner which in-
creases the angle of incidence of air on it. This is called Angle of Attack
(AoA) [13]. Lift force is described by the lift coefficient [3] :
L
A
CL = 1 2
(2.1)
2 ρv
where ρ is the air density, A is the cross sectional area of airfoil, L is the lift
force and v is the wind velocity.
Syed Hammad Zafar 9

Literature Study
Figure 2.3: Rotor blade at AoA = α
Increasing angle of attack after a certain limit (known as Stall) will again
decrease the lift forces experienced by the blades and will dramatically in-
crease the drag force which is parallel to wind flow and pushes the airfoil
in wind direction [4]. Drag force also increases with the increase in angle of
attack (AoA) and coefficient of drag force is described as [3]:
D
A
CD = 1 2
(2.2)
2 ρv
where ρ is the air density, A is the cross sectional area of airfoil, D is the
Drag force and v is the wind velocity.
Thrust force T is the resultant of Lift and Drag forces that rotates rotor
blades. This resultant ratio is also the function of angle of attack for any
L
given airfoil. Maximum value of Lift to Drag ratio ( D ) maintains the max-
imum efficiency of wind turbine [4].
2.3 Tip Speed Ratio

Tip speed ratio (λ) refers to the ratio between rotational speed of tips of
wind turbine blades to the wind speed
T ip speed of blade
λ= (2.3)
W ind speed
If the rotor speed is very slow much of wind will pass through the gaps
without being captured by the blades and power efficiency will decrease.
But if rotor speed is too high then blades will form a solid wall against
air flow and also rotor blade creates turbulence in air as it passes through
Syed Hammad Zafar 10

Literature Study
it. If next blade comes too quickly, it will experience that turbulent wind
and overall efficiency will decrease [3]. Typically used Tip speed ratio for
different number of blades are [14]:
Tip speed ratio Number of blades

12 − 14 2
8 − 10 3
5−6 5
2.4 Turbine Power Output

Wind turbine power output varies with the wind speed but is not propor-
tional to it. Wind speed can be divided into three parts Cut-in speed, Rated
output power/wind speed and Cut-out speed.
Figure 2.4: Power output with steady wind speed
Cut-in speed: Minimum wind speed at which, wind turbine blades ex-
perience some torque exerted by the wind and begin rotation for electrical
m
power generation. This wind speed is typically between 3 − 4 sec . After wind
cut in speed, transition region for the power output starts where both rotor
speed and torque must change separately.
Rated output power/wind speed: Electrical power output increases

with the increase in wind speed until it reaches the limit of generator power
output. This generator power output limit is called Rated power output
and corresponding wind speed is known as Rated power wind speed. Wind
turbine design is arranged in a manner to limit the output power to this
maximum level if wind speed increases beyond this rated power wind speed.

Literature Study
Cut-out speed: Wind speed above then rate output wind speed increases
the forces acting on the turbine structure, at some point there is a high risk
of complete collapse or structural damages. Therefore braking system is
activated to stop turbine blades rotation. This is called Cut-out speed and
m
is normally around 25 sec .
2.5 Maximum Power Captured By Wind Turbine

Wind is kinetic form of energy. Power contained in wind can be expressed
as [3]:
1
P ower = ρAv 3 (2.4)
2
where ρ is the density and Av is the volume flow. This relation shows that
power generated by the wind turbine has a cubic relation with the wind
speed which means it is highly dependent on wind speed. Wind turbulent
nature makes the design of wind turbine control quite a difficult task.
To extract the maximum power output, turbine blades interaction with the
wind within the swept area should be increased as much as possible. Theo-
retically by increasing the number of blades, efficiency of wind turbine should
increase as well. But in reality, by increasing number of blades will also in-
crease the interference among the blades which results that blades passes
through the disturbed wind flow region and decreases the overall turbine
efficiency. For structural stability, number of blades of horizontal axis wind
turbine should be greater then or equal to three which makes the dynamic
properties of turbine similar to disk [3]. Force causing change in momentum
comes entirely from the pressure difference across the actuator disk.
According to the Bet’z limit [3], Maximum 1627 or 59.25% of kinetic energy in
wind can be extracted by wind turbine. Maximum achievable coefficient of
power which is the ratio of power extracted to the energy flow through the
rotor swept area in unit time is 0.59. Coefficient of power can be expressed
as [3]:
P
Cp = 1 3
(2.5)
2 ρAv
where P is the power extracted, ρ is density of air, A is the rotor swept

area and v is the wind velocity. Where as in practice maximum achievable
efficiency is about 45 − 50%.

Literature Study
2.6 Loads Of Wind Turbine

Consider the operation of a wind turbine in nominal wind speed, forces are
acting on elements of the turbine e.g., rotor blades, hub, mainframe, and
tower. Turbine blades are experiencing the bending forces due to wind.
These forces acts in two ways: First, edgewise bending which is in the
direction of the rotor plane and second, flap wise bending in perpendicular
direction to the rotor plane. Wind exerts forces on the hub, mainframe
and tower structures as well. These wind forces have two main effects,
particularly on turbine tower: The yaw moment (Myaw ) which is twisting
of tower and the tilt moment (Mtilt ) which is bending of tower.
Figure 2.5: Tilt and Yaw movement of turbine
The Fourier analysis of the forces discussed above, gives the 1p, 2p, 3p...
frequency components of the loads [15, 16]. This frequency is the excitation
frequency of blades as it passes through the tower cause of tower shadowing
i.e. turbulence in the wind due to the tower. When the blades passes
through this turbulence region, it experiences extra load (excitation pulse)
for a short time. This is referred as Excitation frequency.

Literature Study
2.7 Sustainable Control Concept

In a project called Sustainable Control (SusCon) [7] several new algorithms
were developed at ECN offers an integration of wind turbine control, moni-
toring and supervisory control in one concept for efficient operation of wind
turbines. This concept provides reduction of extreme and fatigue loads on
wind turbine by combining four main components:
1. Optimized Feedback Control (OFC)
2. Fault Tolerant Control (FTC)
3. Extreme Event Control (EEC)
4. Optimal Shutdown Control (OSC)
Figure 2.6: Schematic diagram of Sustainable Control concept
Optimized Feedback Control (OFC) consists of methods developed for

fatigue load reduction of wind turbine under normal operational conditions.
Wind turbine cost and up-scaling limitations are reduced by the reduction
of these loads. Individual pitch control (IPC) is a promising option for load
reduction and is also the focus of this thesis work (See next section).
Fault Tolerant Control (FTC) consists of methods for fault detection

and controller reconfiguration to avoid unnecessary standstill due to minor
sensor or actuator failure.
Extreme Event Control (EEC) consists of the methods for energy pro-
duction during extreme wind conditions which increases the heavy loads on
different turbine components and causes unnecessary standstill.

Literature Study
Optimal Shutdown Control (OSC) consists of the method to bring the

turbine at stand still condition to avoid any damage due to some serious
system failure.
2.8 Cyclic Pitch Control

Larsen [17] in 2005, presented a simple technique for load reduction. Method
states that “ Aerodynamic moment created through cyclic pitch variations
which have phase difference of 120◦ , can compensate the mean of yaw and
tilt moments of three bladed wind turbine ”. In this technique, blade flap-
wise bending and edge-wise bending moments are converted from rotating
frame of reference to fixed frame of reference as in below equations.
Mmeani = Mf lapi cos(θi ) + Medgei sin(θi ) (2.6)

B
X
Mtilt = Mmeani cos(θi ) (2.7)
i=1
B
X
Myaw = Mmeani sin(θi ) (2.8)
i=1
where θi is the blade pitch angle for blade i and Mtilt and Myaw are the tilt
and yaw moments of the rotor. Resulting pitch action is the combination of
both θtilt and θyaw will compensate these flap-wise and edge-wise bending
moments of rotor.
θcyclic = θtilt sin(ψi + φ) + θyaw cos(ψi + φ) (2.9)

where ψi is the azimuth angle of blade i and actuation delay in each blade
is compensated by adding a small phase shift φ.
2.9 Individual Pitch Control (IPC)

Primarily pitch control system is used to limit the aerodynamic power output
of turbine at above rated wind speed and to maximize the energy capture
by the turbine at below rated wind speed. Collective pitch control (CPC)
is a prominent option for this purpose but this technique adjust blade pitch
angles of all three blades to a same angle at same time [8, 18, 19]. But
on large wind turbines, wind speed experienced by the each blade at any
instant may differ significantly. So it is desirable to send the different pitch
angle to each blade for the alleviation of asymmetrical loads of wind turbine.
This technique to control the pitching of each blade individually is called
Individual Pitch Control (IPC) of wind turbine.

Literature Study
2.9.1 Load Reduction Through IPC

As the turbine size increases, the stochastic disturbances on the turbine
due to the turbulence of wind also increases. Recent research in this field
is done by Bossanyi [10]. Bossanyi approach is borrowed from three phase
electrical machine theory in which blade load measurements are transformed
into mean and variations on two orthogonal axis i.e. d-q axis representation
(‘direct’ and ‘quadrature’ axes) [20]. Both these orthogonal axes along with
the transformation matrix are used to transform rotation of the blades into
quadrature axes as in below equation.
 
2π 4π
β1
βd 2 cos(θ) cos(θ + 3 ) cos(θ + 3 )  
= β2 (2.10)
βq 3 sin(θ) sin(θ + 2π 4π
3 ) sin(θ + 3 ) β3
Inverse transformation gives back three rotating blades from d-q axis trans-
formation    
β1 cos(θ) sin(θ)
β2  = cos(θ + 2π ) sin(θ + 2π ) βd (2.11)
3 3 βq
β3 cos(θ + 4π3 ) sin(θ + 4π
3 )
This method treats both d and q axis independently which means that
classical single input single output controller can be implemented to the
both axis separately for control action. However, the interaction between
two axis exists in reality which causes a phase mismatch and to counter act
this problem, an azimuthal phase shift is introduced i.e. adding a constant
offset value in the d-q transformation [20].
2.9.2 Higher Harmonics Control

Technique used for reduction of 2p and 3p load is known as Higher harmonic
control. Previously discussed techniques in section 2.8 and 2.9 are used for
1p load reduction. These load components are caused due to the wind speed
variations over rotor plane.
Van Engelen [18] approach used multi-blade coordinate transformation pro-

posed by Coleman and Feingold [21]. This blade coordinate transformation
approach converts the rotating state variables into fixed coordinate system
and then different feedback loops for 2p and 3p effects are provided.
    
θ1 1 sin(ψ1 ) cos(ψ1 ) θ1cm
θ2  = 1 sin(ψ2 ) cos(ψ2 ) θ2cm  (2.12)
θ3 1 sin(ψ3 ) cos(ψ3 ) θ3cm
1 1 1
    
Mf lap1cm 3 3 3 Mf lap1
Mf lap1  =  2 sin(ψ1 ) 2
sin(ψ1 ) 2
3 sin(ψ1 )
 Mf lap2  (2.13)
cm 3 3
2 2 2
Mf lap1cm 3 cos(ψ1 ) 3 cos(ψ1 ) 3 cos(ψ1 ) Mf lap3

Literature Study
where Mf lapi represents blade flap-wise bending moment and ψi represents

Azimuth angle for blade i. Hence control design for each blade coordinate
angle θicm can be implemented.
Fatigue load reduction has two significant advantages
• Allows lighter blade and tower structure design
• offers longer life span of wind turbine
Reduction of these loads offers cost and material saving for blades and tower
and moreover their working time span is also increased due to this reduction.
2.10 Design Filters and IPC Gain

This section describes the current control design technique used at ECN for
turbine load reduction. As this technique is combination of different filters,
therefore it is named as Design Filters and constant gain value associated
with this technique is named as Design IPC Gain value. Design Filters and
Design IPC Gain are used in this thesis work frequently for analysis.
Wind turbine load reduction is based on how smart and efficient is the
control technique. Several different control techniques has been developed
and implemented in ECN. Current control technique uses the turbine blades
in Mass Spring Damper configuration and turbine tower is considered to be
stationary.
Figure 2.7: Turbine Configuration
Turbine model consists of the pitch actuation dynamics of the blades, ne-
glecting the tower dynamics which eliminates the model complexity. There-
fore all blades are working independently without any coupling effect with
tower. Control activity is achieved through a filter design and a constant
system gain value −8.5193e−09 based on the previous experience.

Literature Study
Figure 2.8: Turbine Configuration
This design filter is the combination of five filters, three notch filters at the
3p, 6p and 9p frequency components and two low pass filters to provide
the required roll off factor to make sure that system stays in design limits.
As discussed earlier, frequency components 3p, 6p and 9p are the higher
harmonics of the tower excitation frequency. Excitation frequency is the
impulse signals felt by tower, whenever a blade passes through it. 1p, 2p
and 3p frequency components occur due the three blades motion and tak-
ing the fourier transform of these components gives the higher frequency
components.
This current design works efficiently for the small wind turbines. But for
large wind turbine with power rating 3 − 7.5M W , tower dynamics cannot
be neglected as it has significant effect on the turbine control system design.
As this thesis work is focused on large wind turbine, therefore a detailed
mathematical model is required which includes turbine blade and tower
dynamics. Then this detailed model will be analysed with current control
design and with a new proposed approach. Detailed mathematical model
design is the next step of this thesis work.

Chapter 3
Modelling of Wind Turbine
3.1 Introduction
Wind turbine is composed of interconnected flexible bodies which undergo
translational or rotational displacement due to the wind forces. Multibody
system approach [5, 6] is followed for the modelling of dynamic behaviour
of wind turbine. This comprehensive model includes all turbine movements
due to the wind force and yields all the useful sensor outputs like flap-wise
bending moments of blades (i.e. bending of blades out of rotor plane), blades
velocities, tower bending and pitch angle of the blades. All these sensor
outputs enables us to implement a feedback loop and control strategy to
adjust the IPC loop gain at optimal value for maximum load reduction.
Equations of motion for turbine dynamics presented in this section are de-
rived form the previous turbine model used in ECN. This ECN model is
presented in a technical report named as ACT Model Calculations [22]. My
contribution in this modelling part is to linearise these equations by solv-
ing them simultaneously and to form a statespace model of these linearised
equations. End results for this simultaneous solution of equations are pre-
sented in this Chapter while complete mathematical calculation is described
in Appendix A.
After having a discussion with University Professor, a further improvement

into this turbine model is also suggested in the Future Work portion of this
thesis work.
3.2 Multibody System

Multibody system is an approach used for the modelling of dynamics of
interconnected bodies due to their linear or angular displacement [5, 23,
24]. These dynamics occur due to the rate of change of momentum of
interconnected bodies caused by external force. In our case , wind turbine is

also considered to be the combination of rigid bodies connected with single

joint and dynamics of wind turbine occur due to the wind forces acting
across the turbine structure.
Figure 3.1: Turbine with blade and tower hinge points
Figure 3.1 shows the schematic diagram of wind turbine. Dotted part shows
the bending of blades and tower due to the wind force. Turbine blades
are similar to the aircraft propellers. Thick portion close to the rotor and
thickness decreases on other end. Bending of turbine blades out of rotor
plane is known as Blade Flapwise Bending and the point from where blade
bends is called Blade Hinge Point. Turbine tower also have similar geometry
where thick portion is close to the base/foundation and thickness decreases
in upward direction. The point from where turbine tower bends due to the
wind forces is called Tower Hinge Point.
Blade construction considered in this model is the combination of two parts.

Blade part attached to the rotor is thicker and rigid i.e. this part doesn’t
bend with the wind forces and other part is attached to this thicker part
which actually bends due to the wind forces. Point between these two blade
parts is Hinge Point. For the sake of simplicity, only one hinge point is
considered for blade modelling. As the bending moment is based on the
bending portion of blade, therefore point of centre of gravity was calculated

for this specific portion of blade to form the equation of motion.
Figure 3.2: Schematic diagram of blade with hinge point and complete
layout
Figure 3.2 shows the construction of turbine blades. Two blade parts are
shown separated by the hinge point. Blade centre of gravity Rcog and the
point at which the wind force is acting Rae is also shown in the line diagram
below. This line diagram shows that the blade bending is just considered to
be the bending at hinge point rather than the curved bending of whole blade
for sake of simplicity. Therefore the reaction force which is actually the joint
force caused due to the bending of blade is also considered to be at hinge
point location in opposite direction of wind force. Further the kinematics of
turbine blades are based on Mass Spring Damper configurations described
in next section.
3.3 Mass Spring Damper Configuration

A detailed mathematical model is developed in which blades and tower of the
turbine are considered to be in Mass Spring Damper (MSD) configuration
to describe all the kinematics of the structure as shown in figure below.

Figure 3.3: Schematic diagram of MSD configuration of turbine
Ideally in Mass Spring Damper configuration, a mass (m) with spring con-
stant (k) and damping coefficient (c) are subjected to an oscillatory force.
Fs = −kx
where x is the distance covered by the mass due to the force.
Damping is an effect that reduces the amplitude of oscillations in an oscil-

latory system and is linearly related to the velocities of the oscillations.
Fd = −cv = −cẋ
Treating mass (m) as a free body and applying Newton’s second law of
motion
Ftotal = ma = mẍ
For Mass Spring Damper configuration:
Ftotal = Fs + Fd
Ftotal = −kx − cẋ
mẍ = −kx − cẋ
k c
ẍ = − m x− m ẋ
All equation of motion for blade and tower are calculated based on this
configuration. Then these equation are linearised to form a complete model
calculation.

3.3.1 Blade Motion

In this case, blades are experiencing the wind force at a point which is
considered to be the 23 of the total blade length. This force not only rotates
the blades around the rotor axis but also bend it in backward direction.
Hinge point mentioned in the figure 3.2 is the point after which blade is
considered to be flexible, below then this point blade is hard enough to with
stand the wind forces without any bending. Blade stiffness is defined as sb ,
blade damping as db . As the blades rotates around rotor axis so mass due to
the inertial moments of the blades relative to the centre of gravity is defined
as Jb .
So the equation of motion for rotation of blade is the function of reaction

force of the joints opposite to the wind force and blade azimuth angle.

Jb θ̈b = −sb θb −db θ̇b +sb cos ψθt +db cos ψ θ̇t −db ω sin ψθt +Fj Rcog +Fax Rae −Rcog
(3.1)
Equation of motion for blade displacement is the function of actual force
acting on blades due to wind and the reaction force exerted by the joint in
opposite direction, which is given by:
mb ẍb = Fax − Fj (3.2)
3.3.2 Tower Motion

Tower motion has no direct relation with the actual wind acting on the
blades but with the sum of all the three joint forces acting in the opposite
direction of wind force and with tilt moment of the blades.
Mtilt X
mt ẍt = −st xt − dt ẋt + 2 + Fj (3.3)
3H i
where Mtilt is the sum of joint forces Fj of three blades and moments of
joints Mj " #
X
Mtilt = Fj Rb − Mj cos(ψ) (3.4)
i
and moment of joint Mj is:
Mj = −sb θb − db θ̇b + sb cos ψθt − db ω sin ψθt + db cos ψ θ̇t (3.5)
3.3.3 Blade Velocities

Sensor outputs for the blade velocities can be calculated on the basis of
blade speed equation which is a function of blade angular motion and tower
displacement.
yb = xt + θb Rae + θt cos ψRb (3.6)

Taking the derivative of the equation and by using the azimuth angles,
velocities of all three blades can be extracted.
vb = ẋt + θ̇b Rae + θ̇t cos ψRb − θt ω sin ψRb (3.7)
3.3.4 Blade Flap-wise Bending Moments

Flap-wise bending is bending of the blade out of rotor plane due to wind
force. Flap-wise bending moments of blades are the function of joint mo-
ments for all three blades and the reaction force applied by the joints in the
opposite direction to the wind force which is given by:
Mf lap = Rb Fj − Mj
Putting the value of moments of joint Mj (as defined earlier) to get the final
equation.
Mf lap = sb θb + db θ̇b − sb cos ψb θt − db cos ψb θ̇t − db ω sin ψb θt + Rb Fj (3.8)
3.3.5 Boundary Condition

“A condition which a quantity that varies throughout a given space or en-
closure must fulfil at every point on the boundary of that space is called a
boundary condition” [25].
In this case, motion of blades will always be the function of tower motion
and the rotational motion of blades. Hence it is the boundary condition for
the model as well.
xt
xb = xt + cos ψRb − θb Rcog (3.9)
Lt
3.3.6 Joint Force

As the wind force try to bend at the joint, in reaction their will be a force
which will act opposite to the wind force. This force of blade joint is called
Joint force Fj and is given by:
Fj = Fax − mb ẍb (3.10)

3.4 Linearised Model Calculations

This Section gives the linearisation of the mathematical equations of blade,
tower, joints etc. Linearisation of these equations first requires the multi-
blade coordinate transformation which helps to integrate the dynamics of
individual blades and express it in a non rotating frame [26]. This multiblade
coordinate transformation is widely used in helicopter industry for stability
and control purpose. In wind turbine field, Coleman and Feingold research
in field of Theory of self-excited mechanical oscillations of helicopter rotors
with hinged blades” [21] is used for wind turbine coordinate transformation.
3.4.1 Coleman Transformation

“Coleman Transformation Method transforms all quantities on the rotating
reference frame to a fixed reference frame and results into a linear time
invariant (LTI) model 1 , which is easy to handle and very suitable for
implementation of different control strategies” [21].
 
1 sin(ψ1 ) cos(ψ1 )
CM = 1 sin(ψ2 ) cos(ψ2 )
1 sin(ψ3 ) cos(ψ3 )
where angle(ψ) is the Azimuth angle of the blade which is defined as:
Azimuth Angle: “ Azimuth angle of the blade is the product of rotational

speed and time. Zero azimuth angle is considered when blade is in vertically
upward direct and is positive in clockwise direction”
For a blades turbine: ψ1 = ψ, ψ2 = ψ + 120◦ , ψ3 = ψ + 240◦ ,
So,
1 1 1
 
3 3 3
CM −1 =  32 sin(ψ) 2
3 sin(ψ + 2π
3 )
2
3 sin(ψ + 4π
3 )

2 2
3 cos(ψ) 3 cos(ψ + 2π
3 )
2
3
4π
cos(ψ + 3 )
In order to translate the blade coordinate from rotating domain to Coleman

domain 2 , following relation will be used:
       
θ1 h i θcollective θcollective h i θ1
θ2  = CM  θyaw  <=>  θyaw  = CM −1 θ2  (3.11)
θ3 θtilt θtilt θ3
1
See Appendix B for Coleman Transformation example
2
See Appendix A for calculations of blade coordinate translation into Coleman domain

Transformation of whole system into the Coleman Domain gives three main
components:
• Collective : Collective component describes the rotational motion of
the blades.
• Yaw : Yaw component describes the twisting of turbine tower due to

the uneven loads on turbine blades.
• Tilt : Tilt component describes the turbine fore-aft motion due to

wind forces.
From above relation we can write:
h i
θcm = CM −1 θ (3.12)
So,
1
θcol = θ1 + θ2 + θ 3 (3.13)
3
2
θyaw = sin ψ1 θ1 + sin ψ2 θ2 + sin ψ3 θ3 (3.14)
3
2
θtilt = cos ψ1 θ1 + cos ψ2 θ2 + cos ψ3 θ3 (3.15)
3
Taking derivative:
1
θ̇col = θ̇1 + θ̇2 + θ̇3 (3.16)
3
2
θ̇yaw − ωθtilt = sin ψ1 θ̇1 + sin ψ2 θ̇2 + sin ψ3 θ̇3 (3.17)
3
2
θ̇tilt + ωθyaw = cos ψ1 θ̇1 + cos ψ2 θ̇2 + cos ψ3 θ̇3 (3.18)
3
3.4.2 Linearisation
Linearisation is carried out by solving all the equations of motion simultane-
ously. This linearisation is implemented after transforming the whole system
into the Coleman domain through Coleman transformation technique.
Rotational motion for all three blades depending on azimuth angles can be
defined as:
Jb θ̈b1 = −sb θb1 − db θ̇b1 + sb cos ψ1 θt + db cos ψ1 θ̇t − db ω sin ψ1 θt

+ Fj R1cog + Fax Rae − Rcog



As the blades are on the rotating frame of reference. Therefore first thing is
to transform them into fixed reference of frame through Coleman transfor-
mation. After further simplification of the blade equations and implementing
the Coleman transformation we get:
 
h i θ̈b1
CM −1 θ̈b2  =
θ̈b3
 
xt
 ẋt 
 
  θ 
0 0 − 13 sb − 31 db − 31 sb − 13 db − 31 sb − 13 db  b1 
1  db θ̇ 
b1 
−
 Lt ω 0 − 3 sb Sψ1 − 3 db Sψ1 − 3 sb Sψ2 − 3 db Sψ2 − 23 sb Sψ3
2 2 2 2 2
− 3 db Sψ3  

Jb θ
 
sb db b2 
− 32 sb Cψ1 − 23 db Cψ1 − 23 sb Cψ2 − 32 db Cψ2 − 23 sb Cψ3 − 32 db Cψ3  

Lt Lt θ̇b2 
 
θb3 
θ̇b3
   
Rcog h i Fj,b1 1 h i Fax,b1
CM −1 Fj,b2  + Rae − Rcog CM −1 Fax,b2  (3.19)

+
Jb Jb
Fj,b3 Fax,b3
where S and C represents sin and cos respectively.
Now translating blade equations into Coleman domain to eliminate the az-
imuth dependency as explained in Section 3.4.1.
Blade dynamics equations:
sb db mb Lb Rae
θ̈col = − 2
θcol − 2
θ̇col − 2
ẍt + 2
Faxcol
Jb + mb Rcog Jb + mb Rcog Jb + mb Rcog Jb + mb Rcog
(3.20)

2 )ω 2 − s
(Jb + mb Rcog
db ω 2mb Rcog Rb ω b
θ̈yaw = − 2
x t + 2
ẋ t + 2
θyaw
Lt (Jb + mb Rcog ) Lt (Jb + mb Rcog ) Jb + mb Rcog
db db ω Rae
− 2
θ̇yaw + 2
θtilt + 2ω θ̇tilt + 2
Faxyaw
Jb + mb Rcog Jb + mb Rcog Jb + mb Rcog
(3.21)
sb + mb Rcog Rb ω 2 db db ω
θ̈tilt = 2
xt + 2
ẋt − 2
θyaw
2 )ω 2 − s
(Jb + mb Rcog b db
− 2ω θ̇yaw + 2
θtilt − 2
θ̇tilt
Jb + mb Rcog Jb + mb Rcog
mb Rcog Rb Rae
− 2
ẍt + 2
Faxtilt (3.22)
Equation for the motion of tower is defined as:
Mtilt X
mt ẍt = −st xt − dt ẋt + 2 + Fj
3H i
Placing
P in values of turbine tilt moment Mtilt and the blade joint forces
i Fj for all three blades and then translation equation into Coleman do-
main as well to eliminate the azimuth dependency.
Tower dynamics equation:

" !
9 mb Rb2 sb + mb Rb2 ω 2 db
mt ẍt = −st xt −dt ẋt + Rb Faxtilt − ẍt − xt − ẋt −db ωθyaw
4H Lt Lt Lt
! #
+ 2mb Rcog Rb ω θ̇yaw + sb − mb Rb Rcog ω 2 θtilt + db θ̇tilt − mb Rb Rcog θ̈tilt
" #
+ 3 Faxcol − mb ẍt − mb Rcog θ̈col (3.23)
As from the above equations, it is clear that both Blade and Tower dynamic
equation are dependent on each other. Next step is to solve the both blade

dynamics and tower dynamics simultaneously to get the linear equations for
the both blade and tower dynamics. Complete calculations are presented
into the Appendix A. Here just the final equations which are in matrix form
are presented.
Linearised Blade Dynamics

 
xt " #
Xblade  ẋt  Yblade
θ̈bcm = θbcm  + Zblade Faxcm
  (3.24)
Zblade
θ̇bcm
Xblade Yblade
where Zblade and Zblade represents the matrices for blade dynamics.
Linearised Tower Dynamics

 
xt " #
 ẋt 
ẍtcm = Xtower 
θbcm  + Ytower Faxcm
 (3.25)
θ̇bcm
where Xtower and Ytower represents the matrices for tower dynamics.
(See Appendix A for complete blade and tower dynamics calculation)
Linearisation of the sensor outputs for the blade velocities are given by:
vb = ẋt + θ̇b Rae + θ̇t cos(ψ)Rb − θt ω sin(ψ)Rb (3.26)
Using above equation for the velocities of all three blades and applying the
Coleman transformation:
 
  ẋt + 31 Rae θ̇1 + θ̇2 + θ̇3
vcol  
− ωR
vyaw  =  b
x + 2
R sin ψ θ̇ + sin ψ θ̇ + sin ψ θ̇ (3.27)

Lt
t 3 ae 1 1 2 2 3 3 
vtilt
 
Rb 2
Lt ẋt + 3 Rae cos ψ1 θ̇1 + cos ψ2 θ̇2 + cos ψ3 θ̇3

Similarly translating the equation of blade velocities into the Coleman do-
main to eliminate the azimuth dependency
Linearised Blade Velocities

 
xt
 ẋt 
vcm = Xvcm 
θbcm 
 (3.28)
θ̇bcm
where Xvcm represents the matrix for the equations of blade velocities.
(See Appendix A for complete blade velocities calculation)
Flap wise bending moments of blades are the function of joint moments for
all three blades and the reaction force applied by the joints in the opposite
direction to the wind force. Linearisation of the sensor output for blade
flap-wise bending moments after some simplification is given by:
" #
sb cos(ψ) + db ω sin(ψ) db
Mf lap =− xt − cos(ψ)ẋt + sb θb + db θ̇b + Rb Fj
Lt Lt
(3.29)
Similarly implementing the Coleman transformation on all the three equa-

tion for each blade flap-wise bending moment and then translating those
equation into the Coleman domain to eliminate the azimuth dependency
Linearised Flapwise bending moments

 
xt " #
 ẋt 
Mf lapcm = Xf lap 
θbcm  + Yf lap Faxcm
 (3.30)
θ̇bcm
where Xf lap and Yf lap represents the matrices for the equations of blade
flap-wise bending moments.
(See Appendix A for complete blade flap-wise bending moments calculation)

After calculation of all the linearised equations, there are the two most
powerful ways to represent a system.
• Statespace form
• Transfer function form
Both techniques are commonly utilized for the system modelling [27]. While
statespace form have several advantage over transfer function form:
• Statespace form provides better insight into the system behaviour like
system controllability and observability.
• Statespace form is not limited to linear and time invariant systems.
• Calculation of the system response with initial condition case is very

simple with statespace form
• Statespace form best suits for theoretical and numerical calculations

for control system analytical and optimization solutions.
3.5 Statespace Representation

State-space representation is used for the mathematical representation of
a physical system. This representation is the replica of the actual system
but less complex than actual system [28, 29]. It provides the liberty to
implement and examen different strategies to get the desired outcome from
the system without any involvement of the actual system. That minimizes
risk, cost and time consumption.
3.5.1 Model Description

State space modelling of the wind turbine involves all the mathematical
calculations from the previous section. It has three main parts:
• Input Side : Wind force Fax will be the input of the system. Wind
force acting on the blades consists of the force due to the free flow
of wind, structural motion of the turbine uax and the blades angular
motion θb (pitch angles).
∂Fax ∂Fax
Fax = δθ + δuax (3.31)
∂θ ∂uax
where
δuax = δuundisturbed + δudue to structural motion (3.32)

Figure 3.4: Turbine Model
• System States : System states are the minimum set of variables

that fully describe the system response to any input signal. In our
case blade and tower dynamics calculation will be the states of the
system.
• Output Side : All sensor outputs for the blade velocities and bade
flap-wise bending moments will be the output of the system. Blade
displacement, tower flapping and all the pitch angle can also be calcu-
lated with a bit more mathematical calculations.
Figure 3.4 shows the complete statespace model configuration of the turbine
model with two feedback loops. One feedback loops provides the structural
dynamics of the system given by the blade velocities and second feedback
loop provides the information for the blades angular motion information.
Filter block mentioned into the figure represents the Design filters used for
the system analysis with current control design technique and IPC gain
block contains the Design IPC Gain value i.e. −8.5193e−09 from the current
control design of ECN.

Order of System: In statespace representation, minimum number of vari-

ables which can completely describe the system response are called States
of the system. Our system has total eight states. Two explaining tower
dynamics and six blade dynamics for the collective, tilt and yaw motion of
turbine. Therefore order of this system is eight.
Velocities Feedback Loop: Reason of making blade velocities feedback

loop can be explained by a simple example. If a person riding a bike will
experience more wind force due to the motion as compare to the person
standing still in the same location. Similarly when the turbine blades ro-
tates, it experience more wind force due to velocity of blades in comparison
of the wind force experience by the same blades in stand still condition. Ve-
locity feedback loop is developed to compensate this additional wind force
experienced by the blades due to their velocities.
Blade Flapwise Bending Moments Feedback Loop: Blade flapwise

bending moments are translated back in to the blade pitch angles by inverse
Coleman transformation. Integrator is added for two main purposes:
• To select only yaw and tilt moment. Reason is that the collective part
is the driving force for the rotational motion of the blades but the yaw
and tilt moments, as discussed earlier, are the source of turbine fatigue
loading.
• To avoid the offset in yaw and tilt component and force them to start
from common zero point which is critical for the system analysis.
Then control technique is implemented for the maximum load reduction
using the blade pitch angles with a loop gain value. Output of the con-
trol system is transformed back into the Coleman domain to complete the
feedback loop.
3.5.2 Matrix Representation

Matrix formation for the model is represented here.
Input Side:

ẋ(t) = A x(t) + B u(t)
 
xt
 ẋt 
 
θ 
col   

 Faxcol
Xtower  θ̇col  Ytower 

ẋ(t) =  + Faxyaw  (3.33)
θ
Xblade  yaw  Yblade

Faxtilt
θ̇yaw 
 
 
 θtilt 
θ̇tilt

Matrix A: System state matrix A is a 8 × 8 matrix and is composed of the

tower and blade dynamics matrices represented as Xblade and Xtower .
• Xblade is calculated by solving equations from Appendix: A.56,A.57,A.58,A.60.
• Xtower is calculated by solving equations from Appendix: A.60,A.61,A.62,A.63.
Matrix B: Similarly Matrix B is 8 × 3 matrix for the system input and is

composed of Yblade and Ytower
• Yblade is calculated by solving equations from Appendix: A.56,A.57,A.58,A.60.
• Ytower is calculated by solving equations from Appendix: A.60,A.61,A.62,A.63.
As these equations were very big and Matlab was used to form matrix A
using these equations. Complete equation calculations are provided in Ap-
pendix A.
Output Side:

y(t) = C x(t) + D u(t)
 
xt
 ẋt 
 
θ 
col   

 Faxcol
Xvcm  θ̇col  0

y(t) =  + Faxyaw  (3.34)
Xf lap θyaw  Yf lap

Faxtilt
θ̇yaw 
 
 
 θtilt 
θ̇tilt
Matrix C: Sensors output matrix C is 6 × 8 matrix composed of Xvcm and

Xf lap , provides the blade velocities and flapwise bending moments.
• Xvcm is calculated by solving equations from Appendix: A.76,A.77.
• Xf lap is calculated by solving equations from Appendix: A.84,A.85,A.86.
Matrix D: Matrix D is 6 × 3 matrix composed of Yvcm and Yf lap .
• Yvcm is zero because blade velocities have no effect on blade angular

motion.
• Yf lap is calculated by solving equations from Appendix: A.84,A.85,A.86.
Similarly calculation for this equations also involved Matlab because of there
size. Complete equation calculations are provided in Appendix A. Above all
equations describes the Statespace model based on the mathematical model
designed for the turbine.

3.6 Model Constraints

Number of constraints were considered in this project to maintain the project
size and to meet the time limitation for completion of work. These con-
straints includes:
• Pitch actuation is not limited for the simplicity in this project. In real
world scenario, maximum achievable pitch actuation is 2 rad sec .
• Wind model designed by the ECN is used for the model input.
Figure 3.5: Wind Model Designed at ECN
• Previous calculations for force due to structural dynamics and force

due to pitch angles is used for model calculations.
∂Fax ∂Fax
Fax = δθ + δUax
∂θ ∂Uax
where δUax is the sum of free flow wind speed and wind due to struc-
tural motion of turbine.
δUax = δUun + δUstr
Partial derivative form of aerodynamic forces in terms of pitch angles

at the equilibrium point p̄ is given by:
(i)
∂Fax 1
= ∇θ Fax (p̄) = ρπR2 (u(i) 2 (i) (i)
ax ) CT (λ , θ )
∂θ(i) 2B
Partial derivative form of aerodynamic forces in terms of blade struc-
tural dynamics at the equilibrium point p̄ is given by:
(i)
∂Fax 1 2 (i)
h
(i) ∂CT (λ(i) , θ(i) ) i
= ∇ u Fax (p̄) = ρπR uax 2C T (λ̄ , θcol ) − λ̄
∂u(i)ax 2B ∂λ(i)

Chapter 4
Model Analysis
4.1 Stability Analysis

This section gives the reader a view of the system response through some
analysis techniques. After completing the turbine modelling task, first thing
is to analyse the stability of the designed model. Stability analysis has been
carried out through Pole-Zero calculations for both open loop and close loop
model configurations. System stability results based on Pole-Zero plotting
are also verified through time and frequency domain analysis. Linearised
stable model is necessary to implement the multivariable control technique
for efficient turbine load reduction.
4.1.1 Open Loop System

For open loop system, model is considered without any feedback loop. Model
with the wind force as the system input which is a combination of force due
to free flow of wind and force due to the structural dynamics of turbine
blades. Blade velocities and blade flapwise bending moments are the model
outputs.
Figure 4.1: Open Loop System

Model Analysis
Figure 4.2: Pole Zero Plot for Open Loop System
For Pole Zero plot Matlab command is utilized i.e.

pzmap(system)
This command creates a Pole Zero plot of the continuous- or discrete-time
dynamic system model. The poles are plotted as x’s and the zeros are plotted
as o’s.
Pole Zero plot in figure 4.2 shows the location of poles in the complex plane.
It is clear that system poles lie on the left hand side of the plane which con-
firms the system stability. As the poles of the system are complex conjugate
pairs, system response from these poles will be decaying sinusoid form. Rate
of decay in system response is determined by the locations of the poles. Poles
of the system far from the origin correspond to the rapid decaying compo-
nents while the poles near to the origin correspond to the slowly decaying
component. Poles closer to the origin are the dominant poles of the system
since there contribution takes the longer time to die out from the system
response. Main task is to check the stability of system which enables the
implementation of the multivariable control technique.
4.1.2 Velocities Feedback Loop

Velocities feedback loop as discussed in previous section, is developed to
compensate this additional wind force experienced by the blades due to

Model Analysis
blade velocities. Pole-Zero plotting for the system feedback loop provides
the following response:
Figure 4.3: Blade Velocities Feedback Loop
Figure 4.4: Pole Zero Plot for System with Blade Velocities Feedback Loop
Pole Zero plot for the system with velocities feedback loop in figure 4.4
shows that system remains into the comfortable region and all poles of the
system still lie on the left hand side of the origin. Only difference made by

Model Analysis
the feedback loop is that it further pushes away the location of the poles
from the origin on the left hand side of the plane which will increase the
rate of decay in system response but system will still maintains stability.
4.1.3 System With Blade Velocities and Flapwise Bending

Moments Feedback Loops
Complete turbine model have two feedback loops, as discussed earlier, one
for blade velocities which compensate the effect of addition wind force expe-
rienced by the blades due to their velocities and second for the blade flapwise
bending moments i.e. bending of blades out of rotor plane. These flapwise
bending moments are translated back to the blade pitch angles and control
strategy is implemented on this loop for turbine load reduction.
Figure 4.5: Turbine Model
Figure 4.5 shows the complete model configuration of the turbine model
with two feedback loops. One feedback loops provides the structural dy-
namics of the system given by the blade velocities and second feedback loop
provides the information for the blades angular motion information. Filter
block mentioned into the figure represents the Design filter used for the sys-
tem analysis and IPC gain block contains the Design IPC Gain value i.e.

Model Analysis
−8.5193e−09 from the current control design. Designed model is analysed

with the current control design used in ECN.
Figure 4.6: Pole Zero plot for Complete Model with and without Design
Filters
Figure 4.6 shows the Pole-Zero plot for the system for two configurations.

Model Analysis
With Design Filter means the model which includes the Design Filters (Cur-
rent control Design) for analysation and Without Design Filter means the
model in which no controller is implemented, a simple feedback loop for the
blade bending moments are translated back into the blade pitch angles to
complete the loop. This comparison enables the model designer to the have
a better insight of the system response to investigate the Current Control
Design.
Pole Zero plotting for model with design filters and without design filters
configuration shows stable system with all poles lying on the left hand side
of the origin. Prominent effect of pole zero cancellation can be observed into
the system response. Therefore the overall effect of response oscillations will
be reduced. This effect will be observed in to the time domain plotting of
system for impulse or unit step input.
4.1.4 Why To Make Analysis Based On Model Configuration

With and Without Design Filters ?
This section gives the reader a clear view of reason behind the system anal-
ysis based on with and without Design Filters configurations and variation
of Design IPC gain value. Both these Design Filters and Design IPC gain
values are adapted from the current control technique utilized in ECN for
turbine load reduction. This analysis gives the critical reason to implement
mutlivariable control technique for turbine load reduction instead of this
current filters based technique.
Stability analysis in previous section gives the stable model in all three
main configurations. Next step is to check that how much of the change this
system can withstand. In other words, robustness of the system. Robustness
of a system is the ability of system to resist change without adapting it’s
initial stable configuration.
Varying Gain
For the Robustness analysis, system response is analysed with varying the
design IPG gain value within a range of ± five times the design gain value
i.e. 5 ÷ IP Cgain ←→ 5 × IP Cgain. This gain range gives a clear view of
system response with the different gain values. Matlab is utilized to perform
this task and 100 equally distance values within this range are kept into the
account. Taking the Pole-Zero plot on each gain value gives the following
output.

Model Analysis
Figure 4.7: Varying Gain with and without Design Filters
Figure 4.7 on the left hand side shows the Pole-Zero plot of Model with
Design Filter Configuration (means including current control design) and
Figure on right hand side shows the Pole-Zero plot of Model without Design
Filter Configuration(Means current control design is not included in model).
Figure 4.7 gives some unexpected results:
• Poles of the system with design filters gives unstable response with
higher IPC gain values.
• Poles of the system without design filters shows better robustness i.e.
stable response with higher IPC gain values.
To have a clear view of the system response, a predefined code in ECN has
been utilized for the system pole sorting. Purpose of this code is to show
the shifting of system poles from their original position to a new position
due to any change within the system. Sorting of the system poles for the
Model with Design Filters Configuration is shown in below figure 4.8. It can
be observed that increase in IPC gain value will make the poles to shift into
the unstable region.

Model Analysis
Figure 4.8: Poles Sorting
Due to such system response, this point onward all system calculations has
been made with design filters and without design filters configurations. That
will enable us to make final conclusion regarding these design filters. This
system instability is highly undesirable because in real world scenario, con-
troller IPC gain value will keep on changing to efficiently cope up with the
effect of blade flapwise bending due to wind turbulent nature.
4.2 Time Domain Analysis

In this section, system stability is evaluated in the time domain analysis and
is compared with the Pole-Zero plotting in previous sections for verification.
Model used for time domain analysis have the two same configuration, with
and without Design Filters. Note that the IPC gain is kept constant at
Design gain value i.e. −8.5193e−09 . Gain variation is not considered in
this analysis. Frequency domain analysis gives better clarity of IPC gain
variation in next section.
Time domain analysis is the change in response of a system with respect to

time. Time domain analysis of the system consists of the response of system
to unit impulse and unit step input.

Model Analysis
Figure 4.9: Impulse Response with and without Design Filters
Figure 4.9 shows the impulse response of Model with Design IPC loop gain
value and in both with and without Design filter configuration. Blue line
for the response with Design Filters configuration and Green line for the
response without Design Filter configuration. Output for the blade bending
moments is shown in three separate plots which are Collective, Yaw and Tilt
moments of the turbine blades.
Figure 4.9 shows the response of the system as a function of time with a unit
impulse signal at the external input. It can be observed that system starts
from zero initial state and gets to the stability with unit impulse input.
Three plots shows the response of collective, yaw and tilt moment outputs
of the MIMO system respectively. Signal oscillations are very low due to
the pole zero cancellation and big oscillation in response at start is due to
the dominant poles close to the origin but afterwards signals get to the zero
reference which confirms the system stability

Model Analysis
Figure 4.10: Unit Step Response with and without Design Filters
Figure 4.10 shows the step response of Model with Design IPC loop gain
value and in both with and without Design filter configuration. Blue line
for the response with Design Filters configuration and Green line for the
response without Design Filter configuration. Output for the blade bending
moments is shown in three separate plots which are Collective, Yaw and Tilt
moments of the turbine blades.
Response of the system with unit step input can be seen in figure 4.10.
Oscillation in start of response is due to the dominant poles of system and
afterwards all outputs converge to the zero reference which confirms the
system stability with the step input. Obvious constant offset can be seen
in the response of collective component output which is the main driving
component of blade rotational motion. Yaw and tilt moment outputs also
shows stable response with step input. System with design filter (Blue line)
and system without design filters (Green line) both shows stable response.

Model Analysis
4.3 Frequency Domain Analysis

This section gives the reader a clear view of the system stability analysis
in frequency domain. As discussed earlier in Pole-Zero plots that Model
becomes unstable at higher gain values, to verify this conclusion three con-
figurations are taken into account:
• Model without Design Filters and Design IPC gain value
• Model with Design Filters and Design IPC gain value
• Model with Design Filters and five times the Design IPC gain value
(i.e. 5 × IP Cgain)
MIMO Nyquist Stability Criteria has been utilized for frequency domain
analysis.
4.3.1 MIMO Nyquist Stability Criteria

Nyquist criteria does not work with close loop system. First thing was to
make the model an open loop system. Following configuration shown in
figure 4.11 has been utilized in which loop for the blade bending moments
is kept open to implement Nyquist stability criteria.
Figure 4.11: Model Configuration for MIMO Nyquist Stability Analysis

Model Analysis
As discussed earlier, three configurations with Design Filters, without Design

Filters and Design Filters with five time the Design IPC gain is utilized.
Following Matlab commands are used to form these three configurations:
temp sys = Statespace(4 : 6, 4 : 6) ∗ IP Cgain ∗ Intg ∗ F ilter;

temp sys2 = Statespace(4 : 6, 4 : 6) ∗ IP Cgain ∗ Intg;
temp sys3 = Statespace(4 : 6, 4 : 6) ∗ (IP Cgain ∗ 5) ∗ Intg ∗ F ilter;
Theorem : The close loop system with loop transfer function L(s) and
negative feedback is stable if and only if the Nyquist plot of det(I + L(s))
[15, 30]
• makes poles anti-clockwise encirclements of the origin.
• does not pass through the origin.
To implement this theorem to our system, first a frequency range is selected

i.e. −1 −→ 3 radsec and then taking the 1000 discrete points into this range
using Matlab. This enables us to calculate the frequency response of the
system for each discrete value by running a loop. Reason for selecting low
frequency range is the visibility of the plot because at higher frequencies it
is very difficult to analyse the direction of encirclement around the origin.
Following code is utilized for this purpose:
tempdata = f reqresp(temp sys, logspace(−1, 3, 1000));

x = 1 : 1000;
f or t = 1 : size(tempdata, 3)
G(t) = det(eye(3) + squeeze(tempdata(:, :, t)));
end
tempdata2 = f reqresp(temp sys2, logspace(−1, 3, 1000));
f or t = 1 : size(tempdata2, 3)
G2(t) = det(eye(3) + squeeze(tempdata2(:, :, t)));
end
tempdata3 = f reqresp(temp sys3, logspace(−1, 3, 1000));
f or t = 1 : size(tempdata3, 3)
G3(t) = det(eye(3) + squeeze(tempdata3(:, :, t)));
end
Matlab Command H = f reqresp(system, w) returns the frequency response

on the real frequency grid specified by the vector w. In our case this vector

Model Analysis
is composed of low frequency discrete (1000) values within range of −1 −→

3 rad
sec . Function det(I + L(s)) is calculated for all three configurations by
running the FOR loop for each configuration separately.
For clarity graph is plotted into two portions
• First to check the encirclements around origin (Above portion of figure

4.12).
• Second with only positive half cycle to check the direction of the en-
circlements around origin (Below portion of figure 4.12).
Following Matlab Code is utilized for plotting purpose:
f igure, subplot(2, 1, 1),

plot(real(G), imag(G)) hold on
plot(real(G), −imag(G)) hold on
plot(real(G2), imag(G2),0 r0 ) hold on
plot(real(G2), −imag(G2),0 r0 ) hold on
plot(real(G3), imag(G3),0 g 0 ) hold on
plot(real(G3), −imag(G3),0 g 0 ) hold on grid on
title(0 M IM O N yquist Stability Criteria0 )
subplot(2, 1, 2),
plot(real(G), imag(G)) hold on
plot(real(G2), imag(G2),0 r0 ) hold on
plot(real(G3), imag(G3),0 g 0 ) hold on grid on
xlabel(0 Only P ositive Real and Imaginary P arts to Check Direction of Encirclements0 )
Where first portion of the code plots the complete (Positive and negative
half) cycle to show the encirclements around the origin which enables us to
check weather it is passing through the origin or not. Second portion of the
code plots only the positive half cycle which enables us to determine the
direction of the encirclements weather it is clock wise or anti clock wise.

Model Analysis
Figure 4.12: Nyquist Plot of Complete Model with and without Design
Filters
It is clear form the figure 4.12 that all three configurations satisfies the
first requirement, as all three encircles the origin point. But the system
with and without Design Filters with Design IPC gain (Blue and Red line
respectively) have anti-clock wise encirclements around origin which means
that both satisfy the second requirement as well and are stable systems.
But graph of system with higher IPC gain value (Green Line) is unstable,
as it has the clock wise encirclement around the origin. This result verifies
our previous argument that this current control design technique becomes
unstable with higher gain value and in other words system with current
control design cannot withstand internal gain variation which is undesirable
for the case of large wind turbines where the controller gain variation will
be the key factor for the turbine efficient load reduction.

Model Analysis
4.4 Model Fatigue and Pitch Activity Calculations

Focus of this thesis work is to reduce the turbine fatigue load reduction
using pitch actuation of turbine blades. This section explains the concept
of turbine fatigue and pitch activity and also the calculation of fatigue and
pitch activity of the designed model using with and without Design Filter
configuration with varying Design IPC gain.
4.4.1 Fatigue Calculation

Maximum amount of stress that an object can withstand without any per-
manent deformation into it’s shape is called Fatigue. Fatigue in simple words
can be explained as if a force is applied on a copper wire at one point, this
force will cause deformation into the shape of wire. This deformation into
the shape can be reversed by applying the same force in opposite direction.
But if this process is repeated several times then the wire will not withstand
this deformation any more and will break (permanent deformation). This
phenomenon of applying force several time causes the Fatigue to copper wire
which ends up into the permanent deformation.
Figure 4.13: Fatigue for Blade Flapwise Bending Moments
Similarly when a wind turbine is exposed to the wind forces, these forces
not only rotates the blades around rotor axis but also bends the blades out
of rotor plane which is known as Blade Bending Moments and bending of

Model Analysis
tower in fore-aft direction due to wind forces is known as Tower Bending

Moments. These bending moments causes the fatigue for blades and tower
and can be catastrophic in extreme wind conditions [31].
Figure 4.13 shows the fatigue graph for the turbine blades with varying
Design IPC gain value. In above portion of figure, solid lines shows the
fatigue calculation with Design Filters configuration while the dotted lines
shows the fatigue calculation without Design Filters configuration. All three
Collective, yaw and tilt components are represented separately in this figure.
Below portion of figure shows the mean value for Collective, yaw and tilt
component where Blue line represents the model with Design Filters and
varying Design IPC gain configuration and Green line represents the model
without Design Filters and varying Design IPC gain configuration.
Figure 4.13 shows the fatigue graph for all three blades with varying gain
into the same range as defined earlier i.e. 5 ÷ IP Cgain ←→ 5 × IP Cgain.
Mean value graph shows that the blade fatigue reduction is better with
design filters in comparison with the fatigue graph of system without design
filters. But as the system gain increases, fatigue graph shoots up. This
behaviour of fatigue graph is due to the fact that increase in IPC-gain value
makes the system unstable. This instability causes the rapid increase in
turbine fatigue which can be catastrophic in real world scenario.
Turbine tower was considered to be stationary in current control method-

ology. But the system designed in this thesis work is for the large wind
turbines where coupling of tower and blades cannot be neglected. Tower
fatigue occurs due to the fore-aft motion of tower due to the wind force.
Figure 4.14: Fatigue for Tower Bending Moments

Model Analysis
Tower fatigue graph in figure 4.14 shows that the system without Design
Filters provides better fatigue load reduction in comparison with the graph
with Design Filters.
4.4.2 Pitch Activity Calculation

Whenever a wind turbine is exposed to the wind flow, it not only gives
the turbine blades the lift force to rotate around the rotor but also the
drag forces which causes the blade flapwise bending. Blade pitch angle is
manipulated in order to change the angle of attack of wind forces to reduce
the drag forces and to increase the lift forces as much as possible. This rate
of change of blade pitch angles of the turbine blades is known as the Pitch
activity.
As discussed earlier in chapter 2, pitch angles are used to manipulate the

angles of attack of the wind force on turbine blades to get the maximum
load reduction. This load reduction results in to the reduction of the blade
fatigue as well. In general convention, turbine load reduction increases with
increase in blade pitch activity.
Wind turbines are huge structures and blade actuators are responsible to
provide the pitch angle demand as per the control requirement. Due to the
physical limitation, maximum achievable pitch activity is 2 rad
sec .
Figure 4.15: Pitch Activity
Figure 4.15 shows the pitch activity graph for the turbine model in both
with and without Design Filters configuration.

Model Analysis
Figure 4.16: Pitch Activity (PSD)
Figure 4.16 shows the Power Spectral Density (PSD) graph for the pitch
activity calculation. A predefined code in ECN is utilized for this calculation.
Figure 4.15 shows the pitch activity plot in rotating domain. Difference in
signal amplitudes is visible but it is more clear in figure 4.16 where spectrum
of pitch activity shows the filtering effect clearly. Power spectral density
(PSD) describes the distribution of signal strength in frequency domain.
First peak is due to the 1p frequency component and afterwards it can be
observed that the signals are smooth with the design filters configuration as
compare to the without filters configurations.
Finally, overall analysis of the system enables us to make the conclusion that
the Current Control Design (Design Filters) with a constant Design IPC
gain value doesn’t provide the satisfactory results for the large wind turbine
model. Although Current Design presents better Pitch Activity calculations
but due to the wind turbulent nature it is highly desirable to vary controller
gain value to efficiently reduce the turbine fatigue. Whereas the current
control method makes the system unstable at higher gain values. Therefore
a new mutlivariable control mechanism is required to replace this current
method to cope up with this problem of turbine fatigue load reduction. Next
Chapter will discuss a new proposed control technique implementation and
it’s advantages over the current method.

Chapter 5
Controller Design
5.1 Multivariable Control

Single Input Single Output or SISO system is referred to the process with
only one output which is controlled by a single input variable. However
most industrial operations have more than one control loop which means
that number of outputs are manipulated through number of input variables
to get the desired results. Systems which such configurations are referred as
Multi Input Multi Output or MIMO systems.
5.1.1 PID Control

Proportional Integral Derivative controller is most commonly used control
technique in industries. PID technique works in close loop system where
an error signal is measured as the difference between the desired set point
(STPT) and process variable (PV). Controller adjust the control inputs of
the system to maintain process variable on desired set point.
Considerable attention has been given to the use of SISO procedures for
the tuning of decentralized PID controllers for MIMO systems. Motivations
comes from the fact that many systems can be made diagonally dominant
by designing appropriated decoupling compensator [32].
Major reasons to avoid PID technique are:
• In presence of an accurate mathematical model, advance control tech-

niques like LQG design offer optimal results as compare to PID tech-
nique.
• In comparison with LQG design, PID technique creates a stable system

without optimizing anything. However in our case, it is required to
minimize the blade bending moments for turbine load reduction.

Controller Design
5.1.2 Model Predictive Control

In Model Predictive Control method, process model and current measure-
ments are used to predict the future values of output variables. Control
signal is calculated through these future predictions and system current
measurements [33, 34].
Major reasons to avoid MPC technique are:

• Empirical model calculations through system identification adds extra
complexity to the model design.
• Additional feedback loop maybe required to compensate the predic-

tion error for MPC implementation which will make the system more
complex to handle.
5.2 Linear Quadratic Gaussian Control

Linear Quadratic Gaussian (LQG) Control Algorithm is one of the most
fundamental optimal control technique. LQG algorithm uses the system
statespace model to minimize the weighted cost function [35]. This technique
provides the unique solution using a linear feedback control law. Calculation
and implementation of this feedback law is fairly easy.
Major advantages of using LQG technique are:

• LQG design gives the best optimal results with accurate mathematical
model of the system.
• Stability is guaranteed if all states of the system are available for feed-
back.
Both the above mentioned properties are achievable in our system design
because a detailed linear mathematical model is developed in Chapter 3
which makes the LQG an obvious choice for controller design.
LQG techniques is the combination of Linear Quadratic Estimator (LQE)

i.e. Kalman Filter and Linear Quadratic Regulator (LQR). This design
gives the freedom to compute estimator and regulator part independently
and then combining them to form complete control design [35, 30]. In our
case, as the turbine model is completely translated in to Coleman domain so
for the sake of simplicity of the control design, collective component of the
bending moment is neglected for both estimation and regulator part, only
tilt and yaw bending moments has been taken into account. Reason is that
the collective component gives the rotational motion of the blades while yaw
and tilt components causes the fatigue due to twisting and for-aft motion of
turbine.

Controller Design
5.2.1 Kalman Filter Design

Kalman filter or state estimator operates recursively over a series of system
measurements observed over time with inaccuracies (Noise) and produces
optimal estimates of system states [36, 30].
Continuous Time Plant Equations
ẋ = Ax + Bu + Gw (5.1)
y = Cx + Du + Hw (5.2)
Kalman filters produces the estimates x̂ of the noisy system to minimize

state error covariance
h i
P = lim E (x − x̂)(x − x̂)T (5.3)
t→∞
Optimal solution is given by:

x̂˙ = Ax̂ + Bu + L y − C x̂ − Du (5.4)

ŷ C D
= x̂ + u (5.5)
x̂ I 0
Solution of Algebric Riccati equation gives the Kalman filter gain L

L = P C T + N̄ R̄−1 (5.6)
In our case, system structural dynamics due to wind are considered to be the
system noise instead of white noise and measurements for blades flap-wise
bending moments are the Kalman filter measured variables.
Figure 5.1: Schematic diagram of Kalman Filter implementation

Controller Design
System in figure 5.1 can be solved in two different ways.
• By duplicating the common inputs for both statespace model and

kalman filter using a 12 × 9 identity matrix
• By individually solving the transfer function for each node of the sys-
tem
As the pitch angles are the input for both statespace model and kalman
filter as shown in figure 5.1 . Statespace model and kalman filter block is
attached using Matlab command and common inputs are duplicated by post
multiplying the system with smart matrix i.e.
 
Vw I 0 0
Vw
θb   0 I 0  ⇒ θb

(5.7)
θb  0 I 0 
BRM
BRM 0 0 I
Other way is by individually solving the transfer function for each node
Figure 5.2: Individual solution for each node
Node A

U
BRM = StatespaceM odel ax (5.8)
θb
Node B

θb
x̂ = KalmanF ilter (5.9)
BRM
Node A+B
 
θb
x̂ = KalmanF ilter  U x  (5.10)
StatespaceM odel a
θb

Controller Design

0 θb
x̂ = KF + KF (5.11)
SS.Uax SS.θb

Zeros(3 × 3) Ones(3 × 3) Uax
x̂ = KF (5.12)
SS.Uax SS.θb θb
Both the approaches gives the correct estimates of the blade flapwise bending
moments and system states.
Figure 5.3: Kalman Filter Estimates for Blade Bending Moments
Figure 5.3 shows the estimates given by the Kalman filter for the yaw and
tilt component of bending moments. Kalman filter shows the high accuracy
as estimated signals are very similar to the actual bending moments of the
system.
5.2.2 Linear Quadratic Regulator Design

Linear Quadratic Regulator is an optimal controller design method which
determines practical feedback gain to minimise the quadratic cost function
of a system [29, 30]. Deviation of system key measurements from their
optimal values defines the cost function of the system. Optimization of the
system involves the minimization of these deviations with weighting factors
developed by the control engineer.
Continuous Time Plant Equation

ẋ = Ax + Bu + Gw (5.13)
Cost Function Z ∞h i
J(u) = xT Qx + uT Ru dt (5.14)
0

Controller Design
State-feedback law defined as: u = −Kx minimizes the cost function of a

continuous time system where K defines the trade off between control effort
and system transient response.
where
K = R−1 B T P (5.15)
P is calculated by solving continuous time Algebric Riccati equation
AT P + P A − P BR−1 B T P + Q = 0 (5.16)
Designing the weighting factors is an iterative process. Control engineer

compares the system response in each iteration with the desired system
response [30]. In our case, cost function is composed of the turbine dynamics
for Yaw and Tilt bending moments. Minimization of these dynamics is
required for the turbine fatigue reduction.
Figure 5.4: Schematic diagram of LQG design
Figure 5.4 shows the system configuration with complete LQG design. Weight-
ing factor calculation has been done in iterative way, where ρ value for the
R matrix is varied in range to get the desired control output. Best starting
point for this iterative process is always to try out value close to the cost
function and then to vary it for certain ranges to get desired results. For

Controller Design
this purpose number of iterations have been carried out with different value
to get the best weighting factors in terms of load reduction.
5.3 Fatigue Calculation With LQG Design

Turbine fatigue as discussed in section 4.4.1. Maximum amount of stress
that an object can withstand without any permanent deformation into it’s
shape is called Fatigue. Fatigue calculation have been made with the LQG
design configuration where weighting factor is varied in order to minimize the
quadratic cost function i.e. Blade yaw-wise and tilt-wise bending moments.
Figure 5.5: System Response with Varying Weighting Factor
Figure 5.5 shows the fatigue calculation of blade bending moments with LQG
configuration. Weighting factor value is varied in the range of 1e3 ←→ 1e15
and taking 400 values with in this range gives us the fatigue reduction bet-
ter than previous design. It can be observed that the increase in weighting
matrix decreases the LQR gain value which results into the increase in the
blade bending moments. This figure also gives the comparison of LQG de-
sign with current control design with designed gain value. It is clear that the

Controller Design
signal amplitude with LQG is significantly smaller than current control de-
sign with both configurations (i.e. With and without design filters). Better
comparison is presented in next section where gain is varied for the current
control design as well.
5.4 Pitch Activity Calculation With LQG Design

Pitch activity as discussed in Section 4.4.2, is the rate of change of pitch
angles of the turbine blades is known as Pitch Activity. Blade pitch angles
are used for the turbine load and fatigue reduction. Below graphs shows the
pitch activity graphs with the LQG control technique.
Figure 5.6: Pitch Activity Comparison

Controller Design
Figure 5.7: Pitch Activity Comparison
From figure 5.6,5.7, it can be observed that the pitch activity signal ampli-
tudes with the LQG design is lower then the previous control design tech-
nique. This is a very healthy result in terms of fatigue reduction because
LQG design utilized less pitch activity as compared to other configuration
and provides better load reduction.
5.5 Fatigue Reduction VS Control Effort

This section describes the turbine fatigue load reduction with variation in
pitch activity. As the turbine blades are experiencing the uneven load due
to the wind force and ideally turbine fatigue decreases with increase in pitch
activity.
Prime target for this thesis work is the turbine fatigue load reduction. Pro-
posed LQG technique fatigue response has been discussed in previous sec-
tion. In order to make a final conclusion about the efficiency of the proposed
technique, fatigue reduction is calculated against the control effort where the

Controller Design
control effort is the amount of pitch activity utilized by the actuators to the
reduce the turbine fatigue. In simple convention, turbine fatigue reduces
with increase in blade pitch activity.
Figure 5.8: Fatigue VS Effort
Figure 5.8 shows the turbine fatigue against pitch activity for all three con-
figurations. It can be observed that the fatigue reduction with the proposed
LQG design is significantly high than the other two configurations. Hence
we can concluded that the proposed control technique offers better load
reduction with the same control effort in comparison with the current tech-
nique. Blue line shows the fatigue graph for current control design and
as discussed earlier system becomes unstable with higher gain value, graph
shoot up shows the system instability. While fatigue graph with LQG tech-
nique is in smooth decreasing fashion which means that the increase in pitch
activity decreases the turbine fatigue.
Reason behind smooth decreasing fashion of fatigue reduction through LGQ

design is Kalman filter optimal state estimation. Optimal in a sense that
it is the only filter which minimizes the variance of estimation error due to
its recursive nature [37]. Kalman filter satisfies two main criteria to be an
optimal state estimator [38, 36]:
• Average value of state estimates should be equal to the average value

of actual state.

Controller Design
• Estimator results must have the smallest possible error variance.
But this optimal solution requires certain assumption about the system
noise. Average value of the system noise must be equal to zero and there
must not exist any correlation between the system noise and measurement
noise [38]. In our case, turbine structural noise due to the wind force is
considered to be the system noise. Cost function optimization with optimal
state estimates of the system given by the Kalman filter provides maxi-
mum reduction into the blade bending moments [29, 30]. LQR controller
calculates the optimal gain value using estimated system states for all the
pitch activity value. Increase in pitch activity provides the controller more
strength which result in smooth reduction of the fatigue in comparison with
the current control design.
As the current control design is composed of the conventional impulse and

low pass filters, it does not posses the ability to actively minimize the ef-
fect of blade bending moments. Due to this reason, fatigue variation with
current design is fluctuating and in high wind speed location, for example
off-shore locations, these bending moments becomes even more intense due
to the turbulent nature of wind itself. This will further increase the irreg-
ular behaviour of turbine fatigue with current control design. High fatigue
variation at off-shore location requires more control effort for load reduction
i.e. high pitch activity of turbine blades and due to the physical limitation
of blade actuators, these high requirement are not achievable in real world
scenario. Therefore fatigue response of current design is not suitable for
large turbine at high wind speed locations.
However fatigue load reduction at one point is almost the same for LQG
design and current control design but still LQG design should be preferred.
Reason is that, fatigue reduction is same for both control techniques for
a very small range of pitch activity. While in real world scenario wind
has a turbulent nature and the amount of force experience by the turbine
blades is also varying all the time. To achieve efficient load reduction, pitch
activity is keep on changing all the time as per the requirement. This pitch
activity requirement cannot be fulfilled with small range whereas the load
reduction offered through LQG design has a smooth response and wind load
variation would not make any big change in the system response. So it can
be concluded that turbine fatigue load optimization is more efficient with
LQG design in comparison with current control design.
5.6 Economic Aspect of Project

This section describes the practical benefits of this thesis work for wind
turbine working in real world scenario.

Controller Design
Economical viability of a wind turbine is based on how long turbine works

efficiently. Period of time in which turbine is able to produce maximum
power output is called Age of wind turbine. Turbine fatigue have inverse
relation to the age of the wind turbine, lower the turbine fatigue higher will
the turbine age.
Figure 5.9: Fatigue VS Effort
In order to draw an economic perspective of this project, graph for turbine

fatigue vs control effort is divided into two regions depending on the maxi-
mum pitch activity offered by the actuation system of the selected turbine.
Region 1 is for the turbine which offers low pitch activity, 8 − 10% approx
fatigue load reduction is achievable in this region and Region 2 is for the tur-
bine which offers more pitch activity, 4 − 11% approx fatigue load reduction
is achievable in this region. 10% reduction of fatigue means 10% increase
in turbine age which means if a turbine age is 20 years, 10% reduction of
fatigue will add 2 extra years to the turbine age. This is highly appreciated
result in terms of fatigue load reduction.

Chapter 6
Conclusions and Future

Work
6.1 Conclusions
Current trend in wind industry is to achieve the maximum cost reduction for
the energy production of wind turbines. Increase in rotor size and placing
the turbine in locations, where the average wind speed is high e.g offshore
locations, offers increase in energy production. But increase in turbine size
and placing them to offshore locations makes the turbine maintenance more
difficult and economically expensive. Hence turbine active fatigue load re-
duction has the vital role to increase the turbine reliability and efficiency
which is also the focus of this thesis work.
Main goals of this thesis work were to create a detailed model for the large
wind turbine which include turbine blade and tower dynamics and to imple-
menting a multivariable control technique for turbine fatigue load reduction.
Following issues were addressed to successfully achieve the prime target :
• Development of a detailed turbine mathematical model including tur-

bine blade and tower dynamics.
• Analysis of current control technique using designed turbine model.
• Development of LQG design for fatigue load reduction
All analysis have been carried out with this designed model for large wind
turbine fatigue load reduction. Comparison of previous control technique
and new proposed LQG design gives clear improvements in terms of tur-
bine fatigue reduction. Previous control technique utilise conventional fil-
ters which does not posses the ability to actively minimize the effect of blade
bending moments. Therefore the fatigue graph shows a fluctuating and ir-
regular behaviour which will further increase at high wind speed locations

Conclusions and Future Work
due to the turbulence of wind itself. Such fatigue behaviour is highly un-
desirable for large wind turbine because it requires rapid and high pitch
actuation for turbine blades. And due to the physical limitations of blade
actuators, this high pitch activity requirement is not achievable in real world
scenario. Where as the fatigue graph for LQG design shows a very smooth
response due to actively minimizing the bending moments of turbine blades.
This smooth response offers the freedom to implement this LQG design to
various turbines which works into different pitch activity ranges.
By examining all the results, it can be concluded that:
• LQG offers better trade-off between load reduction and control effort
in comparison with the current control technique.
• Possibility of 4 − 11% approx fatigue load reduction can be achieved

through LQG technique.
• Turbine age can be improved by efficient load reduction.
• LQG design offers freedom to work with various turbines which works
into different pitch activity ranges.
On the basis of these conclusions, it is highly recommended that the LQG

design should be considered for turbine fatigue reduction. Simulation results
show that the turbine age can also be increased significantly by using LQG
design which will be very helpful in making the wind energy an economically
viable alternative for energy production.
6.2 Future Work

This design model and control approach gives efficient results for the large
wind turbine. Possible addition to this designed model will be:
• Centrifugal force of turbine blades.
Wind turbine is a huge structure and turbine blades and rotor itself possess
heavy weight. Rotational motion of turbine blades in high wind locations
will exert an effect of the centrifugal force due to inertia. That force will
try ot carry the blades away form the body of turbine. In future, it should
be the study point and it will require some practical calculation as well to
make a mathematical formulation for this force effect. This can be added
easily with the designed model which will further improve the turbine control
mechanism.
• Adding extra hinge points in blade construction.

Conclusions and Future Work
For sake of simplicity for this project work, turbines blades were considered
to have a single Hinge point. This assumptions gives the required simplicity
but in real world scenario, dynamics of the turbine blades can be described
more accurately by using more then one hinge points.
Figure 6.1: Blade layout with two hinge points
Adding one or more hinge points for blade construction will increase the
mathematical complexity of the model. When this proposed blade construc-
tion will exposed to the wind forces, there will be the two bending moments
for two reaction forces from each hinge point location. Lift forces which
drives the blades to rotate around rotor are distributed along the blade sur-
face and due to the shape of blades, lift forces are stronger close to the tip
of the blade than the rigid part close to the rotor. Therefore it is a good
idea to add extra hinge point close to the thin portion of blade to explain
the bending moment separately for this portion. Further linearisation of the
model will become more difficult but at the same time the blade dynamics
provided by this proposed construction will be more accurate and close to
the real world turbine blades.

Appendix A
Mathematical Model
Calculations
In order to reduce the loads of wind turbine blades, a system is required

which includes all turbine movements due to the wind force and yields flap-
wise bending moments of blades (i.e. bending of blades out of rotor plane)
and their respective velocities. That will enable us to implement a feedback
loop and control strategy to adjust the IPC loop gain at optimal value for
maximum load reduction. For this purpose a mathematical model is de-
veloped in which blades and tower of the turbine are considered to be in
Mass Spring Damper(MSD) configuration. Equations of motion for turbine
dynamics presented in this Appendix are derived form the previous turbine
model used in ECN. My contribution in this modelling part is to linearise
these equations by solving them simultaneously and to form a Statespace
model of these linearised equations.
Equation of motion for rotation of blade is the function of reaction force of

the joints opposite to the wind force and blade azimuth angle.

Jb θ̈b = −sb θb −db θ̇b +sb cos ψθt +db cos ψ θ̇t −db ω sin ψθt +Fj Rcog +Fax Rae −Rcog
(A.1)
Equation of motion for blade displacement is the function of actual force
acting on blades due to wind and the reaction force exerted by the joint in
opposite direction, which is given by:
mb ẍb = Fax − Fj (A.2)
Tower motion have no direct relation with the actual wind acting on the
blades but with the sum of all the three joint forces acting in the opposite
direction of wind force and with the moment of tilt of the blades.
Mtilt X
mt ẍt = −st xt − dt ẋt + 2 + Fj (A.3)
3H i

Mathematical Model Calculations
where Mtilt is the sum of joint forces Fj of three blades and moments of
joints Mj " #
X
Mtilt = Fj Rb − Mj cos(ψ) (A.4)
i
and moment of joint Mj is:
Mj = −sb θb − db θ̇b + sb cos(ψ)θt − db ω sin(ψ)θt + db cos(ψ)θ̇t (A.5)
Motion of blades will always be the function of tower motion and the rota-
tional motion of blades. Hence it is the boundary condition for the model
as well.
xt
xb = xt + cos ψRb − θb Rcog (A.6)
Lt
Now using Coleman transformation of the blade dynamics to eliminate

the azimuth dependency and to convert it from rotating reference frame
to the static reference frame for the implementation of control technique
Equation(A.1) will be used for all three blades.

Jb θ̈b1 = −sb θb1 − db θ̇b1 + sb cos ψ1 θt + db cos ψ1 θ̇t − db ω sin ψ1 θt + Fj R1cog + Fax Rae − Rcog


Further simplification of the equation gives us:

Jb θ̈b = −sb θb −db θ̇b +(sb cos ψ−db ω sin ψ)θt +db cos ψ θ̇t +Fj Rcog +Fax Rae −Rcog
xt ẋt
We know that θt = and similarly θ̇t =
Lt Lt
x ẋt
t
Jb θ̈b = −sb θb −db θ̇b + sb cos ψ−db ω sin ψ +db cos ψ +Fj Rcog +Fax Rae −Rcog
Lt Lt
Rearranging the above equation
! !
sb cos ψ − db ω sin ψ db cos ψ
Jb θ̈b = xt + ẋt −sb θb −db θ̇b +Fj Rcog +Fax Rae −Rcog
Lt Lt
(A.7)
Putting !
sb cos ψ − db ω sin ψ
=a
Lt
!
db cos ψ
=b
Lt

Final equation for the turbine blades will be :

Jb θ̈b = axt + bẋt − sb θb − db θ̇b + Fj Rcog + Fax Rae − Rcog (A.8)
Writing above equation for all three turbine blades
 
xt
 ẋt 
 
θb1 

   
θ̈b1 a1 b1 −sb −db 0 0 0 0  
1 θ̇ 
θ̈b2  = a2 b2 0 0 −sb −db 0 0   b1 
Jb θ 
θ̈b3 a3 b3 0 0 0 0 −sb −db  b2 
θ̇b2 
 
θb3 
θ̇b3
   
F Fax,b1
Rcog  j,b1  1
+ Fj,b2 + Rae − Rcog Fax,b2  (A.9)
Jb Jb
Fj,b3 Fax,b3
Coleman Transformation of blades:
 
xt
 ẋt 
 
θb1 
   
i θ̈b1 a1 b1 −sb −db 0 0 0 0  
h 1 h i
θ̇ 
CM −1 θ̈b2  = CM −1 a2 b2 0 0 −sb −db 0 0   b1 
Jb θ 
θ̈b3 a3 b3 0 0 0 0 −sb −db  b2 
θ̇b2 
 
θb3 
θ̇b3
   
Fj,b1 Fax,b1
Rcog  1
+ Fj,b2  + Rae − Rcog Fax,b2  (A.10)
Jb Jb
Fj,b3 Fax,b3
h i
where transformation matrix CM −1 is:
1 1 1
 
3 3 3
 2 sin(ψ) 2
sin(ψ + 2π 2
sin(ψ + 4π
3 ) 3 )

3 3 3
2 2
3 cos(ψ) 3 cos(ψ + 2π
3 )
2
3
4π
cos(ψ + 3 )
Taking
 
h i a1 b1 −sb −db 0 0 0 0
= CM −1 a2 b2 0 0 −sb −db 0 0 
a3 b3 0 0 0 0 −sb −db
1 1 1
  
3 3 3 a1 b1 −sb −db 0 0 0 0
=  23 sin(ψt) 23 sin(ψt + 2π
3 )
2 4π  
3 sin(ψt + 3 ) a2 b2 0 0 −sb −db 0 0 
2 2 2π 2 4π
3 cos(ψt) 3 cos(ψt + 3 ) 3 cos(ψt + 3 ) a3 b3 0 0 0 0 −sb −db
(A.11)

Solving first row :
a1 + a2 + a3 & b1 + b2 + b3

1
a1 + a2 + a3 = sb cos ψ1 − db ω sinψ1 + sb cos ψ2 − db ω sinψ2 + sb cos ψ3 − db ω sinψ3
Lt
!
1
= sb cos ψ1 + cos ψ2 + cos ψ3 − db ω sinψ1 + sinψ2 + sinψ3
Lt
a1 + a2 + a3 = 0 (A.12)
&
!
db
b1 + b2 + b3 = cos ψ1 + cos ψ2 + cos ψ3
Lt
b1 + b2 + b3 = 0 (A.13)
Solving second row :

! !
2 2
sin ψ1 a1 + sin ψ2 a2 + sin ψ3 a3 & sin ψ1 b1 + sin ψ2 b2 + sin ψ3 b3
3 3
!
2
= sin ψ1 a1 + sin ψ2 a2 + sin ψ3 a3
3
2 1
= sin ψ1 (sb cos ψ1 − db ω sin ψ1 ) + sin ψ2 (sb cos ψ2 − db ω sin ψ2 )+
3 Lt
!

sin ψ1 (sb cos ψ3 − db ω sin ψ3 )
1
=− db ω (A.14)
Lt
&
!
2
= sin ψ1 b1 + sin ψ2 b2 + sin ψ3 b3
3
!
db
= sin ψ1 cos ψ1 + sin ψ2 cos ψ2 + sin ψ3 cos ψ3
Lt
=0 (A.15)

Solving third row :

! !
2 2
cos ψ1 a1 + cos ψ2 a2 + cos ψ3 a3 & cos ψ1 b1 + cos ψ2 b2 + cos ψ3 b3
3 3
!
2
= cos ψ1 a1 + cos ψ2 a2 + cos ψ3 a3
3
2 1
= cos ψ1 (sb cos ψ1 − db ω sin ψ1 ) + cos ψ2 (sb cos ψ2 − db ω sin ψ2 )+
3 Lt
!

cos ψ1 (sb cos ψ3 − db ω sin ψ3 )
!
2 1
= sb cos21 + cos22 + cos23 − dt ω sin ψ1 cos ψ1 + sin ψ2 cos ψ2 + sin ψ3 cos ψ3
3 Lt
!
2 1 3
= sb − dt ω 0
3 Lt 2
sb
= (A.16)
Lt
&
!
2
cos ψ1 b1 + cos ψ2 b2 + cos ψ3 b3
3
!
2 db 2π 4π
= cos2 (ψt) + cos2 (ψt + ) + cos2 (ψt + )
3 Lt 3 3
!
2 db 3
=
3 Lt 2
db
= (A.17)
Lt

Transformed matrix of :
 
h i θ̈b1
CM −1 θ̈b2  =
θ̈b3
 
xt
 ẋt 
 
  θ 
0 0 − 13 sb − 31 db − 31 sb − 13 db − 31 sb 1
− 3 db  b1 
1  db θ̇ 
b1 
− Lt ω 0 2
− 3 sb Sψ1 − 3 db Sψ1 − 3 sb Sψ2 − 3 db Sψ2 − 23 sb Sψ3
2 2 2 2
− 3 db Sψ3  

Jb θ
 
sb db  b2 
Lt Lt − 32 sb Cψ1 − 23 db Cψ1 − 23 sb Cψ2 − 32 db Cψ2 − 23 sb Cψ3 − 32 db Cψ3  
θ̇b2 
 
θb3 
θ̇b3
   
Rcog h i Fj,b1 1 h i F ax,b1
CM −1 Fj,b2  + Rae − Rcog CM −1 Fax,b2  (A.18)

+
Jb Jb
Fj,b3 Fax,b3
where SandC represents sin and cos respectively
Simplification of above equation

 
h i θ̈b1
CM −1 θ̈b2  =
θ̈b3
 
− 13 sb θb1 + θb2 + θb3 − 13 db θ̇b1 + θ̇b2 + θ̇b3
1  
− dLb tω xt − 32 sb Sψ1 θb1 + Sψ2 θb2 + Sψ3 θb3 − 23 db Sψ1 θ̇b1 + Sψ2 θ̇b2 + Sψ3 θ̇b3
 
Jb
 
 
sb db 2 2
Lt xt + Lt ẋt − 3 sb Cψ1 θb1 + Cψ2 θb2 + Cψ3 θb3 − 3 db Cψ1 θ̇b1 + Cψ2 θ̇b2 + Cψ3 θ̇b3
   
Rcog h i Fj,b1 1 h i F ax,b1
CM −1 Fj,b2  + Rae − Rcog CM −1 Fax,b2  (A.19)

+
Jb Jb
Fj,b3 Fax,b3
Now translating blade dynamics of each blade into the Coleman Domain.
We know that :
       
θcollective h i θ1 θ1 h i θcollective
 θyaw  = CM −1 θ2  <=> θ2  = CM  θyaw  (A.20)
θtilt θ3 θ3 θtilt

From above relation we can write:

h i
θcm = CM −1 θ (A.21)
So,
1
θcol = θ1 + θ2 + θ 3 (A.22)
3
2
θyaw = sin ψ1 θ1 + sin ψ2 θ2 + sin ψ3 θ3 (A.23)
3
2
θtilt = cos ψ1 θ1 + cos ψ2 θ2 + cos ψ3 θ3 (A.24)
3
Taking derivative on both sides
1
θ̇col = θ̇1 + θ̇2 + θ̇3
3
3θ̇col = θ̇1 + θ̇2 + θ̇3 (A.25)
!
2
θ̇yaw = sin ψ1 θ̇1 +sin ψ2 θ̇2 +sin ψ3 θ̇3 +ω cos ψ1 θ1 +cos ψ2 θ2 +cos ψ3 θ3
3
2
θ̇yaw = sin ψ1 θ̇1 + sin ψ2 θ̇2 + sin ψ3 θ̇3 + ωθtilt
3
2
θ̇yaw − ωθtilt = sin ψ1 θ̇1 + sin ψ2 θ̇2 + sin ψ3 θ̇3 (A.26)
3
!
2
θ̇tilt = cos ψ1 θ̇1 +cos ψ2 θ̇2 +cos ψ3 θ̇3 −ω sin ψ1 θ1 +sin ψ2 θ2 +sin ψ3 θ3
3
2
θ̇tilt = cos ψ1 θ̇1 + cos ψ2 θ̇2 + cos ψ3 θ̇3 − ωθyaw
3
2
θ̇tilt + ωθyaw = cos ψ1 θ̇1 + cos ψ2 θ̇2 + cos ψ3 θ̇3 (A.27)
3
Now taking the first derivative of equation(A.34):

h i h ˙ i
θ̇cm = CM −1 θ̇ + CM −1 θ (A.28)
Taking the second derivative of equation(A.34)::

h i h ˙ i h ˙ i h ¨ i
θ̈cm = CM −1 θ̈ + CM −1 θ̇ + CM −1 θ̇ + CM −1 θ (A.29)

Putting values of θ and θ̇

! !
θ̈cm = CM −1 θ̈ + 2CM˙ −1 CM θ̇cm − CM CM˙ −1 CM θcm ¨ −1 CM θcm
+ CM
θ̈cm = CM −1 θ̈ + 2CM˙ −1 CM θ̇cm − 2CM˙ −1 CM CM˙ −1 CM θcm + CM

¨ −1 CM θcm
(A.30)
Now:
   
0 0 1 sin ψ1 cos ψ1
˙ −1 2 2 2 2
CM CM = 3 ω cos ψ1 3 ω cos ψ2 3 ω cos ψ3 × 1 sin ψ2 3 ω cos ψ2
   
2 2 2 2
3 ω sin ψ1 ω sin ψ2 3 ω sin ψ3 1 sin ψ3 3 ω sin ψ3
 3
0 0 0
= 0 0 ω  (A.31)
0 −ω 0
And  
0 0 0
CM˙ −1 CM CM˙ −1 CM = 0 −ω 2 0  (A.32)
0 0 −ω 2
And  
0 0 0
¨ −1 CM = 0 −ω 2
CM 0  (A.33)
0 0 −ω 2
So equation(A.30) will become:
   
" # 0 0 0 0 0 0
θ̈cm = CM −1 θ̈b + 2 0 0 ω  θ̇cm − 0 −ω 2 0  θcm (A.34)
0 −ω 0 0 0 −ω 2

Putting the values from equation(A.25,A.26,A.27,A.34), So equation(A.19)

will give us:
− Jsbb θcol − Jdbb θ̇col

     
0 0
θ̈cm =  − JbdLb t ωxt − Jsbb θyaw − Jdbb θ̇yaw + Jdbb ωθtilt +2  ω θ̇tilt + ω 2 θyaw 
 
sb db sb db db −ω θ̇yaw −ω 2 θtilt
Jb Lt xt + Jb Lt ẋt − Jb θtilt − Jb θ̇tilt − Jb ωθyaw
   
Fjcol Faxcol
Lb  1
+ Fjyaw  + Rae − Rcog Faxyaw  (A.35)
Jb Jb
Fjtilt Faxtilt
So,
 
xt
 ẋt 
 
 sb db
 θ 
0 0 − Jb − Jb 0 0 0 0   col 
 db 2 s d d θ̇ col

θ̈cm = − Jb Lt ω 0 0 0 (ω − Jb ) − Jb
b b
Jb ω
b
2ω  
  
θ

sb db db 2 − sb ) − db  yaw 
 
Jb Lt Jb Lt 0 0 − Jb ω −2ω (ω Jb Jb θ̇yaw 
 
 θtilt 
θ̇tilt
   
F Faxcol
Lb  jcol  1
+ Fjyaw + Rae − Rcog Faxyaw  (A.36)
Jb Jb
Fjtilt Faxtilt
Since boundary condition (Equation (A.6)) gives :
Rb
xb = xt + cos ψxt + θb Rcog
Lt

Coleman transformation of this above equation gives us :

" #
xbcm = CM −1 xb
 Rb

1 1 1 1+ cos ψ1
   !
3 3 3 Lt θb1
Rb
xbcm =  23 sin(ψt) 32 sin(ψt + 2π
3 )
2
sin(ψt + 4π
3 )  +
1 cos ψ2  xt + Rcog θb2 
  
3 Lt
2 2 2π 2 4π Rb
3 cos(ψt) 3 cos(ψt + 3 ) 3 cos(ψt + 3 ) 1+ Lt cos ψ3 θb3
 
xt + θcol Rcog
xbcm =  θyaw Rcog  (A.37)
xt
Lt Rb + θtilt Rcog
 
ẋt + θ̇col Rcog
ẋbcm =  θ̇yaw Rcog  (A.38)
ẋt
Lt Rb + θ̇tilt Rcog
Taking second derivative on both sides
 
ẍt + θ̈col Rcog
ẍbcm =  θ̈yaw Rcog  (A.39)
ẍt
Lt Rb + θ̈tilt Rcog
Joint forces acting in the opposite direction to the wind force are given by:
Fj = Fax − mb ẍb
Taking Coleman Transformation of above equation
h i h i h i
CM −1 Fj = CM −1 Fax − mb CM −1 ẍb
Simplification using equation(A.34)

 
0 0 0
Fjcm = Faxcm − mb ẍbcm − 2 0 0 ω  ẋbcm
0 −ω 0
 
0 0 0 !
− 0 −ω 2 0  xbcm
0 0 −ω 2
Now using equations(A.39)
   !
0 0
Fjcm = Faxcm −mb ẍbcm −2ω  R  2  θyaw Rcog
Lt ẋt + θ̇tilt Rcog −ω
b 
Rb
−ω θ̇yaw Rcog Lt xt + θtilt Rcog
(A.40)

Putting values from above equation into the equation(A.36)

 
xt
 ẋt 
 
 sb db
 θ 
0 0 − Jb − Jb 0 0 0 0  col 
  θ̇col 
 
 db 2 sb db db
θ̈cm = − Jb Lt ω 0 0 0 (ω − Jb ) − Jb Jb ω 2ω  θ 
sb db db 2 − sb ) − db  yaw 
 
Jb Lt Jb Lt 0 0 − Jb ω −2ω (ω Jb Jb θ̇yaw 
 
 θtilt 
θ̇tilt
   !
0 0
mb Rcog Rb 2
− ẍbcm − 2ω Lt ẋt + θ̇tilt Rcog − ω
  θyaw Rcog 
Jb Rb
−ω θ̇yaw Rcog Lt xt + θtilt Rcog
   
F Faxcol
Rcog  axcol  1
+ Faxyaw Rae − Rcog Faxyaw  (A.41)
Jb Jb
Faxtilt Faxtilt
Now putting in the value from equation(A.39) into equation(A.41):
− Jsbb θcol − Jdbb θ̇col

 
θ̈cm =  − JbdLb t ωxt + (ω 2 − Jsbb )θyaw − Jdbb θ̇yaw + Jdbb ωθtilt + 2ω θ̇tilt 
 
sb db db 2 sb db
Jb Lt xt + Jb Lt ẋt − Jb ωθyaw − 2ω θ̇yaw (ω − Jb )θtilt − Jb θ̇tilt
 
ẍt + θ̈col Rcog
mb Rcog  R
−  θ̈yaw Rcog − 2 Ltb ω ẋt − 2ωRcog θ̇tilt − ω 2 Rcog θyaw 

Jb Rb Rb 2 2
Lt ẍt + θtilt Rcog + 2ωRcog θ̇yaw − Lt ω xt − ω Rcog θtilt
 
Faxcol
Rae 
+ Faxyaw  (A.42)
Jb
Faxtilt

Blade dynamics equations :
sb db mb Lb Rae
θ̈col = − 2
θcol − 2
θ̇col − 2
ẍt + 2
Faxcol
Jb + mb Rcog Jb + mb Rcog Jb + mb Rcog Jb + mb Rcog
(A.43)
2 )ω 2 − s
(Jb + mb Rcog
db ω 2mb Rcog Rb ω b
θ̈yaw = − 2
x t + 2
ẋ t + 2
θyaw
db db ω Rae
− 2
θ̇yaw + 2
θtilt + 2ω θ̇tilt + 2
Faxyaw
Jb + mb Rcog Jb + mb Rcog Jb + mb Rcog
(A.44)
sb + mb Rcog Rb ω 2 db db ω
θ̈tilt = 2
xt + 2
ẋt − 2
θyaw
2 )ω 2 − s
(Jb + mb Rcog b db
− 2ω θ̇yaw + 2
θtilt − 2
θ̇tilt
mb Rcog Rb Rae
− 2
ẍt + 2
Faxtilt (A.45)
Now using some symbols in above three equations for the sake of simplicity
and converting them into matrix form
 
xt
 ẋt 
 
    θ 
 col   
θ̈col 0 0 k11 k12 0 0 0 0  k13
 θ̇ 

θ̈yaw  = k21 k22 0 0 k23 k24 k25 k26   col  +  0  ẍt
θyaw 
θ̈tilt k31 k32 0 0 k33 k34 k35 k36  k37
θ̇yaw 

 
 θtilt 
θ̇tilt
 
Faxcol
Rae Faxyaw  (A.46)
+ 2
Jb + mb Rcog
Faxtilt

For the tower dynamics we use the equation (A.3) :
Mtilt X
mt ẍt = −st xt − dt ẋt + 2 + Fj
3H i
where " #
X
Mtilt = Fj Rb − Mj cos(ψ)
i
Applying the Coleman’s transformation on above equation. As equation

have cos(ψ) term in it, so transformation will elminate all the terms which
doesn’t have cos(ψ) in it.
" #
3
Mtilt = Fjtilt Rb − Mjtilt (A.47)
2
Putting in the values of Fjtilt and Mjtilt .

" #
Rb Rb 2
Fjtilt = Faxtilt − mb ẍt + ω xt + Lb ω 2 θ tilt − 2Lb ω θ̇yaw + Lb θ̈tilt
Lt Lt
For Mjtilt
Mj = −sb θb − db θ̇b + sb cos(ψ)θt + db cos(ψ)θ̇t − db ω sin(ψ)θt
So,
 
M j1
2
Mjtilt = cos ψ1 cos ψ2 cos ψ3 Mj2 
3
Mj3
"
2 h
Mjtilt = sb cos2 (ψ1 ) + cos2 (ψ2 ) + cos2 (ψ3 )
3
i
− db omega sin(ψ1 ) cos(ψ1 ) + sin(ψ2 ) cos(ψ2 ) + sin(ψ3 ) cos(ψ3 ) θt

+db cos2 (ψ1 )+cos2 (ψ2 )+cos2 (ψ3 ) θ̇t −sb cos(ψ1 )θ(ψ1 )+cos(ψ2 )θ(ψ2 )+cos(ψ3 )θ(ψ3 )
#

− db cos(ψ1 )θ̇(ψ1 ) + cos(ψ2 )θ̇(ψ2 ) + cos(ψ3 )θ̇(ψ3 )
Mjtilt = sb θt + db θ̇t − sb θbtilt − db θ̇btilt + db ωθbyaw (A.48)

So equation(A.47) will become:

"
3 mb Rb2 mb Rb2 2
Mtilt = Rb Faxtilt − ẍt + ω xt −mb Rb Lb ω 2 θtilt +2mb Lb Rb ω θ̇yaw
2 Lt Lt
#
− mb Lb Rb θ̈tilt − sb θt − db θ̇t + sb θtilt + db θ̇tilt − db ωθyaw
" ! ! ! !
3 mb Rb2 mb Rb2 ω 2 + sb dt 2
Mtilt = Rb Faxtilt − ẍt − xt − ẋt − db θyaw
2 Lt Lt Lt 3
! !
2 2
sb −mb Rb Lb ω 2 θtilt +

+ 2mb Lb Rb ω θ̇yaw + db θ̇tilt − mb Lb Rb θ̈tilt
3 3
" !
Mtilt = Rb Faxtilt − ẍt − xt − ẋt −db ωθyaw +2mb Rcog Rb ω θ̇yaw
2 Lt Lt Lt
! #
+ sb − mb Rb Rcog ω 2 θtilt + db θ̇tilt − mb Rb Rcog θ̈tilt (A.49)
Main tower dynamics equation(A.3) will become:

" !
4H Lt Lt Lt
! #
+2mb Rcog Rb ω θ̇yaw + sb −mb Rb Rcog ω 2 θtilt +db θ̇tilt −mb Rb Rcog θ̈tilt +3Fjcol
(A.50)
As:
Fjcol = Faxcol − mb ẍt + θ̈col Rcog (A.51)
So, tower dynamics equation(A.3) will become:
" !
4H Lt Lt Lt
! # " #
+2mb Rcog Rb ω θ̇yaw + sb −mb Rb Rcog ω 2 θtilt +db θ̇tilt −mb Rb Rcog θ̈tilt +3 Faxcol −mb ẍt −mb Rcog θ̈col
(A.52)

Simplification of the above equation will give us :

" # " # " #
9 mb Rb2 9 sb + mb Rb2 ω 2 9 db
+mt +3mb ẍt = − +st xt − +dt ẋt
4H Lt 4H Lt 4H Lt
" # " #
9 9
− db ω θyaw + 2mb Rcog Rb ω θ̇yaw
4H 4H
" # " #
9 9
+ sb − mb Rb Rcog ω 2 θtilt + db θ̇tilt
4H 4H
" # " # " #
9 9
− 3mb Rcog ẍcol − mb Rcog Rb θ̈tilt + 3Faxcol + Rb Faxtilt
4H 4H
(A.53)
Using symbols for the sake of simplicity
j1 ẍt = j2 xt +j3 ẋt +j4 θyaw +j5 θ̇yaw +j6 θtilt +j7 θ̇tilt +j8 θ̈col +j9 θ̈tilt +3Faxcol +j10 Faxtilt
(A.54)
Writing into the matrix form :

 
xt
 ẋt 
 
θ 
col    
  θ̈col Faxcol
j7  θ̇col 
h i  h i h i
j2 j3 j4 j5 j6 j j9 θ̈yaw + 3 j10
ẍt = j1 j1 0 0 j1 j1 j1 j1 θ
 + j81 0 j1 j1 0 j1
Faxyaw 
yaw 
θ̈tilt Faxtilt
θ̇yaw 
 
 
 θtilt 
θ̇tilt
(A.55)

Now putting in the tower dynamics equation back into the blade dynamics
equations (A.46) to get the final equation of blade dynamics
 
xt
 ẋt 
 
 θcol   
   
θ̈col 0 0 k11 k12 0 0 0 0  k13
 θ̇col   

θ̈yaw  = k21 k22 0 0 k23 k24 k25 k26 
 + 0 ×
θyaw 
θ̈tilt k31 k32 0 0 k33 k34 k35 k36  k37
θ̇yaw 

 
 θtilt 
θ̇tilt
 
xt
 ẋt 
 
θ 
 col  
θ̈col
  
Faxcol !
j7  θ̇col 
h i  h i h i
j2 j3 j4 j5 j6 j j9 θ̈yaw + 3 j10
j1 j1 0 0 j1 j1 j1 j1 θ
 + j18 0 j1 j1 0 j1
Faxyaw 
yaw
Faxtilt

θ̈tilt
θ̇yaw 
 
 
 θtilt 
θ̇tilt
 
Faxcol
Rae Faxyaw  (A.56)
+ 2
Jb + mb Rcog
Faxtilt
Simplification of the equation will give us:

 
k13 j8
1− 0 − k13 j9 
j1 j1 θ̈col
0 1 0  θ̈yaw  =
 

− k37 j8 k17 j9
j1 0 1 − j1 θ̈tilt
 
k13 j2 k 3 j3 k13 j4 k13 j5 k13 j6 k13 j7
j1 j1 k11 k12 j1 j1 j1 j1
k21 k22 0 0 k23 k24 k25 k26
 
 
k31 + k37
j
j2 k37 j3
k32 + j1 0 0 k33 + k37
j1
j4
k34 + k37
j1
j5
k35 + k37
j1
j6 k37 j7
k36 + j1
1 
xt
 ẋt 
 
θ   
 col  3k13 k13 j10
0 !F 
axcol
 j1 j1 Rae
 θ̇ 
 col 
+  0 0 0 + Faxyaw  (A.57)

θyaw 
 2
Jb + mb Rcog
3k37 k37 j10 Faxtilt
j1 0 j1
θ̇yaw 
 
 
 θtilt 
θ̇tilt

Using symbols for simplicity :

 
xt
 ẋt 
 
θ 
  col   
θ̈col 
 θ̇  Faxcol
Zblade θ̈yaw  = Ablade  col  + Bblade Faxyaw  (A.58)
 
θyaw 
θ̈tilt Faxtilt
θ̇yaw 
 
 
 θtilt 
θ̇tilt
 
xt " #
Ablade  ẋt  Bblade
θ̈bcm = θbcm  + Zblade Faxcm
  (A.59)
Zblade
θ̇bcm
Linearised Blade Dynamics
 
xt " #
 ẋt 
θ̈bcm = Xblade 
θbcm  + Yblade Faxcm

θ̇bcm

Putting in the value of θ̈bcm from equation(A.58) back into equation(A.55)

to get the final simplification.
 
xt
 ẋt 
 
θ 
 col   
h i  θ̇  h i Faxcol
j2 j3 j4 j5 j6 j7  col  j
ẍt = j1 j1 0 0 j1 j1 j1 j1   + j31 0 j10 Faxyaw  +
θyaw  1
Faxtilt
θ̇yaw 
 
 
 θtilt 
θ̇tilt
 −1
1 − k13 j8
− k13 j9

h i j1 0 j1
j8 j9
j1 0 j1 0 1 0
 
 
− k37 j8 k17 j9
j1 0 1 − j1
 
k13 j2 k 3 j3 k13 j4 k13 j5 k13 j6 k13 j7
j1 j1 k11 k12 j1 j1 j1 j1
k21 k22 0 0 k23 k24 k25 k26
 
 
k31 + k37 j2 k37 j3
k33 + k37 j4
k34 + k37 j5
k35 + k37 j6
k36 + k37 j7
j1 k32 + j1 0 0 j1 j1 j1 j1
 
xt
 ẋt 
 
θ   −1  
 col  1 − k13 j8
0 − k13 j9 3k13
0 k13 j10 !F !
j1 j1 j1 j1 axcol
 θ̇ 
 col   Rae
+ 0 1 0  0 0 0 + Faxyaw 
  
θyaw 
  2
Jb + mb Rcog
k37 j8 k17 j9 3k37 k37 j10 Faxtilt
− j1 0 1 − j1 j1 0 j1
θ̇yaw 
 
 
 θtilt 
θ̇tilt
(A.60)

Simplification of above equation

 −1
k13 j8
h i h i 1− j1 0 − k13
j1
j9
j2 j3 j4 j5 j6 j7
ẍt = 0 0 + jj81 0 j9
0 1 0
 
j1 j1 j1 j1 j1 j1 j1  
− k37 j8
1 − k17 j9
j1 0 j1
 
k13 j2 k 3 j3 k13 j4 k13 j5 k13 j6 k13 j7
j1 j1 k11 k12 j1 j1 j1 j1
!
k21 k22 0 0 k23 k24 k25 k26
 
 
k31 + k37 j
j2 k37 j3
k32 + j1 0 0 k33 + k37
j1
j4
k34 + k37
j1
j5
k35 + k37
j1
j6
k36 + k37
j1
j7
  1
xt
 ẋt 
 
θ   −1  
k13 j8
 col  1− j1 0 − k13
j1
j9 3k13
j1 0 k13 j10
j1
!!
 θ̇ 
 col 
h
j8 j9
i Rae
+ 0 ‘ 0 1 0  0 0 0 +
   
θyaw  j1 j1 2
Jb + mb Rcog
 
− k37 j8
1 − k17 j9 3k37 k37 j10

j1 0 j1 j1 0 j1
θ̇yaw 
 
 
 θtilt 
θ̇tilt
 
Faxcol
Faxyaw  (A.61)
Faxtilt
Now using symbols for the sake of simplicity

 
xt
 ẋt 
 
θ 
 col   
 θ̇  Faxcol
ẍt = Xtower  col  + Ytower Faxyaw  (A.62)
 
θyaw 
Faxtilt
θ̇yaw 
 
 
 θtilt 
θ̇tilt
Linearised Tower Dynamics

 
xt " #
 ẋt 
ẍtcm = Xtower 
θbcm  + Ytower Faxcm

θ̇bcm

At output side, we will get the velocities for all three blades and the
flapwise bending moments of all three blades. We know that blade speed is
given by the following equation.
yb = xt + θb Rae + θt cos(ψ)Rb (A.63)
Taking the derivative of the equation and converting it for all three blades
vb = ẋt + θ̇b Rae + θ̇t cos(ψ)Rb − θt ω sin(ψ)Rb (A.64)
xt ẋt
Lt Lt
ωR R
b b
vb = − sin(ψ) xt + cos(ψ) ẋt + θ˙b Rae (A.65)
Lt Lt
Using symbols for simplicity
vb = ai xt + bi ẋt + θ˙bi Rae (A.66)
For all three blades

xt
 
   
vb1 a1 b1 Rae 0 0  ẋt 


vb2  = a2 b2 0 Rae 0  θ̇b1  (A.67)
 
vb3 a3 b3 0 0 Rae θ̇b2 
θ̇b3
Now taking the Coleman Transform of both sides
xt
 
   
v
i b1 a
i 1 1b Rae 0 0  ẋt 
h h  
−1  −1 
CM vb2 = CM
 a2 b2 0 Rae 0 θ̇b1 
 
 (A.68)
vb3 a3 b3 0 0 Rae θ̇b2 
θ̇b3
h i
1 1 1
 
3 3 3
 2 sin(ψt) 2
sin(ψt + 2π 2
sin(ψt + 4π
3 ) 3 )

3 3 3
2 2
3 cos(ψt) 3 cos(ψt + 2π
3 )
2
3
4π
cos(ψt + 3 )
Simplification of right hand side gives us:
 
1 1 1
0 1 3 Rae 3 Rae 3 Rae
 ωR 2 2 2
= − Ltb 0 sin ψ1 Rae sin ψ2 Rae sin ψ3 Rae 
3 3 3 
Rb 2 2 2 2

0 Lt 3 cos ψ1b1 + cos ψ2b2 + cos ψ3b3 3 cos ψ1 Rae 3 cos ψ2 Rae 3 cos ψ3 Rae

So equation for the blade velocities will be :

 
  ẋt + 13 Rae θ̇1 + θ̇2 + θ̇3
vcol  
− ωR
vyaw  =  b
x + 2
R sin ψ θ̇ + sin ψ θ̇ + sin ψ θ̇ (A.69)

t ae 1 1 2 2 3 3
 Lt 3 
vtilt

Rb 2
Lt ẋt + 3 Rae cos ψ1 θ̇1 + cos ψ2 θ̇2 + cos ψ3 θ̇3
We know that
h i h i
θcm = CM −1 θbi <=> θbi = CM θcm
So,
1
θcol = θ1 + θ2 + θ 3 (A.70)
3
2
θyaw = sin ψ1 θ1 + sin ψ2 θ2 + sin ψ3 θ3 (A.71)
3
2
θtilt = cos ψ1 θ1 + cos ψ2 θ2 + cos ψ3 θ3 (A.72)
3
1
θ̇col = θ̇1 + θ̇2 + θ̇3
3
3θ̇col = θ̇1 + θ̇2 + θ̇3 (A.73)
!
2
θ̇yaw = sin ψ1 θ̇1 +sin ψ2 θ̇2 +sin ψ3 θ̇3 +ω cos ψ1 θ1 +cos ψ2 θ2 +cos ψ3 θ3
3
2
θ̇yaw = sin ψ1 θ̇1 + sin ψ2 θ̇2 + sin ψ3 θ̇3 + ωθtilt
3
2
θ̇yaw − ωθtilt = sin ψ1 θ̇1 + sin ψ2 θ̇2 + sin ψ3 θ̇3 (A.74)
3
!
2
θ̇tilt = cos ψ1 θ̇1 +cos ψ2 θ̇2 +cos ψ3 θ̇3 −ω sin ψ1 θ1 +sin ψ2 θ2 +sin ψ3 θ3
3
2
θ̇tilt = cos ψ1 θ̇1 + cos ψ2 θ̇2 + cos ψ3 θ̇3 − ωθyaw
3
2
θ̇tilt + ωθyaw = cos ψ1 θ̇1 + cos ψ2 θ̇2 + cos ψ3 θ̇3 (A.75)
3

Putting the values in equation(A.69)

 
  ẋt + 13 Rae 3θ̇col
vcol  
− ωR
vyaw  =  b
x + R θ̇ − ωθ (A.76)

t ae yaw tilt 
 Lt
vtilt

Rb
Lt ẋt + Rae θ̇tilt + ωθyaw
Simplification:
 0
xt
 ẋt 
 
 θcol 
0 1 0 Rae 0 0 0 0 
  
vcol
 
ωR
vyaw  = − L b θ̇ 
col 
0 0 0 0 Rae −ωRae 0 
t
θyaw 
 
Rb
vtilt 0 Lt 0 0 ωRae 0 0 Rae 
θ̇yaw 

 
 θtilt 
θ̇tilt
(A.77)
Linearised Blade Velocities
 
xt
 ẋt 
vcm = Xvcm 
θcm 

θ̇cm

Flap wise bending moments of blades are the function of joint moments
for all three blades and the reaction force applied by the joints in the opposite
direction to the wind force which is given by:
Mf lap = Rb Fj − Mj
Mf lap = sb θb +db θ̇b −sb cos(ψb )θt −db cos(ψb )θ̇t −db ω sin(ψb )θt +Rb Fj (A.78)
xt ẋt
Lt Lt
" #
sb cos(ψ) + db ω sin(ψ) db
Mf lap = − xt − cos(ψ)ẋt + sb θb + db θ̇b + Rb Fj
Lt Lt
(A.79)
Using symbols for simplicity
" #
sb cos(ψ) + db ω sin(ψ)
a=−
Lt
db
b=− cos(ψ)
Lt
Therefore :
 
xt
 ẋt 
 
θb 

0  1

a1 b1 sb db 0 0 0
θ̇ 
Mf lap = a1 b1 0 0 sb db 0
 0   b1  + Rb Fj (A.80)
θ 
a1 b1 0 0 0 0 sb db  b2 
θ̇b2 
 
θb3 
θ̇b3
Coleman Transformation of each side:

 
xt
 ẋt 
 
θb 

" # a
1 b1 sb db 0 0 0 0  1 " #
− θ̇ b1 −
Mf lapcm = CM 1 a1 b1 0 0 sb db 0 0   +Rb CM 1 Fj
  
θ 
a1 b1 0 0 0 0 sb db  b2 
θ̇b2 
 
θb3 
θ̇b3
(A.81)

h i
1 1 1
 
3 3 3
 2 sin(ψt) 2
sin(ψt + 2π 2
sin(ψt + 4π
3 ) 3 )

3 3 3
2 2
3 )
2
3
4π
cos(ψt + 3 )
Simplification by using equation(A.22,A.23,A.24,A.25,A.26,A.27) will give

us :
 
" # sb θcol + db θ̇col
CM − 1 Mf lap =  − dLb tω xt + sb θyaw + db θ̇yaw − db ωθtilt  (A.82)
 
− Lsbt xt − Ldbt ẋt − sb θtilt + db θ̇tilt + db ωθyaw
Putting the above equation back in equation A.81 and also placing in the
value of joint force Fj .
 
sb θcol + db θ̇col
Mf lapcm =  − dLb tω xt + sb θyaw + db θ̇yaw − db ωθtilt 
 
− Lsbt xt − Ldbt ẋt − sb θtilt + db θ̇tilt + db ωθyaw
  
0

" 0 !#
Rb 2
2ω Lt ẋt + Lb θ̇tilt  +  −ω
Lb θyaw 
  
+ Rb Fax − mb ẍbcm − 

−ω R
  2
Lt xt + Lb θb
b
−2ω Lb θ̇yaw
(A.83)
Further simplification will give us :
 
sb θcol+ db θ̇col
2ωmb Rb2

 db ω
− x + ẋt + sb + ω 2 mb Lb Rb θyaw + db θ̇yaw − ωdb θtilt + 2ωmb Lb Rb θ̇tilt 

Mf lapcm = 
 2 Lt 2t Lt

ω mb Rb −sb
Lt xt − Ldbt ẋt + ωdb θyaw − 2ωmb Lb Rb θ̇yaw + sb + ω 2 mb Lb Rb θtilt + db θ̇tilt
 
Rb mb
−  0  ẍt − Rb mb Lb θ̈cm + Rb Faxcm (A.84)
 
mb Rb2
Lt

Now putting the values of Tower dynamics(ẍt ) and Blade dynamics(θ̈b ) from
equation(A.63) and equation(A.60) respectively.
Mf lapcm =
 
0 0 sb db 0 0 0 0
2ωmb Rb2

− dLb tω 0 0 sb + ω 2 mb Lb Rb db −ωdb 2ωmb Lb Rb 
 
 Lt
ω 2 mb Rb2 −sb
 
Lt − Ldbt 0 0 ωdb −2ωmb Lb Rb sb + ω 2 mb Lb Rb db
 
xt
 ẋt 
 
θ    
xt
 
xt

 col  Rb mb ! !
 θ̇ 
 col   0   ẋ t 
 Xblade  ẋt  Yblade
− Xtower  +Ytower Faxcm −Rb mb Lb θbcm + Zblade Faxcm
 
θyaw  θbcm Zblade
 2

mb Rb
 
θ̇yaw  Lt θ̇bcm θ̇bcm
 
 
 θtilt 
θ̇tilt
+ Rb Faxcm (A.85)
Simplification gives us :
 

Rb mb

! x t
 0  Xblade  ẋt 

Mf lapcm = CMf lap −  Xtower − Rb mb Lb
Zblade θbcm 

mb Rb2
Lt θ̇bcm
 
Rb mb !
 0  Yblade
−  Ytower + Rb mb Lb − Rb Faxcm (A.86)
Zblade

mb Rb2
Lt
Linearised Flapwise Bending Moments

 
xt " #
 ẋt 
Mf lapcm  θcm  + Yf lap Faxcm
= Xf lap  
θ̇xm

Appendix B
Coleman’s Transformation
Calculation
Coleman Transformation Method [21] transforms all quantities on the ro-

tating reference frame to a fixed reference frame and results into a linear
time invariant (LTI) model, which is easy to handle and very suitable for
implementation of different control strategies.
Flap wise bending movement of the wind turbine blade is the movement
of the blade out of the rotor plane. These movements are one of the ma-
jor causes of the fatigue loading of the wind turbine and can be catas-
trophic in extreme wind case. These flap wise bending movements of all
three blades have significant importance in a control system design of wind
turbine. These movements are the function of pitch angle and wind speed
experienced by each blade.
(1) (2) (3)

Flap wise bending movements Mf Mf Mf
Controls (Pitch angles) θ(1) θ(2) θ(3)

(1) (2) (3)
Disturbances (wind experienced by each blade) Uax Uax Uax
(i) (i)
∴ Mf = f (θ(i) , Uax )
Linerization
(i) (i)
δMf = a(t)δθ(i) + b(t)δUax
where
(i) (i) (i)
δMf = Mf,equilibrium − Mf
and [a(t), b(t)] are time dependent coefficients.

Coleman’s Transformation Calculation
Here is an example of Coleman Transformation in which three sin waves are

considered to be the three flap wise bending movements of a wind turbine
which are 120◦ apart from each other, so the sine waves are also 120◦ out of
phase from each other.
 (1)  
Mf

sin(ωt)
 (2)  
Mf  = sin(ωt + 2π 3 )

4π
M
(3) sin(ωt + 3 )
f
where ω is the angular frequency of the rotor i.e.
ω = Constant × Azimuth angle(ψ) = Constant × Rotational Speed × time
Rotational speed of the rotor is considered to be 12.1RP M
Coleman’s Transformation Matrix

1 1 1
 
3 3 3
 2 sin(ψt) 2
sin(ψt + 2π 2
sin(ψt + 4π
3 ) 3 )

3 3 3
2 2
3 )
2
3
4π
cos(ψt + 3 )
Multiplication of flap wise bending signal matrix with Coleman’s Transfor-

mation matrix will result into the Tilt and Yaw angles
1 1 1
     
3 3 3 sin(ωt) 0
 2 sin(ψt) 2 sin(ψt + 2π ) 2 sin(ψt + 4π )  × sin(ωt + 2π ) =  T ilt 
3 3 3 3 3 3
2 2 2π 2 4π
3 cos(ψt) 3 cos(ψt + 3 ) 3 cos(ψt + 3 ) sin(ωt + 4π3 ) Y aw
Solution : Taking the right hand side

1 1 1
   
3 3 3 sin(ωt)
=  23 sin(ψt) 2
3 sin(ψt + 2π
3 )
2 4π   2π 
3 sin(ψt + 3 ) × sin(ωt + 3 )
2 2 2π 2 4π
3 cos(ψt) 3 cos(ψt + 3 ) 3 cos(ψt + 3 ) sin(ωt + 4π3 )
1
sin(ωt) + 13 sin(ωt + 2π 1 4π
 
3 3 ) + 3 sin(ωt + 3 )
 2 sin(ψt) × sin(ωt) + 2 sin(ψt + 2π ) × sin(ωt + 2π ) + 2 sin(ψt + 4π ) × sin(ωt + 4π ) 
3 3 3 3 3 3 3
2 2 2π 2π 2 4π 4π
3 cos(ψt) × sin(ωt) + 3 cos(ψt + 3 ) × sin(ωt + 3 ) + 3 cos(ψt + 3 ) × sin(ωt + 3 )

Solving first row of the matrix
1 1 2π 1 4π
R1 = sin(ωt) + sin(ωt + ) + sin(ωt + )
3 3 3 3 3 ! !
1 1 2π 2π 1 4π 4π
R1 = sin(ωt) + sin(ωt) cos( ) + cos(ωt) sin( ) + sin(ωt) cos( ) + cos(ωt) sin( )
3 3 3 3 3 3 3
(((( ((((!
1 1 2π (((((( 2π 1 4π (((((( 4π
R1 = sin(ωt) 1 + sin(ωt) cos(
((() + cos(ωt) sin( ) + sin(ωt) cos(
((() + cos(ωt) sin( )
3 3 ((((( 3
( 3 3 ((((( 3
( 3
1
R1 = sin(ωt) 0
3
R1 = 0
R1 = 0
Solving second row of the matrix
2 2 2π 2π 2 4π 4π
R2 = sin(ωt) sin(ψt) + sin(ωt + ) sin(ψt + ) + sin(ωt + ) sin(ψt + )
3 3 3 3 3 3 3 !
2 2
R2 = sin(ωt) sin(ψt) + (−0.499 sin(ψt) + 0.866 cos(ψt)) × (−0.499 sin(ωt) + 0.866 cos(ωt))
3 3
!
2
+ (−0.5 sin(ψt) − 0.866 cos(ψt)) × (−0.5 sin(ωt) + 0.866 cos(ωt))
3
R2 = sin(ωt) sin(ψt) + cos(ωt) cos(ψt)
R2 = cos t(ω − ψ)
R2 = cos t(ω − ψ)
Solving third row of the matrix
2 2 2π 2π 2 4π 4π
R3 = sin(ωt) cos(ψt) + sin(ωt + ) cos(ψt + ) + sin(ωt + ) cos(ψt + )
3 3 3 3 3 3 3 !
2 2
R3 = sin(ωt) cos(ψt) + (−0.499 cos(ψt) + 0.866 sin(ψt)) × (−0.499 sin(ωt) + 0.866 cos(ωt))
3 3
!
2
+ (−0.5 cos(ψt) + 0.866 sin(ψt)) × (−0.5 sin(ωt) − 0.866 cos(ωt))
3
2 2
R3 = sin(ωt) cos(ψt) 1 + 0.25 + 0.25 − cos(ωt) sin(ψt) 0.749 + 0.749
3 3
R3 = sin(ωt) cos(ψt) − cos(ωt) sin(ψt)
R3 = sin t(ω − ψ)

This is the final Coleman’s transformation matrix which is helpful for

control strategies implementation.
 
 0 
 
T ransf ormed Signal = 
cos t(ω − ψ)

 
sin t(ω − ψ)
Figure B.1: Coleman Transformation & Inverse Coleman Transformation

eaxmple using Matlab script

Matrix Calculations
First Row:
2π 4π
cos ψ1 + cos ψ2 + cos ψ3 = cos(ψt) + cos(ψt + ) + cos(ψt + )
3! 3 !
2π 2π 4π 4π
= cos(ψt) + cos(ψt) cos( ) − sin(ψt) sin( ) + cos(ψt) cos( ) − sin(ψt) sin( )
3 3 3 3
! !
2π 4π 2π 4π
= cos(ψt) 1 + cos( ) + cos( ) − sin(ψt) sin( ) + sin( )
3 3 3 3
=0
&
2π 4π
) + sin(ψt +
sin ψ1 + sin ψ2 + sin ψ3 = sin(ψt) + sin(ψt + )
3 ! 3 !
2π 2π 4π 4π
= sin(ψt) + sin(ψt) cos( ) + cos(ψt) sin( ) + sin(ψt) cos( ) + cos(ψt) sin( )
3 3 3 3
! !
2π 4π 2π 4π
= sin(ψt) 1 + cos( ) + cos( ) + cos(ψt) sin( ) + sin( )
3 3 3 3
=0

Second Row:
sin ψ1 cos ψ1 + sin ψ2 cos ψ2 + sin ψ3 cos ψ3

2π 2π 4π 4π
= sin(ψt) cos(ψt) + sin(ψt + ) cos(ψt + ) + sin(ψt + ) cos(ψt + )
3 3 3 3 !
2π 2π 2π 2π
= sin(ψt) cos(ψt) + sin(ψt) cos( ) + cos(ψt) sin( ) cos(ψt) cos( ) − sin(ψt) sin( ) +
3 3 3 3
!
4π 4π 4π 4π
sin(ψt) cos( ) + cos(ψt) sin( ) cos(ψt) cos( ) − sin(ψt) sin( )
3 3 3 3
!
2π 2 2π 2 4π 2 4π 2
= sin(ψt) cos(ψt) 1 + (cos( )) − (sin( )) + (cos( )) − (sin( )) +
3 3 3 3
!
2π 2π 4π 4π
cos2 (ψt) − sin2 (ψt) sin( ) cos( ) + sin( ) cos( )
3 3 3 3
!
4π 8π 2π 2π
= sin(ψt) cos(ψt) 1 + cos( ) + cos( ) + cos2 (ψt) − sin2 (ψt) sin( ) cos( )+
3 3 3 3
!
4π 4π
sin( ) cos( )
3 3
=0
&
2π 4π
sin2 (ψt) + sin2 (ψt + ) + sin2 (ψt + )
3 3
1 − cos 2(ψt) 1 − cos 2(ψt + 2π 3 ) 1 − cos 2(ψt + 4π
3 )
= + +
2 2 2
3 1 4π 8π 4π 8π
= − cos(2ψ) 1 + cos( ) + cos( ) − sin(2ψ) sin( ) + sin( )
2 2 3 3 3 3
3
=
2
Third Row:
2π 4π
cos2 (ψt) + cos2 (ψt + ) + cos2 (ψt + )
3 3
1 + cos 2(ψt) 1 + cos 2(ψt + 2π 3 ) 1 + cos 2(ψt + 4π
3 )
= + +
2 2 2
3 1 4π 8π 4π 8π
= cos(2ψ) 1 + cos( ) + cos( ) − sin(2ψ) sin( ) + sin( )
22 3 3 3 3
3
=
2

Appendix C
Matlab Script
Detailed scripts are not shown here.
D=dic.get;
opcurve=D(’opcurve’);
Vw = si(12,’m/s’);
pVctr = coreo peratingp oint(opcurve, V w);
Vwc tr = pV ctr.V ;
T hetaCtr = pV ctr.T heta;
LambdaCtr = pV ctr.Lambda;
OmegaCtr = pV ctr.Omega;
Rs = OmegaCtr;
turb=corea ssemblet urb;

turb.pitch.Sys = D(0 P itchActuatorSys0 );
CqApprox=turb.rot.CqApprx;
CtApprox=turb.rot.CtApprx;
, dCt
= coret akec oef f icientd erivatives(CqApprox, CtApprox, T hetaCtr, LambdaCtr);
dFaxd theta = 1/6 ∗ turb.rho ∗ pi ∗ turb.rot.R2 ∗ V w2 ∗ dCt.dT heta;
Ct = coree valapprx(f itCt, double(LambdaCtr), T hetaCtr,0 extrapolate0 );

dF axd uax = 1/6 ∗ turb.rho ∗ pi ∗ turb.rot.R2 ∗ V w ∗ (2 ∗ Ct − (LambdaCtr ∗
(dCt.dLambda)));
%———————————————————

Matlab Script
atower = ( (9/(4*H)) * ( (mb*Rb2 )/Lt)) + mt + (3 ∗ mb);

ptower = −(((9/(4 ∗ H)) ∗ (((mb ∗ Rb2 ∗ Rs2 ) + Sb)/Lt)) + St);
qtower = −(((9/(4 ∗ H)) ∗ (Db/Lt)) + Dt);
rtower = −((9/(4 ∗ H)) ∗ (Db ∗ Rs));
stower = ((9/(4 ∗ H)) ∗ (2 ∗ mb ∗ Rcog ∗ Rb ∗ Rs));
ttower = ((9/(4 ∗ H)) ∗ ((Sb) − (mb ∗ Rb ∗ Rcog ∗ Rs2 )));
utower = ((9/(4 ∗ H)) ∗ (Db));
vtower = −(3 ∗ mb ∗ Rcog);
wtower = −((9/(4 ∗ H)) ∗ (mb ∗ Rcog ∗ Rb));
xtower = ((9/(4 ∗ H)) ∗ Rb);
%———————————————————
k1blade = -Sb / ( Jb +(mb*Rcog2 ));

k2blade = −Db/(Jb + (mb ∗ Rcog 2 ));
k3blade = −(mb ∗ Rcog)/(Jb + (mb ∗ Rcog 2 ));
l1blade = - (Db*Rs) / ( Lt*( Jb + (mb*Rcog2 )));

l2blade = (2 ∗ mb ∗ Rcog ∗ Rb ∗ Rs)/(Lt ∗ (Jb + (mb ∗ Rcog 2 )));
l3blade = ((Jb ∗ Rs2 ) − Sb + (mb ∗ Rcog 2 ∗ Rs2 ))/(Jb + (mb ∗ Rcog 2 ));
l4blade = −Db/(Jb + (mb ∗ Rcog 2 ));
l5blade = (Db ∗ Rs)/(Jb + (mb ∗ Rcog 2 ));
l6blade = 2 ∗ Rs;
m1blade = ( Sb+(mb*Rcog*Rb*Rs2 ))/(Lt ∗ (Jb + (mb ∗ Rcog 2 )));

m2blade = Db/(Lt ∗ (Jb + (mb ∗ Rcog 2 )));
m3blade = −(Db ∗ Rs)/(Jb + (mb ∗ Rcog 2 ));
m4blade = −2 ∗ Rs;
m5blade = ((Jb ∗ Rs2 ) − Sb + (mb ∗ Rcog 2 ∗ Rs2 ))/(Jb + (mb ∗ Rcog 2 ));
m6blade = −Db/(Jb + (mb ∗ Rcog 2 ));
m7blade = −(mb ∗ Rcog ∗ Rb)/(Lt ∗ (Jb + (mb ∗ Rcog 2 )));
%———————————————————
% Blade Dynamics
Xxb lade = [(k3blade∗ptower)/atowerk1blade(k3blade∗rtower)/atower(k3blade∗

ttower)/atower(k3blade∗qtower)/atowerk2blade(k3blade∗stower)/atower(k3blade∗
utower)/atower; l1blade0l3bladel5bladel2blade0l4bladel6blade; m1blade+((m7blade∗
ptower)/atower)0m3blade+((m7blade∗rtower)/atower)m5blade+((m7blade∗
ttower)/atower)m2blade+((m7blade∗qtower)/atower)0m4blade+((m7blade∗
stower)/atower)m6blade + ((m7blade ∗ utower)/atower); ];
Yyb lade = [((3∗k3blade)/atower)+(Rae/(Jb+(mb∗Rcog 2 )))0(k3blade∗
xtower)/atower; 0(Rae/(Jb+(mb∗Rcog 2 )))0; (3∗m7blade)/atower0((m7blade∗

Matlab Script
xtower)/atower) + (Rae/(Jb + (mb ∗ Rcog 2 ))); ];

Zzb lade = [1−((k3blade∗vtower)/atower)0−(k3blade∗wtower)/atower; 010; −(m7blade∗
vtower)/atower01 − ((m7blade ∗ wtower)/atower); ];
% Transformed Blade Dynamics
Aab lade = Zzb lade− 1 ∗ Xxb lade;
Bbb lade = Zzb lade− 1 ∗ Y yb lade;
% Tower Dynamics
Aat ower = [ptower/atower0rtower/atowerttower/atowerqtower/atower0stower/atowerutower
([vtower/atower0wtower/atower] ∗ Aab lade);
Bbt ower = ([vtower/atower0wtower/atower]∗Bbb lade)+[3/atower0xtower/atower];
Xxd ott = [00001000]; T hetad otb = [00000100; 00000010; 00000001; ];
% Velocities
CV o ut = [00001Rae00; −Rs ∗ (Rb/Lt)00 − Rae ∗ Rs00Rae0; 00Rae ∗

Rs0Rb/Lt00Rae; ];
% Flapping moment of blades

CM f fo ut = [0Sb000Db00; −(Rs ∗ Db)/Lt0Sb + (Rs2 ∗ mb ∗ Rcog ∗ Rb) − Rs ∗
Db2 ∗ Rs ∗ mb ∗ Rb2 0Db2 ∗ mb ∗ Rcog ∗ Rb; ((Rs2 ∗ mb ∗ Rb2 ) − Sb)/Lt0Rs ∗
DbSb + (Rs2 ∗ mb ∗ Rcog ∗ Rb) − Db/Lt0 − 2 ∗ mb ∗ Rcog ∗ RbDb; ];
Co ut1 = [mb ∗ Rb; 0; (mb ∗ Rb2 )/Lt; ];
CM f fb lade = CM f fo ut−(Co ut1∗Aat ower)−((Rb∗mb∗Rcog)∗Aab lade);
% Flapping moment of tower
CM f fto ut = (((mt ∗ Rcogt)/Lt) ∗ Aat ower) + (mb ∗ Aat ower) + ((mb ∗

Rcog) ∗ Aab lade(1, :));
CM f ft ower = [−St ∗ Lt000 − Dt ∗ Lt000] + CM f fto ut;
Cxd ott = [00001000];
Ct hetab cm = [zeros(3, 5)eye(3)];
% D matrix for blades

Matlab Script
Do utb lade = −(Co ut1∗Bbt ower)−(Rb∗mb∗Rcog)∗Bbb lade+Rb∗eye(3);
% D matrix for tower

Do ut3 = (((mt ∗ Rcogt)/Lt) ∗ Bbt ower) + (mb ∗ Bbt ower) + ((mb ∗ Rcog) ∗
Bbb lade(1, :));
Do utt ower = Do ut3 − [300];
Dxd ot = zeros(1, 3);
Dt hetab cm = zeros(3, 3);
%———————————————————
% State Space Model
A = [ Xxd ott ; T hetad otb ; Aat ower; Aab lade; ];
B = [ zeros(1,3) ; zeros(3,3) ; Bbt ower; Bbb lade; ];
C = [ CV o ut; CM f fb lade; CM f ft ower; Cxd ott ; Ct hetab cm; ];
D = [ zeros(3,3) ; Do utb lade; Do utt ower; Dxd ot; Dt hetab cm; ];
%———————————————————
Statespacet urbine = ss(A, B, C, D);
%Feedback Velocities
Ac = A -( B * (dFaxd uax ∗ CV o ut));
Bc = [B * dFaxd uaxB ∗ dF axd theta];
Cc = C - ( D * (dFaxd uax ∗ CV o ut));
Dc = [D*dFaxd uaxD ∗ dF axd theta];
Statespacef bl = ss(Ac, Bc, Cc, Dc);
%———————————————————
Intg = [ 0 0 0 ; 0 tf(1,[1,0]) 0 ; 0 0 tf(1,[1,0]) ; ];

Matlab Script
% IPC design filters
a = [1 1 1 0 0];
b = [1 1 1 0.6 0.6];
w = Rs * [3 6 9 9 9];
for j = 1:5
F(j) = tf([a(j) 0 w(j)2 ], [1b(j) ∗ w(j)w(j)2 ]);
end
Filter = F(1) * F(2) * F(3) * F(4) * F(5);
% Adding wind disturbance

Azimuth = zeros(3,length(Winddata));
Wind = zeros(3,length(Winddata));
for k = 1:length(Winddata)
Azimuth= squeeze(Azimuthdata(:,:,k));
ICM= [ 1/3, 1/3, 1/3; (2/3)*(sin (Azimuth)), (2/3)*(sin(Azimuth+(2*pi/3))),

(2/3)*(sin(Azimuth+(4*pi/3))); (2/3)*(cos (Azimuth)), (2/3)*(cos(Azimuth+(2*pi/3))),
(2/3)*(cos(Azimuth+(4*pi/3))); ];
Wind(:,k) = ICM * squeeze(Winddata(:,:,k));
end
%———————————————————
% Pitch Activity Calculations with Regular Gain with Filters
Modf = f eedback(Statespacef bl, −8.5193e − 09 ∗ (ss(Intg)) ∗ F ilter, 4 :

6, 4 : 6);
Modf c omplete = c2d(M odf , T s);
[YM odf c omplete, TM odf c omplete, XM odf c omplete] = lsim(M odf c omplete(:
, 1 : 3), W ind);
Modw f = f eedback(Statespacef bl, −8.5193e − 09 ∗ (ss(Intg)), 4 : 6, 4 :

6);

Matlab Script
Modw fc omplete = c2d(M odw f, T s);
[YM odw fc omplete, TM odw fc omplete, XM odw fc omplete] = lsim(M odw fc omplete(:
, 1 : 3), W ind);
ModelP itchactivityr gf = c2d(−8.5193e−09∗(ss(Intg))∗SelectT iltY aw∗

F ilter, T s);
M odelP itchactivityr gw f = c2d(−8.5193e−09∗(ss(Intg))∗SelectT iltY aw, T s);
[Yp ar gf , Tp ar gf , Xp ar gf ] = lsim(M odelP itchactivityr gf , YM odf c omplete(:

, 4 : 6));
Yp ar gw f, Tp ar gw f, Xp ar gw f
= lsim(M odelP itchactivityr gw f, YM odw fc omplete(:, 4 : 6));
CM = [ 1 sin(Azimuth) cos(Azimuth); 1 sin(Azimuth+(2*pi/3)) cos(Azimuth+(2*pi/3));

1 sin(Azimuth+(4*pi/3)) cos(Azimuth+(4*pi/3)); ];
Yp am odelr gf p itchp osition(:, k) = CM ∗ Yp ar gf (k, :)0 ;

Yp am odelr gw fp itchp osition(:, k) = CM ∗ Yp ar gw f (k, :)0 ;
end
Yp am odelr gf p itchs peedc ol = (1/T s) ∗ dif f (Yp ar gf , 1, 2);

Yp am odelr gw fp itchs peedc ol = (1/T s) ∗ dif f (Yp ar gw f, 1, 2);
Yp am odelr gf p itchs peedr ot = (1/T s)∗dif f (Yp am odelr gf p itchp osition, 1, 2);
Yp am odelr gw fp itchs peedr ot = (1/T s)∗dif f (Yp am odelr gw fp itchp osition, 1, 2);
%———————————————————
% Gain variation
g1= 8.5193e-9/10;
g2= 8.5193e-9 * 5;
g = -logspace(log10(g1),log10(g2),400);
for kgb=1:length(g);

Matlab Script
IPCgain = g(kgb);
IPC = IPCgain *(ss(Intg)*Filter);
Model = feedback(Statespacef bl, IP C, 4 : 6, 4 : 6);
Model2 = feedback(Statespacef bl, IP Cgain ∗ (ss(Intg)), 4 : 6, 4 : 6);
[Am,Bm,Cm,Dm] = ssdata(Model);
ModelD = c2d(Model, Ts);

Yo ut, Tt ime, Xs tates
= lsim(M odelD(:, 1 : 3), W ind);
ModelD2 = c2d(Model2, Ts);

Yo ut2, Tt ime2, Xs tates2
= lsim(M odelD2(:, 1 : 3), W ind);
%———————————————————
% Pitch Activity Calculations with Varying Gain with filters
ModelP itchactivityv g = c2d(IP Cgain ∗ (ss(Intg)) ∗ SelectT iltY aw ∗

F ilter, T s);
M odelP itchactivityv gw f = c2d(IP Cgain ∗ (ss(Intg)) ∗ SelectT iltY aw, T s);
[Yp av gf , Tp av gf , Xp av gf ] = lsim(M odelP itchactivityv g, Yo ut(:, 4 : 6));

Yp av gw f, Tp av gw f, Xp av gw f
= lsim(M odelP itchactivityv gw f, Yo ut2(:, 4 : 6));
% Inverse Coleman Transformation
CM = [ 1 sin(Azimuth) cos(Azimuth); 1 sin(Azimuth+(2*pi/3)) cos(Azimuth+(2*pi/3));

1 sin(Azimuth+(4*pi/3)) cos(Azimuth+(4*pi/3)); ];
Ym odel(:, k) = CM ∗ Yo ut(k, 4 : 6)0 ;

Ym odel2(:, k) = CM ∗ Yo ut2(k, 4 : 6)0 ;
Ym odel3(:, k) = CM ∗ Yo ut(k, 9 : 11)0 ;
Ym odel4(:, k) = CM ∗ Yo ut2(k, 9 : 11)0 ;

Matlab Script
Yp am odelv gf (:, k) = CM ∗ Yp av gf (k, :)0 ;

Yp am odelv gw f (:, k) = CM ∗ Yp av gw f (k, :)0 ;
end
% Blade flapping with Filters
FatigueDatag ain(:, kgb) = cored eql(Ym odel(:, 1500 : end), 50, [], [], 10);
FatigueDatag ainc oleman(:, kgb) = cored eql(Yo ut(1500 : end, 4 : 6)0 , 50, [], [], 10);
FatigueDatag ainc olemanw f (:, kgb) = cored eql(Yo ut2(1500 : end, 4 : 6)0 , 50, [], [], 10);
% Blade flapping without Filters FatigueDatag ain2(:, kgb) = cored eql(Ym odel2(:
, 1500 : end), 50, [], [], 10);
FatigueDatag aint ower(:, kgb) = cored eql(Yo ut(1500 : end, 7)0 , 50, [], [], 4);
FatigueDatag aint ower2(:, kgb) = cored eql(Yo ut2(1500 : end, 7)0 , 50, [], [], 4);
FatigueDatag ainx t(:, kgb) = cored eql(Yo ut(1500 : end, 8)0 , 50, [], [], 10);
FatigueDatag ainx t2(:, kgb) = cored eql(Yo ut2(1500 : end, 8)0 , 50, [], [], 10);
FatigueDatag aint hetab(:, kgb) = cored eql(Ym odel3(:, 1500 : end), 50, [], [], 10);
FatigueDatag aint hetab2(:, kgb) = cored eql(Ym odel4(:, 1500 : end), 50, [], [], 10);
RMSv gf (:, kgb) = rms(50 ∗ dif f (Yp am odelv gf0 ));
RMSv gw f (:, kgb) = rms(50 ∗ dif f (Yp am odelv gw f 0 ));
RMSv gc oleman(:, kgb) = rms(50 ∗ dif f (Yp av gf ));
RMSv gc olemanw f (:, kgb) = rms(50 ∗ dif f (Yp av gw f ));
end
%———————————————————
% Kalman filter for BRM and state estimation
[kalm,L,P] = kalman(Statespacef bl, Qn, Rn, [], 4 : 6, 4 : 6);
trr = [ eye(3) zeros(3) zeros(3); zeros(3) eye(3) zeros(3); zeros(3) eye(3)

zeros(3); zeros(3) zeros(3) eye(3)];

Matlab Script
Systema ppend = append(Statespacef bl, kalm);
Combineds ystem = Systema ppend ∗ trr;
FeedbackB RM = f eedback(Combineds ystem, eye(3), 7 : 9, 4 : 6, +1);
Sbrm = Statespacef bl(4 : 6, :);

SbrmW ind = Sbrm(:, 1 : 3);
SbrmT heta = Sbrm(:, 4 : 6);
kalmani nput = kalm ∗ [zeros(3), eye(3); SbrmW ind, SbrmT heta];
%———————————————————
% Varying Rho for maximum fatigue reduction
Q = Cc(5:6,:)’ * eye(2) * Cc(5:6,:);
Rhor ange = logspace(log10(1e3), log10(1e15), 400);
RMSL QRL QR = zeros(length(Rhor ange), 3);

F atigueL QR = zeros(3, length(Rhor ange));
for zz = 1:length(Rhor ange);
Rho = Rhor ange(zz);
R = eye(2) * Rho;
[K,S,E]=lqr(Statespacef bl.A, Statespacef bl.B(:, 5 : 6), Q, R, []);
kfl qrgain = blkdiag(eye(14), K) ∗ F eedbackB RM ;
lqrf eedback = f eedback(kfl qrgain, eye(2), (5 : 6), (15 : 16), −1);
lqrf eedbackD = c2d(lqrf eedback, T s);
[Yl qr, Tl qr, Xl qr] = lsim(lqrf eedbackD (:, 1 : 3), W ind);
Yl qry awtilt = [zeros(length(W ind), 1)Yl qr(:, 15 : 16)];
% Blade flapping
FatigueL QRC oleman(zz, :) = cored eql(Yl qr(1500 : end, 4 : 6)0 , 50, [], [], 10);
P itchActS T DC oleman(:, zz) = std(50 ∗ dif f (Yl qry awtilt));

Matlab Script
P itchActR M SC oleman(:, zz) = rms(Yl qry awtilt);
FatigueL QR(:, zz) = cored eql(Yl qrr otB RM (:, 1500 : end), 50, [], [], 10);
RM SL QRL QR(zz, :) = rms(Yl qrr otB RM (:, 1500 : end)0 );
P itchActS T D(:, zz) = std(50 ∗ dif f (Yl qrr otp a0 ));
P itchActR M S(:, zz) = rms(Yl qrr otp a0 );
end

Bibliography
[1] T. J. Price, “James blyth britain’s first modern wind power pioneer,”
Wind Engineering, vol. 29, no. 1, pp. 191–200, 2005.
[2] F. Buonsanti, “Promoting renewable energy in longyearbyen: A sus-

tainable means to prevent svalbard s environmental degradation,”
Consilience-The Journal of Sustainable Development, vol. 6, 2011.
[3] N. J. E. B. Toney Burton, David Shrpe, Wind energy handbook. John

Wiley & Sons Ltd, first ed., 2001.
[4] M. O. L. Hansen, Aerodynamics of Wind Turbines. James & James,

first ed., 2000.
[5] J. Wittenburg, “Dynamics of systems of rigid bodies,” Journal of Ap-

plied Mathematics and Mechanics / Zeitschrift fr Angewandte Mathe-
matik und Mechanik, vol. 59, no. 7, pp. 331–331, 1979.
[6] J. Wittenburg, Dynamics of Multibody Systems, vol. First. Springer,

second ed., October 2007.
[7] T. v. E. W. W. X. D. J. V. M. Kanev, S.K.; Engelen, “Sustainable

control,” Technical report ECN-E–12-028, ECN Wind Energy, 2012.
[8] E. A. Bossanyi, “The design of closed loop controllers for wind tur-
bines,” no. DOI: 10.1002/we.34, p. 15, 2000.
[9] E. A. Bossanyi, “Individual blade pitch control for load reduction,”

no. DOI: 10.1002/we.76, p. 10, 2003.
[10] E. A. Bossanyi, “Wind turbine control for load reduction,” no. DOI:
10.1002/we.95, p. 16, 2003.
[11] E. A. Bossanyi, “Further load reductions with individual pitch control,”

no. DOI: 10.1002/we.166, p. 5, 2005.
[12] K. E. Okedu, “Stabilization of a fixed speed wind turbine with a variable

speed wind turbine,” journal of applied science & engineering technol-
ogy, p. 8, 2011.

BIBLIOGRAPHY
[13] J. Rom, High Angle of Attack Aerodynamics: Subsonic, Transonic, and

Supersonic Flows. Springer London, Limited, 2011.
[14] N. Cetin, M. Yurdusev, R. Ata, and A. Ozdemir, “Assessment of op-

timum tip speed ratio of wind turbines,” Mathematical and Computa-
tional Applications, vol. 10, no. 1, pp. 147–154, 2005.
[15] M. PACHTER, T. KOBYLARZ, and C. H. HOUPIS, “Literal nyquist

stability criterion for mimo control systems,” International Journal of
Control, vol. 63, no. 1, pp. 55–65, 1996.
[16] S. Engstrm, “Short term power variations in the output of wind tur-
bines,”
[17] H. M. Larsen, T. and K. Thomson, “Active load reduction using indi-

vidual pitch, based on local blade flowmeasurements,” Wind Energy,
p. 6780, 2005.
[18] P. S. Van der Hooft, E. and T. van Engelen, “Wind turbine control
algorithm,” Wind Engineering, no. ECN-C03-111, 2005.
[19] S. Spada, “The reliance on wind energy depends on advancements in

blade pitch control,” ARC Insights, July 2010.
[20] S. N. Vukosavic, Electrical Machines. Springer New York Heidelberg

Dordrecht London, July 2012.
[21] R. Coleman and A. Feingold, “Theory of self-excited mechanical oscilla-

tions of helicopter rotors with hinged blades,” technical report, USGPO,
1958.
[22] T. v. E. W. Kanev, S.K.; Engelen, “Act model calculation,” tech. rep.,

Energy Research Centre of Netherlands, 2012.
[23] T. Kane and D. Levinson, “Dynamics, theory and applications,”

McGraw- Hill seriez in Mechanical Engineering, 1985.
[24] j. Holierhoek, “Aeroelasticity of large wind turbines,” PrintPartners

Ipskamp, Enschede, The Netherlands, p. 16, 2008.
[25] V. P. S. V. I. Agoshkov, P. B. Dubovsky, Methods for Solving Mathe-

matical Physics Problems. Cambridge Int Science Publishing, 2006.
[26] G. Bir, “Multiblade coordinate transformation and its application to

wind turbine analysis,” 2008.
[27] R. Vaccaro and A. Kot, “A comparison between transfer-function and

state-space approaches to signal modeling,” in Decision and Control,
1985 24th IEEE Conference on, vol. 24, pp. 1633–1634, 1985.

BIBLIOGRAPHY
[28] A. Wright and M. Balas, “Design of state space based control algorithms
for wind turbine speed regulations,” tech. rep., NREL, 2002.
[29] R. Buchi, State Space Control, Lqr and Observer. Books on Demand,
2010.
[30] P. E. E. Ostertag, Mono and multivariable control and extimation.

Springer-Verlag Berlin Heidelberg, first ed., 2011.
[31] W. M. G. Freebury, “Determining equivalent damage loading for full-

scale wind turbine blade fatigue tests,” tech. rep., National Renewable
Energy Laboratory, January 2000. Presented at the 19th American So-
ciety of Mechanical Engineers (ASME) Wind Energy Symposium,Reno,
Nevada.
[32] R. L. D. Garcia, A. Karimi, “Pid controller design for multivariable

systems using gershgorin bands,” tech. rep., Laboratoire dAutomatique,
EPFL, CH-1015 Lausanne, Switzerland, 200.
[33] T. van den Boom, “Model predictive control,”
[34] C. V. R. P. O. M. S. D. Q. Mayne, J. B. Rawlings, “Constrained

model predictive control: Stability and optimality,” Elsevier Science
Ltd., p. 26, 2000.
[35] B. Sohlberg, Applied Model Based Control. first ed., 2008.
[36] G. C. Charles K. Chui, Kalman Filtering: With Real Time Applications.

Springer-Verlag Berlin Heidelberg, third ed., 2009.
[37] L. Kleeman, “Kalman filter understanding and application,” tech. rep.,

Monash University, Clayton, 1996.
[38] D. Simon, “Kalman filtering in embedded system programming,” tech.

rep., Cleveland State University, 2001.

Modelling and Control of Large Wind

Uploaded by

Copyright:

Available Formats

Modelling and Control of Large Wind

Uploaded by

Document Information

Original Description:

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

Modelling and Control of Large Wind

Uploaded by

Copyright:

Available Formats

Modelling and Control of Large Wind

Modellering och reglering av stora vindkraftverk

Syed Hammad Zafar

Faculty of Health, Science and Technology

Syed Hammad Zafar i

Secondly, my sincere thanks to the government of Sweden and Karlstad

I would like to express the deepest appreciation to my external supervisor

I would like to thank my internal supervisor Dr. Jorge Solis, Associate

Syed Hammad Zafar ii

3 Modelling of Wind Turbine 19

Syed Hammad Zafar iii

3.4 Linearised Model Calculations . . . . . . . . . . . . . . . . . . 25

6 Conclusions and Future Work 66

A Mathematical Model Calculations 69

B Coleman’s Transformation Calculation 94

C Matlab Script 100

Syed Hammad Zafar iv

Syed Hammad Zafar v

1.1 Energy Research Centre of the Netherlands . . . . . . . . . . 1

2.1 Horizontal axis upwind wind turbine . . . . . . . . . . . . . . 8

3.1 Turbine with blade and tower hinge points . . . . . . . . . . 20

4.1 Open Loop System . . . . . . . . . . . . . . . . . . . . . . . . 36

Syed Hammad Zafar vi

4.15 Pitch Activity . . . . . . . . . . . . . . . . . . . . . . . . . . . 52

5.1 Schematic diagram of Kalman Filter implementation . . . . 56

6.1 Blade layout with two hinge points . . . . . . . . . . . . . . . 68

B.1 Coleman Transformation & Inverse Coleman Transformation

Syed Hammad Zafar vii

HAWT Horizontal Axis Wind Turbine

Syed Hammad Zafar viii

1.1 Energy Research Centre of the Netherlands

Figure 1.1: Energy Research Centre of the Netherlands

Syed Hammad Zafar 1

design programming for more efficient and reliable turbine working

• Solar power: Latest scientific research is applied in collaboration with

• Biomass: Focus on thermochemical conversion of biomass and biobased

• Energy efficiency: Focused on making industrial process more efficient

• Environment & Energy Engineering: Focused on innovative businesses

• Policy studies: Strategic advices are developed by combined efforts

1.2 Historical Development of Wind Energy

Syed Hammad Zafar 2

• Increase in turbine size.

• Placing turbine at high wind speed location.

Commercial turbine manufacturing companies are focused on large wind tur-

Syed Hammad Zafar 3

• Mass Spring Damper Configuration (MSD)

• Tuned Circuit Configuration (LC circuit)

• Extreme and fatigue load reduction of wind turbine.

• Enables the turbine operation in changing and extreme weather.

Syed Hammad Zafar 4

• Individual pitch control technique improves the trade-off between en-

1.5 Problem Definition

1.6 Research Objectives

Syed Hammad Zafar 5

From academic point of view, previous knowledge and experience gained

• Investigate Current Control Design: Understanding and analysis

• Turbine Modelling: Detailed model designing for the large wind

• Model Analysis: Model analysis using the current control design

• Multivariable Control Implementation: New proposed multivari-